User-Guide-4 4 1

Nektar++: Spectral/hp
Element Framework
Version 4.4.1
User Guide
October 13, 2017
Department of Aeronautics, Imperial College London, UK

Scientific Computing and Imaging Institute, University of Utah, USA
ii
Contents
Introduction x
I Getting Started 1
1 Installation 2
1.1 Installing Debian/Ubuntu Packages . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Installing Redhat/Fedora Packages . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Installing from Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.1 Obtaining the source code . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.2 Linux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.3 OS X . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.4 Windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.5 CMake Option Reference . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Compiling Documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.1 Dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.2 Compiling the User Guide . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.3 Compiling the code documentation . . . . . . . . . . . . . . . . . . 17
1.5 Compiling Tutorials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2 Mathematical Formulation 18
2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Methods overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.1 The finite element method (FEM) . . . . . . . . . . . . . . . . . . 19
2.2.2 High-order finite element methods . . . . . . . . . . . . . . . . . . 19
2.2.3 The Galerkin formulation . . . . . . . . . . . . . . . . . . . . . . . 21
3 XML Session File 23

3.1 Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.1 Vertices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.2 Edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
iii
iv Contents
3.1.3 Faces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.4 Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.5 Curved Edges and Faces . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1.6 Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1.7 Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.1 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.2 Solver Information . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.3 Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.4 Global System Solution Information . . . . . . . . . . . . . . . . . 30
3.3.5 Boundary Regions and Conditions . . . . . . . . . . . . . . . . . . 34
3.3.6 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.7 Quasi-3D approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4.1 FieldConvert checkpoints . . . . . . . . . . . . . . . . . . . . . . . 41
3.4.2 Time-averaged fields . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4.3 Moving average of fields . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.4 Reynolds stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.5 Checkpoint fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4.6 History points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4.7 ThresholdMax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4.8 ThresholdMin value . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4.9 One-dimensional energy . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4.10 Modal energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4.11 Aerodynamic forces . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4.12 Kinetic energy and enstrophy . . . . . . . . . . . . . . . . . . . . . 48
3.5 Forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.1 Absorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.2 Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5.3 Programmatic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5.4 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.6 Analytic Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.6.1 Variables and coordinate systems . . . . . . . . . . . . . . . . . . . 51
3.6.2 Performance considerations . . . . . . . . . . . . . . . . . . . . . . 56
II Preprocessing & Postprocessing 58
4 NekMesh 59
4.1 Exporting a mesh from Gmsh . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2 Defining physical surfaces and volumes . . . . . . . . . . . . . . . . . . . . 60
4.3 Converting the MSH to Nektar++ format . . . . . . . . . . . . . . . . . . 61
4.4 NekMesh modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.4.1 Input modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Contents v
4.4.2 Output modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.4.3 Extract surfaces from a mesh . . . . . . . . . . . . . . . . . . . . . 65
4.4.4 Negative Jacobian detection . . . . . . . . . . . . . . . . . . . . . . 66
4.4.5 Spherigon patches . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.4.6 Periodic boundary condition alignment . . . . . . . . . . . . . . . . 67
4.4.7 Boundary layer splitting . . . . . . . . . . . . . . . . . . . . . . . . 68
4.4.8 High-order cylinder generation . . . . . . . . . . . . . . . . . . . . 69
4.4.9 Linearisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.4.10 Extracting interface between tetrahedra and prismatic elements . . 71
4.4.11 Boundary identification . . . . . . . . . . . . . . . . . . . . . . . . 71
4.4.12 Scalar function curvature . . . . . . . . . . . . . . . . . . . . . . . 71
4.4.13 Link Checking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.14 2D mesh extrusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.15 Variational Optimisation . . . . . . . . . . . . . . . . . . . . . . . 72
4.5 Mesh generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.5.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.5.2 Mesh generation manual . . . . . . . . . . . . . . . . . . . . . . . . 76
5 FieldConvert 78
5.1 Convert .fld / .chk files into Paraview, VisIt or Tecplot format . . . . . . 78
5.2 Convert field files between XML and HDF5 format . . . . . . . . . . . . . 79
5.3 Range option -r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.4 FieldConvert modules -m . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.4.1 Smooth the data: C0Projection module . . . . . . . . . . . . . . . 81
5.4.2 Calculate Q-Criterion: QCriterion module . . . . . . . . . . . . . . 82
5.4.3 Add composite ID: addcompositeid module . . . . . . . . . . . . . 82
5.4.4 Sum two .fld files: addFld module . . . . . . . . . . . . . . . . . . 82
5.4.5 Combine two .fld files containing time averages: combineAvg module 83
5.4.6 Concatenate two files: concatenate module . . . . . . . . . . . . . 83
5.4.7 Equi-spaced output of data: equispacedoutput module . . . . . . . 83
5.4.8 Extract a boundary region: extract module . . . . . . . . . . . . . 83
5.4.9 Compute the gradient of a field: gradient module . . . . . . . . . . 84
5.4.10 Extract a plane from 3DH1D expansion: homplane module . . . . 84
5.4.11 Stretch a 3DH1D expansion: homstretch module . . . . . . . . . . 85
5.4.12 Inner Product of a single or series of fields with respect to a single
or series of fields: innerproduct module . . . . . . . . . . . . . . . . 85
5.4.13 Interpolate one field to another: interpfield module . . . . . . . . . 86
5.4.14 Interpolate scattered point data to a field: interppointdatatofld
module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.4.15 Interpolate a field to a series of points: interppoints module . . . . 87
5.4.16 Isocontour extraction: iscontour module . . . . . . . . . . . . . . . 89
5.4.17 Show high frequency energy of the Jacobian: jacobianenergy module 89
5.4.18 Calculate mesh quality: qualitymetric module . . . . . . . . . . . . 90
5.4.19 Extract mean mode of 3DH1D expansion: meanmode module . . . 90
5.4.20 Project point data to a field: pointdatatofld module . . . . . . . . 90
vi Contents
5.4.21 Print L2 and LInf norms: printfldnorms module . . . . . . . . . . 91

5.4.22 Computes the scalar gradient: scalargrad module . . . . . . . . . . 92
5.4.23 Scale a given .fld: scaleinputfld module . . . . . . . . . . . . . . . . 92
5.4.24 Time-averaged shear stress metrics: shear module . . . . . . . . . 92
5.4.25 Boundary layer height calculation: surfdistance module . . . . . . 93
5.4.26 Calculate vorticity: vorticity module . . . . . . . . . . . . . . . . . 93
5.4.27 Computing the wall shear stress: wss module . . . . . . . . . . . . 93
5.4.28 Manipulating meshes with FieldConvert . . . . . . . . . . . . . . . 94
5.5 FieldConvert in parallel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.6 Processing large files in serial . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.6.1 Using the nprocs and procid options . . . . . . . . . . . . . . . . 95
5.6.2 Using the –part-only and –part-only-overlapping options . . . . . . 96
IIISolver Applications 98
6 Advection-Diffusion-Reaction Solver 99
6.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.3 Session file configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.3.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.3.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.3.3 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.4.1 1D Advection equation . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.4.2 2D Helmholtz Problem . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.4.3 Advection dominated mass transport in a pipe . . . . . . . . . . . 106
7 Acoustic Perturbation Equations Solver 111

7.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.3.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.3.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.3.3 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.4.1 Aeroacoustic Wave Propagation . . . . . . . . . . . . . . . . . . . . 112
8 Cardiac Electrophysiology Solver 115

8.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.1.1 Bidomain Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.1.2 Monodomain Model . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.1.3 Cell Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
8.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Contents vii
8.3.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

8.3.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.3.3 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
8.3.4 Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
8.3.5 Stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
9 Compressible Flow Solver 122

9.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
9.1.1 Euler equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
9.1.2 Compressible Navier-Stokes equations . . . . . . . . . . . . . . . . 123
9.1.3 Numerical discretisation . . . . . . . . . . . . . . . . . . . . . . . . 123
9.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
9.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
9.4.1 Shock capturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
9.4.2 Variable polynomial order . . . . . . . . . . . . . . . . . . . . . . . 132
9.4.3 De-Aliasing Techniques . . . . . . . . . . . . . . . . . . . . . . . . 132
10 Incompressible Navier-Stokes Solver 134

10.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
10.1.1 Velocity Correction Scheme . . . . . . . . . . . . . . . . . . . . . . 134
10.1.2 Direct solver (coupled approach) . . . . . . . . . . . . . . . . . . . 141
10.1.3 Linear Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . 142
10.1.4 Steady-state solver using Selective Frequency Damping . . . . . . . 145
10.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
10.3.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
10.3.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
10.3.3 Womersley Boundary Condition . . . . . . . . . . . . . . . . . . . 149
10.3.4 Forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
10.4 Stability analysis Session file configuration . . . . . . . . . . . . . . . . . . 153
10.4.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
10.4.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
10.4.3 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
10.5 Steady-state solver Session file configuration . . . . . . . . . . . . . . . . . 155
10.5.1 Execution of the classical steady-state solver . . . . . . . . . . . . 156
10.5.2 Execution of the adaptive steady-state solver . . . . . . . . . . . . 156
10.6 Coordinate transformations Session file configuration . . . . . . . . . . . . 157
10.6.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
10.6.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
10.6.3 Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
10.6.4 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
10.6.5 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 159
10.7 Adaptive polynomial order Session file configuration . . . . . . . . . . . . 160
10.7.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
viii Contents
10.7.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

10.7.3 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
10.7.4 Restarting the simulation . . . . . . . . . . . . . . . . . . . . . . . 161
10.8 Advecting extra passive scalar fields . . . . . . . . . . . . . . . . . . . . . 161
10.9 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
10.9.1 Kovasznay Flow 2D . . . . . . . . . . . . . . . . . . . . . . . . . . 162
10.9.2 Kovasznay Flow 2D using high-order outflow boundary conditions 165
10.9.3 Steady Kovasznay Oseen Flow using the direct solver . . . . . . . . 167
10.9.4 Laminar Channel Flow 2D . . . . . . . . . . . . . . . . . . . . . . 168
10.9.5 Laminar Channel Flow 3D . . . . . . . . . . . . . . . . . . . . . . 169
10.9.6 Laminar Channel Flow Quasi-3D . . . . . . . . . . . . . . . . . . . 171
10.9.7 Turbulent Channel Flow . . . . . . . . . . . . . . . . . . . . . . . . 172
10.9.8 Turbulent Pipe Flow . . . . . . . . . . . . . . . . . . . . . . . . . . 175
10.9.9 Aortic Blood Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
10.9.10 finite-strip modeling of flow past flexible cables . . . . . . . . . . . 180
10.9.11 2D direct stability analysis of the channel flow . . . . . . . . . . . 183
10.9.12 2D adjoint stability analysis of the channel flow . . . . . . . . . . . 186
10.9.13 2D Transient Growth analysis of a flow past a backward-facing step190
10.9.14 BiGlobal Floquet analysis of a of flow past a cylinder . . . . . . . 194
11 Linear elasticity solver 198

11.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
11.1.1 The linear elasticity equations . . . . . . . . . . . . . . . . . . . . . 198
11.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
11.3.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
11.3.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
11.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
11.4.1 L-shaped domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
11.4.2 Boundary layer deformation . . . . . . . . . . . . . . . . . . . . . . 201
12 Pulse Wave Solver 203

12.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
12.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
12.3.1 Pulse Wave Sovler mesh connectivity . . . . . . . . . . . . . . . . . 204
12.3.2 Session Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
12.3.3 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
12.3.4 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 207
12.3.5 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
12.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
12.4.1 Human Vascular Network . . . . . . . . . . . . . . . . . . . . . . . 208
12.4.2 Stented Artery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
12.4.3 Stented Artery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
12.5 Further Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
Contents ix
12.6 Future Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
13 Shallow Water Solver 219

13.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
13.1.1 The Shallow Water Equations . . . . . . . . . . . . . . . . . . . . . 219
13.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
13.3.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
13.3.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
13.3.3 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
13.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
13.4.1 Rossby modon case . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
IV Reference 223
14 Optimisation 224
14.1 Operator evaluation strategies . . . . . . . . . . . . . . . . . . . . . . . . . 224
14.1.1 Selecting an operator strategy . . . . . . . . . . . . . . . . . . . . . 225
14.1.2 XML syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
14.1.3 Selecting different operator strategies . . . . . . . . . . . . . . . . . 226
14.2 Collections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
14.2.1 Default implementation . . . . . . . . . . . . . . . . . . . . . . . . 227
14.2.2 Auto-tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
14.2.3 Manual selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
14.2.4 Collection size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
15 Command-line Options 229
16 Frequently Asked Questions 231

16.1 Compilation and Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
16.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Bibliography 235
Introduction
Nektar++ [8] is a tensor product based finite element package designed to allow one
to construct efficient classical low polynomial order h-type solvers (where h is the size
of the finite element) as well as higher p-order piecewise polynomial order solvers. The
framework currently has the following capabilities:
• Representation of one, two and three-dimensional fields as a collection of piecewise

continuous or discontinuous polynomial domains.
• Segment, plane and volume domains are permissible, as well as domains representing
curves and surfaces (dimensionally-embedded domains).
• Hybrid shaped elements, i.e triangles and quadrilaterals or tetrahedra, prisms and
hexahedra.
• Both hierarchical and nodal expansion bases.
• Continuous or discontinuous Galerkin operators.
• Cross platform support for Linux, Mac OS X and Windows.
The framework comes with a number of solvers and also allows one to construct a variety
of new solvers.
Our current goals are to develop:
• Automatic auto-tuning of optimal operator implementations based upon not only

h and p but also hardware considerations and mesh connectivity.
• Temporal and spatial adaption.
• Features enabling evaluation of high-order meshing techniques.
x
Introduction xi
For further information and to download the software, visit the Nektar++ website at
http://www.nektar.info.
Part I
Getting Started
1
Chapter 1
Installation
Nektar++ is available in both a source-code distribution and as pre-compiled binary

packages for a number of operating systems. We recommend using the pre-compiled
packages if you wish to use the existing Nektar++ solvers for simulation and do not need
to perform additional code development.
1.1 Installing Debian/Ubuntu Packages
Binary packages are available for current Debian/Ubuntu based Linux distributions.
These can be installed through the use of standard system package management utilities,
such as Apt, if administrative access is available.
1. Add the appropriate line for the Debian-based distribution to the end of the file
/etc/apt/sources.list
Distribution Repository
Debian 8.0 (jessie) deb http://www.nektar.info/debian-jessie jessie contrib
Debian 7.0 (wheezy) deb http://www.nektar.info/debian-wheezy wheezy contrib
Ubuntu 14.04 (trusty) deb http://www.nektar.info/ubuntu-trusty trusty contrib
2. Update the package lists
apt-get update
3. Install the required Nektar++ packages, or the complete suite with:
apt-get install nektar++
Any additional dependencies required for Nektar++ to function will be automati-

cally installed.
2
1.2 Installing Redhat/Fedora Packages 3
Tip
Nektar++ is split into multiple packages for the different components of
the software. A list of available Nektar++ packages can be found using:
apt-cache search nektar++
1.2 Installing Redhat/Fedora Packages
Add a file to the directory /etc/yum.repos.d/nektar.repo with the following contents
[Nektar]
name=nektar
baseurl=<baseurl>
substituting <baseurl> for the appropriate line from the table below.
Distribution <baseurl>
Fedora 20 http://ww.nektar.info/fedora/20/$basearch
Note
The $basearch variable is automatically replaced by Yum with the architecture
of your system.
1.3 Installing from Source
This section explains how to build Nektar++ from the source-code package.
Nektar++ uses a number of third-party libraries. Some of these are required, others are
optional. It is generally more straightforward to use versions of these libraries supplied
pre-packaged for your operating system, but if you run into difficulties with compilation
errors or failing regression tests, the Nektar++ build system can automatically build
tried-and-tested versions of these libraries for you. This requires enabling the relevant
options in the CMake configuration.
1.3.1 Obtaining the source code

There are two ways to obtain the source code for Nektar++:
• Download the latest source-code archive from the Nektar++ downloads page.
4 Chapter 1 Installation
• Clone the git repository
– Using anonymous access. This does not require credentials but any changes
to the code cannot be pushed to the public repository. Use this initially if you
would like to try using Nektar++ or make local changes to the code.
git clone http://gitlab.nektar.info/clone/nektar/nektar.git nektar++
– Using authenticated access. This will allow you to directly contribute back
into the code.
git clone git@gitlab.nektar.info:nektar/nektar.git nektar++
Tip
You can easily switch to using the authenticated access from anony-
mous access at a later date.
1.3.2 Linux
1.3.2.1 Prerequisites
Nektar++ uses a number of external programs and libraries for some or all of its
functionality. Some of these are required and must be installed prior to compiling
Nektar++, most of which are available as pre-built system packages on most Linux
distributions or can be installed manually by a user. Others are optional and required
only for specific features, or can be downloaded and compiled for use with Nektar++
automatically (but not installed system-wide).
1.3 Installing from Source 5
Installation
Package Req. Sys. User Auto. Note
C++ compiler 3 3 gcc, icc, etc
CMake > 2.8.7 3 3 3 Ncurses GUI optional
BLAS 3 3 3 Or MKL, ACML, OpenBLAS
LAPACK 3 3 3
Boost >= 1.52 3 3 3 3 Compile with iostreams
ModMETIS 3 3
FFTW > 3.0 3 3 3 For high-performance FFTs
ARPACK > 2.0 3 3 For arnoldi algorithms
OpenMPI 3 For parallel execution
GSMPI 3 For parallel execution
HDF5 3 3 3 For large-scale parallel I/O (requires CMake >3.1)
PETSc 3 3 Alternative linear solvers
Scotch 3 3 3 Alternative mesh partitioning
VTK > 5.8 3 3 Visualisation utilities
Warning
Boost version 1.51 has a bug which prevents Nektar++ working correctly. Please
use a newer version.
1.3.2.2 Quick Start

Open a terminal.
If you have downloaded the tarball, first unpack it:
tar -zxvf nektar++-4.4.1.tar.gz
Change into the nektar++ source code directory
mkdir -p build && cd build

ccmake ../
make install
1.3.2.3 Detailed instructions

From a terminal:
1. If you have downloaded the tarball, first unpack it

2. Change into the source-code directory, create a build subdirectory and enter it
3. Run the CMake GUI and configure the build by pressing c
ccmake ../
• Select the components of Nektar++ (prefixed with NEKTAR_BUILD_ ) you would

like to build. Disabling solvers which you do not require will speed up the
build process.
• Select the optional libraries you would like to use (prefixed with NEKTAR_USE_ )
for additional functionality.
• Select the libraries not already available on your system which you wish to be
compiled automatically (prefixed with THIRDPARTY_BUILD_ )
A full list of configuration options can be found in Section 1.3.5.
Note
Selecting THIRDPARTY_BUILD_ options will request CMake to automatically
download thirdparty libraries and compile them within the Nektar++ direc-
tory. If you have administrative access to your machine, it is recommended
to install the libraries system-wide through your package-management
system.
4. Press c to configure the build. If errors arise relating to missing libraries, review
the THIRDPARTY_BUILD_ selections in the configuration step above or install the
missing libraries manually or from system packages.
5. When configuration completes without errors, press c again until the option g to
generate build files appears. Press g to generate the build files and exit CMake.
6. Compile the code
make install
During the build, missing third-party libraries will be automatically downloaded,

configured and built in the Nektar++ build directory.
Tip
If you have multiple processors/cores on your system, compilation can be
significantly increased by adding the -jX option to make, where X is the
number of simultaneous jobs to spawn. For example, use
make -j4 install
on a quad-core system.
7. Test the build by running unit and regression tests.
ctest
1.3.3 OS X
1.3.3.1 Prerequisites
Nektar++ uses a number of external programs and libraries for some or all of its
functionality. Some of these are required and must be installed prior to compiling
Nektar++, most of which are available on MacPorts (www.macports.org) or can be
installed manually by a user. Others are optional and required only for specific features,
or can be downloaded and compiled for use with Nektar++ automatically (but not
installed system-wide).
Note
To compile Nektar++ on OS X, Apple’s Xcode Developer Tools must be
installed. They can be installed either from the App Store (only on Mac
OS 10.7 and above) or downloaded directly from http://connect.apple.com/
(you are required to have an Apple Developer Connection account). Xcode
includes Apple implementations of BLAS and LAPACK (called the Accelerate
Framework).
Installation
Package Req. MacPorts User Auto. Note
Xcode 3 Provides developer tools
CMake > 2.8.7 3 cmake 3 Ncurses GUI optional
BLAS 3 Part of Xcode
LAPACK 3 Part of Xcode
Boost >= 1.52 3 boost 3 3 Compile with iostreams
TinyXML 3 tinyxml 3 3
ModMETIS 3 3
FFTW > 3.0 fftw-3 3 3 For high-performance FFTs
ARPACK > 2.0 arpack 3 For arnoldi algorithms
OpenMPI openmpi For parallel execution
GSMPI 3 For parallel execution
HDF5 3 3 3 For large-scale parallel I/O (requires CMake >3.1)
PETSc petsc 3 3 Alternative linear solvers
Scotch scotch 3 3 Alternative mesh partitioning
VTK > 5.8 vtk 3 Visualisation utilities
Tip
CMake, and some other software, is available from MacPorts (http://macports.
org) and can be installed using, for example,
sudo port install cmake
Package names are given in the table above. Similar packages also exist in
other package managers such as Homebrew.
1.3.3.2 Quick Start

Open a terminal (Applications->Utilities->Terminal).
If you have downloaded the tarball, first unpack it:
Change into the nektar++ source code directory

ccmake ../
make install

From a terminal (Applications->Utilities->Terminal):
1. If you have downloaded the tarball, first unpack it
2. Change into the source-code directory, create a build subdirectory and enter it
3. Run the CMake GUI and configure the build
ccmake ../
Use the arrow keys to navigate the options and ENTER to select/edit an option.
• Select the components of Nektar++ (prefixed with NEKTAR_BUILD_ ) you would

like to build. Disabling solvers which you do not require will speed up the
build process.
• Select the optional libraries you would like to use (prefixed with NEKTAR_USE_ )
for additional functionality.
• Select the libraries not already available on your system which you wish to be
compiled automatically (prefixed with THIRDPARTY_BUILD_ )
•
A full list of configuration options can be found in Section 1.3.5.
Note
Selecting THIRDPARTY_BUILD_ options will request CMake to automatically
download thirdparty libraries and compile them within the Nektar++ direc-
tory. If you have administrative access to your machine, it is recommended
to install the libraries system-wide through MacPorts.
4. Press c to configure the build. If errors arise relating to missing libraries (variables
set to NOTFOUND ), review the THIRDPARTY_BUILD_ selections in the previous step
or install the missing libraries manually or through MacPorts.
5. When configuration completes without errors, press c again until the option g to
generate build files appears. Press g to generate the build files and exit CMake.
6. Compile the code

make install
During the build, missing third-party libraries will be automatically downloaded,

configured and built in the Nektar++ build directory.
Tip
If you have multiple processors/cores on your system, compilation
can be significantly increased by adding the -jX option to make,
where X is the number of simultaneous jobs to spawn. For example,
make -j4 install
7. Test the build by running unit and regression tests.
ctest
1.3.4 Windows
Windows compilation is supported, but the build process is somewhat convoluted at
present. As such, only serial execution is supported with a minimal amount of additional
build packages. These can either be installed by the user, or automatically in the build
process.
Installation
Package Req. User Auto. Note
MS Visual Studio 3 3 2012, 2013 and 2015 known working
CMake ≥ 3.0 3 3
BLAS 3 3 3
LAPACK 3 3 3
Boost ≥ 1.55 3 3 3 Compile with iostreams
ModMETIS 3 3 3
1. Install Microsoft Visual Studio 2015 (preferred), 2013 or 2012 (known to work). This
can be obtained from Microsoft free of charge by using their Community developer
tools from https://www.visualstudio.com/en-us/products/visual-studio-community-vs.
aspx.
2. Install WinRAR from http://www.rarlab.com/download.htm.

3. Install CMake 3.0+ from http://www.cmake.org/download/. When prompted,

select the option to add CMake to the system PATH.
4. (Optional) Install Git from http://git-scm.com/download/win to use the devel-

opment versions of Nektar++. When prompted, select the option to add Git to
the system PATH. You do not need to select the option to add Unix tools to the
PATH.
5. (Optional) If you do not wish to build boost during the compilation process (which
can take some time), then boost binaries can be found at http://sourceforge.
net/projects/boost/files/boost-binaries/1.61.0/. By default these install
into C:\local\boost_1_61_0 . If you use these libraries, you will need to:
• Rename libs-msvc14.0 to lib

• Inside the lib directory, create duplicates of boost_zlib.dll and boost_bzip2.dll
called zlib.dll and libbz2.dll
• Add a BOOST_HOME environment variable. To do so, navigate to Control Panel
> System and Security > System, select Advanced System Settings, and in the
Advanced tab click the Environment Variables. In the System Variables box,
click New. In the New System Variable window, type BOOST_HOME next to
Variable name and C:\local\boost_1_61_0 next toVariable value.
6. Unpack nektar++-4.4.1.tar.gz using WinRAR.
Note
Some Windows versions do not recognise the path of a folder which has
++ in the name. If you think that your Windows version can not handle
path containing special characters, you should rename nektar++-4.4.1
to nektar-4.4.1 .
7. Create a builds directory within the nektar++-4.4.1 subdirectory.
8. Open a Visual Studio terminal. From the start menu, this can be found in All
Programs > Visual Studio 2015 > Visual Studio Tools > Developer Command
Prompt for VS2015.
9. Change directory into the builds directory and run the CMake graphical utility:
cd C:\path\to\nektar\builds
cmake-gui ..
10. Select the build system you want to generate build scripts for. Note that Visual
Studio 2015 is listed as Visual Studio 14 in the drop-down list. If you have a 64-bit
installation of Windows, you should select the Win64 variant, otherwise 32-bit
executables will be generated. Select the option to use the native compilers.
11. Click the Configure button, then the Generate button.
12. Return to the command line and issue the command:
msbuild INSTALL.vcxproj /p:Configuration=Release
To build in parallel with, for example, 12 processors, issue:
msbuild INSTALL.vcxproj /p:Configuration=Release /m:12
13. After the installation process is completed, the executables will be available in
builds\dist\bin .
14. To use these executables, you need to modify your system PATH to include the
library directories where DLLs are stored. To do this, navigate to Control Panel
> System and Security > System, select Advanced System Settings, and in the
Advanced tab click the Environment Variables. In the System Variables box, select
Path and click Edit. To the end of this list, add the full paths to directories:
• builds\dist\lib\nektar++-4.4.1
• builds\dist\bin
• Optionally, if you installed Boost from the binary packages, C:\local\boost_1_61_0 \lib
15. To run the test suite, open a new command line window, change to the builds
directory, and then issue the command
ctest -C Release
1.3.5 CMake Option Reference

This section describes the main configuration options which can be set when building
Nektar++. The default options should work on almost all systems, but additional
features (such as parallelisation and specialist libraries) can be enabled if needed.
1.3.5.1 Components
The first set of options specify the components of the Nektar++ toolkit to compile. Some
options are dependent on others being enabled, so the available options may change.
Components of the Nektar++ package can be selected using the following options:
• NEKTAR_BUILD_DEMOS (Recommended)
Compiles the demonstration programs. These are primarily used by the regression
testing suite to verify the Nektar++ library, but also provide an example of the
basic usage of the framework.
• NEKTAR_BUILD_DOC
Compiles the Doxygen documentation for the code. This will be put in
$BUILDDIR/doxygen/html
• NEKTAR_BUILD_LIBRARY (Required)
Compiles the Nektar++ framework libraries. This is required for all other options.
• NEKTAR_BUILD_SOLVERS (Recommended)
Compiles the solvers distributed with the Nektar++ framework.
If enabling NEKTAR_BUILD_SOLVERS , individual solvers can be enabled or disabled.
See Part III for the list of available solvers. You can disable solvers which are not
required to reduce compilation time. See the NEKTAR_SOLVER_X option.
• NEKTAR_BUILD_TESTS (Recommended)
Compiles the testing program used to verify the Nektar++ framework.
• NEKTAR_BUILD_TIMINGS
Compiles programs used for timing Nektar++ operations.
• NEKTAR_BUILD_UNIT_TESTS
Compiles tests for checking the core library functions.
• NEKTAR_BUILD_UTILITIES
Compiles utilities for pre- and post-processing simulation data.
• NEKTAR_SOLVER_X
Enabled compilation of the ’X’ solver.
A number of ThirdParty libraries are required by Nektar++. There are also optional
libraries which provide additional functionality. These can be selected using the following
options:
• NEKTAR_USE_ACML
Use the optimised BLAS library for AMD processors.
• NEKTAR_USE_ACCELERATE_FRAMEWORK
Use the Mac Osx accelerate framework for BLAS and LAPACK methods. This
option should only be required under in a Mac OSX environment.
• NEKTAR_USE_ARPACK
Build Nektar++ with support for ARPACK. This provides routines used for linear
stability analyses. Alternative Arnoldi algorithms are also implemented directly in
Nektar++.
• NEKTAR_USE_BLAS_LAPACK (Required)
Enables the use of Basic Linear Algebra Subroutines libraries for linear algebra
operations.
• NEKTAR_USE_SYSTEM_BLAS_LAPACK (Recommended)
On Linux systems, use the default BLAS and LAPACK library on the system.
This may not be the implementation offering the highest performance for your
architecture, but it is the most likely to work without problem.
• NEKTAR_USE_CCM
Use the ccmio library provided with the Star-CCM package for reading ccm files.
This option is required as part of NekMesh if you wish to convert a Star-CCM mesh
into the Nektar format. It is possible to read a Tecplot plt file from Star-CCM
but this output currently needs to be converted to ascii format using the Tecplot
package.
• NEKTAR_USE_FFTW
Build Nektar++ with support for FFTW for performing Fast Fourier Transforms
(FFTs). This is used only when using domains with homogeneous coordinate
directions.
• NEKTAR_USE_HDF5
Build Nektar++ with support for HDF5. This enables input/output in the HDF5
parallel file format, which can be very efficient for large numbers of processes. HDF5
output can be enabled by using a command-line option or in the SOLVERINFO
section of the XML file. This option requires that Nektar++ be built with MPI
support with NEKTAR_USE_MPI enabled and that HDF5 is compiled with MPI
support.
• NEKTAR_USE_MKL
Use the Intel MKL library. This is typically available on cluster environments and
should offer performance tuned for the specific cluster environment.
• NEKTAR_USE_MPI (Recommended)
Build Nektar++ with MPI parallelisation. This allows solvers to be run in serial
or parallel.
• NEKTAR_USE_OPENBLAS
Use OpenBLAS for the BLAS library. OpenBLAS is based on the Goto BLAS
implementation and generally offers better performance than a non-optimised
system BLAS. However, the library must be installed on the system.
• NEKTAR_USE_PETSC
Build Nektar++ with support for the PETSc package for solving linear systems.
• NEKTAR_USE_SCOTCH
Build Nektar++ with support for the SCOTCH graph partitioning library. This
provides an alternative mesh partitioning algorithm to METIS. However, METIS
is still required as it is used by the multi-level static condensation algorithm.
• NEKTAR_USE_SMV
Build Nektar++ with support for optimised sparse matrix-vector operations.
• NEKTAR_USE_VTK
Build Nektar++ with support for VTK libraries. This is only needed for specialist
utilities and is not needed for general use.
Note
The VTK libraries are not needed for converting the output of simulations
to VTK format for visualization as this is handled internally.
The THIRDPARTY_BUILD_X options select which third-party libraries are automatically

built during the Nektar++ build process. Below are the choices of X:
• BOOST
The Boost libraries are frequently provided by the operating system, so automatic
compilation is not enabled by default. If you do not have Boost on your system,
you can enable this to have Boost configured automatically.
• GSMPI
(MPI-only) Parallel communication library.
• LOKI
An implementation of a singleton.
• METIS
A graph partitioning library used for substructuring of matrices and mesh parti-
tioning when Nektar++ is run in parallel.
• PETSC
A package for the parallel solution of linear algebra systems.
• SCOTCH
An alternative graph partitioning library used for mesh partitioning when Nektar++
is run in parallel.
• TINYXML
Library for reading and writing XML files.
1.4 Compiling Documentation
Documentation for Nektar++ is provided in a number of forms:
• User Guide (LaTeX, compiled to pdf or html)
• Source code documentation (Doxygen compiled to html)
1.4.1 Dependencies
To build the LaTeX documents (user guide or tutorials), the following dependencies are
required:
• texlive-base
• texlive-latex-extra
• texlive-science
• texlive-fonts-recommended
• imagemagick
To build the Doxygen documentation, the following dependencies are required:
• doxygen
• graphviz
1.4.2 Compiling the User Guide

To compile the User Guide:
1. Configure the Nektar++ build tree as normal.
2. Run
make user-guide-pdf
to make the PDF version, or run
make user-guide-html
to make the HTML version.

1.5 Compiling Tutorials 17
1.4.3 Compiling the code documentation

To compile the code documentation enable the NEKTAR_BUILD_DOC option in the ccmake
configuration tool.
You can then compile the HTML code documentation using:
make doc
1.5 Compiling Tutorials
If you are using a clone of the Nektar++ git repository, you can also download the source
for the Nektar++ tutorials which is available as a git submodule.
1. From a Nektar++ working directory (e.g. $NEKPP ):
git submodule init

git submodule update --remote
2. From your build directory (e.g. $NEKPP/build ), re-run cmake to update the build
system to include the tutorials
cmake ../
3. Compile each required tutorial, for example
make flow-stability-channel
Chapter 2
Mathematical Formulation
2.1 Background
The spectral/hp element method combines the geometric flexibility of classical h-type
finite element techniques with the desirable resolution properties of spectral methods. In
this approach a polynomial expansion of order P is applied to every elemental domain of a
coarse finite element type mesh. These techniques have been applied in many fundamental
studies of fluid mechanics [34] and more recently have gained greater popularity in the
modelling of wave-based phenomena such as computational electromagnetics [17] and
shallow water problems [5] - particularly when applied within a Discontinuous Galerkin
formulation.
There are at least two major challenges which arise in developing an efficient implemen-
tation of a spectral/hp element discretisation:
• implementing the mathematical structure of the technique in a digestible, generic

and coherent manner, and
• designing and implementing the numerical methods and data structures in a matter
so that both high- and low-order discretisations can be efficiently applied.
In order to design algorithms which are efficient for both low- and high-order spectral/hp
discretisations, it is important clearly define what we mean with low- and high-order.
The spectral/hp element method can be considered as bridging the gap between the
high-order end of the traditional finite element method and low-order end of conventional
spectral methods. However, the concept of high- and low-order discretisations can mean
very different things to these different communities. For example, the seminal works by
Zienkiewicz & Taylor [40] and Hughes list examples of elemental expansions only up to
third or possibly fourth-order, implying that these orders are considered to be high-order
for the traditional h-type finite element community. In contrast the text books of the
spectral/hp element community typically show examples of problems ranging from a
18
2.2 Methods overview 19
low-order bound of minimally fourth-order up to anything ranging from 10th -order to

15th -order polynomial expansions. On the other end of the spectrum, practitioners of
global (Fourier-based) spectral methods [15] would probably consider a 16th -order global
expansion to be relatively low-order approximation.
One could wonder whether these different definitions of low- and high-order are just
inherent to the tradition and lore of each of the communities or whether there are more
practical reasons for this distinct interpretation. Proponents of lower-order methods might
highlight that some problems of practical interest are so geometrically complex that one
cannot computationally afford to use high-order techniques on the massive meshes required
to capture the geometry. Alternatively, proponents of high-order methods highlight that
if the problem of interest can be captured on a computational domain at reasonable
cost then using high-order approximations for sufficiently smooth solutions will provide a
higher accuracy for a given computational cost. If one however probes even further it also
becomes evident that the different communities choose to implement their algorithms
in different manners. For example the standard h-type finite element community will
typically uses techniques such as sparse matrix storage formats (where only the non-zero
entries of a global matrix are stored) to represent a global operator. In contrast the
spectral/hp element community acknowledges that for higher polynomial expansions
more closely coupled computational work takes place at the individual elemental level
and this leads to the use of elemental operators rather than global matrix operators. In
addition the global spectral method community often make use of the tensor-product
approximations where products of one-dimensional rules for integration and differentiation
can be applied.
2.2 Methods overview
Here a review of some terminology in order to situate the spectral/hp element method
within the field of the finite element methods.
2.2.1 The finite element method (FEM)

Nowadays, the finite element method is one of the most popular numerical methods in the
field of both solid and fluid mechanics. It is a discretisation technique used to solve (a set
of) partial differential equations in its equivalent variational form. The classical approach
of the finite element method is to partition the computational domain into a mesh of
many small subdomains and to approximate the unknown solution by piecewise linear
interpolation functions, each with local support. The FEM has been widely discussed
in literature and for a complete review of the method, the reader is also directed to the
seminal work of Zienkiewicz and Taylor [40].
2.2.2 High-order finite element methods

While in the classical finite element method the solution is expanded in a series of linear
basis functions, high-order FEMs employ higher-order polynomials to approximate the
20 Chapter 2 Mathematical Formulation
solution. For the high-order FEM, the solution is locally expanded into a set of P + 1
linearly independent polynomials which span the polynomial space of order P . Confusion
may arise about the use of the term order. While the order, or degree, of the expansion
basis corresponds to the maximal polynomial degree of the basis functions, the order of
the method essentially refers to the accuracy of the approximation. More specifically, it
depends on the convergence rate of the approximation with respect to mesh-refinement.
It has been shown by Babuska and Suri [3], that for a sufficiently smooth exact solution
u ∈ H k (Ω), the error of the FEM approximation uδ can be bounded by:
||u − uδ ||E ≤ ChP ||u||k .
This implies that when decreasing the mesh-size h, the error of the approximation
algebraically scales with the P th power of h. This can be formulated as:
||u − uδ ||E = O(hP ).
If this holds, one generally classifies the method as a P th -order FEM. However, for
non-smooth problems, i.e. k < P + 1, the order of the approximation will in general be
lower than P , the order of the expansion.
2.2.2.1 h-version FEM
A finite element method is said to be of h-type when the degree P of the piecewise
polynomial basis functions is fixed and when any change of discretisation to enhance
accuracy is done by means of a mesh refinement, that is, a reduction in h. Dependent
on the problem, local refinement rather than global refinement may be desired. The
h-version of the classical FEM employing linear basis functions can be classified as a
first-order method when resolving smooth solutions.
2.2.2.2 p-version FEM
In contrast with the h-version FEM, finite element methods are said to be of p-type when
the partitioning of domain is kept fixed and any change of discretisation is introduced
through a modification in polynomial degree P . Again here, the polynomial degree
may vary per element, particularly when the complexity of the problem requires local
enrichment. However, sometimes the term p-type FEM is merely used to indicated that
a polynomial degree of P > 1 is used.
2.2.2.3 hp-version FEM
In the hp-version of the FEM, both the ideas of mesh refinement and degree enhancement
are combined.
2.2 Methods overview 21
2.2.2.4 The spectral method

As opposed to the finite element methods which builds a solution from a sequence of local
elemental approximations, spectral methods approximate the solution by a truncated
series of global basis functions. Modern spectral methods, first presented by Gottlieb and
Orzag [15], involve the expansion of the solution into high-order orthogonal expansion,
typically by employing Fourier, Chebyshev or Legendre series.
2.2.2.5 The spectral element method

Patera [28] combined the high accuracy of the spectral methods with the geometric
flexibility of the finite element method to form the spectral element method. The multi-
elemental nature makes the spectral element method conceptually similar to the above
mentioned high-order finite element. However, historically the term spectral element
method has been used to refer to the high-order finite element method using a specific
nodal expansion basis. The class of nodal higher-order finite elements which have become
known as spectral elements, use the Lagrange polynomials through the zeros of the
Gauss-Lobatto(-Legendre) polynomials.
2.2.2.6 The spectral/hp element method

The spectral/hp element method, as its name suggests, incorporates both the multi-
domain spectral methods as well as the more general high-order finite element methods.
One can say that it encompasses all methods mentioned above. However, note that the
term spectral/hp element method is mainly used in the field of fluid dynamics, while the
terminology p and hp-FEM originates from the area of structural mechanics.
2.2.3 The Galerkin formulation

Finite element methods typically use the Galerkin formulation to derive the weak form
of the partial differential equation to be solved. We will primarily adopt the classical
Galerkin formulation in combination with globally C 0 continuous spectral/hp element
discretisations.
To describe the Galerkin method, consider a steady linear differential equation in a

domain Ω denoted by
L(u) = f,
subject to appropriate boundary conditions. In the Galerkin method, the weak form of
this equation can be derived by pre-multiplying this equation with a test function v and
integrating the result over the entire domain Ω to arrive at: Find u ∈ U such that
Z Z
vL(u)dx = vf dx, ∀v ∈ V,
Ω Ω
where U and V respectively are a suitably chosen trial and test space (in the traditional
Galerkin method, one typically takes U = V). In case the inner product of v and L(u)
22 Chapter 2 Mathematical Formulation
can be rewritten into a bi-linear form a(v, u), this problem is often formulated more
concisely as: Find u ∈ U such that
a(v, u) = (v, f ), ∀v ∈ V,
where (v, f ) denotes the inner product of v and f . The next step in the classical Galerkin
finite element method is the discretisation: rather than looking for the solution u in the
infinite dimensional function space U, one is going to look for an approximate solution
uδ in the reduced finite dimensional function space U δ ⊂ U. Therefore we represent the
approximate solution as a linear combination of basis functions Φn that span the space
U δ , i.e.
X
uδ = Φn ûn .
n∈N
Adopting a similar discretisation for the test functions v, the discrete problem to be
solved is given as: Find ûn (n ∈ N ) such that
X
a(Φm , Φn )ûn = (Φm , f ), ∀m ∈ N .
n∈N
It is customary to describe this set of equations in matrix form as
Aû = f̂ ,
where û is the vector of coefficients ûn , A is the system matrix with elements
Z
A[m][n] = a(Φm , Φn ) = Φm L(Φn )dx,
Ω
and the vector f̂ is given by

Z
f̂ [m] = (Φm , f ) = Φm f dx.
Ω
Chapter 3
XML Session File
The Nektar++ native file format is compliant with XML version 1.0. The root element
is NEKTAR which contains a number of other elements which describe configuration for
different aspects of the simulation. The required elements are shown below:
1 <NEKTAR>
2 <GEOMETRY>
3 ...
4 </GEOMETRY>
5 <EXPANSIONS>
6 ...
7 </EXPANSIONS>
8 <CONDITIONS>
9 ...
10 </CONDITIONS>
11 ...
12 </NEKTAR>
The different sub-elements can be split across multiple files, however each file must have a
top-level NEKTAR tag. For example, one might store the geometry information separate
from the remaining configuration in two separate files as illustrated below:
geometry.xml
1 <NEKTAR>
2 <GEOMETRY>
3 ...
4 </GEOMETRY>
5 </NEKTAR>
conditions.xml
1 <NEKTAR>
2 <CONDITIONS>
3 ...
4 </CONDITIONS>
5 <EXPANSIONS>
23
24 Chapter 3 XML Session File
6 ...
7 </EXPANSIONS>
8 ...
9 </NEKTAR>
Note
When specifying multiple files, repeated XML sub-elements are not merged.
The sub-elements from files appearing later in the list will, in general, override
those elements from earlier files.
For example, the NekMesh utility will produce a default EXPANSIONS element
and blank CONDITIONS element. Specifying a custom-written XML file con-
taining these sections after the file produced by NekMesh will override these
defaults.
The exception to this rule is when an empty XML sub-element would override a
non-empty XML sub-element. In this case the empty XML sub-element will be
ignored. If the custom-written XML file containing CONDITIONS were specified
before the file produced by NEKMESH , the empty CONDITIONS tag in the latter
file would be ignored.
3.1 Geometry
This section defines the mesh. It specifies a list of vertices, edges (in two or three
dimensions) and faces (in three dimensions) and how they connect to create the elemental
decomposition of the domain. It also defines a list of composites which are used in the
Expansions and Conditions sections of the file to describe the polynomial expansions and
impose boundary conditions.
The GEOMETRY section is structured as

1 <GEOMETRY DIM="2" SPACE="2">
2 <VERTEX> ... </VERTEX>
3 <EDGE> ... </EDGE>
4 <FACE> ... </FACE>
5 <ELEMENT> ... </ELEMENT>
6 <CURVED> ... </CURVED>
7 <COMPOSITE> ... </COMPOSITE>
8 <DOMAIN> ... </DOMAIN>
9 </GEOMETRY>
It has two (required) attributes:
• DIM specifies the dimension of the expansion elements.
• SPACE specifies the dimension of the space in which the elements exist.
3.1 Geometry 25
These attributes allow, for example, a two-dimensional surface to be embedded in a

three-dimensional space.
Note
The attribute PARTITION may also appear in a partitioned mesh. However,
this attribute should not be explicitly specified by the user.
Each of the VERTEX , EDGE , FACE , ELEMENT and CURVED sections may optionally be
compressed and stored in base64-encoded gzipped binary form, using either little-endian or
big-endian ordering, as specified by the COMPRESSED attribute to these sections. Currently
supported values are:
• B64Z-LittleEndian : Base64 Gzip compressed using little-endian ordering.
• B64Z-BigEndian : Base64 Gzip compressed using big-endian ordering.
Note
The description below explains how the GEOMETRY section is laid out in un-
compressed ascii format. From Jan 2016 the distribution uses the compressed
format for each of the above sections. To convert a compressed xml file into
ascii format use
NekMesh file.msh newfile.xml:xml:uncompress
3.1.1 Vertices
Vertices have three coordinates. Each has a unique vertex ID. In uncompressed form,
they are defined within VERTEX subsection as follows:
1 <V ID="0"> 0.0 0.0 0.0 </V> ...
The VERTEX subsection has optional attributes which can be used to apply a transforma-
tion to the mesh:
XSCALE , YSCALE , ZSCALE , XMOVE , YMOVE , ZMOVE
They specify scaling factors (centred at the origin) and translations to the vertex coordi-
nates. For example, the following snippet
1 <VERTEX XSCALE="5">
2 <V ID="0"> 0.0 0.0 0.0 </V>
3 <V ID="1"> 1.0 2.0 0.0 </V>
4 </VERTEX>
defines two vertices with coordinates (0.0, 0.0, 0.0), (1.0, 2.0, 0.0).
All of these attributes can be arbitrary analytic expressions depending on pre- defined
constants and parameters defined in the XML file and mathematical operations/functions
of the latter. If omitted, default scaling factors 1.0, and translations of 0.0, are assumed.
3.1.2 Edges
Tip
The EDGES section is only necessary when DIM=2 or DIM=3 in the parent
GEOMETRY element and may be omitted for one-dimensional meshes.
Edges are defined by two vertices. Each edge has a unique edge ID. In uncompressed
form, they are defined in the file with a line of the form
1 <E ID="0"> 0 1 </E>
3.1.3 Faces
Tip
The FACES section is only necessary when DIM=3 in the parent GEOMETRY
element and may otherwise be omitted.
Faces are defined by three or more edges. Each face has a unique face ID. They are
defined in the file with a line of the form
1 <T ID="0"> 0 1 2 </T>
2 <Q ID="1"> 3 4 5 6 </Q>
The choice of tag specified (T or Q), and thus the number of edges specified depends on
the geometry of the face (triangle or quadrilateral).
3.1.4 Element
Elements define the top-level geometric entities in the mesh. Their definition depends
upon the dimension of the expansion. For two-dimensional expansions, an element is
defined by a sequence of three or four edges. For three-dimensional expansions, the
element is defined by a list of faces. Elements are defined in the file with a line of the
form
1 <T ID="0"> 0 1 2 </T>
2 <H ID="1"> 3 4 5 6 7 8 </H>
Again, the choice of tag specified depends upon the geometry of the element. The element
tags are:
3.1 Geometry 27
• S Segment
• T Triangle
• Q Quadrilateral
• A Tetrahedron
• P Pyramid
• R Prism
• H Hexahedron
3.1.5 Curved Edges and Faces
Tip
The CURVED section is only necessary if curved edges or faces are present in
the mesh and may otherwise be omitted.
For mesh elements with curved edges and/or curved faces, a separate entry is used
to describe the control points for the curve. Each curve has a unique curve ID and
is associated with a predefined edge or face. The total number of points in the curve
(including end points) and their distribution is also included as attributes. The control
points are listed in order, each specified by three coordinates. Curved edges are defined
in the file with a line of the form
1 <E ID="3" EDGEID="7" TYPE="PolyEvenlySpaced" NUMPOINTS="3">
2 0.0 0.0 0.0 0.5 0.5 0.0 1.0 0.0 0.0
3 </E>
Note
In the compressed form, this section contains different sub-elements to efficiently
encode the high-order curvature data. This is not described further in this
document.
3.1.6 Composites
Composites define collections of elements, faces or edges. Each has a unique composite ID
associated with it. All components of a composite entry must be of the same type. The
syntax allows components to be listed individually or using ranges. Examples include
1 <C ID="0"> T[0-862] </C>
2 <C ID="1"> E[68,69,70,71] </C>
3.1.7 Domain
This tag specifies composites which describe the entire problem domain. It has the form
of
1 <DOMAIN> C[0] </DOMAIN>
3.2 Expansions
This section defines the polynomial expansions used on each of the defined geometric
composites. Expansion entries specify the number of modes, the basis type. The
short-hand version has the following form
1 <E COMPOSITE="C[0]" NUMMODES="5" FIELDS="u" TYPE="MODIFIED" />
or, if we have more then one variable we can apply the same basis to all using
1 <E COMPOSITE="C[0]" NUMMODES="5" FIELDS="u,v,p" TYPE="MODIFIED" />
The expansion basis can also be specified in detail as a combination of one-dimensional

bases, and thus the user is able to, for example, increase the quadrature order. For tet
elements this takes the form:
1 <E COMPOSITE="C[0]"
2 BASISTYPE="Modified_A,Modified_B,Modified_C"
3 NUMMODES="3,3,3"
4 POINTSTYPE="GaussLobattoLegendre,GaussRadauMAlpha1Beta0,GaussRadauMAlpha2Beta0"
5 NUMPOINTS="4,3,3"
6 FIELDS="u" />
and for prism elements:

2 BASISTYPE="Modified_A,Modified_A,Modified_B"
3 NUMMODES="3,3,3"
4 POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre,GaussRadauMAlpha1Beta0"
5 NUMPOINTS="4,4,3"
6 FIELDS="u" />
3.3 Conditions
The final section of the file defines parameters and boundary conditions which define the
nature of the problem to be solved. These are enclosed in the CONDITIONS tag.
3.3.1 Parameters
Parameters may be required by a particular solver (for instance time-integration parame-
ters or solver-specific parameters), or arbitrary and only used within the context of the
session file (e.g. parameters in the definition of an initial condition). All parameters are
enclosed in the PARAMETERS XML element.
3.3 Conditions 29
1 <PARAMETERS>
2 ...
3 </PARAMETERS>
A parameter may be of integer or real type and may reference other parameters defined
previous to it. It is expressed in the file as
1 [PARAMETER NAME] = [PARAMETER VALUE] 
For example,
1 NumSteps = 1000 
2 TimeStep = 0.01 
3 FinTime = NumSteps*TimeStep 
3.3.2 Solver Information

These specify properties to define the actions specific to solvers, typically including
the equation to solve, the projection type and the method of time integration. The
property/value pairs are specified as XML attributes. For example,
1 <SOLVERINFO>
2 
3 
4 
5 </SOLVERINFO>
The list of available solvers in Nektar++ can be found in Part III.
3.3.2.1 Drivers
Drivers are defined under the CONDITIONS section as properties of the SOLVERINFO XML
element. The role of a driver is to manage the solver execution from an upper level.
The default driver is called Standard and executes the following steps:
1. Prints out on screen a summary of all the conditions defined in the input file.
2. Sets up the initial and boundary conditions.
3. Calls the solver defined by SolverType in the SOLVERINFO XML element.
4. Writes the results in the output (.fld) file.
In the following example, the driver Standard is used to manage the execution of the
incompressible Navier-Stokes equations:
1 <SOLVERINFO>
2 
3 
4 
5 
6 
7 </SOLVERINFO>
If no driver is specified in the session file, the driver Standard is called by default. Other
drivers can be used and are typically focused on specific applications. As described in
Sec. 10.3.1 and 10.4.1, the other possibilities are:
• ModifiedArnoldi - computes of the leading eigenvalues and eigenmodes using

modified Arnoldi method.
• Arpack - computes of eigenvalues/eigenmodes using Implicitly Restarted Arnoldi

Method (ARPACK).
• SteadyState - uses the Selective Frequency Damping method (see Sec. 10.1.4)
to obtain a steady-state solution of the Navier-Stokes equations (compressible or
incompressible).
3.3.3 Variables
These define the number (and name) of solution variables. Each variable is prescribed
a unique ID. For example a two-dimensional flow simulation may define the velocity
variables using
1 <VARIABLES>
2 <V ID="0"> u </V>
3 <V ID="1"> v </V>
4 </VARIABLES>
3.3.4 Global System Solution Information

Many Nektar++ solvers use an implicit formulation of their equations to, for instance,
improve timestep restrictions. This means that a large matrix system must be constructed
and a global system set up to solve for the unknown coefficients. There are several
approaches in the spectral/hp element method that can be used in order to improve
efficiency in these methods, as well as considerations as to whether the simulation is run
in parallel or serial. Nektar++ opts for ‘sensible’ default choices, but these may or may
not be optimal depending on the problem under consideration.
This section of the XML file therefore allows the user to specify the global system solution
parameters, which dictates the type of solver to be used for any implicit systems that are
constructed. This section is particularly useful when using a multi-variable solver such as
the incompressible Navier-Stokes solver, as it allows us to select different preconditioning
3.3 Conditions 31
and residual convergence options for each variable. As an example, consider the block
defined by:
1 <GLOBALSYSSOLNINFO>
2 <V VAR="u,v,w">
3 
4 
5 
6 </V>
7 <V VAR="p">
9 
10 
11 </V>
12 </GLOBALSYSSOLNINFO>
The above section specifies that the variables u,v,w should use the IterativeStaticCond
global solver alongside the LowEnergyBlock preconditioner and an iterative tolerance of
10−8 on the residuals. However the pressure variable p is generally stiffer: we therefore
opt for a more expensive FullLinearSpaceWithLowEnergyBlock preconditioner and a
larger residual of 10−6 . We now outline the choices that one can use for each of these
parameters and give a brief description of what they mean.
Note
Defaults for all fields can be defined by setting the equivalent property in
the SOLVERINFO section. Parameters defined in this section will override any
options specificed there.
3.3.4.1 GlobalSysSoln options

Nektar++ presently implements four methods of solving a global system:
• Direct solvers construct the full global matrix and directly invert it using an
appropriate matrix technique, such as Cholesky factorisation, depending on the
properties of the matrix. Direct solvers only run in serial.
• Iterative solvers instead apply matrix-vector multiplications repeatedly, using the

conjugate gradient method, to converge to a solution to the system. For smaller
problems, this is typically slower than a direct solve. However, for larger problems
it can be used to solve the system in parallel execution.
• Xxt solvers use the XX T library to perform a parallel direct solve. This option is
only available if the NEKTAR_USE_MPI option is enabled in the CMake configuration.
• PETSc solvers use the PETSc library, giving access to a wide range of solvers and
preconditioners. See section 3.3.4.4 below for some additional information on how
to use the PETSc solvers. This option is only available if the NEKTAR_USE_PETSC
option is enabled in the CMake configuration.
Both the Xxt and PETSc solvers are considered advanced and are under development –
either the direct or iterative solvers are recommended in most scenarios.
These solvers can be run in one of three approaches:
• The Full approach constructs the global system based on all of the degrees of
freedom contained within an element. For most of the Nektar++ solvers, this
technique is not recommended.
• The StaticCond approach applies a technique called static condensation to instead

construct the system using only the degrees of freedom on the boundaries of the
elements, which reduces the system size considerably. This is the default option
in parallel.
• MultiLevelStaticCond methods apply the static condensation technique repeat-

edly to further reduce the system size, which can improve performance by 25-30%
over the normal static condensation method. It is therefore the default option
in serial. Note that whilst parallel execution technically works, this is under
development and is likely to be slower than single-level static condensation: this is
therefore not recommended.
The GlobalSysSoln option is formed by combining the method of solution with the
approach: for example IterativeStaticCond or PETScMultiLevelStaticCond.
3.3.4.2 Preconditioner options
Preconditioners can be used in the iterative and PETSc solvers to reduce the number
of iterations needed to converge to the solution. There are a number of preconditioner
choices, the default being a simple Jacobi (or diagonal) preconditioner, which is enabled
by default. There are a number of choices that can be enabled through this parameter,
which are all generally discretisation and dimension-dependent:
3.3 Conditions 33
Name Dimensions Discretisations

Null All All
Diagonal All All
FullLinearSpace 2/3D CG
LowEnergyBlock 3D CG
Block 2/3D All
FullLinearSpaceWithDiagonal All CG
FullLinearSpaceWithLowEnergyBlock 2/3D CG
FullLinearSpaceWithBlock 2/3D CG
For a detailed discussion of the mathematical formulation of these options, see the
developer guide.
3.3.4.3 SuccessiveRHS options

The SuccessiveRHS option can be used in the iterative solver only, to attempt to reduce
the number of iterations taken to converge to a solution. It stores a number of previous
solutions, dictated by the setting of the SuccessiveRHS option, to give a better initial
guess for the iterative process.
3.3.4.4 PETSc options and configuration

The PETSc solvers, although currently experimental, are operational both in serial and
parallel. PETSc gives access to a wide range of alternative solver options such as GMRES,
as well as any packages that PETSc can link against, such as the direct multi-frontal
solver MUMPS.
Configuration of PETSc options using its command-line interface dictates what matrix
storage, solver type and preconditioner should be used. This should be specified in a
.petscrc file inside your working directory, as command line options are not currently
passed through to PETSc to avoid conflict with Nektar++ options. As an example, to
select a GMRES solver using an algebraic multigrid preconditioner, and view the residual
convergence, one can use the configuration:
-ksp_monitor
-ksp_view
-ksp_type gmres
-pc_type gamg
Or to use MUMPS, one could use the options:
-ksp_type preonly
-pc_type lu
-pc_factor_mat_solver_package mumps
-mat_mumps_icntl_7 2
A final choice that can be specified is whether to use a shell approach. By default,
Nektar++ will construct a PETSc sparse matrix (or whatever matrix is specified on the
command line). This may, however, prove suboptimal for higher order discretisations.
In this case, you may choose to use the Nektar++ matrix-vector operators, which by
default use an assembly approach that can prove faster, by setting the PETScMatMult
SOLVERINFO option to Shell:
1 
The downside to this approach is that you are now constrained to using one of the
Nektar++ preconditioners. However, this does give access to a wider range of Krylov
methods than are available inside Nektar++ for more advanced users.
3.3.5 Boundary Regions and Conditions

Boundary conditions are defined by two XML elements. The first defines the boundary
regions in the domain in terms of composite entities from the GEOMETRY section of the
file. Each boundary region has a unique ID and are defined as,
1 <BOUNDARYREGIONS>
2 [composite-list] 
3 ...
4 </BOUNDARYREGIONS>
For example,
1 <BOUNDARYREGIONS>
2 C[2] 
3 C[3] 
The second XML element defines, for each variable, the condition to impose on each
boundary region, and has the form,
1 <BOUNDARYCONDITIONS>
2 <REGION REF="[regionID]">
3 <[type1] VAR="[variable1]" VALUE="[expression1]" />
4 ...
5 <[typeN] VAR="[variableN]" VALUE="[expressionN]" />
6 </REGION>
7 ...
8 </BOUNDARYCONDITIONS>
There should be precisely one REGION entry for each B entry defined in the BOUNDARYREGION
section above. For example, to impose a Dirichlet condition on both variables for a
domain with a single region,
3.3 Conditions 35
2 <REGION REF="0">
3 <D VAR="u" VALUE="sin(PI*x)*cos(PI*y)" />
4 <D VAR="v" VALUE="sin(PI*x)*cos(PI*y)" />
5 </REGION>
Boundary condition specifications may refer to any parameters defined in the session file.
The REF attribute corresponds to a defined boundary region. The tag used for each
variable specifies the type of boundary condition to enforce.
3.3.5.1 Dirichlet (essential) condition

Dirichlet conditions are specified with the D tag.
Projection Homogeneous support Time-dependent support Dimensions

CG Yes Yes 1D, 2D and 3D
DG Yes Yes 1D, 2D and 3D
HDG Yes Yes 1D, 2D and 3D
Example:
1 
2 <D VAR="u" VALUE="0" />
3 
4 <D VAR="u" VALUE="x^2+y^2+z^2" />
5 
6 <D VAR="u" USERDEFINEDTYPE="TimeDependent" VALUE="x+t" />
3.3.5.2 Neumann (natural) condition

Neumann conditions are specified with the N tag.
Projection Homogeneous support Time-dependent support Dimensions

CG Yes Yes 1D, 2D and 3D
DG No No 1D, 2D and 3D
HDG ? ? ?
Example:
1 
2 <N VAR="u" VALUE="0" />
3 
4 <N VAR="u" VALUE="x^2+y^2+z^2" />
5 
6 <N VAR="u" USERDEFINEDTYPE="TimeDependent" VALUE="x+t" />
7 
8 <N VAR="u" USERDEFINEDTYPE="H" VALUE="0" />
3.3.5.3 Periodic condition

Periodic conditions are specified with the P tag.
Projection Homogeneous support Dimensions

CG Yes 1D, 2D and 3D
DG No 2D and 3D
Example:
1 <BOUNDARYREGIONS>
2 C[1] 
3 C[2] 
5
7 <REGION REF="0">
8 
9 </REGION>
10 <REGION REF="1">
11 
12 </REGION>
Periodic boundary conditions are specified in a significantly different form to other

conditions. The VALUE property is used to specify which BOUNDARYREGION is periodic
with the current region in square brackets.
Caveats:
• A periodic condition must be set for ”’both”’ boundary regions; simply specifying
a condition for region 0 or 1 in the above example is not enough.
• The order of the elements inside the composites defining periodic boundaries is
important. For example, if ‘C[0]‘ above is defined as edge IDs ‘0,5,4,3‘ and ‘C[1]‘ as
‘7,12,2,1‘ then edge 0 is periodic with edge 7, 5 with 12, and so on.
• For the above reason, the composites must also therefore be of the same size.
• In three dimensions, care must be taken to correctly align triangular faces which
are intended to be periodic. The top (degenerate) vertex should be aligned so that,
if the faces were connected, it would lie at the same point on both triangles.
3.3.5.4 Time-dependent boundary conditions

Time-dependent boundary conditions may be specified through setting the USERDEFINEDTYPE
attribute and using the parameter t where the current time is required. For example,
1 <D VAR="u" USERDEFINEDTYPE="TimeDependent" VALUE="sin(PI*(x-t))" />
3.3 Conditions 37
3.3.5.5 Boundary conditions from file

Boundary conditions can also be loaded from file. The following example is from the
Incompressible Navier-Stokes solver,
1 <REGION REF="1">
2 <D VAR="u" FILE="Test_ChanFlow2D_bcsfromfiles_u_1.bc" />
3 <D VAR="v" VALUE="0" />
4 <N VAR="p" USERDEFINEDTYPE="H" VALUE="0" />
5 </REGION>
Boundary conditions can also be loaded simultaneously from a file and from an analytic
expression. For example in the scenario where a spatial boundary condition is read from
a file, but needs to be modulated by a time-dependent analytic expression:
1 <REGION REF="1">
2 <D VAR="u" VALUE="USERDEFINEDTYPE="TimeDependent" VALUE="sin(PI*(x-t))"
3 FILE="bcsfromfiles_u_1.bc" />
4 </REGION>
In the case where both VALUE and FILE are specified, the values are multiplied together
to give the final value for the boundary condition.
3.3.6 Functions
Finally, multi-variable functions such as initial conditions and analytic solutions may
be specified for use in, or comparison with, simulations. These may be specified using
expressions ( <E> ) or imported from a file ( <F> ) using the Nektar++ FLD file format
1 <FUNCTION NAME="ExactSolution">
2 <E VAR="u" VALUE="sin(PI*x-advx*t))*cos(PI*(y-advy*t))" />
3 </FUNCTION>
4 <FUNCTION NAME="InitialConditions">
5 <F VAR="u" FILE="session.rst" />
6 </FUNCTION>
A restart file is a solution file (in other words an .fld renamed as .rst) where the field data
is specified. The expansion order used to generate the .rst file must be the same as that
for the simulation. .pts files contain scattered point data which needs to be interpolated
to the field. For further information on the file format and the different interpolation
schemes, see section 5.4.14. All filenames must be specified relative to the location of the
.xml file.
With the additional argument TIMEDEPENDENT="1" , different files can be loaded for
each timestep. The filenames are defined using boost::format syntax where the step
time is used as variable. For example, the function Baseflow would load the files
U0V0_1.00000000E-05.fld , U0V0_2.00000000E-05.fld and so on.
1 <FUNCTION NAME="Baseflow">
2 <F VAR="U0,V0" TIMEDEPENDENT="1"FILE="U0V0_%14.8E.fld"/>
3 </FUNCTION>
For .pts files, the time consuming computation of interpolation weights is only performed
for the first timestep. The weights are stored and reused in all subsequent steps, which
is why all consecutive .pts files must use the same ordering, number and location of data
points.
Other examples of this input feature can be the insertion of a forcing term,
1 <FUNCTION NAME="BodyForce">
2 <E VAR="u" VALUE="0" />
3 <E VAR="v" VALUE="0" />
4 </FUNCTION>
5 <FUNCTION NAME="Forcing">
6 <E VAR="u" VALUE="-(Lambda + 2*PI*PI)*sin(PI*x)*sin(PI*y)" />
7 </FUNCTION>
or of a linear advection term

1 <FUNCTION NAME="AdvectionVelocity">
2 <E VAR="Vx" VALUE="1.0" />
3 <E VAR="Vy" VALUE="1.0" />
4 <E VAR="Vz" VALUE="1.0" />
5 </FUNCTION>
3.3.6.1 Remapping variable names

Note that it is sometimes the case that the variables being used in the solver do not match
those saved in the FLD file. For example, if one runs a three-dimensional incompressible
Navier-Stokes simulation, this produces an FLD file with the variables u , v , w and
p . If we wanted to use this velocity field as input for an advection velocity, the
advection-diffusion-reaction solver expects the variables Vx , Vy and Vz .
We can manually specify this mapping by adding a colon to the

2 <F VAR="Vx,Vy,Vz" FILE="file.fld:u,v,w" />
3 </FUNCTION>
There are some caveats with this syntax:
• You must specify the same number of fields for both the variable, and after the
colon. For example, the following is not valid.
2 <F VAR="Vx,Vy,Vz" FILE="file.fld:u" />
3 </FUNCTION>
• This syntax is not valid with the wildcard operator * , so one cannot write for
example:
2 <F VAR="*" FILE="file.fld:u,v,w" />
3 </FUNCTION>
3.3 Conditions 39
3.3.6.2 Time-dependent file-based functions

With the additional argument TIMEDEPENDENT="1" , different files can be loaded for
each timestep. The filenames are defined using boost::format syntax where the step
time is used as variable. For example, the function Baseflow would load the files
U0V0_1.00000000E-05.fld , U0V0_2.00000000E-05.fld and so on.
2 <F VAR="U0,V0" TIMEDEPENDENT="1" FILE="U0V0_%14.8R.fld" />
3 </FUNCTION>
Section 3.6 provides the list of acceptable mathematical functions and other related
technical details.
3.3.7 Quasi-3D approach

To generate a Quasi-3D appraoch with Nektar++ we only need to create a 2D or a 1D
mesh, as reported above, and then specify the parameters to extend the problem to a
3D case. For a 2D spectral/hp element problem, we have a 2D mesh and along with the
parameters we need to define the problem (i.e. equation type, boundary conditions, etc.).
The only thing we need to do, to extend it to a Quasi-3D approach, is to specify some
additional parameters which characterise the harmonic expansion in the third direction.
First we need to specify in the solver information section that that the problem will be
extended to have one homogeneouns dimension; here an example
1 <SOLVERINFO>
2 
3 
4 
5 
6 
7 
8 </SOLVERINFO>
then we need to specify the parameters which define the 1D harmonic expanson along the
z-axis, namely the homogeneous length (LZ) and the number of modes in the homogeneous
direction (HomModesZ). HomModesZ corresponds also to the number of quadrature
points in the homogenous direction, hence on the number of 2D planes discretized with a
specral/hp element method.
1 <PARAMETERS>
2 TimeStep = 0.001 
3 NumSteps = 1000 
4 IO_CheckSteps = 100 
5 IO_InfoSteps = 10 
6 Kinvis = 0.025 
7 HomModesZ = 4 
8 LZ = 1.0 
9 </PARAMETERS>
In case we want to create a Quasi-3D approach starting form a 1D spectral/hp element

mesh, the procedure is the same, but we need to specify the parameters for two harmonic
directions (in Y and Z direction). For Example,
1 <SOLVERINFO>
2 
3 
5 
6 
8 </SOLVERINFO>
9 <PARAMETERS>
10 TimeStep = 0.001 
11 NumSteps = 200 
14 wavefreq = PI 
15 epsilon = 1.0 
16 Lambda = 1.0 
17 HomModesY = 10 
18 LY = 6.5 
20 LZ = 2.0 
21 </PARAMETERS>
By default the opeartions associated with the harmonic expansions are performed with the
Matix-Vector-Multiplication (MVM) defined inside the code. The Fast Fourier Transofrm
(FFT) can be used to speed up the operations (if the FFTW library has been compiled
in ThirdParty, see the compilation instructions). To use the FFT routines we need just
to insert a flag in the solver information as below:
1 <SOLVERINFO>
5 
6 
8 
9 </SOLVERINFO>
The number of homogenenous modes has to be even. The Quasi-3D apporach can be
created starting from a 2D mesh and adding one homogenous expansion or starting form
a 1D mesh and adding two homogeneous expansions. Not other options available. In
case of a 1D homogeneous extension, the homogeneous direction will be the z-axis. In
case of a 2D homogeneous extension, the homogeneous directions will be the y-axis and
the z-axis.
3.4 Filters 41
3.4 Filters
Filters are a method for calculating a variety of useful quantities from the field variables
as the solution evolves in time, such as time-averaged fields and extracting the field
variables at certain points inside the domain. Each filter is defined in a FILTER tag
inside a FILTERS block which lies in the main NEKTAR tag. In this section we give an
overview of the modules currently available and how to set up these filters in the session
file.
Here is an example FILTER :

1 <FILTER TYPE="FilterName">
2 <PARAM NAME="Param1"> Value1 </PARAM>
3 <PARAM NAME="Param2"> Value2 </PARAM>
4 </FILTER>
A filter has a name – in this case, FilterName – together with parameters which are set
to user-defined values. Each filter expects different parameter inputs, although where
functionality is similar, the same parameter names are shared between filter types for
consistency. Numerical filter parameters may be expressions and so may include session
parameters defined in the PARAMETERS section.
In the following we document the filters implemented. Note that some filters are solver-
specific and will therefore only work for a given subset of the available solvers.
3.4.1 FieldConvert checkpoints
Note
This filter is still at an experimental stage. Not all modules and options from
FieldConvert are supported.
This filter applies a sequence of FieldConvert modules to the solution, writing an

output file. An output is produced at the end of the simulation into session_fc.fld ,
or alternatively every M timesteps as defined by the user, into a sequence of files
session_*_fc.fld , where * is replaced by a counter.
The following parameters are supported:
Option name Required Default Description

OutputFile 7 session.fld Output filename. If no extension is
provided, it is assumed as .fld
OutputFrequency 7 NumSteps Number of timesteps after which
output is written, M .
Modules 7 FieldConvert modules to run, sepa-
rated by a white space.
As an example, consider:
1 <FILTER TYPE="FieldConvert">
2 <PARAM NAME="OutputFile">MyFile.vtu</PARAM>
3 <PARAM NAME="OutputFrequency">100</PARAM>
4 <PARAM NAME="Modules"> vorticity isocontour:fieldid=0:fieldvalue=0.1 </PARAM>
5 </FILTER>
This will create a sequence of files named MyFile_*_fc.vtu containing isocontours. The
result will be output every 100 time steps. Output directly to .vtu or .dat is currently
only supported for isocontours. In other cases, the output should be a .fld file.
3.4.2 Time-averaged fields
This filter computes time-averaged fields for each variable defined in the session file.
Time averages are computed by either taking a snapshot of the field every timestep,
or alternatively at a user-defined number of timesteps N . An output is produced at
the end of the simulation into session_avg.fld , or alternatively every M timesteps as
defined by the user, into a sequence of files session_*_avg.fld , where * is replaced by
a counter. This latter option can be useful to observe statistical convergence rates of the
averaged variables.
This filter is derived from FieldConvert filter, and therefore support all parameters
available in that case. The following additional parameter is supported:

SampleFrequency 7 1 Number of timesteps at which the aver-
age is calculated, N .
RestartFile 7 Restart file used as initial average. If no
extension is provided, it is assumed as
.fld
1 <FILTER TYPE="AverageFields">
2 <PARAM NAME="OutputFile">MyAverageField</PARAM>
3 <PARAM NAME="RestartFile">MyRestartAvg.fld</PARAM>
5 <PARAM NAME="SampleFrequency"> 10 </PARAM>
6 </FILTER>
This will create a file named MyAverageField.fld averaging the instantaneous fields
every 10 time steps. The averaged field is however only output every 100 time steps.
3.4 Filters 43
3.4.3 Moving average of fields

This filter computes the exponential moving average (in time) of fields for each variable
defined in the session file. The moving average is defined as:
ūn = αun + (1 − α)ūn−1
with 0 < α < 1 and ū1 = u1 .
The same parameters of the time-average filter are supported, with the output file in
the form session_*_movAvg.fld . In addition, either α or the time-constant τ must be
defined. They are related by:
ts
α=
τ + ts
where ts is the time interval between consecutive samples.
1 <FILTER TYPE="MovingAverage">
2 <PARAM NAME="OutputFile">MyMovingAverage</PARAM>
5 <PARAM NAME="tau"> 0.1 </PARAM>
6 </FILTER>
This will create a file named MyMovingAverage_movAvg.fld with a moving average sam-
pled every 10 time steps. The averaged field is however only output every 100 time
steps.
3.4.4 Reynolds stresses
Note
This filter is only supported for the incompressible Navier-Stokes solver.
This filter is an extended version of the time-average filter. It outputs not only the
time-average of the fields, but also the Reynolds stresses. The same parameters supported
in the time-average case can be used, for example:
1 <FILTER TYPE="ReynoldsStresses">
3 <PARAM NAME="RestartFile">MyAverageRst.fld</PARAM>
6 </FILTER>
By default, this filter uses a simple average. Optionally, an exponential moving average
can be used, in which case the output contains the moving averages and the Reynolds
stresses calculated based on them. For example:
1 <FILTER TYPE="ReynoldsStresses">
3 <PARAM NAME="MovingAverage">true</PARAM>
6 <PARAM NAME="alpha"> 0.01 </PARAM>
7 </FILTER>
3.4.5 Checkpoint fields

The checkpoint filter writes a checkpoint file, containing the instantaneous state of the
solution fields at at given timestep. This can subsequently be used for restarting the
simulation or examining time-dependent behaviour. This produces a sequence of files, by
default named session_*.chk , where * is replaced by a counter. The initial condition
is written to session_0.chk . Existing files are not overwritten, but renamed to e.g.
session_0.bak0.chk . In case this file already exists, too, the chk -file is renamed to
session_0.bak*.chk and so on.
Note
This functionality is equivalent to setting the IO_CheckSteps parameter in the
session file.

OutputFile 7 session Prefix of the output filename to which
the checkpoints are written.
OutputFrequency 3 - Number of timesteps after which output
is written.
For example, to output the fields every 100 timesteps we can specify:
1 <FILTER TYPE="Checkpoint">
2 <PARAM NAME="OutputFile">IntermediateFields</PARAM>
4 </FILTER>
3.4.6 History points

The history points filter can be used to evaluate the value of the fields in specific points
of the domain as the solution evolves in time. By default this produces a file called
session.his . For each timestep, and then each history point, a line is output containing
the current solution time, followed by the value of each of the field variables. Commented
3.4 Filters 45
lines are created at the top of the file containing the location of the history points and
the order of the variables.

the checkpoints are written.
OutputFrequency 7 1 Number of timesteps after which output
is written.
OutputPlane 7 0 If the simulation is homogeneous, the
plane on which to evaluate the history
point. (No Fourier interpolation is cur-
rently implemented.)
Points 3 - A list of the history points. These should
always be given in three dimensions.
For example, to output the value of the solution fields at three points (1, 0.5, 0), (2, 0.5, 0)
and (3, 0.5, 0) into a file TimeValues.his every 10 timesteps, we use the syntax:
1 <FILTER TYPE="HistoryPoints">
2 <PARAM NAME="OutputFile">TimeValues</PARAM>
4 <PARAM NAME="Points">
5 1 0.5 0
6 2 0.5 0
7 3 0.5 0
8 </PARAM>
9 </FILTER>
3.4.7 ThresholdMax
The threshold value filter writes a field output containing a variable m, defined by the
time at which the selected variable first exceeds a specified threshold value. The default
name of the output file is the name of the session with the suffix _max.fld . Thresholding
is applied based on the first variable listed in the session by default.


OutputFile 7 session_max.fld Output filename to which the
threshold times are written.
ThresholdVar 7 first variable name Specifies the variable on which
the threshold will be applied.
ThresholdValue 3 - Specifies the threshold value.
InitialValue 3 - Specifies the initial time.
StartTime 7 0.0 Specifies the time at which to
start recording.
An example is given below:

1 <FILTER TYPE="ThresholdMax">
2 <PARAM NAME="OutputFile"> threshold_max.fld </PARAM>
3 <PARAM NAME="ThresholdVar"> u </PARAM>
4 <PARAM NAME="ThresholdValue"> 0.1 </PARAM>
5 <PARAM NAME="InitialValue"> 0.4 </PARAM>
6 </FILTER>
which produces a field file threshold_max.fld .
3.4.8 ThresholdMin value

Performs the same function as the ThresholdMax filter but records the time at which
the threshold variable drops below a prescribed value.
3.4.9 One-dimensional energy

This filter is designed to output the energy spectrum of one-dimensional elements. It
transforms the solution field at each timestep into a orthogonal basis defined by the
functions
ψp (ξ) = Lp (ξ)
where Lp is the p-th Legendre polynomial. This can be used to show the presence of, for
example, oscillations in the underlying field due to numerical instability. The resulting
output is written into a file called session.eny by default. The following parameters
are supported:

the energy spectrum is written.
is written.
An example syntax is given below:

3.4 Filters 47
1 <FILTER TYPE="Energy1D">
2 <PARAM NAME="OutputFile">EnergyFile</PARAM>
4 </FILTER>
3.4.10 Modal energy
Note
This filter calculates the time-evolution of the kinetic energy. In the case of a two- or
three-dimensional simulation this is defined as
1
Z
Ek (t) = kuk2 dx
2 Ω
However if the simulation is written as a one-dimensional homogeneous expansion so that

N
X
u(x, t) = ûk (t)e2πikx
k=0
then we instead calculate the energy spectrum

1
Z
Ek (t) = kûk k2 dx.
2 Ω
Note that in this case, each component of ûk is a complex number and therefore
N = HomModesZ /2 lines are output for each timestep. This is a particularly useful tool
in examining turbulent and transitional flows which use the homogeneous extension. In
either case, the resulting output is written into a file called session.mdl by default.

the energy spectrum is written.
is written.
An example syntax is given below:

1 <FILTER TYPE="ModalEnergy">
2 <PARAM NAME="OutputFile">EnergyFile</PARAM>
4 </FILTER>
3.4.11 Aerodynamic forces
Note
This filter evaluates the aerodynamic forces along a specific surface. The forces are
projected along the Cartesian axes and the pressure and viscous contributions are
computed in each direction.

OutputFile 7 session Prefix of the output filename to which the
forces are written.
Frequency 7 1 Number of timesteps after which output is
written.
Boundary 3 - Boundary surfaces on which the forces are
to be evaluated.
An example is given below:

1 <FILTER TYPE="AeroForces">
2 <PARAM NAME="OutputFile">DragLift.frc</PARAM>
4 <PARAM NAME="Boundary"> B[1,2] </PARAM>
5 </FILTER>
During the execution a file named DragLift.frc will be created and the value of the
aerodynamic forces on boundaries 1 and 2, defined in the GEOMETRY section, will be
output every 10 time steps.
3.4.12 Kinetic energy and enstrophy
Note
This filter is only supported for the incompressible and compressible Navier-
Stokes solvers in three dimensions.
The purpose of this filter is to calculate the kinetic energy and enstrophy
1 1
Z Z
2
Ek = kuk dx, E= kωk2 dx
2µ(Ω) Ω 2µ(Ω) Ω
3.5 Forcing 49
where µ(Ω) is the volume of the domain Ω. This produces a file containing the time-
evolution of the kinetic energy and enstrophy fields. By default this file is called
session.eny where session is the session name.

OutputFile 7 session.eny Output file name to which the en-
ergy and enstrophy are written.
OutputFrequency 3 - Number of timesteps at which out-
put is written.
To enable the filter, add the following to the FILTERS tag:

1 <FILTER TYPE="Energy">
2 <PARAM NAME="OutputFrequency"> 1 </PARAM>
3 </FILTER>
3.5 Forcing
An optional section of the file allows forcing functions to be defined. These are enclosed
in the FORCING tag. The forcing type is enclosed within the FORCE tag and expressed in
the file as:
1 <FORCE TYPE="[NAME]">
2 ...
3 </FORCE>
The force type can be any of the following:
• "Absorption"
• "Body"
• "Programmatic"
• "Noise"
3.5.1 Absorption
This force type allows the user to apply an absorption layer (essentially a porous region)
anywhere in the domain. The user may also specify a velocity profile to be imposed
at the start of this layer, and in the event of a time-dependent simulation, this profile
can be modulated with a time-dependent function. These velocity functions and the
function defining the region in which to apply the absorption layer are expressed in the
CONDITIONS section, however the name of these functions are defined here by the COEFF
tag for the layer, the REFFLOW tag for the velocity profile, and the REFFLOWTIME for the
time-dependent function.
1 <FORCE TYPE="Absorption">
2 <COEFF> [FUNCTION NAME] <COEFF/>
3 <REFFLOW> [FUNCTION NAME] <REFFLOW/>
4 <REFFLOWTIME> [FUNCTION NAME] <REFFLOWTIME/>
5 </FORCE>
3.5.2 Body
This force type specifies the name of a body forcing function expressed in the CONDITIONS
section.
1 <FORCE TYPE="Body">
2 <BODYFORCE> [FUNCTION NAME] <BODYFORCE/>
3 </FORCE>
3.5.3 Programmatic
This force type allows a forcing function to be applied directly within the code, thus it
has no associated function.
1 <FORCE TYPE="Programmatic">
2 </FORCE>
3.5.4 Noise
This force type allows the user to specify the magnitude of a white noise force. Optional
arguments can also be used to define the frequency in time steps to recompute the noise
(default is never) and the number of time steps to apply the noise (default is the entire
simulation).
1 <FORCE TYPE="Noise">
2 <WHITENOISE> [VALUE] <WHITENOISE/>
3 
4 <UPDATEFREQ> [VALUE] <UPDATEFREQ/>
5 <NSTEPS> [VALUE] <NSTEPS/>
6 </FORCE>
3.6 Analytic Expressions
This section discusses particulars related to analytic expressions appearing in Nektar++.

Analytic expressions in Nektar++ are used to describe spatially or temporally varying
properties, for example
• velocity profiles on a boundary
• some reference functions (e.g. exact solutions)

3.6 Analytic Expressions 51
which can be retrieved in the solver code.
Analytic expressions appear as the content of VALUE attribute of
• boundary condition type tags within <REGION> subsection of <BOUNDARYCONDITIONS> ,

e.g. <D> , <N> etc.
• expression declaration tag <E> within <FUNCTION> subsection.
The tags above declare analytic expressions as well as link them to one of the field
variables declared in <EXPANSIONS> section. For example, the declaration
1 <D VAR="u" VALUE="sin(PI*x)*cos(PI*y)" />
registers expression sin(πx) cos(πy) as a Dirichlet boundary constraint associated with

field variable u .
Enforcing the same velocity profile at multiple boundary regions and/or field variables
results in repeated re-declarations of a corresponding analytic expression. Currently
one cannot directly link a boundary condition declaration with an analytic expression
uniquely specified somewhere else, e.g. in the <FUNCTION> subsection. However this
duplication does not affect an overall computational performance.
3.6.1 Variables and coordinate systems

Declarations of analytic expressions are formulated in terms of problem space-time
coordinates. The library code makes a number of assumptions to variable names and
their order of appearance in the declarations. This section describes these assumptions.
Internally, the library uses 3D global coordinate space regardless of problem dimension.
Internal global coordinate system has natural basis (1,0,0),(0,1,0),(0,0,1) with coordinates
”’x”’,”’y”’ and ”’z”’. In other words, variables ”’x”’,”’y”’ and ”’z”’ are considered to be
first, second and third coordinates of a point (”’x”’,”’y”’,”’z”’).
Declarations of problem spatial variables do not exist in the current XML file format.
Even though field variables are declarable as in the following code snippet,
1 <VARIABLES>
2 <V ID="0"> u </V>
3 <V ID="1"> v </V>
4 </VARIABLES>
there are no analogous tags for space variables. However an attribute SPACE of
<GEOMETRY> section tag declares the dimension of problem space. For example,
1 <GEOMETRY DIM="1" SPACE="2"> ...
2 </GEOMETRY>
specifies 1D flow within 2D problem space. The number of spatial variables presented in
expression declaration should match space dimension declared via <GEOMETRY> section
tag.
The library assumes the problem space also has natural basis and spatial coordinates
have names ”’x”’,”’y”’ and”’z”’.
Problem space is naturally embedded into the global coordinate space: each point of
• 1D problem space with coordinate x is represented by 3D point (x,0,0) in the global

coordinate system;
• 2D problem space with coordinates (x,y) is represented by 3D point (x,y,0) in the
global coordinate system;
• 3D problem space with coordinates (x,y,z) has the same coordinates in the global
space coordinates.
Currently, there is no way to describe rotations and translations of problem space relative
to the global coordinate system.
The list of variables allowed in analytic expressions depends on the problem dimension:
• For 1D problem analytic expressions must make use of variable ”’x”’ only;
• For 2D problem analytic expressions should make use of variables ”’x”’ and ”’y”’.
• 3D problems may use variables ”’x”’, ”’y”’ and ”’z”’ in their analytic expressions.
Violation of these constraints yields unpredictable results of expression evaluation. The

current implementation assigns magic value -9999 to each dimensionally excessive spacial
variable appearing in analytic expressions. For example, the following declaration
2 </GEOMETRY> ...
3 <CONDITIONS> ...
5 <REGION REF="0">
6 <D VAR="u" VALUE="x+y+z" /> <D VAR="v" VALUE="sin(PI*x)*cos(PI*y)" />
7 </REGION>
9 ...
10 </CONDITIONS>
results in expression x + y + z being evaluated at spatial points (xi , yi , −9999) where xi

and yi are the spacial coordinates of boundary degrees of freedom. However, the library
behaviour under this constraint violation may change at later stages of development (e.g.,
magic constant 0 may be chosen) and should be considered unpredictable.
Another example of unpredictable behaviour corresponds to wrong ordering of variables:

2 </GEOMETRY> ...
3 <CONDITIONS> ...
5 <REGION REF="0">
6 <D VAR="u" VALUE="sin(y)" />
7 </REGION>
9 ...
10 </CONDITIONS>
Here one declares 1D problem, so Nektar++ library assumes spacial variable is ”’x”’.
At the same time, an expression sin(y) is perfectly valid on its own, but since it does
not depend on ”’x”’, it will be evaluated to constant sin(−9999) regardless of degree of
freedom under consideration.
3.6.1.1 Time dependence
Variable ”’t”’ represents time dependence within analytic expressions. The boundary con-
dition declarations need to add an additional property USERDEFINEDTYPE="TimeDependent"
in order to flag time dependency to the library.
3.6.1.2 Syntax of analytic expressions
Analytic expressions are formed of
• brackets (). Bracketing structure must be balanced.
• real numbers: every representation is allowed that is correct for boost::lexical_cast<double>() ,

e.g.
1 1.2, 1.2e-5, .02
• mathematical constants
Identifier Meaning Real Value

Fundamental constants
E Natural Logarithm 2.71828182845904523536
PI π 3.14159265358979323846
GAMMA Euler Gamma 0.57721566490153286060
DEG deg/radian 57.2957795130823208768
PHI golden ratio 1.61803398874989484820
Derived constants
LOG2E log2 e 1.44269504088896340740
LOG10E log10 e 0.43429448190325182765
LN2 loge 2 0.69314718055994530942
π
PI_2 2 1.57079632679489661923
π
PI_4 4 0.78539816339744830962
1
1_PI π 0.31830988618379067154
2
2_PI π 0.63661977236758134308
2_SQRTPI √2 1.12837916709551257390
√π
SQRT2 2 1.41421356237309504880
SQRT1_2 √1 0.70710678118654752440
2
• parameters: alphanumeric names with underscores, e.g. GAMMA_123 , GaM123_45a_ ,

_gamma123 are perfectly acceptable parameter names. However parameter name
cannot start with a numeral. Parameters must be defined with <PARAMETERS>...</PARAMETERS> .
Parameters play the role of constants that may change their values in between of
expression evaluations.
• variables (i.e., x, y, z and t )
• unary minus operator (e.g. -x )
• binary arithmetic operators +, -, *, /, ˆ Powering operator allows using real

exponents (it is implemented with std::pow() function)
• boolean comparison operations <, <=, >, >=, == evaluate their sub-expressions
to real values 0.0 or 1.0.
• mathematical functions of one or two arguments:

Identifier Meaning
abs(x) absolute value |x|
asin(x) inverse sine arcsin x
acos(x) inverse cosine arccos x
ang(x,y) computes polar coordinate θ = arctan(y/x) from (x, y)
atan(x) inverse tangent arctan x
atan2(y,x) inverse tangent function (used in polar transformations)
ceil(x) round up to nearest integer dxe
cos(x) cosine cos x
cosh(x) hyperbolic cosine cosh x
exp(x) exponential ex
fabs(x) absolute value (equivalent to abs)
floor(x) rounding down bxc
log(x) logarithm base e, ln x = log x
log10(x) logarithm base 10, log10 x p
rad(x,y) computes polar coordinate r = x2 + y 2 from (x, y)
sin(x) sine sin x
sinh(x) hyperbolic sine sinh x
√
sqrt(x) square root x
tan(x) tangent tan x
tanh(x) hyperbolic tangent tanh x
These functions are implemented by means of the cmath library: http://www.

cplusplus.com/reference/clibrary/cmath/. Underlying data type is double
at each stage of expression evaluation. As consequence, complex-valued expressions
(e.g. (−2)0 .123) get value nan (not a number). The operator ˆ is implemented
via call to std::pow() function and accepts arbitrary real exponents.
• random noise generation functions. Currently implemented is awgn(sigma) -

Gaussian Noise generator, where σ is the variance of normal distribution with zero
mean. Implemented using the boost::mt19937 random number generator with
boost variate generators (see http://www.boost.org/libs/random)
3.6.1.3 Examples
Some straightforward examples include
• Basic arithmetic operators: 0.5*0.3164/(3000ˆ0.25)
• Simple polynomial functions: y*(1-y)
• Use of values defined in PARAMETERS section: -2*Kinvis*(x-1)
• More complex expressions involving trigonometric functions, parameters and con-

stants: (LAMBDA/2/PI)*exp(LAMBDA*x)*sin(2*PI*y)
• Boolean operators for multi-domain functions: (y<0)*sin(y) + (y>=0)*y
3.6.2 Performance considerations

Processing analytic expressions is split into two stages:
• parsing with pre-evaluation of constant sub-expressions,
• evaluation to a number.
Parsing of analytic expressions with their partial evaluation take place at the time of
setting the run up (reading an XML file). Each analytic expression, after being pre-
processed, is stored internally and quickly retrieved when it turns to evaluation at given
spatial-time point(s). This allows to perform evaluation of expressions at a large number
of spacial points with minimal setup costs.
3.6.2.1 Pre-evaluation details

Partial evaluation of all constant sub-expressions makes no sense in using derived constants
from table above. This means, either make use of pre-defined constant LN10ˆ2 or
straightforward expression log10(2)ˆ2 results in constant 5.3018981104783980105
being stored internally after pre-processing. The rules of pre-evaluation are as follows:
• constants, numbers and their combinations with arithmetic, analytic and comparison
operators are pre-evaluated,
• appearance of a variable or parameter at any recursion level stops pre-evaluation of

all upper level operations (but doesn’t stop pre-evaluation of independent parallel
sub-expressions).
For example, declaration

1 <D VAR="u" VALUE="exp(-x*sin(PI*(sqrt(2)+sqrt(3))/2)*y )" />
results in expression exp(-x*(-0.97372300937516503167)*y ) being stored internally:

sub-expression sin(PI*(sqrt(2)+sqrt(3))/2) is evaluated to constant but appearance
of x and y variables stops further pre-evaluation.
Grouping predefined constants and numbers together helps. Its useful to put brackets to
be sure your constants do not run out and become factors of some variables or parameters.
Expression evaluator does not do any clever simplifications of input expressions, which is
clear from example above (there is no point in double negation). The following subsection
addresses the simplification strategy.
3.6.2.2 Preparing analytic expression

The total evaluation cost depends on the overall number of operations. Since evaluator
is not making simplifications, it worth trying to minimise the total number of operations
in input expressions manually.
Some operations are more computationally expensive than others. In an order of increasing
complexity:
• +, -, <, >, <=, >=, ==,
• *, /, abs, fabs, ceil, floor,
• ˆ, sqrt, exp, log, log10, sin, cos, tan, sinh, cosh, tanh, asin, acos, atan .
For example,
• x*x is faster than xˆ2 — it is one double multiplication vs generic calculation of

arbitrary power with real exponents.
• (x+sin(y))ˆ2 is faster than (x+sin(y))*(x+sin(y)) - sine is an expensive

operation. It is cheaper to square complicated expression rather than compute it
twice and add one multiplication.
• An expression exp(-41*( (x+(0.3*cos(2*PI*t)))ˆ2 + (0.3*sin(2*PI*t))ˆ2 )) makes

use of 5 expensive operations ( exp , sin , cos and power ˆ twice) while an
equivalent expression exp(-41*( x*x+0.6*x*cos(2*PI*t) + 0.09 )) uses only 2
expensive operations.
If any simplifying identity applies to input expression, it may worth applying it, provided
it minimises the complexity of evaluation. Computer algebra systems may help.
3.6.2.3 Vectorized evaluation

Expression evaluator is able to calculate an expression for either given point (its space-
time coordinates) or given array of points (arrays of their space-time coordinates, it uses
SoA). Vectorized evaluation is faster then sequential due to a better data access pattern.
Some expressions give measurable speedup factor 4.6. Therefore, if you are creating your
own solver, it worth making vectorized calls.
Part II
Preprocessing & Postprocessing
58
Chapter 4
NekMesh
NekMesh is a utility bundled with Nektar++ which has two purposes:
• allow foreign mesh file formats to be converted into Nektar++’s XML format;
• aide in the generation of high-order meshes through a series of supplied processing
modules.
Note
NekMesh replaces a previous utility called MeshConvert. This change is to
reflect the fact that the program no longer only converts and manipulates
meshes but can now also generate them from a CAD definition. This mesh
generator is in an early stage of development and as such is disabled by default.
For the time being those not using the mesh generator can use NekMesh as they
would have used MeshConvert, none of the functionality or methodology has
changed.
There is also some limited support for other output formats. We begin by running
through a basic example to show how a mesh can be converted from the widely-used
mesh-generator Gmsh to the XML file format.
Note
The default since January 2016 is to output the .xml files in a compressed form
where the VERTEX, EDGES, FACES, ELEMENTS and CURVED information
is compressed into binary format which is then converted into base64. This is
identified for each section by the attribute COMPRESSED="B64Z-LittleEndian” .
To output in ascii format add the module option “:xml:uncompress” to the
.xml file, i.e.
NekMesh file.msh newfile.xml:xml:uncompress
59
60 Chapter 4 NekMesh
4.1 Exporting a mesh from Gmsh
To demonstrate how NekMesh works, we will define a simple channel-like 3D geometry.

First, we must define the Gmsh geometry to be used. The Gmsh definition is given below,
and is visualised in figure 4.1.
1 Point(1) = {-1, 0, 0, 1.0};
2 Point(2) = {-0.3, 0, 0, 1.0};
3 Line(3) = {1, 2};
4 s[] = Extrude {0, 0, 7} {
5 Line{3}; Layers{5}; Recombine;
6 };
7 v[] = Extrude {{0, 0, 1}, {0, 0, 0}, Pi} {
8 Surface{s[1]}; Layers{10}; Recombine;
9 };
Whilst a full tutorial on Gmsh is far beyond the scope of this document, note the use
of the Recombine argument. This allows us to generate a structured hexahedral mesh;
remove the first Recombine to generate a prismatic mesh and both occurances to generate
a tetrahedral mesh. Increasing the Layers numbers refines the mesh in the radial and
azimuthal direction respectively.
4.2 Defining physical surfaces and volumes
Figure 4.1 Geometry definition in Gmsh (left) and resulting high-order mesh visualised in
ParaView (right).
In order for us to use the mesh, we need to define the physical surfaces which correspond
to the inflow, outflow and walls so that we can set appropriate boundary conditions.
The numbering resulting from the extrusions in this case is not straightforward. In the
graphical interface, select Geometry > Physical Groups > Add > Surface , and then
hover over each of the surfaces which are shown by the dashed gray lines. The numbering
will be revealed in the toolbar underneath the geometry as a ruled surface. In this case:
• Walls: surfaces 7, 8, 28, 29.
• Inflow: surface 16.

4.3 Converting the MSH to Nektar++ format 61
• Outflow: surface 24.
We also need to define the physical volumes, which can be done in a similar fashion. For
this example, there is only one volume having ID 1. Adding these groups to the end of
the .geo file is very straightforward:
1 Physical Volume(0) = {1};
2 Physical Surface(1)= {7,8,28,29};
3 Physical Surface(2) = {16};
4 Physical Surface(3) = {24};
Either choose the option File->Save Mesh or, assuming this is saved in a file named
test.geo , run the command
gmsh -3 test.geo
which will produce the resulting MSH file test.msh . One can generate a high-order
mesh by specifying the order on the command line, for example
gmsh -3 -order 6 test.geo
will generate a sixth-order mesh. Note that you will need to use a current version of
Gmsh in order to do this, most likely from subversion.
4.3 Converting the MSH to Nektar++ format
Assuming that you have compiled Nektar++ according to the compilation instructions,
run the command
NekMesh test.msh test.xml
to generate the XML file.
Note
This file contains only the geometry definition (and a default EXPANSIONS
definition). In order to use this mesh, a CONDITIONS section must be supplied
detailing the solver and parameters to use.
To validate the mesh visually, we can use a utility such as Paraview or VisIt. To do this,
run the command
XmlToVtk test.xml
which generates an unstructured VTK file test.vtu .
It is possible that, when the high-order information was inserted into the mesh by Gmsh,
invalid elements are generated which self intersect. In this case, the Jacobian of the
mapping defining the curvature will have negative regions, which will generate warnings
such as:
Warning: Level 0 assertion violation

3D deformed Jacobian not positive (element ID = 48) (first vertex ID = 105)
This tells you the element ID that is invalid, and the ID of the first vertex of the element.
Whilst a resulting simulation may run, the results may not be valid because of this
problem, or excessively large amounts of time may be needed to solve the resulting linear
system.
4.4 NekMesh modules
NekMesh is designed to provide a pipeline approach to mesh generation. To do this, we

break up tasks into three different types. Each task is called a module and a chain of
modules specifies the pipeline.
• Input modules read meshes in a variety of formats;

• Processing modules modify meshes to aide in generation processes;
• Output modules write meshes in a variety of formats.
The figure below depicts how these might be coupled together to form a pipeline: On the
Input Process 1 Process 2 Output
Figure 4.2 Illustrative pipeline of the NekMesh process.
command line, we would define this as:
NekMesh -m process1 -m process2 input.msh output.xml
Process modules can also have parameters passed to them, that can take arguments, or
not.
NekMesh -m process1:p1=123:booleanparam input.msh output.xml
To list all available modules use the -l command line argument:

4.4 NekMesh modules 63
Available classes:
Input: dat:
Reads Tecplot polyhedron ascii format converted from Star CCM (.dat).
...
and then to see the options for a particular module, use the -p command line argument:
Options for module detect:

vol: Tag identifying surface to process.
Note
Module names change when you use the -p option. Input modules should be
preceded by in: , processing modules by proc: and output modules by out: .
4.4.1 Input modules
Input and output modules use file extension names to determine the correct module to
use. Not every module is capable of reading high-order information, where it exists. The
table below indicates support currently implemented.
Format Extension High-order Notes

Gmsh msh 3 Only reads nodes, elements and physical
groups (which are mapped to composites).
Nektar rea 3 Reads elements, fluid boundary conditions.
Most curve types are unsupported: high-
order information must be defined in an
accompanying .hsf file.
Nektar++ xml 3 Fully supported.
PLY ply 7 Reads only the ASCII format..
Semtex sem 3 Reads elements and boundary conditions.
In order to read high-order information,
run meshpr session.sem > session.msh
and place in the same directory as the
session file.
Star-CCM+ dat 7 Star outputs plt file which currently needs
to be coverted to ascii using Tecplot.
Reads mesh only, only support for quads
and triangles (2D) and hexes, prisms,
tetrahedra (3D).
Star-CCM+ ccm 7 Reads start ccm format. Reads mesh only,
only support for quads and triangles (2D)
and hexes, prisms, tetrahedra (3D). Re-
quires NEKTAR_USE_CCM option to
be activated in cmake and then requires
ccmio library to be compiled by user.
VTK vtk 7 Experimental support. Only ASCII trian-
gular data is supported.
Note that you can override the module used on the command line. For example, Semtex
session files rarely have extensions. So for a session called pipe-3d we can convert this
using the syntax
NekMesh pipe-3d:sem pipe-3d.xml
Typically, mesh generators allow physical surfaces and volumes to contain many element
types; for example a cube could be constructed from a mixture of hexes and prisms. In
Nektar++, a composite can only contain a single element type. Whilst the converter will
attempt to preserve the numbering of composites from the original mesh type, sometimes
a renumbering will occur when a domain contains many element types. For example, for
a domain with the tag 150 containing quadrilaterals and triangles, the Gmsh reader
will print a notification along the lines of:
Multiple elements in composite detected; remapped:

- Tag 150 => 150 (Triangle), 151 (Quadrilateral)
The resulting file therefore has two composites of IDs 150 and 151 respectively, con-
taining the triangular and quadrilateral elements of the original mesh.
4.4.2 Output modules

The following output formats are supported:
Format Extension High-order Notes

Gmsh msh 3 High-order hexes, quads, tetrahedra and tri-
angles are supported up to arbitrary order.
Prisms supported up to order 4, pyramids
up to order 1.
Nektar++ xml 3 Most functionality supported.
VTK vtk 7 Experimental. Only ASCII triangular data
is supported.
Note that for both Gmsh and VTK, it is highly likely that you will need to experiment
with the source code in order to successfully generate meshes since robustness is not
guaranteed.
The default for xml is into binary data which has been converted into base64. If you
wish to see an ascii output you need to specify the output module option uncompress
by executing:
NekMesh Mesh.msh output.xml:xml:uncompress
Finally, both the Gmsh and Nektar++ output modules support an order parameter,
which allows you to generate a mesh of a uniform polynomial order. This is used in the
same manner as the above, so that the command
NekMesh Mesh.msh output.msh:msh:order=7
will generate an order 7 Gmsh mesh. In the rest of these subsections, we discuss the
various processing modules available within NekMesh.
4.4.3 Extract surfaces from a mesh

Often one wants to visualise surfaces of a 3D mesh, or extract the values of variables on
the surface and visualise them. To support this, NekMesh can extract two-dimensional
surfaces which can be visualised using FieldConvert in order to extract the value of a
3D field on a given surface.
As an example, we can extract composite surfaces 2 and 3-5 from a mesh using the
extract module:
NekMesh -m extract:surf=2,3-5 Mesh.xml output.xml
If you also wish to have the boundaries of the extracted surface detected add the
detectbnd option
NekMesh -m extract:surf=2,3-5:detectbnd Mesh.xml output.xml
which will produce new composites for the extracted boundary.
4.4.4 Negative Jacobian detection

To detect elements with negative Jacobian determinant, use the jac module:
NekMesh -m jac Mesh.xml output.xml
To get a detailed list of elements which have negative Jacobians, one may use the list
option:
NekMesh -m jac:list Mesh.xml output.xml
and to extract the elements for the purposes of visualisation within the domain, use the
extract boolean parameter:
NekMesh -m jac:extract Mesh.xml MeshWithNegativeElements.xml
To turn off curvature associated with negative jacobians one can try to use the linearise
module:
NekMesh -m linerise:invalid Mesh.xml output.xml
This option will remove all high order curvature on all element types with singular
jacobians.
4.4.5 Spherigon patches

Where high-order information is not available (e.g. when using meshes from imaging
software), various techniques can be used to apply a smoothing to the high-order element.
In NekMesh we use spherigons, a kind of patch used in the computer graphics community
used for efficiently smoothing polygon surfaces.
Spherigons work through the use of surface normals, where in this sense ‘surface’ refers
to the underlying geometry. If we have either the exact or approximate surface normal
at each given vertex, spherigon patches approximate the edges connecting two vertices
by arcs of a circle. In NekMesh we can either approximate the surface normals from the
linear elements which connect to each vertex (this is done by default), or supply a file
which gives the surface normals.
To apply spherigon patches on two connected surfaces 11 and 12 use the following
command:
NekMesh -m spherigon:surf=11,12 \
MeshWithStraighEdges.xml MeshWithSpherigons.xml
If the two surfaces "11" and "12" are not connected, or connect at a sharp edge which is
C 0 continuous but not C 1 smooth, use two separate instances of the spherigon module.
NekMesh -m spherigon:surf=11 -m spherigon:surf=12 \

This is to avoid the approximated surface normals being incorrect at the edge.
If you have a high-resolution mesh of the surfaces 11 and 12 in ply format it can be
used to improve the normal definition of the spherigons. Run:
NekMesh -m spherigon:surf=11,12:usenormalfile=Surf_11-12_Mesh.ply \
This can be useful, for example, when meshing the Leading edge of an airfoil. Starting
from a linear mesh (left figure) the spherigon patches curve the surface elements producing
leading edge closer to the underlying geometry:
4.4.6 Periodic boundary condition alignment

When using periodic boundary conditions, the order of the elements within the boundary
composite determines which element edges are periodic with the corresponding boundary
composite.
To counteract this issue, NekMesh has a periodic alignment module which attempts
to identify pairs of mutually periodic edges. Given two surfaces surf1 and surf2 ,
which for example correspond to the physical surface IDs specified in Gmsh, and an axis
which defines the periodicity direction, the following command attempts to reorder the
composites:
NekMesh -m peralign:surf1=11:surf2=12:dir=y \
-m peralign:surf1=13:surf2=14:dir=z Mesh.xml Mesh_aligned.xml
Figure 4.3 (a) Leading edge without spherigons, (b) Leading edge with spherigons
Here the surfaces with IDs 11 and 12 will be aligned normal to the y-axis and the surfaces
13 and 14 will be aligned normal to the z-axis.
Note that this command cannot perform magic – it assumes that any given edge or face
lying on the surface is periodic with another face on the opposing surface, that there are
the same number of elements on both surfaces, and the corresponding edge or face is the
same size and shape but translated along the appropriate axis.
In 3D, where prismatic or tetrahedral elements are connected to one or both of the
surfaces, additional logic is needed to guarantee connectivity in the XML file. In this
case we append the orient parameter:
NekMesh -m peralign:surf1=11:surf2=12:dir=y:orient input.dat output.xml
Note
One of the present shortcomings of orient is that it throws away all high-order
information and works only on the linear element. This can be gotten around
if you are just doing e.g. spherigon patches by running this peralign module
before the spherigon module.
4.4.7 Boundary layer splitting

Often it is the case that one can generate a coarse boundary layer grid of a mesh. NekMesh
has a method for splitting prismatic and hexahedral elements into finer elements based
on the work presented in [25] and [26]. You must have a prismatic mesh that is O-type –
that is, you can modify the boundary layer without modifying the rest of the mesh.
Given n layers, and a ratio r which defines the relative heights of elements in different
layers, the method works by defining a geometric progression of points
2(1 − r)
xk = xk−1 + ark , a=
1 − rn+1
in the standard segment [−1, 1]. These are then projected into the coarse elements to
construct a sequence of increasingly refined elements, as depicted in figure 4.4.
∆n χe (ξ)
ξ3
ξ1 ξ2
Figure 4.4 Splitting Ωst and applying the mapping χe to obtain a high-order layer of prisms
from the macro-element.
To split a prism boundary layer on surface 11 into 3 layers with a growth rate of 2 and 7
integration points per element use the following command:
NekMesh -m bl:surf=11:layers=3:r=2:nq=7 MeshWithOnePrismLayer.xml \

MeshWith3PrismsLayers.xml
Note
You can also use an expression in terms of coordinates (x, y, z) for r to make
the ratio spatially varying; e.g. r=sin(x) . In this case the function should be
sufficiently smooth to prevent the elements self-intersecting.
4.4.8 High-order cylinder generation

Generating accurate high-order curved geometries in Gmsh is quite challenging. This
module processes an existing linear cylindrical mesh, with axis aligned with the z-
coordinate axis, to generate accurate high-order curvature information along the edges.
NekMesh -m cyl:surf=2:r=1.0:N=5 LinearCylinder.xml HighOrderCylinder.xml
The module parameters are:

Figure 4.5 (a) LE with Spherigons but only one prism layer for resolving the boundary layer,
(b) LE with Spherigons with 3 growing layers of prisms for better resolving the boundary layer.
• surf : Surface on which to apply curvature. This should be the outer surface of
the cylinder.
• r : Radius of the cylinder.
• N : Number of high-order points along each element edge.
Note
The module could also be used to apply curvature along the interior of a hollow
cylinder. However, there are no checks to ensure the resulting elements are not
self-intersecting.
4.4.9 Linearisation
The ability to remove all the high-order information in a mesh can be useful at times
when mesh generation is tricky or impossible in the presence of curvature
To do this in NekMesh use the command:
NekMesh -m linearise:all high-order-mesh.xml linear-mesh.xml
The output will contain only the linear mesh information, all curved information is
removed. Alternatively
NekMesh -m linearise:invalid high-order-mesh.xml linear-mesh.xml

attempts to remove curvature from elements only where necessary. This is a simple
algorithm that removes curvature from invalid elements and repeats until all elements
are valid. Either all or invalid must be specified.
• all : remove curvature from all elements.
• invalid : remove curvature from invalid elements.
• prismonly : consider only prisms when removing curvature. This is useful in the
presence of a prismatic boundary layer.
4.4.10 Extracting interface between tetrahedra and prismatic elements

When the mesh is three-dimensional and comprised of a prismatic boundary layer with
tetrahedra in the interior of the domain, this module extracts the prismatic elements only,
and constructs a boundary region for the interface between the tetrahedra and prisms.
This is useful in, for example, the study of aortic flows, where the prismatic boundary
layer can be extracted and refined to study unsteady advection-diffusion problems on a
more refined grid inside the boundary layer.
To use this module you therefore use the command:
NekMesh -m extracttetprisminterface input.xml output.xml
There are no configuration options for this module, as it is highly specific to a certain
class of meshes.
4.4.11 Boundary identification

Some mesh formats lack the ability to identify boundaries of the domain they discretise.
NekMesh has a rudimentary boundary identification routine for conformal meshes, which
will create a composite of edges (2D) or faces (3D) which are connected to precisely one
element. This can be done using the detect module:
NekMesh -m detect volume.xml volumeWithBoundaryComposite.xml
4.4.12 Scalar function curvature

This module imposes curvature on a surface given a scalar function z = f (x, y). For
example, if on surface 1 we wish to apply a surface defined by a Gaussian z = exp[−(x2 +
y 2 )] using 7 quadrature points in each direction, we may issue the command
NekMesh -m scalar:surf=1:nq=7:scalar=exp$x*x+y*y$ mesh.xml deformed.xml

Note
This module makes no attempt to apply the curvature to the interior of the
domain. Elements must therefore be coarse in order to prevent self-intersection.
If a boundary layer is required, one option is to use this module in combination
with the splitting module described earlier.
4.4.13 Link Checking

It is quite possible that a mesh contains some sort of hanging entity or element connectivity
error. The check link module is a fast check that, a) elements are correctly connected
and b) the boundary entities (composites) match the interior domain:
NekMesh -m linkcheck mesh.xml mesh2.xml
This module should be added to the module chain if the user suspects there may be a
mesh issue. The module will print a warning if there is a connectivity error.
4.4.14 2D mesh extrusion

This module allows a 2D mesh, quads, triangles or both, to be extruded in the z direction
to make a simple 3D mesh made of prisms and hexahedra. It is also capable of extruding
the high-order curvature within the 2D mesh. The module requires two parameters:
NekMesh -m extrude:layers=n:length=l 2D.xml 3D.xml
length which determines how long the z extrusion will be and layers, the number of
elements in the z direction.
4.4.15 Variational Optimisation

This module can correct invalid and improve the quality of elements in high-order meshes
by applying curvilinear deformation to the interiors of domains. It achieves this by
solving a solid mechanics system which, using variational calculus has been cast is a
non-linear energy optimsation problem. It is basis of the work in [36].
It works by considering the boundary (curved) mesh entities to be fixed and moving the
interior nodes to a lower energy configuration. This new configuration in most scenarios
is a higher quality mesh. The energy is evaluated depending on which functional is
chosen. We find hyperleastic to be the most reliable but it can also model the mesh
and a linearelastic solid as well as functionals based on the Winslow equation and the
distortion method proposed by Roca et al. [14].
There are a large number of options which can be viewed using the help function but the
basic usage is:
4.5 Mesh generation 73
NekMesh -m varopti:type inital.xml optimised.xml
where type can be hyperelastic, linearelastic, winslow or roca.
4.5 Mesh generation
In addition to the functionality described previously, NekMesh is capable of generating

high-order meshes directly from a CAD definition. By default this functionality is not
activated, a user wishing to utilise the mesh generation capability of NekMesh must
compile Nektar++ with the NEKTAR_USE_MESHGEN option on. As well as compiling the
relevant routines into NekMesh it will also download a number of other packages which
are required.
The most critical dependancy of the mesh generation routines is OpenCascade which
powers the CAD engine. NekMesh is capable of finding and using existing installations
of OpenCascade 6.8 or OCE 0.17. If either are not present on the installation machine
NekMesh will install OCE 0.17 from source. This is a very big installation and will take
some time so it is advised that the user ensures OpenCascade is availble on the machine.
As with all tasks within NekMesh the mesh generation capability exists as its own separate
module which is of type Input. Due to the vast amount of code associated with the
generation of high-order meshes and the comparatively small nature of modules in the
NekMesh program a new library has been created for Nektar++ called NekMeshUtils,
which contains all the core routines and classes for the NekMesh mesh format as well as a
series of classes for the generation of meshes. This library also contains the CAD API
for Nektar++ which is used to generate the meshes.
4.5.1 Methodology
This section outlines the approach taken by NekMesh to generate high-order meshes. To
simplify the sometimes very complicated high-order mesh generation processes in other
programs, NekMesh executes all the stages required to produce a high-order mesh in one
single pipeline which once started requires no interaction from the user. In broad terms
these stages are:
• Specification of the element sizes in the mesh,
• Coarse linear mesh generation of the domain,
• Generation of optimised high-order surface on the geometric boundary,
and are outlined in more detail in the following sections.

4.5.1.1 CAD Interaction

At the core of all the ideas in the NekMesh generator is that the final mesh is a high
quality representation of the underlying geometry. As such all of the entities in the
mesh must know where they are located with respect to the CAD and the system to
be able to query any geometric information at any point in the domain easily and with
accuracy. To handle this NekMesh has been interfaced with the third-party suite of CAD
libraries called OpenCascade. In its normal state OpenCascade is a very large collection
of libraries with tens of thousands of functions which are simply not needed for our
purposes, because of this its installation is a very arduous and long process. Combine
this with the fact that there are dozens of versions and types of OpenCascade, such as
OpenCascade Community Edition, it is simply impossible for NekMesh to use already
existing OpenCascade installations on a given machine. To solve these issues, when
installing Nektar++ with the mesh generator it will download pre-compiled binaries for
the relevant OS and link against those, any previously installed versions of OpenCascade
will not be searched for and therefore ignored. To reduce the massively complex libraries
in OpenCascade down to a manageable set of functions to be used in NekMesh a set of
interface classes have been created which act as buffer between it and Nektar++. These
CAD classes mean that development of mesh generation routines is significantly easier
and in the future Nektar++ developers will be able to utilise CAD information in all
aspects of the framework without having to learn OpenCascade. Another advantage with
this approach is that adding support for other CAD engines, as well as OpenCascade, in
the future should be relativity simple and will not require the rewriting of any of the
NekMesh code.
4.5.1.2 Automatic specification of the mesh

One of the key challenges of generating a high-order mesh is the creation of a suitable
coarse linear mesh. It is quite difficult for a user to define a full set spacings over a whole
domain which will produce a good quality especially when aiming for coarseness. This is
tackled in NekMesh with a system for automatically defining a set of smooth and coarse
mesh spacings throughout the whole domain. This is achieved using an octree description
of the domain. The domain is recursively subdivided into octants which each describe
a small portion of the domain. The level to which the domain subdivides is based on
the curvature of the geometric boundary. Higher curvature regions will subdivide to
a finer level allowing for increased control on the mesh specification and smoothness.
The geometric curvature is then related to a mesh sizing parameter and propagated
throughout the domain ensuring a smooth mesh. For those unfamiliar with octrees, it is
best to think of it as a non-conforming hexahedral mesh
4.5.1.3 Linear Mesh Generation

The first challenge mentioned in the previous section is addressed with the NekMesh
approach to linear mesh generation. Primarily because of the difficulties in interfacing
existing linear mesh generators for high-order applications the decision was made to
include a bespoke linear mesh generator within the program. Compared with the mesh
generators included in commercial packages this linear mesh generator takes the quite
unconventional and more historic approach in building the mesh in a bottom up fashion
from 0D to 3D. Using this approach means it is possible to guarantee a level of boundary
conformity which direct to 3D approaches cannot at the desired level of coarseness. In
this approach, first mesh nodes are placed on the vertices of the CAD model (0D), then
the curves in the CAD are meshed in 1D using the vertex nodes as boundaries, then the
surfaces are meshed in their 2D parameter plane using the curve meshes as boundaries
and finally the 3D volume is meshed using the surface mesh as the boundary to complete
the linear mesh. In NekMesh, to achieve greater robustness, the 2D mesh generation
library Triangle is used and the TetGen library for the 3D. Both of which are highly
developed Delaunay based mesh generators. As with all additional libraies in Nektar++
these are automatically downloaded and installed if needed.
4.5.1.4 High-order Surface Generation

Addition of the high-order nodes to and the curving of the mesh is very open problem,
no high-order mesh generator has solved this and while the methods used in NekMesh
are not 100% full-proof, the system currently in place can create good quality high-order
curved meshes with a reasonable robustness. This area will receive the greatest level
of development in the future. The most critical part of defining the high-order mesh is
the addition of high-order nodes on the geometric surface. The mesh generator must
achieve the greatest level of geometric accuracy as it can otherwise it will greatly affect
the final flow solutions. If the linear surface triangulation is taken to be fixed during this
process, the problem can be addressed in a element by element fashion. If the high-order
nodes are placed by simply using an affine mapping to the CAD surface and back the
resulting high-order triangle will inherit the same distortions as the CAD surface. To
solve this NekMesh uses a system node location optimisation in the parameter plane of
the CAD surface to ensure the high-order triangles have as little distortion as possible
while remaining exactly on the geometric surface. To do this the system models the
high-order edges and triangles as a network of springs with an associated spring energy
which is minimised using a multidimensional Newton type optimisation procedure with a
Gauss-Seidel matrix solver.
4.5.1.5 Mesh Correction

Due to the fact that, for the time being, no consideration is given to the curving of mesh
interior entities explicitly in the mesh generation process, the curving the geometric
surface can produce meshes with invalid elements, especially in the case of Euler type
(Tetrahedra only) meshes. Three strategies exist within Nektar++ to correct these
elements. Firstly removing the curvature, by removing the curvature of invalid elements
they become valid. However this has the massive downside of compromising the geometric
accuracy of the mesh but is quick and effective, this can be enacted using the command:
NekMesh -m linearise:invalid invalidMesh.xml validMesh.xml

An alternative to this is to use the linear elastic solver within Nektar++ to deform the
mesh interior entities. Its use is very computationally expensive, as with all PDE solvers,
and is also not particularly robust. It can be used with the set of commands outlined in
the FieldConvert deform and displacement modules and the section on the Linear Elastic
Solver.
The final and possibly most useful approach is to use the Variational Optimsation module
to curve the interior of the domain. This is explained in 4.4.15.
4.5.2 Mesh generation manual

The mesh generation is executed with the command:
NekMesh session.mcf mesh.xml
where session.mcf is a mesh configuration file which contains all the options and parameters
needed for mesh generation. Below is an example of a simple example which generates a
2D NACA wing.
1 <NEKTAR>
2 <MESHING>
3
4 <INFORMATION>
5 
6 
7 </INFORMATION>
8
9 <PARAMETERS>
10 
11 
12 
13 
14
15 
16 
17 
18 
19 
20 
21 </PARAMETERS>
22
23 </MESHING>
24 </NEKTAR>
In all cases the mesh generator needs two pieces of information and four parameters. It
firstly needs to know the CAD file with which to work. In the example above this is
listed as a 4 digit number, this is because the mesh generator is equiped with a NACA
wing generator. In all other cases this parameter would be a STEP file. Secondly, what
type of mesh to make, the options are EULER and BL for 3D meshes and 2D and
2DBL for 2D meshes. In the case of EULER the mesh will be made with only tetrahedra.
For BL the mesh generator will attempt to insert a single macro prism layer onto the
geometry surface. This option requires additional parameters. This is similar for the 2D
scenarios. The automatic mesh specification system requires three parameters to build
the specification of a smooth, curvature refined mesh. Firstly MinDelta which is the
size of the smallest element to be found in the final mesh. Secondly MaxDelta which
is the maximum size of an element in the mesh and lastly EPS which is a sensitivity
to curvature parameter with a range 1 ≥ ε > 0 which heuristically controls the size of
the elements for a given degree of curvature on the geometric surface. Order is the
polynomial order of the mesh to be generated. When generating a boundary layer mesh
a few additional parameters must be given. An example is shown.
1 <NEKTAR>
2 <MESHING>
3
4 <INFORMATION>
5 
6 
7 </INFORMATION>
8
9 <PARAMETERS>
10 
11 
12 
13 
14
15 
16 
17 
18 
19 
20
21 
22 
23 
24 
25 
26 
27 </PARAMETERS>
28
29 </MESHING>
30 </NEKTAR>
A list of the CAD surfaces which will have a prism generated on is described by BLSurfs
and the thickness of the boundary to aim for is BLThick . The mesh generator has been
created with a range of error messages to aid in debugging. If you encounter an error
and the mesh generator fails, run NekMesh with the verbose -v flag and send the stdout
with the .mcf and .step files to m.turner14@imperial.ac.uk . Without the feedback this
functionality cannot improve.
Chapter 5
FieldConvert
FieldConvert is a utility embedded in Nektar++ with the primary aim of allowing the
user to convert the Nektar++ output binary files (.chk and .fld) into a format which
can be read by two common visualisation softwares: Paraview/VisIt (.vtu format) or
Tecplot/VisIt (.dat or .plt formats). FieldConvert also allows the user to manipulate the
Nektar++ output binary files by using some additional modules which can be called with
the option -m which stands for m odule. Note that another flag, -r (which stand for
r ange) allows the user to specify a sub-range of the domain on which the conversion or
manipulation of the Nektar++ output binary files will be performed.
Almost all of the FieldConvert functionalities can be run in parallel if Nektar++ is

compiled using MPI (see the installation documentation for additional info on how to
implement Nektar++ using MPI). 1
5.1 Convert .fld / .chk files into Paraview, VisIt or Tecplot format
To convert the Nektar++ output binary files (.chk and .fld) into a format which can be
read by two common visualisation softwares: Paraview (.vtu format), VisIt (.vtu format)
or Tecplot (.dat or .plt format) the user can run the following commands:
• Paraview or VisIt (.vtu format)
FieldConvert test.xml test.fld test.vtu
• Tecplot (.dat format)
FieldConvert test.xml test.fld test.dat
• Tecplot or VisIt(.plt format)

1
Modules that do not have parallel support will be specified in the appropriate section.
78
5.2 Convert field files between XML and HDF5 format 79
FieldConvert test.xml test.fld test.plt
where FieldConvert is the executable associated to the utility FieldConvert, test.xml

is the session file and test.vtu , test.dat , test.plt are the desired format outputs,
either Paraview, VisIt or Tecplot formats.
Tip
Note that the session file is also supported in its compressed format
test.xml.gz .
5.2 Convert field files between XML and HDF5 format
When Nektar++ is compiled with HDF5 support, solvers can select the format used for
output of .fld files. FieldConvert can be used to convert between these formats using
an option on the .fld output module. For example, if in.fld is stored in the default
XML format, it can be converted to HDF5 format by issuing the command
FieldConvert in.fld out.fld:fld:format=Hdf5
5.3 Range option -r
The Fieldconvert range option -r allows the user to specify a sub-range of the mesh
(computational domain) by using an additional flag, -r (which stands for r ange and
either convert or manipulate the Nektar++ output binary files. Taking as an example
the conversion of the Nektar++ binary files (.chk or .fld) shown before and wanting to
convert just the 2D sub-range defined by −2 ≤ x ≤ 3, −1 ≤ y ≤ 2 the additional flag
-r can be used as follows:
• Paraview or VisIt (.vtu format)
FieldConvert -r -2,3,-1,2 test.xml test.fld test.vtu
• Tecplot (.dat format)
FieldConvert -r 2,3,-1,2 test.xml test.fld test.dat
where -r defines the range option of the FieldConvert utility, the two first numbers
define the range in x direction and the the third and fourth number specify the y range.
A sub-range of a 3D domain can also be specified. For doing so, a third set of numbers
has to be provided to define the z range.
80 Chapter 5 FieldConvert
5.4 FieldConvert modules -m
FieldConvert allows the user to manipulate the Nektar++ output binary files (.chk and
.fld) by using the flag -m (which stands for m odule).. Specifically, FieldConvert has
these additional functionalities
1. C0Projection : Computes the C0 projection of a given output file;
2. QCriterion : Computes the Q-Criterion for a given output file;
3. addcompositeid : Adds the composite ID of an element as an additional field;
4. addFld : Sum two .fld files;
5. combineAvg : Combine two Nektar++ binary output (.chk or .fld) field file contain-
ing averages of fields (and possibly also Reynolds stresses) into single file;
6. concatenate : Concatenate a Nektar++ binary output (.chk or .fld) field file into
single file;
7. equispacedoutput : Write data as equi-spaced output using simplices to represent

the data for connecting points;
8. extract : Extract a boundary field;
9. homplane : Extract a plane from 3DH1D expansions;
10. homstretch : Stretch a 3DH1D expansion by an integer factor;
11. innerproduct : take the inner product between one or a series of fields with another
field (or series of fields).
12. interpfield : Interpolates one field to another, requires fromxml, fromfld to be

defined;
13. interppointdatatofld : Interpolates given discrete data using a finite difference

approximation to a fld file given an xml file;
14. interppoints : Interpolates a set of points to another, requires fromfld and fromxml
to be defined, a line or plane of points can be defined;
15. isocontour : Extract an isocontour of “fieldid” variable and at value “fieldvalue”.

Optionally “fieldstr” can be specified for a string defiition or “smooth” for smoothing;
16. jacobianenergy : Shows high frequency energy of Jacobian;
17. qualitymetric : Evaluate a quality metric of the underlying mesh to show mesh
quality;
18. meanmode : Extract mean mode (plane zero) of 3DH1D expansions;

5.4 FieldConvert modules -m 81
19. pointdatatofld : Given discrete data at quadrature points project them onto an
expansion basis and output fld file;
20. printfldnorms : Print L2 and LInf norms to stdout;
21. scalargrad : Computes scalar gradient field;
22. scaleinputfld : Rescale input field by a constant factor;
23. shear : Computes time-averaged shear stress metrics: TAWSS, OSI, transWSS,
TAAFI, TACFI, WSSG;
24. surfdistance : Computes height of a prismatic boundary layer mesh and projects
onto the surface (for e.g. y + calculation).
25. vorticity : Computes the vorticity field.
26. wss : Computes wall shear stress field.
The module list above can be seen by running the command
FieldConvert -l
In the following we will detail the usage of each module.
5.4.1 Smooth the data: C0Projection module

To smooth the data of a given .fld file one can use the C0Projection module of Field-
Convert
FieldConvert -m C0Projection test.xml test.fld test-C0Proj.fld
where the file test-C0Proj.fld can be processed in a similar way as described in section
5.1 to visualise the result either in Tecplot, Paraview or VisIt.
The option localtoglobalmap will do a global gather of the coefficients and then scatter
them back to the local elements. This will replace the coefficients shared between two
elements with the coefficients of one of the elements (most likely the one with the highest
id). Although not a formal projection it does not require any matrix inverse and so is
very cheap to perform.
The option usexmlbcs will enforce the boundary conditions specified in the input xml
file.
The option helmsmoothing=L will perform a Helmholtz smoothing projection of the form
2 ! 2
2π 2π

2
∇ + ûnew = ûorig
L L
which can be interpreted in a Fourier sense as smoothing the original coefficients using a
low pass filter of the form
1 2π
ûnew
k = 2 2 ûorig
k where K0 =
(1 + k /K0 ) L
and so L is the length scale below which the coefficients values are halved or more. Since
this form of the Helmholtz operator is not possitive definite, currently a direct solver is
necessary and so this smoother is mainly of use in two-dimensions.
5.4.2 Calculate Q-Criterion: QCriterion module

To perform the Q-criterion calculation and obtain an output data containing the Q-
criterion solution, the user can run
FieldConvert -m QCriterion test.xml test.fld test-QCrit.fld
where the file test-QCrit.fld can be processed in a similar way as described in section
5.4.3 Add composite ID: addcompositeid module

When dealing with a geometry that has many surfaces, we need to identify the composites
to assign boundary conditions. To assist in this, FieldConvert has a addcompositeid
module, which adds the composite ID of every element as a new field. To use this we
simply run
FieldConvert -m addcompositeid mesh.xml out.dat
In this case, we have produced a Tecplot file which contains the mesh and a variable that
contains the composite ID. To assist in boundary identification, the input file mesh.xml
should be a surface XML file that can be obtained through the NekMesh extract module
(see section 4.4.3).
5.4.4 Sum two .fld files: addFld module

To sum two .fld files one can use the addFld module of FieldConvert
FieldConvert -m addfld:fromfld=file1.fld:scale=-1 file1.xml file2.fld \

file3.fld
In this case we use it in conjunction with the command scale which multiply the values
of a given .fld file by a constant value . file1.fld is the file multiplied by value ,
file1.xml is the associated session file, file2.fld is the .fld file which is summed to
file1.fld and finally file3.fld is the output which contain the sum of the two .fld
files. file3.fld can be processed in a similar way as described in section 5.1 to visualise
the result either in Tecplot, Paraview or VisIt.
5.4.5 Combine two .fld files containing time averages: combineAvg module
To combine two .fld files obtained through the AverageFields or ReynoldsStresses filters,
use the combineAvg module of FieldConvert
FieldConvert -m combineAvg:fromfld=file1.fld file1.xml file2.fld \

file3.fld
file3.fld can be processed in a similar way as described in section 5.1 to visualise the
result either in Tecplot, Paraview or VisIt.
5.4.6 Concatenate two files: concatenate module

To concatenate file1.fld and file2.fld into file-conc.fld one can run the following
command
FieldConvert -m concatenate file.xml file1.fld file2.fld file-conc.fld
where the file file-conc.fld can be processed in a similar way as described in section
5.4.7 Equi-spaced output of data: equispacedoutput module

This module interpolates the output data to an truly equispaced set of points (not
equispaced along the collapsed coordinate system). Therefore a tetrahedron is represented
by a tetrahedral number of poinst. This produces much smaller output files. The points
are then connected together by simplices (triangles and tetrahedrons).
FieldConvert -m equispacedoutput test.xml test.fld test.dat
or
FieldConvert -m equispacedouttput test.xml test.fld test.vtu
Note
Currently this option is only set up for triangles, quadrilaterals, tetrahedrons
and prisms.
5.4.8 Extract a boundary region: extract module

The boundary region of a domain can be extracted from the output data using the
following command line
FieldConvert -m extract:bnd=2:fldtoboundary=1 test.xml \

test.fld test-boundary.fld
The option bnd specifies which boundary region to extract. Note this is different to
NekMesh where the parameter surf is specified and corresponds to composites rather
boundaries. If bnd is not provided, all boundaries are extracted to different fields. The
fldtoboundary is an optional command argument which copies the expansion of test.fld
into the boundary region before outputting the .fld file. This option is on by default.
If it turned off using fldtoboundary=0 the extraction will only evaluate the boundary
condition from the xml file. The output will be placed in test-boundary-b2.fld. If more
than one boundary region is specified the extension -b0.fld, -b1.fld etc will be outputted.
To process this file you will need an xml file of the same region. This can be generated
using the command:
NekMesh -m extract:surf=5 test.xml test-b0.xml
The surface to be extracted in this command is the composite number and so needs to
correspond to the boundary region of interest. Finally to process the surface file one can
use
FieldConvert test-b0.xml test-b0.fld test-b0.dat
This will obviously generate a Tecplot output if a .dat file is specified as last argument.
A .vtu extension will produce a Paraview or VisIt output.
5.4.9 Compute the gradient of a field: gradient module

To compute the spatial gradients of all fields one can run the following command
FieldConvert -m gradient test.xml test.fld test-grad.fld
where the file file-grad.fld can be processed in a similar way as described in section
5.4.10 Extract a plane from 3DH1D expansion: homplane module

To obtain a 2D expansion containing one of the planes of a 3DH1D field file, use the
command:
FieldConvert -m homplane:planeid=value file.xml file.fld file-plane.fld
If the option wavespace is used, the Fourier coefficients corresponding to planeid are
obtained. The command in this case is:
FieldConvert -m homplane:wavespace:planeid=value file.xml \

file.fld file-plane.fld
The output file file-plane.fld can be processed in a similar way as described in section
5.1 to visualise it either in Tecplot or in Paraview.
5.4.11 Stretch a 3DH1D expansion: homstretch module

To stretch a 3DH1D expansion in the z-direction, use the command:
FieldConvert -m homstretch:factor=value file.xml file.fld file-stretch.fld
The number of modes in the resulting field can be chosen using the command-line
parameter output-points-hom-z . Note that the output for this module should always
be a .fld file and this should not be used in combination with other modules using a
single command.
The output file file-stretch.fld can be processed in a similar way as described in

section 5.1 to visualise it either in Tecplot or in Paraview.
5.4.12 Inner Product of a single or series of fields with respect to a single

or series of fields: innerproduct module
You can take the inner product of one field with another field using the following
command:
FieldConvert -m innerproduct:fromfld=file1.fld file2.xml file2.fld \

out.stdout
This command will load the file1.fld and file2.fld assuming they both are spatially
defined by files.xml and determine the inner product of these fields. The input option
fromfld must therefore be specified in this module.
Optional arguments for this module are fields which allow you to specify the fields
that you wish to use for the inner product, i.e.
FieldConvert -m innerproduct:fromfld=file1.fld:fields=’’0,1,2’’ file2.xml \

file2.fld out.stdout
will only take the inner product between the variables 0,1 and 2 in the two fields files.
The default is to take the inner product between all fields provided.
Additional options include multifldids and allfromflds which allow for a series of
fields to be evaluated in the following manner:
FieldConvert -m innerproduct:fromfld=file1.fld:multifldids=’’0-3’’\
file2.xml file2.fld out.stdout
will take the inner product between a file names field1_0.fld, field1_1.fld, field1_2.fld
and field1_3.fld with respect to field2.fld.
Analogously including the options allfromflds , i.e.
FieldConvert -m innerproduct:fromfld=file1.fld:multifldids=’’0-3’’:\
allfromflds file2.xml file2.fld out.stdout
Will take the inner product of all the from fields, i.e. field1_0.fld,field1_1.fld,field1_2.fld
and field1_3.fld with respect to each other. This option essentially ignores file2.fld. Only
the unique inner products are evaluated so if four from fields are given only the related
trianuglar number 4 × 5/2 = 10 of inner products are evaluated.
This option can be run in parallel.
5.4.13 Interpolate one field to another: interpfield module

To interpolate one field to another, one can use the following command:
FieldConvert -m interpfield:fromxml=file1.xml:fromfld=file1.fld \
file2.xml file2.fld
This command will interpolate the field defined by file1.xml and file1.fld to the
new mesh defined in file2.xml and output it to file2.fld . The fromxml and
fromfld must be specified in this module. In addition there are two optional ar-
guments clamptolowervalue and clamptouppervalue which clamp the interpolation
between these two values. Their default values are -10,000,000 and 10,000,000.
Tip
This module can run in parallel where the speed is increased not only due to
using more cores but also, since the mesh is split into smaller sub-domains, the
search method currently adopted performs faster.
5.4.14 Interpolate scattered point data to a field: interppointdatatofld mod-

ule
To interpolate discrete point data to a field, use the interppointdatatofld module:
FieldConvert -m interppointdatatofld file1.xml file1.pts file1.fld

This command will interpolate the data from file1.pts to the mesh and expansions
defined in file1.xml and output the field to file1.fld . The file file.pts is of the
form:
1 <?xml version="1.0" encoding="utf-8" ?>
2 <NEKTAR>
3 <POINTS DIM="1" FIELDS="a,b,c">
4 1.0000 -1.0000 1.0000 -0.7778
5 2.0000 -0.9798 0.9798 -0.7980
6 3.0000 -0.9596 0.9596 -0.8182
7 4.0000 -0.9394 0.9394 -0.8384
8 </POINTS>
9 </NEKTAR>
where DIM="1" FIELDS="a,b,c specifies that the field is one-dimensional and contains
three variables, a, b, and c. Each line defines a point, while the first column contains
its x-coordinate, the second one contains the a-values, the third the b-values and so on.
In case of n-dimensional data, the n coordinates are specified in the first n columns
accordingly. In order to interpolate 1D data to a nD field, specify the matching coordinate
in the output field using the interpcoord argument:
FieldConvert -m interppointdatatofld:interpcoord=1 3D-file1.xml \

1D-file1.pts 3D-file1.fld
This will interpolate the 1D scattered point data from 1D-file1.pts to the y-coordinate
of the 3D mesh defined in 3D-file1.xml . The resulting field will have constant values
along the x and z coordinates. For 1D Interpolation, the module implements a quadratic
scheme and automatically falls back to a linear method if only two data points are
given. A modified inverse distance method is used for 2D and 3D interpolation. Linear
and quadratic interpolation require the data points in the .pts -file to be sorted by
their location in ascending order. The Inverse Distance implementation has no such
requirement.
5.4.15 Interpolate a field to a series of points: interppoints module

You can interpolate one field to a series of given points using the following command:
FieldConvert -m interppoints:fromxml=file1.xml:fromfld=file1.fld \
file2.pts file2.dat
This command will interpolate the field defined by file1.xml and file1.fld to the
points defined in file2.pts and output it to file2.dat . The fromxml and fromfld
must be specified in this module. The format of the file file2.pts is of the same form
as for the interppointdatatofld module:
2 <NEKTAR>
3 <POINTS DIM="2" FIELDS="">

4 0.0 0.0
5 0.5 0.0
6 1.0 0.0
7 </POINTS>
8 </NEKTAR>
There are three optional arguments clamptolowervalue , clamptouppervalue and defaultvalue

the first two clamp the interpolation between these two values and the third defines the
default value to be used if the point is outside the domain. Their default values are
-10,000,000, 10,000,000 and 0.
In addition, instead of specifying the file file2.pts , a module list of the form
FieldConvert -m interppoints:fromxml=file1.xml:fromfld= \
file1.fld:line=npts,x0,y0,x1,y1
can be specified where npts is the number of equispaced points between (x0, y0) to
(x1, y1) which can also be used in 3D by specifying (x0, y0, z0) to (x1, y1, z1).
An extraction of a plane of points can also be specified by
FieldConvert -m interppoints:fromxml=file1.xml:fromfld=file1.fld:\
plane=npts1,npts2,x0,y0,z0,x1,y1,z1,x2,y2,z2,x3,y3,z3
where npts1,npts2 is the number of equispaced points in each direction and (x0, y0, z0),
(x1, y1, z1), (x2, y2, z2) and (x3, y3, z3) define the plane of points specified in a clockwise
or anticlockwise direction.
In addition an extraction of a box of points can also be specified by
FieldConvert -m interppoints:fromxml=file1.xml:fromfld=file1.fld:\
box=npts1,npts2,npts3,xmin,xmax,ymin,ymax,zmin,zmax
where npts1,npts2,npts3 is the number of equispaced points in each direction and

(xmin, ymin, zmin) and (xmax, ymax, zmax3) define the limits of the box of points.
For the plane and box interpolation there is an additional optional argument cp=p0,q
which adds to the interpolated fields the value of cp = (p − p0)/q and cp0 = (p − p0 +
0.5u2 )/q where p0 is a reference pressure and q is the free stream dynamics pressure. If
the input does not contain a field “p” or a velocity field “u,v,w” then cp and cp0 are not
evaluated accordingly
Note
This module runs in parallel for the plane and box extraction of points. In
this case a series of .dat files are generated that can be concatinated together.
Other options do not run in parallel.
5.4.16 Isocontour extraction: iscontour module

Extract an isocontour from a field file. This option automatically take the field to an
equispaced distribution of points connected by linear simplicies of triangles or tetrahedrons.
The linear simplices are then inspected to extract the isocontour of interest. To specify
the field fieldid can be provided giving the id of the field of interest and fieldvalue
provides the value of the isocontour to be extracted.
FieldConvert -m isocontour:fieldid=2:fieldvalue=0.5 test.xml test.fld \

test-isocontour.dat
Alternatively fieldstr =”u+v” can be specified to calculate the field u2 and extract
its isocontour. You can also specify fieldname =”UplusV” to define the name of the
isocontour in the .dat file, i.e.
FieldConvert -m isocontour:fieldstr="u+v":fieldvalue=0.5:\
fieldname="UplusV" test.xml test.fld test-isocontour.dat
Optionally smooth can be specified to smooth the isocontour with default values
smoothnegdiffusion =0.495, smoothnegdiffusion =0.5 and smoothiter =100. This op-
tion typically should be used wiht the globalcondense option which removes multiply
defined verties from the simplex definition which arise as isocontour are generated element
by element. The smooth option preivously automatically called the globalcondense
option but this has been depracated since it is now possible to read isocontour files
directly and so it is useful to have these as separate options.
In addition to the smooth or globalcondense options you can specify removesmallcontour =100
which will remove separate isocontours of less than 100 triangles.
Note
Currently this option is only set up for triangles, quadrilaterals, tetrahedrons
and prisms.
5.4.17 Show high frequency energy of the Jacobian: jacobianenergy module

FieldConvert -m jacobianenergy file.xml file.fld jacenergy.fld
The option topmodes can be used to specify the number of top modes to keep.
The output file jacenergy.fld can be processed in a similar way as described in section
5.4.18 Calculate mesh quality: qualitymetric module

The qualitymetric module assesses the quality of the mesh by calculating a per-element
quality metric and adding an additional field to any resulting output. This does not
require any field input, therefore an example usage looks like
FieldConvert -m qualitymetric mesh.xml mesh-with-quality.dat
Two quality metrics are implemented that produce scalar fields Q:
• By default a metric outlined in [14] is produced, where all straight sided elements
have quality Q = 1 and Q < 1 shows the deformation between the curved element
and the straight-sided element. If Q = 0 then the element is invalid. Note that
Q varies over the volume of the element but is not guaranteed to be continuous
between elements.
• Alternatively, if the scaled option is passed through to the module, then the
scaled Jacobian
minξ∈Ωst J(ξ)
Js =
maxξ∈Ωst J(ξ)
(i.e. the ratio of the minimum to maximum Jacobian of each element) is calculated.
Again Q = 1 denotes an ideal element, but now invalid elements are shown by
Q < 0. Any elements with Q near zero are determined to be low quality.
5.4.19 Extract mean mode of 3DH1D expansion: meanmode module

To obtain a 2D expansion containing the mean mode (plane zero in Fourier space) of a
3DH1D field file, use the command:
FieldConvert -m meanmode file.xml file.fld file-mean.fld
The output file file-mean.fld can be processed in a similar way as described in section
5.1 to visualise the result either in Tecplot or in Paraview or VisIt.
5.4.20 Project point data to a field: pointdatatofld module

To project a series of points given at the same quadrature distribution as the .xml file
and write out a .fld file use the pointdatatofld module:
FieldConvert --noequispaced -m pointdatatofld file.pts file.xml file.fld

This command will read in the points provided in the file.pts and assume these
are given at the same quadrature distribution as the mesh and expansions defined in
file.xml and output the field to file.fld . If the points do not match an error will be
dumped.
The file file.pts which is assumed to be given by an interpolation from another source
is of the form:
2 <NEKTAR>
3 <POINTS DIM="3" FIELDS="p">
4 1.70415 -0.4 -0.0182028 -0.106893
5 1.70415 -0.395683 -0.0182028 -0.106794
6 1.70415 -0.3875 -0.0182028 -0.106698
7 1.70415 -0.379317 -0.0182028 -0.103815
8 </POINTS>
9 </NEKTAR>
where DIM="3" FIELDS="p specifies that the field is three-dimensional and contains one
variable, p. Each line defines a point, the first, second, and third columns contains the
x, y, z-coordinate and subsequent columns contain the field values, in this case the p-value
So in the general case of n-dimensional data, the n coordinates are specified in the first
n columns accordingly followed by the field data.
The default argument is to use the equipapced (but potentially collapsed) coordinates
which can be obtained from the command.
FieldConvert file.xml file.dat
In this case the pointdatatofld module shoudl be used without the –noequispaced
option. However this can lead to problems when peforming an elemental forward
projection/transform since the mass matrix in a deformed element can be singular as
the equispaced points do not have a sufficiently accurate quadrature rule that spans the
polynomial space. Therefore it is adviseable to use the set of points given by
FieldConvert --noequispaced file.xml file.dat
which produces a set of points at the gaussian collapsed coordinates. In this case one
must also use the –noequispaced option when projecting to a field.
Finally the option setnantovalue=0 can also be used which sets any nan values in the
interpolation to zero or any specified value in this option.
5.4.21 Print L2 and LInf norms: printfldnorms module

FieldConvert -m printfldnorms test.xml test.fld out.stdout
This module does not create an output file which is reinforced by the out.stdout option.
The L2 and LInf norms for each field variable are then printed to the stdout.
5.4.22 Computes the scalar gradient: scalargrad module

The scalar gradient of a field is computed by running:
FieldConvert -m scalargrad:bnd=0 test.xml test.fld test-scalgrad.fld
boundaries. If bnd is not provided, all boundaries are extracted to different fields. To
process this file you will need an xml file of the same region.
5.4.23 Scale a given .fld: scaleinputfld module

To scale a .fld file by a given scalar quantity, the user can run:
FieldConvert -m scaleinputfld:scale=value test.xml test.fld test-scal.fld
The argument scale=value rescales of a factor value test.fld by the factor value.
The output file file-conc.fld can be processed in a similar way as described in section
5.4.24 Time-averaged shear stress metrics: shear module

Time-dependent wall shear stress derived metrics relevant to cardiovascular fluid dynamics
research can be computed using this module. They are
• TAWSS: time-averaged wall shear stress;
• OSI: oscillatory shear index;
• transWSS: transverse wall shear stress;
• TACFI: time-averaged cross-flow index;
• TAAFI: time-averaged aneurysm formation index;
• |WSSG|: wall shear stress gradient.
To compute these, the user can run:
FieldConvert -m shear:N=value:fromfld=test_id_b0.fld test.xml test-multishear.fld

The argument N and fromfld are compulsory arguments that respectively define the
number of fld files corresponding to the number of discrete equispaced time-steps,
and the first fld file which should have the form of test_id_b0.fld where the first
underscore in the name marks the starting time-step file ID.
The input .fld files are the outputs of the wss module. If they do not contain the
surface normals (an optional output of the wss modle), then the shear module will not
compute the last metric, |WSSG|.
5.4.25 Boundary layer height calculation: surfdistance module

The surface distance module computes the height of a prismatic boundary layer and
projects this value onto the surface of the boundary, in a similar fashion to the extract
module. In conjunction with a mesh of the surface, which can be obtained with NekMesh ,
and a value of the average wall shear stress, one potential application of this module is
to determine the distribution of y + grid spacings for turbulence calculations.
To compute the height of the prismatic layer connected to boundary region 3, the user
can issue the command:
FieldConvert -m surfdistance:bnd=3 input.xml output.fld
Note that no .fld file is required, since the mesh is the only input required in order
to calculate the element height. This produces a file output_b3.fld , which can be
visualised with the appropriate surface mesh from NekMesh .
5.4.26 Calculate vorticity: vorticity module

To perform the vorticity calculation and obtain an output data containing the vorticity
solution, the user can run
FieldConvert -m vorticity test.xml test.fld test-vort.fld
where the file test-vort.fld can be processed in a similar way as described in section
5.1.
5.4.27 Computing the wall shear stress: wss module

To obtain the wall shear stres vector and magnitude, the user can run:
FieldConvert -m wss:bnd=0:addnormals=1 test.xml test.fld test-wss.fld
boundaries. If bnd is not provided, all boundaries are extracted to different fields. The
addnormals is an optional command argument which, when turned on, outputs the
normal vector of the extracted boundary region as well as the shear stress vector and
magnitude. This option is off by default. To process the output file(s) you will need an
xml file of the same region.
5.4.28 Manipulating meshes with FieldConvert
FieldConvert has support for two modules that can be used in conjunction with the linear
elastic solver, as shown in chapter 11. To do this, FieldConvert has an XML output
module, in addition to the Tecplot and VTK formats.
The deform module, which takes no options, takes a displacement field and applies it to
the geometry, producing a deformed mesh:
FieldConvert -m deform input.xml input.fld deformed.xml
The displacement module is designed to create a boundary condition field file. Its
intended use is for mesh generation purposes. It can be used to calculate the displacement
between the linear mesh and a high-order surface, and then produce a fld file, prescribing
the displacement at the boundary, that can be used in the linear elasticity solver.
Presently the process is somewhat convoluted and must be used in conjunction with
NekMesh to create the surface file. However the bash input below describes the pro-
cedure. Assume the high-order mesh is in a file called mesh.xml , the linear mesh
is mesh-linear.xml that can be generated by removing the CURVED section from
mesh.xml , and that we are interested in the surface with ID 123.
# Extract high order surface

NekMesh -m extract:surf=123 mesh.xml mesh-surf-curved.xml
# Use FieldConvert to calculate displacement between two surfaces

FieldConvert -m displacement:id=123:to=mesh-surf-curved.xml \
mesh-linear.xml mesh-deformation.fld
# mesh-deformation.fld is used as a boundary condition inside the

# solver to prescribe the deformation conditions.xml contains
# appropriate Nektar++ parameters (mu, E, other BCs, ...)
LinearElasticSolver mesh-linear.xml conditions.xml
# This produces the final field mesh-linear.fld which is the

# displacement field, use FieldConvert to apply it:
FieldConvert-g -m deform mesh-linear.xml mesh-linear.fld mesh-deformed.xml
5.5 FieldConvert in parallel 95
5.5 FieldConvert in parallel
To run FieldConvert in parallel the user needs to compile Nektar++ with MPI support
and can employ the following command
mpirun -np <nprocs> FieldConvert test.xml test.fld test.dat
mpirun -np <nprocs> FieldConvert test.xml test.fld test.plt
or
mpirun -np <nprocs> FieldConvert test.xml test.fld test.vtu
replacing <nprocs> with the number of processors. For the .dat and .plt outputs
the current version will proudce a single output file. However it is also sometimes useful
to produce multiple output files, one for each partition, and this can be done by using
the writemultiplefiles option, i.e.
mpirun -np <nprocs> FieldConvert test.xml test.fld \

test.dat:dat:writemultiplefiles
mpirun -np <nprocs> FieldConvert test.xml test.fld \

test.plt:plt:writemultiplefiles
For the .vtu format multiple files will by default be produced of the form test_P0.vtu ,
test_P1.vtu , test_P2.vtu . For this format an additional file called .pvtu is written
out which allows for parallel reading of the individual .vtu files.
FieldConvert functions that produce a .fld file output will also be created when running
in parallel. In this case when producing a .fld file a directory called test.fld (or the
specified output name) is created with the standard parallel field files placed within the
directory.
5.6 Processing large files in serial
When processing large files, it is not always convenient to run in parallel but process
each parallel partition in serial, for example when interpolating a solution field from one
mesh to another.
5.6.1 Using the nprocs and procid options

One option is to use the –nprocs and –procid command line options. For example,
the following command will interpolate partition 2 of a decomposition into 10 partitions
of fiile2.xml from file1.fld
FieldConvert --nprocs 10 --procid 2 \

-m interpfield:fromxml=file1.xml:fromfld=file1.fld \
file2.xml file2.fld
This call will only therefore consider the interpolation process across one partition
(namely, partition 2). To create the full interpolated field requires a loop over each of the
partitions, which, in a bash shell can be run as
for n in ‘seq 0 9‘; do

FieldConvert --nprocs 10 --procid $n \
-m interpfield:fromxml=file1.xml:fromfld=file1.fld \
file2.xml file2.fld
done
The resulting output will lie in a directory called file2.fld , with each of the different
parallel partitions in files with names P0000000.fld , P0000001.fld , . . . , P0000009.fld .
This is nearly a complete parallel field file. However, the Info.xml file, which contains
the information about which elements lie in each partition, is missing. This can be
generated by using the command
FieldConvert --nprocs 10 file2.xml file2.fld/Info.xml:info
Note the final :info extension on the last argument is necessary to tell FieldConvert
that you wish to generate an info file, but with the extension .xml . This syntax allows
the routine not to get confused with the input/output XML files.
5.6.2 Using the –part-only and –part-only-overlapping options

Another approach to serially proessing a large file is to initially process the file into
multiple partitions. This can be done with the –part-only option. So the command
FieldConvert --part-only 10 file.xml file.fld
will partition the mesh into 10 paritions and write each partition into a directory called
file_xml . If you enter this directory you will find partitioned XML files P0000000.xml ,
P0000001.xml , . . . , P0000009.xml which can then be processed individually as outlined
above.
There is also a –part-only-overlapping option, which can be run in the same fashion.
FieldConvert --part-only-overlapping 10 file.xml file.fld
In this mode, the mesh is partitioned into 10 partitions in a similar manner, but the
elements at the partition edges will now overlap, so that the intersection of each partition
5.6 Processing large files in serial 97
with its neighbours is non-empty. This is sometime helpful when, for example, producing a
global isocontour which has been smoothed. Applying the smoothed isocontour extraction
routine with the –part-only option will produce a series of isocontour where there will be
a gap between partitions, as the smoother tends to shrink the isocontour within a partition.
using the –part-only-overlapping option will still yield a shrinking isocontour, but the
overlapping partitions help to overlap the partiiton boundaries.
Part III
Solver Applications
98
Chapter 6
Advection-Diffusion-Reaction Solver
6.1 Synopsis
The ADRSolver is designed to solve partial differential equations of the form:

∂u
α + λu + ν∇u + ∇ · (D∇u) = f (6.1)
∂t
in either discontinuous or continuous projections of the solution field. For a full list of
the equations which are supported, and the capabilities of each equation, see the table
below.
Equation to solve EquationType Dimensions Projections
u=f Projection All Continuous/Discontinuous
∇2 u = 0 Laplace All Continuous/Discontinuous
2
∇ u=f Poisson All Continuous/Discontinuous
∇2 u + λu = f Helmholtz All Continuous/Discontinuous
∇2 u + V∇u = f SteadyAdvectionDiffusion 2D only Continuous/Discontinuous
2
∇ u + λu = f SteadyDiffusionReaction 2D only Continuous/Discontinuous
∇2 u + V∇u + λu = f SteadyAdvectionDiffusionReaction 2D only Continuous/Discontinuous

∂u
+ V∇u = f UnsteadyAdvection All Continuous/Discontinuous
∂t
∂u
= ∇2 u UnsteadyDiffusion All Continuous/Discontinuous
∂t
∂u
+ V∇u = ∇2 u UnsteadyAdvectionDiffusion All Continuous/Discontinuous
∂t
∂u
+ u∇u = 0 UnsteadyInviscidBurger 1D only Continuous/Discontinuous
∂t
Table 6.1 Equations supported by the ADRSolver with their capabilities.
6.2 Usage
99
100 Chapter 6 Advection-Diffusion-Reaction Solver
ADRSolver session.xml
6.3 Session file configuration
The type of equation which is to be solved is specified through the EquationType

SOLVERINFO option in the session file. This can be set as in table 6.1. At present, the
Steady non-symmetric solvers cannot be used in parallel.
6.3.1 Solver Info

The solver info are listed below:
• Eqtype: This sets the type of equation to solve, according to the table above.
• TimeIntegrationMethod: The following types of time integration methods have

been tested with each solver:
EqType Explicit Diagonally Implicit IMEX Implicit
UnsteadyAdvection X
UnsteadyDiffusion X X
UnsteadyAdvectionDiffusion X
UnsteadyInviscidBurger X
• Projection: The Galerkin projection used may be either:
– Continuous for a C0-continuous Galerkin (CG) projection.

– Discontinuous for a discontinous Galerkin (DG) projection.
• DiffusionAdvancement: This specifies how to treat the diffusion term. This will
be restricted by the choice of time integration scheme:
– Explicit Requires the use of an explicit time integration scheme.

– Implicit Requires the use of a diagonally implicit, IMEX or Implicit scheme.
• AdvectionAdvancement: This specifies how to treat the advection term. This

will be restricted by the choice of time integration scheme:
– Explicit Requires the use of an explicit or IMEX time integration scheme.

– Implicit Not supported at present.
• AdvectionType: Specifies the type of advection:

6.4 Examples 101
– NonConservative (for CG only).

– WeakDG (for DG only).
• DiffusionType:
– LDG .
• UpwindType:
– Upwind .
6.3.2 Parameters
The following parameters can be specified in the PARAMETERS section of the session file:
• epsilon : sets the diffusion coefficient .

Can be used in: SteadyDiffusionReaction, SteadyAdvectionDiffusionReaction, Un-
steadyDiffusion, UnsteadyAdvectionDiffusion.
Default value: 0.
• d00 , d11 , d22 : sets the diagonal entries of the diffusion tensor D.
Can be used in: UnsteadyDiffusion
Default value: All set to 1 (i.e. identity matrix).
• lambda : sets the reaction coefficient λ.

Can be used in: SteadyDiffusionReaction, Helmholtz, SteadyAdvectionDiffusionRe-
action
Default value: 0.
6.3.3 Functions
The following functions can be specified inside the CONDITIONS section of the session file:
• AdvectionVelocity : specifies the advection velocity V.
• InitialConditions : specifies the initial condition for unsteady problems.
• Forcing : specifies the forcing function f.
6.4 Examples
Example files for the ADRSolver are provided in solvers/ADRSolver/Examples
6.4.1 1D Advection equation

In this example, it will be demonstrated how the Advection equation can be solved on a
one-dimensional domain.
6.4.1.1 Advection equation

We consider the hyperbolic partial differential equation:
∂u ∂f
+ = 0, (6.2)
∂t ∂x
where f = au is the advection flux.
6.4.1.2 Input file

The input for this example is given in the example file Advection1D.xml
The geometry section defines a 1D domain consisting of 10 segments. On each segment

an expansion consisting of 4 Lagrange polynomials on the Gauss-Lobotto-Legendre points
is used as specified by
1 <EXPANSIONS>
2 <E COMPOSITE="C[0]" FIELDS="u" TYPE="GLL_LAGRANGE_SEM" NUMMODES="4"/>
3 </EXPANSIONS>
Since we are solving the unsteady advection problem, we must specify this in the solver
information. We also choose to use a discontinuous flux-reconstruction projection and
use a Runge-Kutta order 4 time-integration scheme.
1 
2 
3 
4 
5 
We choose to advect our solution for 20 time units with a time-step of 0.01 and so provide
the following parameters
1 FinTime = 20 
2 TimeStep = 0.01 
3 NumSteps = FinTime/TimeStep 
We also specify the advection velocity. We first define dummy parameters

1 advx = 1 
2 advy = 0 
and then define the actual advection function as

2 <E VAR="Vx" VALUE="advx" />
3 </FUNCTION>
Two boundary regions are defined, one at each end of the domain, and periodicity is
enforced
6.4 Examples 103
1 <BOUNDARYREGIONS>
2 C[1] 
3 C[2] 
5
7 <REGION REF="0">
8 
9 </REGION>
10 <REGION REF="1">
11 
12 </REGION>
Finally, we specify the initial value of the solution on the domain

2 <E VAR="u" VALUE="exp(-20.0*x*x)" />
3 </FUNCTION>
4
6 <E VAR="u" VALUE="exp(-20.0*x*x)" />
7 </FUNCTION>
6.4.1.3 Running the code
ADRSolver Advection1D.xml
To visualise the output, we can convert it into either TecPlot or VTK formats
FieldConvert Advection1D.xml Advection1D.fld Advection1D.dat

FieldConvert Advection1D.xml Advection1D.fld Advection1D.vtu
6.4.2 2D Helmholtz Problem

In this example, it will be demonstrated how the Helmholtz equation can be solved on a
two-dimensional domain.
6.4.2.1 Helmholtz equation

We consider the elliptic partial differential equation:
∇2 u + λu = f (6.3)
where ∇2 is the Laplacian and λ is a real positive constant.

6.4.2.2 Input file

The input for this example is given in the example file Helmholtz2D_modal.xml
The geometry for this problem is a two-dimensional octagonal plane containing both
triangles and quadrilaterals. Note that a mesh composite may only contain one type of
element. Therefore, we define two composites for the domain, while the rest are used for
enforcing boundary conditions.
1 <COMPOSITE>
2 <C ID="0"> Q[22-47] </C>
3 <C ID="1"> T[0-21] </C>
4 <C ID="2"> E[0-1] </C>
5 .
6 .
7 <C ID="10"> E[84,75,69,62,51,40,30,20,6] </C>
8 </COMPOSITE>
9
10 <DOMAIN> C[0-1] </DOMAIN>
For both the triangular and quadrilateral elements, we use the modified Legendre basis
with 7 modes (maximum polynomial order is 6).
1 <EXPANSIONS>
4 </EXPANSIONS>
Only one parameter is needed for this problem. In this example λ = 1 and the Continuous
Galerkin Method is used as projection scheme to solve the Helmholtz equation, so we
need to specify the following parameters and solver information.
1 <PARAMETERS>
2 Lambda = 1 
3 </PARAMETERS>
4
5 <SOLVERINFO>
6 
8 </SOLVERINFO>
All three basic boundary condition types have been used in this example: Dirichlet,
Neumann and Robin boundary. The boundary regions are defined, each of which
corresponds to one of the edge composites defined earlier. Each boundary region is then
assigned an appropriate boundary condition.
1 <BOUNDARYREGIONS>
2 C[2] 
3 .
4 .
5 C[10] 
7
6.4 Examples 105
9 <REGION REF="0">
10 <D VAR="u" VALUE="sin(PI*x)*sin(PI*y)" />
11 </REGION>
12 <REGION REF="1">
13 <R VAR="u" VALUE="sin(PI*x)*sin(PI*y)-PI*sin(PI*x)*cos(PI*y)"
14 PRIMCOEFF="1" />
15 </REGION>
16 <REGION REF="2">
17 <N VAR="u" VALUE="(5/sqrt(61))*PI*cos(PI*x)*sin(PI*y)-
18 (6/sqrt(61))*PI*sin(PI*x)*cos(PI*y)" />
19 </REGION>
20 .
21 .
We know that for f = −(λ+2π 2 )sin(πx)cos(πy), the exact solution of the two-dimensional
Helmholtz equation is u = sin(πx)cos(πy). These functions are defined specified to
initialise the problem and verify the correct solution is obtained by evaluating the L2
and Linf errors.
1 <FUNCTION NAME="Forcing">
2 <E VAR="u" VALUE="-(Lambda + 2*PI*PI)*sin(PI*x)*sin(PI*y)" />
3 </FUNCTION>
4
6 <E VAR="u" VALUE="sin(PI*x)*sin(PI*y)" />
7 </FUNCTION>
ADRSolver Test_Helmholtz2D_modal.xml
This execution should print out a summary of input file, the L2 and Linf errors and the
time spent on the calculation.
6.4.2.4 Post-processing
Simulation results are written in the file Helmholtz2D_modal.fld. We can choose to

visualise the output in Gmsh
FieldConvert Helmholtz2D_modal.xml Helmholtz2D_modal.fld Helmholtz2D_modal.vtu
which generates the file Helmholtz2D_modal.vtu which can be visualised and is shown
in Fig. 6.1
Figure 6.1 Solution of the 2D Helmholtz Problem.
6.4.3 Advection dominated mass transport in a pipe

The following example demonstrates the application of the ADRsolver for modelling
advection dominated mass transport in a straight pipe. Such a transport regime is
encountered frequently when modelling mass transport in arteries. This is because the
diffusion coefficient of small blood borne molecules, for example oxygen or adenosine
triphosphate, is very small O(10−10 ).
6.4.3.1 Background
The governing equation for modelling mass transport is the unsteady advection diffusion
equation:
∂u
+ v∇u + ∇2 u = 0
∂t
For small diffusion coefficient, , the transport is dominated by advection and this leads
to a very fine boundary layer adjacent to the surface which must be captured in order to
get a realistic representation of the wall mass transfer processes. This creates problems
not only from a meshing perspective, but also numerically where classical oscillations are
observed in the solution due to under-resolution of the boundary layer.
The Graetz-Nusselt solution is an analytical solution of a developing mass (or heat)

transfer boundary layer in a pipe. Previously this solution has been used as a benchmark
for the accuracy of numerical methods to capture the fine boundary layer which develops
for high Peclet number transport (the ratio of advection to diffusion). The solution
is derived based on the assumption that the velocity field within the mass transfer
boundary layer is linear i.e. the Schmidt number (the relative thickness of the momentum
to mass transfer boundary layer) is sufficiently large. The analytical solution for the
6.4 Examples 107
non-dimensional mass transfer at the wall is given by:
24/3 (P eR/z)1/3
Sh(z) = ,
g 1/3 Γ(4/3)
where z is the streamwise coordinate, R the pipe radius, Γ(4/3) an incomplete Gamma
function and P e the Peclet number given by:
2U R
Pe =

In the following we will numerically solver mass transport in a pipe and compare the
calculated mass transfer at the wall with the Graetz-Nusselt solution. The Peclet number
of the transport regime under consideration is 750000, which is physiologically relevant.
6.4.3.2 Input file

The geometry under consideration is a pipe of radius, R = 0.5 and length l = 0.5
Figure 6.2 Pipe.
Since the mass transport boundary layer will be confined to a very small layer adjacent
to the wall we do not need to mesh the interior region, hence the mesh consists of a layer
of ten prismatic elements over a thickness of 0.036R. The elements progressively grow
over the thickness of domain.
In this example we utilise heterogeneous polynomial order, in which the polynomial

order normal to the wall is higher so that we avoid unphysical oscillations, and hence the
incorrect solution, in the mass transport boundary layer. To do this we specify explicitly
the expansion type, points type and distribution in each direction as follows:
1 <EXPANSIONS>
3 NUMMODES="3,5,3"
4 BASISTYPE="Modified_A,Modified_A,Modified_B"
5 NUMPOINTS="4,6,3"
6 POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre,GaussRadauMAlpha1Beta0"
7 FIELDS="u" />
8 </EXPANSIONS>
The above represents a quadratic polynomial order in the azimuthal and streamwise
direction and 4th order polynomial normal to the wall for a prismatic element.
We choose to use a continuous projection and an first-order implicit-explicit time-

integration scheme. The DiffusionAdvancement and AdvectionAdvancement parameters
specify how these terms are treated.
3 
4 
We integrate for a total of 30 time units with a time-step of 0.0005, necessary to keep
the simulation numerically stable.
1 TimeStep = 0.0005 
2 FinalTime = 30 
3 NumSteps = FinalTime/TimeStep 
The value of the parameter is = 1/P e

1 epsilon = 1.33333e-6 
The analytical solution represents a developing mass transfer boundary layer in a pipe. In
order to reproduce this numerically we assume that the inlet concentration is a uniform
value and the outer wall concentration is zero; this will lead to the development of the
mass transport boundary layer along the length of the pipe. Since we do not model
explicitly the mass transfer in the interior region of the pipe we assume that the inner
wall surface concentration is the same as the inlet concentration; this assumption is valid
based on the large Peclet number meaning the concentration boundary layer is confined
to the region in the immediate vicinity of the wall. The boundary conditions are specified
as follows in the input file:
1 <BOUNDARYREGIONS>
2 C[3] 
3 C[4] 
4 C[2] 
5 C[5] 
7
6.4 Examples 109
9 <REGION REF="0">
10 <D VAR="u" VALUE="1" />
11 </REGION>
12 <REGION REF="1">
13 <N VAR="u" VALUE="0" />
14 </REGION>
15 <REGION REF="2">
16 <D VAR="u" VALUE="0" />
17 </REGION>
18 <REGION REF="3">
19 <D VAR="u" VALUE="1" />
20 </REGION>
The velocity field within the domain is fully devqeloped pipe flow (Poiseuille flow), hence
we can define this through an analytical function as follows:
2 <E VAR="Vx" VALUE="0" />
3 <E VAR="Vy" VALUE="0" />
4 <E VAR="Vz" VALUE="2.0*(1-(x*x+y*y)/0.25)" />
5 </FUNCTION>
We assume that the initial domain concentration is uniform everywhere and the same as
the inlet. This is defined by,
3 </FUNCTION>
6.4.3.3 Results
To compare with the analytical expression we numerically calculate the concentration
gradient at the surface of the pipe. This is then plotted against the analytical solution
by extracting the solution along a line in the streamwise direction, as shown in Fig. 6.3.
Figure 6.3 Concentration gradient at the surface of the pipe.

Chapter 7
Acoustic Perturbation Equations
Solver
7.1 Synopsis
The aim of APESolver is to predict aerodynamic sound generation. Through the

application of a splitting technique, the flow-induced acoustic field is totally decoupled
from the underlying incompressible hydrodynamic field. The acoustic perturbation
equations (APE-1/APE-4) proposed by Ewert and Shroeder are employed as the governing
equations of the acoustic field and they assure stable aeroacoustic simulation due to
the suppression of the term related to the production of perturbed vorticity. These
equations are similar to the linearised perturbed compressible equations, but while in the
original formulation the flow decomposition is based on solenoidal vortical perturbations
as well as irrotational acoustic perturbations, in this case perturbations are assumed to
be exclusively of acoustic nature.
∂p0 ∂γpu0i ∂ui p0

+ + = c2 qc
∂t ∂xi ∂xi
∂u0i ∂uj u0j ∂p0 /ρ
+ + = Ui
∂t ∂xi ∂xi
where (ui , p, ρ, c2 = γp/ρ) represents the base flow and (u0i , p0 ) the perturbations. c2 qc is
the acoustic source term.
7.2 Usage
APESolver session.xml
111
112 Chapter 7 Acoustic Perturbation Equations Solver
7.3.1 Solver Info
• Eqtype Specifies the equation to solve. This should be set to APE .
• UpwindType Specifies the numerical interface flux scheme. Currently, only APEUpwind
supported.
7.3.2 Parameters
• Gamma : Ratio of specific heats
7.3.3 Functions
• BaseFlow Baseflow (ui , p, ρ) defined by the variables u0,v0,w0,p0,rho0
• Source Source term c2 qc

• InitialConditions
7.4 Examples
7.4.1 Aeroacoustic Wave Propagation

In this section we explain how to set up a simple simulation of aeroacoustics in Nektar++.
We will study the propagation of an acoustic wave in the simple case where the base
flow is ui = 0, p = p∞ = 106 , ρ = ρ0 = 1.204. The geometry consists of 64 quadrilateral
elements, as shown in Fig. 7.1.
7.4.1.1 Input file

We require a discontinuous Galerkin projection and use an explicit fourth-order Runge-
Kutta time integration scheme. We therefore set the following solver information:
1 
2 
4 
To maintain numerical stability we must use a small time-step. The total simulation
time is 150 time units. Finally, we set the density, heat ratio and ambient pressure.
1 TimeStep = 0.00001 
2 NumSteps = 150 
3 FinTime = TimeStep*NumSteps 
4 Rho0 = 1.204 
5 Gamma = 1.4 
6 Pinfinity = 100000 
7.4 Examples 113
Figure 7.1 Geometry used for the example case of modelling propagation of acoustic waves
where ui = 0, p = p∞ = 106 , ρ = ρ0 = 1.204
Let us note that to solve efficiently this problem a discontinuous Garlerkin approach was
used. The system is excited via the initial conditions putting a Gaussian pulse for pulse
fluctuations. Finally, it is necessary to specify the base flow and the eventual source
terms using the following functions:
2 <E VAR="u0" VALUE="0" />
3 <E VAR="v0" VALUE="0" />
4 <E VAR="p0" VALUE="Pinfinity" />
5 <E VAR="rho0" VALUE="Rho0" />
6 </FUNCTION>
7
8 <FUNCTION NAME="Source">
9 <E VAR="S" VALUE="0" />
10 </FUNCTION>
11
12 
14 <E VAR="p" VALUE="100*exp(-32*(((x)*(x))+((y)*(y))))" />
15 <E VAR="u" VALUE="0" />
16 <E VAR="v" VALUE="0" />
17 </FUNCTION>

114 Chapter 7 Acoustic Perturbation Equations Solver
APESolver Test_pulse.xml
7.4.1.3 Results
Fig. 7.2 shows the pressure profile at different time steps, showing the acoustic propaga-
tion.
Figure 7.2
It is possible to show the profile of the pressure perturbations with respect to the spatial
coordinate. The pressure fluctuations, that are concentrated in a specific locations at the
beginning (as specified by the initial conditions), propagate with time and for sufficiently
large time the decay is exponential as predicted by literature [12].
Figure 7.3
Chapter 8
Cardiac Electrophysiology Solver
8.1 Synopsis
The CardiacEPSolver is used to model the electrophysiology of cardiac tissue, specifically

using the monodomain or bidomain model. These models are continuum models and
represent an average of the electrical activity over many cells. The system is a reaction-
diffusion system, with the reaction term modeling the flow of current in and out of the
cells using a separate set of ODEs.
8.1.1 Bidomain Model

The Bidomain model is given by the following PDEs,
∂ 2 Vi ∂ 2 Vi ∂(Vi − Ve )

gix + g iy = χ Cm + Gm (Vi − Ve )
∂x2 ∂y 2 ∂t
∂ 2 Ve ∂ 2 Ve ∂(Vi − Ve )

gex + gey = −χ Cm + Gm (Vi − Ve ) .
∂x2 ∂y 2 ∂t
However, when solving numerically, one often rewrites these equations in terms of the
transmembrane potential and extracellular potential,
∂Vm ∂ 2 Ve ∂ 2 Ve

χ Cm + Jion = gex + gey
∂t ∂x2 ∂y 2
2
∂ Ve 2
∂ Ve 2
∂ Vm ∂ 2 Vm
(gix + gex ) 2 + (giy + gey ) 2 = −gix − g iy
∂x ∂y ∂x2 ∂y 2
8.1.2 Monodomain Model

In the case where the intracellular and extracellular conductivities are proportional, that
is gix = kgex for some k, then the above two PDEs can be reduced to a single PDE:
∂Vm

χ Cm + Jion = ∇ · (σ∇Vm )
∂t
115
116 Chapter 8 Cardiac Electrophysiology Solver
8.1.3 Cell Models

The action potential of a cardiac cell can be modelled at either a biophysical level of
detail, including a number of transmembrane currents, or as a phenomenological model,
to reproduce the features of the action potential, with fewer variables. Each cell model
will include a unique system of ODEs to represent the gating variables of that model.
A number of ionic cell models are currently supported by the solver including:
• Courtemanche, Ramirez, Nattel, 1998
• Luo, Rudy, 1991
• ten Tusscher, Panfilov, 2006 (epicardial, endocardial and mid-myocardial variants)
Phenomological cell models are also supported:
• Aliev-Panfilov
• Fitzhugh-Nagumo
It is important to ensure that the units of the voltage and currents from the cell model
are consistent with the units expected by the tissue level solver (monodomain/bidomain).
We will show as an example the Courtemanche, Ramirez, Nattel, 1998 human atrial
model.
The monodomain equation:

∂Vm

χ Cm + Jion = ∇ · (σ∇Vm )
∂t
8.2 Usage
CardiacEPSolver session.xml
8.3.1 Solver Info
• Eqtype Specifies the PDE system to solve. The following values are supported:
– Monodomain : solve the monodomain equation.

– BidomainRoth : solve the bidomain equations using the Roth formulation.
• CellModel Specifies the cell model to use. Available cell models are
8.3 Session file configuration 117
Value Description No. of Var. Ref.

AlievPanfilov Phenomological 1 [2]
CourtemancheRamirezNattel98 Human atrial 20 [22]
FitzHughNagumo
Fox02
LuoRudy91 Mammalian ventricular 7 [21]
PanditGilesDemir03
TenTusscher06 Human ventricular 18 [35]
Winslow99
• Projection Specifies the Galerkin projection type to use. Only Continuous has
been extensively tested.
• TimeIntegrationMethod Specifies the time integration scheme to use for advancing

the PDE system. This must be an IMEX scheme. Suitable choices are: IMEXOrder1 ,
IMEXOrder2 , IMEXOrder3 , IMEXdirk_3_4_3 . The cell model state variables are
time advanced using Forward Euler for the ion concentrations, and Rush-Larsen
for the cell model gating variables.
• DiffusionAdvancement Specifies whether the diffusion is handled implicitly or

explicitly in the time integration scheme. The current code only supports Implicit
integration of the diffusion term. The cell model is always integrated explicitly.
8.3.2 Parameters
The following parameters can be specified in the PARAMETERS section of the session file.
Example values are taken from [13].
• Chi sets the surface-to-volume ratio (Units: mm−1 ).

Example: χ = 140mm−1
• Cm sets the specific membrane capacitance (Units: µF mm−2 ).

Example: Cm = 0.01µF mm−2
• Substeps sets the number of substeps taken in time integrating the cell model for
each PDE timestep.
Example: 4
• d_min , d_max , o_min , o_max specifies a bijective map to assign conductivity

values σ to intensity values µ when using the IsotropicConductivity function. The
intensity map is first thresholded to the range [dmin , dmax ] and then the conductivity
is calculated as
omax − omin
σ= (1 − µ) + omin
dmax − dmin
8.3.3 Functions
The following functions can be specified inside the CONDITIONS section of the session file.
If both are specified, the effect is multiplicative. Example values are taken from [13].
• IsotropicConductivity specifies the conductivity σ of the tissue.

Example: σ = 0.13341 mS mm−1 , based on σ = σσi +σ
i σe
e
, σi = 0.17, σe = 0.62mS mm−1
The variable name to use is intensity since the conductivity may be derived from
late-Gadolinium enhanced MRA imaging. Example specifications are
1 <E VAR="intensity" VALUE="0.13341" />
2 <F VAR="intensity" FILE="scarmap.con" />
where scarmap.con is a Nektar++ field file containing a variable intensity

describing the conductivity across the domain.
• AnisotropicConductivity specifies the conductivity σ of the tissue.
8.3.4 Filters
• CellHistoryPoints writes all cell model states over time at fixed points. Can be
used along with the HistoryPoints filter to record all variables at specific points
during a simulation.
1 <FILTER TYPE="CellHistoryPoints">
2 <PARAM NAME="OutputFile">crn.his</PARAM>
5 0.00 0.0 0.0
6 </PARAM>
7 </FILTER>
– OutputFile specifies the filename to write history data to.

– OutputFrequency specifies the number of steps between successive outputs.
– Points lists coordinates at which history data is to be recorded.
• CheckpointCellModel checkpoints the cell model. Can be used along with the
Checkpoint filter to record complete simulation state and regular intervals.
1 <FILTER TYPE="CheckpointCellModel">
2 <PARAM NAME="OutputFile"> session </PARAM>
4 </FILTER>
– OutputFile (optional) specifies the base filename to use. If not specified, the
session name is used. Checkpoint files are suffixed with the process ID and
the extension ‘.chk‘.
– OutputFrequency specifies the number of timesteps between checkpoints.
• Electrogram Computes virtual unipolar electrograms at a prescribed set of points.

1 <FILTER TYPE="Electrogram">
2 <PARAM NAME="OutputFile"> session </PARAM>
5 0.0 0.5 0.7
6 1.0 0.5 0.7
7 2.0 0.5 0.7
8 </PARAM>
9 </FILTER>
– OutputFile (optional) specifies the base filename to use. If not specified, the
session name is used. The extension ‘.ecg‘ is appended if not already specified.
– OutputFrequency specifies the number of resolution of the electrogram data.
– Points specifies a list of coordinates at which electrograms are desired. They
must not lie within the domain.
• Benchmark Records spatially distributed event times for activation and repolarisa-
tion (recovert) during a simulation, for undertaking benchmark test problems.
1 <FILTER TYPE="Benchmark">
2 <PARAM NAME="ThresholdValue"> -40.0 </PARAM>
3 <PARAM NAME="InitialValue"> 0.0 </PARAM>
4 <PARAM NAME="OutputFile"> benchmark </PARAM>
5 <PARAM NAME="StartTime"> 0.0 </PARAM>
6 </FILTER>
– ThresholdValue specifies the value above which tissue is considered to be

depolarised and below which is considered repolarised.
– InitialValue specifies the initial value of the activation or repolarisation
time map.
– OutputFile specifies the base filename of activation and repolarisation maps
output from the filter. This name is appended with the index of the event and
the suffix ‘.fld‘.
– StartTime (optional) specifies the simulation time at which to start detecting
events.
8.3.5 Stimuli
Electrophysiological propagaion is initiated through the stimulus current Iion . The
STIMULI section describes one or more regions of stimulus and the time-dependent
protocol with which they are applied.
1 <STIMULI>
2 ...
3 </STIMULI>
A number of stimulus types are available
8.3.5.1 Stimulus types
• StimulusRect stimulates a cuboid-shaped region of the domain, specified by two

coordinates (x1 , y1 , z1 ) and (x2 , y2 , z2 ). An additional parameter specifies the
"smoothness" of the boundaries of the region; higher values produce a sharper
boundary. Finally, the maximum strength of the stimulus current is specified in
µA/mm3
1 <STIMULUS TYPE="StimulusRect" ID="0">
2 <p_x1> -15.24 </p_x1>
3 <p_y1> 14.02 </p_y1>
4 <p_z1> 6.87 </p_z1>
5 <p_x2> 12.23 </p_x2>
6 <p_y2> 16.56 </p_y2>
7 <p_z2> 8.88 </p_z2>
8 <p_is> 100.00 </p_is>
9 <p_strength> 50.0 </p_strength>
10 </STIMULUS>
• StimulusCirc stimulates a spherical region of the domain, as specified by a centre

and radius. The smoothness and strength parameters are also specified as for
‘StimulusRect‘.
1 <STIMULUS TYPE="StimulusCirc" ID="0">
2 <p_x1> -15.24 </p_x1>
3 <p_y1> 14.02 </p_y1>
4 <p_z1> 6.87 </p_z1>
5 <p_r1> 12.23 </p_r1>
6 <p_is> 100.00 </p_is>
7 <p_strength> 50.0 </p_strength>
8 </STIMULUS>
8.3.5.2 Protocols
A protocol specifies the time-dependent function indicating the strength of the stimulus
and one such PROTOCOL section should be included within each STIMULUS . This can be
expressed as one of:
• ProtocolSingle a single stimulus is applied at a given start time and for a given
duration
1 <PROTOCOL TYPE="ProtocolSingle">
2 <START> 0.0 </START>
3 <DURATION> 2.0 </DURATION>
4 </PROTOCOL>
• ProtocolS1 a train of pulses of fixed duration applied at a given start time and
with a given cycle length.
1 <PROTOCOL TYPE="ProtocolS1">
4 <S1CYCLELENGTH> 300.0 </S1CYCLELENGTH>
5 <NUM_S1> 5 </NUM_S1>
6 </PROTOCOL>
• ProtocolS1S2 same as ‘ProtocolS1‘ except with an additional single pulse applied

at a different cycle length at the end of the train of S1 pulses.
1 <PROTOCOL TYPE="ProtocolS1S2">
5 <NUM_S1> 5 </NUM_S1>
7 </PROTOCOL>
Chapter 9
Compressible Flow Solver
9.1 Synopsis
The CompressibleFlowSolver allows us to solve the unsteady compressible Euler and

Navier-Stokes equations for 1D/2D/3D problems using a discontinuous representation
of the variables. In the following we describe both the compressible Euler and the
Navier-Stokes equations.
9.1.1 Euler equations

The Euler equations can be expressed as a hyperbolic conservation law in the form
∂q ∂fi ∂gi ∂hi

+ + + = 0, (9.1)
∂t ∂x ∂y ∂z
where q is the vector of the conserved variables, fi = fi (q), gi = gi (q) and hi = hi (q)
are the vectors of the inviscid fluxes
       

 ρ 
 
 ρu 
 
 ρv 
 
 ρw 

 ρu  p + ρu2 ρuv ρuw

 
 
 
 
 
 
 

  
 
 
 
 
 

q= ρv , fi = ρuv , gi = p + ρv 2 , hi = ρvw ,


 ρw





 ρuw





 ρvw





 p + ρw 2 


 
 
 
 
 
 
 

E u(E + p) v(E + p) w(E + p)
       
(9.2)
where ρ is the density, u, v and w are the velocity components in x, y and z directions, p
is the pressure and E is the total energy. In this work we considered a perfect gas law
for which the pressure is related to the total energy by the following expression
p 1
E= + ρ(u2 + v 2 + w2 ), (9.3)
γ−1 2
where γ is the ratio of specific heats.
122
9.1 Synopsis 123
9.1.2 Compressible Navier-Stokes equations

The Navier-Stokes equations include the effects of fluid viscosity and heat conduction and
are consequently composed by an inviscid and a viscous flux. They depend not only on
the conserved variables but also, indirectly, on their gradient. The second order partial
differential equations for the three-dimensional case can be written as:
∂q ∂f ∂g ∂h
+ + + = 0, (9.4)
∂t ∂x ∂y ∂z
where q is the vector of the conserved variables, f = f (q, ∇(q)), g = g(q, ∇(q)) and
h = h(q, ∇(q)) are the vectors of the fluxes which can also be written as:
f = fi − fv , g = gi − gv , h = hi − hv , (9.5)
where fi , gi and hi are the inviscid fluxes of Eq. (9.2) and fv , gv and hv are the viscous
fluxes which take the following form:
   

 0 
 
 0 

τxx τxy

 
 
 


 
 
 

fv = τyx , gv = τyy ,
   



 τzx 






 τzy 



uτxx + vτyx + wτzx + kTx uτxy + vτyy + wτzy + kTy
   
  (9.6)

 0 

τxz

 


 

hv = τyz ,
 



 τzz 



uτxz + vτyz + wτzz + kTz
 
where τxx , τxy , τxz , τyx , τyx , τyy , τyz , τzx , τzy and τzz are the components of the stress
tensor1
u +v +w u +v +w
τxx = 2µ ux − x 3y z , τyy = 2µ vy − x 3y z ,

ux +vy +wz
τzz = 2µ wz − 3 , τxy = τyx = µ(vx + uy ), (9.7)
τyz = τzy = µ(wy + vz ), τzx = τxz = µ(uz + wx ).
where µ is the dynamic viscosity calculated using the Sutherland’s law and k is the
thermal conductivity.
9.1.3 Numerical discretisation

In Nektar++ the spatial discretisation of the Euler and of the Navier-Stokes equations is
projected in the polynomial space via a discontinuous projection. Specifically we make
use either of the discontinuous Galerkin (DG) method or the Flux Reconstruction (FR)
1
Note that we use Stokes hypothesis λ = −2/3.
124 Chapter 9 Compressible Flow Solver
approach. In both the approaches the physical domain Ω is divided into a mesh of N non-
overlapping elements Ωe and the solution is allowed to be discontinuous at the boundary
between two adjacent elements. Since the Euler as well as the Navier-Stokes equations are
defined locally (on each element of the computational domain), it is necessary to define a
term to couple the elements of the spatial discretisation in order to allow information to
propagate across the domain. This term, called numerical interface flux, naturally arises
from the discontinuous Galerkin formulation as well as from the Flux Reconstruction
approach.
For the advection term it is common to solve a Riemann problem at each interface of the
computational domain through exact or approximated Riemann solvers. In Nektar++
there are different Riemann solvers, one exact and nine approximated. The exact Riemann
solver applies an iterative procedure to satisfy conservation of mass, momentum and
energy and the equation of state. The left and right states are connected either with
the unknown variables through the Rankine-Hugoniot relations, in the case of shock,
or the isentropic characteristic equations, in the case of rarefaction waves. Across the
contact surface, conditions of continuity of pressure and velocity are employed. Using
these equations the system can be reduced to a non-linear algebraic equation in one
unknown (the velocity in the intermediate state) that is solved iteratively using a Newton
method. Since the exact Riemann solver gives a solution with an order of accuracy that
is related to the residual in the Newton method, the accuracy of the method may come
at high computational cost. The approximated Riemann solvers are simplifications of
the exact solver.
Concerning the diffusion term, the coupling between the elements is achieved by using a
local discontinuous Galerkin (LDG) approach as well as five different FR diffusion terms.
The boundary conditions are also implemented by exploiting the numerical interface
fluxes just mentioned. For a more detailed description of the above the interested reader
can refer to [9] and [24].
9.2 Usage
CompressibleFlowSolver session.xml
In the following we describe the session file configuration. Specifically we consider the
sections under the tag <CONDITIONS> in the session (.xml) file.
Parameters
Under this section it is possible to set the parameters of the simulation.
1 <PARAMETERS>
2 TimeStep = 0.0000001 
3 FinTime = 1.0 

7 Gamma = 1.4 
8 pInf = 101325 
9 rhoInf = 1.225 
10 TInf = pInf/(287.058*rhoInf) 
11 Twall = pInf/(287.058*rhoInf)+15.0 
12 uInf = 147.4 
13 vInf = 0.0 
14 wInf = 0.0 
15 mu = 1e-5 
16 Pr = 0.72 
17 thermalConductivity = 0.02 
18 </PARAMETERS>
• TimeStep is the time-step we want to use;
• FinTime is the final physical time at which we want our simulation to stop;
• NumSteps is the equivalent of FinTime but instead of specifying the physical final
time we specify the number of time-steps;
• IO_CheckSteps sets the number of steps between successive checkpoint files;
• IO_InfoSteps sets the number of steps between successive info stats are printed
to screen;
• Gamma ratio of the specific heats. Default value = 1.4;
• pInf farfield pressure (i.e. p∞ ). Default value = 101325 P a;
• rhoInf farfield density (i.e. ρ∞ ). Default value = 1.225 Kg/m3 ;
• TInf farfield temperature (i.e. T∞ ). Default value = 288.15 K;
• Twall temperature at the wall when isothermal boundary conditions are employed
(i.e. Tw ). Default value = 300.15K;
• uint farfield X-component of the velocity (i.e. u∞ ). Default value = 0.1 m/s;
• vInf farfield Y -component of the velocity (i.e. v∞ ). Default value = 0.0 m/s;
• wInf farfield Z-component of the velocity (i.e. w∞ ). Default value = 0.0 m/s;
• mu dynamic viscosity (i.e. µ∞ ). Default value = 1.78e-05 P as;
• Pr Prandtl number. Default value = 0.72;

• thermalConductivity thermal conductivity (i.e. κ∞ ). This can be set as an

alternative to Pr , in which case the Prandtl number is calculated from κ∞ (it is
only possible to set one of them). By default, this is obtained from the Prandtl
number;
Solver info
Under this section it is possible to set the solver information.
1 <SOLVERINFO>
2 
4 
5 
7 
8 
9 
10 </SOLVERINFO>
• EQType is the tag which specify the equations we want solve:
– NavierStokesCFE (Compressible Navier-Stokes equations);

– EulerCFE (Compressible Euler equations).
– IsentropicVortex (Isentropic vortex test case).
– RinglebFlow (Ringleb flow test case).
• Projection is the type of projection we want to use:
– DisContinuous .
Note that the Continuous projection is not supported in the Compressible
Flow Solver.
• AdvectionType is the advection operator we want to use:
– WeakDG (classical DG in weak form);

– FRDG (Flux-Reconstruction recovering nodal DG scheme);
– FRSD (Flux-Reconstruction recovering a spectral difference (SD) scheme);
– FRHU (Flux-Reconstruction recovering Huynh (G2) scheme);
– FRcmin (Flux-Reconstruction with c = cmin );
– FRcinf (Flux-Reconstruction with c = ∞).
• DiffusionType is the diffusion operator we want to use for the Navier-Stokes

equations:
– LDGNS (LDG);
– LFRDGNS (Flux-Reconstruction recovering nodal DG scheme);

– LFRSDNS (Flux-Reconstruction recovering a spectral difference (SD) scheme);
– LFRHUNS (Flux-Reconstruction recovering Huynh (G2) scheme);
– LFRcminNS (Flux-Reconstruction with c = cmin );
– LFRcinfNS (Flux-Reconstruction with c = ∞).
• TimeIntegrationMethod is the time-integration scheme we want to use. Note that

only an explicit discretisation is supported:
– ForwardEuler ;
– RungeKutta2_ImprovedEuler ;
– ClassicalRungeKutta4 .
• UpwindType is the numerical interface flux (i.e. Riemann solver) we want to use
for the advection operator:
– AUSM0 ;
– AUSM1 ;
– AUSM2 ;
– AUSM3 ;
– Average ;
– ExactToro ;
– HLL ;
– HLLC ;
– LaxFriedrichs ;
– Roe .
• ViscosityType is the viscosity type we want to use:
– Constant (Constant viscosity);

– Variable (Variable viscosity through the Sutherland’s law.);
Boundary conditions
In this section we can specify the boundary conditions for our problem. First we need to
define the variables under the section VARIABLES . For a 1D problem we have:
1 <VARIABLES>
2 <V ID="0"> rho </V>
3 <V ID="1"> rhou </V>
4 <V ID="4"> E </V>
5 </VARIABLES>
For a 2D problem we have

1 <VARIABLES>
2 <V ID="0"> rho </V>
3 <V ID="1"> rhou </V>
4 <V ID="2"> rhov </V>
5 <V ID="4"> E </V>
6 </VARIABLES>
For a 3D problem we have:

1 <VARIABLES>
2 <V ID="0"> rho </V>
3 <V ID="1"> rhou </V>
4 <V ID="2"> rhov </V>
5 <V ID="3"> rhow </V>
6 <V ID="4"> E </V>
7 </VARIABLES>
After having defined the variables depending on the dimensions of the problem we want
to solve it is necessary to specify the boundary regions on which we want to define the
boundary conditions:
1 <BOUNDARYREGIONS>
2 C[100] 
Finally we can specify the boundary conditions on the regions specified under BOUNDARYREGIONS .
In the following some examples for a 2D problem:
• Slip wall boundary conditions:

2 <REGION REF="0">
3 <D VAR="rho" USERDEFINEDTYPE="Wall" VALUE="0" />
4 <D VAR="rhou" USERDEFINEDTYPE="Wall" VALUE="0" />
5 <D VAR="rhov" USERDEFINEDTYPE="Wall" VALUE="0" />
6 <D VAR="E" USERDEFINEDTYPE="Wall" VALUE="0" />
7 </REGION>
• No-slip wall boundary conditions:

2 <REGION REF="0">
3 <D VAR="rho" USERDEFINEDTYPE="WallViscous" VALUE="0" />
4 <D VAR="rhou" USERDEFINEDTYPE="WallViscous" VALUE="0" />
5 <D VAR="rhov" USERDEFINEDTYPE="WallViscous" VALUE="0" />
6 <D VAR="E" USERDEFINEDTYPE="WallViscous" VALUE="0" />
7 </REGION>
• Adiabatic wall boundary conditions:

2 <REGION REF="0">
3 <D VAR="rho" USERDEFINEDTYPE="WallAdiabatic" VALUE="0" />
4 <D VAR="rhou" USERDEFINEDTYPE="WallAdiabatic" VALUE="0" />
5 <D VAR="rhov" USERDEFINEDTYPE="WallAdiabatic" VALUE="0" />
6 <D VAR="E" USERDEFINEDTYPE="WallAdiabatic" VALUE="0" />
7 </REGION>
• Farfield boundary conditions (including inviscid characteristic boundary conditions):

2 <REGION REF="0">
3 <D VAR="rho" VALUE="rhoInf" />
4 <D VAR="rhou" VALUE="rhoInf*uInf" />
5 <D VAR="rhov" VALUE="rhoInf*vInf" />
6 <D VAR="E"
7 VALUE="pInf/(Gamma-1)+0.5*rhoInf*(uInf*uInf+vInf*vInf+wInf*wInf)"/>
8 </REGION>
• Pressure outflow boundary conditions:

2 <REGION REF="0">
3 <D VAR="rho" USERDEFINEDTYPE="PressureOutflow" VALUE="0" />
4 <D VAR="rhou" USERDEFINEDTYPE="PressureOutflow" VALUE="0" />
5 <D VAR="rhov" USERDEFINEDTYPE="PressureOutflow" VALUE="0" />
6 <D VAR="E" USERDEFINEDTYPE="PressureOutflow" VALUE="0" />
7 </REGION>
Initial conditions and exact solution

Under the two following sections it is possible to define the initial conditions and the
exact solution (if existent).
2 <E VAR="rho" VALUE="rhoInf"/>
3 <E VAR="rhou" VALUE="rhoInf*uInf" />
4 <E VAR="rhov" VALUE="rhoInf*vInf" />
5 <E VAR="E"
7 </FUNCTION>
8
10 <E VAR="rho" VALUE="rhoInf" />
11 <E VAR="rhou" VALUE="rhoInf*uInf" />
12 <E VAR="rhov" VALUE="rhoInf*vInf" />
13 <E VAR="E"
15 </FUNCTION>
9.4 Examples
9.4.1 Shock capturing

Compressible flow is characterised by abrupt changes in density within the flow domain
often referred to as shocks. These discontinuities lead to numerical instabilities (Gibbs
phenomena). This problem is prevented by locally adding a diffusion term to the equations
to damp the numerical fluctuations. These fluctuations in an element are identified using
a sensor algorithm which quantifies the smoothness of the solution within an element.
The value of the sensor in an element is defined as
||ρpe − ρp−1e ||L2
Se = p (9.8)
||ρe ||L2
An artificial diffusion term is introduced locally to the Euler equations to deal with flow
discontinuity and the consequential numerical oscillations. Two models are implemented,
a non-smooth and a smooth artificial viscosity model.
9.4.1.1 Non-smooth artificial viscosity model

For the non-smooth artificial viscosity model the added artificial viscosity is constant in
each element and discontinuous between the elements. The Euler system is augmented
by an added laplacian term on right hand side of equation 9.1. The diffusivity of the
system is controlled by a variable viscosity coefficient . The value of is dependent
on 0 , which is the maximum viscosity that is dependent on the polynomial order (p),
the mesh size (h) and the maximum wave speed and the local sensor value. Based on
pre-defined sensor threshold values, the variable viscosity is set accordingly

 0
 if se < sκ − κ
π(Se −sκ )
= 0 0.5 1 + sin 2κ if sκ − κ < Se < sκ + κ (9.9)

1 if se > sκ + κ

To enable the non-smooth viscosity model, the following line has to be added to the
SOLVERINFO section:
1 <SOLVERINFO>
2 
3 <SOLVERINFO>
The diffusivity is controlled by the following parameters:

1 <PARAMETERS>
2 Skappa = -1.3 
3 Kappa = 0.2 
4 mu0 = 1.0 
5 </PARAMETERS>
where mu0 is the maximum values for the viscosity coefficient, Kappa is half of the width
of the transition interval and Skappa is the value of the centre of the interval.
9.4 Examples 131
Mach ADViscCoeff
1.3 0.28
1.2 0.26
0.5 1.1 0.5 0.24
1 0.22
0.9 0.2
0.8 0.18
0.7 0.16
0.6 0.14
0.5
y
y
0.12
0.4 0.1
0.3 0.08
0 0 0.06
0.04
0.02
-0.5 -0.5
0 0.5 1 0 0.5 1
x x
Figure 9.1 (a) Steady state solution for M = 0.8 flow at α = 1.25◦ past a NACA 0012 profile,
(b) Artificial viscosity () distribution
9.4.1.2 Smooth artificial viscosity model

For the smooth artificial viscosity model an extra PDE for the artificial viscosity is
appended to the Euler system
∂ 1 h

= ∇ · (∇) + λmax Sκ − on Ω
∂t τ p
(9.10)
∂
=0 on Γ
∂n
where Sκ is a normalised sensor value and serves as a forcing term for the artificial
viscosity. A smooth artificial viscosity distribution is obtained.
To enable the smooth viscosity model, the following line has to be added to the
SOLVERINFO section:
1 <SOLVERINFO>
2 
3 <SOLVERINFO>
Furthermore, the extra viscosity variable eps has to be added to the variable list:
1 <VARIABLES>
2 <V ID="0"> rho </V>
3 <V ID="1"> rhou </V>
4 <V ID="2"> rhov </V>
5 <V ID="4"> E </V>
6 <V ID="5"> eps </V>
7 </VARIABLES>
A similar addition has to be made for the boundary conditions and initial conditions.
The tests that have been run started with a uniform homogeneous boundary condition
and initial condition. The following parameters can be set in the xml session file:
1 <PARAMETERS>
2 Skappa = -1.3 
3 Kappa = 0.2 
4 mu0 = 1.0 
5 FH = 3 
6 FL = 0.01*FH 
7 C1 = 0.03 
8 C2 = 5/3*C1 
9 </PARAMETERS>
where for now FH and FL are used to tune which range of the sensor is used as a forcing
term and C1 and C2 are fixed constants which can be played around with to make the
model more diffusive or not. However these constants are generally fixed.
9.4.2 Variable polynomial order

A sensor based p-adaptive algorithm is implemented to optimise the computational cost
and accuracy. The DG scheme allows one to use different polynomial orders since the
fluxes over the elements are determined using a Riemann solver and there is now further
coupling between the elements. Furthermore, the initial p-adaptive algorithm uses the
same sensor as the shock capturing algorithm to identify the smoothness of the local
solution so it rather straightforward to implement both algorithms at the same time.
The polynomial order in each element can be adjusted based on the sensor value that
is obtained. Initially, a converged solution is obtained after which the sensor in each
element is calculated. Based on the determined sensor value and the pre-defined sensor
thresholds, it is decided to increase, decrease or maintain the degree of the polynomial
approximation in each element and a new converged solution is obtained.


 pe − 1 if se > sds

 p +1
e if ssm < se < sds
pe = (9.11)
 pe
 if sf l < se < ssm


pe − 1 if se < sf l
For now, the threshold values se , sds , ssm and sf l are determined empirically by looking
at the sensor distribution in the domain. Once these values are set, two .txt files are
outputted, one that has the composites called VariablePComposites.txt and one with the
expansions called VariablePExpansions.txt. These values have to copied into a new .xml
file to create the adapted mesh.
9.4.3 De-Aliasing Techniques

Aliasing effects, arising as a consequence of the nonlinearity of the underlying problem,
need to be address to stabilise the simulations. Aliasing appears when nonlinear quantities
9.4 Examples 133
are calculated at an insufficient number of quadrature points. We can identify two types
of nonlinearities:
• PDE nonlinearities, related to the nonlinear and quasi-linear fluxes.
• Geometrical nonlinearities, related to the deformed/curves meshes.
We consider two de-aliasing strategies based on the concept of consistent integration:
• Local dealiasing: It only targets the PDE-aliasing sources, applying a consistent

integration of them locally.
• Global dealiasing: It targets both the PDE and the geometrical-aliasing sources. It
requires a richer quadrature order to consistently integrate the nonlinear fluxes,
the geometric factors, the mass matrix and the boundary term.
Since Nektar++ tackles separately the PDE and geometric aliasing during the projection
and solution of the equations, to consistently integrate all the nonlinearities in the
compressible NavierStokes equations, the quadrature points should be selected based on
the maximum order of the nonlinearities:
max(2Pexp , Pgeom ) 3
Qmin = Pexp + + (9.12)
2 2
where Qmin is the minimum required number of quadrature points to exactly integrate
the highest-degree of nonlinearity, Pexp being the order of the polynomial expansion
and Pgeom being the geometric order of the mesh. Bear in mind thatwe are using a
discontinuous discretisation, meaning that aliasing effect are not fully controlled, since
the boundary terms introduce non-polynomial functions into the problem.
To enable the global de-aliasing technique, modify the number of quadrature points by:
1 <E COMPOSITE="[101]"
2 BASISTYPE="Modified_A,Modified_A"
3 NUMMODES="7,7"
4 POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre"
5 NUMPOINTS="14,14"
6 FIELDS="rho,rhou,rhov,E"
7 />
where NUMMODES corresponds to P +1, where P is the order of the polynomial used to
approximate the solution. NUMPOINTS specifies the number of quadrature points.
Chapter 10
Incompressible Navier-Stokes Solver
10.1 Synopsis
A useful tool implemented in Nektar++ is the incompressible Navier Stokes solver that
allows one to solve the governing equation for viscous Newtonians fluids governed by:
∂u
+ u · ∇u = −∇p + ν∇2 u + f (10.1a)
∂t
∇·u=0 (10.1b)
where V is the velocity, p is the specific pressure (including density) and ν the kinematic
viscosity.
10.1.1 Velocity Correction Scheme

The first approach uses a splitting/projection method where the velocity system and
the pressure are typically decoupled. Splitting schemes are typically favoured for their
numerical efficiency since the velocity and pressure are handled independently, requiring
the solution of three (in two dimensions) elliptic systems of rank N (opposed to a single
system of rank 3N solved in the Stokes problem). However, a drawback of this approach
is the splitting scheme error which is introduced when decoupling the pressure and the
velocity system, although this can be made consistent with the overall temporal accuracy
of the scheme by appropriate discretisation of the pressure boundary conditions.
10.1.1.1 High order splitting scheme

In the original approach a stiffly-stable time integration was proposed in the work of
Karniadakis, Israeli and Orszag [19]. This was then later fully analysed in the work of
Guermond and Shen [16].
134
10.1 Synopsis 135
Briefly, high order splitting scheme was originally proposed in three steps involving
explicit advection of the non-linear terms, followed by the solution of the pressure Poisson
system and finally solving a Helmholtz problem to enforce the viscous terms and velocity
boundary conditions. In the following however we briefly formulate this scheme as a two
steps using a formulation outline by Guermond and Shen.
1. In the first step we formulate a weak pressure Poisson problem by taking the inner
product over the solution domain Ω of equation (10.1a) with respect to the gradient
of the test basis, ∇q, i.e.
∂u
Z Z Z Z
∇q · + ∇q · N(u) = − ∇q · ∇p + ∇q · ν∇2 u (10.2)
Ω ∂t Ω Ω Ω
R
where N (u) = u · ∇u. We recall that the term Ω ∇q · ∇p is the weak approximation
to the Laplacian operator for pressure. To decouple this term from the velocity
system a few steps are necessary. Using the identity
∇2 u = −∇ × ∇ × u + ∇(∇ · u)
we can enforce the divergence to be zero by setting the last term to zero. If we now
integrate the 1st, 2nd and last term in equation (10.2) by parts we can obtain the
weak pressure equation
!
∂u n+1
Z Z
n+1
∇q · ∇p = q∇· + N(u)n+1
Ω Ω ∂t
!
∂u n+1
Z
− q + N(u)n+1 − ν∇ × ∇ × un+1 · n (10.3)
∂Ω ∂t
where ∂Ω is the boundary of the domain and we have used the factor that ∇ · (∇ ×
∇ × u) = 0. To get the final form of the weak pressure Poisson equation we can
use a backward approximation of the time derivative to obtain
∂u n+1 γ0 ũn+1 − û
' (10.4)
∂t ∆t
where ũn+1 is an intermediate velocity upon which to decouple the system we
impose that ∇ · ũn+1 = 0 and
( (
1, if J = 1 un , if J = 1
γ0 = 3 û = n 1 n−1
2 , if J = 2 2u − 2 u , if J = 2.,
Finally we introduce a consistent extrapolation for the non-linear terms and the
curl of vorticity terms of the form:
(
∗,n+1 Nn , if J = 1
N =
2Nn − Nn−1 , if J = 2.
136 Chapter 10 Incompressible Navier-Stokes Solver
A similar extrapolation can be used on the curl-curl term to end up with the final
weak pressure approximation
û
Z Z
∇q · ∇pn+1 = q∇· − + N(u)∗,n+1
Ω Ω ∆t
!
∂u n+1
Z
− q + N(u)∗.n+1 − ν(∇ × ∇ × u)∗,n+1 · n (10.5)
∂Ω ∂t
We note this can be recast into an equivalent strong form of the pressure Poisson
equation of the form
û
∇2 pn+1 = ∇ · − N∗,n+1 (10.6)
∆t
with consistent Neumann boundary conditions prescribed as
∂p n+1 h ∂u n+1 i
=− − ν(∇ × ∇ × u)∗,n+1 + N∗,n+1 · n (10.7)
∂n ∂t
2. The second step is discretise equation (10.1a) at time level n + 1, use the pressure
at n + 1 from the first step and solve for the velocity un+1 .
In this step now approximate the time derivative using
∂u n+1 γ0 un+1 − û
' (10.8)
∂t ∆t
which leads us to the Helmholtz problem

γ0 n+1 γ
0 1
∆− u =− û + ∇pn+1 (10.9)
ν∆t ν∆t ν
This scheme is activated in the SolverInfo section with the SolverType specification:
10.1.1.2 Velocity Correction Scheme with a Weak Pressure formulation

As presented in the previous section in the work of Guermond and Shen [?] and subsequent
work they formulate the pressure in a weak rather than strong form to obtain the pressure
Poisson system. Therefore if we take the inner product of equation (10.1a) with respect
to the gradient of the test space, ∇q, we obtain equation (10.2)
We again make the approximation
∂u n+1 γ0 ũn+1 − û
' . (10.10)
∂t ∆t
However this time we only integrate by parts the last term and do not integrate the
non-linear term by parts. However we still need to enforce the condition that ∇ · ũn+1 = 0
10.1 Synopsis 137
and so we also integrate just this part of the time derivate by parts to arrive at a weak
pressure system of the form:
γ0 û
Z Z Z
∇q · ∇pn+1 + q ũn+1 · n = − N?,n+1 )
∇q · (
Ω ∆t ∂Ω0 Ω ∆t
γ0
Z Z
∗,n+1
− q ν(∇ × ∇ × u) · n + qwn+1 · n (10.11)
∆t ∂Ωd
S
∂Ωd ∂Ω0
where ∂Ωd is the Dirichlet boundary conditions for the velocity and ∂Ω0 is the outflow
boundary.
This scheme is activated in the SolverInfo section with the SolverType specification:
1 
10.1.1.3 Specifying pressure boundary conditions

In order to specify the pressure boundary conditions given by equation (10.7) or for the
equivalent conditions in the VCSWeakPressure scheme the USERDEFINEDTYPE condition
“H” can be used. Therefore a zero velocity wall boundary condition on boundary region
0 in two-dimensions can be specified as
2 <REGION REF="0">
6 </REGION>
10.1.1.4 Outflow boundary conditions

The most straightforward outflow condition is to specify fully developed conditions of
∇un+1 · n = 0 and p = 0 which can be specified as
2 <REGION REF="0">
3 <N VAR="u" VALUE="0" />
4 <N VAR="v" VALUE="0" />
5 <D VAR="p" VALUE="0" />
6 </REGION>
However when energetic vortices pass through an outflow region one can experience
instabilities as identified by the work of Dong, Karnidakis and Chryssostomidis [11]. In
this paper they suggest to impose a pressure Dirichlet outflow condition of the form
1
pn+1 = ν∇u∗,n+1 · n − | u∗,n+1 |2 So (n · u∗,n+1 ) + fbn+1 · n (10.12)
2
with a step function defined by So (n · u) = 12 (1 − tanh un·u

0δ
), where u0 is the characteristic
velocity scale and δ is a non-dimensional positive constant chosen to be sufficiently
small. fb is the forcing term in this case the analytical conditions can be given but
if these are not known explicitly, it is set to zero, i.e. fb = 0. (see the test Ko-
vaFlow_m8_short_HOBC.xml for a non-zero example). Note that in the paper [11]
they define this term as the negative of what is shown here so that it could be use
used to impose a default pressure values. This does however mean that the forcing
term is imposed through the velocity components u, v by specifying the entry VALUE
(An example can be found in ChanFlow_m3_VCSWeakPress_ConOBC.xml). For the
velocity component one can specify
1 h n+1 1 i
∇un+1 · n = p n + | u∗,n+1 |2 So (n · u∗,n+1 ) − ν(∇ · u∗,n+1 )n − fbn+1 (10.13)
ν 2
This condition can be enforced using the USERDEFINEDTYPE “HOutflow”, i.e.

2 <REGION REF="0">
3 <N VAR="u" USERDEFINEDTYPE="HOutflow" VALUE="0" />
4 <N VAR="v" USERDEFINEDTYPE="HOutflow" VALUE="0" />
5 <D VAR="p" USERDEFINEDTYPE="HOutflow" VALUE="0" />
6 </REGION>
Note that in the moving body work of Bao et al. [4] some care must be made to identify
when the flow over the boundary is incoming or outgoing and so a modification of the
term
1
| u∗,n+1 |2 So (n · u∗,n+1 )
2
is replaced with
1
(θ + α2 ) | u∗,n+1 |2 +(1 − θ + α1 )(u∗,n+1 · n)u∗,n+1 So (n · u∗,n+1 )
2
where the default values are given by θ = 1, α1 = 0, α2 = 0 and these values can be set
through the parameters OutflowBC_theta , OutflowBC_alpha1 and OutflowBC_alpha2 .
Dong has also suggested convective like outflow conditions in [10] which can be enforced
through a Robin type specification of the form
∂un+1 γ0 D0 n+1 1h D0
+ u = f n+1 + E(n, u∗,n+1 ) + pn+1 n − ν(∇ · u∗,n+1 )n + û (10.14)
∂n ∆t ν ∆t
10.1 Synopsis 139
∂pn+1 1 n+1
+ p = − −ν(∇ × ∇ × u)∗,n+1 + N∗,n+1 · n
∂n νD0
1 h n+1
+ E(n, u∗,n+1 + pn+1 n − ν(∇ · u∗,n+1 )n

− f (10.15)
νD0
2 <REGION REF="0">
3 <R VAR="u" USERDEFINEDTYPE="HOutflow" VALUE="0" PRIMCOEFF="D0/TimeStep"/>
4 <R VAR="v" USERDEFINEDTYPE="HOutflow" VALUE="0" PRIMCOEFF="D0/TimeStep"/>
5 <R VAR="p" USERDEFINEDTYPE="HOutflow" VALUE="0" PRIMCOEFF="1.0/(D0*Kinvis)"/>
6 </REGION>
10.1.1.5 Substepping/subcycling the Velocity Correction Scheme

It is possible to use difference time steps in the velocity correction scheme using a
substepping (also known as subcycling) [31] or auxiliary semi-Lagrangian approach [39].
Originally the scheme was proposed by Maday, Patera and Ronquist who referred to as
an operator-integration-factor splitting method [23]
Figure 10.1 Schematic representation of the substepping approach. (a) Making an explicit time
step the hyperbolic solution, travelling with a speed a, can be understood as being related to the
solution at point xd (the departure point). (b) Making smaller explicit time steps we can evaluate
the solution φ(xd ) at the departure point and then use this value to make a semi-Lagrangian
discretisation of the implicit components usually associated with diffusion.
A schematic of the approach can be understood from figure 10.1.1.5 where we observe
that smaller time steps can be used for the explicit advection steps whilst a larger overall
time step is adopted for the more expensive implicit solve for the diffusion operator. More
details of the implementation can be found in [39] and [31]. In the following sections
we outline the parameters that can be used to set up this scheme. Since the explicit
part is advanced using a DG scheme it is necessary to use a Mixed_CG_Discontinuous
expansion with this option.
Note
Example of the substepping scheme can be found in the regression
tests directory under $NEKHOME/Solver/IncNavierStokesSolver/Tests/ di-
rectory. For exmaple the test cases KovaFlow_SubStep_2order.xml,
Hex_Kovasnay_SubStep.xml and Tet_Kovasnay_SubStep.xml
10.1.1.6 Approximation spaces for the velocity correction scheme

For well resolved simulations it appears that often using the same polynomial space for
the pressure and velocity does give suitable answer but this does not satisfy the so-called
LBB or inf-sup condition. Therefore it is potentially better to specify an equivalent of
the Taylor Hood approximation and use one higher polynomial order for velocity than
the pressure with a continuous expansion. To specify this type of expansion you can use
an expansion section of the form:
1 <EXPANSIONS>
2 <E COMPOSITE="C[0]" NUMMODES="8" FIELDS="u,v" TYPE="MODIFIED" />
3 <E COMPOSITE="C[0]" NUMMODES="7" FIELDS="p" TYPE="MODIFIEDQUADPLUS1" />
4 </EXPANSIONS>
In the above example the “u,v” fields are specified to have a polynomial order of 8 using
a modified expansion. Implicitly this form of the expansion definition uses a quadrature
order of 9. The above definition then also uses a modified expansion for pressure but of
order 7. Since currently for this solver to run we need to use a consistent quadrature
order for both the velocity and pressure fields we specify the MODIFIEDQUADPLUS1 to tell
the solver to use an addition quadrature point and therefore also use 9 quadrature points
in each 1D direction for the pressure.
In other cases it is sometimes useful to run with an even higher quadrature order, for
example to handle highly deformed elements where the Jacobian is represented by a
polynomial expansion. This can be done by using a more detailed definition of the
expansion of the form:
1 <EXPANSIONS>
2 <E COMPOSITE="C[0]" BASISTYPE="Modified_A,Modified_B" NUMMODES="8,8" POINTSTYPE="GaussLobattoLegen
3 <E COMPOSITE="C[0]" BASISTYPE="Modified_A,Modified_B" NUMMODES="7,7" POINTSTYPE="GaussLobattoLegen
4 </EXPANSIONS>
In this example we have specified an 8th order expansion for “u,v” and a 7th order
expansion for “p”. The BasisType is given as “Modified_A, Modified_B” which is
for a triangular expansion (note a quadrilateral expansion would have been “Modi-
fied_A,Modified_A”) and so the number of quadrature points in this case is 9 in the first
direction which uses Gauss Lobatto Legendre points but only 8 in the second direction
since this uses a Gauss Radau formula with α = 1, β = 0 weights (see ?? for details on
why).
10.1 Synopsis 141
10.1.2 Direct solver (coupled approach)

The second approach consists of directly solving the matrix problem arising from the
discretization of the Stokes problem. The direct solution of the Stokes system introduces
the problem of appropriate spaces for the velocity and the pressure systems to satisfy
the inf-sup condition and it requires the solution of the full velocity-pressure system.
However, if a discontinuous pressure space is used then all but the constant mode of the
pressure system can be decoupled from the velocity. When implementing this approach
with a spectral/hp element discretization, the remaining velocity system may then also
be statically condensed to decouple the so called interior elemental degrees of freedom,
reducing the Stokes problem to a smaller system expressed on the elemental boundaries.
The direct solution of the Stokes problem provides a very natural setting for the solution
of the pressure system which is not easily dealt with in a splitting scheme. Further, the
solution of the full coupled velocity system allows the introduction of a spatially varying
viscosity, which arise for non-Newtonian flows, with only minor modifications.
We consider the weak form of the Stokes problem for the velocity field boldsymbolu =
[u, v]T and the pressure field p:
(∇φ, ν∇u) − (∇ · φ, p) = (φ, f ) (10.16a)

(q, ∇ · u) = 0 (10.16b)
where the components of A,B and C are ∇φb , ν∇ub , ∇φb , ν∇ui and ∇φi , ν∇ui and the
components Db and Di are q, ∇ub and q, ∇ui . The indices b and i refer to the degrees of
freedom on the elemental boundary and interior respectively. In constructing the system
we have lumped the contributions form each component of the velocity field into matrices
A,B and C. However, we note that for a Newtonian fluid the contribution from each field
is decoupled. Since the interior degrees of freedom of the velocity field do not overlap,
the matrix C is block diagonal and to take advantage of this structure we can statically
condense out the C matrix to obtain the system:
    
A − BC −1 B T DbT − BC −1 Di 0 ub fb − BC −1 fi
 Db − DiT C −1 B T −DiT C −1 Di 0   p  =  −DiT C −1 fi  (10.17)
     
BT Di C ui fi
To extend the above Stokes solver to an unsteady Navier-Stokes solver we first introduce
the unsteady term, ∂u/∂t, into the Stokes problem. This has the principal effect
of modifying the weak Laplacian operator ∇φ, ν∇u] into a weak Helmholtz operator
∇φ, ν∇u) − λ(φ, u where λ depends on the time integration scheme. The second
modification requires the explicit discretisation of the non-linear terms in a similar
manner to the splitting scheme and this term is then introduced as the forcing term f .
For more details see [1, 32].
10.1.3 Linear Stability Analysis
Hydrodynamic stability is an important part of fluid-mechanics that has a relevant role

in understanding how an unstable flow can evolve into a turbulent state of motion with
chaotic three-dimensional vorticity fields and a broad spectrum of small temporal and
spatial scales. The essential problems of hydrodynamic stability were recognised and
formulated in 19th century, notably by Helmholtz, Kelvin, Rayleigh and Reynolds.
Conventional linear stability assumes a normal representation of the perturbation fields

that can be represented as independent wave packets, meaning that the system is self-
adjoint. The main aim of the global stability analysis is to evaluate the amplitude of the
eigenmodes as time grows and tends to infinity. However, in most industrial applications,
it is also interesting to study the behaviour at intermediate states that might affects
significantly the functionality and performance of a device. The study of the transient
evolution of the perturbations is seen to be strictly related to the non-normality of the
linearised Navier-Stokes equations, therefore the normality assumptiong of the system is
no longer assumed. The eigenmodes of a non-normal system do not evolve independently
and their interaction is responsible for a non-negligible transient growth of the energy.
Conventional stability analysis generally does not capture this behaviour, therefore other
techniques should be used.
A popular approach to study the hydrodynamic stability of flows consists in performing

a direct numerical simulation of the linearised Navier-Stokes equations using iterative
methods for computing the solution of the associated eigenproblem. However, since
linearly stable flows could show a transient increment of energy, it is necessary to extend
this analysis considering the combined effect of the direct and adjoint evolution operators.
This phenomenon has noteworthy importance in several engineering applications and it
is known as transient growth.
In Nektar++ it is then possible to use the following tools to perform stability analysis:
• direct stability analysis;
• adjoint stability analysis;
• transient growth analysis;
10.1.3.1 Direct stability analysis
The equations that describe the evolution of an infinitesimal disturbance in the flow
can be derived decomposing the solution into a basic state (U, p) and a perturbed state
U + εu0 with ε 1 that both satisfy the Navier-Stokes equations. Substituting into
the Navier Stokes equations and considering that the quadratic terms u0 · ∇u0 can be
10.1 Synopsis 143
neglected, we obtain the linearised Navier-Stokes equations:
∂u0
+ U · ∇u0 + u0 · ∇U = −∇p + ν∇2 u0 + f (10.18a)
∂t
∇ · u0 (10.18b)
The linearised Navier-Stokes equations are identical in form to the non-linear equation,
except for the non-linear advection term. Therefore the numerical techniques used for
solving Navier-Stokes equations can still be applied as long as the non-linear term is
substituted with the linearised one. It is possible to define the linear operator that
evolved the perturbation forward in time:
u0 (x, t) = A(U)u0 (x, 0) (10.19)
Let us assume that the base flow U is steady, then the perturbations are autonomous
and we can assume that:
u0 (x, t) = q0 (x) exp(λt) where λ = σ + iω (10.20)
Then we obtain the associated eigenproblem:
A(U)q0 = λq0 (10.21)
The dominant eigenvalue determines the behaviour of the flow. If it the real part is
positive there exists exponentially growing solutions, conversely if every single eigenvalues
has negative real part then the flow is linearly stable. If the real part of the eigenvalue is
zero, it is present a bifurcation point.
10.1.3.2 Adjoint Stability Analysis

The adjoint of a linear operator is one of the most important concept in functional
analysis and has an it has important role to tackle the transition to turbulence. Let us
write the linearised Navier-Stokes equation in a compact form:
!
1
−∂t − (U · ∇) + (∇U) · + Re ∇2 −∇
Hq = 0 where H = (10.22)
∇· 0
The adjoint operator (H∗ is defines as:

hHq, qi = hq, H∗ q∗ i (10.23)
Integrating by parts and employing the divergence theorem, it is possible to express the
adjoint equations:
∂u∗ 1 2
− + (U · ∇)u∗ + (∇U)T · u∗ = −∇p∗ + ∇ u (10.24a)
∂t Re
∇ · u∗ = 0 (10.24b)
The adjoint fields are in fact related to the concept of receptivity. The value of the
adjoint velocity at a point in the flow indicates the response that arises from an unsteady
momentum source at that point. The adjoint pressure and the adjoint stream function
play instead the same role for mass and vorticity sources respectively. Therefore, the
adjoint modes can be seen as a powerful tool to understand where to act in order to
ease/inhibit the transition.
10.1.3.3 Transient Growth Analysis

Transient growth is a phenomenon that occurs when a flow that is linearly stable, but
whose perturbations exhibit a non-negligible transient response due to regions of localised
convective instabilities. This situation is common in many engineering applications, for
example in open flows where the geometry is complex, producing a steep variation of the
base flow. Therefore, the main question to answer is if it exists a bounded solution that
exhibit large growth before inevitably decaying. Let us introduce a norm to quantify the
size of a perturbation. It is physically meaningful to use the total kinetic energy of a
perturbation on the domain Ω. This is convenient because it is directly associated with
the standard-L2 inner product:
A(τ )v = σu, kuk = 1 (10.25)
where σ = ku0 (τ )k. This is no other that the singular value decomposition of A(τ ). The
phenomenology of the transient growth can be explained considering the non-normality of
the linearised Navier-Stokes evolution operator. This can be simply understood using the
simple geometric example showed in figure 10.1.3.3. Let us assume a unit-length vector
f represented in a non-orthogonal basis .This vector is defined as the difference of the
nearly collinear vectors Φ1 and Φ2 . With the time progression, the component of these
two vectors decrease respectively by 20% and 50%. The vector f increases substantially
in length before decaying to zero. Thus, the superposition of decaying non-orthogonal
eigenmode can produce in short term a growth in the norm of the perturbations.
10.1 Synopsis 145
Figure 10.2 Geometric interpretation of the transient growth. Adapted from Schmid, 2007
10.1.4 Steady-state solver using Selective Frequency Damping

To compute linear stability analysis, the choice of the base flow, around which the
system will be linearised, is crucial. If one wants to use the steady-state solution of the
Navier-Stokes equations as base flow, a steady-state solver is implemented in Nektar++.
The method used is the encapsulated formulation of the Selective Frequency Damping
method [18]. Unstable steady base flows can be obtained with this method. The SFD
method is based on the filtering and control of unstable temporal frequencies within the
flow. The time continuous formulation of the SFD method is
(
q̇ = N S(q) − χ(q − q̄),
(10.26)
q̄˙ = q−q̄
∆ .
where q represents the problem unknown(s), the dot represents the time derivative, N S
represents the Navier-Stokes equations, χ ∈ R+ is the control coefficient, q̄ is a filtered
version of q, and ∆ ∈ R∗+ is the filter width of a first-order low-pass time filter. The
steady-state solution is reached when q = q̄.
The convergence of the method towards a steady-state solution depends on the choice of
the parameters χ and ∆. They have to be carefully chosen: if they are too small, the
instabilities within the flow can not be damped; but if they are too large, the method
may converge extremely slowly. If the dominant eigenvalue of the flow studied is known
(and given as input), the algorithm implemented can automatically select parameters
that ensure a fast convergence of the SFD method. Most of the time, the dominant
eigenvalue is not know, that is why an adaptive algorithm that adapts χ and ∆ all along
the solver execution is also implemented.
Note that this method can not be applied for flows with a pure exponential growth of
the instabilities (e.g. jet flow within a pipe). In other words, if the frequency of the
dominant eigenvalue is zero, then the SFD method is not a suitable tool to obtain a
steady-state solution.
10.2 Usage
IncNavierStokesSolver session.xml
In the following the possible options are shown for the incompressible Navier-Stokes.
The Expansion section for an incompressible flow simulation can be set as for other
solvers regardless of the projection type. Here an example for a 3D simulation (for 2D
simulations the specified fields would be just u,v,p ).
1 <EXPANSIONS>
2 <E COMPOSITE="C[0]" NUMMODES="6" FIELDS="u,v,w,p" TYPE="MODIFIED" />
3 </EXPANSIONS>
In case of a simulation using the Direct Solver we need to set FIELDS=u,v as the pressure
expansion order will be automatically set to fulfil the inf-sup condition. Possible choices
for the expansion TYPE are:
Basis TYPE
Modal MODIFIED
Nodal GLL_LAGRANGE
Nodal SEM GLL_LAGRANGE_SEM
10.3.1 Solver Info

The following parameters can be specified in the SOLVERINFO section of the session file:
• EqType : sets the kind of equations we want to solve on the domain as:
Possible values are:
Equations EQTYPE Dim. Projections Alg.
Steady Stokes (SS) SteadyStokes All CG VCS

Steady Oseen (SO) SteadyOseen All CG DS
Unsteady Stokes (US) UnsteadyStokes All CG VCS
Steady Linearised NS (SLNS) SteadyLinearisedNS All CG DS
Unsteady Linearised NS (ULNS) UnsteadyLinearisedNS All CG VCS,DS
Unsteady NS (UNS) UnsteadyNavierStokes All CG,CG-DG VCS
• SolverType : sets the scheme we want to use to solve the set of equations as
Algorithm SolverType Dimensions Projections

Velocity Correction Scheme (VCS) VelocityCorrectionScheme 2D, Quasi-3D, 3D CG, CG-DG
VCS with weak pressure VCSWeakPressure 2D, Quasi-3D, 3D CG, CG-DG
Direct solver CoupledLinearisedNS 2D, Quasi-3D, 3D CG
• Driver : this specifies the type of problem to be solved:
Driver Description Dimensions Projections

Standard Time integration of the equations All CG, DG
SteadyState Steady-state solver (see Sec. 10.1.4) All CG, DG
• Projection : sets the Galerkin projection type as

Galerkin Projection Projection Dimensions Equations Algorithms

Continuous (CG) Continuous All All All
Discontinuous (DG) DisContinuous All ... ...
Mixed CG and DG (CG-DG) Mixed_CG_Discontinuous 2D,3D just UNS VCS-substepping
• TimeIntegrationMethod : sets the time integration method as

Possible values are
Time-Integration Method TimeIntegrationMethod Dimensions Equations Projections

IMEX Order 1 IMEXOrder1 all US, UNS CG
Backward Euler BackwardEuler all US, UNS CG-DG
BDF Order 1 BDFImplicitOrder1 all US, UNS CG-DG
BDF Order 2 BDFImplicitOrder2 all US, UNS CG-DG
• Extrapolation : Specify the extrapolation method (standard or substetping) to

be used in velocity correction scheme. Essentially this activates the sub-stepping
routine which requires the mixed CG-DG projection
1 
Possible values are SubStepping or Standard with “Standard” being the default
value if nothing is specifiied.
• SubStepIntScheme : choose the substep DG time integration scheme so that a

different order schems can be used as compared to the overal time integraiton
scheme.
1 
Possible values are
Time-Integration Method SubStepIntScheme

Forward Euler ForwardEuler
RK 2 RungeKutta2_ImprovedEuler
This option is useful if you wish to use an overall scheme that is first order accurate
for example with TimeIntegrationMethod as BDFImplicitOrder1 but using a second
order RungeKutta2_ImprovedEuler for greater stability in the substep.
• GlobalSysSoln : sets the approach we use to solve the the linear systems of the
type Ax = b appearing in the solution steps, such as the Poisson equation for the
pressure in the splitting-scheme. It can be set as
1 
Possible values are
System solution GlobalSysSoln Parallel

Direct Solver (DS) DirectFull just quasi-3D
DS with Static Condensation DirectStaticCond just Quasi-3D
DS with Multilevel Static Condensation DirectMultiLevelStaticCond just Quasi-3D
Iterative Solver (IS) IterativeFull just Quasi-3D
IS with Static Condensation IterativeStaticCond quasi-3D
IS with Multilevel Static Condensation IterativeMultiLevelStaticCond quasi-3D
Default values are DirectMultiLevelStaticCond in serial and IterativeStaticCond

in parallel.
• SmoothAdvection : activates a stabilization technique which smooths the advection

term using the pressure inverse mass matrix. It can be used just in combination
with nodal expansion basis for efficiency reasons.
1 
• SpectralVanishingViscosity : activates a stabilization technique which increases

the viscosity on the modes with the highest frequencies.
1 
In a Quasi-3D simulation, this will affect both the Fourier and the spectral/hp expan-
sions. To activate them independently, use SpectralVanishingViscositySpectralHP
and SpectralVanishingViscosityHomo1D .
• DEALIASING : activates the 3/2 padding rule on the advection term of a Quasi-3D
simulation.
1 
• SPECTRALHPDEALIASING : activates the spectral/hp dealiasing to stabilize the simu-

lation. This method is based on the work of Kirby and Sherwin [7].
1 
10.3.2 Parameters
• TimeStep : sets the time-step for the integration in time formula
• NumSteps : sets the number of time-steps
• Kinvis : sets the cinematic viscosity coefficient formula
• SubStepCFL : sets the CFL safety limit for the sub-stepping algorithm (default
value = 0.5)
• MinSubSteps : perform a minimum number of substeps in sub-stepping algorithm

(default is 1)
• MaxSubSteps : perform a maxmimum number of substeps in sub-stepping algorithm

otherwise exit (default is 100)
• SVVCutoffRatio : sets the ratio of Fourier frequency not affected by the SVV
technique (default value = 0.75, i.e. the first 75% of frequency are not damped)
• SVVDiffCoeff : sets the SVV diffusion coefficient (default value = 0.1)
10.3.3 Womersley Boundary Condition

It is possible to define the time-dependent Womersley velocity profile for pulsatile flow in
a pipe. The modulation of the velocity profile is based on the desired peak or centerline
velocity which can be represented by a Fourier expansion Umax = A(ωn )eiωn t where A
are the Fourier modes and ω the frequency. The womersely solution is then defined as:
N
J0 (i3/2 αn r/R) iωn t
Añ [1 −
X
w(r, t) = A0 (1 − (r/R)2 ) + ]e
n=1
J0 (i3/2 α)
where the womersley number α is defined:

r
2πn
αn = R
Tν
and Añ (n = 1 : N )are the Fourier coefficients scaled in the following way:
1
Añ = 2An /[1 − ]
J0 (i3/2 α)
The Womersley velocity profile is implemented in the following way:

1 <REGION REF="0">
2 <D VAR="u" USERDEFINEDTYPE="Womersley:WomParams.xml" VALUE="0" />
3 <D VAR="v" USERDEFINEDTYPE="Womersley:WomParams.xml" VALUE="0" />
4 <D VAR="w" USERDEFINEDTYPE="Womersley:WomParams.xml" VALUE="0" />
6 </REGION>
A file containing the Fourier coefficients, Ã, must be in the directory where the solver
is called from. The name of the file is defined by the string given in the attribute
USERDEFINEDTYPE after the “:” and contains the real and imaginary coefficients. This
file has the format
1 <NEKTAR>
2 <WOMERSLEYBC>
3 <WOMPARAMS>
4 <W PROPERTY="Radius" VALUE="0.5" />
5 <W PROPERTY="Period" VALUE="1.0" />
6 <W PROPERTY="axisnormal" VALUE="0.0,0.0,1.0" />
7 <W PROPERTY="axispoint" VALUE="0.0,0.0,0.0" />
8 </WOMPARAMS>
9
10 <FOURIERCOEFFS>
11 <F ID="0"> 0.600393641193, 0.0 </F>
12 <F ID="1"> -0.277707172935, 0.0767582715413 </F>
13 <F ID="2"> -0.0229953131146, 0.0760936232478 </F>
14 <F ID="3"> 0.00858135174058, 0.017089888642 </F>
15 <F ID="4"> 0.0140332527651, 0.0171575122496 </F>
16 <F ID="5"> 0.0156970122129, -0.00547357750345 </F>
17 <F ID="6"> 0.00473626554238, -0.00498786519876 </F>
18 <F ID="7"> 0.00204434981523, -0.00614566561937 </F>
19 <F ID="8"> -0.000274697215201, 0.000153571881197 </F>
20 <F ID="9"> -0.000148037910774, 2.68919619581e-05 </F>
21 </FOURIERCOEFFS>
22 </WOMERSLEYBC>
23 </NEKTAR>
Each value of Ã is provided in the FOURIERCOEFFS section and provided as separate

entries containing the real and imaginary components, i.e. the mean component provided
above is 0.600393641193, 0.0.
Similarly in the WOMPARAMS section the key parameters of the boundary condition are
also provided as:
• RADIUS is the radius of the boundary.
• PERIOD is the cycle time period,
• AXISNORMAL defines the normal direction to the boundary,
• AXISPOINT defines a coordinate in the boundary centre,
10.3.4 Forcing
10.3.4.1 MovingBody
Note
This force type is only supported for the Quasi-3D incompressible Navier-Stokes
solver.
This force type allows the user to solve the interaction system of an incompressible fluid
flowing past a flexible moving bodies [27]. By this forcing function, one can eliminate
the difficulty of moving mesh by using body-fitted coordinates, so that an additional
acceleration term(i.e., forcing term) is introduced to the momentum equations by the
non-inertial transform from the deformed and moving coordinate system to non-deformed
and stationary one.
1 <FORCE TYPE="MovingBody">
2 </FORCE>
Available options of the motion type for the moving body include free, constrained and
forced vibrations, which can be specified in the SOLVERINFO section. The free type of
motion allows the body to move in both streamwise and crossflow directions, while the
constrained type limits the motion only in the crossflow direction. For the forced type,
the vibration profiles of the body should be specified as a given function or read from
input file in MovingBody section. here an example,
1 <SOLVERINFO>
4 
6 

10 
11 </SOLVERINFO>
A moving body type boundary condition should be specified in BOUNDARYCONDITIONS for

the velocities on the moving body,
2 <REGION REF="0">
3 <D VAR="u" USERDEFINEDTYPE="MovingBody" VALUE="0" />
4 <D VAR="v" USERDEFINEDTYPE="MovingBody" VALUE="0" />
5 <D VAR="w" VALUE="0" />
7 </REGION>
For the simulation of low mass ratio, there is an option to activate fictitious mass method
for stabilizing explicit coupling between the fluid solver and structural dynamic solver.
Here we need to specify the values of fictitious mass and damping in PARAMETERS , for
example,
1 <SOLVERINFO>
2 
3 </SOLVERINFO>
4 <PARAMETERS>
5 FictDamp = 1000 
6 FictMass = 1.5 
7 </PARAMETERS>
A filter calles as MovingBody is encapsulated in this module to evaluate the aerodynamic

forces along the moving body surface. The forces for each computational plane are
projected along the Cartesian axes and the pressure and viscous contributions are
computed in each direction.

OutputFile 7 session Prefix of the output filename to which the
forces are written.
Frequency 7 1 Number of timesteps after which output is
written.
Boundary 3 - Boundary surfaces on which the forces are
to be evaluated.
To enable the filter, add the following to the FORCE tag::

10.4 Stability analysis Session file configuration 153
1 <FORCE TYPE="MovingBody">
2 <PARAM NAME="OutputFile">DragLift.frc</PARAM>
4 <PARAM NAME="Boundary"> B[0] </PARAM>
5 </FORCE>
During the execution a file named DragLift.frc will be created and the value of the
aerodynamic forces on boundary 0, defined in the GEOMETRY section, will be output every
10 time steps.evaluates the aerodynamic forces along the moving body surface. The forces
for each computational plane are projected along the Cartesian axes and the pressure
and viscous contributions are computed in each direction.
Also, to use this module a MAPPING needs to be specified, as described in section 10.6.
In the case of free and constrained motion presented here, the functions defined by the
mapping act as initial conditions. Also, when using the MovingBody forcing, it is not
necessary to set the TIMEDEPENDENT property of the mapping.
10.4 Stability analysis Session file configuration
The type of equation which is to be solved is specified through the EqType option in the
session file. This can be set to any of the following:
Equation to solve
∂u0
∂t
+ L(U, u0 ) = −∇p + ν∇2 u0
Equation Type Dimensions Projections Algorithms

UnsteadyNavierStokes 2D, Quasi-3D Continuous VCS,DS
10.4.1 Solver Info

• Eqtype : sets the type of equation to solve, according to the table above.
• TimeIntegrationMethod : the following types of time integration methods have

been tested with each solver:
Explicit Diagonally Implicit IMEX Implicit

UnsteadyNavierStokes X X
• Projection : the Galerkin projection used may be either
– Continuous : for a C0-continuous Galerkin (CG) projection;

– Discontinuous : for a discontinous Galerkin (DG) projection.
• EvolutionOperator :
– Nonlinear (non-linear Navier-Stokes equations).

– Direct (linearised Navier-Stokes equations).
– Adjoint (adjoint Navier-Stokes equations).
– TransientGrowth ((transient growth evolution operator).
• Driver : specifies the type of problem to be solved:
– Standard (time integration of the equations)

– ModifiedArnoldi (computations of the leading eigenvalues and eigenmodes
using modified Arnoldi method)
– Arpack (computations of eigenvalues/eigenmodes using Implicitly Restarted
Arnoldi Method (ARPACK) ).
• ArpackProblemType : types of eigenvalues to be computed (for Driver Arpack only)
– LargestMag (eigenvalues with largest magnitude).

– SmallestMag (eigenvalues with smallest magnitude).
– LargestReal (eigenvalues with largest real part).
– SmallestReal (eigenvalues with smallest real part).
– LargestImag (eigenvalues with largest imaginary part).
– SmallestIma (eigenvalues with smallest imaginary part ).
• Homogeneous : specifies the Fourier expansion in a third direction (optional)
– 1D (Fourier spectral method in z-direction).
• ModeType : this specifies the type of the quasi-3D problem to be solved.
– MultipleMode (stability analysis with multiple modes).

– SingleMode (BiGlobal Stability Analysis: full-complex mode).
– HalfMode (BiGlobal Stability Analysis: half-complex mode u.Re v.Re w.Im
p.Re).
10.4.2 Parameters
• Re : sets the Reynolds number
• Kinvis : sets the kinematic viscosity ν.
• kdim : sets the dimension of the Krylov subspace κ. Can be used in: ModifiedArnoldi
and Arpack . Default value: 16.
10.5 Steady-state solver Session file configuration 155
• evtol : sets the tolerance of the eigenvalues. Can be used in:
• imagShift : provide an imaginary shift to the direct sovler eigenvalue problem by

the specified value. lttModifiedArnoldi and Arpack . Default value: 10−6 .
• nits : sets the maximum number of iterations. Can be used in: ModifiedArnoldi
and Arpack . Default value: 500.
• LZ : sets the length in the spanswise direction Lz . Can be used in Homogeneous

set to 1D . Default value: 1.
• HomModesZ : sets the number of planes in the homogeneous directions. Can be used
in Homogeneous set to 1D and ModeType set to MultipleModes .
• N_slices : sets the number of temporal slices for Floquet stability analysis.
• period : sets the periodicity of the base flow.
• realShift : provide a real shift to the direct sovler eigenvalue problem by the
specified value.
10.4.3 Functions
When using the direct solver for stability analysis it is necessary to specify a Forcing
function “StabilityCoupledLNS” in the form:
1 <FORCING>
2 <FORCE TYPE="StabilityCoupledLNS">
3 </FORCE>
4 </FORCING>
This is required since we need to tell the solver to use the existing field as a forcing
function to the direct matrix inverse as part of the Arnoldi iteration.
Note
Examples of the set up of the direct solver stability analysis (and other incom-
pressible Navier-Stokes solvers) can be found in the regression test directory
NEKTAR/solvers/IncNavierStokesSolver/Tests . See for example the files
PPF_R15000_ModifiedArnoldi_Shift.tst and PPF_R15000_3D.xml noting
that some parameters are specified in the .tst files.
10.5 Steady-state solver Session file configuration
In this section, we detail how to use the steady-state solver (that implements the selective
frequency damping method, see Sec. 10.1.4). Two cases are detailed here: the execution
of the classical SFD method and the adaptive SFD method, where the control coefficient
χ and the filter width ∆ of the SFD method are updated all along the solver execution.
For the second case, the parameters of the SFD method do not need to be defined by the
user (they will be automatically calculated all along the solver execution) but several
session files must be defined in a very specific way.
10.5.1 Execution of the classical steady-state solver

10.5.1.1 Solver Info
The definition of Eqtype , TimeIntegrationMethod and Projection is similar as what is
explained in 10.4.1. The use of the steady-state solver is enforced through the definition
of the Driver which has to be SteadyState . EvolutionOperator does not need to be
defined to run the unadapted SFD method (by default, it is set to Nonlinear ).
10.5.1.2 Parameters
• Kinvis : sets the kinematic viscosity ν. It is typically 1/Re if both the characteristic
velocity and characteristic length are chosen to be 1.
• ControlCoeff : sets the control coefficient χ of the SFD method. Default value: 1.
• FilterWidth : sets the filter width ∆ of the SFD method. Default value: 2.
• GrowthRateEV and FrequencyEV : if the growth rate and the frequency of the
dominant eigenvalue are known, they can be given given as input and the code will
automatically select the optimum parameters χ and ∆ (and overwrite the values of
ControlCoeff and GrowthRateEV that may be given in the session file)
• TOL : sets the tolerance of the SFD method. The code will run until ||q − q̄||inf <
T OL. Default value: 10−8 .
Note that for the steady-state solver, the parameter NumSteps is not taken into account.
The solver will run until a steady-state solution is found and not for a pre-defined number
of time steps.
10.5.2 Execution of the adaptive steady-state solver

Running the adaptive selective frequency damping method requires to set up the session
files in a very specific manner. First, the Geometry section must be in a separated
archive file. If the test case studied is called "Session", the mesh file must be called
Session.xml.gz (the linux command "gzip" can be used to obtain this file).
The requirements for the file Session.xml are similar as for the ones for the classical
SFD method. The Geometry section being removed and placed in Session.xml.gz .
10.6 Coordinate transformations Session file configuration 157
This file defines the properties of the nonlinear problem solved (i.e. the flow for which
we want a steady-state). Also, the SOLVERINFO section must contain the line:
1 
The adaptive SFD method used is coupled with a stability analysis method. Then kdim ,
nvec , evtol and nits should be defined into the PARAMETERS section of Session.xml .
If not, these parameters will take the default values presented in Sec. 10.4.
The goal of running the stability analysis is to evaluate the dominant eigenvalue of a
“partially converged” steady base flow. This approximation is then used by the steady-
state solver to select a control coefficient χ and a filter width ∆ then ensure a fast
convergence towards a steady-state solution.
To define the linear stability problem, another file, that must be called Session_LinNS.xml ,
has to be defined. This file must be an exact copy/paste of Session.xml , only
three things have to be modified:
1. The boundary conditions must be modified to be homogeneous (i.e. equal to zero)

at all inflow boundaries.
2. A non-zero function InitialConditions has to be defined.
3. A random function BaseFlow has to be defined (it will be overwritten all along
the solver execution). We recommend it to be a copy of InitialConditions .
Once these three files (the Geometry in Session.xml.gz , the nonlinear problem defini-
tion in Session.xml and the homogeneous linear problem in Session_LinNS.xml ) are
correctly defined, the adaptive SFD method must be executed using:
IncNavierStokesSolver Session.xml.gz Session.xml
10.6 Coordinate transformations Session file configuration
This section describes how to include a coordinate transformation to the solution of the
incompressible Navier-Stokes equations. In some cases, this approach allows a slightly
deformed geometry to be mapped into a geometry with a homogeneous direction, which
can be treated using a quasi-3D method. It is also useful for problems with a moving
body, where otherwise a moving mesh would have to be employed.
10.6.1 Solver Info

To activate the mapping technique, SolverType needs to be set as:
1 
Also, there are other optional properties in the SolverInfo section:

1 
2 
3
4 
the first two options determine if the pressure and viscous terms resulting from the
coordinate transformation are treated implicitly using an iterative procedure. If the last
option is set to true, the viscous terms in the mapping are not computed. This leads to
a faster solution, but the effect on the results need to be determined for the specific case.
By default, all mapping terms are computed and treated explicitly.
10.6.2 Parameters
When treating the mapping terms implicitly, the following parameters can be set:
1 MappingPressureTolerance = 1e-8 
2 MappingViscousTolerance = 1e-8 
3 MappingPressureRelaxation = 0.9 
4 MappingViscousRelaxation = 0.9 
They determine the tolerance of the iterative solution of the equations, and a relaxation
parameter which can improve the numerical stability of the method.
10.6.3 Mapping
The particular transformation employed is specified by:
1 <MAPPING TYPE="XYofZ">
2 <COORDS>Mapping</COORDS>
3 <VEL>MappingVel</VEL>
4 <TIMEDEPENDENT>True</TIMEDEPENDENT> 
5 </MAPPING>
where TIMEDEPENDENT indicates if the transformation varies with time.
The available values for TYPE , and the transformations they represent, are:
Mapping type x̄ ȳ z̄
Translation x + f (t) y + g(t) z + h(t)
XofZ x + f (z, t) y z
XofXZ f (x, z, t) y z
XYofZ x + f (z, t) y + g(z, t) z
XYofXY f (x, y, t) g(x, y, t) z
General f (x, y, z, t) g(x, y, z, t) h(x, y, z, t)
10.6 Coordinate transformations Session file configuration 159
where (x̄, ȳ, z̄) are the Cartesian (physical) coordinates and (x, y, z) are the transformed
coordinates. Note that for quasi-3D problems, the z coordinate cannot be transformed.
10.6.4 Functions
The function COORDS (and VEL for time dependent mappings) indicated in the MAPPING
section need to be defined, for example as:
1 <FUNCTION NAME="Mapping">
2 <E VAR="x" VALUE="x + cos(PI*z)" />
3 <E VAR="y" VALUE="y + cos(2*PI*t)" />
4 </FUNCTION>
5
6 <FUNCTION NAME="MappingVel">
7 <E VAR="vx" VALUE="0.0" />
8 <E VAR="vy" VALUE="-1.0*2*PI*sin(2*PI*t)" />
9 </FUNCTION>
the transformation defined by these functions need to be valid (non-zero Jacobian). By

default, any component of COORDS that is not specified is taken as a trivial transformation,
e.g. x̄ = x, and any velocity not specified is considered to be zero.
10.6.5 Boundary conditions

In case of a time-dependent mapping, a moving body boundary condition is available:
2 <REGION REF="0">
3 <D VAR="u" USERDEFINEDTYPE="MovingBody" VALUE="0" />
4 <D VAR="v" USERDEFINEDTYPE="MovingBody" VALUE="0" />
5 <D VAR="w" VALUE="0" />
7 </REGION>
When using the MovingBody boundary condition, the Dirichlet condition is relative to
the boundary, while the regular Dirichlet boundary condition is taken in an absolute
sense.
All Dirichlet boundary conditions are specified in the Cartesian (physical) space, and are
automatically transformed to the computational frame of reference.
Note
Examples of the use of mappings can be found in the test direc-
tory NEKTAR/solvers/IncNavierStokesSolver/Tests . See for exam-
ple the files KovaFlow_3DH1D_P8_16modes_Mapping-implicit.xml and
CylFlow_Mov_mapping.xml .
10.7 Adaptive polynomial order Session file configuration
An adaptive polynomial order procedure is available for 2D and Quasi-3D simulations.

This procedure consists of the following steps:
• Advance the equations for a determined number of time steps

• Use the sensor defined in equation 9.8 to estimate an error measure (the variable
used in the sensor can be specified). The error is defined here as the square of the
sensor.
• Use the error to determine if the order in each element should be increased by one,
decreased by one, or left unaltered.
• Project the solution in each element to the new polynomial order and use it as
an initial condition to restart the equation, repeating all steps a given number of
times.
It is important to note that restarting the simulation after the refinement can be an
expensive operation (in a typical case 200 times the cost of a single time step). Therefore,
the number of steps between successive refinements needs to be carefully chosen, since
if this value is too low the procedure becomes inefficient, while if it is too high the
refinement might not capture accurately structures that are convected by the flow.
10.7.1 Solver Info

The use of the adaptive polynomial order procedure is enforced through the definition of
the Driver which has to be Adaptive .
10.7.2 Parameters
• NumSteps : when using the adaptive order procedure, this parameter determines
the number of time steps between successive refinements.
• NumRuns : this parameter defines the number of times the sequence of solving the
equation and refining is performed. Therefore, the total number of time steps in
the simulation is N umSteps × N umRuns.
• AdaptiveMaxModes : sets the maximum number of modes (in each direction) that
can be used in an element during the adaptive procedure. The solution will not be
refined beyond this point, even if the error is higher than the tolerance. Default
value: 12.
• AdaptiveMinModes : sets the minimal number of modes (in each direction) that can
be used in an element during the adaptive procedure. Default value: 4.
10.8 Advecting extra passive scalar fields 161
• AdaptiveUpperTolerance : defines a constant tolerance. The polynomial order in

an element is increased whenever the error is higher than this value. This can be
replaced by a spatially-varying function, as described below. Default value: 10−6 .
• AdaptiveLowerinModes : defines a constant tolerance. The polynomial order in an

element is decreased whenever the error is lower than this value. This can also be
replaced by a spatially-varying function. Default value: 10−8 .
• AdaptiveSensorVariable : integer defining which variable will be considered when

calculating the error. For example, if this parameter is set to 1 in the Incompressible
Navier-Stokes Solver, the error will be estimated using the v velocity. Default value:
0.
10.7.3 Functions
Spatially varying tolerances can be specified by defining the functions AdaptiveLowerinModes
and/or AdaptiveUpperTolerance . In this case, the tolerance in an element is taken as
the average of the function in the quadrature points of the element. If these functions
are defined, the values set for the tolerances in the PARAMETERS section are ignored.
10.7.4 Restarting the simulation

The simulation can be restarted using the final distribution of polynomial orders obtained
from the adaptive procedure by setting the expansions as
1 <EXPANSIONS>
2 <F FILE="restartfile.fld" />
3 </EXPANSIONS>
note that this will only affect the polynomial order. The initial condition still needs to
be set correctly, and does not need to come from the same file used for the expansions.
10.8 Advecting extra passive scalar fields
In some cases, it might be useful to advect passive scalar fields with the velocity obtained
from the solution of the Navier-Stokes equation. For example, in study of mass transfer
or heat transfer problems where getting analytical expression for advection velocity is
not possible, the transport (advection-diffusion) equation needs to be solved along with
the Navier-Stokes equation to get the scalar concentration or temperature distribution in
the flow field.
In the input file, the extra field variables that are being advected need to be defined
after the variables representing the velocity components. The pressure needs to be at the
end of the list. For example, for a 2D simulation the expansions and variables would be
defined as
1 <EXPANSIONS>
2 <E COMPOSITE="C[0]" NUMMODES="5" FIELDS="u,v,c1,c2,p" TYPE="MODIFIED" />
3 </EXPANSIONS>
4
5 <VARIABLES>
6 <V ID="0"> u </V>
7 <V ID="1"> v </V>
8 <V ID="2"> c1 </V>
9 <V ID="3"> c2 </V>
10 <V ID="4"> p </V>
11 </VARIABLES>
where u, v are the velocity components, c1 and c2 are extra fields that are being advected
and p is the pressure.
In addition, diffusion coefficients for each extra variable can be specified by adding a
function DiffusionCoefficient
1 <FUNCTION NAME="DiffusionCoefficient">
2 <E VAR="c1" VALUE="0.1" />
3 <E VAR="c2" VALUE="0.01" />
4 </FUNCTION>
Boundary conditions for the extra fields are set up in the same way as the velocity and
pressure
2 <REGION REF="0">
5 <D VAR="c1" VALUE="1" />
6 <D VAR="c2" VALUE="0" />
8 </REGION>
It should be noted that if the diffusion coefficient is too small, the transport equation
becomes advection dominated. This leads to small grid spacing required to resolve all
physical scales associated with the transport equation (the ratio of resolution required for
transport to Navier Stokes equation scales with Sc3/4 , where Sc is the Schmidt number
= kinematic viscosity/diffusion coefficient). Hence, small diffusion coefficient might lead
to spurious oscillations if the mesh spacing is not small enough.
10.9 Examples
10.9.1 Kovasznay Flow 2D
This example demonstrates the use of the velocity correction o solve the 2D Kovasznay
flow at Reynolds number Re = 40. In the following we will numerically solve for the two
dimensional velocity and pressure fields with steady boundary conditions.
10.9 Examples 163
10.9.1.1 Input file

The input for this example is given in the example file KovaFlow_m8.xml . The mesh
consists of 12 quadrilateral elements.
We will use a 7th-order polynomial expansions (N = 8 modes) using the modified

Legendre basis and therefore require the following expansion definition.
1 <EXPANSIONS>
3 </EXPANSIONS>
We next specify the solver information for our problem. In particular, we select the
velocity correction scheme formulation, using a continuous Galerkin projection. For this
scheme, an implicit-explicit ime-integration scheme must be used and we choose one of
second order.
1 <SOLVERINFO>
4 
7 </SOLVERINFO>
The key parameters are listed below. Since the problem is unsteady we prescribe the
time step and the total number of time steps. We also know the required Reynolds
number, but we must prescribe the kinematic viscosity to the solver. We first define a
dummy parameter for the Reynolds number, and then define the kinematic viscosity as
the inverse of this. The value of λ is used when defining the boundary conditions and
exact solution. Note that PI is a pre-defined constant.
1 <PARAMETERS>
2 TimeStep = 0.001 
4 Re = 40 
5 Kinvis = 1/Re 
6 LAMBDA = 0.5*Re-sqrt(0.25*Re*Re+4*PI*PI)
7 </PARAMETERS>
We choose to impose a mixture of boundary condition types as defined below.

2 <REGION REF="0">
3 <D VAR="u" VALUE="1-exp(LAMBDA*x)*cos(2*PI*y)" />
4 <D VAR="v" VALUE="(LAMBDA/2/PI)*exp(LAMBDA*x)*sin(2*PI*y)" />
6 </REGION>
7 <REGION REF="1">
8 <D VAR="u" VALUE="1-exp(LAMBDA*x)*cos(2*PI*y)" />
9 <D VAR="v" VALUE="(LAMBDA/2/PI)*exp(LAMBDA*x)*sin(2*PI*y)" />
10 <D VAR="p" VALUE="0.5*(1-exp(2*LAMBDA*x))" />
11 </REGION>
12 <REGION REF="2">
13 <N VAR="u" VALUE="0" />
14 <D VAR="v" VALUE="0" />
15 <N VAR="p" VALUE="0" />
16 </REGION>
Initial conditions are obtained from the file KovaFlow_m8.rst, which is a Nektar++ field
file. This is the output of an earlier simulation, renamed with the extension rst to
avoid being overwritten, and is used in this case to reduce the integration time necessary
to obtain the steady flow.
2 <F FILE="KovaFlow_m8.rst" />
3 </FUNCTION>
Note the use of the F tag to indicate the use of a file. In contrast, the exact solution is
prescribed using analytic expressions which requires the use of the E tag.
2 <E VAR="u" VALUE="1-exp(LAMBDA*x)*cos(2*PI*y)" />
3 <E VAR="v" VALUE="(LAMBDA/2/PI)*exp(LAMBDA*x)*sin(2*PI*y)" />
4 <E VAR="p" VALUE="0.5*(1-exp(2*LAMBDA*x))" />
5 </FUNCTION>
10.9.1.2 Running the simulation

Launch the simulation using the following command
IncNavierStokesSolver KovaFlow_m8.xml
After completing the prescribed 100 time-steps, the difference between the computed
solution and the exact solution will be displayed. The actual mantissas may vary slightly,
but the overall magnitude should be as shown.
L 2 error (variable u) : 3.75296e-07

L inf error (variable u) : 5.13518e-07
L 2 error (variable v) : 1.68897e-06
L inf error (variable v) : 2.23918e-06
L 2 error (variable p) : 1.46078e-05
L inf error (variable p) : 5.18682e-05
The output of the simulation is written to KovaFlow_m8.fld . This can be visualised

by converting it to a visualisation format. For example, to use ParaView, convert the
output into VTK format using the tility.
FieldConvert KovaFlow.xml KovaFlow.fld KovaFlow.vtu

10.9 Examples 165
The result should look similar to that shown in Figure 10.3.
Figure 10.3 Velocity profiles for the Kovasznay Flow (2D).
10.9.2 Kovasznay Flow 2D using high-order outflow boundary conditions

In this example, we solve the same case of 2D Kovasznay flow on severely-truncated
computational domain but using a high-order outflow boundary condition, which is
much more accurate and robust for unbounded flows [11]. The solver information and
parameters used here are similar to the previous one. What only we need to modify in
the input file is just the boundary condition type upon the outlet region shown as below
2 <REGION REF="0">
3 <D VAR="u" VALUE="1-exp(KovLam*x)*cos(2*PI*y)" />
4 <D VAR="v" VALUE="(KovLam/2/PI)*exp(KovLam*x)*sin(2*PI*y)" />
6 </REGION>
7 <REGION REF="1">
8 <N VAR="u" USERDEFINEDTYPE="HOutflow"
9 VALUE="-Kinvis*KovLam*exp(KovLam*x)*cos(2*PI*y)
10 - 0.5*(1-exp(2*KovLam*x))-0.5*(((1-exp(KovLam*x)*cos(2*PI*y))
11 *(1-exp(KovLam*x)*cos(2*PI*y))+(KovLam/(2*PI)*exp(KovLam*x)
12 *sin(2*PI*y))*(KovLam/(2*PI)*exp(KovLam*x)*sin(2*PI*y))))
13 *(0.5*(1.0-tanh((1-exp(KovLam*x)*cos(2*PI*y))*20)))" />
14 <N VAR="v" USERDEFINEDTYPE="HOutflow"
15 VALUE="Kinvis*KovLam*KovLam/(2*PI)*exp(KovLam*x)*sin(2*PI*y)" />
16 <D VAR="p" USERDEFINEDTYPE="HOutflow"
17 VALUE="-Kinvis*KovLam*exp(KovLam*x)*cos(2*PI*y)
18 - 0.5*(1-exp(2*KovLam*x)) -0.5*(((1-exp(KovLam*x)*cos(2*PI*y))
19 *(1-exp(KovLam*x)*cos(2*PI*y))+(KovLam/(2*PI)*exp(KovLam*x)
20 *sin(2*PI*y))*(KovLam/(2*PI)*exp(KovLam*x)*sin(2*PI*y))))
21 *(0.5*(1.0-tanh((1-exp(KovLam*x)*cos(2*PI*y))*20)))" />
22 </REGION>
23 <REGION REF="2">
24 <N VAR="u" VALUE="0" />

25 <D VAR="v" VALUE="0" />
26 <N VAR="p" VALUE="0" />
27 </REGION>
We note that in this example the “VALUE” property is set based on the analytic solution
but this is not typically known and so often a VALUE of zero will be specified.
Instead of loading an initial condition from a specified file, we initialized the flow fields
in this example by using following expressions
2 <E VAR="u" VALUE="(1-exp(KovLam*x)*cos(2*PI*y))" />
3 <E VAR="v" VALUE="(KovLam/(2*PI))*exp(KovLam*x)*sin(2*PI*y)" />
4 <E VAR="p" VALUE="0.5*(1-exp(2*KovLam*x))" />
5 </FUNCTION>

We then launch the simulation by the same solver as that in the previous example
IncNavierStokesSolver KovaFlow_m8_short_HOBC.xml
The solution with errors displayed as below

The physical solution visualized in velocity profiles is also illustrated in Figure 10.4.
Figure 10.4 Velocity profiles for the Kovasznay Flow in truncated domain (2D).
10.9 Examples 167
10.9.3 Steady Kovasznay Oseen Flow using the direct solver

In this example, we instead compute the steadhy Kovasznay Oseen flow using the direct
solver. In contrast to the velocity correction scheme in which we time-step the solution
to the final time, the direct solver computes the solution with a single solve.
10.9.3.1 Input file

We can begin with the same input file as for the previous example, but with the
following modifications. For reference, the modified version is provided in the example
Oseen_m8.xml .
In the solver information, we must instead select the Steady-Oseen equation type and
choose to use the coupled linearised Navier-Stokes
1 
2 
Note
Since we are using a coupled system, we are not solving for the pressure.
We should therefore remove all references to the variable p in the ses-
sion. In particular, it should be removed from the EXPANSIONS , VARIABLES ,
BOUNDARYCONDITIONS and FUNCTIONS sections of the file.
Instead of loading an initial condition from file, we can simply prescribe a zero field.
4 </FUNCTION>
We must also provide an advection velocity.

2 <E VAR="u" VALUE="(1-exp(-LAMBDA*x)*cos(2*PI*y))" />
3 <E VAR="v" VALUE="(-LAMBDA/(2*PI))*exp(-LAMBDA*x)*sin(2*PI*y)" />
4 </FUNCTION>

Run the simulation using
IncNavierStokesSolver Oseen_m8.xml
The resulting flow field should match the solution from the previous example.
10.9.4 Laminar Channel Flow 2D

In this example, we will simulate the flow through a channel at Reynolds number 1 with
fixed boundary conditions.
10.9.4.1 Input file

The input file for this example is given in ChanFlow_m3_SKS.xml . The geometry is a
square channel with height and length D = 1, discretised using four quadrilateral elements.
We use a quadratic expansion order, which is sufficient to capture the quadratic flow
profile. In this example, we choose to use the skew-symmetric form of the advection term.
This is chosen in the solver information section:
1 
A first-order time integration scheme is used and we set the time-step and number of time
integration steps in the parameters section. We also prescribe the kinematic viscosity
ν = 1/Re = 1.
Boundary conditions are defined on the walls (region 0) and at the inflow (regions 1) as
Dirichlet for the velocity field and as high-order for the pressure. At the outflow the
velocity is left free using Neumann boundary conditions and the pressure is pinned to
zero.
2 <REGION REF="0">
6 </REGION>
7 <REGION REF="1">
8 <D VAR="u" VALUE="y*(1-y)" />
11 </REGION>
12 <REGION REF="2">
13 <N VAR="u" VALUE="0" />
14 <N VAR="v" VALUE="0" />
15 <D VAR="p" VALUE="0" />
16 </REGION>
Initial conditions are set to zero. The exact solution is a parabolic profile with a pressure
gradient dependent on the Reynolds number. This is defined to allow verification of the
calculation.
2 <E VAR="u" VALUE="y*(1-y)" />
4 <E VAR="p" VALUE="-2*Kinvis*(x-1)" />
5 </FUNCTION>
10.9 Examples 169
10.9.4.2 Running the solver
IncNaverStokesSolver ChanFlow_m3_SKS.xml
The error in the solution should be displayed and be close to machine precision

The solution should look similar to that shown in Figure 10.5.
Figure 10.5 Pressure and velocity profiles for the laminar channel flow (2D).
10.9.5 Laminar Channel Flow 3D
We now solve the incompressible Navier-Stokes equations on a three-dimensional domain.

In particular, we solver the three-dimensional equivalent of the previous example. We
will also solve the problem in parallel.
Note
In order to run the example, you must have a version of Nektar++ compiled
with MPI. This is the case for the packaged binary distributions.
10.9.5.1 Input file

The input file for this example is given in Tet_channel_m8_par.xml . In this example
we use tetrahedral elements, indicated by the A element tags in the geometry section.
All dimensions have length D = 1. We will use a 7th-order polynomial expansion. Since
we now have three dimensions, and therefore three velocity components, the expansions
section is now
1 <EXPANSIONS>
3 </EXPANSIONS>
The solver information and parameters are similar to the previous example. Boundary
conditions must now be defined on the six faces of the domain. Flow is prescribed in the
z-direction through imposing a Poiseulle profile on the inlet and side walls. The outlet is
zero-Neumann and top and bottom faces impose zero-Dirichlet conditions.
1 <BOUNDARYREGIONS>
2 C[1] 
3 C[6] 
4 C[2] 
5 C[3] 
6 C[4] 
7 C[5] 
9
11 <REGION REF="0">
12 <D VAR="u" VALUE="0" />
13 <D VAR="v" VALUE="0" />
14 <D VAR="w" VALUE="y*(1-y)" />
16 </REGION>
17 <REGION REF="1">
18 <N VAR="u" VALUE="0" />
19 <N VAR="v" VALUE="0" />
20 <N VAR="w" VALUE="0" />
21 <D VAR="p" VALUE="0" />
22 </REGION>
23 <REGION REF="2">
24 <D VAR="u" VALUE="0" />
25 <D VAR="v" VALUE="0" />
26 <D VAR="w" VALUE="0" />
28 </REGION>
29 <REGION REF="3">
30 <D VAR="u" VALUE="0" />
31 <D VAR="v" VALUE="0" />
32 <D VAR="w" VALUE="y*(1-y)" />
34 </REGION>
35 <REGION REF="4">
36 <D VAR="u" VALUE="0" />
37 <D VAR="v" VALUE="0" />
10.9 Examples 171
38 <D VAR="w" VALUE="0" />

40 </REGION>
41 <REGION REF="5">
42 <D VAR="u" VALUE="0" />
43 <D VAR="v" VALUE="0" />
44 <D VAR="w" VALUE="y*(1-y)" />
46 </REGION>
Initial conditions and exact solutions are also prescribed.

To run the solver in parallel, we use the mpirun command.
mpirun -np 2 IncNaverStokesSolver Tet_channel_m8_par.xml
The expected results are shown in Figure 10.6.
Figure 10.6 Pressure and velocity profiles for the laminar channel flow (full 3D).
10.9.6 Laminar Channel Flow Quasi-3D

For domains where at least one direction is geometrically homogeneous, a more efficient
discretisation is to use a pure spectral discretisation, such as a Fourier expansion, in
these directions. We use this approach to solve the same problem as in the previous
example. We reuse the two-dimensional spectral/hp element mesh from the nd couple
this with a Fourier expansion in the third component.
10.9.6.1 Input file

The input file for this example is ChanFlow_3DH1D_MVM.xml . We indicate that we
are coupling the spectral/hp element domain with a pure spectral expansion using the
following solver information
We must also specify parameters to describe the particular spectral expansion

2 LZ = 1.0 
The parameter HomModesZ specifies the number of Fourier modes to use in the homoge-
neous direction. The LZ parameter specifies the physical length of the domain in that
direction.
Note
This example uses an in-built Fourier transform routine. Alternatively, one
can use the FFTW library to perform Fourier transforms which typically offers
improved performance. This is enabled using the following solver information
As with the the spectral/hp element mesh consists of four quadrilateral elements with a
second-order polynomial expansion. Since our domain is three-dimensional we have to
now include the third velocity component
The remaining parameters and solver information is similar to previous examples.
Boundary conditions are specified as for the two-dimensional case (except with the
addition of the third velocity component) since the side walls are now implicitly periodic.
The initial conditions and exact solution are prescribed as for the fully three-dimensional
case.

IncNaverStokesSolver ChanFlow_3DH1D_MVM.xml
The results can be post-processed and should match those of the fully three-dimensional
case as shown in Figure 10.6.
10.9.7 Turbulent Channel Flow

In this example we model turbulence in a three-dimensional square channel at a Reynolds
number of 2000.
10.9 Examples 173
Note
This example requires the FFTW Fast-Fourier transform library to be selected
when compiling Nektar++.
10.9.7.1 Input file

The input file for this example is TurbChFl_3DH1D.xml . The geometry makes used of
the homogeneous extension discussed in the previous example. The channel has height
D = 2 and length L = 4π and is discretised using quadratic quadrilateral elements in
the spectral/hp element plane and a Fourier basis in the third coordinate direction. The
elements are non-uniformly distributed so as to best capture the flow features with fewest
degrees of freedom and is shown in Figure 10.7.
Figure 10.7 Mesh used for the turbulent channel flow.
The spanwise length of the channel is set using the LZ parameter and discretised with
32 Fourier modes by setting the value of HomModesZ .
2 LZ = 4*PI/3 
A second-order IMEX scheme is used for time-integration scheme is used with a time-step
of 0.0001. The length of the simulation is 1 time-unit (10,000 steps).
Periodicity is naturally enforced in the spanwise direction, so boundary conditions need

only be provided for the upper and lower walls, inlet and outlet as denoted by the
following BOUNDARYREGIONS .
1 <BOUNDARYREGIONS>
2 C[1] //walls
3 C[2] //inflow
4 C[3] //outflow
In this example, we will use a body force to drive the flow and so, in addition to the
spanwise periodicity, enforce periodicity in the streamwise direction of the spectral/hp
element mesh. This is achieved by imposing the following boundary conditions
1 <REGION REF="1">
2 
3 

4 
5 
6 </REGION>
7 <REGION REF="2">
8 
9 
10 
11 
12 </REGION>
Here, we use P to denote the boundary type is periodic, and the value in square brackets
denotes the boundary region to which the given boundary is periodic with. In this case
regions 1 and 2 are denoted periodic with each other.
A streamwise plug-profile initial condition is prescribed such that u = 1 everywhere,

except the wall boundaries. The body force requires two components in the XML file.
The first specifies the type of forcing to apply and appears directly within the NEKTAR
tag.
1 <FORCING>
2 <FORCE TYPE="Body">
3 <BODYFORCE> BodyForce </BODYFORCE>
4 </FORCE>
5 </FORCING>
The second defines the BodyForce function which will be used and is located within the
CONDITIONS section,
1 <FUNCTION NAME="BodyForce">
2 <E VAR="u" VALUE="2*Kinvis" />
4 <E VAR="w" VALUE="0" />
5 </FUNCTION>
To improve numerical stability, we also enable dealising of the advection term. This uses
additional points to perform the quadrature and then truncates the higher-order terms
when projecting back onto the polynomial space, thereby removing spurious oscillations.
It is enabled by setting the solver information tag
1 
This feature is only available when using the FFTW library is used, so we enable this
using
1 

To run the solver, we use the following command
10.9 Examples 175
IncNaverStokesSolver TurbChFl_3DH1D.xml
The result after transition has occurred is illustrated in Figure 10.8.
Figure 10.8 Velocity profile of the turbulent channel flow (quasi-3D).
10.9.8 Turbulent Pipe Flow

In this example we simulate flow in a pipe at Reynolds number 3000 using a mixed
spectral/hp element and Fourier discretisation. The Fourier expansion is used in the
streamwise direction in this case and the spectral/hp elements are used to capture the
circular cross-section.
10.9.8.1 Input File

The circular pipe has diameter D = 1, the 2D mesh is composed of 64 quadrilateral ele-
ments and the streamwise direction is discretised with 128 Fourier modes. An illustrative
diagram of the discretisation is given in Figure 10.9.
The input file for this example is Pipe_turb.xml . We use 7th-order lagrange polynomials
through the Gauss-Lobatto-Legendre points for the quadrilateral expansions.
1 <E COMPOSITE="C[0]" NUMMODES="8" FIELDS="u,v,w,p" TYPE="GLL_LAGRANGE_SEM" />
We set the Fourier options, as in the previous example, except using 128 modes and a
length of 5 non-dimensional units. A small amplitude noise is also added to the initial
condition, which is a plug profile, to help stimulate transition. Since the streamwise
direction is the Fourier direction, we must necessarily use a body force to drive the flow.
Figure 10.9 Domain for the turbulent pipe flow problem.

In this example we will run the solver in parallel. Due to the large number of Fourier
modes and relatively few elements, it is more efficient to parallelise in the streamwise
direction. We can specify this by providing an additional flag to the solver, –npz . This
indicates the number of partitions in the z-coordinate. In this example, we will only run
two processes. We therefore would specify –npz 2 to ensure parallelisation only occurs
in the Fourier direction.
To improve the efficiency of the solver further, we would prefer to solve the Helmholtz
problems within the spectral/hp element planes using a direct solver (since no paralleli-
sation is necessary). The default when running in parallel is to use an iterative solver, so
we explicitly specify the type of algorithm to use in the session file solver information:
1 
The solver can now be run as follows
mpirun -np 2 IncNavierStokesSolver --npz 2 Pipe_turb.xml
When the pipe transitions, the result should look similar to Figure 10.10.
10.9.9 Aortic Blood Flow

The following example demonstrates the application of the incompressible Navier-Stokes
solver using the Velocity Correction Scheme algorithm for modelling viscid Newtonian
blood flow in a region of a rabbit descending thoracic aorta with intercostal branch pairs.
Such studies are necessary to understand the effect local blood flow changes have on
cardiovascular diseases such as atherosclerosis.
10.9 Examples 177
Figure 10.10 Velocity profile of the turbulent pipe flow (quasi-3D).
In the following we will numerically solve for the three dimensional velocity and pressure
field for steady boundary conditions. The Reynolds number under consideration is 300,
which is physiologically relevant.
Geometry
The geometry under consideration is a segment of a rabbit descending aorta with two
pairs of intercostal arteries branching off. The inlet has a diameter D = 3.32mm.
Figure 10.11 Reduced region of rabbit descending thoracic aorta.
In order to capture the physics of the flow in the boundary layer, a thin layer consisting
of prismatic elements is created adjacent to the surface, and curved using spherigons.
The interior consist of tetrahedral elements.
Input parameters
Figure 10.12 Surface mesh indicating curved surface elements at a branch location.
10.9.9.1 Expansion:
In this example we will use a fourth order polynomial expansion. There are two composites
defined here since we have both prismatic and tetrahedral elements.
1 <EXPANSIONS>
2 <E COMPOSITE="C[0]" NUMMODES="5" TYPE="MODIFIED" FIELDS="u,v,w,p" />
4 </EXPANSIONS>
10.9.9.2 Solver information:

1 <SOLVERINFO>
4 
7 </SOLVERINFO>
10.9.9.3 Parameters:
Since we are prescribing a Reynolds number of 300, and to simplify the problem definition,
we set the mean inlet velocity to 1, this allows us to define the kinematic viscosity as
ν = URe
D
= 3.32
300 = 1/90.36.
1 <PARAMETERS>
2 TimeStep = 0.0005 
3 NumSteps = 1600 
6 Kinvis = 1.0/90.36 
7 </PARAMETERS>
10.9 Examples 179
10.9.9.4 Boundary conditions:

For the purpose of this example a blunted inlet velocity profile has been prescribed.
Ideally to obtain more significant results, the velocity profile at the inlet would be
obtained from previous simulations on the complete rabbit aorta (including aortic root,
aortic arch, and descending aorta with all 5 pairs of intercostal arteries), where a blunted
profile at the aortic root is a better representation of reality.
Dirichlet boundary conditions are imposed for the velocity at the inlet, as well as on the
wall to account for the no-slip condition. Neumann boundary conditions are imposed for
the velocity field at the outlets where fully developed flow is imposed.
1 <BOUNDARYREGIONS>
2 C[2] 
3 C[3,4,5,6] 
4 C[7] 
5 C[8] 
7
9 <REGION REF="0">
10 <D VAR="u" VALUE="0.024" />
11 <D VAR="v" VALUE="-0.064" />
12 <D VAR="w" VALUE="-0.998" />
14 </REGION>
15 <REGION REF="1">
16 <N VAR="u" VALUE="0" />
17 <N VAR="v" VALUE="0" />
18 <N VAR="w" VALUE="0" />
19 <D VAR="p" VALUE="0" />
20 </REGION>
21 <REGION REF="2">
22 <N VAR="u" VALUE="0" />
23 <N VAR="v" VALUE="0" />
24 <N VAR="w" VALUE="0" />
25 <D VAR="p" VALUE="0" />
26 </REGION>
27 <REGION REF="3">
28 <D VAR="u" VALUE="0" />
29 <D VAR="v" VALUE="0" />
30 <D VAR="w" VALUE="0" />
32 </REGION>
10.9.9.5 Functions:
4 <E VAR="w" VALUE="0" />
5 <E VAR="p" VALUE="0" />
6 </FUNCTION>
10.9.9.6 Results
We can visualise the internal velocity field by applying a volume render filter in ParaView.
Figure 10.13 The solved-for velocity field.
It is possible to visualise the wall shear stress distribution by running the FldAddWSS
utility.
Figure 10.14 Non-dimensional wall shear stress distribution.
10.9.10 finite-strip modeling of flow past flexible cables

As a computationally efficient model, strip theory-based modeling technique has been
proposed previously to predict vortex-induced vibration (VIV) for higher Reynolds
number flows. In the strip theory-based model, the fluid flow solution is obtained on a
series of 2D computational planes (also called as “strips”) along the riser’s axis direction.
These strips then are coupled with each other through structural dynamic model of the
riser, and then VIV response prediction is achieved by the strip-structure interactions.In
the 2D strip theory, it is assumed that the flow is purely two-dimensional without spanwise
correlation, which allows the problems to be split into various 2D planes. A consequence
10.9 Examples 181
of 2D strip solution under this assumption is that it is unable to reflect the influence
of spanwise wake turbulence on the structural dynamics. In order to overcome this
shortcoming, we proposed a new module in the framework of Nektar++, in which a
spanwise scale is locally allocated to each one of the strips, so that the spanwise velocity
correlation is reconstructed in the flow field within each strips. In particular, this model
lets the fluid domain to be divided in N strips with thickness ratio of Lz /D and evenly
distributed along the spanwise (z) direction. The gap between the neighboring strips,
represented by Lg , satisfies relation Lc = N (Lz + Lg ). Since the strip in this model has
finite scale in the z-direction, we named it as finite strip to distinguish from traditional 2D
strip plane. Next, the flow dynamics within each individual strips are modeled by viscous
incompressible Navier-Stokes equations, while a tensioned beam model is employed to
govern the dynamics of the flexible structures. In this example, we will show how to
perform a finite-strip model to predict the vortex-induced vibration responses of flexible
cables. Let us consider a vortex-induced vibration of a slender cable with an aspect ratio
of Lz /D=4π, which is immersed in uniform flows at Re=100.
10.9.10.1 Input File

The cable with a mass ratio (defined as the ratio of the total oscillating mass to the mass
of displaced fluid) of 1 has diameter D = 1, the 2D mesh is composed of 284 quadrilateral
elements. The spanwise direction is split in 16 strips with thickness ratio of Lc /D=π/8
and one pair of complex Fourier modes for each one of the strips. We will use a sixth
order polynomial expansion for the spectral element and the input file for this example
is CylFlow_HomoStrip.xml .
To use the finite strip routines we need just to insert a flag of "HomoStrip" in the solver
information as below, in addition, we need to specify the types of vibration and support
ends for the cables. In this case, the vibration type is specified as VALUE="CONSTRAINED" ,
which means that the cable’s vibration is constrained only in the crossflow direction.
Other options include VALUE="FREE" and "FORCED" , respectively corresponding to the
free vibrations in both streamwise and crossflow directions and forced vibration by
specified functions given in input file. For the support ends of the cable, another option
of VALUE="PINNED-PINNED" is available for the simulations, which satisfies the condition
of zero values of displacements on the support ends.
10.9.10.2 Solver information:

1 <SOLVERINFO>
2 
3 
4 
5 </SOLVERINFO>
10.9.10.3 Parameters
All the simulation parameters are specified in the section as follows.
1 <PARAMETERS>
2 LZ = PI/8 
3 LC = 4*PI 
4 A = 0.025 
5 omega = 1.0 
6 PROC_Z = 16 
7 Strip_Z = 16 
8 DistStrip = PI/4 
9 StructStiff = 0.02 
10 StructRho = 2.0 
11 CableTension = 8.82 
12 BendingStiff = 0.0 
13 FictDamp = 0.0 
14 FictMass = 3.0 
15 </PARAMETERS>
16
In this example we will run the solver in parallel. We can specify the number of the
strips by providing an additional flag to the solver, –nsz. In this example, we will run 16
strips, therefore it would be specified as –nsz 16. The solver can now be run as follows
mpirun -np 16 IncNavierStokesSolver CylFlow_HomoStrip.xml --npz 16 --nsz 16
The simulation results are illustrated in spanwise vorticity contours in Figure 10.15. The
wake response of the cable appears as standing wave patter in the earlier stage and then
it transitions into traveling wave response, as shown in this figure.
Figure 10.15 Spanwise vorticity contours in standing wave and traveling wave patterns predicted
in finite strip modeling.
10.9 Examples 183
10.9.11 2D direct stability analysis of the channel flow

In this example, it will be illustrated how to perform a direct stability analysis using
Nektar++. Let us consider a canonical stability problem, the flow in a channel which
is confined in the wall-normal direction between two infinite plates (Poiseuille flow) at
Reynolds number 7500. This problem is a particular case of the stability solver for the
IncNavierStokesSolver.
10.9.11.1 Background
We consider the linearised Incompressible Navier-Stokes equations:
∂u0
+ U · ∇u0 + u0 · ∇U = −∇p + ν∇2 u0 + f (10.27a)
∂t
∇ · u0 = 0 (10.27b)
We are interested to compute the leading eigenvalue of the system using the Arnoldi
method.
10.9.11.2 Geometry
The geometry under consideration is a 2D channel.
10.9.11.3 Mesh Definition

In the GEOMETRY section, the dimensions of the problem are defined. Then, the coordinates
(XSCALE, YSCALE, ZSCALE) of each vertices of each element are specified. As this input
file defines a two-dimensional problem: ZSCALE = 0.
2 <VERTEX>
3 <V ID="0">3.142e+00 1.000e+00 0.000e+00</V>
4 ...
5 <V ID="62">-3.142e+00 -1.000e+00 0.000e+00</V>
6 </VERTEX>
Edges can now be defined by two vertices.

1 <EDGE>
2 <E ID="0"> 0 1 </E>
3 ...
4 <E ID="109"> 62 55 </E>
5 </EDGE>
In the ELEMENT section, the tag T and Q define respectively triangular and quadrilateral
element. Triangular elements are defined by a sequence of three edges and quadrilateral
elements by a sequence of four edges.
1 <ELEMENT>
2 <Q ID="0"> 0 1 2 3 </Q>
3 ...
4 <Q ID="47"> 107 108 109 95 </Q>
5 </ELEMENT>
6
Finally, collections of elements are listed in the COMPOSITE section and the DOMAIN section
specifies that the mesh is composed by all the triangular and quadrilateral elements. The
other composites will be used to enforce boundary conditions.
1 <COMPOSITE>
2 <C ID="0"> Q[0-47] </C>
3 <C ID="1"> E[17,31,44,57,70,83,96,109,0,19,32,45,58,71,84,97] </C> //wall
4 <C ID="2"> E[3,6,9,12,15,18] </C>//inflow
5 <C ID="3"> E[98,100,102,104,106,108] </C> //outflow
6 </COMPOSITE>
7 <DOMAIN> C[0] </DOMAIN>
8 </GEOMETRY>
9
10.9.11.4 Expansion
This section defines the polynomial expansions used on each composites. For this example
we will use a 10th order polynomial, i.e. P = 11.
1 <EXPANSIONS>
3 </EXPANSIONS>
4

In this example the EvolutionOperator must be Direct to consider the linearised
Navier-Stokes equations and the Driver was set up to ModifiedArnoldi for the solution
of the eigenproblem.
1 <SOLVERINFO>
4 
6 
8 </SOLVERINFO>
9
All the stability parameters are specified in this section.
10.9 Examples 185
1 <PARAMETERS>
2 TimeStep = 0.002 
6 Re = 7500 
7 Kinvis =1./Re 
8 kdim =16 
9 nvec =2 
10 evtol =1e-5
11 nits =5000 
12 </PARAMETERS>
13
10.9.11.7 Boundary Conditions

1 <BOUNDARYREGIONS>
2 C[1] 
3 C[2] 
4 C[3] 
6
8 <REGION REF="0">
10 <D VAR="v" VALUE="0" />
12 </REGION>
13 <REGION REF="1">
14 
15 
16 
17 </REGION>
18 <REGION REF="2">
19 
20 
21 
22 </REGION>
24
10.9.11.8 Function
We need to set up the base flow that can be specified as a function BaseFlow. In case
the base flow is not analytical, it can be generated by means of the Nonlinear evolution
operator using the same mesh and polynomial expansion. The initial guess is specified in
the InitialConditions functions and can be both analytical or a file. In this example
it is read from a file.
1 <FUNCTION NAME="BaseFlow">
2 <F VAR="u,v,p" FILE="ChanStability.bse" />
3 </FUNCTION>
4
6 <F VAR="u,v,p" FILE="ChanStability.rst" />

7 </FUNCTION>
8
10.9.11.9 Usage
IncNavierStokesSolver ChanStability.xml
10.9.11.10 Results
The stability simulation takes about 250 iterations to converge and the dominant eigen-
values (together with the respective eigenvectors) will be printed. In this case it was
found λ1,2 = 1.000224 × e±0.24984i . Therefore, since the magnitude of the eigenvalue is
larger than 1, the flow is absolutely unstable. It is possible to visualise the eigenvectors
using the post-processing utilities. The figure shows the profile of the two eigenmode
component, which shows the typical Tollmien - Schlichting waves that arise in viscous
boundary layers.
Figure 10.16
10.9.12 2D adjoint stability analysis of the channel flow

In this example, it will be illustrated how to perform an adjoint stability analysis using
Nektar++. Let us consider a canonical stability problem, the flow in a channel which
is confined in the wall-normal direction between two infinite plates (Poiseuille flow) at
Reynolds number 7500
We consider the equations:
∂u∗ 1 2
− + (U · ∇)u∗ + (∇U)T · u∗ = −∇p∗ + ∇ u (10.28a)
∂t Re
10.9 Examples 187
Figure 10.17
∇ · u∗ = 0 (10.28b)
We are interested in computing the leading eigenvalue of the system using the Arnoldi
method.
10.9.12.2 Geometry & Mesh

The geometry and mesh are the same ones used for the direct stability analysis in the
previous example.

This sections defines the problem solved. In this example the EvolutionOperator must
be Adjoint to consider the adjoint Navier-Stokes equations and the Driver was set up
to ModifiedArnoldi for the solution of the eigenproblem.
1 <SOLVERINFO>
4 
8 </SOLVERINFO>
9 \end{subequations}
10
11 \textbf{Parameters}
12
13 \begin{lstlisting}[style=XMLStyle]
14 <PARAMETERS>
15 TimeStep = 0.002 
16 NumSteps = 500 

19 Re = 7500 
20 Kinvis =1./Re 
21 kdim =16 
22 nvec =2 
23 evtol =1e-5
24 nits =5000 
25 </PARAMETERS>
26

1 <BOUNDARYREGIONS>
2 C[1] 
3 C[2] 
4 C[3] 
6
8 <REGION REF="0">
10 <D VAR="v" VALUE="0" />
12 </REGION>
13 <REGION REF="1">
14 
15 
16 
17 </REGION>
18 <REGION REF="2">
19 
20 
21 
22 </REGION>
24
10.9.12.5 Functions
operator using the same mesh and polynomial expansion.
2 <F VAR="u,v,p" FILE="ChanStability.bse" />
3 </FUNCTION>
4
The initial guess is specified in the InitialConditions functions and can be both
analytical or a file. In this example it is read from a file.
2 <F VAR="u,v,p" FILE="ChanStability.rst" />
3 </FUNCTION>
10.9 Examples 189
10.9.12.6 Usage
IncNavierStokesSolver ChanStability_adj.xml
10.9.12.7 Results
The equations will then be evolved backwards in time (consistently with the negative
sign in front of the time derivative) and the leading eigenvalues will be computed after
about 300 iterations. It is interesting to note that their value is the same one computed
for the direct problem, but the eigenmodes present a different shape.
Figure 10.18
Figure 10.19
10.9.13 2D Transient Growth analysis of a flow past a backward-facing step

In this section it will be described how to perform a transient growth stability analysis
using Nektar++. Let us consider a flow past a backward-facing step (figure 10.20). This
is an important case because it allows us to understand the effects of separation caused
by abrupt changes in the geometry and it is a common geometry in several studies of
flow control and turbulence of separated flow.
Figure 10.20
Transient growth analysis allows us to study the presence of convective instabilities that
can arise in stable flows. Despite the fact that these instabilities will decay for a long time
(due to the stability of the flow), they can produce significant increases in the energy of
perturbations. The phenomenon of transient growth is associated with the non-normality
of the linearised Navier-Stokes equations and it consists in computing the perturbation
that leads to the highest energy growth for a fixed time horizon.
10.9.13.2 Input Parameters

In the GEOMETRY section, the dimensions of the problem are defined. Then, the coordinates
(XSCALE, YSCALE, ZSCALE) of each vertices are specified. As this input file defines a two-
dimensional problem: ZSCALE = 0.
2 <VERTEX>
3 <V ID="0">3.000e+00 -1.000e+00 0.000e+00</V>
4 ...
5 <V ID="399">-1.000e+01 0.000e+00 0.000e+00</V>
6 </VERTEX>
7
Edges can now be defined by two vertices.

1 <EDGE>
2 <E ID="0"> 0 1 </E>
3 ...
4 <E ID="828"> 399 394 </E>
5 </EDGE>
6
10.9 Examples 191
In the ELEMENT section, the tag T and Q define respectively triangular and quadrilateral
element. Triangular elements are defined by a sequence of three edges and quadrilateral
elements by a sequence of four edges.
1 <ELEMENT>
2 <T ID="0"> 0 1 2 </T>
3 ...
4 <T ID="209"> 333 314 332 </T>
5 <Q ID="210"> 334 335 336 0 </Q>
6 ...
7 <Q ID="429"> 826 827 828 818 </Q>
8 </ELEMENT>
9
Finally, collections of elements are listed in the COMPOSITE section and the DOMAIN section
specifies that the mesh is composed by all the triangular and quadrilateral elements. The
other composites will be used to enforce boundary conditions.
1 <COMPOSITE>
2 <C ID="0"> T[0-209] </C>
3 <C ID="1"> Q[210-429] </C>
4 <C ID="2"> E[2-3,7,10,16,21,2,...,828] </C>
5 <C ID="3"> E[821,823,825,827] </C>
6 <C ID="4"> E[722,724,726,728] </C>
7 </COMPOSITE>
8
9 <DOMAIN> C[0,1] </DOMAIN>
10 </GEOMETRY>
11
10.9.13.3 Expansion
For this example we will use a 6th order polynomial, i.e. P = 7:
1 <EXPANSIONS>
4 </EXPANSIONS>
5
10.9.13.4 Solver Information

This sections defines the problem solved. In this example the EvolutionOperator must
be TransientGrowth and the Driver was set up to Arpack for the solution of the
eigenproblem.
1 <SOLVERINFO>
3 
6 
7 
8 

9 </SOLVERINFO>
10
1 <PARAMETERS>
2 FinalTime = 0.1 
3 TimeStep = 0.005 
4 NumSteps = FinalTime/TimeStep 
5 IO_CheckSteps = 1/TimeStep 
7 Re = 500 
8 Kinvis = 1.0/Re 
9 kdim = 4 
10 nvec = 1 
11 evtol = 1e-4 
12 </PARAMETERS>
13

1 <BOUNDARYREGIONS>
2 C[2] 
3 C[3] 
4 C[4] 
6
8 <REGION REF="0">
10 <D VAR="v" VALUE="0" />
12 </REGION>
13 <REGION REF="1">
14 <D VAR="u" VALUE="0" />
15 <D VAR="v" VALUE="0" />
17 </REGION>
18 <REGION REF="2">
19 <D VAR="u" VALUE="0" />
20 <D VAR="v" VALUE="0" />
22 </REGION>
24
10.9.13.7 Functions
operator using the same mesh and polynomial expansion.
10.9 Examples 193
2 <F VAR="u,v,p" FILE="bfs_tg-AR.bse" />

3 </FUNCTION>
4
The initial guess is specified in the InitialConditions functions and in this case is read
from a file.
2 <F VAR="u,v,p" FILE="bfs_tg-AR.rst" />
3 </FUNCTION>
4
10.9.13.8 Usage
IncNavierStokesSolver bfs_tg-AR.xml
10.9.13.9 Results
The solution will be evolved forward in time using the operator A, then backward in
time through the adjoint operator A∗ . The leading eigenvalue is λ = 3.236204). This
corresponds to the largest possible transient growth at the time horizon τ = 1. The
leading eigenmode is shown below. This is the optimal initial condition which will lead
to the greatest growth when evolved under the linearised Navier-Stokes equations.
Figure 10.21
It is possible to visualise the transient growth plotting the energy evolution over time
if the system is initially perturbed with the leading eigenvector. This analysis was
performed for a time horizon τ = 60. It can be seen that the energy grows in time
Figure 10.22
reaching its maximum value at x = 24 and then decays, almost disappearing after 100
temporal units.
Figure 10.23
10.9.14 BiGlobal Floquet analysis of a of flow past a cylinder

In this example it will be described how to run a BiGlobal stability analysis for a time-
periodic base flow using Nektar++. Let us consider a flow past a circular cylinder at
10.9 Examples 195
Re = 220 has a 2D time-periodic wake that is unstable to a 3D synchronous "mode A"

instability.
Figure 10.24
The numerical solution of the fully three- dimensional linear eigenvalue problem is often
computationally demanding and may not have significant advantages over performing
a direct numerical simulation. Therefore, some simplifications are required; the most
radical consist in considering that the base flow depends only on one spatial coordinate,
assuming that the other two spatial coordinates are homogenous. While this method
offers a good prediction for the instability of boundary layers, it is not able to predict
the instability of Hagen-Poiseuille flow in a pipe at all Reynolds numbers. Between
a flow that depends upon one and three-spatial directions, it is possible to consider a
steady or time-periodic base flow depending upon two spatial directions and impose three-
dimensional disturbances that are periodic in the the third homogeneous spatial direction.
This approach is known as BiGlobal stability analysis and it represents the extension
of the classic linear stability theory; let us consider a base flow U that is function of
only two spatial coordinates: mathbf U (x, y, t). The perturbation velocity can u0 can be
expressed in a similar form, but with the dependence on the third homogeneous direction
incorporated through the Fourier mode: u0 = û0 (x, y, t)eiβz , where β = 2π/L)and L is
the length in the homogeneous direction.
10.9.14.2 Input parameters

In this example we use a mesh of 500 quadrilateral elements with a 6th order polynomial
expansion. The base flow has been computed using the Nonlinear evolution operator
with appropriate boundary conditions. From its profile, it was possible to determine the
periodicity of the flow sampling the velocity profile over time. In order to reconstruct the
temporal behaviour of the flow, 32 time slices were considered over one period. Using
these data it is possible to set up the stability simulation for a specified β, for example
β = 1.7. Let us note that while the base flow is 2D, the stability simulation that we are
performing is 3D.
10.9.14.3 Expansion
In this example we will use a 6th order polynomial expansion, i.e. P = 7.
1 <EXPANSIONS>
2 <E COMPOSITE="C[0]" NUMMODES="7" TYPE="GLL_LAGRANGE_SEM" FIELDS="u,v,w,p" />
3 </EXPANSIONS>
4
1 <SOLVERINFO>
4 
6 
10 </SOLVERINFO>
11
10.9.14.5 Functions
2 <F VAR="u,v,p" FILE="cyinder_floq" />
3 </FUNCTION>
4
10.9.14.6 Usage
IncNavierStokesSolver session.xml
10.9.14.7 Results
The stability simulation takes about 20 cycles to converge and the leading eigenvalue is
λ = 1.2670 with a growth rate σ = 4.7694e − 02. The figure below shows the profile of
the magnitude of the eigenmode at z = 2.
10.9 Examples 197
Figure 10.25
Chapter 11
Linear elasticity solver
11.1 Synopsis
The LinearElasticSolver is a solver for solving the linear elasticity equations in two and
three dimensions. Whilst this may be suitable for simple solid mechanics problems, its
main purpose is for use for mesh deformation and high-order mesh generation, whereby
the finite element mesh is treated as a solid body, and the deformation is applied at the
boundary in order to curve the interior of the mesh.
Currently the following equation systems are supported:
Value Description
LinearElasticSystem Solves the linear elastic equations.
IterativeElasticSystem A multi-step variant of the elasticity solver,
which breaks a given deformation into multiple
steps, and applies the deformation to a mesh.
11.1.1 The linear elasticity equations

The linear elasticity equations model how a solid body deforms under the application of
a ‘small’ deformation or strain. The formulation starts with the equilibrium of forces
represented by the equation
∇·S+f =0 in Ω (11.1)
where S is the stress tensor and f denotes a spatially-varying force. We further assume
that the stress tensor S incorporates elastic and, optionally, thermal stresses that can
be switched on to assist in mesh deformation applications. We assume these can be
decomposed so that S is written as
S = Se + St ,
198
11.2 Usage 199
where the subscripts e and t denote the elastic and thermal terms respectively. We adopt
the usual linear form of the elastic stress tensor as
Se = λTr(E) I + µE,
where λ and µ are the Lamé constants, E represents the strain tensor, and I is the
identity tensor. For small deformations, the strain tensor E is given as
1
E= ∇u + ∇ut (11.2)
2
where u is the two- or three-dimensional vector of displacements. The boundary conditions
required to close the problem consist of prescribed displacements at the boundary ∂Ω,
i.e.
u = û in ∂Ω. (11.3)
We further express the Lamé constants in terms of the Young’s modulus E and Poisson
ratio ν as
νE E
λ= , µ= .
(1 + ν)(1 − 2ν) 2(1 + ν)
The Poisson ratio, valid in the range ν < 21 , is a measure of the compressibility of the
body, and the Young’s modulus E > 0 is a measure of its stiffness.
11.2 Usage
LinearElasticSolver [arguments] session.xml [another.xml] ...
11.3.1 Solver Info
• EqType Specifies the PDE system to solve, based on the choices in the table above.
• Temperature Specifies the form of the thermal stresses to use. The choices are:
– None : No stresses (default).

– Jacobian : Sets St = βJI, where β is a parameter defined in the parameters
section, J is the elemental Jacobian determinant and I is the identity matrix.
– Metric : A more complex term, based on the eigenvalues of the metric
tensor. This can only be used for simplex elements (triangles and tetrahedra).
Controlled again by the parameter β.
• BCType Specifies the type of boundary condition to apply when the IterativeElasticSystem
is being used.
200 Chapter 11 Linear elasticity solver
– Normal : The boundary conditions are split into NumSteps steps, as defined
by a parameter in the session file (default).
– Repeat : As the geometry is updated, re-evaluate the boundary conditions.
This enables, for example, a cirlce to be rotated continuously.
11.3.2 Parameters
• nu : sets the Poisson ratio ν.

Default value: 0.25.
• E : sets the Young’s modulus E.

Default value: 1.
• beta : sets the thermal stress coefficient β.

Default value: 1.
• NumSteps : sets the number of steps to use in the case that the iterative elastic
system is enabled. Should be greater than 0.
Default value: 0.
11.4 Examples
11.4.1 L-shaped domain

The first example is the classic L-shaped domain, in which an exact solution is known,
which makes it an ideal test case [20]. The domain is the polygon formed from the
vertices
(−1, −1), (0, −2), (2, 0), (0, 2), (−1, −1), (0, 0).
The exact solution for the displacements is known in polar co-ordinates (r, θ) as
rα
ur (r, θ) = [C1 (C2 − α − 1) cos((α − 1)θ) − (α + 1) cos((α + 1)θ)]
2µ
rα
uθ (r, θ) = [(α + 1) sin((α + 1)θ) + C1 (C2 + α − 1) sin((α − 1)θ)]
2µ
where α ≈ 0.544483737 . . . is the solution of the equation α sin(2ω) + sin(2ωα) = 0,
cos((α + 1)ω) λ + 2µ
C1 = − , C2 = 2
cos((α − 1)ω) λ+µ
with λ and µ being the Lamé constants and ω = 3π/4. Boundary conditions are set to
be the exact solution and f = 0. The solution has a singularity at the origin, and so in
order to test convergence h-refinement is required.
11.4 Examples 201
Figure 11.1 Solution of the u displacement field for the L-shaped domain.
A simple example of how the linear elastic solver can be set up can be found in the
Tests/L-shaped.xml session file in the linear elastic solver directory. A more refined
domain with the obtained u solution is shown in figure 11.1. The solver can be run using
the command:
LinearElasticSolver L-domain.xml
The obtained solution L-domain.fld can be applied to the mesh to obtain a deformed
XML file using the deform module in FieldConvert :
FieldConvert -m deform L-domain.xml L-domain.fld L-domain-deformed.xml
11.4.2 Boundary layer deformation

In this example we use the iterative elastic system to apply a large deformation to a
triangular boundary layer mesh of a square mesh Ω = [0, 1]2 . At the bottom edge, we
apply a Dirichlet condition g = 12 sin(πx) that is enforced by splitting it into N substeps,
so that at each step we solve the system with the boundary condition g n (x) = g(x)/N .
The process is depicted in figure 11.2.
The setup is very straightforward. The geometry can be found inside the file Examples/bl-mesh.xml
202 Chapter 11 Linear elasticity solver
Figure 11.2 Figures that show the initial domain (left), after 50 steps (middle) and final
deformation of the domain (right).
and the conditions inside Examples/bl-conditions.xml . The solver can be set up

using the following parameters, with NumSteps denoting N :
1 <SOLVERINFO>
2 
3 </SOLVERINFO>
4
5 <PARAMETERS>
6 nu = 0.3 
7 E = 1.0 
9 </PARAMETERS>
To identify the boundary that we intend to split up into substeps, we must assign the
WALL tag to our boundary regions:
2 <REGION REF="0">
3 <D VAR="u" VALUE="0" USERDEFINEDTYPE="Wall" />
4 <D VAR="v" VALUE="0.5*sin(PI*x)" USERDEFINEDTYPE="Wall" />
5 </REGION>
6 <REGION REF="1">
9 </REGION>
The solver can then be run using the command:
LinearElasticSolver bl-mesh.xml bl-conditions.xml
This will produce a series of meshes bl-mesh-%d.xml , where %d is an integer running

between 0 and 100. If at any point the mesh becomes invalid, that is, a negative Jacobian
is detected, execution will cease.
Chapter 12
Pulse Wave Solver
12.1 Synopsis
1D modelling of the vasculature (arterial network) represents and insightful and efficient
tool for tackling problems encountered in arterial biomechanics as well as other engineering
problems. In particular, 3D modelling of the vasculature is relatively expensive. 1D
modelling provides an alternative in which the modelling assumptions provide a good
balance between physiological accuracy and computational efficiency.To describe the flow
and pressure in this network we consider the conservation of mass and momentum applied
to an impermeable, deformable tube filled with an incompressible fluid, the nonlinear
system of partial differential equations presented in non-conservative form is given by
∂U ∂U
+H =S (12.1)
∂t ∂x
" # " # " #

U=
U
, H=
U A
, S= 0
∂p 1 f
A ρ ∂A U ρ A −s
in which A is the Area (related to pressure), x is the axial coordinate along the vessel,
U (x, t) the axial velocity, P (x, t) is the pressure in the tube, ρ is the density and finally
f the frictional force per unit length. The unknowns in equation 12.1 are u, A and p;
hence we must provide an explicit algebraic relationship to close this system. Typically,
closure is provided by an algebraic relationship between A and p. For a thin elastic tube
this is given by
√
√
p πhE
p = p0 + β A− A0 , β= (12.2)
(1 − ν 2 )A0
203
204 Chapter 12 Pulse Wave Solver
where p0 is the external pressure, A0 is the initial cross-sectional area, E is the Young’s
modulus, h is the vessel wall thickness and ν is the Poisson’s ratio. Other more elaborate
pressure - area relationships are currently being implemented into the framework. Appli-
cation of Riemann’s method of characteristics to equations 12.1 and 12.2 indicates that
velocity and area are propagated through the system by forward and backward travelling
waves. These waves are reflected and within the network by appropriate treatment of
interfaces and boundaries. In the following, we will explain the usage of the blood flow
solver on the basis of a single-artery problem and also on an arterial network consisting
of 55 arteries.
12.2 Usage
PulseWaveSolver session.xml
12.3.1 Pulse Wave Sovler mesh connectivity

Typically 1D arterial networks are made up of a connection of different base units:
segments, bifurcations and merging junctions. The input format in the PulseWaveSolver
means these connections are handle naturally from the mesh topology; hence care must
be taken when designing the 1D domain. The figure below outlines the structure of a
bifurcation, which is a common reoccurring structure in the vasculature.
Figure 12.1 Model of bifurcating artery. The bifurcation is made of three domains and 15
vertices. Vertex V[0] is the inlet and vertices V[10] and V[15] the outlets.
To represent this topology in the xml file we specify the following vertices under the
section VERTEX (the extents are: −100 ≥ x ≤ 100 and −100 ≥ y ≤ 100 )
1 <VERTEX>
2 <V ID="0">-1.000e+02 0.000e+00 0.000e+00</V>
3 <V ID="1">-8.000e+01 0.000e+00 0.000e+00</V>
4 <V ID="2">-6.000e+01 0.000e+00 0.000e+00</V>
5 <V ID="3">-4.000e+01 0.000e+00 0.000e+00</V>
6 <V ID="4">-2.000e+01 0.000e+00 0.000e+00</V>

7 <V ID="5"> 0.000e+00 0.000e+00 0.000e+00</V>
8
9 <V ID="6"> 2.000e+01 2.000e+01 0.000e+00</V>
10 <V ID="7"> 4.000e+01 4.000e+01 0.000e+00</V>
11 <V ID="8"> 6.000e+01 6.000e+01 0.000e+00</V>
12 <V ID="9"> 8.000e+01 8.000e+01 0.000e+00</V>
13 <V ID="10"> 1.000e+02 1.000e+02 0.000e+00</V>
14
15 <V ID="11"> 2.000e+01 -2.000e+01 0.000e+00</V>
16 <V ID="12"> 4.000e+01 -4.000e+01 0.000e+00</V>
17 <V ID="13"> 6.000e+01 -6.000e+01 0.000e+00</V>
18 <V ID="14"> 8.000e+01 -8.000e+01 0.000e+00</V>
19 <V ID="15"> 1.000e+02 -1.000e+02 0.000e+00</V>
20 </VERTEX>
The elements from these vertices are then constructed under the section ELEMENT by
defining
1 <ELEMENT>
2 
3 <S ID="0"> 0 1 </S>
4 <S ID="1"> 1 2 </S>
5 <S ID="2"> 2 3 </S>
6 <S ID="3"> 3 4 </S>
7 <S ID="4"> 4 5 </S>
8 
9 <S ID="5"> 5 6 </S>
10 <S ID="6"> 6 7 </S>
11 <S ID="7"> 7 8 </S>
12 <S ID="8"> 8 9 </S>
13 <S ID="9"> 9 10 </S>
14 
15 <S ID="11"> 5 11 </S>
16 <S ID="12"> 11 12 </S>
17 <S ID="13"> 12 13 </S>
18 <S ID="14"> 13 14 </S>
19 <S ID="15"> 14 15 </S>
20 </ELEMENT>
The composites, which represent groups of elements and boundary regions are defined
under the section COMPOSITE by
1 <COMPOSITE>
2 <C ID="0"> S[0-4] </C> 
3 <C ID="1"> V[0] </C> 
4
5 <C ID="3"> S[5-9] </C> 
6 <C ID="4"> V[10] </C> 
7
8 <C ID="6"> S[11-15] </C> 
9 <C ID="8"> V[15] </C> 
10 </COMPOSITE>
Each of the segments can be then represented under the section DOMAIN by
1 <DOMAIN>
2 <D ID="0"> C[0] </D> 
3 <D ID="1"> C[3] </D> 
4 <D ID="2"> C[6] </D> 
5 </DOMAIN>
We will use the different domains later to define variable material properties and cross-
sectional areas.
12.3.2 Session Info

The PulseWaveSolver is sqpecified through the EquationType option in the session file.
This can be set as follows:
• Projection : Only a discontinuous projection can be specified using the following

option:
– Discontinuous for a discontinous Galerkin (DG) projection.
• TimeIntegrationMethod
• UpwindTypePulse :
– UpwindPulse
12.3.3 Parameters
The following parameters can be specified in the PARAMETERS section of the session file.
• TimeStep is the time-step size;
• FinTime is the final physical time at which the simulation will stop;
• NumSteps is the equivalent of FinTime but instead of specifying the physical final
time the number of time-steps is defined;
• IO_CheckSteps sets the number of steps between successive checkpoint files;
• IO_InfoSteps sets the number of steps between successive info stats are printed
to screen;
• rho density of the fluid. Default value = 1.0;
• nue Poisson’s ratio. Default value = 0.5 ;
• pest external pressure. Default value = 0;
• h0 wall thickness Default value = 1.0;

12.3.4 Boundary conditions

In this section we can specify the boundary conditions for our problem. First we need to
define the variables under the section VARIABLES .
1 <VARIABLES>
2 <V ID="0"> A </V>
3 <V ID="1"> u </V>
4 </VARIABLES>
The composites that we want to apply out boundary conditions then need to be defined
in the BOUNDARYREIONS , for example if we had three composites (C[1], C[4] and C[8])
that correspond to three vertices of the computational mesh we would define:
1 <BOUNDARYREGIONS>
2 C[1] 
3 C[4] 
4 C[8] 
Finally we can specify the boundary conditions on the regions specified under BOUNDARYREGIONS .
The Pulse Wave Solver comes with a number of boundary conditions that are unique to
this solver. Boundary conditions must be provided for both the area and velocity at the
inlets and outlets of the domain. Examples of the different boundary conditions will be
provided in the following.
12.3.4.0.1 Inlet boundary condition: Typically at the inlet of the domain a flow
profile ( Q-inflow ) is specified through a USERDEFINEDTYPE boundary conditioning . An
example inlet condition for the parent artery of the previously bifurcation example is
1 <REGION REF="0">
2 <D VAR="A" USERDEFINEDTYPE="Q-inflow" VALUE="(7.112e-4)*(sin(7.854*t)
3 -0.562)*(1/(1+exp(-400*(sin(7.854*t)-0.562))))" />
4 <D VAR="u" USERDEFINEDTYPE="Q-inflow" VALUE="1.0" />
5 </REGION>
12.3.4.0.2 Terminal boundary conditions: At the outlets of the domain there are
four possible boundary conditions: reflection ( Terminal ), terminal resistance R-terminal ,
Two element windkessel (CR) CR-terminal , and three element windkessel (RCR)
RCR-terminal . An example of the outflow boundary condition of the RCR terminal is
given below
1 <REGION REF="1">
2 <D VAR="A" USERDEFINEDTYPE="RCR-terminal" VALUE="RT" />
3 <D VAR="u" USERDEFINEDTYPE="RCR-terminal" VALUE="C" />
4 </REGION>
Where RT is the total peripheral resistance used in the the R-terminal , CR-terminal
and RCR-terminal models
12.3.5 Functions
The following functions can be specified inside the CONDITIONS section of the session file:
• MaterialProperties : specifies the material properties for each domain.
• A_0 : Initial area of each domain.
• AdvectionVelocity : specifies the advection velocity v.
• InitialConditions : specifies the initial condition for unsteady problems.
• Forcing : specifies the forcing function f
As an example to specify the material properties for each domain in the previous
bifurcation example we would enter:
1 <FUNCTION NAME="MaterialProperties">
2 <E VAR="beta" DOMAIN="0" VALUE="97" />
5 </FUNCTION>
The values of beta are used in the pressure-area relationship (equation 12.2).
12.4 Examples
12.4.1 Human Vascular Network

The Pulse Wave Solver is also capable of handling more complex networks, such as a
complete human arterial tree proposed by Westerhof et al. [38]. In this example, we will
use the refined data from [33] and set up the network shown in the figure in the right.
We will explain how bifurcations are set correctly and how each arterial segment gets its
correct physiological data.
First, we will set up the mesh where each arterial segment is represented by one element
and two vertices respectively. Then, we will subdivide the mesh into different subdomains
by using the <COMPOSITE> section. Here, each arterial segment is described by the
contained elements and its first and last vertex.
The mesh connectivity is specified during the creation of elements by indicating the
starting vertex and ending vertex of each individual artery segment. Shared vertices are
used to describe bifurcations, junctions and mergers between different artery segments in
the network.
The composites are then used to specify the two adjoining segments of an artery, where
the first segment merely allows for description of the connectivity.
12.4 Examples 209

2 <VERTEX>
3 <V ID="0"> 0.000e+00 0.000e+00 0.000e+00</V> 
4 <V ID="1"> 4.000e+00 0.000e+00 0.000e+00</V>
5
6 <V ID="2"> 4.000e+00 0.000e+00 0.000e+00</V> 
7 <V ID="3"> 6.000e+00 0.000e+00 0.000e+00</V>
8
9 <V ID="4"> 4.000e+00 0.000e+00 0.000e+00</V> 
10 <V ID="5"> 7.400e+00 0.000e+00 0.000e+00</V>
11 .
12 .
13 .
14 <V ID="108"> 109.100e+00 -45.000e+00 0.000e+00</V> 
15 <V ID="109"> 143.500e+00 -45.000e+00 0.000e+00</V>
16 </VERTEX>
17 <ELEMENT>
18 <S ID="0"> 0 1 </S>
19 <S ID="1"> 1 2 </S>
20 <S ID="2"> 1 4 </S>
21 <S ID="3"> 2 3 </S>
22 <S ID="4"> 4 5 </S>
23 <S ID="5"> 5 6 </S>
24 <S ID="6"> 5 8 </S>
25 <S ID="7"> 6 7 </S>
26 <S ID="8"> 8 9 </S>

27 .
28 .
29 .
30 <S ID="106"> 103 108 </S>
31 <S ID="107"> 108 109 </S>
32 <S ID="108"> 85 98 </S>
33 <ELEMENT>
34 <COMPOSITE>
35 <C ID="0"> S[0] </C> 
36 <C ID="1"> V[0] </C>
37 <C ID="2"> V[1] </C>
38
39 <C ID="3"> S[1,3] </C> 
40 <C ID="4"> V[2] </C>
41 <C ID="5"> V[3] </C>
42
43 <C ID="6"> S[2,4] </C> 
44 <C ID="7"> V[4] </C>
45 <C ID="8"> V[5] </C>
46 .
47 .
48 .
49 <C ID="162"> S[106,107] </C> 
50 <C ID="163"> V[108] </C>
51 <C ID="164"> V[109] </C>
52 </COMPOSITE>
53 </GEOMETRY>
Then the choice of polynomial order, solver information, area of the arteries and other
parameters are specified.
1 <EXPANSIONS>
2 <E COMPOSITE="C[0]" NUMMODES="5" FIELDS="A,u" TYPE="MODIFIED" />
4 ...
5
7 </EXPANSIONS>
8
9 <CONDITIONS>
10
11 <PARAMETERS>
12
13 TimeStep = 1e-4 
14 FinTime = 1.0 
16 IO_CheckSteps = NumSteps/50 
17 ...
18 A53 = 0.126 
19 A54 = 0.110 
20 A55 = 0.060 
21 </PARAMETERS>
22
23 <SOLVERINFO>
12.4 Examples 211
24 

26 
27 
28 </SOLVERINFO>
29
30 <VARIABLES>
31 <V ID="0"> A </V>
32 <V ID="1"> u </V>
33 </VARIABLES>
The vertices where the network terminates are specified as boundary regions based on
their subsequent composite ids.
1 <BOUNDARYREGIONS>
2 C[1] C[17] C[23] 
3 ...
4 C[164] 
In the boundary conditions section the inflow and outflow conditions are set up. Here we
use an inflow boundary condition for the area at the beginning of the ascending aorta
taken from [33] and plotted on the right. Potential choices for inflow boundary conditions
include Q-Inflow and Time-Dependent inflow. The outflow conditions for the terminal
regions of the network could be specified by different models including eTerminal, R, CR,
RCR and Time-Dependant outflow.
2 <REGION REF="0"> 
3 <D VAR="A" USERDEFINEDTYPE="TimeDependent"
4 VALUE="5.983*(1+0.597*(sin(6.28*t + 0.628) - 0.588)*
5 (1./(1+exp(-2*200*(sin(6.28*t + 0.628) - 0.588)))))" />
6 <D VAR="u" USERDEFINEDTYPE="TimeDependent" VALUE="0.0" />
7 </REGION>
8 <REGION REF="1">
9 <D VAR="A" USERDEFINEDTYPE="TimeDependent" VALUE="A6" />
11 </REGION>
12 <REGION REF="2">

15 </REGION>
16 <REGION REF="3">
19 </REGION>
20 ....
21 <REGION REF="28">
24 </REGION>
Again, for the initial conditions we start our simulation from static equilibrium conditions
A = A0 and for u being initially at rest. The following lines show how we specify A0 and
β for different arterial segments.
2 <E VAR="A" DOMAIN="0" VALUE="5.983" />
3 <E VAR="u" DOMAIN="0" VALUE="0.0" />
4 </FUNCTION>
5 ...
7 <E VAR="A" DOMAIN="54" VALUE="A55" />
9 </FUNCTION>
10
11 <FUNCTION NAME="A_0">
12 <E VAR="A_0" DOMAIN="0" VALUE="A1" />
13 ...
14 <E VAR="A_0" DOMAIN="54" VALUE="A55" />
15 </FUNCTION>
16
19 ...
21 </FUNCTION>
Our simulation is started as described before and the results show the time history for
the conservative variables A and u, as well as for the characteristic variables W1 and W2
at the beginning of the ascending aorta (Artery 1). We can see that physically correct the
shape of the inflow boundary condition appears in the forward traveling characteristic
W1. As we do not have a terminal resistance at the outflow, one would normally expect
W2 to be constant. However this is not the case, as bifurcations cause reflections if the
radii of parent and daughter vessels are not well matching, leading to changes in W2.
The shapes of A and u result from this facts and show the values for the physiological
variables during one cardiac cycle. We may annotate that this values slightly differ from
in vivo measurements due to the missing terminal resistance, which will be added in
future.
12.4 Examples 213
These short examples should give an insight to the functionality of our PulseWaveSolver
and show that results such as luminal area and pressure within the artery can be simulated.
These results can contribute to understanding the physiology of the human vascular
system and they can be used for patient-specific planning of medical interventions.
12.4.2 Stented Artery
12.4.3 Stented Artery
In the following we will explain the usage of the Pulse Wave solver to model the flow and
pressure variation through a stented artery - a cardiovascular procedure in which a small
mesh tube is inserted into an artery to restore blood flow through a constricted region.
Due to the implantation of the stent this region will have different material properties
compared to the the surrounding unstented tissue; hence will influence the propagation
of waves through this system. The stent scenario to be modelled is a straight arterial
segment with a stent situated between x = a1 and x = a2 as shown below.
Figure 12.2 Model of straight artery with a stent in the middle.

12.4.3.0.1 Geometry: In the following we describe the geometry setup for modelling
1D flow in a stent. This is done by defining vertices, elements and composites. The
vertices of the domain are shown below, consisting of 30 elements (Ω) and 31 vertices
(V[n]).
Figure 12.3 1D arterial domain consisting of 30 elements and 31 vertices.
To represent the above in the xml file, we define 31 vertices as follows:

1 <VERTEX>
2 <V ID="0"> 0.000e+00 0.000e+00 0.000e+00</V>
3 .
4 .
5 .
6 <V ID="30">30.000e+00 0.000e+00 0.000e+00</V>
7 </VERTEX>
and the connectivity of these vertices to make up the 30 elements:

1 <ELEMENT>
2 <S ID="0"> 0 1 </S>
3 .
4 .
5 .
6 <S ID="29"> 29 30 </S>
7 </ELEMENT>
These elements are combined to three different composites (shown below): composite 0
represents all the elements; composite 1 the inflow boundary and composite 2 the outflow
boundary.
Figure 12.4 Three composites (C[0], C[1] and C[2]) for the stunted artery.
The above composites are specified as follows:

1 <COMPOSITE>
2 <C ID="0"> S[0-29] </C>
3 <C ID="1"> V[0] </C>
4 <C ID="2"> V[30] </C>
5 </COMPOSITE>
12.4 Examples 215
Finally the domain is specified by the first composite by

1 <DOMAIN>
2 <D ID="0"> C[0] </D>
3 </DOMAIN>
12.4.3.0.2 Expansion: For the expansions we use 4th-order polynomials which define
our two variables A and u on the domain.
1 <EXPANSIONS>
3 </EXPANSIONS>
12.4.3.0.3 Solver Information: The Discontinuous Galerkin Method is used as

projection scheme and the time-integration is performed by a simple Forward Euler
scheme. A full list of possible time integration scheme is given in the parameter section
of the Pulse Wave Solver
1 <SOLVERINFO>
2 
4 
5 
6 </SOLVERINFO>
12.4.3.0.4 Parameters: Parameters used for the simulation are taken from [33]
1 <PARAMETERS>
2 TimeStep = 2e-6 
3 FinTime = 0.25 
5 IO_CheckSteps = NumSteps/50 
7 T = 0.33 
8 h0 = 1.0 
9 rho = 1.0 
10 nue = 0.5 
11 pext = 0.0 
12 a1 = 10.0 
13 a2 = 20.0 
14 kappa = 100.0 
15 Y0 = 1.9099e+5 
16 k = 2 
17 k1 = 200 
18 </PARAMETERS>
12.4.3.0.5 Boundary conditions: At the inflow we apply a pressure boundary con-

dition as shown in the figure below. This condition models the pressure variation during
one heartbeat. A simple absorbing outflow boundary condition is applied the right end
of the tube.
These are defined in the xml file as follows,

Figure 12.5 Pressure profile applied at the inlet of the artery
1 <BOUNDARYREGIONS>
2 C[1] 
3 C[2] 
5
7 <REGION REF="0">
8 <D VAR="A" USERDEFINEDTYPE="TimeDependent" VALUE=
9 "(2000*sin(2*PI*t/T)*1./(1+exp(-2*k1*(T/2-t))-pext)/451352+1)^2" />
11 </REGION>
12 <REGION REF="1">
13 <D VAR="A" VALUE="1.0" />
14 <D VAR="u" VALUE="1.0" />
15 </REGION>
The simulation starts from the static equilibrium of the vessel with normalised area and
velocity.
4 </FUNCTION>
5
6 <FUNCTION NAME="A_0">
8 </FUNCTION>
12.4.3.0.6 Functions: The stent is introduced by applying a variable material proper-

ties function (β - see equation 12.2) along the vessel in the x direction, shown graphically
below and defined in the xml file by
2 <E VAR="E0" DOMAIN="0" VALUE=
12.5 Further Information 217
Figure 12.6 material property variation along the artery. The stiff region in the middle represents
the stent.
3 "Y0*(1.0-kappa/(1+exp(-2*k*(a1-x)))+kappa/(1+exp(-2*k*(a2-x))))" />
4 </FUNCTION>
12.4.3.1 Simulation
The simulation is started by running
PulseWaveSolver Test_1.xml
It will take about 60 seconds on a 2.4GHz Intel Core 2 Duo processor and therefore is
computationally realisable at every clinical site.
12.4.3.2 Results
As a result we get a 3-dimensional interpretation of the aortic cross-sectional area varying
in axial direction both for the stented and non-stented vessel. In case of the stent, the
rigid metal mesh will restrict the deformation of the area in that specific part of the
artery compared to the normal vessel (Fig. 12.7).
Also, if we look at the pressure at three points within the artery (P, M, D) we will
recognize that there are major differences between the stented and normal vessel. While
in the normal vessel (left) the pressure wave applied at the inflow is propagated without
any losses, this does not hold for the stented artery (right). Here, the stiffening at the
stent causes reflections and thus there are losses for total pressure at the medial (M) and
distal (D) point.
12.5 Further Information
The PulseWaveSolver has beqen developed with contributions by various students and
researchers at the Department of Aeronautics, Imperial College London. Further in-
Figure 12.7
formation on the solver and its underlying mathematical framework can be found in
[30, 29].
12.6 Future Development
The PulseWaveSolver is a useful tool for computational modelling of one-dimensional

blood flow in the human body. However, there are several ideas for future development
which include:
1. Inclusion of a pre-processor and post-processor.
2. Profiling the code to improve performance.
3. Cleaning up the input file to make the input format more user-friendly.
4. Modelling of valves and alternative pressure-area laws for models of venous flow.
5. Incorporating a model of the heart.

Chapter 13
Shallow Water Solver
13.1 Synopsis
The ShallowWaterSolver is a solver for depth-integrated wave equations of shallow water

type. Presently the following equations are supported:
Value Description
LinearSWE Linearized SWE solver in primitive variables
(constant still water depth)
NonlinearSWE Nonlinear SWE solver in conservative variables
(constant still water depth)
13.1.1 The Shallow Water Equations

The shallow water equations (SWE) is a two-dimensional system of nonlinear partial
differential equations of hyperbolic type that are fundamental in hydraulic, coastal and
environmental engineering. In deriving the SWE the vertical velocity is considered
negligible and the horizontal velocities are assumed uniform with depth. The SWE are
hence valid when the water depth can be considered small compared to the characteristic
length scale of the problem, as typical for flows in rivers and shallow coastal areas. Despite
the limiting restrictions the SWE can be used to describe many important phenomena,
for example storm surges, tsunamis and river flooding.
The two-dimensional SWE is stated in conservation form as

∂U
+ ∇ · F(U ) = S(U )
∂t
where F(U ) = [E(U ) , G(U )] is the flux vector and the vector of conserved variables read
U = [H , Hu , Hv]T . Here H(x, t) = ζ(x, t) + d(x) is the total water depth, ζ(x, t) is the
free surface elevation and d(x) is the still water depth. The depth-averaged velocity is
219
220 Chapter 13 Shallow Water Solver
denoted by u(x, t) = [u, v]T , where u and v are the velocities in the x- and y-directions,
respectively. The content of the flux vector is
   
Hu Hv
E(U ) =  Hu + gH 2 /2  ,
2 G(U ) =  Hvu ,
   
Huv Hv 2 + gH 2 /2
in which g is the acceleration due to gravity. The source term S(U ) accounts for, e.g.,
forcing due to bed friction, bed slope, Coriolis force and higher-order dispersive effects
(Boussinesq terms). In the distributed version of the ShallowWaterSolver only the Coriolis
force is included.
13.2 Usage
ShallowWaterSolver session.xml
13.3.1 Solver Info
• Eqtype : Specifies the equation to solve. This should be set to NonlinearSWE .
• UpwindType
• Projection
• TimeIntegrationScheme
13.3.2 Parameters
• Gravity
13.3.3 Functions
• Coriolis : Specifies the Coriolis force (variable name: ‘f‘)
• WaterDepth : Specifies the water depth (variable name: ‘d‘)
13.4 Examples
13.4.1 Rossby modon case

This example, provided in RossbyModon_Nonlinear_DG.xml is of a discontinuous Galerkin
simulation of the westward propagation of an equatorial Rossby modon.
13.4 Examples 221
13.4.1.1 Input Options
For what concern the ShallowWaterSolver the <SOLVERINFO> section allows us to specify
the solver, the type of projection (continuous or discontinuous), the explicit time inte-
gration scheme to use and (in the case the discontinuous Galerkin method is used) the
choice of numerical flux. A typical example would be:
1 <SOLVERINFO>
2 
3 
4 
5 
6 </SOLVERINFO>
In the <PARAMETERS> section we, in addition to the normal setting of time step etc., also
define the acceleration of gravity by setting the parameter "Gravity":
1 <PARAMETERS>
2 TimeStep = 0.04 
3 NumSteps = 1000 
6 Gravity = 1.0 
7 </PARAMETERS>
We specify f which is the Coriolis parameter and d denoting the still water depth as
analytic functions:
1 <FUNCTION NAME="Coriolis">
2 <E VAR="f" VALUE="0+1*y" />
3 </FUNCTION>
4
5 <FUNCTION NAME="WaterDepth">
6 <E VAR="d" VALUE="1" />
7 </FUNCTION>
Initial values and boundary conditions are given in terms of primitive variables (please note
that also the output files are given in terms of primitive variables). For the discontinuous
Galerkin we typically enforce any slip wall boundaries weakly using symmetry technique.
This is given by the USERDEFINEDTYPE="Wall" choice in the <BOUNDARYCONDITIONS>
section:
2 <REGION REF="0">
3 <D VAR="eta" USERDEFINEDTYPE="Wall" VALUE="0" />
4 <D VAR="u" USERDEFINEDTYPE="Wall" VALUE="0" />
5 <D VAR="v" USERDEFINEDTYPE="Wall" VALUE="0" />
6 </REGION>
222 Chapter 13 Shallow Water Solver

After the input file has been copied to the build directory of the ShallowWaterSolver
the code can be executed by:
./ShallowWaterSolver Rossby_Nonlinear_DG.xml
13.4.1.3 Post-proceesing
After the final time step the solver will write an output file RossbyModon_Nonlinear_DG.fld .
We can convert it to tecplot format by using the FieldConvert utility. Thus we execute
the following command:
FieldConvert RossbyModon_Nonlinear_DG.xml RossbyModon_Nonlinear_DG.fld \

RossbyModon_Nonlinear_DG.dat
This will generate a file called RossbyModon_Nonlinear_DG.dat that can be loaded directly
into tecplot:
Part IV
Reference
223
Chapter 14
Optimisation
One of the most frequently asked questions when performing almost any scientific
computation is: how do I make my simulation faster? Or, equivalently, why is my
simulation running so slowly?
The spectral element method is no exception to this rule. The purpose of this chapter
is to highlight some of the easiest parameters that can be tuned to attain optimum
performance for a given simulation.
Details are kept as untechnical as possible, but some background information on the
underlying numerical methods is necessary in order to understand the various options
available and the implications that they can have on your simulation.
14.1 Operator evaluation strategies
When discretising a PDE using most variants of the spectral element method, the resulting
problem is usually expressed as a matrix equation. In traditional linear finite element
codes, the matrix is usually represented as a large sparse global matrix, which represents
the action of a particular operator such as the Laplacian matrix across the whole domain.
However, when we consider spectral element methods, in which the polynomial order
representing the expansion can be far higher, this method becomes far less optimal. We
can instead consider the action of an operator locally on each element, and then perform
an assembly operation. This is mathematically equivalent to the global matrix approach
and gives exactly the same answer, but at high polynomial orders it is far more efficient
on modern CPU architectures.
Furthermore, this local approach can be represented in one of two ways: either as a dense
matrix for each element, which is typically more efficient at intermediate polynomial
orders, or in the hp element case as a tensor product of smaller dense matrices via
an approach deemed sum-factorisation, which is used at very high polynomial orders.
Figure ?? gives an overview of these three different operator strategies.
224
14.1 Operator evaluation strategies 225
A goal of Nektar++ is to support not only high order expansions, but all orders from low
(where element size h is the dominant factor) to high (where p dominates); a procedure
we have dubbed “from h to p efficiently”.
14.1.1 Selecting an operator strategy

An obvious question is: “which strategy should I select?” The most important factors in
this decision are:
1. what the operator is;
2. polynomial order p;
3. element type and dimension of the problem;
4. underlying hardware architecture;
5. the number of operator calls in the solver;
6. BLAS implementation speed.
Generally you can use results from three publications [37, 7, 6] which outline results for
two- and three-dimensional elements.
In general, the best approach is to perform some preliminary timings by changing the
appropriate variables in the session file, which is outlined below. As a very rough guide,
for 1 ≤ p ≤ 2 you should use the global approach; for 3 ≤ p ≤ 7 use the local approach;
and for p ≥ 8 use sum-factorisation. However, these guidelines will vary due to the
parameters noted above. In future releases of Nektar++ we hope to tune these variables
automatically to make this decision easier to make.
14.1.2 XML syntax

Operator evaluation strategies can be configured in the GLOBALOPTIMISATIONPARAMETERS
tag, which lies inside the root NEKTAR tag:
1 <NEKTAR>
2 <GLOBALOPTIMIZATIONPARAMETERS>
3 <BwdTrans>
4 <DO_GLOBAL_MAT_OP VALUE="0" />
5 <DO_BLOCK_MAT_OP TRI="1" QUAD="1" TET="1"
6 PYR="1" PRISM="1" HEX="1" />
7 </BwdTrans>
8 <IProductWRTBase>
11 PYR="1" PRISM="1" HEX="1" />
12 </IProductWRTBase>
13 <HelmholtzMatrixOp>
226 Chapter 14 Optimisation

16 PYR="1" PRISM="1" HEX="1" />
17 </HelmholtzMatrixOp>
18 <MassMatrixOp>
21 PYR="1" PRISM="1" HEX="1" />
22 </MassMatrixOp>
23 </GLOBALOPTIMIZATIONPARAMETERS>
24 </NEKTAR>
14.1.3 Selecting different operator strategies

Operator evaluation is supported for four operators: backward transform, inner product,
Helmholtz and mass operators. It is possible to specify the following optimisation flags
for different operators:
1. DO_GLOBAL_MAT_OP : If VALUE is 1 , the globally assembled system matrix will

be used to evaluate the operator. If VALUE is 0 , the operator will be evaluated
elementally.
2. DO_BLOCK_MAT_OP : If VALUE is 1 , the elemental evaluation will be done using

the elemental/local matrices (which are all concatenated in a block matrix, hence
the name). If VALUE is 0 , the elemental evaluation will be done using the sum-
factorisation technique.
Each element type (triangle, quadrilateral, etc) has its own VALUE, since break-even
points for sum-factorisation and the local matrix approach will differ depending on
element type. Note that due to a small shortcoming in the code, all element types
must be defined; so three-dimensional elements must still be defined even if the
simulation is two-dimensional.
Note that global takes precendence over block, so if VALUE is set to 1 for both then the
operator will be global.
For very complex operators – in particular HelmholtzMatrixOp – always set DO_BLOCK_MAT_OP

to 1 as sum-factorisation for these operator types can be costly.
14.2 Collections
The Collections library adds optimisations to perform certain elemental operations

collectively by applying an operator using a matrix-matrix operation, rather than a
sequence of matrix-vector multiplications. Certain operators benefit more than other
from this treatment, so the following implementations are available:
14.2 Collections 227
• StdMat: Perform operations using collated matrix-matrix type elemental operation.
• SumFac: Perform operation using collated matrix-matrix type sum factorisation

operations.
• IterPerExp: Loop through elements, performing matrix-vector operation.
• NoCollections: Use the original LocalRegions implementation to perform the

operation.
All configuration relating to Collections is given in the COLLECTIONS XML element within
the NEKTAR XML element.
14.2.1 Default implementation

The default implementation for all operators may be chosen through setting the DEFAULT
attribute of the COLLECTIONS XML element to one of StdMat , SumFac , IterPerExp or
NoCollection . For example, the following uses the collated matrix-matrix type elemental
operation for all operators and expansion orders:
1 <COLLECTIONS DEFAULT="StdMat" />
14.2.2 Auto-tuning
The choice of implementation for each operator, for the given mesh and expansion orders,
can be selected automatically through auto-tuning. To enable this, add the following to
the Nektar++ session file:
1 <COLLECTIONS DEFAULT="auto" />
This will collate elements from the given mesh and given expansion orders, run and time
each implementation strategy in turn, and select the fastest performing case. Note that
the selections will be mesh- and order- specific. The selections made via auto-tuning are
output if the –verbose command-line switch is given.
14.2.3 Manual selection

The choice of implementation for each operator may be set manually within the COLLECTIONS
tag as shown in the following example. Different implementations may be chosen for
different element shapes and expansion orders. Specifying * for ORDER sets the default
implementation for any expansion orders not explicity defined.
1 <COLLECTIONS>
2 <OPERATOR TYPE="BwdTrans">
3 <ELEMENT TYPE="T" ORDER="*" IMPTYPE="IterPerExp" />
4 <ELEMENT TYPE="T" ORDER="1-5" IMPTYPE="StdMat" />
5 </OPERATOR>
6 <OPERATOR TYPE="IProductWRTBase">
7 <ELEMENT TYPE="Q" ORDER="*" IMPTYPE="SumFac" />
228 Chapter 14 Optimisation
8 </OPERATOR>
9 </COLLECTIONS>
Manual selection is intended to document the optimal selections on a given hardware

platform after extensive prior testing, to avoid the need to run the auto-tuning for each
run.
14.2.4 Collection size

The maximum number of elements within a single collection can be enforced using the
MAXSIZE attribute.
Chapter 15
Command-line Options
--verbose
Displays extra info.
--version
Displays software version, and source control information if applicable.
--help
Displays help information about the available command-line options for the
executable.
--parameter [key]=[value]
Override a parameter (or define a new one) specified in the XML file.
--solverinfo [key]=[value]
Override a solverinfo (or define a new one) specified in the XML file.
--io-format [format]
Determines the output format for writing Nektar++ field files that are used to
store, for example, checkpoint and solution field files. The default for format is
Xml , which is an XML-based format, which is written as one file per process. If
Nektar++ is compiled with HDF5 support, then an alternative option is Hdf5 ,
which will write one file for all processes and can be more efficient for very
large-scale parallel jobs.
--npx [int]
When using a fully-Fourier expansion, specifies the number of processes to use
in the x-coordinate direction.
--npy [int]
When using a fully-Fourier expansion or 3D expansion with two Fourier direc-
tions, specifies the number of processes to use in the y-coordinate direction.
229
230 Chapter 15 Command-line Options
--npz [int]
When using Fourier expansions, specifies the number of processes to use in the
z-coordinate direction.
--part-info
Prints detailed information about the generated partitioning, such as number
of elements, number of local degrees of freedom and the number of boundary
degrees of freedom.
--part-only [int]
Partition the mesh only into the specified number of partitions, write to file
and exit. This can be used to pre-partition a very large mesh on a single
high-memory node, prior to being executed on a multi-node cluster.
--use-metis
Forces the use of METIS for mesh partitioning. If Nektar++ is compiled with
Scotch support, the default is to use Scotch.
--use-scotch
Forces the use of Scotch for mesh partitioning.
Chapter 16
Frequently Asked Questions
16.1 Compilation and Testing
Q. I compile Nektar++ successfully but, when I run ctest, all the tests fail.
What might be wrong?
On Linux or Mac, if you compile the ThirdParty version of Boost, rather than using
version supplied with your operating system (or MacPorts on a Mac), the libraries will
be installed in the ThirdParty/dist/lib subdirectory of your Nektar++ directory.
When Nektar++ executables are run, the Boost libraries will not be found as this path
is not searched by default. To allow the Boost libraries to be found set the following
environmental variable, substituting $NEKTAR_HOME with the absolute path of your
Nektar++ directory:
• On Linux (sh, bash, etc)
export LD_LIBRARY_PATH=${NEKTAR_HOME}/ThirdParty/dist/lib
or (csh, etc)
setenv LD_LIBRARY_PATH ${NEKTAR_HOME}/ThirdParty/dist/lib
• On Mac
export DYLD_LIBRARY_PATH=${NEKTAR_HOME}/ThirdParty/dist/lib
Q. How to I compile Nektar++ to run in parallel?
Parallel execution of all Nektar++ solvers is available using MPI. To compile using MPI,
enable the NEKTAR_USE_MPI option in the CMake configuration. On recent versions of
231
232 Chapter 16 Frequently Asked Questions
MPI, the solvers can still be run in serial when compiled with MPI. More information on
Nektar++ compilation options is available in Section 1.3.5.
Q. When compiling Nektar++, I receive the following error:
CMake Error: The following variables are used in this

oject, but they are set to NOTFOUND.
Please set them or make sure they are set and tested
rrectly in the CMake files:
NATIVE_BLAS (ADVANCED)
linked by target "LibUtilities" in directory
/path/to/nektar++/library/LibUtilities
NATIVE_LAPACK (ADVANCED)
linked by target "LibUtilities" in directory
/path/to/nektar++/library/LibUtilities
This is caused by one of two problems:
• The BLAS and LAPACK libraries and development files are not installed. On
Linux systems, both the LAPACK library package (usually called liblapack3 or
lapack) and the development package (usually called liblapack-dev or lapack-devel)
must be installed. Often the latter is missing.
• An alternative BLAS/LAPACK library should be used. HPC systems frequently

use the Intel compilers (icc, icpc) and the Intel Math Kernel Library (MKL). This
software should be made available (if using the modules environment) and the
option NEKTAR_USE_MKL should be enabled.
Q. When I compile Nektar++ I receive an error
error: #error "SEEK_SET is #defined but must not be for

the C++ binding of MPI. Include mpi.h before stdio.h"
This can be fixed by including the flags
-DMPICH_IGNORE_CXX_SEEK -DMPICH_SKIP_MPICXX
in the CMAKE_CXX_FLAGS option within the ccmake configuration.
Q. After installing Nektar++ on my local HPC cluster, when I run the ’ctest’
command, all the parallel tests fail. Why is this?
The parallel tests are those which include the word parallel or par . On many HPC
systems, the MPI binaries used to execute jobs are not available on the login nodes, to
prevent inadvertent parallel runs outside of the queuing system. Consequently, these
16.2 Usage 233
tests will not execute. To fully test the code, you can submit a job to the queuing system
using a minimum of two cores, to run the ctest command.
Q. When running any Nektar++ executable on Windows, I receive an error

that zlib.dll cannot be found. How do I fix this?
Windows searches for DLL files in directories specified in the PATH environmental
variable. You should add the location of the ThirdParty files to your path. To fix this
(example for Windows XP):
• As an administrator, open ”System Properties” in control panel, select the ”Ad-

vanced” tab, and select ”Environment Variables”.
• Edit the system variable ‘path‘ and append
C:\path\to\nektar++\ThirdParty\dist\bin
to the end, replacing path\to\nektar++ appropriately.
Q. When compiling Nektar++ Thirdparty libraries I get an error “CMake

Error: Problem extracting tar”
Nektar++ tries to download the appropriate ThirdParty libraries. However if the

download protocols are restricted on your computer this may fail leading to the error
“‘CMake Error: Problem extracting tar”. These libraries are available from
http://www.nektar.info/thirdparty/
and can be downloaded directly into the $NEKTAR_HOME/ThirdParty directory
16.2 Usage
Q. How do I run a solver in parallel?
In a desktop environment, simply prefix the solver executable with the mpirun helper.
For example, to run the Incompressible Navier-Stokes solver on a 4-core desktop computer,
you would run
mpirun -np 4 IncNavierStokesSolver Cyl.xml
In a cluster environment, using PBS for example, the mpiexec command should be used.
Q. How can I generate a mesh for use with Nektar++?
Nektar++ supports a number of mesh input formats. These are converted to the
Nektar++ native XML format (see Section 3) using the NekMesh utility (see Section 4.
Supported formats include:
234 Chapter 16 Frequently Asked Questions
• Gmsh (.msh)
• Polygon (.ply)
• Nektar (.rea)
• Semtex (.sem)
Bibliography
[1] M Ainsworth and S Sherwin. Domain decomposition preconditioners for p and hp

finite element approximation of stokes equations. COMPUTER METHODS IN
APPLIED MECHANICS AND ENGINEERING, 175:243–266, 1999.
[2] R. R. Aliev and A. V. Panfilov. A simple two-variable model of cardiac excitation.

Chaos, Solitons & Fractals, 7:293–301, 1996.
[3] Ivo Babuška and Manil Suri. The p and h-p versions of the finite element method,
basic principles and properties. SIAM review, 36(4):578–632, 1994.
[4] Y. Bao, R. Palacios, M. Graham, and S.J. Sherwin. Generalized “thick” strip
modelling for vortex-induced vibration of long flexible cylinders. J. Comp. Phys,
321:1079–1097, 2016.
[5] P-E Bernard, J-F Remacle, Richard Comblen, Vincent Legat, and Koen Hillewaert.
High-order discontinuous galerkin schemes on general 2d manifolds applied to the
shallow water equations. Journal of Computational Physics, 228(17):6514–6535,
2009.
[6] C. D. Cantwell, S. J. Sherwin, R. M. Kirby, and P. H. J. Kelly. From h to p efficiently:

selecting the optimal spectral/hp discretisation in three dimensions. Math. Mod.
Nat. Phenom., 6:84–96, 2011.
[7] C. D. Cantwell, S. J. Sherwin, R. M. Kirby, and P. H. J. Kelly. From h to p efficiently:

strategy selection for operator evaluation on hexahedral and tetrahedral elements.
Computers & Fluids, 43:23–28, 2011.
[8] CD Cantwell, D Moxey, A Comerford, A Bolis, G Rocco, G Mengaldo, D De Grazia,

S Yakovlev, J-E Lombard, D Ekelschot, et al. Nektar++: An open-source spectral/hp
element framework. Computer Physics Communications, 192:205–219, 2015.
[9] D. De Grazia, G. Mengaldo, D. Moxey, P. E. Vincent, and S. J. Sherwin. Connections

between the discontinuous galerkin method and high-order flux reconstruction
235
236 Bibliography
schemes. International Journal for Numerical Methods in Fluids, 75(12):860–877,

2014.
[10] S. Dong. A convective-like energy-stable open boundary condition for simulation of

incompressible flows. Journal of Computational Physics, 302:300–328, 2015.
[11] S. Dong, G. E. Karniadakis, and C. Chryssostomidis. A robust and accurate

outflow boundary condition for incompressible flow simulations on severely-truncated
unbounded domains. Journal of Computational Physics, 261:95–136, 2014.
[12] AP Dowling and Ffowcs-Williams JE. Sound and sources of sound. Ellis Horwood
series in engineering science, 1983.
[13] Niederer ”et al.”. Verification of cardiac tissue electrophysiology simulators using an
n-version benchmark. Philos Transact A Math Phys Eng Sci, 369:4331–51, 2011.
[14] Abel Gargallo-Peiró, Xevi Roca, Jaime Peraire, and Josep Sarrate. Distortion
and quality measures for validating and generating high-order tetrahedral meshes.
Engineering with Computers, 31(3):423–437, 2015.
[15] David Gottlieb, Steven A Orszag, and CAMBRIDGE HYDRODYNAMICS INC

MA. Numerical analysis of spectral methods. SIAM, 1977.
[16] J.L. Guermond and J. Shen. Velocity-correction projection methods for incompress-
ible flows. SIAM J. Numer. Anal., 41:112–134, 2003.
[17] Jan S Hesthaven and Tim Warburton. Nodal high-order methods on unstructured
grids: I. time-domain solution of maxwell’s equations. Journal of Computational
Physics, 181(1):186–221, 2002.
[18] B. E. Jordi, C. J. Cotter, and S. J. Sherwin. Encapsulated formulation of the

selective frequency damping method. Phys. Fluids, 2014.
[19] G. E. Karniadakis, M. Israeli, and S. A. Orszag. High-order splitting methods for

the incompressible Navier–Stokes equations. 97(2):414–443, 1991.
[20] Jonas Koko. Vectorized matlab codes for linear two-dimensional elasticity. Scientific
Programming, 15(3):157–172, 2007.
[21] C. H. Luo and Y. Rudy. A model of the ventricular cardiac action potential.
depolarization repolarization and their interaction. Circulation research, 68:1501–
1526, 1991.
[22] R. J. Ramirez M. Courtemanche and S. Nattel. Ionic mechanisms underlying human

atrial action potential properties: insights from a mathematical model. American
Journal of Physiology-Heart and Circulatory Physiology, 275:H301–H321, 1998.
[23] Y. Maday, A. T. Patera, and E.M. Ronquist. An operator-integration-factor splitting

method for time-dependent problems: Application to incompressible fludi flow. J.
Sci. Comp., 4:263–292, 1990.
Bibliography 237
[24] Gianmarco Mengaldo, Daniele De Grazia, Freddie Witherden, Antony Farrington,

Peter Vincent, Spencer Sherwin, and Joaquim Peiro. A Guide to the Implemen-
tation of Boundary Conditions in Compact High-Order Methods for Compressible
Aerodynamics. American Institute of Aeronautics and Astronautics, 2014/08/10
2014.
[25] D. Moxey, M. Hazan, J. Peiró, and S. J. Sherwin. An isoparametric approach to

high-order curvilinear boundary-layer meshing. Comp. Meth. Appl. Mech. Eng.,
2014.
[26] D. Moxey, M. Hazan, J. Peiró, and S. J. Sherwin. On the generation of curvilinear

meshes through subdivision of isoparametric elements. to appear in proceedings of
Tetrahedron IV, 2014.
[27] David J Newman and George Em Karniadakis. A direct numerical simulation study
of flow past a freely vibrating cable. Journal of Fluid Mechanics, 344:95–136, 1997.
[28] Anthony T Patera. A spectral element method for fluid dynamics: laminar flow in a
channel expansion. Journal of computational Physics, 54(3):468–488, 1984.
[29] N Pignier. One-dimensional modelling of blood flow in the cardiovascular system,

2012.
[30] CJ Roth. Pulse wave propagation in the human vascular system, 2012.
[31] S Sherwin. A substepping navier-stokes splitting scheme for spectral/hp element

discretisations. pages 43–52. Elsevier Science, 2003.
[32] SJ Sherwin and M Ainsworth. Unsteady navier-stokes solvers using hybrid spec-
tral/hp element methods. APPLIED NUMERICAL MATHEMATICS, 33:357–363,
2000.
[33] SJ Sherwin, L Formaggia, J Peiró, and V Franke. Computational modelling of 1d

blood flow with variable mechanical properties and its application to the simulation
of wave propagation in the human arterial system. Int. J. Numer. Meth. Fluids,
43:673–700, 2003.
[34] SJ Sherwin and G Em Karniadakis. Tetrahedral hp finite elements:

Algorithms and flow simulations. Journal of Computational Physics, 124(1):14–45,
1996.
[35] K. H. W. J. ten Tusscher and A. V. Panfilov. Alternans and spiral breakup in

a human ventricular tissue model. American Journal of Physiology-Heart and
Circulatory Physiology, 291:H1088–H1100, 2006.
[36] M Turner, J Peiró, and D Moxey. A Variational Framework for High-Order Mesh
Generation. In 25th International Meshing Roundtable, volume 163, pages 340–352,
2016.
238 Bibliography
[37] Peter EJ Vos, Spencer J Sherwin, and Robert M Kirby. From h to p efficiently:
Implementing finite and spectral/hp element methods to achieve optimal perfor-
mance for low-and high-order discretisations. Journal of Computational Physics,
229(13):5161–5181, 2010.
[38] N Westerhof. Anatomic studies of the human systemic arterial tree. J. Biomech.,
2:121–143, 1969.
[39] D Xiu, SJ Sherwin, S Dong, and GE Karniadakis. Strong and auxiliary forms of the
semi-lagrangian method for incompressible flows. J. Sci. Comp., 25:323–346, 2005.
[40] Olgierd Cecil Zienkiewicz and Robert Leroy Taylor. Basic formulation and linear
problems. McGraw-Hill, 1989.

User-Guide-4 4 1

Uploaded by

Copyright:

Available Formats

User-Guide-4 4 1

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

User-Guide-4 4 1

Uploaded by

Copyright:

Available Formats

Nektar++: Spectral/hp

Department of Aeronautics, Imperial College London, UK

3 XML Session File 23

II Preprocessing & Postprocessing 58

4.4.2 Output modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5.4.21 Print L2 and LInf norms: printfldnorms module . . . . . . . . . . 91

7 Acoustic Perturbation Equations Solver 111

8 Cardiac Electrophysiology Solver 115

8.3.1 Solver Info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

9 Compressible Flow Solver 122

10 Incompressible Navier-Stokes Solver 134

10.7.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

11 Linear elasticity solver 198

12 Pulse Wave Solver 203

12.6 Future Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218

13 Shallow Water Solver 219

15 Command-line Options 229

16 Frequently Asked Questions 231

• Representation of one, two and three-dimensional fields as a collection of piecewise

• Both hierarchical and nodal expansion bases.

• Continuous or discontinuous Galerkin operators.

• Cross platform support for Linux, Mac OS X and Windows.

Our current goals are to develop:

• Automatic auto-tuning of optimal operator implementations based upon not only

• Temporal and spatial adaption.

• Features enabling evaluation of high-order meshing techniques.

Nektar++ is available in both a source-code distribution and as pre-compiled binary

1.1 Installing Debian/Ubuntu Packages

2. Update the package lists

3. Install the required Nektar++ packages, or the complete suite with:

apt-get install nektar++

Any additional dependencies required for Nektar++ to function will be automati-

1.2 Installing Redhat/Fedora Packages

Add a file to the directory /etc/yum.repos.d/nektar.repo with the following contents

1.3 Installing from Source

1.3.1 Obtaining the source code

• Clone the git repository

git clone http://gitlab.nektar.info/clone/nektar/nektar.git nektar++

git clone git@gitlab.nektar.info:nektar/nektar.git nektar++

1.3.2.2 Quick Start

If you have downloaded the tarball, first unpack it:

tar -zxvf nektar++-4.4.1.tar.gz

Change into the nektar++ source code directory

mkdir -p build && cd build

1.3.2.3 Detailed instructions

1. If you have downloaded the tarball, first unpack it

tar -zxvf nektar++-4.4.1.tar.gz

mkdir -p build && cd build

3. Run the CMake GUI and configure the build by pressing c

• Select the components of Nektar++ (prefixed with NEKTAR_BUILD_ ) you would

A full list of configuration options can be found in Section 1.3.5.

6. Compile the code

During the build, missing third-party libraries will be automatically downloaded,

7. Test the build by running unit and regression tests.

sudo port install cmake

1.3.3.2 Quick Start

If you have downloaded the tarball, first unpack it:

tar -zxvf nektar++-4.4.1.tar.gz