Chapter 3

Quantum Mechanics and First-Principles

Molecular Dynamics Selection of Polymer
Sensing Materials

Mario Blanco, Abhijit V. Shevade, and Margaret A. Ryan

Abstract We present two first-principles methods, density functional theory (DFT)

and a molecular dynamics (MD) computer simulation protocol, as computational
means for the selection of polymer sensing materials. The DFT methods can yield
binding energies of polymer moieties to specific vapor bound compounds, quan-
tities that were found useful in materials selection for sensing of organic and
inorganic compounds for designing sensors for the electronic nose (ENose) that
flew on the International Space Station (ISS) in 2008–2009. Similarly, we present
an MD protocol that offers high consistency in the estimation of Hildebrand and
Hansen solubility parameters (HSP) for vapor bound compounds and amorphous
polymers. HSP are useful for fitting measured polymer sensor responses with
physically rooted analytical models. We apply the method to the JPL electronic
nose (ENose), an array of sensors with conducting leads connected through thin
film polymers loaded with carbon black. Detection relies on a change in electric
resistivity of the polymer film as function of the amount of swelling caused by the
presence of the analyte chemical compound. The amount of swelling depends upon
the chemical composition of the polymer and the analyte molecule. The pattern is
unique and it unambiguously identifies the compound. Experimentally determined
changes in relative resistivity of fifteen polymer sensor materials upon exposure to
ten vapors were modeled with the first-principles HSP model.

M. Blanco (*)
Division of Chemistry and Chemical Engineering, California Institute of Technology, BI 139-74,
Pasadena, CA 91125, USA
e-mail: mario@wag.Caltech.edu
A.V. Shevade and M.A. Ryan
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA

M.A. Ryan et al. (eds.), Computational Methods for Sensor Material Selection, 71
Integrated Analytical Systems,
DOI 10.1007/978-0-387-73715-7_3, # Springer ScienceþBusiness Media, LLC 2009
72 M. Blanco et al.

3.1 Introduction

In the context of sensor development work, the main objective of molecular

modeling is to provide theoretical support for sensor material selection. Material
selection for the JPL electronic nose (ENose) made use of computer modeling. The
JPL (ENose) is an array of polymer–carbon composite sensors designed to monitor
breathing air quality for chemical contaminants aboard space shuttles/space stations
[1–6]. The JPL ENose is under development as an array–based sensing system,
which can run continuously and monitor for the presence of toxic chemicals in the
air in near real time. The sensing array in the JPL ENose is made from polymer–
carbon composite sensing films. The previous two generations of the JPL ENose
have successfully detected organic contaminants at parts-per-million (ppm) to sub
ppm concentration levels. In general, the methods outlined here are applicable to
amorphous materials principally made of individual monomer units, synthetic or of
biological origin.
Methods that correlate polymer–carbon composite sensor response with molec-
ular descriptors [4, 6] are quite helpful. These quantitative structure property
relationships (QSPR) models require experimental measurements to create a pre-
dictive model. In addition, there are limitations regarding predictions outside the
initial training set.
Ideally, a first principles approach can be used to predict the interactions
between any vapor bound analyte and individual or collective organic moieties
built into a polymer sensor, without the need to perform experiments. These
predictions can then be fed back into the experimental effort to optimize the
material selection process. This is the approach we take in this chapter.
Although direct computational estimates of the energetics of analyte/polymer
interactions with simulated bulk samples of the full polymer environment could
eventually become quite useful in this field [7], we have chosen to focus on
a methodology that involves calculating binding energies for individual organics-
analyte binary systems, including common classes of organic compounds such as
alkanes, alkenes, aromatics, primary, secondary and tertiary amines, aldehydes,
carboxylic acids, and esters. We have also found useful to model the physical
properties of the polymer, particularly Hansen and Hildebrand solubility para-
meters (HSP) as critical factors in creating predictive algebraic models [8]. How-
ever, the direct estimates of the free energy of binding of analytes in simulated
polymer bulk samples await the application of more advanced methods [9].

3.1.1 First Principles Modeling

First-principles modeling refers to the estimation of physical properties without

the use of experimental information. Typically this involves the solution of the
Schrödinger equation (quantum mechanics) or if a suitable force field (an algebraic
representation of the energy of the molecules involved as a function of geometry)
3 Quantum Mechanics and First-Principles Molecular Dynamics Selection 73

is available it might also involve solutions to Hamilton’s equations of motion

(Newtonian classical molecular dynamics). First principle properties that are now
available for computation include electronic energies, binding or interaction ener-
gies and entropies, electronic charge distributions, dipole moments, dielectric
properties, swelling, densities, free volume distributions, solubility parameters,
surface tension, and diffusion coefficients.
Quantum interaction energies are typically calculated using quantum mechanics
(QM) density functional theory (DFT) to solve the Schrödinger equation for all
electrons in the system:

HC ¼ EC (3.1)

where H is the Hamiltonian of the electronic degrees of freedom (all atoms in the
system).

X h2 X h2 X X Zk Zl e2 X X e2 X Zk e2
H¼ r2  r2 þ þ  :
k
2Mk i
2mi k l>k
Rkl r
i j>i ij kj
rkj
(3.2)

Here, k, l run over all the nuclei, and i, j over all electrons. The first two terms give
the kinetic energy of the nuclei and electrons, respectively, and the third term yields
the repulsion between the atomic nuclei, while the fourth and fifth terms are the
most difficult terms to solve, representing the repulsion of all electrons and the
attraction between electrons and nuclei, respectively.
We recommend B3LYP for routine calculations [10, 11]. For DFT methods that
might offer advantages in estimating dispersion (van der Waals) interactions
we refer the reader to the X3LYP method [12] or to the most current M06 suite
of DFT functionals [13–16]. Often it is necessary to use basis superposition error
(BSSE) corrections in these calculations. Superposition errors arise from the use of
a finite basis set. The number of functions that represent the electron densities must
be manageable, an unavoidable process in estimating numerical solutions of the
Schrödinger equation. The polymer moiety and the analyte are represented with
different basis sets; particularly the locations of the centers of these basis functions
are different. To avoid errors coming from these differing basis sets we include in
the calculation of the polymer moiety the basis functions (as ghost functions, not the
electrons or nuclei, just the numerical basis functions) for the analyte and vice
versa. Thus, both energy calculations use exactly the same basis set for the
combined polymer/analyte calculation and the error is avoided.
The quantum calculations yield mainly three types of sensor useful information:
1. The relative energies of polymer/analyte together vs. polymer and analyte
(separately), which measure the binding energy of the pair
2. The atomic charges (typically Mulliken population charges) for each atom
3. The energy as a function of polymer/analyte distance at various orientations
74 M. Blanco et al.

These results are useful in evaluating the potential use of a given material for
detection of a particular analyte. In addition, these results are also of practical
importance in creating an accurate “force field” that can be employed to represent
the energies, forces, and molecular dynamics of analyte/polymer interactions using
classical (Newtonian) dynamics. We refer to the process of using quantum mechan-
ically derived information to build a suitable force field for molecular dynamics
also as first principles modeling because no experimental data are required.

3.2 Identifying Chemical Functionalities in Polymers

for Analyte Detection: SO2 Polymer Sensor Example

The target compound list for the third generation JPL ENose, now undergoing
deployment in the International Space Station (ISS), included a set of inorganic
compounds (notably mercury, Hg, and sulfur dioxide, SO2) in addition to a subset
of previously detected organics (Table 3.1).
We use B3LYP DFT calculations [10, 11] to estimate binding energies of typical
organic functionalities present in homo- and copolymers of interest (see Table 3.2)
with these analytes. Subsequently the QM results were used to develop a first
principles force field for use in the calculation of interaction energies of SO2
molecules with various organic functional groups. As an example, the quantum
binding energies for methylamine/SO2 system at various scan distances, measured
between the sulfur atom in SO2 and the nitrogen atom in methylamine are shown in
Table 3.1. Scan distances from 1.5 to 6 Å in increments of 0.1 Å were chosen in all
cases. In most cases the molecular orientation was fixed around the most optimal
approach found with a generic force field [17], using Mulliken population quantum
charges, and a docking algorithm previously developed by one of the authors [18].
The QM binding energy for the methylamine/SO2 system is 10.7 kcal mol1. In
this case BSSE error corrections are quite small, on the order of 0.1 kcal mol1.
Nonetheless, we included them in all energy calculations. The molecules are shown
in the optimal, fixed, molecular approach orientation (Fig. 3.1).

Table 3.1 Calculated analyte properties for third generation JPL ENose
Hansen Solubilities (cal cm3)1/2 Molecular volume (Å3)
Electrostatics Dispersion H-bond
Acetone 4.79 7.53 0 65.7
Ammonia 6.66 3.83 8.31 22.2
Dichloromethane 2.31 8.75 0 57.8
Ethanol 5.72 6.82 4.33 53.2
Freon 218 (C3F8) 1.5 5.18 0 104.1
Mercury 0 20.11 0 15.5
Methanol 7.33 6.16 5.2 35.9
2-Propanol 4.42 7.14 3.53 71.9
Sulfur Dioxide 19.07 6.95 0 40.7
Toluene 0.99 8.92 0 98.98
Table 3.2 Polymer candidates for the Third Generation JPL ENose
Cohesive energy components Hansen solubilities
Polymer Electrostatic Dispersion H-Bond Electrostatic Dispersion H-Bond
(cal cm3) (cal cm3)1/2
Poly(4-vinyl phenol) 5.06 46.78 3.24 2.25 6.84 1.80
Methyl vinyl ether/maleic acid, 50/50 22.05 29.65 6.88 4.70 5.45 2.62
Poly(styrene-co-maleic acid) 7.62 29.99 2.65 2.76 5.48 1.63
Polyamide resin 7.32 45.51 1.46 2.71 6.75 1.21
Poly(N-vinyl pyrrolidone) 16.73 41.36 0 4.09 6.43 0.00
Vinyl alcohol/vinyl butyral, 20/80 15.39 52.29 2.54 3.92 7.23 1.59
Ethyl cellulose 8.97 27.05 0.14 2.99 5.20 0.37
Poly(2,4,6-tribromostyrene), 66% 0.55 33.37 0 0.74 5.78 0.00
Poly(vinyl acetate) 22.31 29.17 0 4.72 5.40 0.00
Poly(caprolactone) 20.92 69.12 0 4.57 8.31 0.00
Soluble polyimide, Matrimid 7.27 54.89 0 2.70 7.41 0.00
Poly(epichlorohydrin-co-ethylene oxide) 10.48 56.7 0 3.24 7.53 0.00
Poly(vinylbenzyl chloride) 1.81 48.07 0 1.35 6.93 0.00
Styrene/isoprene, 14/86 ABA block polymer 1.69 39.08 0 1.30 6.25 0.00
3 Quantum Mechanics and First-Principles Molecular Dynamics Selection

Ethylene-propylene diene terpolymer 0.57 55.77 0 0.75 7.47 0.00

Polyethylene oxide, PEO100 18.38 72.61 0 4.29 8.52 0.00
75
76 M. Blanco et al.

Fig. 3.1 Binding energy as a function of the distance between analyte (SO2) and a potential
polymer moiety (methylamine). Insert shows the varied distance (white arrow) between the
nitrogen in methylamine and sulfur in SO2

Figure 3.2 shows B3LYP DFT results for a variety of other moieties of interest
and SO2. Binding energies range between 0.0 and 15.0 kcal mol1 (customarily we
drop the sign when referring to these energies as “binding” energies). PL1 refers to
an approach from above the plane of the molecule, parallel to the plane of
symmetry, which appeared more stable than an in plane approach for both benzene
and ethylene (not shown). An axial approach to ethylene was also tried and found
even less favorable. We note that aliphatic moieties (ethane, ethylene) as well as
aromatics (benzene) are nonbinding. These results exclude polymers that contain
ethylene and styrene as their main constituents for the detection of SO2. On the
other hand, amine-containing polymers, particularly secondary and tertiary amines,
are predicted to have good binding energies (ca. 10–15 kcal mol1), suitable for
their reversible detection. Secondary cyclic amines, such as methylpyrrolidone, and
carboxylate moieties, including formic acid, formaldehyde, and formamide, are
predicted to be intermediate in binding (between 4 and 8 kcal mol1).
The fundamental understanding on the basis of the quantum mechanical molecu-
lar interactions between various organic moieties/SO2 systems was useful to priori-
tize the selection of polymers sensor materials for the JPL ENose for SO2 detection.
Two polymers were selected and made into polymer carbon black composite
sensors [19]. These two polymers are both poly-4-vinyl pyridine derivatives with
a quaternary and a primary amine. The polymers were designated EYN2 and
EYN7; the structures are shown in Fig. 3.3. The polymers were synthesized from
poly-4-vinyl pyridine and made into polymer-carbon composite sensing films using
3 Quantum Mechanics and First-Principles Molecular Dynamics Selection 77

Fig. 3.2 Binding energy as a function of distance between analyte (SO2) and a host of potential
polymer moieties

protocols which have been previously described [1, 2]. These films were loaded
with 8–10% carbon by weight and solution deposited onto micro-hotplate sensor
substrates with a sensor area of 200 mm by 200 mm (4  10–8 cm2). The baseline
resistance of each sensor was 10 kO.
Previously tested polymers, shown in Fig. 3.4, yielded less optimal changes in
resistance, on the order of less than 2% at higher (e.g.,15 ppm) concentrations. The
molecular level predictions (weak/strong binding) when compared to experimental
sensor responses follow closely the same order:
Tertiary amine > primary amine > carboxylate > aliphatic polyimide > poly-
amide resin > polycaprolactone > ethylene–propylene.
New sensor responses are shown in Fig. 3.5. Primary and cyclic amines provide
sufficient binding energy for the detection of SO2 in the gas phase, as predicted.
Changes in base line resistance are on the order of 6% at 9 ppm SO2 concentrations.
On the basis of the quantum binding energy predictions, even higher sensitivities
would have been obtained if secondary and/or pure tertiary amines (without sur-
rounding carbonyl moieties as is the case in polyimide) had been used in the
polymer sensor synthesis.
78 M. Blanco et al.

Fig. 3.3 Monomer structures for the preparation of polymers and co-polymers selected for SO2
detection

Fig. 3.4 Sulfur dioxide response of six polymer sensors: 1 polyimide, 2 polyamide resin,
3 polycaprolactone, 4 ethylene-propylene, 5 poly 4-vinylphenol, and 6 polyvinyl acetate
3 Quantum Mechanics and First-Principles Molecular Dynamics Selection 79

Fig. 3.5 Response of two sensors, both made from polymer EYN2, to 0.2–9 ppm SO2 in air

3.3 Molecular Dynamics Principles for Modeling Polymer

and Analyte Bulk Properties

Following the quantum mechanical calculations, which often yield enough data to
make informed decisions for sensing materials, we pursue the modeling of the prop-
erties of the polymer in the bulk. We hope to achieve a first-principles understand-
ing of the interactions that give rise to the full gamut of molecular interactions with
the analyte. We use molecular dynamics (MD), solving the classical Hamilton’s
equations of motion. Beginning with the Lagrangian,
L ¼ T  VðqÞ: (3.3)

X @H @H @L
H¼ pi q_ i  L; ¼ p_ i ; ¼ q_ i ; pi ¼ : (3.4)
j
@qi @pi @ q_ i

Here q represents the atomic coordinates p their momenta, and T the kinetic energy
of all the atoms (here we no longer deal with individual electrons). V(q) is the
potential energy function, an algebraic expression often referred to as the “force
field”, giving the energy of the molecules for a given set of geometric positions of
their constituent atoms. V(q) should be on the basis of solutions to the Schrödinger
equation, as much as possible. In practice a generic force field is used for valence
terms (covalently bound atoms) while the nonbond interactions (electrostatics,
80 M. Blanco et al.

hydrogen bonds, and van der Waals or dispersion interactions) are calculated from
quantum mechanical scans as those shown in Sect. 3.2. If the state (coordinates and
momenta) of the system is well known at time t, we can find the position and
velocity of all atoms at time t + dt by integration of the partial differential (3.4). For
this we typically use the Verlet algorithm:
_
pðtÞ
qðt þ DtÞ ¼ 2qðtÞ  qðt  DtÞ  Dt2 : (3.5)
m

It gives accurate results to O (Dt4). For a full description of MD the reader is

referred to standard molecular dynamics textbooks [21, 22].
The experimentally determined changes in relative resistivity, DR/R, of the
polymer sensors upon exposure to vapors have been correlated with computer
calculated Hansen solubility parameters [9].
" #
X
3  s p
DR=R ¼ R0 expðgVs Þexp bi di  di : (3.6)
i¼1

gVs is the activation energy of diffusion of the solute in the polymer, proportional to
the molar volume of the odorant, Vs. The exponential factor g is a best-fit parameter.
We base this relation on the experimental observation that the diffusion coefficient
of various molecules is linearly related to the molar volume of the solute above the
glass transition temperature (Tg) of the polymer [22]. dsi are the Hansen solubility
parameters (HSP) of the solvent s, where i = 1, 2, and 3 and refer to the electrostatic,
dispersion, and hydrogen bond components, respectively. Similarly, dpi is the ith
HSP component of the polymer sensor p. The exponential coefficients bi are treated
as best fit parameters and so is the pre-exponential term R0. It should be noted that
we preserve the sign of the energy components in (3.6), which is usually lost in the
definition of Hansen and Hildebrand parameters. This is important because such
interactions can be attractive or repulsive, depending on the polymer/odorant
mixture in question.
The Hildebrand solubility parameter for a pure liquid substance is defined as the
square root of the cohesive energy density.

d ¼ ½ðDHv  RTÞ=Vm 1=2 ; (3.7)

DHv is the heat of vaporization and Vm the molar volume. RT is the ideal gas pV
term and it is subtracted from the heat of vaporization to obtain the energy of
vaporization. Typical units are as follows:
1 hildebrand = 1 cal1/2 cm3/2 = 0.48888  MPa1/2 = 2.4542  10–2 (kcal
mol1)1/2 A3/2
Hansen [23] proposed an extension of the Hildebrand parameter to estimate the
relative miscibility of polar and hydrogen bonding systems:

d2 ¼ d2d þ d2p þ d2h ; (3.8)

3 Quantum Mechanics and First-Principles Molecular Dynamics Selection 81

where, dd, dp, and dh are the dispersion, electrostatic, and hydrogen bond compo-
nents of d, respectively. For molecules whose heats of vaporization can be
measured, or calculated, one can easily determine the value of d. The Hansen
solubility parameters in (3.8) are typically determined empirically on the basis of
multiple experimental solubility observations, including observed solubility or
swelling with a series of solvents, NMR and IR signals, elution times in chro-
matographic columns, etc. The reported values in the literature can vary over a large
range, however, owing to the multiple experiments used to determine these com-
ponents. Instead, we rely on a MD protocol to estimate the HSPs.

3.4 First-Principles Force Field

The MD protocol is defined as a sequence of individual modeling tasks which begin

with the proper selection of a force field. The energy of the polymers, V(q) in (3.3),
is represented as a sum of valence and non-bond terms as follows:
X X X X
VðqÞ ¼ Vr þ Vy þ V’ þ Vcoul ðRij ; qij Þ
bonds angles torsions charges
X X
þ Vcoul ðRij ; qij Þ þ VHbond ðRij Þ: (3.9)
charges O;N;S

Valence force field: we use a generic force field, Dreiding [17] to estimate valence
terms. Bond stretch terms are given by

1
Vr ¼ Kr ðR  R0 Þ2 : (3.10)
2

Bond angle distortions are given by

1 Ky
Vy ¼ ðcosy  cosy0 Þ2 (3.11)
2 sin2 y0

and dihedral angle torsions by

1
V’ ¼ Vð1  d cos3fÞ: (3.12)
2

We also approximate the noncovalent hydrogen bond term, which runs over
heteroatoms such as O, N, and S, using the published Dreiding Lennard-Jones
12–10 potential form
"   10 #
R0 12 R0
VHbond ðRij Þ ¼ D0  : (3.13)
Rij Rij
82 M. Blanco et al.

We employ D0 ¼ 3.2 kcal mol1 and R0 ¼ 2.5 Å, which give better agreement with
the heat of vaporization of water (580 cal cm3) than the published Dreiding
hydrogen bond values when MP2/6 – 31g** atomic charges are used for the isolated
water molecule [q(O) ¼ 0.72866, q(H) ¼ 0.36433].
Electrostatics: all electrostatics interaction pairs were included in the calculation
without the use of cutoffs or spline functions.
X X qi qj
Vcoul ðRij Þ ¼ c : (3.14)
j i>j
Rij

Quantum charges, qi, were calculated at the quantum optimized (minimum energy)
geometry of each molecule (polymer segment and analytes) using polarized
Mulliken charges at the minimum conformation. We use the Jaguar suite of
programs [24], B3LYP DFT with a good (6 – 31g**) basis set. e ¼ 1 is the dielectric
constant, qi and qj are the atomic charges in electron units, and Rij is the distance
in Å; c ¼ 322.0637 converts the electrostatic energy to kcal mol1. For SO2,
the Mulliken population charges are q(S) ¼ 0.83914 and q(O1, O2) ¼ 0.41957
electrons.
The noncovalent terms in (3.9) play a crucial role in determining the structure of
the polymer and the HSP values for both polymer and analyte. Thus, the force field
is further refined for the dispersion (van der Waals) noncovalent interactions.

3.4.1 Refinement of van der Waal Interactions

The van der Waals interactions can be as important in determining the binding
energetics of analyte/polymer sensor as much as the electrostatics, particularly
when nonpolar chemical groups are involved at either end.
The binding energies for molecular pairs in Fig. 3.2, typical organic molecules
interacting with SO2, were calculated as follows:

Ebind ¼ EAB  ðEAB  EBA Þ: (3.15)

Here Ebind is the calculated energy of the AB pair, and EAB* and EBA* are the
energies of molecules A and B. These energies are calculated with the basis set of B
and A present (B*, A*, respectively); the process described above is basis set
superposition error correction (BSSE). As mentioned above we carried out a
distance sweep between the organic molecule and SO2 calculating the quantum
energy of interaction between 1.5 and 6 Å in increments of 0.1 Å. In some cases,
e.g., ethane, ethylene, and benzene we used various directions for the sulfur dioxide
approach to the organic compounds, such as parallel (PL) or perpendicular (PR) to
the molecular axis of symmetry. The binding energies exclude all covalent inter-
actions (as the individual molecular energies have been subtracted) but include all
3 Quantum Mechanics and First-Principles Molecular Dynamics Selection 83

nonbond terms, i.e., electrostatics, hydrogen bond, and dispersion. After subtraction
of the electrostatic terms we fit the remaining binding energy with a Morse potential:

VðRij Þ ¼ De ð1  exðRij Re Þ Þ2  De ; (3.16)

where De represents the van der Waals contribution to the total binding energy at
the equilibrium distance Re and x is related to the harmonic force constant k
associated with the noncovalent binding mode by
rffiffiffiffiffiffiffiffiffiffi
k
x¼ : (3.17)
2De

Rij is the distance between the two atom centers. Because elements have different
environments we need to estimate a set of van der Waals dispersion parameters for
each atom type pair. Thus, an sp2 carbon, such as a carbon in ethylene, will have its
own specific set of values (De, Re, and x). The process is somewhat involved, but for
each geometry present in the distance sweep between the analyte and the organic
moiety all the interatomic distances are recorded and used to fit the Morse para-
meters after subtraction of the electrostatic Coulomb energy. A full least square
procedure on all Morse parameters combined is carried out.
The force field represents the calculated quantum binding energies quite well. We
obtained the force-field energy curves shown in Fig. 3.6 for the secondary amine
with SO2. Small open circles are the quantum binding energies given by (3.15)
Tables 3.3a and 3.3b contain the set of (De, Re, and x) for each type of atom pairs
involving SO2 and for the atom types present in the organic functionalities included

Fig. 3.6 Quantum binding energies and the force field van der Waals and Coulomb components
for dimethylamine interacting with SO2
84 M. Blanco et al.

Table 3.3a van der Waals force field parameters determined from DFT (B3LYP) binding ener-
getics. Morse parameters for sulfur (S_3) Dreiding atom type in SO2 with other atom types present
in ten organic compounds
Atom types Hybridization Re De x
C_3 sp3 3.132617 0.749458 11.99488
C_2 sp2 2.737065 1.340003 13.17681
C_R sp2 aromatic 5.344190 0.002262 14.72481
H_$ sp3 3.973471 0.029141 12.00958
$
H_ sp2 2.961596 2.245861 13.49589
H_Aa 3.499816 0 11.99997
H_Ab 3.954983 0.000694 12.00027
H_R sp2 aromatic 3.077340 0.193491 13.22020
N_3 sp3 2.230163 8.252832 11.92240
N_R sp2 aromatic 3.498603 0 11.91701
O_2 sp2 3.459701 0.000602 11.81915
O_R sp2 aromatic 5.089213 0.167158 12.62331
a
Attached to heteroatoms (e.g., O)
b
Attached to N atom only

Table 3.3b Morse parameters for oxygen atoms of SO2 (O_2) with common Dreiding atom types
present in ten organic compounds
Atom types Hybridization Re De x
3
C_3 sp 3.450184 0.000471 11.99898
C_2 sp2 4.053726 0 11.07503
C_R sp2 aromatic 3.485616 0 12.00207
H_$ sp3 3.900214 0.013218 12.03888
H_$ sp2 2.822144 0 12.03854
H_Aa 4.001108 0 13.15887
H_Ab 3.969541 0.296846 11.84432
H_R sp2 aromatic 3.028448 0.427043 12.66140
N_3 sp3 3.345748 0.846320 11.87265
N_R sp2 aromatic 3.499287 0 13.63104
O_2 sp2 3.728343 0.106303 15.86427
O_R sp2 aromatic 2.999365 0 12.37445
a
Attached to hetero-atoms (e.g., O)
b
Attached to N atom only

in this study. A brief glance at this table shows that the sp3 Nitrogen atom in the amine
interacts the strongest (De = 8.25 kcal mol1) with the sulfur in sulfur dioxide.

3.5 Molecular Dynamics Protocol for Polymer Sensor Responses

As shown in the previous section quantum mechanics can provide sufficient infor-
mation to guide material sensor selection successfully. A wealth of information is
generated in the process, which includes the force field parameters in Tables 3.3a
3 Quantum Mechanics and First-Principles Molecular Dynamics Selection 85

and 3.3b. These parameters can be used to estimate polymer and analyte material
properties that can provide the basis for developing predictive models of sensor
responses, such as the model given by (3.6). Thus, in this section we aim at
estimating without experimental input Hansen solubility parameters. The most
common problem in computer simulations of polymers is the long time required
to obtain an equilibrated simulated sample. This is a common problem with
amorphous condensed phases. We have developed a method that overcomes the
common equilibration problems with condensed phase molecular dynamics, i.e.,
how to choose initial molecular configurations not far from equilibrium at normal
densities. Significant amounts of simulation time are usually required to equilibrate
the initially random packed molecules often generated with Monte Carlo methods.
In particular, densely packed simulated polymers often lead to highly nonequili-
brated dihedral populations. Thus, care must be taken to generate an ensemble of
thermally accessible conformations not far from equilibrium. These two require-
ments, condensed phase densities and equilibrated molecular conformations, are
satisfied through the following MD protocol:
1. A cubic periodic unit cell containing a given number of molecules is built at
a low density, rlow, typically 50% of the target density. Generally four
polymer chains are sufficient, although for very high molecular weights
even one chain can be adequate. For solvents 16–64 solvent molecules are
adequate. We find that for packing the structure, it is useful to scale van der
Waals radii by a factor of 0.30 to get initial structures that will eventually lead
to a good ensemble. In cases where the compounds are polymers, or a
molecule with a large number of torsional degrees of freedom, we use the
Amorphous Builder in Cerius2 [25] to create the initial low-density sample.
The initial polymer amorphous structures are constructed using the rotational
isomeric state (RIS) table [26] and a suitable Monte Carlo procedure to
achieve a correct distribution of conformational states in the low-density
sample. The Amorphous Builder converts an existing model into an amor-
phous structure by manipulating the model’s rotatable bonds. Each unique
torsion can be defined using a Monte Carlo procedure with statistical weights
given by a previously built rotational isomeric state table determined with
well established molecular mechanics dihedral sampling procedures. Confor-
mations are rejected if two or more atoms come closer than a van der Waals
scale distance. In polymer calculations, the number of monomers in each
chain is usually determined such that the total volume of the four chains is
at least 6,000 Å [3]. Alternatively, a degree of polymerization of 30 suffices to
give values comparable to those from experiment. In such polymer samples,
the minimum number of atoms is at least 1,000. Larger samples are recom-
mended whenever possible.
2. For convenience we used the experimental densities of the solvents and
polymers as target values as these are commonly available in the literature.
For liquid systems with unknown densities we typically run a preliminary MD
calculation with a rough “trial” density, such as that predicted from group
86 M. Blanco et al.

additivity methods, to obtain a good initial estimate. The procedure below will
increase the density to a maximum, rhigh, typically 125% of the target density.
The resulting amorphous structure is then relaxed, resulting in a predicted
target density for the start of the definitive MD calculation.
3. The charges of the isolated solvent or polymer molecules are defined using
those obtained from quantum mechanical calculations (Mulliken population
charges) as previously explained.
4. The force field parameters are taken from a generic force field, such as the
generic Dreiding force field with modifications of the van der Waals para-
meters for higher accuracy.
5. Minimization: The potential energy of the bulk system is minimized for M
steps, typically M = 5,000 steps, or until the atom rms force converges to 0.10
kcal (mol Å)1, whichever comes first.
6. Annealing dynamics to allow the structures to equilibrate typically 750 steps of
MD (1 fs per step) at high temperature (typically between 400 and 800 K, with
700 K generally adequate) using canonical fixed volume dynamics (NVT) are
carried out to anneal the sample.
7. Compression: The reduced cell coordinates are shrunk such that the density is
increased by (rhighrlow)/N, where N is typically 5.
8. The atomic coordinates are minimized and dynamics run on the system with
the previously described procedure holding the cell fixed (steps 5–6).
9. A total of N compression, minimization, and dynamics cycles are performed
until the density reaches rhigh, typically 125% of the target density, steps 5–8.
10. The cell parameters are then increased in N cycles of expansion, minimization,
and dynamics, until the target density is reached
11. The sample is allowed to relax in M steps of minimization allowing both the
cell and the atomic coordinates to relax.
12. Molecular dynamics are performed for a time to thermalize and then to
measure properties. Typically we perform these for as few as 20 ps but longer
times are recommended for high molecular weight compounds. The first 10 ps
are used for thermalization of the sample at the desired temperature. The last 10
ps are used for averaging of cell volume and potential energy components: van
der Waals (dispersion), electrostatic (polar), and hydrogen bonding.
13. The Hansen enthalpy components are calculated by subtracting the potential
energy of the bulk system from the sum of the potential energies of the
individual molecules in vacuum.

X
n  
d2k ¼ Eki  Ekc =N0 hVc =ni: (3.18)
i¼1

14. Here < > indicates a time average over the duration of the dynamics, n the
number of molecules, k ¼ 1,2,3 for coulomb (polar), van der Waals (disper-
sion), and hydrogen bond components, and N0 is Avogadro’s number.
15. This process is repeated P times with different initial random conformations
and packing. Typically P ¼ 10 is adequate but higher values are recommended.
3 Quantum Mechanics and First-Principles Molecular Dynamics Selection 87

Fig. 3.7 A polymer or solvent sample is put through a series of compression and expansion steps
until the proper density and packing are obtained. On the left the initial density is 40% of the target
density. After compression, second step, the sample is overcompressed by 20%. Finally the sample
is allowed to relax. Through NPT molecular dynamics a final prediction of the density and
cohesive energy of the sample is obtained. The process is repeated for several samples and
statistics are gathered

16. Hansen solubility parameters and molar volumes are computed as well as their
standard deviations. We use the 95% confidence limit of an F statistical
distribution test, with two standard deviations from the average value, to
identify outliers. Typically a P ¼ 10 sample run will have no outliers; more
than two outliers are rare.
The overall procedure is schematically illustrated in Fig. 3.7. Tables 3.1 and 3.2
show the results of this procedure, the Hansen solubility parameters for analytes,
and polymer candidates for the JPL ENose.
At this point we introduce experimental information to complete the model
in (3.6). Once these free parameters in (3.6) are determined we can employ this
expression to make sensor response predictions for new analytes, not initially
included in the training set. Tables 3.1 and 3.2 contain the calculated (using
the MD protocol outlined above) Hansen solubilities for analytes and polymers,
respectively. Figure 3.8 shows the experimental vs. predicted sensor responses for
some typical polymers. Figure 3.9 shows the model predictions vs. experimental
data for some of the polymer sensors.
We next tested the predictive power of (3.6) by calculating the responses of the
JPL ENose sensor array to an analyte not originally included in the training set. The
results are shown in Fig. 3.10. There is good agreement between the model’s
prediction and the measured JPL ENose responses to Freon113. This is an impor-
tant test because it shows that these model have true predictive power, beyond
simple data regression or statistical correlations.

3.6 Discussion and Conclusions

We have presented a first-principles approach to materials selection for polymer

sensors for specific analytes. The quantum (B3LYP/DFT) binding energies
of analyte with a few representative organic moieties can yield a great deal
88 M. Blanco et al.

Fig. 3.8 Calculated (3.6) vs. experimental sensor responses to four analytes in the third generation
JPL ENose compound list

of information. The quantum binding energies, as shown in the sulfur dioxide

example, are strongly correlated with the measured responses, and therefore, the
quantum calculated energy data can significantly aid the selection of sensing
materials. Furthermore, the data can be employed to successfully model the self-
interactions, in the form of Hansen solubility parameters, of analytes and polymers.
These first-principles MD-derived solubilities can be used to model the responses
of an array of polymer sensors to a wide variety of analytes and provide the basis
for extrapolations outside the original training set.
Table 3.4 shows the values of the fitting parameters in (3.6), obtained using
a least square fit of the experimental data and the calculated Hansen parameters in
Tables 3.1 and 3.2 for each of analyte and polymer sensor pairs in the study. Over
half of the polymer sensors follow (3.6), with a Pearson’s correlation coefficient
above 0.75. Clearly there are some sensors that deviate significantly from this
equation, particularly copolymers such as methyl vinyl ether/maleic acid (50/50)
and ethyl cellulose. Overall, the model represents the data with an average
Pearson’s correlation coefficient of R2 ¼ 0.73. In a previous study, the experimental
responses of a different set of polymer sensors for the ENose had been fitted with
the same expression and methodology [8]. Comparison between theory, (3.6), and
experimental changes in resistivity of seven polymer sensors exposed to 24 solvents
3 Quantum Mechanics and First-Principles Molecular Dynamics Selection 89

poly(styrene-co-maleic acid) poly (vynyl acetate)

1.2 0.25
1
Calculated DR/R

Calculated DR/R
0.2
0.8
0.15
0.6
0.1
0.4
0.2 0.05
R2 = 0.8855 R2 = 0.8694
0 0
0 0.5 1 0 0.05 0.1 0.15 0.2 0.25 0.3
Experimental DR/R Experimental DR/R

poly (caprolactone) poly (styrene-isoprene)

0.3 1.2
0.25
Calculated DR/R

1
0.2 Calculated DR/R
0.8
0.15 0.6
0.1 0.4
0.05 R2 = 0.8547 0.2 R2 = 0.9647
0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.2 0.4 0.6 0.8 1 1.2
Experimental DR/R Experimental DR/R

Fig. 3.9 Calculated (3.6) vs. experimental sensor responses of four polymer sensors to ten analytes
in the third generation JPL ENose compound list

Fig. 3.10 Calculated (dark) vs. experimental (light) pattern of changes in resistivity by Freon113
on the JPL ENose array. Hansen solubility parameters for Freon113 are 2.82, 8.54, and 0 (cal per
cm3)1/2 for electrostatic, dispersion, and hydrogen bonding respectively. The molar volume is
110.91 Å3

gave a correlation coefficient R2 ¼ 0.89. The model was used to predict the response
of Freon 113, an analyte not present in the training set. The predicted pattern of
changes in resistivity agrees quite well with the measured polymer responses.
 90

Table 3.4 Fitting model parameters, (3.6), for the JPL Enose polymer sensors.
Polymera Ro g be bd bh Pearson’s R2
2 2 0 0 0
1 Poly(4-vinyl phenol) 9.4311  10 9.2379  10 1.4550  10 2.4467  10 1.1473  10 0.75
2 Methyl vinyl ether/maleic acid, 50/50 2.0412  101 6.4933  103 4.3703  102 2.7624  103 2.7005  101 1.10
3 Poly(styrene-co-maleic acid) 1.4201  101 1.6597  102 5.7604  101 5.6279  101 2.0015  102 0.89
4 Polyamide resin 3.6871  102 2.9980  102 5.9101  102 3.2918  101 2.0119  101 0.57
5 Vinyl alcohol/vinyl butyral, 20/80 1.9976  102 3.0485  102 1.1009  103 1.0789  100 6.9556  101 0.68
6 Ethyl cellulose 2.7476  102 3.8149  102 9.4130  101 2.5777  101 5.3688  101 0.48
7 Poly(2,4,6-tribromostyrene), 9.2849  100 2.1263  102 7.4222  102 4.0145  101 1.7336  101 0.58
66%
8 Poly(vinyl acetate) 2.6723  102 2.7333  102 2.0736  101 4.3585  102 2.4584  103 0.87
9 Poly(caprolactone) 1.3872  102 3.3264  102 1.4970  101 9.2264  102 1.7486  101 0.85
10 Soluble polyimide, Matrimid 6.3407  100 1.3727  101 4.9892  101 1.0439  101 2.3117  100 0.68
11 Poly(epichlorohydrin- 2.1315  102 1.1886  101 8.3737  102 5.1148  101 1.6265  101 0.64
co-ethylene oxide)
12 Poly(vinylbenzyl chloride) 1.1729  103 3.1200  102 1.5584  100 2.5691  100 5.7381  101 0.54
13 Styrene/isoprene, 14/86 ABA block polymer 9.7786  106 9.6770  102 4.2760  101 1.7732  103 4.1881  101 0.96
14 Ethylene-propylene diene terpolymer 4.6806  106 1.2221  101 4.1192  101 1.3660  100 3.2794  102 0.92
15 Polyethylene oxide, PEO100 4.4420  102 1.6183  102 1.4877  101 3.6562  102 2.9365  101 0.77
Averageb 0.73
a
Poly(N-vinyl pyrrolidone) data were not available. Sensor was dropped from experimental consideration
b
Excludes methyl vinyl ether/maleic acid
c
Pearson’s correlation coefficient ¼ 1 for a perfect correlation
M. Blanco et al.
3 Quantum Mechanics and First-Principles Molecular Dynamics Selection 91

Ideally, one might wish to predict directly the swelling of polymer sensors,
loaded with carbon black, and transform this into a change in resistivity. This is
however a great challenge for various reasons, from the ability to estimate the free
energy of mixing analytes and polymers, to good knowledge of the true chemical
composition of carbon black and how adsorbed molecules change its intrinsic
conductivity. Nonetheless, we have shown that a combination of quantum mechan-
ics with first-principles MD can afford a great deal of information that it is useful in
designing and selecting materials for specific analytes. Future work involves the use
of a newly developed MD method for the direct estimation of free energies [9].

Acknowledgments This work was supported in part by the Materials and Process Simulation
Center, Beckman Institute at the California Institute of Technology and by a grant from NASA.

References

1. Ryan, M. A.; Shevade, A. V.; Zhou, H.; Homer, M. L., Polymer-carbon black composite
sensors in an electronic nose for air-quality monitoring, Mrs Bull. 2004, 29, 714–719
2. Ryan, M. A.; Zhou, H. Y.; Buehler, M. G.; Manatt, K. S.; Mowrey, V. S.; Jackson, S. R.;
Kisor, A. K.; Shevade, A. V.; Homer, M. L., Monitoring space shuttle air quality using the jet
propulsion laboratory electronic nose, IEEE Sensors J. 2004, 4, 337–347
3. Zhou, H. Y.; Homer, M. L.; Shevade, A. V.; Ryan, M. A., Nonlinear least-squares based
method for identifying and quantifying single and mixed contaminants in air with an elec-
tronic nose, Sensors 2006, 6, 1–18
4. Shevade, A. V.; Homer, M. L.; Taylor, C. J.; Zhou, H. Y.; Jewell, A. D.; Manatt, K. S.; Kisor,
A. K.; Yen, S. P. S.; Ryan, M. A., Correlating polymer–carbon composite sensor response
with molecular descriptors, J. Electrochem. Soc. 2006, 153, H209–H216
5. Shevade, A. V.; Ryan, M. A.; Homer, M. L.; Kisor, A. K.; Manatt, K. S.; Lin, B.; Fleurial, J.
P.; Manfreda, A. M.; Yen, S. P. S., Calorimetric measurements of heat of sorption in polymer
films: A molecular modeling and experimental study, Anal. Chim. Acta 2005, 543, 242–248
6. Shevade, A. V., Developing sensor activity relationships for the JPL electronic nose sensors
using molecular modeling and QSAR techniques, 2005 IEEE Sensors (IEEE Cat.
No.05CH37665C) 2005, 4 pp.
7. Cozmuta, I.; Blanco, M.; Goddard, W. A., Gas sorption and barrier properties of polymeric
membranes from molecular dynamics and Monte Carlo simulations, J. Phys. Chem. B 2007,
111, 3151–3166
8. Belmares, M.; Blanco, M.; Goddard, W. A.; Ross, R. B.; Caldwell, G.; Chou, S. H.; Pham, J.;
Olofson, P. M.; Thomas, C., Hildebrand and Hansen solubility parameters from molecular
dynamics with applications to electronic nose polymer sensors, J. Comput. Chem. 2004, 25,
1814–1826
9. Lin, S. T.; Blanco, M.; Goddard, W. A., The two-phase model for calculating thermodynamic
properties of liquids from molecular dynamics: Validation for the phase diagram of Lennard-
Jones fluids, J. Chem. Phys. 2003, 119, 11792–11805
10. Becke, A. D., Density-functional thermochemistry. 3. The role of exact exchange, J. Chem.
Phys. 1993, 98, 5648–5652
11. Lee, C. T.; Yang, W. T.; Parr, R. G., Development of the Colle–Salvetti correlation-energy
formula into a functional of the electron-density, Phys. Rev. B 1988, 37, 785–789
12. Xu, X.; Goddard, W. A., The X3LYP extended density functional for accurate descriptions of
nonbond interactions, spin states, and thermochemical properties, Proc. Natl Acad. Sci. USA
2004, 101, 2673–2677
92 M. Blanco et al.

13. Zhao, Y., Development and assessment of a new hybrid density functional model for thermo-
chemical kinetics, J. Phys. Chem. A 2004, 108, 2715–2719
14. Zhao, Y., A density functional that accounts for medium-range correlation energies in organic
chemistry, Org. Lett. 2006, 8, 5753–5755
15. Zhao, Y., Comparative DFT study of van der Waals complexes: Rare-gas dimers, alkaline-
earth dimers, zinc dimer, and zinc-rare-gas dimers, J. Phys. Chem. A 2006, 110, 5121–5129
16. Zhao, Y., Density functionals with broad applicability in chemistry. Acc. Chem. Res. 2008, 41,
157–167
17. Mayo, S. L.; Olafson, B. D.; Goddard, W. A., Dreiding – A generic force-field for molecular
simulations. J. Phys. Chem. 1990, 94, 8897–8909
18. Blanco, M., Molecular silverware.1. General-solutions to excluded volume constrained pro-
blems. J. Comput. Chem. 1991, 12, 237–247
19. Ryan, M. A.; Shevade, A. V.; Taylor, C. J.; Homer, M. L.; Jewell, A. D.; Kisor, A. K.; Manatt,
K. S.; Yen, S. P. S.; Blanco, M.; Goddard III, W. A. In expanding the capabilities of the JPL
electronic nose for an international space station technology demonstration, In Proceedings of
36th International Conference on Environmental Systems 2006, 2006–01–2179, Norfolk,
Virginia, USA
20. Allen, M. P.; Tildesley, D. J., Computer Simulations of Liquids, Oxford University Press,
Oxford, 1987
21. Frenkel, D., Computer Simulation in Chemical Physics, Kjeuver, New York, 1993
22. van Krevelen, D. W., Properties of Polymers: Their Correlation with Chemical Structure;
their Numerical Estimation and Prediction from Group Contributions, Elsevier Science,
New York, 1990
23. Hansen, C. M., The three dimensional solubility parameter – Key to paint component affinities
I. – Solvents, plasticizers, polymers, and resins. J. Paint Technol. 1967, 39, 104–117
24. Jaguar, 7.207; Schrodinger, LLC, NY, 2007
25. Accelrys, I. Cerius2, 4.01, Accelrys, Inc.: San Diego, CA, 2005
26. Lin, S. T.; Blanco, M. Rotational Isomeric State Table Algorithm, California Institute of
Technology, Pasadena, CA, 2003