0% found this document useful (0 votes)
270 views

CEE 471, Fall 2019: HW2 Solutions: Bhavesh Shrimali, Aditya Kumar

This document contains solutions to homework problems from a civil engineering course. It includes: 1) Calculations of the gradient, divergence, and curl of various vector and tensor expressions. 2) Derivations of the derivatives of the functions f(A) = tr(A^-1A^-T) and g(A) = det(A). It is shown that the Gâteaux derivative of g(A) is Dg(A)[U] = det(A)tr(A^-1U). 3) Expressions for the second derivatives of f(A). So in summary, it provides detailed solutions to calculus of tensors homework problems involving derivatives of functions of tensors

Uploaded by

Jonathan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
270 views

CEE 471, Fall 2019: HW2 Solutions: Bhavesh Shrimali, Aditya Kumar

This document contains solutions to homework problems from a civil engineering course. It includes: 1) Calculations of the gradient, divergence, and curl of various vector and tensor expressions. 2) Derivations of the derivatives of the functions f(A) = tr(A^-1A^-T) and g(A) = det(A). It is shown that the Gâteaux derivative of g(A) is Dg(A)[U] = det(A)tr(A^-1U). 3) Expressions for the second derivatives of f(A). So in summary, it provides detailed solutions to calculus of tensors homework problems involving derivatives of functions of tensors

Uploaded by

Jonathan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

CEE 471, Fall 2019: HW2 Solutions

Bhavesh Shrimali, Aditya Kumar


Department of Civil and Environmental Engineering, University of Illinois, Urbana–Champaign, IL 61801, USA

1. u(x) = (x1 − x3 )e1 + (x2 − x3 )e2 + (x1 + x2 − x3 )e3 .

(a) ∇u = grad(u) or in indicial notation ui,j

∂u1 ∂u1 ∂u1


 
 ∂x1 ∂x2 ∂x3 
 ∂u
 2 ∂u2 ∂u2 
∇u = 


 ∂x1 ∂x2 ∂x3 
 ∂u3 ∂u3 ∂u3 
∂x1 ∂x2 ∂x3
 
1 0 −1
= 0 1 −1 . ♣ ... ( 5 points)
1 1 −1

Note that for the indicial notation ui,j , the first index i goes with the rows and the second index
j goes with the columns.
(b) We have x = x1 e1 + x2 e2 + x3 e3 . Then in indicial notation,x · u can be expressed as xj uj .
Therefore grad(x · u) is

(xj uj ),i = uj xj,i + xj uj,i


= ui + xj uj,i

Computing the three components, we get


Ñ é
x1
grad(x · u) = 2 x2 . ♣ ... ( 5 points)
−x3

(c) x ⊗ u can be expressed in indicial notation as xi uj . Therefore div(x ⊗ u) is

(xi uj ),j = uj xi,j + xi uj,j


= ui + xi uj,j
= ui + xi (uj,j = 1 + 1 − 1 = 1)

Computing the three components, we get


Ñ é
2x1 − x3
div(x ⊗ u) = 2x2 − x3 . ♣ ... ( 5 points)
x1 + x2

Email addresses: bshrima2@illinois.edu (Bhavesh Shrimali), akumar51@illinois.edu (Aditya Kumar)


(d)
Ñ é Ñ é Ñ é
x1 x1 − x3 x2 (x1 + x2 − x3 ) − x3 (x2 − x3 )
x×u= x2 × x2 − x3 = x3 (x1 − x3 ) − x1 (x1 + x2 − x3 )
x3 x1 + x2 − x3 x1(x2 − x3 ) − x2 (x1 − x3 )
x1 x2 + x22 − 2x2 x3 + x23
Ñ é

= −x21 − x1 x2 + 2x1 x3 − x23 .


x2 x3 − x1 x3
Hence
Ñ é
x2 x1 + 2x2 − 2x3 −2x2 + 2x3
∇(x×u) = −2x1 − x2 + 2x3 −x1 2x1 − 2x3 . ♣ ... ( 5 points)
−x3 x3 x2 − x1
(e) Let A be a constant second-order tensor. Then, x · Ax can be expressed in indicial notation as
xi Aij xj . Hence ∇(x · Ax) reads as
(Akj xk xj ),i = Akj (xk xj ),i (A is constant)
= Akj (xk,i xj + xk xj,i )
= Akj (δki xj + xk δji )
= Aij xj + Aki xk .

Thus ∇(x · Ax) = (A + AT ) · x. Finally, we can write div(∇(x · Ax)) as


(Aij xj ),i + (Aki xk ),i = Aij δij + Aki δki = 2Aii , ... ( 5 points)
or div(∇(x · Ax)) = 2tr(A). ♣

2. (a) f (A) = tr(A−1 A−T )


Recall A−1 A−T = A−1 −1 −1 −1
ik Ajk . Hence f (A) = Aik Aik . Therefore

∂A−1 −1
kl Akl ∂A−1
= 2A−1
kl
kl
(1)
∂Aij ∂Aij
∂A−1 ∂(Aim A−1
mj )
To compute kl
, we make use of the identity Aik A−1
kj = δ ij . Differentiating it gives =
∂Aij ∂Akl
0. Expanding with the product rule yields to
−1
∂Aim −1 ∂Amj
Amj + Aim =0
∂Akl ∂Akl
∂A−1
mj
δik δml A−1
mj + A im =0
∂Akl
∂A−1
mj
δik A−1
lj + A im =0 (2)
∂Akl
Multiplying both side of Equation ((2)) by A−1
ni (note the choice of indices) gives

∂A−1
nj
= −δik A−1 −1 −1 −1
ni Alj = −Ank Alj
∂Akl
∂A−1
ij
or = −A−1 −1
ik Alj .
∂Akl
2
∂A−1
Therefore kl
= −A−1 −1
ki Ajl . Plugging it back in Equation ((1)), we get
∂Aij

∂A−1 −1
kl Akl
= −2A−1 −1 −1
ki Akl Ajl ... ( 5 points)
∂Aij

∂f
Then = −2A−T A−1 A−T . ♣
∂A
It follows that
∂2f
= 2(A−1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1
km Ani Ajl Akl +Aki Ajm Anl Akl +Aki Ajl Akm Anl ). ♣ ... ( 5 points)
∂Aij ∂Amn

(b) (i) Using the definition of Gâteux Derivative: Dg(A)

det (I + S) = 1 + trS + o(S) (3)

Hence, for A and U ∈ Lin, where Lin is the set of all second order tensors, we have

det(A + U) = det A I + A−1 U


 

=⇒ det(A + U) = (detA) 1 + tr A−1 U + o(A−1 U)


  

=⇒ det(A + U) − det(A) = det(A)tr A−1 U + o(U) ∵ det (A) o(A−1 U) = o (U)


 

Thus, det(A + U)−det(A) is equal to a term linear in U plus a term that approaches 0 faster
than U. Finally we have:

=⇒ Dg(A)[U] = det(A)tr A−1 U + o(U)




As U −→ 0, we have
∂g
· U = det(A)tr A−1 U

Dg(A)[U] =
∂A
∂g
=⇒ · U = det(A)A−T · U
∂A
Å ã
∂g
− det(A)A−T · U = 0
∂A
∂g
=⇒ = det(A)A−T ♣ ... ( 7.5 points) (4)
∂A
(ii) Recall from HW1-Problem 7 that
1î 3
ó
(trA) − 3 (trA) trA2 + 2trA3 ,

g(A) = detA =
6
so that, using the product rule for derivation, we have

∂(trA2 ) ∂(trA3 )
ï ò
∂g 1 2 ∂(trA) ∂(trA) 2
= 3 (trA) −3 trA − 3trA +2 , (5)
∂A 6 ∂A ∂A ∂A ∂A

∂(trA) ∂(trA2 ) ∂(trA3 )


In the sequel we compute , and using indicial notation and the chain
∂A ∂A ∂A
rule.

3
• trA = Akk . Thus
∂(trA) ∂Akk
=
∂A ∂Aij
= δik δkj
= δij = I. (6)

• A2 = Aik Akj =⇒ trA2 = Aik Aki . Therefore

∂(trA2 ) ∂Akl Alk


=
∂A ∂Aij
∂Akl ∂Alk
= Alk + Akl
∂Aij ∂Aij
= δik δjl Alk + Akl δil δjk
= 2Aji = 2AT . (7)

• A3 = Aik Akl Alj =⇒ trA3 = Aik Akl Ali . Therefore

∂(trA3 ) ∂Amk Akl Alm


=
∂A ∂Aij
∂Amk ∂Akl ∂Alm
= Akl Alm + Amk Alm + Amk Akl
∂Aij ∂Aij ∂Aij
= δim δjk Akl Alm + Amk δki δjl Alm + Amk Akl δli δmj
= 3Ajk Aki = 3(A2 )T . (8)

Plugging Equations ((6))-((8)) in Equation ((5)) yields to


∂g 1î 2
ó
= 3((trA) − trA2 )I − 6trA AT + 6(A2 )T
∂A 6
= (A2 )T − I1 (A)AT + I2 (A)I (using the definition of the invariants of A)
= I3 (A)A−T (Cayley-Hamilton theorem, see HW1 Pbm 8) (9)

∂ det(A) ∂ det(A)
Therefore, = det(A)A−T or with indicial notation = det(A)A−1
ji . ♣
∂A ∂Aij

We now take the second derivative of det(A). To do so, we compute the derivative of the previous
expression with respect with respect to Akl because the first derivative was taken earlier with respect
to Aij (see Equations ((6))-((8))). Using the product rule and the derivative of A−1 with respect to
A derived before, we have
∂2g ∂2g
= (10)
∂A2 ∂Aij ∂Akl
∂ det(A) −1 ∂A−1
ji
= Aji + det(A) (11)
∂Akl ∂Akl
= det(A)A−1 −1 −1 −1
lk Aji − det(A)Ajk Ali (12)
= det(A)(A−1 −1
lk Aji − A−1 −1
jk Ali ) ... ( 7.5 points) (13)

4
3. We consider a vector field v(x) = (x1 e1 + x2 e2 ) ln(x21 + x22 ) defined over a cylinder Ω of radius R and
height H as represented in Figure (fig. 1). The boundary of the cylinder ∂Ω is the union of the lateral

Figure 1: Cylindrical body of interest.

face ∂Ω1 and the top and bottom faces ∂Ω2 and ∂Ω3 . Therefore the outward normal unit vector n(i)
of ∂Ωi is given by
x1 x2
• n(1) = e1 + e2
R R
• n(2) = e3
• n(3) = −e3
Following the definition, we compute div(v(x))

∂v1 (x) ∂v2 (x)


div(v(x)) = vi,i = +
x1 x2
2
2x 2x2
= ln(x21 + x22 ) + 2 1 2 + ln(x21 + x22 ) + 2 2 2 (product rule)
x1 + x2 x1 + x2
= 2[ln(x21 + x22 ) + 1]. ♣ ... ( 10 points)

Next
ZZZ ZZ
div(v(x)) dx = v(x) · n dx (Divergence Theorem)

Z Z∂Ω ZZ ZZ
= v(x) · n(1) dx + v(x) · n(2) dx + v(x) · n(3) dx
∂Ω1 ∂Ω2 ∂Ω3
ZZ
(1) (i)
= v(x) · n dx (v · n = 0 for i = 2, 3)
∂Ω1
x21 x2
ZZ Å ã
= ln(R2 ) + 2 dx (x11 + x22 = R2 on ∂Ω1 )
R R
Z∂Ω
Z1
= 2R ln(R) dx
∂Ω1
= 4πHR2 ln(R). ♣ ... ( 10 points)

The volume integral of div(v(x)) can be computed without using the divergence theorem. However it
would be more difficult.

5
4. Let Ω denote the volume occupied by a unit sphere, the surface of which is denoted by ∂Ω and has an
outward normal unit vector n. Consider a frame of reference in which origin coincides with the center
of the sphere. In this reference frame, the position vector is aligned with the normal vector.
RR
(a) We write ∂Ω
x ⊗ n dx in indicial notation as
ZZ ZZZ
∂xi
xi nj dx = dx (Divergence Theorem)
∂Ω ∂xj
Z Z ZΩ
= δij dx
ZΩZ Z
= δij dx


= δij . ♣ ... ( 10 points)
3

(b) Following the same derivation, we have


ZZ ZZZ
∂(ni nj )
ni nj nk dx = dx (Divergence Theorem)
∂Ω ∂xk
Z Z ZΩ
∂(xi xj )
= dx (ni = xi for a unit sphere)
Ω ∂xk
ZZZ
= δik xj + δjk xi dx
ZΩZ Z ZZZ
= δik xj dx + δjk xi dx.
Ω Ω
RRR
We now just need to evaluate x dx and we use spherical coordinates to do so. We have
Ω i
therefore,
Ñ é Ñ é
x1 r cos(φ) sin(θ)
x= x2 = r sin(φ) sin(θ) and dx = r2 sin(θ) dr dθ dφ for the volume element.
x3 r cos(θ)

Then, we have for i = 1


ZZZ Z 1 Z π Z 2π
x1 dx = r3 cos(φ) sin(θ) sin(θ) dr dθ dφ
Ω r=0 θ=0 φ=0
ÇZ 2π å
=0 cos(φ) dφ = 0
0

RRR
Similarly, Ω xi dx = 0 for i = 2, 3 respectively because the problem is symmetric with respect
to the cartesian basis. Therefore,
ZZZ ZZ
xi dx = 0 =⇒ ni nj nk dx = 0. ♣ . . . ( 10 points)
Ω ∂Ω

6
(c) Following the same derivation again, we have
ZZ ZZZ
∂(ni nj nk )
ni nj nk nl dx = dx (Divergence Theorem)
∂Ω ∂xl
Z Z ZΩ
∂(xi xj xk )
= dx (ni = xi for a unit sphere)
∂xl
Z Z ZΩ
= δil xj xk + δjl xi xk + δkl xi xj dx

ZZ Z ZZZ ZZZ
= δil xj xk dx + δjl xi xk dx + δkl xi xj dx.
Ω Ω Ω
RRR
We now just need to evaluate x x dx and we use again the spherical coordinates to do so.
Ω i j
2 2 2
We then have 6 combinations of xi xj to investigate:
RRRx1 , x2 , x3 , x1 x2 , x1 x3 , and x2 x3 . It is easy to
show as before, by choosing i = 1 and j = 2, that Ω xi xj dx = 0 if i 6= j. Finally, for i = j = 1
ZZZ Z 1 Z π Z 2π
x21 dx = r4 cos(φ)2 sin(θ)3 dr dθ dφ
Ω r=0 θ=0 φ=0
ÇZ 1
å ÅZ π ã ÇZ 2π
å
4 2 1 + cos(2φ)
= r dr [1 − cos(θ) ] sin(θ) dθ dφ
0 0 0 2
ÇZ π ï 3 òπ å
π cos (θ)
= sin(θ) dθ +
5 0 3 0

= .
15
This result being correct as well for Ω x22 dx and Ω x23 dx, we have
RRR RRR

ZZZ

xi xj dx = δij
15
ZZ Ω

=⇒ ni nj nk nl dx = (δil δjk + δjl δik + δkl δij ). ♣ ... ( 10 points)
∂Ω 15

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy