Anda di halaman 1dari 155

UNIVERSITY OF CALGARY

Molecular Modelling of Proton Transport and Structure in the Short-Side-Chain

Perfluorosulfonic Acid Polymer

by

Iordan Hristov Hristov

A THESIS

SUBMITTED TO THE FACULTY OF GRADUATE STUDIES

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE

DEGREE OF DOCTOR OF PHILOSOPHY

DEPARTMENT OF CHEMISTRY

CALGARY, ALBERTA

April, 2010

© Iordan Hristov Hristov 2010


UNIVERSITY OF CALGARY

FACULTY OF GRADUATE STUDIES

The undersigned certify that they have read, and recommend to the Faculty of Graduate

Studies for acceptance, a thesis entitled “Molecular Modelling of Proton Transport and

Structure in the Short-Side-Chain Perfluorosulfonic Acid Polymer” submitted by Iordan

Hristov Hristov in partial fulfillment of the requirements for the degree of DOCTOR OF

PHILOSOPHY.

Neutral Chair, Dr. Jürgen Gailer Dr. Arvi Rauk


Department of Chemistry Department of Chemistry

Supervisor, Dr. Reginald Paul “Internal” External, Dr. Barry Sanders


Department of Chemistry Department of Physics and Astronomy

Co-Supervisor, Dr. Stephen J. Paddison External, Dr. Raymond Kapral


University of Tennessee University of Toronto

Dr. Peter Kusalik


Department of Chemistry

Date
“Ere wa rudy fir tha beg shew?”

“Are we ready for the big show? Yes, Mr. Gotchchauk,

I certainly think we definitely have a good chance to be

almost completely ready.”

On the Air, ABC, 1992


Abstract

The subject of this thesis is the development of the necessary tools and their applica-

tion for a better understanding of the morphology and proton transport in the perflu-

orosulfonic acid (PFSA) short-side-chain (SSC) membranes. During recent years these

membranes have been the subject of enhanced interest as potential fuel cell electrolytes

replacing the relatively, better known Nafion membranes. In order to achieve these ends

we developed the mathematical formalism and the necessary algorithm for computing the

force fields that are unique to this material and which reproduce the available data for a

2-unit polymer. Furthermore, this algorithm allows the construction of three dimensional

polymeric structures possessing high molecular weights (MW) and specific morphologies

resulting by the imposition of optional restraints. The polymers thus built were placed

and equilibrated in a periodic simulation cell subject to periodic boundary conditions

(PBC). Special attention was given to the effects of PBC on the virial and the pressure

of the system in view of the fact that the experimentally measured densities of the hy-

drated polymers are not available. We were able to derive an analytic expression capable

of predicting the pore radii in the hydrated membranes as a function of the equivalent

weight (EW) and hydration level.

Having obtained systems with the correct morphologies and under suitable ambient

conditions our next goal was to investigate the transport of the proton through the

aforementioned hydrated pores. In the first instance we assumed the proton to exist

as a hydronium ion and thus enabling the motion of such a relatively heavy particle

to be studied by the application of a classical mechanical molecular dynamics (MD)

technique. The primary object of such a calculation is the diffusion coefficient since these

are experimentally available transport parameters. The results of our calculations showed

excellent agreement with experiment only in the case of the pores with low hydration,

iii
iv

however, at a higher level of hydration the predicted diffusion coefficients are too low

indicating the existence of an alternative more rapid mode of transport.

The obvious alternative mechanism is the well known hopping or Grotthuss mecha-

nism. The most convenient approach is to employ the Empirical Valence Bond (EVB)

method, where the force field is generated on the fly. The generation of the force field in

its most general form is, with the currently available electronic computational facilities,

a prohibitively expensive task since, in principle; it requires a quantum mechanical (QM)

calculation of the electronic energy at a continuum of points on the entire potential en-

ergy surface (PES). In this thesis we have developed a new methodology which we refer

to as the Just-In-Time EVB (JIT-EVB) method that requires such a computation to be

carried out on a much smaller grid of points. Such a computer program, that we have

developed, not only provides an important means for studying the transport of protons in

a highly acidic medium but has also enabled us to garner valuable structural insight. For

example we have found that an ion cage structure composed of sulfonate groups clamps

hydronium ions thereby impeding their diffusion rate.

The numerous advancements in the simulation methodology presented here are ex-

pected to result in significantly improved reliability of the simulations, allowing for accu-

rate structure-property modelling that will ultimately enable the targeted design of new

polymer systems.
Acknowledgements

I thank Dr. Paul for all his support and trust in me during the last six years. A

significant part of this work would have been impossible without the guidance of Dr.

Paddison, for which I am greatly indebted to him. Continuous financial support from

the Alberta Ingenuity Fund and the Natural Science and Engineering Research Council

is gratefully acknowledged.

v
Table of Contents

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Introduction to Fuel Cells . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Review of Key Experimental Work . . . . . . . . . . . . . . . . . . . . . 8
1.4 Review of Theoretical Work . . . . . . . . . . . . . . . . . . . . . . . . . 12

I Methodology Development 29
2 Force Field Development . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3 Virial Formulation For Periodic Systems . . . . . . . . . . . . . . . . . . 41
4 Just-in-Time Empirical Valence Bond Method . . . . . . . . . . . . . . . 55

II Polymer System Studies 68


5 Constructing The Polymer Systems . . . . . . . . . . . . . . . . . . . . . 69
6 Designer Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7 Morphology of Hydrated SSC Polymer Systems . . . . . . . . . . . . . . 92
8 Proton Diffusion in SSC Polymer Systems . . . . . . . . . . . . . . . . . 104

III Conclusions and Future Work 113


9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
10 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

IV Appendixes and Bibliography 116


A Short-Side-Chain Force Field . . . . . . . . . . . . . . . . . . . . . . . . . 117
B Cross Section Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
C Molecular Volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
D JIT-EVB Simulation Details . . . . . . . . . . . . . . . . . . . . . . . . . 124
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

vi
List of Tables

1.1 Density of Hydrated Nafion at 300 and 350 K . . . . . . . . . . . . . . . 22

3.1 Virial comparison for a 256 particle Lennard-Jones model . . . . . . . . . 53

5.1 Examples of random polymer sequences generated with different seeds . . 70

6.1 Pore radius as a function of the repeat unit formula (CF2 CF2 )n −CF2 CF −
(OCF2 CF2 SO3 H) and the hydration level λ . . . . . . . . . . . . . . . . 91

8.1 Diffusion coefficients obtained from a JIT-EVB simulation of excess proton


with 64 water molecules . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
8.2 Diffusion coefficients obtained from a JIT-EVB simulation of five triflic
acid molecules and 30 water molecules . . . . . . . . . . . . . . . . . . . 110

A.1 SSC force field parameters for the harmonic stretching potential Ebond . . 118
A.2 SSC force field parameters for the harmonic bending potential Eangle . . 119
A.3 SSC force field parameters for the torsion potential Edih . . . . . . . . . . 119
A.4 SSC force field parameters for the non-bonding interactions ECoulomb and
ELJ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

D.1 Force field parameters for the anharmonic stretching potential EM orse used
in JIT-EVB simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
D.2 Atom charges used in JIT-EVB simulations . . . . . . . . . . . . . . . . 125

vii
viii

List of Figures

1.1 Different types fuel cells . . . . . . . . . . . . . . . . . . . . . . . . . . . 4


1.2 PFSA based polymers for fuel cell membranes . . . . . . . . . . . . . . . 6
1.3 Conductivity plot for Nafion and the SSC polymer as a function of the
hydration level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Cluster-network model for the morphology of hydrated Nafion . . . . . . 9
1.5 Two rows of the hexagonal lattice formed by the polymer chains (shown
as circles) seen end-on . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Morphologies of Nafion at different volume fractions of water . . . . . . . 10
1.7 Schematic representation of the nano-phase separation in the hydrated
morphology of Nafion and sulfonated polyetherketone derived from exper-
iments and modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.8 Parallel water-channel (inverted-micelle cylinder) model of Nafion . . . . 13
1.9 Idealized membrane pore showing the hydronium ion, water molecules, and
radially symmetric axially periodic distribution of sulfonate SO3− fixed sites 14
1.10 Structure of four-unit perfluorosulfate oligomer, optimized in vacuum, wa-
ter and methanol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.11 Two monomeric sequences of Nafion 117 with different monomer clustering 18
1.12 Fully optimized global minimum energy structures of the C6 two sidechain
fragment showing hydration and proton dissociation as additional water
molecules are added . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.13 Fully optimized polymeric fragment and the potential energy profile for
rotation about the F2 C − CF2 bond along the backbone . . . . . . . . . 21
1.14 Representative configurations of the solvation structures observed in the
simulations using the classical hydronium potential and the EVB potential,
which were common for both hydration levels (λ = 7, 15) . . . . . . . . . 24
1.15 Proton mean-square-displacement (MSD) in Nafion and water . . . . . . 26
1.16 van Hove space-time correlation function for the hydronium-sulfonate ion
pair, given the sulfonate anion as the space-time origin . . . . . . . . . . 27
1.17 Morphological models of Nafion . . . . . . . . . . . . . . . . . . . . . . . 27

2.1 The seven dihedral angles of the SSC force field illustrated on polymer
segments with a protonated sidechains . . . . . . . . . . . . . . . . . . . 34
2.2 Comparison of the ab initio and classic torsion profiles (with contributions
from all MM terms) around the C1O-O3 bond . . . . . . . . . . . . . . . 37
2.3 Comparison of the ab initio and classic torsion profiles (with contributions
from all MM terms) around the C2S-S6 bond . . . . . . . . . . . . . . . 38
2.4 Energy profile of the dihedral potential for the angle H3–O3H–S6–C2S: in
the SSC force field and a Newman projection of the dihedral angle . . . . 39
ix

3.1 Local region (solid sphere) replicated in a spherical shell. The image
spheres fill up the entire volume of the 3D shell. A special arrangement of
the images ensures zero forces at the local region boundary . . . . . . . . 42
3.2 If L becomes smaller than 2Rc the interactions in the primary cell, as well
as all its neighbors have to be evaluated explicitly . . . . . . . . . . . . . 43
3.3 Hamiltonian conservation in a short MD trajectory of a Lennard-Jones
system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4 The distance between the central particle i and the image particle j 0 in
the m-th shell is a function of the angle θ (determining the position of the
image sphere in the shell), the i − j distance r in the primary cell and the
distance between the sphere centers 2mRc . . . . . . . . . . . . . . . . . 50

4.1 Reproduction of the original idea of Grotthuss for proton shuttling between
two electrodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.2 Schematic depiction of the “special pair dance” occurring in the first
solvation-shell of the hydronium during the long trajectory segments with-
out PT events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.3 Probability distribution of the proton as a function of the Oa −Ob distance
and the asymmetric stretch coordinate δ = ROa H − ROb H . . . . . . . . . 58
4.4 Scatter plot of the proton distribution in a Zundel ion (represented classi-
cally as hydronium ion plus water) at 300 K . . . . . . . . . . . . . . . . 59
4.5 A Zundel ion (represented classically as hydronium ion plus water) can ex-
ist in two resonance forms, obtained by interconversion between a covalent
OH bond (solid line) and a hydrogen bond (dashed line) . . . . . . . . . 61
4.6 PT between two sulfonate groups influenced by a neighbouring Zundel ion
in a triflic acid monohydrate solid . . . . . . . . . . . . . . . . . . . . . . 63
4.7 Molecules whose conformations are inside the reactive trigger zone (green
boundary) will be subject to JIT-EVB mixing of the resonance forms to
reproduce the ab initio forces along the grid . . . . . . . . . . . . . . . . 66
4.8 At point A a molecule enters the trigger zone. An ab initio calculation
has to be performed to determine the correct mixing of the resonance forms 67

5.1 Flowchart representing the creation of a SSC polymer with random monomer
sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.2 Perfluorocyclohexane converging to two distinct local minima depending
on the conformation seed . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.3 Flowchart representing the stepwise process of finding a low energy con-
formation for the polymer . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.4 The interactions needed to be considered in order to optimize the coordi-
nates of the new atoms (set A) are those within the set and with the built
atoms (set B) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

6.1 Cross section view of an ideal pore with the shape restraint represented as
an outer layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.2 Perfluoropentane built with a straight backbone along an axis . . . . . . 81
6.3 Single perfluoropentane chain built in a periodic cell with a harmonic bond
restraint between the end carbon atoms . . . . . . . . . . . . . . . . . . . 82
6.4 2D channels formed by two polymer chains (black and green) in a periodic
cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.5 Perfluorocyclohexane built in a periodic cell with MIC applied to both
bonding and non-bonding potentials . . . . . . . . . . . . . . . . . . . . . 85
6.6 Two perfluorohexane chains built in an infinite simulation universe without
any restraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.7 Schematic representation of the approach of two polymer strands in the
anti conformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.8 Schematic representation of a pore wall as formed by straight polymer chains 88
6.9 Cross section view of the pore and two polymer chains. . . . . . . . . . . 90

7.1 Polymer morphology snapshots at the end of the production run for the
three levels of hydration . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.2 Hydrogen bond chain from the snapshot in Fig. 7.1a (λ = 3) using MIC . 95
7.3 Polymer chain of Fig. 7.1a (λ = 3) without MIC . . . . . . . . . . . . . . 96
7.4 S-S pair correlation plot (between the ionized sulfonic group sulfur atoms)
for the three hydration levels . . . . . . . . . . . . . . . . . . . . . . . . . 97
7.5 An ion cage that exhibits very short S-S distances . . . . . . . . . . . . . 98
7.6 S-O pair correlation plot (between the ionized sulfonic acid sulfur atom
and the hydronium ion oxygen atom) for each of the three hydration levels 100
7.7 Fraction of hydronium ions with a given number of sulfonate neighbours
in the SSC polymer for different hydration levels λ . . . . . . . . . . . . 101
7.8 Fraction of hydronium ions with a given number of sulfonate neighbours
in the Nafion for different hydration levels λ . . . . . . . . . . . . . . . . 102

8.1 MSD of hydronium ions in the SSC polymer as a function of time and
hydration level λ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
8.2 MSD of the CEC in triflic acid solution as a function of time. The hydra-
tion level is λ = 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

A.1 Atom type labels for the SSC force field . . . . . . . . . . . . . . . . . . 117

B.1 Cross section view of a pore formed by polymer chains . . . . . . . . . . 121

D.1 When a hydronium ion and a water molecule satisfy the trigger conditions
for hydrogen bond h1 they both become part of a cluster . . . . . . . . . 126

x
Acronyms

ABNR Adopted Basis Newton-Raphson

CEC Center of Excess Charge

CM Center of Mass

DFT Density Functinal Theory

EVB Empirical Valence Bond

EW Equivalent Weight

F3C Flexible Three Center

IPS Isotropic Periodic Sum

JIT-EVB Just-in-Time Empirical Valence Bond

LRC Long Range Correction

MD Molecular Dynamics

MIC Minimum Image Convention

MM Molecular Mechanics

MSD Mean Square Displacement

MW Molecular Weight

PBC Periodic Boundary Conditions

PEM Proton Exchange Membrane

xi
xii

PES Potential Energy Surface

PFSA Perfluorosulfonic Acid

PT Proton Transfer

QM Quantum Mechanics

RDF Radial Distribution Function

RMS Root Mean Square

SASA Solvent-Accessible Surface Area

SD Steepest Descent

SMILES Simplified Molecular Input Line Entry Specification

SSC Short-Side-Chain
1

Chapter 1

Introduction
2

1.1 Introduction to Fuel Cells

Over the last hundred years since the beginning of the industrial revolution the consump-

tion of crude oil has exponentially increased. In the past twenty years alone the world has

consumed more than two thirds of the total amount of oil extracted since the 1880 [1].

According to reliable forecasts there now exists a small window of time (20-50 years) in

which our society will be faced with acute fuel shortages [2]. We are, therefore, at the

present in a transition period in which the global economy must reduce its dependence

on the consumption of non-renewable resources.

While we could envision home and business consumers switching to renewable sources

for their energy needs [3] the situation with the auto industry is far more challenging.

Transportation currently accounts for one third of the energy consumption in the United

States alone and the trends predict it becoming the largest consumer by 2020 [4]. Thus far

no alternative fuel technology has come anywhere near the gasoline/combustion engine

in price or reliability. The challenge is, therefore, not just developing efficient ways of

using alternative fuels in cars but also to have a profitable distribution network for these

fuels along with an enormous industrial capacity to produce them.

The United States is already in the process of laying down the foundations of this

colossal transition by setting forth a plan for a hydrogen-based economy. In the ini-

tial stages the hydrogen can be produced by reforming fossil fuels or by water splitting

employing nuclear energy with the ultimate goal of switching over time to renewable

sources like biomass or water splitting by sunlight, wind energy etc. The attractiveness

of hydrogen starts with its localized production – from nuclear reactors to any of the

alternative energy sources which can be situated in close proximity to the end user. Its

transportation can, in part, utilize the existing infrastructure for distributing natural

gas. When hydrogen makes its way down to our cities and vehicles we will have the
3

cleanest, most environmentally friendly chemical fuel1 . This is of paramount importance

as burning hydrogen does not lead to any CO2 emission. Hydrogen is a carbon-neutral

fuel.

Currently most of the hydrogen is used as a mixture with natural gas for combustion

engine vehicles. Hydrogen however has a special place among the other fuels in that its

reaction with O2 can be harvested to produce electricity directly, in an apparatus known

as the fuel cell. The attractiveness of using electric vehicles comes from their highly

efficient engines – they waste half as much energy when compared to a combustion

engine. The list of other fuels that can be converted to electricity in a similar manner is

limited to methanol and ethanol.

The invention of the fuel cell is attributed to Sir William Grove, a Welsh scientist who

in 1839 demonstrated that the electrolysis of water can be reversed, producing electricity

from hydrogen and oxygen [5]. The advancements in other competing technologies like

the dynamo generator which were successfully brought to market in the following years

meant little interest for an electrochemical generator. The interest in converting fuel

into electricity directly was resurrected over a century later with the advent of space

exploration. Commercializing the technology started in the 1980s by Ballard Power

Systems.

Classifying fuel cells can be done either by the type of fuel or, more commonly, by

the type of the electrolyte. A summary of the different types is presented in Fig. 1.1.

Unlike ordinary batteries that have to be recharged by another source of electricity fuel

cells can continuously provide electric current when fed with fuel and oxygen. Proton

exchange membrane (PEM) fuel cells employ a proton-conducting polymer membrane

as the electrolyte between the anode and the cathode. Oxidation of hydrogen on the
1
Nowadays hydrogen fuelling stations can be found scattered across parts of Europe and North
America. In British Columbia a “Hydrogen Highway” was built to link Victoria and Whistler in time
for the 2010 Winter Olympic Games.
4

Figure 1.1: Different types fuel cells. Figure reproduced from Ref. [6].
5

anode releases electrons into the outer circuit and protons into the surrounding medium.

The protons migrate to the cathode side to preserve the electroneutrallity of the system

where they recombine with the electrons and oxygen from the air to produce water. The

membrane separating the two electrodes allows passage of the protons while preventing

the flow of fuel, oxygen or electrons. An alternative to the proton conduction electrolytes

is found in the solid oxide fuel cell. Here the electrolyte system allows passage of oxygen

ions between the electrodes.

The electrolyte is the cornerstone of the fuel cell since the requirements for effective ion

transport dictate the design and operating conditions of the whole cell. The temperature

range that enables ion transport affects the kinetics of the electrochemical reactions

on the electrodes, the choice of catalysts and the acceptable level of impurities in the

fuel. Fuel cells operating under harsh conditions (i.e., corrosive electrolytes, very high

temperatures) would be impractical for mobile applications like powering electric cars or

consumer electronics. On the other hand, low-temperature fuel cells require expensive

catalysts which also makes them unsuitable. The current trend in fuel cell research is

to bridge the gap between these extreme cases in a manner that will provide a safe,

affordable solution for all applications.

The electrolyte systems that we are interested in are polymer based, with a Teflon

backbone and hydrophilic side chains. At the end of the side chains are protogenic groups

including sulfonic acid groups (−SO3 H). In order for these membranes to exhibit proton

conductivity they require water. The most well known representatives of this class of

polymers are Nafion (developed by DuPont) and the SSC polymer (developed by Dow)

whose chemical structures are given in Fig. 1.2. Here is an excerpt from the DuPont

announcement from 1969: “A new thermoplastic polymer family - offering features of

both fluorocarbon polymers and ion exchange resins - has been developed by DuPont’s

Plastics Departments. The new composition is currently called XR. It is expected to pro-
6

SSC 3M Nafion

Figure 1.2: PFSA based polymers for fuel cell membranes. The fragment in the Nafion
structure shown in red is absent in the SSC polymer, hence the name “short-side-chain”.
Typical n values for the SSC polymer are 3 to 6.

vide unique advantages for electrochemical, aerospace, and chemical industries. Based

on advanced fluorocarbon chemistry, the polymer exhibits such features of fluorocarbons

and ionomers as ionic conductivity, permeability, transparency, toughness, chemical in-

ertness, flexibility and adhesion to most substrates . . . The fluorinated polymer contains

pendant sulfonic acid groups, which produce an exceptionally strong acid resin. The

number of sulfonic acid groups can be varied to provide different ion exchange capacity,

electrical and mechanical properties.” [7].

1.2 Motivation

In the mid 1980s Ballard Power Systems demonstrated a significant improvement in fuel

cell performance with the new type of SSC PFSA membrane [8]. The current that the fuel

cell can provide is limited by diffusion processes which supply reactants to the electrodes,

including the trans-membrane shuttling of protons, seen in Fig. 1.1. Polymer systems

that exhibit proton conductivity with limited need for hydration are highly desirable as

this will allow an increase in the operating temperature of the fuel cell above 80◦ C.
7

In turn, this will allow for cheaper catalysts to be employed on the electrodes. In this

respect, the SSC polymer shows very promising conductivities with only a fraction of the

water required by Nafion (see Fig. 1.3).


12. Proton Conduction in PEMs 387

1e-3 0.15

1e-4
Conductivity (S/cm)

0.1
1e-5

pure water
DH O
2
1e-6
0.05
Nafion
1e-7 Dow SSC

0
1e-8
0 5 10 15 20 25 30
1
ction XV Waters per Sulfonate

(b)the SSC polymer as a function of the


Figure 1.3: Conductivity plot for Nafion and
hydration level. Figure source: Ref. [9].
ndence of proton mobility on water content. (a) Proton self-
D!) of Nafion and sulfonated polyetherketone membranes at
Furthermore, this practical observation shows that the membrane structure has a
tion of the water volume fraction showing the substantially
profound effect on its conductivity and raises the important question whether further
in the PFSA membrane as low to intermediate water contents.
Proton conductivity
improvementsof could
Nafion and through
be achieved low EW (!800)
structural Dow SSC
modifications. For instance – varying

a function of the
thestructure
waterof content expressed
the side chain, the type of as the number
the protogenic of number of CF2
group, the

onic acid group showing


groups substantially
in the backbone higherreliable
and so on. However, conductivity
prediction of at
structure-property
levels. Taken relationships
from Ref. [12].systems is a formidable challenge. Localized proton dissociation and
in these

shuttling requires an accurate QM description, which unfortunately cannot be applied

to the entire membrane channel where these events occur. Long simulation runs and
er, when theextensive
twosampling
membranes are fully hydrated the
are required to diminish the effect of fluctuations on the predicted
coefficients are very similar to one another. A further
ifferences that polymer chemistry may make on proton
igure 12.1(b) where the proton conductivity is plotted
and the short side chain (i.e. SSC) PFSA ionomer
8

dynamical properties. This is especially true when determining the effect of small, subtle

changes in the polymer structure like an additional CF2 group in the polymer backbone.

The goal of my Ph.D. work was to develop the simulation methodology that would

allow a better understanding of the properties of these systems. In particular, the effect

of structural parameters like the EW and the MW on the morphology of the hydrated

polymers and the proton diffusion rate. The key aspects of these developments are

presented in Part II of the thesis. This is followed in Part III with the practical aspects of

creating the hydrated polymer for computer simulations and the results of our structural

and dynamical property modelling.

Attempts of this nature have been made in the past and a vast amount of literature is

now available. However, for the purpose of this thesis we will review some of the previous

work that is directly related to the present objectives.

1.3 Review of Key Experimental Work

The morphology of the PFSA polymers has been the subject of numerous experimental

studies over the last few decades. A comprehensive review of the current understanding

of the morphology of these systems can be found for example in Ref. [10]. From a his-

torical perspective, the introduction of the inverted micelle model, for the structure of

the PFSA systems by Gierke et al. [11], is no doubt one of great significance. Based on

both wide-angle and small-angle x-ray diffraction data, as well as transmission electron

micrograph imaging and using a process of elimination of other morphologies these au-

thors have arrived at the cluster-network, depicted in Fig. 1.4, as the one best fitting the

experimental data.

Within the context of this model the ionic clustering of the strong sulfonic acid groups,

results in spherical vesicles containing the adsorbed water to produce structures that
9

Figure 1.4: Cluster-network model for the morphology of hydrated Nafion. Figure
source: Ref. [10].

resemble an inverted micelle. Percolation between these clusters is assumed to occur

through narrow channels that interconnect the micelles. However, one important aspect

of the membrane morphology that is not accounted for by the Gierke model is the partial

crystallinity brought about by the polytetrafluoroethylene backbone. Hence, it has been

suggested that a separate crystalline phase exists in the domain where the polymer chains

assemble in a lamellar hexagonal conformation [12], as seen in Fig. 1.5. The side chains

extend perpendicular to the polymer backbone into the ionic cluster domains.

One of the most important the properties of the perfluorosulfonic acid (PFSA) mem-

branes is their hydration level. As the water content increases to a critical fraction of 0.1,

an insulator-to-conductor transition occurs in the membrane, which is attributed to trans-

port through the channel connections between the inverted micelle clusters [13]. Further

swelling of the polymer results in a structural inversion that leads to rod-like polymer

structures suspended in the water phase, as shown in Fig. 1.6. A coherent mechanis-

tic scheme that explains the dilution and swelling of the membranes emerges when the

morphology of the system is described by elongated bundles of polymer chains [15]. The

diameter of these bundles is estimated to be on the order of 40 Å with lengths larger

than 1000 Å.


10

SO3−
SO3−

SO3−
SO3−

SO3−
SO3−

SO3−
SO3−

SO3−
! SO3− ""!

suggested
Figure 1.5: Twobyrows
Gebelof the41,hexagonal
et al. our evaluation of surface-to-volume
lattice formed by the ratios of the fluoropolymer
polymer chains (shown as
circles) seen end-on.
(described Figure
in Chapter reproduced
3) shows from Ref. of
that the morphology [12].
Nafion evolves from a matrix with
dispersed water molecules at low hydration level, to the water channel model at medium
hydration level, and to polymer aggregates dispersed in water at high hydration level (Figure
1.6).

Figure 1.6: FMorphologies of Nafion


igure 1.6. Morphologies at different
of Nafion at differentvolume fractions
volume fraction of(Iwater.
of H2O H2O).
Based on
surface-to-volume ratios of the fluoropolymer the morphology of Nafion evolves from a
matrix with dispersed water molecules at low hydration level, to the water channel model
at medium hydration
1.6 Improved P E level,
M withand, finally,
inorganic to polymer aggregates dispersed in water at high
nanoparticles
hydration level. Figure source: Ref. [14].
As mentioned in section 1.3, the working temperature of normal PFSA membranes is limited
at around 80 ~ 90 oC, which constrains the tolerance of Pt catalysts to contaminants.
Furthermore, poor proton conductivity at low hydration levels complicates the design of
67
water management systems. Materials based on e.g. sulfonated poly-arylene ethers , poly-
68,69 70,71
sulfones , and poly-benzimidazoles have been developed as possible alternatives, but
most of them have either lower proton conductivity or insufficient stability.
11

Figure 1 Schematic representation


Figure 1.7: Schematicof therepresentation
nano-phase separation
of thein the hydrated morphology
nano-phase separation of Nafion
in theand sulfonatedmor-
hydrated PEEKK derived
from experiments and modeling. This scheme illustrates the distinctions in the hydrophilic/hydrophobic separation, connectivity of the water
phology of Nafion and sulfonated polyetherketone derived from experiments and mod-
and ion domains, and separation of the −SO3− groups. Taken from Reference (6) with permission from Elsevier.
elling. This scheme illustrates the distinctions in the hydrophilic/hydrophobic separation,
connectivity of the water and ion domains, and separation of the SO3− groups. Figure
source: Ref. [16].

The effect of the polymers microstructure and the acidity of the protogenic groups on

the transport properties of the membranes has been recently investigated in the seminal

work of Kreuer [16]. Substitution of the perfluoro-polymer with a less hydrophobic aro-

matic polyetherketone system resulted in narrower, less connected hydrophilic channels

and to larger separations between the sulfonic acid functional groups (see Fig. 1.7). Al-

though the protogenic groups were found to be less acidic, the proton conductivity has

remained high, while the undesirable elecroosmotic drag of water is significantly reduced.

The work also explores polymer blends as a way of improving the characteristics of the
12

membrane, as well as the possibilities of using heterocycles as the proton solvent. Such

heterocyclic compounds (e.g., imidazole) can be immobilized by grafting onto a polymer

backbone, thus providing a non-volatile, proton-conducting medium that operates in the

complete absence of water.

The most solid evidence for the morphology of the hydrated Nafion polymers has

come from the recent work of Schmidt-Rohr et al. [17]. By performing a new analysis on

the existing small-angle scattering data that includes contributions from the crystalline

phase, these authors have been able to show, unequivocally, the occurrence of cylindrical

inverted micelles. The walls of the micelles are formed from straight helical polymer

chains and are lined with the hydrophilic side chains. At a hydration level of 20vol%

the diameter of the channels is about 2.4 nm. Furthermore, not only is the inverted

micelle cylindrical but so is the general shape of the crystalline phase. A schematic

representation of these structures is shown in Fig. 1.8.

1.4 Review of Theoretical Work

In this section we will review some of the modelling work carried out in the last couple

of decades that is most relevant to our own simulations. A much broader-scope review

of the previous computational work can be found in Refs. [10, 18].

One of the earliest proton transport studies to include structural level details was

the nonequilibrium statistical mechanical model of Paddison et al. [19]. The idea of

the model is to calculate the hydronium ion diffusion coefficient through the Einstein

relationship D = kT /ζα , where ζα is the Stokes friction coefficient of the hydronium ion.

This friction coefficient can be conveniently determined from a sum of four friction terms,

each calculated as a the ensemble average of a force correlation function:


13

a b c d
H2 O 100

10 1

4 nm
H2O
10 0

I (a.u.)
10 –1
50 nm

10 –2

H2O channel Crystallite

Figure 1.8: Parallel water-channel (inverted-micelle cylinder) model of Nafion. a, Two


views of an inverted-micelle cylinder, with the polymer backbones on the outside and
the ionic2 Parallel
Figure side groups lining the (inverted-micelle
water-channel water channel. Shading
cylinder)ismodel
used oftoNafion.
distinguish
a, Twochains in an inverted-m
views of
front and in the back. b, Schematic diagram of the approximately hexagonal packing
the ionic side groups lining the water channel. Shading is used to distinguish chains in front and in the back. b,
of several inverted-micelle cylinders. c, Cross-sections through the cylindrical water
of several(white)
channels inverted-micelle
and the cylinders. c, Cross-sections
Nafion crystallites (black)through
in the the cylindrical water
non-crystalline channels
Nafion (white) and the N
matrix
(dark grey),
(dark grey),asasused
usedininthe simulation
the of the
simulation small-angle
of the scattering
small-angle scattering in d. d,Figure
curvescurves. Small-angle scattering data
source:
Nafion
Ref. at 20 vol% of H2 O, and our simulated curve from the model shown in c (solid line). The inset shows the
[17].
curves from the water channels and the crystallites by themselves (in a structureless matrix) are shown dashed

structure was 13% by volume (15% of the dry polymer). Other and simulated dat
simulations with crystallinities between 9 and 15% also gave the excess scatteri
acceptable results. The crystallinity from the straight-cylinder are discussed in th
model is probably an overestimate because simulations for
undulating channels (see Supplementary Information, Fig. S3) SAXS SIMULATIONS
show that correlated undulations of crystallites and water
channels reduce the scattered intensity at intermediate and For comparison
high q values, while leaving the small-angle upturn due to scattering for o
crystallites unchanged. namely Gierke’s
14
J. Chem. Phys., Vol. 115, No. 16, 22 October 2001 Proton diffusion in p

Beardmore53 using Lekn


actions in three-dimensi
one and two dimensions
chosen anionic charge di
The third term in Eq
water–water interactions
dipole interactions accor
2
V ss " ! ri "r j ! # !
3" 4'
where, once again, a th
over all rotational angles
The final term descr
molecules experience du
der the assumption that th
Figure 1.9: Idealized membrane
FIG. 1. Idealized pore pore
membrane showingshowing thethehydronium
hydronium ion, ion, water
water field dueand
mol- molecules, to the fixed sit
ecules,axially
and radially symmetric axially periodic distribution SO −
of sulfonate
radially symmetric periodic distribution of sulfonate 3 fixed with
sites. the expression
Figure
"–SO3"# fixed sites.
source: Ref. [20].
"2 ' & * 0
V sp " ri # %
eL
collection of momentum and position vectors of the N water
molecules each with a mass m, and V(r! ,r) is the total po- It should be clear a
Z∞ tential energy Z∞of the system. The latter Z∞ is assumed to consist Z∞ scribed, is an (N#1)-bo
ζ

ofFthe following



molecules and a single h
= Fαs e−iLt αs dt + Ffour
αs e
terms:
−iLt
Fps dt + Fαp e−iLt Fps dt + Fαp e−iLt Fas dt
β N
pore there is one proton
0 0 0 0 the presence of the ‘‘oth
V " r! ,r# !
i!1
$ V ! s " ! r! "ri ! # #V ! p " r! # (1.1)
certain ramifications. Per
In this equation L is the Liouville N operator ofNthe system, and the forcepresence subscripts of the other pro
ing of the interaction of
α, s and p designate the hydronium # $
i$ j
" ! ri "r
V ssion, the $
j ! solvent
##
i!1
sp " ri # . molecules"2#
Vwater and nium
the ion and the water m
pendant
tons will result in ov
groups (i.e., theThe first term
sulfonate is the respectively.
anions), interaction potential energy between
For example, the first the
term on+calculated
the right-in Eqs. "4# a
hydronium ion and the ith water molecule and is assumed to experienced by the hydr
hand-side of Eq.be1.1 corresponds
a typical ion–dipoleto theinteraction.
average force experienced
If the rotational by the hydronium ion
contri- cific information concern
butions are ignored one obtains the simplified expression the pore,ion the effects of th
due to the solvent molecules, the next term is the average force on the hydronium
2 2 in the model at this point
& e 1
due to the pendantVgroups! s " ! r! "rviai ! #the
%"solvent 2 2medium and 4 , so on. For a "3# simpleproton
cylindrical
interactions are n
48' ( kT ! r! "ri !
contribution of these repu
pore model withwhere
regular ( is pendant chain distribution
the permittivity of the water (shown
in the schematically
pore, k the in Fig. 1.9) the
efficient will be small. C
Boltzmann constant, and T the temperature. error"s# introduced by ign
last three terms of Eq. 1.1 can be calculated exactly.
The second term is the potential energy experienced by significant.
the hydronium ion due to the sulfonate
The predictive capability of the method was assessed by separate calculations groups. As indicated Thecar-time-dependent
earlier, these pendant groups are distributed periodically in mentum of all th
the pore,
ried out with Nafion and if thepores
membrane lengthhydrated
of their intrusion
with 6 and within the poremolecules,
13 waters is f N#1respec-
(p! ,r! ,p,r;t), satis
R" ) "thus ) is the radial separation distance of the hydro-
nium ion from the fixed sites# and axial spacing L/n, and one - f N#1
i !L T f N#1 ,
assumes that the hydronium ion is transported along the axial -t
center of the pore, then this potential energy term is assumed
where L T is the Hermiti
to have the form
Poisson bracket
15

tively, associated with each fixed anionic site (pendant chain). When combined with

experimentally estimated parameters, the model-predicted proton diffusion coefficients

of 5.05×10−10 and 8.36×10−10 m2 /s, are in good agreement with experimental values.

For a Nafion membrane pore with an hydration level of six water molecules per sulfonic

acid group, the model was used to compute friction coefficients for various distributions of

the pendant sites, and for different side chain lengths [20]. The model showed substantial

sensitivity to these parameters and predicted that for pores of fixed volume and a constant

total number of sulfonate groups, the friction on the hydrated proton is the greatest for

distributions with high local sulfonate density. When the radius and length of the pore

were varied, the model demonstrated that the proton diffusion increases with increasing

channel diameter. These calculations, therefore, demonstrate the important predictive

capability of this molecular-based, nonequilibrium statistical mechanical model.

More recently the model was used to predict the diffusion coefficient of protons in a

fully hydrated Nafion membrane with 22.5 water molecules per sulfonic acid group [21].

In this work profiles of the friction and diffusion coefficients were determined across the

radius of the pore, demonstrating that these parameters vary by a full order of magnitude

across the radial cross-section of the pore. The model calculated a diffusion coefficient

for a proton moving along the pore center of 1.92×10−9 m2 /s in good agreement with

experimental measurements. In addition, the model also identified that the region within

4 Å of the pore center exhibits an intermolecular transfer (Grotthuss mechanism) of the

proton between the water molecules. This is in contrast with the region lying within 8 Å

of the wall of the pore, where the transport of the proton is predominantly vehicular in

nature. Although the agreement with the experimental diffusion measurements has been

very good for diffusion along the center of the pore, inclusion of the contributions from a

distribution of protons across the radius of the pore will certainly correspond to a more

realistic modelling of proton transport.


16

In one of the earliest atomistic simulations of the PFSA system the conformations

and hydrophilicity of the side chain in Nafion were examined using Density Functinal

Theory (DFT) [22]. The study has shown that the ether portion of the side chain is

hydrophobic and stiff, while the SO3− group is strongly hydrophilic and more flexible. In

the absence of explicit solvent molecules the preferred side chain conformation is a folded

(curled up) one.

Similar results were later obtained on for a much larger system using molecular me-

chanics (MM) [23]. In this latter study the force field parameters for the polymer were

chosen to reproduce some of the experimental data, such as the liquid-vapour equilibrium

and the self-diffusion coefficient. A highly folded spiral-like configuration was obtained

for a 10 unit polymer when the randomly bent chain was taken as the initial configura-

tion. MD simulations of shorter oligomers solvated in water and methanol have revealed

a noticeable difference between the backbone conformations in different solvents, as seen

in Fig. 1.10. The skeleton structure in water was observed to be substantially more folded

than in methanol. Additionally, the side chain of the Nafion oligomers was found to be

quite stiff; with only a few conformational transitions being detected. Examining the

hydrogen bond propensity of the side chain, the authors have found that both water and

methanol form stable hydrogen bonds with the oxygens of SO3− group. All other parts

of the side chain were hydrophobic, including ester oxygens, which showed practically no

tendency for hydrogen bonding. On average, each SO3− group formed approximately five

hydrogen bonds to the solvent water and four bonds to methanol. The solvent molecules

bonded to the sulfonate group form a pronounced anisotropic first solvation shell.

Following this, the same authors investigated the microphase segregation in hydrated

Nafion membranes at different water contents [24]. As the degree of solvation increased,

the formation of water clusters containing up to about 100 water molecules was observed.

In contrast to the conventional network models, the water clusters did not form a con-
17

. Phys. Chem. B, Vol. 104, No. 18, 2000 Vishnyakov and Neimark

Optimized structures of ten-unit perfluorosulfate oligomer


by potential energy minimization in a vacuum (a) stretched
tion with tortuosity of 2.4, obtained from the regular config-
ith all CCCC dihedrals in trans position, (b) strongly folded
e configuration obtained from a random structure.

3a). The folding is caused by intramolecular van der


nd electrostatic interactions, which turned out to be
nough to overcome the gain of the torsional potential
ue to folding. None of the optimized structures shown
es 2 and 3a are likely to represent the global minimum
tential energy of the oligomers in a vacuum. However,
ysis shows that even in a vacuum molecular geometry
gomers is not entirely dominated by the torsional term.
on, the intramolecular nonbonded interactions compete
termolecular solute-solvent interactions, which favor
hing of the oligomer. However, the intramolecular
contribution to the free energy always favors to folded
of chain molecules; for example, this effect is well
nted for lipid membranes.50 The three main contributions
ee energy of conformational transition of a macromol-
solution are (1) intramolecular terms, (2) change in
alpy of solvent-solute interactions, and (3) change in
alpy and entropy of the solvent. The latter is especially
nt for hydrophilic molecules in self-associated solvents Figure 3. (a) Structure of four-unit perfluorosulfate oligomer, opti-
hose considered in the present work. This term cannot mized in aofvacuum (b) snapshot of the molecular configuration of the optimized in
Figure 1.10: (a)
ctly taken into account in static simulations, like energy
Structure four-unit perfluorosulfate oligomer,
four-unit oligomer in water and (c) the same in methanol.
vacuum
ation. On the one hand, water (b)shows
snapshot of the molecular configuration of the four-unit oligomer in water
more extensive and (c)
n bond network compared tothe methanol.
sameOninthe other
methanol. ofFigure
Nafion conformational
source: Ref.transitions
[23]. in polymer chains are
ethanol is expected to show stronger van der Waals sterically hindered. Therefore, Nafion chains in the membrane
ons with the hydrophobic polytetrafluoroethylene skel- matrix are not supposed to exhibit the conformations similar to
Nafion. Thus, methanol is a better solvent for Nafion; those observed for the oligomers in a vacuum. Yet, the energy
solvated Nafion membrane, the microphase segregation minimization in a vacuum allows one to compare different
pronounced in the case of water. intramolecular contributions into the total internal energy of the
uld be noted, also, that the energy barriers of confor- system.
transitions in condensed phases are much higher In our molecular dynamics simulations of the four-unit
d to those in a vacuum. In the dense membrane matrix oligomer in water and methanol, the fluorocarbon skeleton
18

Figure 1.11: Two monomeric sequences of Nafion 117 with different monomer clustering:
(top) blocky polymer with low degree of randomness (bottom) more dispersed polymer
with high degree of randomness. Figure source: Ref. [25].

tinuous hydrophilic subphase. The cluster size distribution was found to be wide and

evolved in time due to formation and break-up of temporary bridges between the clus-

ters. This dynamic behaviour of the cluster system allowed for the macroscopic transfer

of water and counterions. The calculated diffusion coefficients of water were found to be

of the same order of magnitude as those experimentally measured.

The properties of hydrated Nafion are attributed to its nanophase-segregated struc-

ture in which hydrophilic clusters are embedded in a hydrophobic matrix. However,

prior to the work of Jang et al. (Ref. [25]) there has been little characterization of how

the monomeric sequence of the Nafion chain affects the nanophase-segregation structure

and transport in hydrated Nafion. Using atomistic MD simulations, the authors have

investigated these effects on a hydrated Nafion system with 15 water molecules per sul-

fonate group. Two extreme monomeric sequences were examined, one very blocky and

other very dispersed, as illustrated in Fig. 1.11. Both monomeric sequences produce

a nanophase-segregated structure with hydrophilic and hydrophobic domains. The cal-

culated structure factor shows that the monomer sequence of the polyelectrolyte has a

noticeable effect on the extent of phase-segregation: the blocky sequence has better phase

segregation than the dispersed case. The characteristic dimension of the simulated hy-

drophilic clusters is 50 Å for the blocky case and 20-30 Å for the dispersed case, which are

in good agreement with the 40-50 Å obtained from small-angle scattering experimental
19

observations. This comparison suggests that the real Nafion structure is intermediate

but closer to the blocky case. The interface between the water and polymer phases was

also analyzed to determine how the sulfonate groups are arranged at the interface. It

was found that the latter has a heterogeneous structure, consisting of hydrophobic and

hydrophilic patches. The degree of segregation and size of the patches is larger in the

blocky sequence than the dispersed case. Water transport in these systems was shown

to depend on these structural differences caused by the monomeric sequence. Since the

blocky case leads to larger clusters and channels the observed water diffusion was higher.

This result is consistent with the experimental studies on the difference between Nafion

and sulfonated PEEK. However, no significant difference in the hydronium vehicular dif-

fusion was observed indicating that the proton hopping mechanism must be responsible

for the experimentally observed differences in proton diffusion.

Thereafter, the focus of most studies shifted to include the promising new SSC poly-

mer system under minimal hydration conditions [26]. The number of CF2 groups in the

backbone that separates the side chains affects the connectivity of the terminal sulfonic

acid groups. Specifically, with more than four CF2 groups no hydrogen bonding occurs

between neighbouring sulfonic acid groups on the same backbone in the absence of water.

The number of water molecules required to form a continuous hydrogen-bonded network

between the terminal sulfonic acid groups is also a function of the number of CF2 groups

on the backbone separating the side chains. It has been shown that one, two, and three

water molecules bridged the sulfonic acid groups when five, seven, and nine CF2 units

separated the chains, respectively. The separation along the polymeric backbone of the

side chains affects the minimum amount of water necessary to observe the transfer of

protons to the first hydration shell as demonstrated in Fig. 1.12.

In order to understand the flexibility of both the backbone and side chain in the SSC

PFSA system, a computational first principles based study was conducted on a dry, two
20

Modeling of Perfluorosulfonic Acid Membrane J. Phys. Chem. A, Vol. 109, No. 33, 2005 7589

(a) (b)

Figure 1.12: Fully optimized (B3LYP/6-311G**) global minimum energy structures of


the C6 two sidechain fragment showing hydration and proton dissociation as additional
water molecules are added: (a) no dissociation of either acidic proton with four H2 Os,
although the O-H bond distance in one of the sulfonic acid groups is lengthened 1.10 Å;
(b) both protons dissociation with the hydration of five H2 Os. Figure source: Ref. [26].

side chain oligomer [27]. The rotational PES of the various C-C, C-O, C-S, and S-O bonds

were examined with the help of DFT, revealing that the polymer backbone is relatively

stiff, with a barrier of nearly 7.0 kcal/mol (see Fig. 1.13). This barrier corresponds to
Figure 4. Fully optimized (B3LYP/6-311G**) global minimum energy structures of the C6 two side chain fragment showing hydration and proton
the energy difference between the staggered trans and planar cis conformations of the
dissociation as additional water molecules are added: (a) no dissociation of either acidic proton with four H2Os, although the O-H bond distance
in one of the sulfonic acid groups is lengthened 1.10 Å; (b) both protons dissociation with the hydration of five H2Os; (c and d) the hydrated
carbon atoms.
protons in Zundel-ion-like Furthermore,
configurations that further the calculations
separate have
with hydration shown
of six thatHthe
and seven stiffest portion of the
2Os, respectively.

C6 Fragment side+chain
4-7 H is2Os.
nearThe
its B3LYP/6-311G**
attachment to the backbone
fully with
water the CF per
molecules − Osulfonic
and O −acid CFrequired
2 barriersto stabilize the
optimized structures for the C6 fragment with nine carbon atoms protonic charge in the first hydration shell. Of additional
in the backboneofare9.1 and 8.0inkcal/mol,
displayed Figure 4a-d,respectively. The mostsignificance
with the energies flexible portion of the
is the result thatside
bothchain occursprotons appear
dissociated
and selected structural data reported in Tables 1 and 4, +
as ‘Zundel ion’-like (i.e. H5O2 ) in the global minimum
at the pointofoftheattachment
respectively. Examination of the sulfonic
calculated binding energies acid conformation,
group wherewith the one
rotational barrier of
of the hydrated the bridging the
protons
reveals that the trend in the binding energy calculated per water two sulfonate groups. This result is very similar to that
carbon-sulfur
molecule is consistent bond
among the was
three determined
methods. to be only determined
Specifically, 2.1 kcal/mol. via AIMD as a stable protonic defect in the
the binding energy per water molecule decreased from that with trifluoromethanesulfonic acid monohydrate solid, where in the
The polymer flexibility studies were later extended
only a single water molecule (Figure 3a) as the water molecules
with extensive searches for mini-
unit cell (i.e. [CF3SO3H]4) one proton is ‘shuttled’ between two
were added, until proton dissociation occurred. With proton sulfonate groups and another ‘shuttled’ between a pair of water
mum energy structures with 4-7 explicit water molecules [28, 29]. It was13shown that the
dissociation, and subsequent separation of the hydronium ions molecules as a Zundel ion. The global minimum energy
from the sulfonate groups, the magnitude of the binding energy
perfluorocarbon backbone may adopt either anstructures elongated of the same oligomeric fragment with six and seven
geometry, with all carbons in
increased to a value of 14.7 kcal/mol for the fragment with seven added water molecules are shown Figure 4c and 4d and indicated
water molecules a when
transcomputed on the CP-corrected
configuration, or a foldedpotential
conformation thatasthea additional
result of water has not appreciably
the hydrogen bondingchanged
of either the
energy surface. The lowest energy conformation of the oligo- presence or position of the dissociated protons: they remain
meric fragment with four added water molecules (Figure 4a) Zundel-like. The tabulated structural data (Table 4) indicates
showed no proton dissociation but an increase in the oxygen- that the additional water molecules have resulted in a greater
hydrogen bond distance of nearly 0.1 Å for both sulfonic acid separation of the transferred protons from their conjugate bases
protons and some increased separation of the terminal side chain and a bringing together of the sulfonate groups.
groups (>0.3 Å). The distance (along the backbone) between C8 Fragment + 3-7 H2Os. Binding energies computed from
the tertiary carbons remained essentially constant in the equi- both uncorrected and ZPE corrected total electronic B3LYP/6-
librium structures as the water molecules were added, indicating 311G** energies, along with those corrected for BSSE with
21

!
!
!
!
!
!
!
!
!
!
!
!
!
!
"#$%&'!
Figure ()! ! "*++,!Fully
1.13: -./010234! 56789:;<=7($>>?!
optimized .-+,13@0A! B@CD13E/)!
(B3LYP/6-31G**) :-/3E/0C+!fragment
polymeric 3E3@D,! .@-B0+3!
andB-@!the
@-/C/0-E! CF-*/! /G3!
potential
"HIíI"H!F-E4!C+-ED!/G3!FCAJF-E3)!
energy profile for rotation about the F2 C − CF2 bond along the backbone. Figure source:
!
Ref. [29].
:-/3E/0C+!3E3@D,!.@-B0+3K!L3@3!43/3@10E34!B-@!/G3!.-0E/!-B!C//CAG13E/!-B!/G3!K043!AGC0E!/-!
/G3!.3@B+*-@-=!FCAJF-E3!C/!/G3!6789:;<=7($>>!+3M3+N!0EA+*40ED!/G3!"IOP!CE4!PO!I"H!
F-E4K! CE4! C@3! 40K.+C,34! 0E! "0D*@3! H)! ! #/! GCK! D3E3@C++,! F33E! F3+03M34! /GC/! /G3! 3/G3@!
the terminal
+0EJCD3K! sulfonic
0E! /G3! acids with
K043! AGC0EK! the water.
.@-M043! These
B+3Q0F0+0/,! electronic
CE4! structureB@334-1!
A-EB-@1C/0-EC+! calculations show
-B! /G3! K043!
AGC0EK)!!R!.@0-@!5(S?!B0@K/!.@0EA0.+3K!0EM3K/0DC/0-E!F,!-E3!-B!/G3!C*/G-@K!@3M3C+34!/GC/!/G3!
that the fragments displaying the latter ’kinked’ backbone possessed stronger binding of
@-/C/0-EC+!FC@@03@!-B!/G3!-*/3@1-K/!3/G3@!+0EJCD3!0E!/G3!TCB0-E!K043!AGC0E!LCK!CF-*/!U)<!
JAC+;1-+)!!'QC10EC/0-E!-B!/G3!K*@BCA3K!0E!"0D*@3!HN!G-L3M3@N!0E40AC/3!/GC/!/G3!FC@@03@K!-B!
the water to the sulfonic acid groups, and also undergo proton dissociation with fewer
/G3!5K-+3?!3/G3@!+0EJCD3!-B!/G0K!VVI!/-!/G3!FCAJF-E3!0K!K*FK/CE/0C++,!D@3C/3@!L0/G!FC@@03@K!
water molecules.
-B!C..@-Q01C/3+,!W)(!CE4!<)X!JAC+;1-+!B-@!/G3!"IOP!CE4!POI" H!F-E4KN!@3K.3A/0M3+,)!!Y3!
4-! .-0E/! -*/! /GC/!the
Subsequently, F-/G! K*@BCA3K!
hydration of C@3! Z*0/3!PFSA
the SSC [@-*DG\! CE4! C/!has
membrane K-13!
been.-0E/K! C..3C@!
explored /-! F3!
through
40K]-0E/34)!!^G3K3!@3K*+/K!C@3!+0J3+,!4*3!/-!/G3!0ECF0+0/,!-B!/G3!10E0102C/0-E!KAG313!5_H?!
/-!+-AC/3!C!D+-FC+!10E01*1!B-@!/G3!K043!AGC0E!5i.e.!0/!F3A-10ED!3E/CED+34!CK!/G3!40G34@C+!
comparing the energetics of a three side chain oligomeric fragment of the polymer [30].
CED+3!0K!A-EK/@C0E34?!CE4!C@3!43K.0/3!C!M3@,!K*FK/CE/0C+!3BB-@/!/-![K1--/G!-*/\!/G3!K*@BCA3!
Extensive searches for minimum energy conformations with between 6 and 9 water
F,! .3@B-@10ED! E*13@-*K! KACEK! -B! /G3! CED+3! F3D0EE0ED! B@-1! 40BB3@3E/! .-0E/K! -E! /G3!
K*@BCA3)!#/!0K!C+K-!01.-@/CE/!/-!E-/3!/GC/!43K.0/3!/G3K3!K0DE0B0ACE/!FC@@03@K!B-@!/G3!A-1.+3/3!!
molecules revealed that at the lower range of the examined hydration (i.e., 2H2 O/SO3 H)

the uniform hydration of the sulfonic acid groups results in the lowest energy and there-

fore most favourable state of the system. The calculations have shown that as the degree

of hydration is increased the energetic preference for uniform hydration decreases, disap-

pearing altogether at 3H2 O/SO3 H. Furthermore, it was found that water distributions

that facilitate a higher degree of dissociation and separation of the protons are important

factors in stabilizing the fragments.

The effects of hydration level and temperature on the nanostructure of an atomistic

!!
"#$%&'!H)!!:-/3E/0C+!3E3@D,!.@-B0+3K!B-@!@-/C/0-E!CF-*/!/G3!"IíP!CE4!PíI"H!F-E4K!C/!/G3!C//CAG13E/!-B!/G3!K043!AGC0E!
/-!/G3!FCAJF-E3)!
!
@-/C/0-E!-B!/G3!K043!AGC0E!C/!0/K!C//CAG13E/!/-!/G3!FCAJF-E3N!/G3@3!0K!C!K0DE0B0ACE/!.-@/0-E!
22

Table 1.1: Density of Hydrated Nafion at 300 and 350 K. Table source: Ref. [31].

hydration level (λ) density at 300 K, [g/cm3 ] density at 350 K, [g/cm3 ]

3.5 1.700 ± 0.038 1.703 ± 0.034


6 1.698 ± 0.051 1.677 ± 0.028
11 1.686 ± 0.025 1.662 ± 0.021
16 1.640 ± 0.023 1.606 ± 0.018

model of a Nafion membrane, as well as the vehicular transport of hydronium ions and

water molecules were examined using classical MD simulations in the paper of Venkat-

nathan et al. [31]. Through the determination and analysis of structural and dynamical

parameters such as radial distribution functions (RDF), coordination numbers, mean

square deviations, and diffusion coefficients, the authors have shown that hydronium

ions play an important role in modifying the hydration structure near the sulfonate

groups. In the regime of low level of hydration, short hydrogen bonded linkages made of

water molecules and sometimes hydronium ions alone give a more constrained structure

among the sulfonate side chains. The work has also examined the density of the Nafion

systems as a function of the hydration level and temperature (see Table 1.1), which is of

paramount importance for determining the correct diffusion coefficients. At 300 and 350

K, the density of the hydrated Nafion system gradually decreases with an increase in the

level of hydration of Nafion. This was attributed to structural relaxation with increasing

hydration that leads to a swelling of the membrane at a given temperature. Temperature

was found to have a significant effect on the diffusion coefficients for both water and

hydronium ions. The diffusion coefficient for water agreed well with experimental data,

while the diffusion coefficient of the hydronium ions was much smaller (6-10 times) which

was attributed to the lack of inclusion of the proton hopping mechanism in the study.

Subsequently, the work was significantly extended in a two part paper examining

the effect of hydration on the membrane nanostructure and the dynamics of water and
23

hydronium ion [32,33]. It was found that at λ less than 7, most of the water molecules and

hydronium ions are bound to the sulfonate groups. The strong binding of hydronium ions

to sulfonate groups prevents vehicular transport of protons. Multiple sulfonate groups

surrounding the hydronium ion offer steric hindrance to hydration of the hydronium ion,

which hinders structural diffusion of protons. Water molecules were mainly found in

the vicinity of the sulfonate groups, while the ether oxygen and backbone were strongly

hydrophobic. In addition to calculating diffusion coefficients as a function of hydration

level, the authors have also determined mean residence time of water and hydronium ions

in the first solvation shell of SO3− groups. The mean residence time of water decreases

with increasing membrane hydration from 1 ns at a low hydration level to 75 ps at the

highest hydration level studied. The mean residence time of hydronium ions is larger than

the corresponding values for water molecules by a factor of 2.5-4.5. The work provides an

explanation for the experimentally observed characteristic time of slow proton dynamics

in hydrated Nafion in terms of the residence of hydronium ions and water molecules in the

first solvation shell of SO3− groups. These dynamical changes are related to the changes

in membrane nanostructure.

More recently, a comparative study was carried out to determine the hydrated mor-

phology and proton diffusion coefficients in two different PFSA membranes as functions

of water content [34]. Classical MD simulations were performed on a 1143 EW Nafion

and a SSC PFSA polymer with an EW of 977. The water cluster distributions displayed

distinctive differences at the lower water contents (with 4.4 and 6.4 H2 O molecules per

side chain) where hydration of the SSC PFSA membrane tends to produce a more dis-

persed cluster distribution, and thus enhance the connectivity of the clusters by water

channels. On the other hand, the Nafion system characterized by longer and more flexible

side chains, is more amenable to aggregate and form clusters that are more disconnected.

At higher water contents, the cluster differences between the two systems become very
solvent reorganization because th
through the hydrogen bond netw
cation configuration predominates
is adjacent to a sulfonic acid grou
cation is hydrogen bonded to a sulf
is located away from SO3- (bottom
24
easily moves away from the sulfo
shuttle between these two wate
evidence exists for this kind of
Letters J. Phys. Chem. B, Vol. 109, No. 9, 2005 solvation 3729 structure in favor of
concentrated strong acids, i.e., con
for the transition between the CIP and the SSIP is clearly a
Simulations recently performed w
deficiency, since that model does not allow for rearrangement iterative multi-state empirical va
of the covalent and hydrogen bond configurations. A goal of 20
method, which is capable of
this study is to contrast the classical and MS-EVB potentials in has replicated this enhance
protons,
order to show the need for a model that allows bonding structure over the Eigen-type. In
rearrangements (Grotthuss shuttling), so we have chosenmental not to results, a classical hydron
investigate these artificial peaks further. deficient.
The MS-EVB proton CIP to SSIP transition requires little The bottom panel of Figure 2 sho
solvent reorganization because the excess proton can shuttle the classical and MS-EVB2 hydro
through the hydrogen bond network. A Zundel-like (Hthe +)
5O2sulfonic acid oxygen for the lo
cation configuration predominates when the MS-EVB2 proton distinguishing feature of the MS
is adjacent to a sulfonic acid group. One water of the Zundel feature, which replaces the peak
cation is hydrogen bonded to a sulfonic oxygen, while thesimulation. other This peak in the class
is located away from SO3- (bottom panel, Figure 3). The of proton
the hydronium hydrogens of the
easily moves away from the sulfonic ion through a Grotthuss directly hydrogen bonded to the s
shuttle between these two water molecules. Experimental as the hydrogens of the hydronium
evidence exists for this kind of depletion of the Eigen-type (top panel, Figure 3) as described a
solvation structure in favor of the Zundel-type for predominates other in the CIP region fo
concentrated strong acids, i.e., concentrated HCl solutions. 18,20
the distinction between the hydro
Simulations recently performed with the new self-consistent hydronium and those from the
Figure 3. (topiterative
panel) Representative configuration
multi-state empirical of the “intermedi-
valence bond (SCI-MS-EVB) (bottom panel, Figure 3). There is
Figure 1.14: Representative configurations of the solvation structures observed in the
ate” solvation method,
structures20observed in the simulations using the classical
which is capable of simulating multiple excess tion to the MS-EVB2 radial distrib
hydronium potential. 22,23 Note the water molecule hydrogen bonded
simulations using the classical hydronium potential (left) and the EVB potential (right), protons, has replicated this enhancement of the Zundel solvation
(H-bonds in red) to the oxygen adjacent to the inspected oxygen (aster-
structure over thewater
Eigen-type. In and
the polymer
contextwere
hydronium configurations.
of these experi-
isk). Extraneous hydrogen bonds, molecules, Figure 4 illustrates the diffusio
which were common for both hydration levels (λ = 7, 15). Figure source: Ref. [35]. mental (bottom
excluded for clarity. results,panel)
a classical hydronium
Representative potential ofseems clearly
configuration containing species as a function
deficient.
the Zundel structure observed in the simulations using the MS-EVB2 previously shown15 that the self-di
potential. Relative
Theto the inspected
bottom paneloxygen
of Figure(asterisk)
2 shows onethe
water of the
distribution functions
Zundel cation is located in the contact ion pair (CIP) position, while used in the MS-EVB2 model is ap
the classical and MS-EVB2 hydronium hydrogen atomsthe from experimental value and, no
the other is located in the solvent separated ion pair (SSIP) position.
the sulfonic acid oxygen for the low hydration simulation. The for this Nafion simulatio
diffusion
distinguishing
a weakfeature of the MS-EVB2 curveofis the mentbroad 6 for both degrees of hydratio
small. The diffusion coefficient of water and hydronium ions are both slightly lower in
trajectories indicates
feature,and which
correlation
replaces
between
the hydronium
peak around
the dipole
the sulfonic group the dipole of the ion3 inÅ this
in the classical
of the MS-EVB2 excess proton i
region, with asimulation. This peakforinantiparallel
marginal preference the classical simulation is composed
configurations. earlier work to be roughly half of
A distance ofof
the SSC membrane when compared to Nafion, suggesting that structural diffusion by 3.2the
Å hydronium hydrogens
is certainly too of the direct
long to allow CIP hydronium
hydrogen that are and,not in turn, the classical hydron
directly hydrogen bonded to the
bonding to the hydronium ion and the sulfonic acid oxygen. sulfonic acid oxygen, as well
half of the MS-EVB2 excess proton
as the hydrogens of the hydronium
Further inspection reveals that these configurations are typifiedin the intermediate position
also extends to these simulatio
(top panel, Figure 3)(Figure
as described above. Since
thattheareZundelexperimental
cation
proton hopping may account for the observed higher conductivities in the SSC PFSA
by “bridging” water
predominates
molecules
in the CIP
3, top
region for
panel)
the MS-EVB2 simulation,
6 data. In the present s

hydrogen bonded to both the classical hydronium and a for the apparent under estimation
neighboring theoxygendistinction between
of the same the hydrogens
sulfonic acid group,coming
with thefrom the theCIP
Grotthuss shuttling proton by th
hydronium and those from the SSIP hydronium is blurred
membrane. sulfonic
Figure 3. (top panel) Representative configuration of the “intermedi-
oxygen participating
(bottom panel,
in no hydrogen bonds. When a in the least hydrated simulation)
solvating water
ate” solvation structures observed in the simulations using the classical occupies theFigure
hydrogen 3). There
bondingis also no appreciable
position on this contribu-the classical hydroniums. The MS
hydronium potential.22,23 Note the water molecule hydrogen bondedsulfonic oxygention tothe
the hydronium
MS-EVB2 radial distribution
is unable from these
to occupy this intermediate
trapped in a cage formed by classi
hydronium configurations.
One of the most popular methods to incorporate the Grotthuss mechanism in such
(H-bonds in red) to the oxygen adjacent to the inspected oxygenintermediate
(aster- position and instead is found in the solvent cannot participate in the Grotthus
isk). Extraneous hydrogen bonds, water molecules, and polymer were
separated Figure
position. We4 suggest
illustratesthatthethese
diffusion coefficients
configurations mayfor all proton
for the proton transport. When the
excluded for clarity. (bottom panel) Representative configuration of
represent containing
local minima species
betweenasthea CIP function
and theof hydration.
SSIP. The It hasincreased,
been thereby increasing the
the Zundel structure observed in the simulations using the MS-EVB2
large scale simulations is through the EVB method. One such approach has examined
solvent reorganization
potential. Relative to the inspected oxygen (asterisk) one water of the
shown15by
previouslyrequired thatthetheclassical
self-diffusion
model of to the water potential
allow and the extent of the bonding net
Zundel cation is located in the contact ion pair (CIP) position, while used in the MS-EVB2 model is approximately 30% larger than
the other is located in the solvent separated ion pair (SSIP) position. the experimental value and, not unexpectedly, the water
the solvation properties of the hydrated excess proton in a water cluster of the Nafion diffusion for this Nafion simulation is also larger than experi-
trajectories indicates a weak correlation between the dipole of ment6 for both degrees of hydration. Conversely, the diffusion
the sulfonic group and the dipole of the hydronium ion in this of the MS-EVB2 excess proton in bulk water was shown in
117 membrane [35]. MD simulations were performed with both classical, nondissociable
region, with a marginal preference for antiparallel configurations. earlier work to be roughly half of the bulk experimental value,
A distance of 3.2 Å is certainly too long to allow direct hydrogen and, in turn, the classical hydronium was shown to be about
bonding to the hydronium ion and the sulfonic acid oxygen. half of the MS-EVB2 excess proton diffusion value.15 This trend
hydronium cations and with a single excess proton that was treated by the EVB method.
Further inspection reveals that these configurations are typified also extends to these simulations and the corresponding
by “bridging” water molecules (Figure 3, top panel) that are experimental6 data. In the present simulations, one likely reason
hydrogen bonded to both the classical hydronium and a for the apparent under estimation of the diffusion constant of
Two levels of hydration were studied (λ = 7, 15), revealing the same marked difference
neighboring oxygen of the same sulfonic acid group, with the the Grotthuss shuttling proton by the MS-EVB2 model (certainly
sulfonic oxygen participating in no hydrogen bonds. When a in the least hydrated simulation) is the effect of “caging” by
solvating water occupies the hydrogen bonding position on this the classical hydroniums. The MS-EVB2 proton is artificially
between the hydronium ion solvation structures, as illustrated in Fig. 1.14. As only a
sulfonic oxygen the hydronium is unable to occupy this trapped in a cage formed by classical hydroniums that, in turn,
intermediate position and instead is found in the solvent cannot participate in the Grotthuss hopping process important
separated position. We suggest that these configurations may for the proton transport. When the level of Nafion hydration is
single excess proton was treated with the EVB formalism, at the low hydration level its
represent local minima between the CIP and the SSIP. The increased, thereby increasing the distance between the cations
solvent reorganization required by the classical model to allow and the extent of the bonding network, the ratio between the

diffusion was artificially reduced by caging from the other classical hydroniums ions in

the simulation cell.

Thereafter, a self-consistent variant of the EVB method was used to allow all excess
25

protons to shuttle via the Grotthuss mechanism [36]. The total proton diffusion was,

then, decomposed into vehicular and Grotthuss components which were found to be of

the same relative magnitude, but with a strong negative correlation, resulting in a smaller

overall diffusion for the Nafion system. By contrast, Grotthuss diffusion accounts for 70%

of the total hydronium diffusion in bulk water, with negligible negative correlation of the

two components as seen in Fig. 1.15. Furthermore, correlated motions between the ion

pair were also examined through the distinct portion of the van Hove correlation function,

as shown in Fig. 1.16. At approximately 425 ps, the function develops a peak that is

two to three times the average hydronium density. So, given that a sulfonate anion

occupied a given position 425 ps earlier, the likelihood of finding a hydronium ion in this

same position is nearly three times greater than that of the uniform hydronium density.

This demonstrates a significant correlation in the local ion pair diffusion. The sulfonate

ions effectively act as proton “traps”, limiting the hydronium diffusion primarily to the

long time correlated ion pair motions. This may in part explain why side chain length

variants of Nafion-like polymers, such as the SSC membrane or Aciplex, exhibit varying

transport rates. A shorter pendant chain may restrain the sulfonate groups from deeply

penetrating the hydrophobic phase and trapping the excess protons in the bulk water

region where transport could be the greatest.

In the largest atomistic simulation carried up to date, Knox et al. investigated six of

the most significant morphological models of hydrated Nafion to compare their structural

properties and behaviour [37]. These models shown schematically in Fig. 1.16 are the

cluster-channel model, the parallel cylinder model, the local order model, the lamellar

model, the rod network model, and a “random” model that does not directly assume any

particular morphology. In order to probe multiple hydrophilic clusters and to accurately

measure the signature scattering these authors used, for the first time in this field, large-

scale systems (∼ 2 million atoms and a box length of ∼ 30 nm). Each system was initially
ajectory developed along
not project onto any given
strate the interesting and
ajectory developed along Figure 6. x-Coordinate of a representative trajectory for the CEC; the
wo components of the total
strate the interesting and total trajectory (black), the continuous (vehicular, red), and discrete 26
Figure 6. x-Coordinate
(Grotthuss, of a representative
blue) components. The inset trajectory for the
more clearly CEC; the
displays the
o components of the total
discrete and continuous total trajectory (black), the continuous (vehicular, red), and
stepwise nature of the discrete portion and the continuous nature discrete
of
d through the MSD plots (Grotthuss,
the vehicularblue) components. The inset more clearly displays the
portion.
discrete and continuous stepwise nature of the discrete portion and the continuous nature of
e total diffusion less than
d through the MSD plots the vehicular portion.
sion of either component
e total diffusion less than
he strong negative overlap
sion of either component
wo displacement vectors
e strong negative overlap
less than that of either
wo displacement vectors
usion accounts for ∼70%
less than that of either
fusion (Figure 8) in bulk
usion accounts for ∼70%
ation of the vehicular and
fusion (Figure 8) in bulk
tion of the vehicular and
n. It has been previously
ulation that the diffusion
n. It has been previously
ced by the motion of the
ulation that the diffusion
observed here significant
ced by the motion of the
have quantified the time
observed here significant
ough the distinct portion
have quantified the time
given by34
ough the distinct portion
given by34
Figure 7. Total mean-squared displacement (black) and the continuous
|ri(0) - rj(t)|)" ) (red) and discrete (blue) components of the mean-squared displacement
Figure 7. Total mean-squared displacement (black) and the continuous
in Nafion.
ri(0) - rj(t)|)" ) (red) and discrete (blue) components of the mean-squared displacement
FgRβ(r, t) (6) in Nafion.
Fg (r, t) (6)

nction (normalized by the
ate anions are chosen as
nction (normalized by the
ate anions are chosen as
tion develops a peak that
onium density. So, given
ion develops a peak that
en position 425 ps earlier,
onium density. So, given
um cation in this same
n position 425 ps earlier,
than that of the uniform
um cation in this same
s a significant correlation
than that of the uniform
a characteristic period of
s a significant correlation
noted that, for long times
a characteristic period of
ajectory and short radial
oted that, for long times
ively sparse. For example,
jectory and short radial
and a radial distance of
vely sparse. For example,
Longer times and shorter
and a radial distance of
ly less reliable. Figure 1.15: Figure
onger times and shorter Proton8. MSD
Total in
mean-squared displacement
Nafion (top) and water(black) and the
(bottom). continuous
Figure source: Ref. [36].
(red) and discrete (blue) components of the mean-squared displacement
ted motion, it is easy to Figure
y less reliable. in bulk8.water
Total mean-squared displacement (black) and the continuous
iffusion seen by Spohr et (red) and discrete (blue) components of the mean-squared displacement
ted motion, it is easy to
ed to a flexible model for in bulk water
diffusion of both the pendant chain and the associated hydro-
iffusion seen by Spohr et
ong characteristic period nium ion, it seems inappropriate to generalize this increase in
ed to a flexible model for diffusion of both the pendant chain and the associated hydro-
nd the comparatively low local diffusion to an increase in macroscopic proton transport.
ong characteristic period nium ion, it seems inappropriate to generalize this increase in
nd the comparatively low local diffusion to an increase in macroscopic proton transport.
27

Perfluorosulfonic Acid Membrane Nafion J. Phys.

Figure 9. Distinct portion of the van Hove space-time correlation Figure 10. Radial distr
Figure 1.16: function
van Hoveeq 6space-time correlation function
for the hydronium-sulfonate forgiven
ion pair the the
hydronium-sulfonate
sulfonate ion (black) and h
backbone
pair, given theanion
sulfonate anion as the
as the space-time space-time origin. Figure source: Ref. [36].restricted radial distri
origin.
polymer backbone (red
It may very well be that the perceived increase in diffusion is with an apex formed f
hydrogen center-of-ma
simply an artifact of the correlated motion of the ion pair; that
region).
is, the more labile sulfonate ion of the flexible chain simply
Molecular
drags Dynamics of
the hydronium Nafion
cation as it diffuses about some mean J. Phys.
significant preferenti
position. However, because the pendant chain is ultimately
with all
them, thesurrounded
hydrophobi b
bound to the comparatively static polymer backbone, the motion
of this putative mean position may be inaccessible to the arranged on a lattice an
4. Conclusions
size distribution. Howe
available molecular dynamics time scales.
been randomly placed
3.4. Amphiphilic Association of the Hydronium Cation The proton transp
from a broad, slanted d
and the Hydrophobic Domain. It has recently been demon- region was found in
for the cylinder model
strated that the amphiphilic-like character of the hydrated proton Grotthuss shuttling
amorphous systems m a
observed near the water liquid-vapor interface35 and water nium MSD
scattering shows
spectra, th
espe
clusters36-38 extends to other mixed dielectrics such as methanol- correlated
includes exchange
proposed con
water solutions.39 Although the degree of amphiphilic associa- fusion of the
the spheres, it may transien
be
tion may be somewhat potential dependent,40,41 there is com- attempts to explain pei
topology, resulting
pelling experimental support 42-45 for the surface enhancement more, theseparation
to-center distinct po d
Figure
Figure 1.17: observed 1. Morphological
Morphological modelsmodels of Nafion
of force
Nafion shown
shown asasschematics.
schematics. The Dark shows
represents the ion pair d
model namesin both
usedempirical
in this work field
are
35,38
labeled and
withabtheir
initio simula-
respective sphere and connecting
water; for clarity,
tions.polymerinisparentheses.
36,37
abbreviations not shown. Figure
Dark source:
represents water;Ref. [37]. polymer
for clarity, time scale of
respectively. We usedsevera
isRadial
not shown.
distributions were therefore calculated between the polymer shell
In total, our(∼5.5 Å
results
hydronium cation and the hydrophobic polymer backbone the spheres were
influences not al
the diffus
(including all morphological
carbon and fluorine randomly
pocket ofplaced
Nafion. onAst
six selected modelatoms
systems butofexcluding
hydrated those
Nafionof same replacement
diffuses away from and t
the pendant chain) as well as between
will now be individually described. water and the hydrophobic
model
ologyas“resets”
well astheperio
po
backbone. It has model
A “random” been previously
with randomlydemonstrated
placed water that molecules
the aniso-
the average
mean separation
position relati
tropic solvation
in a box of the
of Nafion hydronium
polymer chainscation results
(basically in a preferential
a model of random linked
fonatetogether with r
ions effectivel
hydrophobic association in the lone pair
6,29 region
water shapes and sizes in the polymer) has been studied. of the This
ion’s
39 “slanted” distribution
nium diffusion prim
solvation
model hasshells.
been usedWith thisininthe
before mind, the radial
literature distribution
to try to predict a which basically results
28

built to closely approximate the proposed hydrophilic cluster structure in a given model.

Formation of connecting bridges between clusters and the resulting percolation was ob-

served at the molecular level in all of the models, with significant involvement of the

sulfonate groups. The solvent-accessible surface area (SASA) was used to measure the

hydrophilic-hydrophobic interfacial area of each model, revealing for the first time large

magnitude SASA values (∼ 1000 m2 /g) in the rod model. The authors have pointed out

that such a large interfacial area underscores the strong nanophase segregation behavior

of Nafion. It was suggested that the high interfacial area may also enhance proton trans-

port because of the amphiphilic nature of excess protons, which have been observed to

prefer hydrophilic-hydrophobic interfaces. Most interestingly, the structural and scatter-

ing spectra of the nonrandom models were found to be closely comparable, emphasizing

the insensitivity of the characteristic scattering peak to widely varying geometry and

model differences.

Finally, a dissipative particle dynamics study was carried out recently to elucidate

the role of MW on the hydrated morphology of SSC PFSA membrane [38]. The increase

of MW induces aggregation of the fluorocarbon backbone that minimizes chain bend-

ing forces, while maintaining a phase-separated structure, and results in larger, more

elongated water domains.


Part I

Methodology Development

29
30

Chapter 2

Force Field Development

The quality of any MM simulation is largely determined by the suitability of its force

field. Unfortunately, there is no single, universal force field that can correctly predict all

properties, across the vast array of known compounds. Ideally, any application of a given

force field to a new class of compounds should follow a careful investigation of its validity

within the new context. However, this route was not adopted in the only theoretical study

thus far carried out on a large-scale SSC system [39]. There the force field parameters

were taken directly from the force field employed for Nafion. The inconclusive results

of this study fail to even qualitatively distinguish the proton mobilities in the SSC and

Nafion systems. Hence, for our modelling we have chosen to first modify and tailor the

Nafion force field prior to its application to the SSC polymer.

When examining the structural or dynamic differences between the SSC polymer and

Nafion the question may well be raised whether these differences are indeed present in

the real systems and not just the artefacts of the different force fields. Let us consider

the backbone lengths, LSSC and LN af , of a SSC polymer and a Nafion polymer strand

with the same number of backbone carbon atoms. This length will be measured at the

minimum energy conformation as found by MM. Since this equilibrium conformation

minimizes the total MM energy of the polymer it is a function of all the terms in the force

field, including the torsion potentials and the non-bonding interactions. The difference

in the lengths can be written as LSSC − LN af = (L0SSC + δSSC ) − L0N af + δN af =

L0SSC − L0N af +(δSSC − δN af ). Here the primed lengths are the exact lengths (e.g., from

ab initio calculations) and the δ symbols represent the errors of the MM estimates. Let

us now determine how well the exact difference in the lengths L0SSC − L0N af is reproduced
31

by the force field. From the above formula we can see that a good representation will be

achieved in two cases: if δSSC , δN af are close in magnitude and of the same sign, or if they

are both close to zero. Therefore, if we have two well parameterized classical force fields,

one for the SSC polymer and one for Nafion (i.e., both δSSC , δN af are close to zero), we

can get an accurate estimate for the difference in backbone lengths of the two polymers.

On the other hand, one universal force field with the same number of parameters will

necessarily have larger errors. In this case a good estimate can still be achieved, but only

if δSSC , δN af are, fortuitously, of the same magnitude and sign. Therefore a unique SSC

force field will improve, rather than impede the comparisons with Nafion.

The key feature that our force field aims to reproduce is the conformation of the

polymer molecule. The leading role in determining the conformation is played by the

torsion potential. While the bond lengths and angles of an atom group are restricted

about a single equilibrium value, dihedral potentials are modeled with periodic functions

allowing for a number of low-energy conformations. By specifying the sequence of dihedral

angles one can uniquely determine the morphology of the polymer, an approach widely

used in proteins [40]. Recent ab initio modelling of the SSC system has suggested that

the conformation of the backbone has important effect on chemical properties under low

hydration conditions. Extensive electronic structure calculations of oligomeric fragments

possessing two side chains revealed that the extent of the separation of the sulfonic acid

groups along with the conformation of the backbone significantly affect the propensity

of the acid groups to dissociate [26, 28]. Kinked conformations of the backbone give

rise to closer proximity of the acid groups, stronger binding of the water molecules and

enhanced proton dissociation at lower degrees of hydration. As these properties needed

to be correctly reproduced in our SSC specific force field, we take special care in our

treatment of the torsion potentials (described below).

The force field we present here is based on the generic DREIDING force field devel-
32

oped by Mayo et al. [41] and that has been used and supplemented by several authors

investigating the Nafion system [25,36,35]. Our adaptation of the DREIDING force field

to the SSC polymer is based on extensive ab initio calculations performed by Paddison

and Elliott [29]. Six different torsion profiles were computed in their study of a two-side

chain oligomeric fragment altogether comprising more than 200 geometries. The average

bond lengths and angles were extracted from these structures and their values were used

as the equilibrium parameters r0 and θ0 in the force field potentials. The force constants

of the stretching and bending motion were assumed to be the same as in Nafion (see

Appendix A, Fig. A.1, Tables A.1 and A.2). The atom charges were calculated through

a Mulliken population analysis of the ab initio structures of the SSC fragment, while

the Lennard-Jones parameters  and σ were assumed to remain unchanged from Nafion

(Appendix A, Table A.4).

The last component of the SSC force field is the torsion potential and it is the corner

stone of our force field since it ties together all the other potentials and ensures agreement

with the ab initio torsion barriers. From Paddison and Elliott’s oligomer geometries one

can further extract any structural information like dihedral angles, including those whose

profiles were not explicitly studied in their work. One such dihedral is the O4:-S6:-C2S:-

C2O (where the colon indicates a wildcard1 ensuring that both the neutral and ionized

side-chains are covered, see Fig. 2.1g). Obtaining a MM representation of the profile of

this dihedral angle and further assuming that it does not change after proton dissociation

makes it possible to go beyond the neutral system modeled in the ab initio work. Another

compelling reason for including this torsion potential in the MM energy balance is the

accompanying exclusion of the 1-4 non-bonding interactions (i.e., between the atoms O4:

and C2O). The Coulomb interaction between these two atoms, especially in the case of an

ionized chain, is quite large which makes fitting the ab initio data very difficult. For that
1
In computer (software) technology, a wildcard character can be used to substitute for any other
character or characters in a string.
33

reason we have seven torsion potentials as shown in Fig. 2.1) and Appendix A, Table A.3.

It should be noted that our force field does not include all possible dihedral angles that

one can find in the polymer. One such group of dihedral angles that are absent is that

which involves fluorine atoms. The reason we limit the number of dihedral angle types in

the force field is to minimize the computational work during the MD simulation. As our

force field (including the torsion potential) completely replaces the potentials defined in

the generic DREIDING force field [41], any terms from the latter that do not carry over

to the SSC force field should be considered zero. This implies, however, that such terms

are not simply missing but that their effect has been adsorbed into the parameters of

the remaining terms. Our force field should be considered complete in the sense that we

have obtained the best possible correspondence to the ab initio data given the number of

fitting parameters in the torsion potential. In the next few paragraphs we show exactly

how these parameters’ values were obtained.

We define the torsion energy of an oligomeric fragment with atom coordinates Ri as

the difference of the molecule’s ab initio energy and the sum of the energies of the other

bonding and non-bonding potentials:

Etors (Ri ) = EQM − Ebonds − Eangles − ECoulomb − ELJ (2.1)

The index i runs over all of the 200 SSC oligomer structures, e.g. R1 represents the

coordinates of the atoms of the SSC structure in the first data point of the first torsion

profile. This assignment bestows an additional role to the torsion term in enabling it to

absorb the error of the model, which arises from the assumption that an ab initio energy

can be split into a sum of MM terms. Alternatively, the torsion energy in Eq. (2.1) can

be viewed as the residual energy after all other classical terms have been accounted. The

torsion energy is modeled as a sum of dihedral potentials according to:


X
7 X
Nm
Ef it (Ri ) = am cos(bm ϕim,l − cm ) (2.2)
m=1 l=1
34

C1O C1O C1O C1O

O3 O3 O3 O3

F2O C2O F2O F2O C2O F2O F2O C2O F2O F2O C2O F2O

F2S C2S F2S F2S C2S F2S F2S C2S F2S F2S C2S F2S

O4 S6 O4 O4 S6 O4 O4 S6 O4 O4 S6 O4

O3H O3H O3H O3H

H3 H3 H3 H3

(a) (b) (c) (d)

F1

C1 C1O C1O

F1 O3 F1 F1 F1 F1 O3

F2O C2O F2O C1 C1 C1 C1 F2O C2O F2O

F2S C2S F2S F1 F1 F1 F1 F2S C2S F2S

O4 S6 O4 (f) O4 S6 O4

O3H O3H

H3 H3

(e) (g)

Figure 2.1: The seven dihedral angles of the SSC force field illustrated on polymer
segments with a protonated sidechains. Only the first two dihedral angles (a,b) actually
require the protonated form of the sidechain as they include the hydrogen atom or its
adjacent oxygen. A backbone dihedral angle (f) is composed of any four backbone atoms,
which can also be terminal or branch carbon atoms. The last dihedral angle (g) is the
additional one defined in this work.
35

where the outer sum is over the seven distinct types of dihedral angles shown in Fig. 2.1

and the inner sum is over all instances of the given type in the oligomer2 . We choose to

optimize the set of parameters {a, b, c} so that the difference between any two oligomer

structures i and j is zero:


Efi it − Efjit − Etors
i j
− Etors = ∆Efi,jit − ∆Etors
i,j
≡ ∆Efi,jit−tors (2.3)

In the ideal case, where the ab initio torsion profiles match our force field, the ∆Efi,jit−tors

must vanish. Hence, the absolute value (or the square) of ∆Efi,jit−tors is a measure of the

quality of the fit for the oligomer structures i and j. This allows us to define a penalty

function for the entire set as:


Ns 
X 2
i,j
P (a, b, c) = ∆Ef it−tors (2.4)
i>j

As we do not limit i and j to belong to data points from the same torsion profile our fit

will ensure that the ab initio energy differences are evenly matched across all the profiles.

The form of our dihedral potential, Edih = a cos(bϕ − c) used in calculating Ef it ,

warrants further description. The constant term that is present in other definitions of

a dihedral potential [42] is redundant in our case owing to the fact that we are only

fitting to energy differences between molecules possessing the same type and number of

dihedral angles (i.e., different conformations of the same oligomer). Even if a constant

term was included in the potential it cannot improve the fit, as it will cancel in Eq. (2.3).

Finally, we note that the absence of such a term has no effect on the forces. When

the penalty function is at a minimum its derivatives (hereafter referred to as “forces”)

with respect to the parameters am , bm , and cm will be zero for all m = 1, 2, ..7. By

deriving analytical expressions for these forces that act on the parameters we can employ

an efficient minimization scheme for P , e.g. a steepest descent (SD) minimization [43].
2
For example, there are two instances of the dihedral angle shown in Fig. 2.1g as there are two O4
atoms.
36

The first derivative of the potential with respect to each of the unknown parameters is

straight forwardly obtained:


i,j
!
∂P X s
i,j
N
∂∆Ef it X s N
∂Efi it ∂Efj it
Fam =− = −2 ∆Ef it−tors = −2 ∆Efi,jit−tors −
∂am i>j
∂am i>j
∂am ∂am
X
Nm X
Ns

= −2 ∆Efi,jit−tors cos(bm φim,l − cm ) − cos(bm φjm,l − cm )
l=1 i>j

∂P X m X sN N

Fbm =− =2 ∆Efi,jit−tors am sin(bm ϕim,l − cm )ϕim,l − sin(bm ϕjm,l − cm )ϕjm,l
∂bm l=1 i>j

∂P X m X s N N

Fcm = − = −2 ∆Efi,jit−tors am sin(bm ϕim,l − cm ) − sin(bm ϕjm,l − cm ) (2.5)
∂cm l=1 i>j

Since an SD algorithm had already been developed in the context of our general MD

code, it was expedient to use this algorithm for the present purposes. The actual im-

plementation involved treating the 7×3 unknowns am , bm , and cm as the coordinates of

seven dummy atoms in 3D space. The forces that act on these atoms were calculated

by Eq. (2.5) with the “energy” of the system being the penalty function P in Eq. (2.4).

The lowest possible value for P along with the corresponding values of the dummy atom

coordinates (i.e., the force field parameters) were obtained at the completion of the min-

imization run and the latter are collected in Appendix A, Table A.3.

The most difficult torsion profile to accurately match was that around the C1O-

O3 bond [29] where the side chain attaches to the polymer backbone (see Fig. 2.1).

This profile has the highest rotational barrier being more than 38 kJ/mol and highly

asymmetric. The latter suggests significant contributions arising from Coulomb and

Lennard-Jones interactions resulting from the crowding of the atoms near the point of

attachment. A comparison of the ab initio results with our MM force field energies is

shown in Fig. 2.2 and indicates very good agreement in most regions of the PES. A

second torsion profile for rotation about the C2S-S6 bond is displayed in Fig. 2.3 and

likewise shows good correspondence between the ab initio data and the MM results.
37

Figure 2.2: Comparison of the ab initio and classic torsion profiles (with contributions
from all MM terms) around the C1O-O3 bond.
38

Figure 2.3: Comparison of the ab initio and classic torsion profiles (with contributions
from all MM terms) around the C2S-S6 bond.
39

10
"

5
a cos(bϕ − c) kJ mol−1

C2S:
!

0 H3O
4O O4
H3
−5

−10
−200 −100 0 100 200
ϕ [degrees]

Figure 2.4: Energy profile of the dihedral potential for the angle H3–O3H–S6–C2S: in
the SSC force field (left) and a Newman projection of the dihedral angle (right). The
small discontinuity of the potential at ±180◦ is due to the unconstrained optimization of
the periodicity parameter in our force field (see Appendix A, Table A.3).

So far we have seen that the simplified dihedral potential a cos(bϕ − c) can be success-

fully parametarized to reproduce the relative energy of SSC conformers. When defining

the torsion energy in Eq. (2.1) we noted the fact that it can also be viewed as the residual

energy of the molecule after all other terms have been accounted for. Therefore, it is not

yet clear if we can assign any physical meaning to the periodicity and phase parameters

b and c, respectively, as we could normally do for torsion potentials [42]. With this in

mind let us now take a closer look at the periodicity and phase parameters shown in

Appendix A, Table A.3. If we plot the dihedral potential for the first angle in the table

that includes OH group of sulfonic acid (shown top left in Fig. 2.1) we get the energy

curve in Fig. 2.4. Two low energy conformations can be seen here at about −150◦ and

119◦ separated by a transition state at −16◦ . The location of the minima is close to what

we can predict based solely on the electrostatic interactions. The hydrogen atom is most

strongly attracted to the lone pairs on the double bonded oxygens in the SO3 H group.
40

From the Newman projection in Fig. 2.4 we can see that the closest approach to the oxy-

gen atoms takes place when the H3–O3H–S6–C2S: angle is ±120◦ . This coincides well

with what is known from other systems where the energy profile of one fold periodicity

(i.e., for b = 1) is explained in terms of the orientational preference of dipole pairs [42].

Hence, even though our dihedral potentials have exclusively adsorbed the error of the

MM representation they still behave like the regular torsions.

One unexpected result is the value of the periodicity parameter b in the last torsion

profile in Appendix A, Table A.3. For the O4:–S6:–C2S:–C2O dihedral angle (shown

bottom right in Fig. 2.1) we get periodicity of about one half. Such periodicity does

not make sense if we think of the dihedral potential as a truncated Fourier expansion.

It also represents a perplexing case of a dihedral angle which has to be turned 4π be-

fore its energy repeats. The PES of this particular angle was not part of the ab initio

calculations of Paddison and Elliott [29] since for unionized side chains this dihedral is

redundant. Obviously the parameter values obtained here minimize the penalty function

P in Eq. (2.4), which does not include any periodicity considerations. However the one

half periodicity of the O4:–S6:–C2S:–C2O dihedral angle is never actually an issue. Since

ϕ in the potential expression a cos(bϕ − c) is always given in the range [−π, π] the longer

than usual period of this potential function never comes into play. This is not unusual.

The limited range of the angle is what allows us to model a bending angle potential, like

H–O–H in water, with a harmonic potential which is not periodic.


41

Chapter 3

Virial Formulation For Periodic Systems

Later, in this work we present the methodology needed to simulate polymeric structures

of infinite length by ensuring the continuity of the polymer chains across periodic bound-

aries. The rationale for using periodic boundaries in the first place is to minimize the

boundary effects experienced by the atoms near the cell surface. Furthermore, the prop-

erty that we want to obtain from our simulations is the proton diffusion coefficient, or in

other words, the distance that the protons have travelled, and PBC allow seamless move-

ment of all species. Evaluating the interactions in these periodic systems is not trivial

due to the infinite number of interacting pairs. Currently, there is a dichotomy in the

evaluation methods with the Lennard-Jones potentials treated with the cutoff plus Long

Range Correction (LRC) method and the electrostatic potential treated with the Ewald

summation or similar periodic methods [44]. Recently, a distinctively new approach was

introduced in the Isotropic Periodic Sum (IPS) method where a cutoff is combined with

spherical periodic images [45, 46]. This method is simple to implement, and most impor-

tantly, valid for all interaction potentials. In this chapter we present the IPS method in

some detail and derive a more accurate expression for its virial1 .

Cutoff methods enable the study of large systems by considering only a small, local

region where the interactions are computed exactly. Beyond the cutoff the interactions

are computed based on an approximate model which embodies some of the features of

the system. The LRC method, for example, uses the number density to construct a

structureless, infinite region. The IPS method goes one step further by using the density

along with the exact atom positions of the local region. Thus, in IPS the infinite region
1
Here we use Clausius’ meaning of “virial of force” as a function of the forces in the system.
42

i Rc

Figure 3.1: Local region (solid sphere) replicated in a spherical shell. The image spheres
fill up the entire volume of the 3D shell. A special arrangement of the images ensures
zero forces at the local region boundary (see Ref. [45]).

is represented as local region replicas (referred to as images), arranged uniformly in

spherical shells, see Fig. 3.1. Exact, as well as approximate formulations are available

for general 1/rn and exponential potentials. The formalism developed in Ref. [45] can be

applied to other potentials as well, even those where the summation over the image shells

does not converge (e.g., 1/r). In the simpler case of 3D isotropic systems (i.e., periodic

in all three directions) the more recent paper [46] provides convenient fitting polynomials

that are straightforward to implement in a computer code. The work also extends the

IPS method with a discrete fast Fourier transform for heterogeneous systems.

A comprehensive comparison between the LRC and IPS methods applied to the

Lennard-Jones model can be found in Ref. [47]. The study has shown that the two

methods predict nearly identical results for both thermodynamic and transport proper-

ties when the cutoff is sufficiently long. One interesting exception was the pressure, where

for short cutoffs the results were substantially different and good agreement required a

notably longer cutoff. Conventionally, the choice of the cutoff radius Rc is based on some
43

Rc

Figure 3.2: If L becomes smaller than 2Rc the interactions in the primary cell, as well
as all its neighbors have to be evaluated explicitly.

system specific information (e.g., the σ parameter of the Lennard-Jones model) and com-

putational considerations (like the number of steps that can be computed in a given time

interval). The original version of the IPS method assumes that Rc is constant and not

explicitly dependent on the simulation cell dimensions. This assumption has important

repercussions for the virial and pressure regardless of the simulated ensemble. In NPT

simulations with constant Rc the number of particles in the local region will increase

upon compression of the simulation cell, resulting in unnecessarily detailed and long

calculations. If the dimensions of the simulation cell fall below 2Rc we must explicitly

consider the primary cell along with its neighbors, (see Fig. 3.2), a complication rarely

implemented in a computer code. On the other hand, an expansion of the simulation cell

combined with a constant Rc will result in loss of detail, potentially leading to a local re-

gion that is no longer representative of the whole system. However, it is entirely possible

to minimize such side effects resulting from constant cutoff NPT simulations by ensuring

a good initial guess for the density as well as small volume fluctuations. The bigger
44

question, explored in this chapter, is whether, for any simulated ensemble, a cutoff not

explicitly dependent on the simulation cell dimensions would produce the correct virial.

This chapter is organized as follows: We first look into the different contributions to the

potential energy in the IPS method and the modifications necessary for the boundary

term. We then introduce the virial and present two ways of calculating it, and finally

demonstrate the merits of the new virial formulation in two test cases.

The instantaneous potential energy of a system represented by two regions, local

and infinite, can be split into two terms. The energy contribution from the local region

Ulocal is equivalent for any cutoff method since it is calculated exactly. The energy of

the infinite region is labeled here as Uperiodic , to imply that it has been calculated with

the IPS method2 . At the boundary between the two regions it is desirable to maintain

continuity of the energy, forces3 , and virial in order to avoid pressure artifacts and improve

the numerical stability in MD simulations [44, 48]. Such side effects can be avoided by

shifting the potential at the boundary, which is an approach analogous to the well known

truncated and shifted potentials [44]. The shifting is corrected later by a restoring term.

These two artificial contributions to the energy are collected in Uboundary , giving a total

potential energy:

U = Ulocal + Uperiodic + Uboundary (3.1)

An illustration of the effect of the boundary term on the total energy fluctuations is

presented in Fig. 3.3. Since Uperiodic is calculated from replicas of the local region, it will

suffer from the same boundary truncation inherent in Ulocal . Therefore, the shifting and

restoring of the potentials that happens in Uboundary has to be carried out for both the
2
A more explicit notation was used
P in Ref. [45]. For the potential
P designated there as εij (rij ) we use
1/rn . Our Ulocal corresponds to 12 ε (rij ) and Uperiodic to 21 φ (rij , Ωi ).
rj ∈Ωi rj ∈Ωi
3
In the IPS method the forces with respect to atom distances −∂U /∂rij are continuous at the
boundary by construction. There are also other forces we have to consider with respect to the dimensions
of the simulation cell and the cutoff radius in order to obtain a correct virial. Therefore the force
continuity built in the IPS method does not automatically lead to a virial continuity.
45

603.6
Uboundary 0
Uboundary =0
603.8

604.0
H [kJ/mol]

604.2

604.4

604.6

604.80 100 200 300 400 500


step

Figure 3.3: Hamiltonian conservation in a short MD trajectory of a Lennard-Jones


system. The curves show the total energy (i.e., kinetic and potential) with (solid line)
and without (dashed line) boundary correction. Including the Uboundary term in the
potential energy results in smaller oscillations and a smoother curve.
46

local and the periodic energy terms. As the system virial includes contributions from

the boundary correction we will first examine the nature of Uboundary in more detail.

The notation and grouping of terms adopted here is slightly different4 from the original

work in order to facilitate the connection with the virial introduced later. However,

the concepts remain the same. The pair-wise sums, Ulocal and Uperiodic , by construction

include the ij-pairs where atom j is inside the local region of i, i.e., rij < Rc (where

rij are calculated after applying the minimum image convention (MIC) with respect to

atom i). For these pairs only, we evaluate the boundary energy after setting atom j at

the boundary of i, i.e., for rij = Rc . Thus, both the restoring and shifting components

of Uboundary can be expressed as:

X
N
[Ulocal (rij = Rc ) + Uperiodic (rij = Rc )] (3.2)
i<j

Next, the restoring term will be approximated by removing the rij < Rc requirement.
∀ ∀
We introduce Ulocal and Uperiodic which represent the same potential functions, but with

extended domains. This implies that all j atoms are considered whether or not they

reside inside the local sphere of i. For the restoring energy we can then write:

4πRc3 X  ∀
N


Urestoring = Ulocal (rij = Rc ) + Uperiodic (r ij = Rc ) (3.3)
3L3 i<j

The basis of this approximation is that the local region is representative of the whole

system, and the interactions (both local and periodic) of the atoms within the cutoff are

a fraction of the interactions in the system (no limits on rij being imposed). This fraction

is equal to the volume of the sphere divided by the cell volume. Effectively, the distance

dependence has been removed from the restoring term as all rij have been set equal to

Rc , regardless of their actual value. For the shifting term, on the other hand, we can
4
In the original formulation the shifting term is grouped with the local and periodic energy terms to
give, what the authors call configurational energy. Only the restoring term is referred to as the boundary
energy Ebound . Thus, the definitions only differ in the grouping of the terms.
47

adopt a more convenient form, with the Heaviside step function used to emphasize the

shifting energy’s depends on rij :

X
N
 ∀ 

Ushifting = Ulocal (rij = Rc ) + Uperiodic (rij = Rc ) Θ (Rc − rij ) (3.4)
i<j

Finally, the boundary energy is calculated as Uboundary = Urestoring − Ushifting .

An important difference from the original IPS work lies in the atom pairs being con-

sidered in the boundary correction. There the electrostatic and Lennard-Jones restoring

terms (i.e., Ebound ) are derived by including all self-pairs, and correspondingly, these pairs

are included in the shifting terms. However, there is no physical basis for considering

the self-pairs in Uboundary . The continuity that the latter brings is only relevant for one

central atom i and a different atom j, moving in and out of the local region of i5 . In-

cluding the self-pairs in the boundary correction, however, results in a spurious energy of
 3  
4πRc ∀ ∀
3L3
− 1 Ulocal (rii = Rc ) + Uperiodic (rii = Rc ) per atom, as seen from Eqs. (3.3) and

(3.4). In most practical cases all bonded interactions of atom i (such as bonds, angles and

dihedrals) will be entirely within its local region. Regardless of the type of the bonded

ij-interaction there is never a continuity issue if rij is always less than Rc . Likewise,

all intramolecular interactions in small molecules (like solvents) can be excluded from

Uboundary if these molecules would always fit inside the cutoff sphere. A demonstration

of the effects of an “all-pair” boundary correction will be presented later, following the

calculation of the virial. Since all components of the potential energy are now known we

proceed with the calculation of the virial and pressure.

The thermodynamic pressure is most conveniently derived in the NVT ensemble. A

precise derivation can be found for example in Ref. [49]. Here we follow the simpler

approach presented in Ref. [50] that allows us to focus on periodic systems. Thermody-

namic pressure is the volume derivative of the total energy (sum of potential and kinetic
However, the self-pairs must be included in the calculation of Uperiodic in Eq. (3.1) in order for the
5

images to have the same charge and particle density as the local region.
48

energies) at constant temperature: − (dE/dV )T . The kinetic energy contribution to the

pressure is identical to ideal gas pressure resulting in:


 
dU
P V = N kT − V (3.5)
dV

As usual, for MD simulations, we will assume ergodicity which allows us to consider

instantaneous values for the energy, virial and pressure. The ensemble average of these

quantities can then be calculated from the instantaneous values averaged over the given

trajectory. Following Ref. [50] the negative of the instantaneous virial, V (dU/dV ), can

be written as:
!
dU (r, L, Rc ) 1 X ∂U drij ∂U ∂U dRc dL
V =V + +
dV 2 i,j ∂rij dL ∂L ∂Rc dL dV
!
1 X drij ∂U ∂U dRc 1
=V − fij + +
2 i,j dL ∂L ∂Rc dL 3L2
!
1 1X drij ∂U ∂U dRc
= − fij L + L+ L (3.6)
3 2 i,j dL ∂L ∂Rc dL

The derivation is valid for pairwise additive potentials, with rij calculated after applying

MIC (with respect to atom i). Assuming that the ij-atom distances, rij , scale linearly6

with the dimension of the simulation box L then according to Euler’s theorem we get

L (drij /dL) = rij . Similarly, (for convenience) we choose Rc to be linearly dependent on

L, we get L (dRc /dL) = Rc . This allows us to obtain the virial as a scalar product of the

forces and the arguments of the potential energy:


!
dU (r, L, Rc ) 1 1X ∂U ∂U
−V = fij rij − L− Rc (3.7)
dV 3 2 i,j ∂L ∂Rc
P
As explained in Ref. [50] the standard virial expression fi ri /3 is incomplete for

systems with PBC. An additional −L∂U/∂L term must be included in order to take

into account pressure contributions from changes in the image cell separations. However,
6
This choice affects only the instantaneous pressure, but not the ensemble-averaged one. For an
in-depth explanation please see Ref. [51].
49

in IPS it is not L that determines the image separations (as in rectangular cells) but

rather the diameter of the local sphere 2Rc . Therefore, a physically correct formulation

of the virial in the IPS method requires the presence of the extra −Rc ∂U/∂Rc term. In

order to obtain this term we have to ensure that dRc /dL in Eq. (3.6) is nonzero and

Rc is linearly dependent on L. Accordingly, the cutoff must be an explicit, homogenous

function of the side of the simulation cell of order one, i.e., Rc = Rc (L). It is evident that

the cutoff contribution to the virial will also be found in any constant volume simulation

(i.e., NVE or NVT) since no difficulties are encountered in calculating the derivative of

Rc (L) for a fixed value of the argument. Let us note that the derivation of Eq. (3.7)

does not assume any specific form for U (r, L, Rc ), including the periodic and boundary

contribution (if such is present). Consequently it is not just limited to the IPS method.

Examples of this can be found in Refs. [52, 53] where the authors have employed the

cutoff contribution to the virial in LRC, constant volume simulations. Before moving to

the practical illustrations of the theory we will consider whether there is a simpler way

of calculating the virial from Eq. (3.7).

To evaluate Uperiodic in IPS it is necessary to integrate over an angle θ that determines

the location of the image sphere in a given shell and then sum over all shells - from the

one adjacent to the local sphere (with index m = 1) to infinity. Thus, for a 1/rn potential

the IPS energy of one interacting pair in the m-th shell is determined from the integral:

1 sin θdθ
(3.8)
2 (r2 + 4m2 Rc2 + 4mrRc cos θ)n/2
0

The denominator originates from applying the cosine theorem to the angle π − θ in

Fig. 3.4. The integral depends on both r and Rc , but the integration variable can be

changed to the reduced distances r/Rc obtaining, thereby, a factor of 1/Rcn outside the

integration. From the assumption that r is linearly dependent on L, and the requirement

that this be true for Rc as well, it follows that the IPS energy is homogenous in L of
50

m-th shell image

r
π−θ j’
2mRc

θ
i r j

Figure 3.4: The distance between the central particle i and the image particle j 0 in the
m-th shell is a function of the angle θ (determining the position of the image sphere in
the shell), the i − j distance r in the primary cell and the distance between the sphere
centers 2mRc .
51

order −n. The result is valid for 3D isotropic systems regardless of any approximations

and fittings applied to the integral or the subsequent shell summation over m. The

same homogeneity in L is present in the local energy Ulocal , and can be proven for the

boundary correction as well7 . Since all components of the potential energy of Eq. (3.1)

are homogenous in L with an order of −n (for a 1/rn potential) we can calculate the

virial as follows, (see Ref. [49]):

dU (r, L, Rc ) dU (r, L, Rc ) dL 1 dU (r, L, Rc ) n


−V = −L3 =− L = U (r, L, Rc ) (3.9)
dV dL dV 3 dL 3

Any representation of the infinite region that ensures energy homogeneity equivalent to

that of the local region can make use of Eq. (3.9). Consequently, the above result is valid

for both the IPS and the LRC methods.

The benefit of the above equation over Eq. (3.7) is that only energy fractions are

needed instead of energy derivatives. Thus, a Monte Carlo simulation will not have to

resort to any force evaluations in order to obtain the virial. Moreover, the cutoff force

−∂Uperiodic /∂Rc in the IPS method is as laborious to calculate as the atom pair forces.

In a MD simulation, where the atom pair forces are required to propagate the system,

considerable effort can be saved by avoiding the calculation of the cutoff forces and

employing Eq. (3.9). This remarkably simple result is well known for the local energy,

while its validity for the total energy, including the complicated periodic and boundary

terms, is not obvious. Hence, allowing for a cutoff dependent on system dimensions, and

a more physically accurate picture, simplifies the calculation of the virial.

We now consider a simple analytical example, which is an adaptation of the one

appearing in Ref. [50] for rectangular simulation and image cells. Consider a physical

system containing a spherical region of radius Rc with a single particle located at its
7
Since Rc3 is a homogenous function in L of order three, the volume fraction 4πRc3 /3L3 in Eq. (3.3)
will be zero order in L. The Heaviside step function in Eq. (3.4) has the property Θ (Lx) = Θ (x) for
any positive L so it is also zero order in L. Therefore, the boundary correction will have the same order
as the corrected energy.
52

center, along with infinitely many copies distributed in spherical shells of the same density

(as in Fig. 3.1 with i being the only particle). For a 1/rn potential the virial due to the

single particle in the center can be evaluated directly by a summation over the shells.

Instead of integrating over the angle θ (see Fig. 3.4), we use the fact that there are

24m2 + 2 images in the m-th shell (see Ref. [45]), each of them at 2mRc from the center.

Assuming homogeneity of all atom positions in the physical system the Euler theorem

gives the virial as:



n n X 24m2 + 2 n
Wdirect = U (r, Rc ) = n = Uperiodic (r = 0, Rc ) (3.10)
3 3 m=1 (2mRc ) 3

In the end result Uperiodic (r = 0, Rc ) corresponds to the IPS periodic energy calculated

with r set equal to zero. On the other hand, with the original formulation of the virial

we would get:

1X
N
4πRc3
Woriginal = ri fi + Ebound = 0 + [Ulocal (r = Rc ) + Uperiodic (r = Rc , Rc )]
3 i=1 3L3
 
4πRc3 1
= + Uperiodic (r = Rc , Rc ) (3.11)
3L3 Rcn

The only remaining terms above are those coming from the boundary correction. As

we have emphasized before, the boundary correction in the original paper includes the

self-pairs, which when combined with a system dimensions independent cutoff produce

the unexpected result of Eq. (3.11). Let us now see if our modifications can recover the

exact virial. With Eq. (3.9) the IPS virial for the center particle becomes:

n n
Wnew = U (r, L, Rc ) = [Uperiodic (r = 0, Rc ) + Uboundary ]
3 3
n
= Uperiodic (r = 0, Rc ) (3.12)
3

All boundary terms vanish since our formulation excludes the self-pairs from the bound-

ary correction. We see that this result obtained from an IPS periodic energy calculation

is equivalent to the direct summation result in Eq. (3.10).


53

Table 3.1: Virial comparison for a 256 particle Lennard-Jones model. In the second
column use is made of the virial and boundary terms from the original IPS formulation
[45], the third column utilizes the formulation described here, i.e., Eq. (3.9) with self-
pairs excluded from the boundary term. Fourth column shows the virials obtained with
the LRC method [53]. All three simulations are constant volume simulations (canonical,
canonical and microcanonical, respectively) hence Rc remains constant. The density
corresponds to the liquid region of a supercritical isotherm with 2.5 times the critical
temperature. All numerical values are in reduced units.

parameters Woriginal a Wnew WLRC b

ρ = 0.8 T = 3 Rc = 3.4 2108±1c 2079±1 2085


a
Simulation details: Our IPS implementation is based on the fitting polynomials appearing in
Ref. [46]. Starting from random initial positions each system was optimized for one hundred steps
with the SD method [43], followed by one hundred steps of Adopted Basis Newton-Raphson (ABNR)
optimization [54]. Random velocities with zero net momentum were assigned corresponding to the
target temperature of 3. Each system was equilibrated for 2000 steps with the generalized Nose-
Hoover method [55] using a thermostat period of 0.0625 and a time step of 0.003. The NVT
simulation was continued in the production run for another 4×105 steps.
b
The virial has been calculated as P V − (N − 1) kT using the values from Table IV in Ref. [53].
c
Lower bounds on the uncertainties were estimated with the method described in Ref. [56].

For our next example we examine a many particle Lennard-Jones model in the state

displayed in Table 3.1. The IPS virial calculated with the new formulation is only about a

quarter of a percent different from the LRC value, whereas with the original formulation

the difference is nearly four times larger (i.e., a 1% difference). This result indicates that

the IPS and LRC virials (and pressures) are actually much closer than had been previously

seen for a state of similar density [47]. Most importantly, the excellent agreement has

been obtained for the relatively short cutoff of 3.4σ. In the LRC method the assumption

is that the pair-correlation function g(r) is unity beyond the cutoff, hence the probability

of finding an atom in the thin spherical layer between R and R + dR is determined by

the number density alone. The IPS method operates on a similar, coarse-grained version

of g (r) = 1, applied to the image spheres, instead of the individual atoms. Thus, in

every spherical shell between (2m − 1) Rc and (2m + 1) Rc , one finds the exact number

of image spheres (i.e., 24m2 + 2) dictated by the density. This close similarity between
54

the key assumptions of the two methods results in nearly identical virials when they are

calculated in a consistent manner. The IPS method has the noteworthy advantage of

handling electrostatic potentials, allowing a simple, universal way of modelling all types

of interaction in the infinite region.

To summarize, we have seen that in a case where the virial can be calculated exactly

the original formulation of the boundary energy and virial fail. Atom pairs that, due

to their connectivity, remain within the cutoff should not be included in the boundary

correction. In simulations where the system dimensions change the cutoff radius should

be scaled accordingly. Additionally, a −Rc ∂U/∂Rc virial term is required to account

for pressure contributions from the image separations. This extra term is necessary

even when the system dimensions are constant. The complete virial can be obtained

conveniently, as a fraction of the potential energy avoiding the calculation of derivatives.

The new virial formulation in the IPS method considerably improves the agreement with

LRC results even for short cutoffs.


55

Chapter 4

Just-in-Time Empirical Valence Bond Method

The computational approach adopted in my work is MM where the motion of the atoms is

treated classically (i.e., using Newton’s laws). Instead of using wave functions and solving

Schrodinger’s equation MM employs fixed force fields where forces are computed from

a mixture of bonding and non-bonding potentials (e.g., bond stretching, angle bend-

ing, dihedral torsion, electrostatic and Lennard-Jones potentials). The success of this

approach relies on good parametrization of these potentials, as discussed in Chapter 2.

Even though the wave function formalism is entirely absent in the classical treatment,

electron density effects such as the polarization can be readily incorporated in the clas-

sical framework. However, force fields are rarely developed to reproduce ab initio data

alone. Physical constants of the substances can also be taken into account. A good

force field for water, for example, would be expected to closely predict the correct dipole

moment, dimer geometry, energy, density, diffusion coefficient etc. In the case of the

hydronium ion mobility, however, MM does not provide the means for describing proton

hopping via Grotthuss-type mechanism, sketched out in Fig. 4.1.

In pure water the mobility of the proton is determined almost entirely by the rate

of this proton hopping [58]. If the estimates are based solely on the vehicular mode of

proton transport even the qualitative predictions of the diffusion constant for different

polymers can be incorrect. In this chapter we examine some of the methods used to

incorporate proton hopping in classical simulations and describe a new approach that we

have developed, referred to as the JIT-EVB method.

The general mechanism of proton hopping was proposed by Grotthuss over 200 years

ago [57]. With the advent of computer simulation and ab initio MD simulation, in par-
56
Grotthuss

dissociating and reforming


hange with the making and
ogen bonds. This process,
ng hydrogen bonds, occurs
f the favorable energy scale
s qualitatively explained in
rom Leipzig, Theodor Chris-
Figure 4.1: Reproduction of the original idea of Grotthuss for proton shuttling between
(1785–1822) obviously had
two electrodes [57]. Hydrogen (h) and oxygen (o) are represented by ⊕ and signs,
he published in 1806: “It is
respectively, and water is represented by neighbouring ⊕ pairs, i.e., by oh. The water
he molecules of water, situ-
molecules are arranged along a linear chain such as to form a wire and the electrolytic
ductor wires, will alone be
decomposition of liquid water manifests itself by giving the rightmost oxygen away to
placed intermediately will
the cathode (marked by a + sign), thus breaking apart a water molecule. The leftover
ely their componenthydrogen,
princi- in turn, combines itself with the oxygen of its left water neighbour and so
forth.
” and “… all the molecules
Figure 6. Reproduction of the page containing Figures I and II of the pam-
would be decomposed and
phlet printed 1805 in Rome.[66c] Oxygen, o, and hydrogen, h, are represented
r to explain electrolysis of by !have
and a signs, respectively,
" clearer and water is represented bylead
neighboring
ticular, we now understanding of the events that to the translocation
d from the English transla- !" pairs, that is, by oh. Note that the distinction between atoms and mole-
in French,[66a] see of
Figure 5
the excesscules is notinclear
charge in the
water [60,text: “… The
18]. the molecule of waterpostulated
traditionally represented solvation
by o,h species,
…” whereas “… the molecules of oxygen situated in …” (cited according to
order to illustrate his idea, + [66b] +
namely the Eigen ion (H
the English 9 O4 ) and the
translation). TheZundel ion (H5 Oare
water molecules 2 )arranged
are shown along toabe transient and
linear
tches where he drew linear chain such as to form a wire (called fil in the French original) [66a]
and the
ode with the anodemobilepart of
due to the interactions
electrolytic with of
decomposition their own
liquid solvation
water manifestsshells. Statistical
itself according analysis of the
to Fig-
on for an explanation). This ure I by giving the rightmost oxygen away to the cathode (marked by a +
processes leading up to
sign) thus the proton
breaking apart transfer (PT) event
a water molecule. have demonstrated
The left-over hydrogen, in turn,a “special pair
combines itself with the oxygen of its left water neighbor, the hydrogen of
dance” in which a distorted Eigen cation constantly changes its closest oxygen atom as
which forms with the oxygen (marked r) of its left neighbor a new water
molecule and so forth. The reverse process happens at the anode (marked
shown in Fig. 4.2. This “dance” represents repeated trials to locate a suitable partner
by a # sign): hydrogen Q is taken up by the anode, thus breaking apart
water
for the PT act andmolecule
occurs, QP,
on sothethat oxygen every
average, P can form
40 f as.new
This water moleculethe
resembles to-description of
gether with hydrogen X from its right neighbor and so forth. Clearly, a con-
PT events intinuous wireab
the early of inito
water MDmolecules connects
reports, exceptin Figure I the cathode
that partner with the
exchange takes place on
anode, which allows for a continuous process of breaking and making of
the femtosecond
waterscale, whereas
molecules alongitthetakes
chainseveral
(comparepicoseconds for a version
with the modern PT tointake place. The
Figure 4). Figure II is an extension of this idea to two coupled Galvanic cells.
outcome of the “special pair dance” is equivalent to diffusional rotation of the H3 O+ ,

except that no rotational motion is involved.


perception
Earlier studies of thethus
PES led to translocation
of the the alternative designation
process “Grotthuss that the
have demonstrated
diffusion” for the concept of structural diffusion and to the in-
barrier resides in the initial
troduction dissociation
of the of a water
term “Grotthuss molecule from
wires/bridges” for the
suchsecond
one- solvation
dimensional
shell of the hydronium ion.hydrogen-bonded
Once this event occurs waterthewires.
translocation of the proton through
Although the paradigm of Grotthuss diffusion has already
been around for 200 years, the Grotthuss mechanism as such
was unclear until fairly recently. There was, however, the
notion of preferred solvation structures of hydrated protons in
the literature, in particular the complexes proposed by Eigen
and collaborators,[67, 68] H3O + ·ACHTUNGRE(H2O)3, on the one hand, and by
segments of a “resting” hydronium.
Participation of A1 cleavage in the faster partner exchange
events (rather than in the slower PT) is in line with previous
calculations showing that A1 HBs are much weaker than typical
57
water-water HBs,48 apparently due to the unfavorable interac-
tion of a water hydrogen with the positive center. For this reason,
the frequency of partner switches depends more weakly on
temperature than PT, so we expect the importance of the SP-
dance to diminish at higher temperatures (we have verified this
qualitatively for the qMS-EVB3 potential). Likewise, simula-
tions with faster PT (such as the MS-EVB2 and qMS-EVB3)
exhibit less of an SP-dance.
The outcome of the SP-dance is equivalent to diffusional
rotation of the H3O+, as suggested by Hückel,12 except that
Figure 12. Same as Figure 11 for a quantal MS-EVB3 trajectory. See no rotational motion is involved. This phenomenon rapidly
Figures S7-S9 in Supporting Information for the other water models.
randomizes the proton hop direction, so that proton mobility
in water is diffusive rather than coherent (i.e., it does not
involve correlated hopping over long HB water chains as
forming a SP,20,21 whereas another HB is longer than average.42 previously anticipated).14 Indeed, time-resolved fluorescence
This widens the first peak in the RDF but not sufficiently to measurements from molecules exhibiting excited-state PT to

Figure 13. Schematic depiction of the “special pair dance” occurring in the first solvation-shell of the hydronium in its “resting state” (during the
Figure
long trajectory 4.2: Schematic
segments of depiction
no-PT). Oxygens in red, except of thehydronium
for the “special oxygenpair dance” occurring in the first
in magenta.
solvation-shell of the hydronium during the long trajectory segments without PT events.
Oxygens are shown in red, except for the hydronium oxygen in magenta. Figure source:
Ref. [60].
��� ���� ���� � ������ ��������� ����������� �� ��� ������� ����� �������� ��� �������
�������� ���� ������� ������������ ������� ���� ����� ������� ��� �� ������ ��� ���
�������� ��� ������ ��� �� ������� ���� ���� �� ��� ���� ��� ����� ������ ���� �������
������ �������� �� ��� ��������� ������ ��� ���������� ����� ��������� ����������
58 ��������
���������������� ��
������� ������ ���
���� ������ ����� �
a b
8.0 8.0
��� �������� ����
��� �������� ����
5.6 5.6
�������� ������ ���
P 3.2 3.2 ���������� �����
0.8 0.8
��������� �� �����
�� �������� �� ��
–1.6 –1.6 ������� �� ��� ��
–4.0 –4.0 ����������� �� ���
2.2 2.2

2.5 2.5
R oo (Å)
2.8 0 1 2.8 0 1
–1 –1
δ (Å) δ (Å)

Figure 4.3:������ � ��������������


Probability distribution �� ��������
of the ������
proton����������
as a function��������of the������������
Oa − Ob distance
and the asymmetric ���� �� ���
�������� ����stretch ������ ������������
coordinate δ = ROa H� �−�R � O� 2 � � ��������
. �Both �� ��� ��and
electrons ��� nuclei are
bH
� �

���� �������� ����������� ������� ����������� ��� ��� �������������


treated by QM in a (using path integrals), while in b the nuclei are treated classically. The �������
������ ����������
colouring scheme ��� �������
in a represents ��� ��� ������������
the coordination number������ (see the����� �� ���
figure �� Ref. [62]
source
���� ������� ��� ��� ��������� ��� ����������� �� ���
for more details) decreasing from about 4 (yellow) in the Eigen ion regions �� ���������� �� ��� � ��
� to� about 3.5

(blue) in the Zundel ion region. The quantum fluctuations of the nuclei ��
������� � � �� ��� ������ �� ��� � � � � ���� ���������� �� ��������� ����� ������ ���
obliterate the

distinction�������
between ������the ����������
peaks seen � ��
in��� �� ��� �����to�������
b, pointing a smooth �������and ���� �����barrierless
���virtually
���� ����
inter-conversion ��� � ���
between the��Eigen
����� ���
and�������
Zundel ����� �� ��� �� ���� �������������� �� ��
ions.
��� ������� �������� ����� ������������ ������� ���� ����� �� �� �� ������
������� �� ���� ����� ��� �������� ���� ���� �� �������� �� ��������� ���
���
the Zundel������
ion is �� ������ ���������� �� �� ������ � ������ �� ������ � ��� ����� �
fast, with a barrier which is less than the available thermal energy at
���
��� ���� � � �� ���������� �� ��� ���� ������� �� ��� �������������
� ���� �
������ ������������
room temperature [62]. Quantum ����� �� ���� ������
�������� �fluctuations �������
further ����� �
stabilize ���������
the Zundel ����
ion leading to
������� ���� �� ��� �� ����� �� ���� �������� ��� ������������� ��� �������� ���
a continuous proton distribution
������������ along
��� ����������� the �reaction
����� path as ���
� �� ���������
� � �� seen in�Fig.
���� 4.3.
��� ���������

����� ��� ���������� �� ����� ��� ����� �� ��� ���� ������ ���������� �������� ��
As a consequence of a practically non-existent

barrier, proton shuttling is not a rare
�������������� ��� ��� ����� ��� < ��� �� ��� ���� ���� �� ��� ������������� ������
������ ��be � ��������� ���� ��� �� ��� ���� ������� ��� ���� ���� �� ������ ��
� �observed
��� �
event and should by routine MD simulations (i.e., such simulations that do
������ ���� �� ������ �� ��� ����� ����� �� ���� ������� � ��������� ���� ������
not artificially
�����enhance
�� ���� ��the probability
�������� of some
���� �� ��� events).
������� Whether
�� ���������� a PT ��
�������� will�������
be observed or� �������� ����
������
� ������ ��� ���� �������������� ���� ���
��������������� ������� ������ �� � � ��� � ���� ���� � ������ ��
not depends indeed on how accurately the interactions are calculated along the reaction
���� ����� ����� �� ���������� ���� ������� ��� ����� ��� �� ���������� ���� �� �������������� ��� ��
� ��
path. The � ��
issue �
poses��a� �������� �� ����� �for
serious problem ������
MM �� ������� ������as
simulations ������� ��� ������ potentials
the interaction ����������� ���� �����
����� ���������� ��� ��� ����� ��� . ��� �� � ������ ����� ���� ��� ��� ��� ��� ���� ����������� �������� ��
are fitted to equilibrium
���� �� ������ ��� ��structures (i.e.,��hydronium
���� ����� ����� ��� �� ������ ��ion and������
�������� water)
����rather
��� ���than������
along�� �������� �����
� � �

��������� �� ��� ������ ���� ����������� �� �� �� �� ��� ��� �������� ����� ���
������ ����� �� �� � �
reaction paths.� In this regard it is instructive to compare the ab initio proton distribution
��� ���� �������� �� ������� ��������� ������������ ��� ���������� ��� ���� ������ �� ����� �� ��
������
functions seen ��� ����������
above with that������������� �������� ������������
from a MM simulation ���� � � � �������������
presented in Fig. 4.4. Even though

������ � ��� ��� � �� �������� ���� � ��������������


© 1999 Macmillan Magazines Ltd
59

� �
ROO Å

� �
δ Å

Figure 4.4: Scatter plot of the proton distribution in a Zundel ion (represented classically
as hydronium ion plus water) at 300 K. The distribution is plotted as a function of the
Oa − Ob distance and the asymmetric stretch coordinate δ = ROa H − ROb H . Bond
stretching in the hydronium ion is modelled by a Morse potential.
60

the hydronium ion was modelled by a Morse potential which facilitates the Oa H bond

dissociation δ values above -0.4 Å or oxygen atom separations below 2.5 Å were never

observed. Thus, the areas of the plot that would correspond to a symmetrical Zundel ion

structure (i.e., δ = 0 and ROO = 2.38 Å) remain empty [63]. The region that the MM

potentials effectively sample corresponds to only the lower left corner of the ab initio

distribution (see the contour plots in Fig. 4.3).

Unfortunately, it is currently not feasible to carry out high quality ab initio calcula-

tions that reliably represent the hydrogen bond on nanoscale structures like the pore in

the hydrated SSC polymer. This is particularly true when long trajectories are required

as is the case when modelling a dynamic process like the diffusion of a proton or water

molecules. In the not too distant past even much smaller systems were beyond the scope

of ab initio calculations thus leading to the development of alternative, classical methods

to model the proton hopping. As will be explained later, the classical approach can offer

another unique advantage that goes beyond the increase in length and time scales.

In order for a proton translocation to occur in classical simulations it will be neces-

sary to manually alter the identity of the species by updating their connectivity matrix,

interaction potentials and parameters. Ideally this switch should take place when δ ≈ 0

to ensure energy and force continuity. However, as we have seen above, the common MM

potentials strongly disfavour δ values near zero. A reactive MD simulation approach cur-

rently under development [64] employs a set of geometrical triggers to identify structures

that are likely to lead to a PT if the molecules move on the correct ab initio energy

surface. When such a structure occurs during a MD run the proton is transferred to the

accepting oxygen atom and the interaction potentials are switched accordingly. Since

this does not occur at δ ≈ 0 the discontinuity in energy and forces requires a short equi-

libration so that the transfer process remains thermoneutral1 . The success of such an
1
Both the reactant and product of the PT process are Zundel ions, hence the process is thermoneutral.
61

H H H H H H H H

O H OO H O O H OO H O

H H H H H H H H
a b

Figure 4.5: A Zundel ion (represented classically as hydronium ion plus water) can
exist in two resonance forms, a and b, obtained by interconversion between a covalent
OH bond (solid line) and a hydrogen bond (dashed line). Note the two structures have
identical geometries.

approach depends on the careful selection of the trigger values so that the reaction rates

can be reproduced.

The interconversion between the hydronium ion and water need not be as abrupt, as

in the case of the reactive MD method. Instead of occurring at some particular value of

δ we can imagine a gradual transition between the identities of the two species. At any

instance in time the interactions in the Zundel ion can be represented by a superposition

of two resonance forms, displayed in Fig. 4.5. The mixing between the a and b characters

of the classical Zundel ion can be calibrated such that its overall properties are similar

to that of an ab initio Zundel ion. The original EVB formulation developed thirty years

ago [65] determines the mixing coefficients from the ground state solution of the secular

equation:

Ha − E Hab

=0 (4.1)

Hab Hb − E
The potential energy of resonance forms a and b, Ha and Hb respectively, are calculated

classically using valence bond theory (i.e., as sum of stretching and bending potentials

plus non-bonding Lennard-Jones and Coulombic potentials). When the ground state

energy E is known one can solve for the empirical off-diagonal term Hab and vice versa.

Thus, the ab initio PES of the Zundel ion needs to be obtained first in order to determine
62

the appropriate value for Hab . Once this is done the properties (e.g., the location of

the positive charge) of any Zundel ion occurring in our simulation can be obtained by

summing the coefficients of the resonance forms.

Quite early in the development it was realized that a single, global value for the

empirical parameter Hab does not lead to a good agreement between the ab initio and

EVB PES. It is possible to overcome this problem if the Hab term is determined on the

basis of the current reaction coordinates ROO and δ. As is the case with any classical

approach, it is often desirable to reproduce experimentally determined properties, such

as the proton diffusion coefficient. This can be achieved by adjusting the parameters of

the intermolecular interaction potentials (such as the charge and the Lennard-Jones pa-

rameters ε and σ) which affect the diagonal terms Ha and Hb in the secular determinant.

The EVB method has currently evolved to include features such as more resonance forms

that include the solvating water molecules [66], solvent polarization [67], self consistency

in multiproton systems [68], nuclear quantum effects [69, 70] and so on. Another closely

related method that has gained considerable popularity is the Multi-Configuration MM

method [71].

Despite these advancements modelling the structural diffusion of proton in the SSC

polymer with the EVB method remains challenging. There is nothing, inherently limiting

the applicability of EVB to PT between a hydronium ion and water only. The side

chains of the polymer carry the strong sulfonic acid group −SO3 H which are rendered

deprotonated in the presence of water. Typically, the ab initio PES can be determined

from a 10x10 grid in the ROO , δ plane. However, in the highly concentrated, proton rich

medium present in the hydrated polymer there are not only different pairs of donors and

acceptors but also coupled PTs as shown in Fig. 4.6. The increase in the number of

resonance forms means a higher order EVB matrix and more off-diagonal terms to fit.

Determination of the ab initio PES must now be carried out in a grid of 10x10x10x10
63

O O O O
O S O S
H H

S O O S O O

O O
H H H H
O O

H H

O O
H H H H

a b

O O O O
O S O S
H H

S O O S O O

O O
H H H H
O O

H H

O O
H H H H

c d

Figure 4.6: PT between two sulfonate groups influenced by a neighbouring Zundel ion
(see Ref. [72] for more details) in a triflic acid monohydrate solid. Here we see the four
possible resonance forms of the process obtained by interchanging covalent bonds (solid
lines) with hydrogen bonds (dashed lines).
64

points. Coupled three-PTs would be impossible to parametrize since a grid with 1×106

points is now required. Hence, computing the EVB matrices of a SSC system would

require more calculations than a direct ab initio MD run. In the next few paragraphs

we shall introduce the JIT-EVB method demonstrating how a significant reduction in

the number of grid points can be achieved. Firstly, we investigate the possibility of

developing a way of simplifying the manner in which EVB mixes the resonance forms.

The solution of the secular equation in the EVB method yields both the ground state

energy and the corresponding mixing coefficients for the resonance forms. Alternatively,

we could directly fit these coefficients such that the desired ab initio energy (or forces)

are reproduced. For instance, in the case of two resonance forms a and b the classical

energy could be calculated as ca Ea + cb Eb , this time ca and cb being the unknown empir-

ical parameters. Just like the off-diagonal terms in EVB are dependent on the reaction

coordinates ROO , δ so are the mixing coefficients ca and cb . The convenience such sim-

plification brings becomes apparent for higher order EVB matrices where the number

of unknown off-diagonal terms starts to exceed the number of the resonance forms. We

should not, however, be concerned about such an approach causing a deterioration of the

final result. Since, even though the EVB mixing might be more sophisticated it is not

more accurate.

The success of the EVB method in reproducing the ab initio PES is entirely due

to the reasonable interaction potentials and the extensive parametrization, both in the

diagonal and off-diagonal terms. Since the diffusion coefficient is a dynamic property

obtainable by MD, our primary concern is reproducing the ab initio forces, rather than

the energy. For this reason it is preferable to fit the parameters ca and cb such that

the total force ca Fa + cb Fb matches the ab initio force2 . It is important to address a


2
In our method each resonance form coefficient is calculated as a product of two polynomial functions
P
4
j P
4
j
of the reaction coordinates ROO and δ, e.g., c0i = sj ROO tj (1 − δ/ROO ) . The primed coefficients
j=0 j=0
65

commonly held misconception about the forces in the EVB method regarding the assumed

“orthonormality” of the resonance states. Such an assumption allows the forces to be

calculated through the Hellmann-Feynman theorem. However, as explained above, the

false consequences of even the most serious assumption (i.e., the correspondence between

the classical potentials in the resonance forms and the QM states) can be mitigated by

extensive parametrization. The problem is that the parametrization that reproduces the

ab initio energies along the grid is not the same one that will reproduce the forces.

Borrowing the idea of geometrical triggers from the reactive MD method, in JIT-

EVB we use these triggers to switch from the unique, classical potentials of the reacting

species to a linear combination of such potentials in the resonance forms. In other

words, we do not automatically transfer the proton but merely allow for this to occur

gradually. Since the conformations outside the trigger zone are considered unreactive, no

grid points are required there, see Fig. 4.7. However, the fact that a certain conformation

(or a grid point) belongs to the reactive zone does not automatically mean that such a

conformation will ever be encountered during simulation. We can reduce the number

of grid points even further by excluding such areas of the reactive zone that have low

probability of occurring. To see how this can be achieved let us examine Fig. 4.8. With

our EVB method the ab initio PES and the fitting are performed on the fly, just as soon

as a molecule requires it. Hence the name of the method “Just-in-Time” EVB. Once

a grid point has been evaluated by an ab initio calculation and the fitting obtained the

information is saved to a database. Subsequently, whenever a system finds itself in the

vicinity of the same grid point the information is looked up from the database without

the need to perform another ab initio calculation. With the JIT-EVB method a system

where coupled PT events can occur can be studied immediately, without the prerequisite
are then squared and normalized such that ci are always positive and correspond to the probability of
finding resonance form i. Hence, the fitting parameters are actually the polynomial coefficients s and t.
For more details, please refer to Appendix D.
66

Figure 4.7: Molecules whose conformations are inside the reactive trigger zone (green
boundary) will be subject to JIT-EVB mixing of the resonance forms to reproduce the
ab initio forces along the grid. Outside of the reactive zone a single resonance form is
sufficient, hence, JIT-EVB is not invoked.

of tens of thousands of grid point evaluations.

Finally for this chapter we will address an issue that is of paramount importance in

classical proton hopping simulations. Pure ab inito methods can handle PBC through

periodic wave functions. Here it does not matter if the proton donor and the proton

acceptor belong to the same periodic cell as the MIC always applies. In a classical

simulation, however, this is not the case. Let us consider Fig. 4.5. If the acceptor oxygen

is in a different periodic cell then resonance form b (which depicts a hydronium ion on

the left) will be misleading – the new covalent OH bond will actually span across periodic

cells. Thus, resonance form b will have a very high potential energy. In this situation the

standard EVB method which uses the ground state solution of Eq. (4.1) will invariably

lean toward resonance form a. It is not surprising then that all EVB results reported to

date underpredict the proton diffusion coefficient. This flaw of the classical simulation

is not limited to EVB. Even in a simple QM/MM approach once a molecule leaves

the (periodic) QM zone there will be an immediate jump in the potential energy and
67

B
A

Figure 4.8: At point A a molecule enters the trigger zone (top). An ab initio calculation
has to be performed to determine the correct mixing of the resonance forms. As with
the regular EVB method, the fitting obtained at this point is considered valid within
some neighbourhood determined by the grid spacing. The system is propagated until the
molecule either exits the trigger zone or drifts outside the validity radius. When the latter
happens (say at point B) a new ab initio calculation and parametrization are carried out
(bottom) and so on. The molecule explores the PES using the fitting parameters of its
closest grid point - A, B etc. Beyond the trigger zone boundary the molecule returns to
a unique classical representation in the resonance form with the highest contribution.

forces, unless all its hydrogen atoms are in the same periodic cell as their oxygen anchor.

For this reason QM/MM studies of proton transport in bulk water have been unable to

incorporate PT events [73].

The problem is not unique to the OH stretching potentials. Both bending and dihe-

dral potentials that involve mobile hydrogen atoms can also yield spurious results. The

solution is quite simple by introducing what we call MIC bonding potentials which we

describe in Chapter 6. Application of the JIT-EVB method to water and triflic acid is

discussed in Chapter 8.
Part II

Polymer System Studies

68
69

Chapter 5

Constructing The Polymer Systems

It is evident that an experimental chemist wishing to study the properties of a new com-

pound first needs to obtain a sample of it. If computational chemistry does, indeed,

provide the tools and methods to explore the properties of compounds then a given

compound of interest must be a well defined chemical entity that can be modelled. Con-

sequently, one would expect to know the formula of the compound, its structure and

conformation in order to uniquely predict its properties. For the hydrated polymer sys-

tems of my study the exact formula, structure and conformation are unknown. The goal

of this chapter is to present the methodology we have developed to create a “sample” of

the SSC polymer that will be later subjected to computational testing. In the next few

paragraphs we proceed toward this goal starting from determining the polymer formula,

then building its chemical structure and finally generating plausible 3D conformations.

The SSC polymer is a copolymer of tetrafluoroethylene and a sulfonyl fluoride vinyl

ether with the general formula CF3 [(CF2 CF2 )n − CF2 CF (OCF2 CF2 SO3 H)]N CF3 . The

synthesis of the polymer does not allow perfect control of the monomer ratio in the

polymer chain, resulting in a lack of a well defined value for n. However, a typical MW

of the polymer unit (i.e., its EW) is 830 ± 20g/mol [74], corresponding to the presence of

between five and six tetrafluoroethylene groups in the backbone for every sulfonic acid

group. Hence, a reliable model of the SSC polymer needs to incorporate some degree of

randomness in generating the molecular formula and structure.

With the advent of computer databases new chemical notations have emerged that

make it possible to uniquely represent and digitize chemical structures. In the mid 60s

H. L. Morgan introduced the Chemical Abstract’s registration system that assigns an al-
70

Table 5.1: Examples of random polymer sequences generated with different seeds.

seed na N Polymerb

1 5 2 TTSTSSTTTTTT
2 5 2 TSTTTTTTTTTT
3 6 3 TTTTTTSTSTSTTTTTSTTTT
a
The general formula of the polymer is CF3 [(CF2 CF2 )n − CF2 CF (OCF2 CF2 SO3 H)]N CF3 .
b
The shorthands T and S are used in place of tetrafluoroethylene and sulfonyl fluoride vinyl ether,
respectively

phanumeric sequence to chemical structures [75]. Another notations system that emerged

nearly twenty years later is the SMILES system [76]. It has gained considerable popular-

ity due to its simple notation rules. For example, tetrafluoroethylene is represented by

the string sequence FC(F)=C(F)F, with brackets used to indicate side chains attached on

the left. If we know the name of the compound or its CAS registry number the SMILES

representation can be conveniently obtained using a simple on-line converter [77]. An-

other useful on-line tool incorporates a graphical structure editor that directly generates

the SMILES sequence [78]. With the help of the SMILES system one can reduce the

problem of randomness in the polymer formula to randomness of a string. The algorithm

we use to determine the SMILES sequence of a random polymer is presented in the flow

chart in Fig. 5.1. Even though randomness is a favourable feature in the monomer se-

quence the final result should be reproducible. This is accomplished by specifying a seed

for the random number generator. When the program is executed with the same seed it

will produce the same random sequence of monomers regardless of the hardware or the

operating system it runs on. Some examples are shown in Table 5.1.

The algorithm we have presented here for the generation of random comb copolymers

can be extended to higher levels of branching by manipulating the SMILES representation

directly. For example, introducing a branch in the polymer chain can be carried out

by a simple replacement of one atom by the SMILES formula of the branch. Such
71

input the
random
seed

input the
chemical
formula as
n, N

SMILES = 'C(F)(F)F-'

iterate over the N


polymer units

iterate over the


'-C(F)(F)F' appended
n+1 monomer
to SMILES
units

generate a random
number between 0
and 1

Yes No
number > 1/(n+1)?

'-C(F)(F)-C(F)(F)' '-C(F)(F)-C(F)(O-C(F)(F)-C(F)(F)-S(=O)(=O)O)'
appended to SMILES appended to SMILES

Figure 5.1: Flowchart representing the creation of a SSC polymer with random monomer
sequence and an average formula of CF3 [(CF2 CF2 )n CF2 CF (OCF2 CF2 SO3 H)]N CF3 .
The flowchart is traversed starting from the top (always traversing the right sub-flowchart
before proceeding downward). The style used is a variation of the one described in
Ref. [79].
72

manipulations are easily scriptable in any high-level language like Python, or can be

carried out in a text editor.

Having determined the polymer formula and represented its structure in the SMILES

notation it is now necessary to translate this information into a set of bonds, angles and

dihedrals which collectively constitute what is known as the connectivity matrix. This

information is used by the interaction evaluators in our code to calculate the energy,

forces and virial of the system. Generating the connectivity matrix is done from the

SMILES formula with the help of the OpenBabel [80] chemical expert system.

The hardest part, however, lies in the last step of the building process i.e., obtaining

the 3D conformation of the polymer. The nature of the hydrated polymer systems poses

an immense challenge for the prediction of their conformation, perhaps even eclipsing

those encountered with proteins. Determining the conformation from X-ray measure-

ments is not possible as only a small part of the polymer is crystalline. In fact, unlike

proteins in their native state, the hydrated polymers are characterized as non-equilibrium,

“living systems” [81]. It is useful to estimate the size of the conformational phase space

for our polymer systems. The MW of Nafion has been estimated to be in the range

of 105 − 106 g/mol [10]. Assuming a similar range for the SSC polymer and a typical

EW of 850 g/mol the smallest number of polymer units N is found to be 118. In each

polymer unit we have n tetrafluoroethylene monomers and one side chain monomer with

a total of 2(n + 1) backbone dihedral angles. A simple estimate of the number of con-

formations of the entire polymer can be done if we assume that each dihedral angle has

a limited number of preferred conformations (e.g. trans, gauche plus and minus). Thus,

we get the astronomical number of about 10788 conformers1 for even the shortest SSC

structure. This enormous conformational phase space cannot be traversed by any com-
1
If each of the 2(n + 1)N dihedral angles in the polymer backbone has c possible conformations then
the total number of conformations for the polymer would be c2(n+1)N . Taking c = 3, n = 6 and N =
118 we get approximately 10788 .
73

Figure 5.2: Perfluorocyclohexane converging to two distinct local minima depending


on the conformation seed. For seed=1 a twisted chair structure is obtained (left). For
seed=5 an ideal chair structure emerges whose energy is about 12.7 kJ/mol lower (right).

putational means in the time that the universe has existed! More importantly, even the

real polymer would never be able to sample the entire phase space. Consequently, the

polymer will explore only a small part of the entire phase space volume, limited to some

neighbourhood of the initial configuration. This implies that for any sufficiently long

polymer the 3D conformation will depend on the initial conditions. Here, again, it is

desirable to have reproducibility of the conformations obtained by the polymer builder,

which can be achieved by introducing another random seed. In fact, even short systems

like perfluorinated cyclohexane already settle into different local minima, displayed in

Fig. 5.2. In summary, the random monomer sequence issue is a separate one from the

issue of the random 3D conformation of the polymer. For a given monomer sequence we

can generate a large number of conformations by specifying unique initial conditions for

the 3D conformation search. Thus, an appropriate estimation of the properties of the

polymer system should necessarily include a statistical average over both the monomer

sequences (in random polymers) and the initial conditions.

As demonstrated above, the enormity of the size of the phase space severely limits the
74

usefulness of statistical averaging over randomly picked conformations. For this reason

the 3D conformation search algorithm we have developed aims at building and folding

the growing polymer chain as it may occur during the polymer synthesis. It is based on

the following rules:

• A newly added group of atoms to the growing end of the polymer (e.g., CF2 , SO3

etc.) will have time to find its energy minimum before the next addition takes

place.

• The part of the polymer already built is much larger than the newly added group,

so the interactions with the latter will not lead to conformational changes in the

rest of the polymer.

• In the case of steric clashes the entire polymer will relax its structure but only to

the point of avoiding the particular clash allowing a new addition to take place.

The algorithm we use to determine the 3D conformation of the polymer according to the

above rules is presented in Fig. 5.3. The general idea is to split the optimization of the

whole polymer into a sequence of short optimizations performed only on few of atoms at

a time. These optimizations are fast, not because the rest of polymer is kept frozen, but

due to the reduced number of interactions being considered, see Fig. 5.4. Our MD code

was specifically developed to accommodate the interaction set switching that takes place

in the polymer building process.

A distinct advantage of our polymer builder is that the desired polymer can be built

directly into the periodic simulation box. Even though a polymer chain can span many

simulation cells it always experiences the interactions of the periodic images and folds

accordingly at every addition step. This is in stark contrast to the technique that is

currently employed where a polymer chain is built in an infinite simulation box first and

then put into a periodic box. The latter technique complicates the folding of the polymer
75

input the
conformation
random seed

bundle all dihedrals with the


same central bond

find the new atoms in the


iterate over all
bundle (with yet unknown
dihedral bundles
coordinates)

run a final SD/ABNR set the interaction set the interaction


optimization of the whole calculators to consider only calculators to consider only
polymer (new atoms, new atoms + (built atoms, built atoms)
built atoms) interaction pairs interaction pairs

run a short SD optimization run a short SD optimization


to find the new atom to find the built atom
coordinates, built atoms are coordinates
frozen

add the new atoms to the list


of built atoms

No
energy below
threshold?

Figure 5.3: Flowchart representing the stepwise process of finding a low energy con-
formation for the polymer. The bundling of atoms by dihedral angles ensures that at
each addition step the correct local conformation is obtained. The optimization of the
coordinates of the new atoms starts from randomly chosen values controlled by the con-
formation seed. At the end of the stepwise process the polymer structure is further
optimized until its gradient has been sufficiently reduced allowing an MD simulation to
commence.
76

C A

F
O

B
F

Figure 5.4: The interactions needed to be considered in order to optimize the coordinates
of the new atoms (set A) are those within the set and with the built atoms (set B). Only
in the case of steric clashes it is necessary to relax the entire built polymer, in which case
all interactions are considered (i.e., including those within set B).

enormously due to the sheer size of the conformational phase space. One solution that

alleviates the problem is to use multiple copies of a single, much shorter polymer chain,

which can be equilibrated more rapidly. However, the resulting increase in polymer chain

mobility in this segmentation approach may affect the transport properties of both water

and hydronium ions.

Finally, the polymer system needs to be hydrated in order for any proton transport

to take place. We know, a priori, the amount of water needed from the λ parameter

(i.e., the number of water molecules per sulfonic acid group). The way we achieve the

hydration is by superimposing the built polymer onto a simulation box filled with water

molecules. The size of the water box coincides with the size of the box in which the

polymer was built and is determined by the density of the desired system. The potential

energy of each water molecule, in the box, is calculated by taking into account both the

water-water and water-polymer interactions. The water molecules that overlap with the
77

polymer will possess very high potential energy due to the repulsive term in the Lennard-

Jones potential. To reduce the number of the water molecules to the desired hydration

level the excess water molecules are removed, starting with those with the highest energy.

Once the correct composition of the hydrated polymer has been achieved we carry out

two cycles of optimization. In the first one the coordinates of the water molecules are

optimized, while the polymer structure remains frozen. This is done in order to avoid

rupturing the polymer. In the second optimization cycle the entire system is allowed to

move.

So far we have seen the approach adopted for the building of 3D conformations that

correspond to a specific local minimum determined by the seed of the conformation search

algorithm. In the next chapter we show how we can simulate polymer systems with a

specific morphology.
78

Chapter 6

Designer Structures

The methods described in the previous chapter were aimed at building polymer systems

whose morphology was uniquely determined by the monomer sequences and the initial

conditions. The most interesting morphologies are seen in a class of polymers known

as block copolymers where each chain contains long, covalently bonded homopolymer

units. Such polymers are able to spontaneously form ordered phases with nanometer

scale structures [82]. The large variety of shapes accessible to the copolymer opens

the door to designing membranes with unique morphologies and potentially desirable

properties. Theoretical studies have already addressed the effect of blockiness of Nafion

on the observed morphology and properties [25]. Blocky structures with high degree of

clustering of the same monomers were found to exhibit larger size hydrophilic domains

and better proton conductivity.

As useful as it is, the bottom up approach that relates monomer sequences to mor-

phology and properties has a flip side. For instance, it would be a daunting task to

isolate the effect of a single feature, like the depth of the side chain protrusion into the

pore on the proton diffusion rate. Numerous polymer systems will have to be tested,

with their side chain protrusion depths averaged over the production runs in the hope

that statistically significant differences can be observed. In order to avoid a complicated

cross-correlation analysis, the selected systems must only differ in the side chain depths

and maintain equivalence amongst all other features, i.e. they should posses the same

pore radius, side chain density and so on. In this chapter we will introduce a different

approach that allows us to work directly at the monomer morphology level.

Previously, a kinetic model was developed based on a non-equilibrium statistical me-


79

chanical approach [19, 20, 83, 21] for direct determination of the effects of geometrical

parameters of the membrane pores on the diffusion of hydronium ions. This top down

approach relating membrane morphology to measurable properties may be applied to

any hydrated polymer electrolyte system. The only system-specific parameters used in

the model are the radius of the pore, the protrusion of the anionic side chains into the

pore and their distribution within the pore. These studies revealed that a larger pore

radius, shorter side chains and more uniform side chain distribution will lead to higher

proton diffusion coefficients [20].

An alternative approach was presented by Spohr et al. [81,84,85] in which the polymer

system is divided into structureless pore walls, represented by a simple volume exclusion

potential and the explicit side-chain fragments, treated with an all atom potential. Ex-

cluding the motion of a large part of the polymer has allowed simulation times of tens of

nanoseconds revealing slow conformational changes in the side-chain dihedral angles.

By the very nature of these models the picture for the proton transport will be

severely altered by changes of the interaction potentials. Therefore, our goal here will be

to preserve all the interactions in the real system, without simplifying any of the bonding

or non-bonding potentials. Thus, the effect of the polymer on the proton transport

inside the pores will be brought about by the actual backbone and side chains. The

only modification we will introduce will be an external potential whose function is to

restrain the polymer backbone to the desired shape of the pore. A comparison with the

traditional restraining scheme is depicted in Fig. 6.1. As we will find out later in the

chapter, once the polymer system has been built the shape restraining potential can be

turned off, in some special cases.

We begin by looking at a simple example of a polymer with a straight backbone. As

this is the preferred conformation arising from the low energy anti conformation of the

backbone dihedral angles, such a restraint may at first appear to be redundant. However,
80

water shape water polymer shape


phase restraint phase wall restraint

Figure 6.1: Cross section view of an ideal pore with the shape restraint represented as
an outer layer. Traditionally the pore shape is enforced by restraints acting directly on
the water phase, where the proton transport occurs (left). In our scheme the restraining
potential acts only on the polymer backbone leaving the side chains and the water phase
shielded from the external field (right).

any sufficiently long strand of the polymer will collapse onto itself as a consequence of

the action of the Lennard-Jones and Coulombic forces. To prevent such a collapse and,

furthermore, to orient the backbone along a specific axis we introduce a restraining

potential proportional to r2 , where r is the distance from the axis in the normal plane.

Because of the zig-zag conformation of the tetrahedral carbon atoms the restraint is

applied only to every alternate carbon atom, see Fig. 6.2. Despite the presence of the

straight backbone restraint which prevents each strand from twisting and intertwining

with itself, we can, by placing other strands in a parallel fashion ensure that the favourable

inter-strand interactions still persist. Therefore, we build bundles of straight polymer

chains, analogous to the ones constituting the membrane pores [17].

In principle, it is possible to apply the restraining potential of Fig. 6.2 individually

to each strand in the bundle. However, this particular form of restraint gives rise to

some computational artefacts. Due to the distance r being defined with respect to the x

coordinate axis, the restraining potential acts as an external field. In the presence of such

fields (be it restraining, electric, or gravitational) the system is no longer invariant with
81

Figure 6.2: Perfluoropentane built with a straight backbone along an axis. A harmonic
restraining potential of the form kr2 (where r is the distance from the x-axis in the
yz-plane) is applied to the bottom three carbon atoms (shown in green).

respect to translations and rotations. Therefore, neither the linear momentum nor the

angular momentum will be conserved [86]. These complications become apparent when

the system requires temperature or pressure control1 . Therefore, it is highly desirable

to use restraints that conserve the total linear momentum of the system. One way of

achieving this objective, for the straight backbone polymer, is to restrain the end-to-end

distance in the polymer chain. Such a restraint corresponds to a new bonding potential

being defined between the end atoms. However, as with any bonding potential, the

orientation of the participating atoms is undefined. Hence the end-to-end restraint will

produce a straight polymer with an arbitrary orientation in space, see Fig. 6.3. The

problem is that in order to create the bundles all polymer strands need to have one

consistent orientation, regardless of what that orientation may be. With the end-to-end

restraint we cannot easily enforce the consistency in orientation between the different
1
Only the relative momenta of the individual atoms with respect to the center of mass (CM) should
be thermostated [55]. Similarly, the ideal gas component of pressure which is proportional to the kinetic
energy of the atoms should exclude the CM contribution [87]. Such methods are well known but not
widely implemented in computer code due to the non-symplectic coordinate transformations that need
to take place. Furthermore, calculating transport properties also becomes more complicated in such
system as explained in Ref. [88].
82

Figure 6.3: Single perfluoropentane chain built in a periodic cell with a harmonic bond
restraint between the end carbon atoms (shown in green). Also shown is one periodic
image in the +x direction to emphasize the lack of continuity across periodic boundaries
(see the text). The conformation deviates slightly from the ideally straight one due to
the non-bonding interactions with the periodic images.
83

Figure 6.4: 2D channels formed by two polymer chains (black and green) in a periodic
cell. Routine simulations will exhibit features whose length is bound by the dimensions
of the cell (left). However, when the polymer chains are continuous across the periodic
boundaries the pores they form will be infinite (right). In the latter case, the absence of
end effects may permit smaller dimensions for the periodic cell.

strands without resorting back to using external potentials.

The size of the simulated systems should be as large as the methods and the time

constraints allow. As we are interested in the proton transport occurring in the membrane

pores we can achieve long, realistic size pores in two cases. The first and obvious one is

when the simulation cell itself is large. However, the more appealing one is when there is

continuity at the boundary, such that an infinite pore can be obtained. These two case

are shown in Fig. 6.4. Unfortunately, the end-to-end restraint does not automatically

lead to boundary continuity. To conclude, the end-to-end restraint is a step in the right

direction as it removes the external field and conserves the linear momentum. But we

still need to go even further in order to control the orientation of the polymer strands.

When calculating the non-bonding interactions in a periodic system we know that

we first have to apply MIC, then calculate the distances r, and finally, calculate the

corresponding 1/rn potential. On the other hand, for bonding potentials multiples of the
84

periodic cell dimensions (say Lx , Ly and Lz ) are not subtracted before obtaining r. The

only reason it is done this way is because of the assumption that the bond lengths are

always much smaller than the cell’s dimensions, hence MIC is automatically satisfied.

Nevertheless, let us see what happens when we apply the convention to the familiar

example of Fig. 5.2 with the molecule (perfluorocyclohexane) placed in a periodic cell.

The result is shown in Fig. 6.5. Amazingly, the two types of structures depicted have

the exact same connectivity matrix. Applying MIC for the bonding potentials allows for

bonds, angles and dihedrals2 to be calculated with the image atoms, across the periodic

boundary. We have already seen the need for such potentials when proton hopping is

involved. Here it is possible to obtain an infinite, straight polymer chain (bottom) that is

significantly more stable than the cyclic structure (top). The infinite polymers that can

be generated in this manner will have a straight conformation only when the dimension

of the periodic cell is commensurate with the straight chain length. Looking more closely

at the bottom conformation we note that:

• Since the straight polymer conformation is enforced by the dimensions of the pe-

riodic cell simulations where the dimension can change (e.g., NPT ensemble) can

disrupt any structure modelled with the MIC bonding potentials.

• Once the initial bias to connect with the image atoms has been removed, the

simulation can proceed as a normal, unbiased and unrestrained one that conserves

the total linear momentum.

• The orientation of the straight polymer is uniquely specified by setting one dimen-

sion (e.g., Lx ) to be commensurate with the straight chain length. Thus, consistent

orientation will be found in bundles of the same chain.


2
Since all angles are calculated from bond distances using the cosine theorem, the MIC bending and
dihedral potentials are the same potentials but with all distances involved calculated after applying MIC.
85

Figure 6.5: Perfluorocyclohexane built in a periodic cell with MIC applied to both
bonding and non-bonding potentials. Closest periodic images are also shown. The same
random seed and optimization procedure was used in the top and bottom cases, i.e.
brute force SD optimization followed by an ABNR minimization. In the bottom case we
have picked two linked carbon atoms (shown in green), set their coordinates to equal the
end-to-end distance in a straight C6 chain and kept these two atoms frozen during the
SD stage. All atoms were allowed to move during the ABNR stage. Not only does the
straight conformation persist, it is also about 114 kJ/mol lower in energy than the cyclic
conformation shown at the top.
86

F
C
C
F

C a F F b C

F
F F
F

a c

Figure 6.6: Two perfluorohexane chains built in an infinite simulation universe without
any restraints. Their closest segments are highlighted in green. The two segments are
slightly offset vertically, with the carbon atom a being near the midpoint of the b − c
bond. The distances are 4.91 Å between the ab atoms and 5.13 Å between the ac atoms.
Shown on top are the approximate Newman projections revealing the presence of the
gauche conformation in both segments.

• Pores formed by the infinite polymers (whether straight or not) will exhibit bound-

ary continuity (as in Fig. 6.4, right), allowing unimpeded proton transport across

periodic boundaries.

Simulating structures formed by polymer bundles requires knowledge of the effective

width of the polymer chains. In this paragraph we will try to determine this width.

Additionally this will allow us to establish the link between the polymer formula, the pore

size and the hydration level. Let us examine the conformation of two perfluorohexane

chains as they approach one another, as depicted in the bottom of Fig. 6.6. As we
87

Figure 6.7: Schematic representation of the approach of two polymer strands in the anti
conformation. When these are straight perfluorocyclohexane chains, in a periodic cell
(built with MIC potentials as in Fig. 6.5), the separation between the chains is about
6.24 Å.

can see from the Newman projections the preferred conformation is, sadly, not the anti

conformation. The two chains are interlocked with the less bulky fluorine side, while

the carbon atoms are on the opposite side. The gauche conformation that these strands

adopt enable the close approach of one fluorine atom from one strand to the opposing

carbon-carbon bond (e.g., the fluorine atom at a to the b − c bond). On the other

hand, all carbon-carbon bonds in a purely anti conformation strand will recede behind

the fluorines, as illustrated in Fig. 6.7. The crowding that is present here makes the

interlocking more difficult. Indeed, two perfectly straight chains can be expected to be

significantly farther from each other.

The motivation for this chapter was to develop the methodology that will link specific

geometrical factors (e.g., pore radius) to phenomena that occur inside the pore, like

proton diffusion. Unfortunately, we do not know how to maintain a perfect cylindrical

pore (with an unambiguous pore radius) using realistic, twisted polymer chains. Since the

pore walls only provide the boundary conditions for water and hydronium ion movement

in the pore, we could neglect the details of the intra-wall interactions. Thus, we will

assume that whether the pore walls are composed of twisted or straight polymer chains,
88

b
a

Figure 6.8: Schematic representation of a pore wall as formed by straight polymer chains
(shown as columns). In the vertical direction the MIC potentials can be used to enable
pore continuity across the periodic boundaries. To control the lateral movements of the
chains an additional end-to-end restraints can be applied, i.e. between carbon atoms a
and b.

their effect on the water phase will be the same. However, this requires us to hold the

straight chains closer together so to achieve the same inter-chain separation as it would

be found with twisted chains. This can be easily accomplished by employing end-to-end

type restraints (similar to the ones in Fig. 6.3) across the straight chains that form the

pore walls, Fig. 6.8. The pore walls we construct in this manner are not impermeable to

small molecules like water and hydronium ion. Placing additional bare polymer chains

(i.e., without sulfonic acid groups) in the four corners of the simulation cell will provide a

realistic model for the pore surroundings where any small, polar species will be naturally

excluded.

Finally, in this chapter, we establish the connection between the repeat unit of the

polymer, the pore radius and the hydration level. The volume of the pore can be found

from the volume of the water phase and the volume of the polymer forming the pore
89

walls:

Vpore = V water + Vpolymer (6.1)


phase

Let us label the pore radius R and choose the pore length L such that it contains

only one repeat unit (i.e., one sulfonic acid group) with the formula (CF2 CF2 )n −

CF2 CF (OCF2 CF2 SO3 H). The value of L can be calculated as L = 2(n + 1)lc where lc

is the chain length per backbone carbon atom. Thus, for a cylindrical pore we obtain:

Vpore = πR2 2(n + 1)lc (6.2)

If the pore walls are comprised of m polymer chains distributed along the circumference

of the pore and pointing inward then the total number of sulfonic acid groups in the pore

will also be m. With the hydration level λ defined as the number of water molecules per

sulfonic acid group we arrive at mλ as the total number of water molecules. Accordingly,

using VH2 O as the volume of a single water molecule we can write:

V water = mλVH2 O (6.3)


phase

The volume of the polymer can be found from the number of chains m, the volume of

a single polymer chain Vchain and a factor f (R, w) which represents the cross section of

the chain that is inside the pore:

Vpolymer = mVchain f (R, w) + mV side = mπ (w/2)2 Lf (R, w) + mV side


chain chain

= mπ (w/2)2 2(n + 1)lc f (R, w) + mV side (6.4)


chain

In estimating Vchain in the above equation we have assumed a cylindrical shape of radius

w/2 and length L. In our model w was the effective width of a bare (i.e., without side

chains) polytetrafluoroethylene molecule. Accordingly, the volume of the side chain has

to be explicitly included in Eq. (6.4). As the side chains are expected to be entirely

inside the pore, their volume is not subject to the f factoring. The form of f (R, w)
90

A |OA| = |OB| = R

R w |AB| = w
2
O 2α �AOB = 2α = 2π/m
w w
sin α =
2 2R
�w �
B ∴ m = π/ arcsin
2R

Figure 6.9: Cross section view of the pore and two polymer chains.

can be obtained from the intersection of the two circles defining the pore and polymer

chain cylinders. The details are given in Appendix B. For brevity here we will keep the

shorthand notation f (R, w). When the last three equations are substituted into Eq. (6.1)

we get the following relation:

πR2 2(n + 1)lc = mλVH2 O + mπ (w/2)2 2(n + 1)lc f (R, w) + mV side (6.5)
chain

The number of polymer chains m along the circumference of the pore can be linked to R

and w as shown in Fig. 6.9. After substitution of the expression for m in Eq. (6.5)

and some rearranging we finally arrive at the following transcendental equation for
 
R n, λ, lc , w, VH2 O , V side :
chain

w λVH2 O + V side


R2 arcsin − π (w/2)2 f (R, w) = chain
(6.6)
2R 2(n + 1)lc

The only free variables in this expression are n and λ representing the polymer formula

and the hydration level. Numerical solutions for a number of cases are listed in Table 6.1.

Not surprisingly, increasing the hydration level swells the pores leading to larger pore
91

Table 6.1: Pore radius as a function of the repeat unit formula (CF2 CF2 )n − CF2 CF −
(OCF2 CF2 SO3 H) and the hydration level λ, calculated by numerically solving Eq. (6.6).

a
n EW [g/mol] λ R [nm]

3 578 3 1.32
6 1.68
13 2.52
5 778 3 1.00
6 1.24
13 1.80
7 978 3 0.83
6 1.02
13 1.44
a
The following parameter values were used: lc = 0.131 nm, w = 0.5 nm, VH2 O = 0.0312 nm3 ,
V side = 0.156 nm3 . The chain length per backbone atom lc is calculated as the projection of a
chain
carbon-carbon bond along the cylinder axis, i.e. lc = r0 sin (θ0 /2), with the equilibrium values for
the bond length and angle taken from Tables A.1 and A.2 in Appendix A. The estimate for the
effective width of the polymer chain cylinder w is based on the values we have seen in Fig. 6.6. A
Monte Carlo method was used to calculate the volumes of a water molecule VH2 O and the side chain
V side as explained in Appendix C.
chain

radii. On the other hand, increasing the EW results in longer pores with extra volume

to fill along their axes. As the water molecules fill up this space the pores get thinner.

The numerical solutions we have obtained here for a SSC polymer with n = 7 are slightly

below the pore diameters found in a comparable Nafion system [17]. This is indeed

expected as Nafion has a longer side chain that requires more volume. With Eq. (6.6)

it is now possible to predict the effect of different factors on the pore radius in any

applicable polymer system. In particular, it allows us to use the appropriate values for

the dimensions of the pore and simulation cell when modelling the diffusion of species.
92

Chapter 7

Morphology of Hydrated SSC Polymer Systems

Previously we introduced the idea of designer systems where the morphology of the poly-

mer was controlled by some external restraint. The employed restraints were geometrical

in nature and they acted on the polymer backbone. However, the currently available syn-

thetic methods allow only variations in the EW and hydration level of the SSC polymer.

Thus, it is of great practical interest to establish the effect of these parameters on the

morphology and transport properties of the polymer. In this chapter we examine the local

morphology of a hydrated SSC polymer as a function of the hydration level λ. Given that

our modelling is based on a force field developed specifically for the SSC polymer, the

morphology discussed below is expected to reflect that of the real polymer system. For the

system studied here the EW was set at 578 g/mol corresponding to three tetrafluoroethy-

lene units and a general formula CF3 [(CF2 CF2 )3 − CF2 CF (OCF2 CF2 SO3 H)]40 CF3 .

The complete simulation details are given below.

Simulation details: All atom simulations of the above polymer strand were carried out at three

hydration levels of 3, 6 and 13 water molecules per sulfonic acid group. The simulation procedure begins

with building the polymer with a random conformation in a 3D periodic box, as described in Chapter 5.

The size of the simulation box is determined by the desired density of 1.67 g/cm3 for the hydrated

polymer systems, a value that was taken from Nafion [25]. The non-bonding interactions are calculated

with the IPS method [45,46] with the cut off of the Coulomb and Lennard-Jones interactions at 1.5 nm.

All optimizations performed in this chapter (unless otherwise stated) consist of first carrying out a SD

minimization [43] followed by an ABNR minimization [54]. After the polymer structure was optimized

it was introduced into a simulation box filled with flexible three center (F3C) water molecules [89]. The

interaction energy between each water molecule and the polymer was then calculated and all but the
93

required 40λ water molecules removed, beginning with the H2 O molecules possessing the highest energy

(i.e., those that overlap with the polymer). This procedure results in three hydrated polymer systems

with 120, 240 and 520 water molecules and a total of 1768, 2128 and 2968 atoms, respectively. The

water phase is first optimized (SD only) keeping the polymer frozen, followed by optimization of the

whole system (polymer and water). All sulfonic acid protons are then transferred to their second closest

neighbour water molecule producing solvent separated hydronium ions and the sulfonate terminated

side chains. Following proton dissociation, the atoms in the side chain are renamed according to our

scheme (see Fig. A.1), and the corresponding new potentials are used henceforth. The system is again

optimized in stages first the hydronium ions (SD only, keeping water and polymer frozen), then water

and hydronium ions (SD only, keeping the polymer frozen) and finally, the whole system. The optimized

hydrated polymer systems are then equilibrated at 300 K for up to 450 ps using an NVT simulation.

The NVT run is stopped when the predicted change in the potential energy is less than 1 % per ns. All

data is collected from a 2 ns NVE production run with data points saved every 200 steps. The time

step in all MD simulations is 1 f s.

Snapshots of typical morphologies at the three different water contents (λ = 3, 6, 13)

are shown in Fig. 7.1. Cursory examination of Fig. 7.1 shows the increasing distinctions

in the distribution of the water and hydronium ions through the polymer as the water

content is increased; an observation which is consistent with studies of Devanathan et

al. in Nafion membranes over a similar hydration range [32]. The water appears to only

form isolated clusters or domains with the sulfonate groups and hydronium ions at the

lowest hydration level (λ = 3). However, connectivity between these domains begins

to emerge at λ = 6 and might be considered as an appearance of channels. There is

also the emergence of a separation of the hydronium ions from the sulfonate groups by

several water molecules. At the highest hydration level (λ = 13) these channels appear to

permeate the morphology of the system. The sulfonate groups and hydronium ions are

found at much greater average separations from each other. However, contact ion pairs
94

Figure 7.1: Polymer morphology snapshots at the end of the production run (2 ns)
for the three levels of hydration: a) λ = 3, b) λ = 6, c) λ = 13. Each system shown
represents a 3 Å deep cross section of a 3 × 3 supercell so that the continuity can be
observed in the x and y directions. The three supercells are drawn to scale. Hydronium
ion oxygen atoms (blue) and sulfur atoms (yellow) are emphasized. The other atoms
seen are oxygen (red), hydrogen (white), carbon (light blue) and fluorine (green).
95

Figure 7.2: Hydrogen bond chain from the snapshot in Fig. 7.1a (λ = 3) using MIC.
The sulfonic acid group oxygens have been left out. The numbers on the sulfur atoms
represent the index of the side chain (1 through 40). Colouring scheme as in Fig. 7.1.

between one sulfonate group and two hydronium ions are still visible in this snapshot

(Fig. 7.1c).

Careful examination of the central section of Fig. 7.1a reveals the presence of a hydro-

gen bond chain that spans about 1/3 of the supercell. This particular chain is composed

of the following species: (going clockwise) H2 O − H3 O − SO3 − H3 O − H2 O − SO3 −

H3 O − H2 O − SO3 − H3 O − SO3 − H3 O − H2 O, which is shown in Fig. 7.2. The side

chain index numbers (on the sulfur atoms) indicate that in this particular snapshot the

sulfonate groups are not from neighbouring side chains but are 2 side chains apart in

the case of the 1-4 hydronium/water bridge, and 5 side chains apart in the case of the

31-37 hydronium ion bridge. One should note that the distances between the sulfur

atoms shown in Fig. 7.2 are measured after applying MIC. These distances are also used

in the pair-correlation plots presented later. However, since the polymer spans several

simulation cells the separation of the same sulfonate groups in the actual polymer is

quite different. The single polymeric fragment of Fig. 7.1a is shown in Fig. 7.3 without
96

Figure 7.3: Polymer chain of Fig. 7.1a (λ = 3) without MIC. Water and hydronium ions
are left out for clarity. The sulfur atoms from the hydrogen bond bridges in Fig. 7.2 are
emphasized.

applying MIC and with both the water molecules and hydronium ions left out for clarity.

This rendition of the macromolecule reveals much greater separation between the sulfur

atoms (e.g. 22.7 Å for 1-4 and 25.6 Å for 31-37). Furthermore it is evident that the poly-

mer backbone consists of a few almost perfectly straight segments. This preference in

the conformation of the backbone with the CF2 groups forming a helical pitch is similar

to that determined in the ab initio calculations of Paddison and Elliott for a two unit

oligomer [26] and a three unit oligomer [30] and this result in our classical simulations is

similar despite the polymer being placed in a periodic simulation box. The periodicity

of the model probably allows for the formation of hydrogen bond bridges while avoiding

any significant bending of the backbone. Similarly, the real polymer may organize itself

into straight bands with different threads held together by intermolecular hydrogen bond

bridges.
97

Figure 7.4: S-S pair correlation plot (between the ionized sulfonic group sulfur atoms)
for the three hydration levels.

Further quantitative information concerning molecular interactions in the hydrated

SSC polymer systems is obtained in examining the RDF, denoted as g(R) . The in-

teractions of the sulfonate groups are quantified through the sulfur-sulfur g(R) and are

plotted in Fig. 7.4 for all three hydration levels. These pair correlation plots are derived

from trajectory snapshots taken every 200 steps during the entire production run. One

striking feature in this data is the intense peak for λ = 3 at approximately 4.5 Å. This

separation between the sulfur atoms is significantly smaller than the one found in the

case of only a single bridging hydronium ion (see Fig. 7.2), and is actually very close

to the separation between a sulfur atom and a hydronium ion in a contact ion pair (see

Fig. 7.6 and the discussion below). It is also somewhat smaller than observed in the

simulations of Devanathan et al. [32] with 1134 EW Nafion where at a similar degree

of hydration they observed a broader peak with a shoulder at 4.6 Å and a maximum at
98

Figure 7.5: An ion cage that exhibits very short S-S distances. The rest of the side chain
atoms in the polymer have been omitted for clarity. The two lower hydronium ions are
also fully coordinated with three sulfonic acid groups (only two are shown).

about 5.2 Å. Scanning the trajectories in this part of the plot revealed a close packing

structure that is shown in a wire-frame rendition of the atoms in Fig. 7.5 (similar colour

convention as before).

The hydronium ions in these structures are found in ion cages with all three hydrogen

atoms hydrogen-bonded to the oxygen atoms in three different sulfonate groups. There

is no water to mitigate the strong ionic interactions in such clusters and this results in

very small separations of the sulfur atoms. Since the hydronium ions in these cages are

immobilized to a large extent it is insightful to know what fraction of the hydronium

ions exist in such a state and its dependence on the degree of hydration. This analysis

is presented later on in the chapter. The intensity of the 4.5 Å peak greatly diminishes

for λ = 6 broadening into two peaks at around 5.0 and 7.5 Å with a probability of

about one. This may be viewed as being entirely determined by the number density of

the sulfur atoms and suggests there is little interaction between the sulfonates at this
99

water content. The peak completely disappears at the highest hydration level with the

majority of the sulfonate groups seemingly very well separated by over 10 Å, the average

separation of the tertiary carbons (i.e., CFO) when the backbone carbon atoms are in an

anti conformation [26]. Another type of contact ion pair between the sulfonate groups

and the hydronium ions are those where the ionic cages are more labile - either due to

the incorporation of water or to missing hydronium ions. This is the case of the hydrogen

bond bridges in Fig. 7.2. The observed S-S separation in this type of bridges gives rise to

the broad peak near 7.5 Å that is visible in the pair-correlation plot for both λ = 3, 6. Not

surprisingly, these peaks of the labile ionic cages also disappear in the highest hydration

case.

As we have seen from the S-S pair correlation plot with an increase in hydration the

ion cages release some of the trapped hydronium ions. Another effect is the increase

in the spatial separation between the hydronium ion and the sulfonate groups once the

cages break down. This can be seen from the S-O pair correlation plot in Fig. 7.6. The

plot for λ = 3 exhibits a sharp peak for the contact ion pair just below 4 Å. A shoulder

on the left side of the peak is also visible that can be attributed to the fully “sulfonated”

hydronium ions in the ion cages. The typical S-O separation there is between 3.58 and

4.0 Å. Another broad peak for the more hydrated systems is visible at about 6 Å that

corresponds to the solvent separated species sulfonic acid group/water/hydronium ion.

As the ion cages disappear with the increase in λ, the intensity of the first peak goes

down and it becomes more Gaussian. At the same time it becomes more likely to find

the hydronium ions away from the sulfonate groups.

Now we turn our attention to a way of quantifying these observations. What we are

interested in is the mobility of the hydronium ions which is directly linked to the resilience

of the ion cages. As we have seen above any defects in the ion cage composition lead

to much greater separations between the hydronium ion and the sulfonic acid group
100

Figure 7.6: S-O pair correlation plot (between the ionized sulfonic acid sulfur atom and
the hydronium ion oxygen atom) for each of the three hydration levels.
101

Figure 7.7: Fraction of hydronium ions with a given number of sulfonate neighbours in
the SSC polymer for different hydration levels λ. The numbers have been averaged from
the production run trajectory (2 ns).

making the hydronium ion more effective as a charge carrier in the polymer membrane.

In Fig. 7.7 we show a histogram of the number of sulfonate groups present in the first

solvation shell (4.3 Å) of a hydronium ion. When the S-O separation between the two

species is within this cut off the sulfonic acid is considered a neighbour of the hydronium

ion. The histogram is drawn in a way so that a direct comparison can be made with the

corresponding plot for Nafion [32], as seen in Fig. 7.8. For the SSC polymer at λ = 3 about

1/4 of the hydronium ions are trapped in ion cages with three sulfonic acid neighbours.

Roughly, equal numbers of hydronium ions are found with one sulfonic acid neighbour,

while slightly more have two neighbours. The trend in the histogram indicates that as

the hydration level increases there is a dramatic drop in the number of fully “sulfonated”

hydronium ions (with three sulfonic group neighbours) and a significant increase in the
102

l. 111, No. 28, 2007 Devanathan et al.

ordination Numbers of
fur (nsh) and Water Molecules
n as a Function of λ
nsh nsw
2.46 0.00
2.05 2.23
1.59 3.65
1.14 4.26
0.97 4.81
0.77 5.34
0.76 5.33
0.49 5.79

13.5 and 20. The position of the


iffers from that in united-atom
observed a sharp first peak at ∼4
d nonmonotonic trends in gS-S(r)
ydration. In contrast, Urata et al.11
Figure 5. Distribution of sulfonate neighbors of hydronium ions in
ng from ∼5 Å at λ ) 2.8 to ∼67.8:
Figure Å Fraction
Nafion of forhydronium ions levels,
various hydration with λ,a given number
indicated of sulfonate neighbours in
by the legend.
coming shorter and broader
Nafionwith for different hydration levels λ. Figure source: Ref. [32].
consistent with our finding that
ge S-S coordination number at a
e average distance from a sulfur
lfur atom can be foundnumber
(coordina- of free hydronium ions (with zero neighbours). It is intriguing that the SSC
5.3, 5.6, 6.0, 6.5, 6.6, 7.0, 7.1, and
curves
3, 5, 7, 9, 11, 13.5, and for λ = 3, 6, 13 resemble the Nafion plots of Devanathan et al. but for a higher
20. Urata
creasing from 4.6 Å at λ ) 2.8-
nited-atom simulation.hydration
Thus, our level - namely λ = 5, 9, 20 water molecules. This allows the SSC system to
ulfonate groups move away from
have the same amount of free hydronium ions as Nafion but at a lower water content.
membrane hydration in qualitative
of previous simulations. 9,11,15 The
Another feature that emerges is that the correspondence between the two membranes
fur distance for different levels of
resent work establisheschanges
a classical with the hydration level - while at the lowest water content the improvement
ulfonate separation in hydrated
is ∆λ = 2 (i.e. Figure3 → 5) 6. at
Percentages of hydronium
the highest contentions
thethat have at least 1 sulfonate
improvement is already ∆λ = 7 (i.e.
anges in gOh-Oh(r) with increasing neighbor (square) and multiple sulfonate neighbors (triangle) as a func-
es in gS-S(r), especially13 → at 20).
low These tionsimulation
of hydration level, λ, The
results canpercentage
be usedofto nondiffusing
explain hydrogen atoms
the differences found in the
obtained from neutron scattering experiment25 is represented by circles.
sts that hydronium ions may be
roups at low λ consistentexperimental
with the conductivity plots for the two polymer systems [18]. In particular, it was
these first two peaks decreases with increasing λ, while the area
s seen in Figure 3a. The area under
found under the curve for the third neighbor shell located between
ncreasing λ indicating that that
the a low EW the SSC polymer system has a superior conductivity to Nafion at
∼5.2 and 7.2 Å increases with increasing λ. This indicates
onium oxygen atoms around each
the same hydration level, and
increasing with increasing
separation with increasing λ. In agreement
the water content thiswithsuperiority
Cui becomes
The average distance from a 14
h another hydronium even ion can et al., we do not observe an “apparently artificial peak” at 3.2
be pronounced.
more Å that Petersen et al.17 observed in their gOs-Oh(r) from classical
5.9, 6.1, 6.4, 6.6, 6.8, 7.1, and 7.6
, 5, 7, 9, 11, 13.5, and 20. simulations, which they attributed to the limitations of the
In With
conclusion, by carefully
classical developing
hydronium a SSCthespecific
ion. While force
classical field we ion
hydronium have obtained new
on, the hydronium ions move away
apart. into themodel
ulfonate groups move insights does not incorporate proton shuttling (see ref 21), its
morphology of the SSC polymer systems as a function of the hydration
epresent the sulfonate-hydronium limitations may have been overstated in the literature. Our
simulations show that the hydronium ion can move away from
-Oh(r) and gOs-Oh(r), respectively.
shows a dominant peak between the sulfonate group without the use of potentials that allow
t, the united-atom simulations of proton transfer.
k at 4 Å for the four λ values they The second peak in gOs-Oh(r) has been attributed by Cui et
14
al. to the sulfonate group and hydronium ion being separated
e is an additional peak at 3.2 Å
103

level. The polymer backbone is able to assume straight configurations while at the same

time forming strong hydrogen bond bridges. Ion cages with fully “sulfonated” hydronium

ions are typical of the low hydration level and explain the previously unresolved peaks

(in Nafion) in the S-S and S-O pair correlation plots. The trends seen in a structural

parameter representing the number of sulfonic groups around a hydronium ion allows us

to explain the differences between the proton mobility of Nafion and the SSC polymer,

which is discussed in the next chapter.


104

Chapter 8

Proton Diffusion in SSC Polymer Systems

In this chapter we present the results of the work we have carried out on the simulation

of vehicular proton diffusion in the SSC polymer. In addition, two other systems were

examined – one involving a single excess proton in water and a second one consisting of a

multiproton triflic acid solution. These latter systems were considered, solely, as tests for

our JIT-EVB method. The lessons we have learned from these experimental runs, and

in particular from triflic acid, give us the confidence needed to apply the new method to

polymer system.

First, we present a brief description of the manner whereby the diffusion process was

followed in the simulations. In the polymer systems all side-chains were considered ion-

ized, hence the simulations were performed on a mixture comprised of tethered sulfonate

groups, hydronium ions and water molecules. In the absence of proton hopping events

the proton diffusion is naturally determined from the diffusion of the hydronium ions,

more specifically, from the trajectory of the oxygen atom in H3 O+ . Similarly, in the

case of the excess proton in water studies, carried out with the JIT-EVB method, proton

diffusion was examined through the oxygen atom associated with the excess charge. This

representation of the proton diffusion is commonly referred to as the Center of Excess

Charge (CEC) representation. Finally, in the simulation of triflic acid solution (which,

like the polymer systems was considered to be fully ionized) an oxygen atom from a

sulfonate group could potentially be a CEC as well, if a hydronium ion donates a proton

back to SO3− .

In all of our studies we have chosen to evaluate the diffusion constant from the Einstein
105

relation using the MSD of the CEC:




|R(t) − R(0)|2
D = lim (8.1)
t→∞ 6t

We have found that the alternative Green-Kubo relation [86] which employs the velocity

autocorrelation functions is inconvenient to use owing to the poor convergence in the

t → ∞ limit.

The proton diffusion data for the SSC polymer system is based on the same simulation

runs1 discussed in Chapter 7. The average of the MSD of the hydronium ions for the

three hydration levels is shown in Fig. 8.1. One should note the crucial importance of the

initial configuration averaging in order to obtain accurate results. The predicted values

for the diffusion coefficient are 2.84×10−7 , 1.36×10−6 , and 3.47×10−6 cm2 /s for water

contents of 3, 6, and 13, water molecules per sulfonic acid group. The agreement with

the experimental results of Kreuer et al. [90] is nearly quantitative at the lowest water

content. This is a strong indication that the dominating transport mechanism under

minimal hydration may be vehicular. Still, even under these conditions proton hopping

does occur, as demonstrated recently by accurate ab initio MD simulations [91]. The

proton hopping is only localized and does not lead to significant displacement of the

protons. On the other hand, the diffusion coefficients for λ = 6 and λ = 13 are much

lower than those obtained from the experimental measurements suggesting an increased

contribution from structural diffusion in these systems.

Since the force field used in the present calculation is based on the one used in Ref. [25]

to study the hydronium ion diffusion in Nafion, it is, indeed, remarkable to see how well

the two polymer systems are differentiated in the simulations. The only hydration level

investigated for the Nafion system was λ = 15 where one finds an average hydronium

diffusion coefficient about 2.5 times smaller than the one we found here for λ = 13. This
1
Additional simulation details: The H3 O+ diffusion coefficients were calculated from an average over
five initial configurations R(0), taken from the same trajectory about 25 ps apart, using the Einstein
relation, Eq. (8.1). A second averaging is done over the diffusion coefficients of the 40 hydronium ions.
106

Figure 8.1: MSD of hydronium ions in the SSC polymer as a function of time and
hydration level λ. The slope of the curves gives directly 6D in units cm2 /s. The average
temperature from the 2 ns NVE production runs was 315 K.
107

is in qualitative agreement with the conductivity difference between a SSC polymer of

EW 800 and Nafion [18, 90]. As the conductivity of the polymer samples is proportional

to the total diffusion coefficient (and not just the vehicular component) it is important to

know if the vehicular contribution changes significantly between the two polymer systems.

In the case of Nafion the computed vehicular diffusion coefficient corresponds to 19.5%

of the total measured diffusion, whereas in the case of the SSC polymer the computed

value is about 20.4%. Since the hydration level of the Nafion system is a bit higher (i.e.,

λ = 15) it is reasonable to expect a higher contribution from the structural diffusion

and a concomitant reduction of the vehicular component. We can conclude that the

two polymer systems exhibit practically the same partition between their vehicular and

structural modes of proton transport at least in the case of λ between 13 and 15 water

molecules per sulfonic acid group.

The failure of the classical MM approach to account for the major part of the diffusion

coefficient was the motivation that obliged us to develop the the JIT-EVB method. While

reproducing experimental results is the ultimate goal we have chosen not to resort to

any system-specific fitting parameters, including specially developed force fields for the

species involved. This is in stark contrast to the alternative EVB implementations which

rely heavily on extensive, system-specific parametrization. The only kinds of parameters

used in the JIT-EVB method are those which determine the quality of the ab initio

PES fit. These, for example, are the validity radius around a grid point (i.e., the grid

point spacing), the order of the polynomials used in the fitting etc. Consequently, the

agreement between the JIT-EVB simulations and experiment is controlled by the ab

initio method used.

The first system that we tested the JIT-EVB method on was an excess proton with

64 water molecules. This is a very small simulation system and the goal here is not to

extract any physical constants, but rather to determine the simulation parameters (e.g.,
108

Table 8.1: Diffusion coefficients obtained from a JIT-EVB simulation of excess proton
with 64 water molecules in a cubic periodic box. Five runs were started with different
initial configurations. The diffusion constant including both the vehicular and Grotthuss
components appears in the second column. The third column shows the number of grid
points in the reactive trigger zone (i.e., the number of ab initio calculations and fittings
performed). Last column shows the root mean square (RMS) difference between the QM
and classical forces in the superposition of resonance forms.

Run # D [cm2 /s] grid points Frms [kJ mol−1 nm−1 ]

1 9.3×10−5 6 843
2 1.9×10−5 5 819
3 3.7×10−6 3 853
4 8.9×10−8 6 733
5 1.7×10−5 4 883
a
Simulation details: the forcefield parameters used for all JIT-EVB simulations are given in Ap-
pendix D. Proton hopping was not allowed during the equilibration. Accordingly, the MIC bonding
potentials were not invoked in this stage. Non-bonding interactions were treated with the IPS
method [46] using the maximum cutoff (i.e., half a box length). Excess proton and 64 F3C water
molecules [89] were placed randomly in a periodic box of 1.2429 nm length on all sides. The systems
were optimized first with the SD method [43]. Random velocities with zero net momentum were then
assigned corresponding to the target temperature of 298 K. Each system was equilibrated for 5×103
steps with the generalized Nose-Hoover method [55] in the canonical ensemble using a thermostat
period of 72.0 f s and a time step of 1 f s. After the systems were equilibrated proton hopping
was turned on and MIC applied to the bonding potentials. The details of the JIT-EVB simulation
are given in Appendix D. Diffusion data was collected during a 50 ps production run with a time
step of 0.2 f s. In each of the five trajectories an averaging of the CEC MSD was done over four
initial configurations, taken 250 steps apart. One Berendsen thermostat [93] was used for the atoms
outside any reactive zones and separate thermostats were used for each of the reactive zones. All
thermostats had the same target temperature of 298 K, while their periods were 1 ps for the former
and 0.1 ps for the latter kind.

time step, thermostat frequency, grid spacing etc.) that lead to stable MD trajectories.

Still, a comparison can be made between our results (shown in Table 8.1) and diffusion

results from a Car-Parrinello ab inito MD [92] study of the same system.

Not surprisingly, due to the small system size we observe a high variance in the

predicted diffusion constant – nearly three orders of magnitude difference in some cases

which underlines the need for extensive sampling (or larger systems) before sound results

can be obtained. Never the less, in some cases we see significant enhancement of the

diffusion constant compared to the experimental vehicular diffusion in bulk water (about
109

2.3×10−5 cm2 /s). A more sensible comparison, however, is with the results of ab initio

MD simulations as our JIT-EVB method is designed to match those forces. Depending on

the functional and the initial conditions Car-Parrinello MD results estimate the proton

diffusion constant between 0.5×10−5 and 2.1×10−5 cm2 /s [92]. For comparison, the

average of the data in Table 8.1 is 2.6×10−5 cm2 /s. It can be expected that with an

increase of the simulation size our results will further increase [94] bringing diffusion

constant closer to the experimental value of 9.307×10−5 cm2 /s [58]. Despite the high

statistical noise in the diffusion constant these preliminary tests of the JIT-EVB method

reveal a highly consistent number of grid points and fit quality (last two columns). In

some cases as few as three grid points are all that is needed to represent the active region

of the PES.

In our EVB method the interacting species are represented as a superposition of

resonance forms, with the coefficients selected such that the desired (i.e., ab initio) forces

are reproduced. The last column in the table shows the difference with the QM forces

remaining after the fit. In order to put the values in perspective, we note that the largest

difference seen in the first entry (883 kJ mol−1 nm−1 ) has the magnitude of the restoring

force in a water OH bond stretched to 1.04 Å. The limitations of DFT in accurately

describing hydrogen bonds are well known and they stem from non-local correlations

in the van der Waals forces [95]. As the PBE functional that was used here does not

incorporate any van der Waals correction terms we can conclude that at least some of the

difference between the DFT and the classical forces is due to these terms. Accordingly,

by representing the DFT forces as a superposition of classical forces in the resonance

forms, the JIT-EVB method can improve the agreement with experiment, particularly

in systems with hydrogen bonds where the van der Waals interactions are important.

Lastly we show the diffusion results for triflic acid at the intermediate hydration level

of λ = 6, see Table 8.2. The MSD of the CEC as a function of time is shown in Fig. 8.2.
110

Table 8.2: Diffusion coefficients obtained from a JIT-EVB simulation of five triflic acid
molecules and 30 water molecules. The final density obtained for the system after relax-
ation appears in the first column. The diffusion constant including both the vehicular
and Grotthuss components appears in the second column. The third column shows the
number of grid points in the reactive trigger zone (i.e., the number of ab initio calcula-
tions and fittings performed). Last column shows the RMS difference between the QM
and classical forces in the superposition of resonance forms.

ρ [g/cm3 ] D [cm2 /s] grid points Frms [kJ mol−1 nm−1 ]

1.556 1.1×10−5 101 881


a
Simulation details: the forcefield parameters used for all JIT-EVB simulations are given in Ap-
pendix D. Proton hopping was not allowed during the equilibration. Accordingly, the MIC bonding
potentials were not invoked in this stage. Non-bonding interactions were treated with the IPS
method [46] using the maximum cutoff (i.e., half a box length). Five triflate ions F3 CSO3− , five
hydronium ions H3 O+ and thirty F3C water molecules [89] were placed randomly in a periodic box
of 1.2893 nm length on all sides, corresponding to a density of 1 g/cm3 . The system was optimized
first with the SD method [43] for 1500 steps. Random velocities with zero net momentum were then
assigned corresponding to the target temperature of 298 K. Each system was equilibrated for 5×103
steps with the generalized Nose-Hoover method [55] in the canonical ensemble using a thermostat
period of 81.6 f s and a time step of 1 f s. This was followed by 3×104 equilibration steps in the
NPT ensemble performed with the generalized Nose-Hoover method [87] using a barostat period of
1 ps. The final density for the system was 1.556 g/cm3 . After the system was equilibrated, proton
hopping was turned on and MIC applied to the bonding potentials. The details of the JIT-EVB
simulation are given in Appendix D. Diffusion data was collected during a 60.6 ps production run
with a time step of 0.2 f s. In each of the five trajectories an averaging of the CEC MSD was done
over four initial configurations, taken 250 steps apart. One Berendsen thermostat [93] was used for
the atoms outside any reactive zones and separate thermostats were used for each of the reactive
zones. All thermostats had the same target temperature of 298 K, while their periods were 1 ps for
the former and 0.1 ps for the latter kind.
111

[nm2 ]

2
[R(t) − R(0)]

t [ps]
Figure 8.2: MSD of the CEC in triflic acid solution as a function of time. The hydration
level is λ = 6.

The diffusion constant obtained here is nearly one order higher compared to the vehicular

diffusion constant calculated for the SSC polymer. The JIT-EVB predicted diffusion is

also higher than the experimental one of the SSC polymer at this hydration level. As our

simulation lacks the polymer component that tethers the end acid groups and obstructs

the free movement of hydronium ion such an increased mobility is expected. Of note is

the significantly higher number of grid points required in this system. This is due to the

wide variety of hydrogen bond complexes that occur in this multiproton, concentrated

acid solution. The acid dissociation is a favourable, though not barrierless process. For

this reason the acid was introduced as triflate and hydronium ions. In this manner the

diffusion can be sampled immediately (after some equilibration) without requiring that
112

we first observe the rare proton dissociation events.

In these preliminary examples of the JIT-EVB method we have obtained qualitatively

reasonable results. Application of the method to larger systems, including the SSC

polymer is currently being carried out.


Part III

Conclusions and Future Work

113
114

Chapter 9

Conclusions

The subject of this thesis was the development of the necessary tools and their application

for exploring the morphology and proton transport in the PFSA SSC membranes. A

major part of the work related to the forces in the modelled systems since they are of

paramount importance, affecting the dynamics, kinetic energy, temperature, pressure in

the system, and so on. As the simulation methodology was MM, all forces are in principle

derived from the force field. For our studies we have used a special force field that works

well for the SSC polymer. Furthermore, the traditional view that a force field needs to

be known beforehand has been replaced by the method we have developed for creating

the force field parametrization on the fly, during a MD simulation. With this approach

we were able to simulate bond breaking and making events by switching between force

field parameters, depending on the current value of the reaction coordinates. It is hoped

that these developments will broaden the concept of classical MM simulations.

The second pivotal point of the work was overcoming the problem of high dimen-

sionality. For example, a conformational search algorithm (like the brute force geometry

optimization) that works well for a small number of degrees of freedom is absolutely

useless for a system like a polymer. Another example is the parametrization of the PES

of a reaction with multiple degrees of freedom. An important conclusion here is that

even the real systems, that exist on a much larger timescale, do not necessarily sample

the entire available phase space. Accordingly, we have developed the methodology that

only operates on small, tractable regions of the phase space. This approach has allowed

us to build systems of thousands of atoms in conformations that may resemble the real

systems and parametarize a PES surface only in the most frequently visited regions.
115

Chapter 10

Future Work

The wide scope of the methodology developed and presented in this thesis has, unfor-

tunately, presented serious time limitations on the number cases we were able to study

in the time allowed. A natural extension of the work, in particular the proton hopping

studies, would include bigger systems where the polymer backbone is also included in

the JIT-EVB simulation. Establishing a structure–properties relationship in the SSC

polymer can then be undertaken by varying the monomer ratios, EW distribution, the

total MW of the polymer etc. As we have seen in the last chapter, a relatively high

number of grid points may be required in the multi-proton, concentrated acid systems.

Simulations that are run in parallel on equivalent systems will be beneficial if the code

allows pooling of the grid points between the computers. Thus the PES will be required

to be parameterized only once. Once the grid points are known for a particular system

it will be possible to smoothly interpolate the forces in the reactive regions of the PES.

This will allow for energy conservation in the JIT-EVB method, and possibly, an increase

of the time step. Since the JIT-EVB method is quite general it may find applications in

other areas, particularly biological systems.


Part IV

Appendixes and Bibliography

116
117

Appendix A

Short-Side-Chain Force Field

F1T F1 F1O
O1
F1T C1T C1 C1O
H1 H1
F1T F1 O3 water
terminal group
F2O C2O F2O

H2 F2S C2S F2S C2SI

O4 S6 O4 O4I S6I O4I


O2
H2 H2 O3H O4I
H3
hydronium ion ionized side-chain

Figure A.1: Atom type labels for the SSC force field. Water and hydronium ion are
shown for completeness only, their force field parameters are as described in Ref. [25]
118

Table A.1: SSC force field parameters for the harmonic stretching potential Ebond .

Atom types r0 [nm] k [kJ mol−1 nm−2 ]

H1 O1 0.1000 209200
H2 O2 0.0982 454364
H3 O3H 0.0974 292880
O3H S6 0.1628 292880
O4: S6: 0.1451 585760
S6: C2S: 0.1886 292880
C2S: C2O 0.1555 292880
C: F: 0.1348 253240
C:O O3 0.1392 292880
C1:O C1: 0.1566 292880
C1: C1: 0.1564 179627
2
a
Ebond = 0.5k (r − r0 )
b
Force constants k are taken from Ref. [25].
c
A colon (:) in the atom names represents a wildcard (e.g. C: covers C1, C1T, C1O, C2O, C2S, C2SI,
etc.).
119

Table A.2: SSC force field parameters for the harmonic bending potential Eangle .

Atom types θ0 k [kJ mol−1 rad−2 ]

H1 O1 H1 109.47 502.08
H2 O2 H2 113.40 330.64
C: C: C: 113.64 471.45
F1O C1O C1: 106.43 472.04
O3 C1O C1: 112.23 470.70
O3 C1O F1O: 105.60 470.70
C: C: F: 108.62 472.04
F: C: F: 109.02 470.70
F2O C2O O3 111.82 470.70
C2S: C2O O3 106.96 470.70
C2O C2S: S6: 114.52 490.91
F2S C2S: S6: 106.76 478.15
O4: S6: O4: 123.71 509.36
O4 S6 O3H 108.07 509.36
C2S: S6 O3H 100.36 456.31
C2S: S6: O4: 107.09 456.31
C1O O3 C2O 126.40 470.70
2
a
Eangle = 0.5k (θ − θ0 )
b
Force constants k are adapted from Refs. [25] and 23.
c
A colon (:) in the atom names represents a wildcard (e.g. C: covers C1, C1T, C1O, C2O, C2S, C2SI,
etc.).

Table A.3: SSC force field parameters for the torsion potential Edih .

Atom types a [kJ mol−1 ] b c [deg]

H3 O3H S6 C2S: 7.4185 1.3349 -21.037


O3H S6 C2S: C2O -9.5835 1.1453 17.673
S6: C2S: C2O O3 -34.0433 1.0306 -0.931
C2S: C2O O3 C1O 8.4275 1.1671 14.264
C2O O3 C1O C1: 38.2339 0.9381 32.494
C1: C1: C1: C1: 5.6356 1.0258 -49.553
O4: S6: C2S: C2O 12.0085 0.4842 -6.642
a
Edih = a cos(bϕ − c)
b
A colon (:) in the atom names represents a wildcard (e.g. C: covers C1, C1T, C1O, C2O, C2S, C2SI,
etc.).
120

Table A.4: SSC force field parameters for the non-bonding interactions ECoulomb and ELJ .

Atom type q ε [kJ mol−1 ] σ [nm]

H1 0.4100 0.0418400 0.08018


O1 -0.8200 0.7732032 0.31655
H2 0.4606 0.0418400 0.08018
O2 -0.3818 0.7732032 0.31655
F1 -0.2709 0.2075264 0.30249
C1O 0.4462 0.3978984 0.34730
F1O -0.2741 0.2075264 0.30249
C2S 0.3715 0.3978984 0.34730
C2SI 0.3234 0.3978984 0.34730
H3 0.3882 0.0004184 0.28464
F2O -0.2569 0.2075264 0.30249
F2S -0.2414 0.2075264 0.30249
O3H -0.5140 0.4004088 0.30332
S6 1.2412 1.4392956 0.35903
S6I 1.0237 1.4392956 0.35903
C1 0.5497 0.3531296 0.34599
O4 -0.4512 0.4004088 0.30332
O4I -0.5876 0.4004088 0.30332
O3 -0.5392 0.4004088 0.30332
a
ECoulomb = k q1rq2 , with k = 138.93547 kJ mol−1 nm
√ h   i
+σ2 12 +σ2 6
b
ELJ = 4 ε1 ε2 σ12r − σ12r
c
The Lennard-Jones parameters and are obtained from Ref. [25].
121

Appendix B

Cross Section Factor

The pore model presented in Fig. B.1 emphasizes that each polymer chain (shown as

cylinder) is considered only partially inside the pore. When two circles of radii R and

Figure B.1: Cross section view of a pore formed by polymer chains. In our model each
polymer chain (represented by a cylinder) is bisected by the radius of the pore. The
part of the polymer chains on the inside is shown in yellow, the part outside the pore in
brown. The remaining pore volume is filled with side chains and water molecules (not
shown).

r separated by a distance d intersect each other the common area can be found as in

Eq.(14) of Ref. [96]:

2 d2 + r 2 − R 2 2 d2 − r2 + R2
A (R, r, d) = r ArcCos + R ArcCos
2dr 2dR
1p
− (d + r − R) (d − r + R) (−d + r + R) (d + r + R) (B.1)
2
122

Since the centers of the polymer chains lie on the pore surface we can make the substi-

tution d = R. For the radius r we will use half the effective width of the polymer chain,

i.e. w/2. This leads to a cross section area of:


 
1 2 4R 2 w2 w√
A (R, w) = w ArcSec + R ArcCos 1 − − 16R2 − w2 (B.2)
4 w 8R2 8

We define the cross section factor f (R, w) as the ratio of the cross section area A (R, w)

and the polymer chain area π (w/2)2 resulting in:


  r
A (R, w) 1 4R 4R2 w2 1 16R2
f (R, w) = = ArcSec + ArcCos 1 − − − 1 (B.3)
π (w/2)2 π w πw2 8R2 2π w2
123

Appendix C

Molecular Volume

Here we explain the method used to determine the molecular volume of a water molecule

and the polymer side chain. A trial point will be considered lying within the molecule if

it is within σ/2 from any atom, σ being the Lennard-Jones parameter. For every atom

type we use the corresponding value of σ from Table A.4 in Appendix A. Trial points are

picked randomly within the smallest rectangular box that the molecule will fit in. For

a trial point within the molecule the counter Nhit is increased by one. The molecular

volume is then calculated as Vmol = (Nhit /Ntotal ) Vbox .

The molecule conformation used for the volume estimate was obtained from a geom-

etry optimization in an infinite simulation universe, with no other species present. The

total number of trial points Ntotal was 1×105 for water and 2×106 for the side chain1 ,

resulting in molecular volumes of VH2 O = 0.0312 nm3 and V side = 0.156 nm3 . To
chain

check the validity of this approach one can compare the known density of water, about

1.0 g/cm3 , to the value of 0.96 g/cm3 that follows from the above molecular volume.

1
For the volume estimate of the side chain we consider only the CF2 CF2 SO3 H fragment, i.e. without
the branching oxygen atom, as the latter falls within the volume of the backbone cylinder.
124

Appendix D

JIT-EVB Simulation Details

In our previous calculations the harmonic stretching potentials were used for all molec-

ular bonds. However, In order to facilitate the proton hopping mechanism in EVB a

more suitable potential is the anharmonic Morse potential. This new potential has been

employed for modelling the OH bonds in hydronium ion and the sulfonic acid group.

Furthermore, the atomic charges in H3 O+ are also different from those that were used in

our vehicular diffusion studies, and have been updated in accordance with the hydronium

ion model of Ref. [67]. All other force field parameters not listed here remain unchanged

from those found in Appendix A. Of particular interest may be the potential of water,

for which we employ the standard F3C model [89] without any modifications. Even

when proton hopping is not allowed (e.g., during the initial equilibration stage) the force

field parameters listed below are still used in the conventional manner (i.e. without such

complications as mixing of resonance forms or applying MIC to the bonding potentials).

Table D.1: Force field parameters for the anharmonic stretching potential EM orse used
in JIT-EVB simulations. Atom labels appear in Fig. A.1.

Atom types D [kJ mol−1 ] a [nm−1 ] r0 [nm]

H2 O2 603.0 11.85 0.098


H3 O3H 603.0 11.85 0.098
2
a
EM orse = D 1 − ea(r0 −r)
b
Force field parameters for hydronium ion are taken from Ref. [67]. The same parameters are assumed
valid for the OH bond in the sulfonic acid group.

In Chapter 4 the idea of geometrical triggers were introduced. These triggers, that

define the extent of the reactive zone of the PES, are delineated by the following criteria:

• distance between the hopping proton Hd and the accepting oxygen Oa below 2.2 Å.
125

• distance between the donor and acceptor oxygens Od and Oa below 2.85 Å.

• angle Hd Od Oa below 30◦ .

• angle Od Oa Ga between 110◦ and 180◦ (where Ga is along the bisector of the angle

in the accepting water, or the sulfur atom in SO3− ).

At every step during the simulation the molecular configurations are checked against

these triggers. If no structures satisfy the trigger conditions the properties of the whole

system are calculated conventionally, without invoking the JIT-EVB method. If, on the

other hand, a structure does satisfy the trigger conditions it is denoted as a cluster,

illustrated in Fig. D.1.

Once we have determined the clusters that represent the reactive zones the next step

is to determine the resonance forms for each of those clusters. Combinatorics gives 2n

resonance forms for a cluster with n reactive hydrogen bonds. For example, in Fig. D.1a

we would have two resonance forms, four resonance forms in the case of sulfonate (b)

and eight in the case of water wire (c). However, resonance forms that correspond to

rare events like the ionization of water will, in general, have a very small contribution.

Thus, we choose to exclude all combinations of PTs that will lead to over-protonation of

the accepting species or under-protonation of the donors. For each cluster we go through

each of the acceptable resonance forms, change the atom types according to the new

Table D.2: Atom charges used in JIT-EVB simulations. Atom labels appear in Fig. A.1.

Atom type q

H2 0.33
O2 0.01
F2S -0.2107
a
ECoulomb = k q1rq2 , with k = 138.93547 kJ mol−1 nm
b
Charges on the hydronium ion are taken from Ref. [67]. The charge on the fluorine atoms is adjusted
such that the total charge of the triflate ion F3 CSO3− is minus one.
126

H O H
H O H
H O h2 S O H O
H O S O H O
H O H
H O H
h1 h1
a O b O
O H O
H H
H H
H H

H H
h1 h2 h3
c O H O H O H O

H H H H

Figure D.1: When a hydronium ion and a water molecule satisfy the trigger conditions
for hydrogen bond h1 they both become part of a cluster (a). If the same hydronium
ion is hydrogen-bonded to a sulfonate group and the bond h2 also satisfies the trigger
conditions, the cluster will include both the sulfonate group and the water (b). A third
example with three hydrogen bonds is the water wire (c). Hence a cluster contains all
species connected by reactive hydrogen bonds. For simplicity the charges on the species
have been omitted. Hydrogen bonds are depicted with dashed lines.
127

identity of the atoms (e.g. the hydronium ion oxygen O2 becomes a water oxygen O1

etc.), update all force field parameters and calculate the properties of the resonance form

(i.e., energy, gradient and virial). The resonance form properties are calculated from the

bonding terms entirely within the cluster, and the non-bonding interactions within the

cluster plus its periodic images.

Having determined the properties of the resonance forms we now have a basis in which

to expand the ab initio forces. At this point an ab initio calculation is carried out for

each cluster, supplemented with capping atoms if needed1 . There is no restriction on the

QM method used to obtain the forces, other than that it must be done with the same

PBC. Here we have used the ASE library [97] combined with the GPAW calculator [98]

that performs a DFT calculation using plane-waves. The PBE functional was used with

the default atomic setups. This ab initio calculation gives us the first grid point in the

reactive zone of the PES.


P
Now we fit the QM forces to a superposition of resonance form forces, i.e. F = ci F i .
i
The fit will be most accurate for the current conformation of the cluster. As the geometry

of the cluster changes, the fit coefficients ci must be updated accordingly. To overcome

this problem each coefficient ci (where i designates the resonance form) are considered
P4
j P
4
product of two polynomials of the reaction coordinates: c0i = sj ROO tj wj , where
j=0 j=0
w = 1 − δ/ROO , δ = ROd Hd − ROa Hd . The primed coefficient c0i have to be squared and

normalized in order to correspond to a probability, which is achieved through the relation



0 2
P 0 2
ci = (ci ) (ci ) . When resonance form i requires PTs across multiple hydrogen bonds
i
(e.g., the resonance form corresponding to a hydronium ion on the right in the water wire

of Fig. D.1c) the reaction coordinates ROO and w are averaged over all participating

hydrogen bonds. Accordingly, in this example ROO = (ROO,h1 + ROO,h2 + ROO,h3 ) /3

and w = (wh1 + wh2 + wh3 ) /3. Thus the unknown fitting coefficients are the sj and tj in
1
In the case of sulfonic groups the carbon atom is part of the cluster, while the ab initio calculation
includes three extra capping fluorines.
128

the expansion polynomials of each resonance form coefficient c0i . The fitting is performed

using the basic particle swarm optimization method [99]. Twenty walkers were employed

in the method for 50 optimization cycles.

The cluster properties used for the force fitting have been calculated with the as-

sumption that clusters are isolated from each other and the rest of the atoms. This

however is not true, so in order to obtain the properties of the whole system additional

interactions have to be taken into account. Once the resonance form coefficients ci are

known each cluster is put in the resonance form with the highest contribution. This is

done to ensure that upon exit from the reactive zone the clusters will be in their correct

resonance form. The charge and Lennard-Jones parameters of the clusters are updated as

a superposition of the resonance form parameters, e.g. the charge on a particular cluster
P
atom is calculated as q = ci qi . The non-bonding interactions are now recalculated to
i
include all atoms, whether in a cluster or not, employing the exclusion lists of the highest

contribution resonance states. The bonding interactions are calculated as sum of two

classes. In the first class are terms that are entirely within some cluster. Such terms are

subject to the resonance form mixing. The second class of bonding terms are those that,

at least partially, are outside any clusters. Such terms are unaffected by the resonance

forms and are evaluated and summed directly, without any weighting.

As explained in Chapter 4 the parametrization obtained for the first grid point is

assumed valid in some neighbourhood around it, determined by the validity radius. Here

we use a value of 0.387 Å for this radius. One should note that the described fitting

procedure is not performed for equivalent clusters. Thus, a system with many hydronium

ions may need only a couple of clusters to be parametrized, for instance, one for Zundel

ion type clusters and another one for Eigen ion clusters. All parametrization is kept on

file. As the simulation progresses more grid points are accumulated. Accordingly, there

is a build-up stage of the simulation where most of the ab initio work is done, while in
129

the latter stages hardly any QM calculations are necessary. As the molecules explore the

reactive zone PES they employ the parameters of their closest grid point.
Bibliography

[1] Michael Ball. The Hydrogen Economy: Opportunities and Challenges, chapter

Fossil fuels - supply and future availability. Cambridge University Press, 2009.

[2] Daniel G. Nocera. On the future of global energy. Daedalus, 135(4):112–115, 2006.

[3] Daniel G. Nocera. Personalized Energy: The Home as a Solar Power Station and

Solar Gas Station. ChemSusChem, 2(5):387–390, 2009.

[4] The increase in oil consumption, 2005. [Online; accessed 9-March-2010].

[5] Jiujun Zhang. PEM Fuel Cell Electrocatalysts and Catalyst Layers: Fundamentals

and Applications. Springer, 2008.

[6] Elna Schirrmeister Frank Marscheider-Weidemann and Annette Roser. The Hy-

drogen Economy: Opportunities and Challenges, chapter Key role of fuel cells.

Cambridge University Press, 2009.

[7] P. R. Resnick. A short history of Nafion®. L’Actualité Chimique, 301:144–147,

2006.

[8] A. Ghielmi, P. Vaccarono, C. Troglia, and V. Arcella. Proton exchange membranes

based on the short-side-chain perfluorinated ionomer. Journal of Power Sources,

145(2):108 – 115, 2005.

[9] Stephen J. Paddison. Device and Materials Modeling in PEM Fuel Cells, chap-

ter Proton Conduction in PEMs: Complexity, Cooperativity and Connectivity.

Springer, 2009.

[10] Kenneth A. Mauritz and Robert B. Moore. State of Understanding of Nafion.

Journal of Chemical Information and Computer Sciences, 104:4535–4586, 2004.

130
131

[11] T. D. Gierke, G. E. Munn, and F. C. Wilson. The morphology in nafion perfluori-

nated membrane products, as determined by wide- and small-angle x-ray studies.

Journal of Polymer Science: Polymer Physics Edition, 19(11):1687–1704, 1981.

[12] Howard W. Starkweather. Crystallinity in perfluorosulfonic acid ionomers and

related polymers. Macromolecules, 15(2):320–323, 1982.

[13] Romuald Wodzki, Anna Narebska, and Wojciech Kwas Nioch. Percolation conduc-

tivity in Nafion membranes. Journal of Applied Polymer Science, 30(2):769–780,

1985.

[14] Xueqian Kong. Characterization of Proton Exchange Membrane Materials for Fuel

Cells by Solid State Nuclear Magnetic Resonance. PhD thesis, Iowa State Univer-

sity, 2010.

[15] Laurent Rubatat, Anne Laure Rollet, Gerard Gebel, and Olivier Diat. Evidence

of Elongated Polymeric Aggregates in Nafion. Macromolecules, 35(10):4050–4055,

2002.

[16] K. D. Kreuer. On the development of proton conducting polymer membranes for

hydrogen and methanol fuel cells. Journal of Membrane Science, 185(1):29 – 39,

2001.

[17] Klaus Schmidt-Rohr and Qiang Chen. Parallel cylindrical water nanochannels in

Nafion fuel-cell membranes. Nature Materials, 7(1):75–83, 2008.

[18] Klaus-Dieter Kreuer, Stephen J. Paddison, Eckhard Spohr, and Michael Schuster.

Transport in Proton Conductors for Fuel-Cell Applications: Simulations, Elemen-

tary Reactions, and Phenomenology. Chemical Reviews, 104(10):4637–4678, 2004.


132

[19] Stephen J. Paddison, Reginald Paul, and Jr. Thomas A. Zawodzinski. A Sta-

tistical Mechanical Model of Proton and Water Transport in a Proton Exchange

Membrane. Journal of The Electrochemical Society, 147(2):617–626, 2000.

[20] Stephen J. Paddison, Reginald Paul, and Jr. Thomas A. Zawodzinski. Proton fric-

tion and diffusion coefficients in hydrated polymer electrolyte membranes: Com-

putations with a non-equilibrium statistical mechanical model. The Journal of

Chemical Physics, 115(16):7753–7761, 2001.

[21] S.J. Paddison and R. Paul. The nature of proton transport in fully hydrated

Nafion®. Physical Chemistry Chemical Physics, 4(7):1158–1163, 2002.

[22] Stephen J. Paddison and Thomas A. Zawodzinski Jr. Molecular modeling of the

pendant chain in Nafion®. Solid State Ionics, 113-115:333 – 340, 1998.

[23] Aleksey Vishnyakov and Alexander V. Neimark. Molecular simulation study of

Nafion membrane solvation in water and methanol. The Journal of Physical Chem-

istry B, 104(18):4471–4478, 2000.

[24] Aleksey Vishnyakov and Alexander V. Neimark. Molecular Dynamics Simulation

of Microstructure and Molecular Mobilities in Swollen Nafion Membranes. The

Journal of Physical Chemistry B, 105(39):9586–9594, 2001.

[25] Seung Soon Jang, Valeria Molinero, Tahir Çaǧin, and William A. Goddard.

Nanophase-Segregation and Transport in Nafion 117 from Molecular Dynamics

Simulations: Effect of Monomeric Sequence. The Journal of Physical Chemistry B,

108(10):3149–3157, 2004.

[26] Stephen J. Paddison and James A. Elliott. Molecular Modeling of the Short-Side-

Chain Perfluorosulfonic Acid Membrane. The Journal of Physical Chemistry A,

109:7583–7593, 2005.
133

[27] Stephen J. Paddison and James Elliott. Molecular Flexibility in the Short-Side-

Chain Perfluorosulfonic Acid Membrane. ECS Transactions, 1(6):207–214, 2006.

[28] Stephen J. Paddison and James A. Elliott. The effects of backbone conforma-

tion on hydration and proton transfer in the short-side-chain perfluorosulfonic acid

membrane. Solid State Ionics, 177:2385–2390, 2006.

[29] Stephen J. Paddison and James A. Elliott. On the consequences of side chain

flexibility and backbone conformation on hydration and proton dissociation in per-

fluorosulfonic acid membranes. Physical Chemistry Chemical Physics, 8:2193–2203,

2006.

[30] Stephen J. Paddison and James A. Elliott. Selective hydration of the short-side-

chain perfluorosulfonic acid membrane. An ONIOM study. Solid State Ionics,

178:561–567, 2007.

[31] Arun Venkatnathan, Ram Devanathan, and Michel Dupuis. Atomistic simulations

of hydrated Nafion and temperature effects on hydronium ion mobility. The Jour-

nal of Physical Chemistry B, 111(25):7234–7244, 2007.

[32] R. Devanathan, A. Venkatnathan, and M. Dupuis. Atomistic Simulation of Nafion

Membrane: I. Effect of Hydration on Membrane Nanostructure. The Journal of

Physical Chemistry B, 111(28):8069–8079, 2007.

[33] Ram Devanathan, Arun Venkatnathan, and Michel Dupuis. Atomistic simulation

of nafion membrane. 2. Dynamics of water molecules and hydronium ions. The

Journal of Physical Chemistry B, 111(45):13006–13013, 2007.

[34] Shengting Cui, Junwu Liu, Myvizhi Esai Selvan, Stephen J. Paddison, David J.

Keffer, and Brian J. Edwards. Comparison of the Hydration and Diffusion of Pro-
134

tons in Perfluorosulfonic Acid Membranes with Molecular Dynamics Simulations.

The Journal of Physical Chemistry B, 112(42):13273–13284, 2008.

[35] Matt K. Petersen, Feng Wang, Nick P. Blake, Horia Metiu, and Gregory A. Voth.

Excess Proton Solvation and Delocalization in a Hydrophilic Pocket of the Proton

Conducting Polymer Membrane Nafion. The Journal of Physical Chemistry B,

109:3727–3730, 2005.

[36] Matt K. Petersen and Gregory A. Voth. Characterization of the Solvation and

Transport of the Hydrated Proton in the Perfluorosulfonic Acid Membrane Nafion.

The Journal of Physical Chemistry B, 110:18594–18600, 2006.

[37] Craig K. Knox and Gregory A. Voth. Probing Selected Morphological Models of

Hydrated Nafion Using Large-Scale Molecular Dynamics Simulations. The Journal

of Physical Chemistry B, 114(9):3205–3218, 2010.

[38] Dongsheng Wu, Stephen J. Paddison, and James A. Elliott. Effect of Molecular

Weight on Hydrated Morphologies of the Short-Side-Chain Perfluorosulfonic Acid

Membrane. Macromolecules, 42(9):3358–3367, 2009.

[39] Daniel Brandell, Jaanus Karo, Anti Liivat, and John O. Thomas. Molecular dy-

namics studies of the Nafion®, Dow® and Aciplex® fuel-cell polymer membrane

systems. Journal of Molecular Modeling, 13:1039–1046, 2007.

[40] A. M. Lesk. Introduction to Protein Science Architecture, Function and Genomics.

”Oxford University Press”, 2004.

[41] Stephen L. Mayo, Barry D. Olafson, and William A. Goddard. DREIDING: a

generic force field for molecular simulations. The Journal of Physical Chemistry,

94:8897–8909, 1990.
135

[42] Christopher J. Cramer. Essentials of computational chemistry: theories and mod-

els. ”John Wiley & Sons”, 2004.

[43] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery. Numeri-

cal recipes in Fortran 77: the art of scientific computing. ”Cambridge University

Press”, 1992.

[44] Daan Frenkel and Berend Smit. Understanding Molecular Simulation: From Algo-

rithms to Applications. Academic Press, 2002.

[45] Xiongwu Wu and Bernard R. Brooks. Isotropic periodic sum: A method for

the calculation of long-range interactions. The Journal of Chemical Physics,

122(4):044107, 2005.

[46] Xiongwu Wu and Bernard R. Brooks. Using the isotropic periodic sum method to

calculate long-range interactions of heterogeneous systems. The Journal of Chem-

ical Physics, 129(15):154115, 2008.

[47] Kazuaki Takahashi, Kenji Yasuoka, and Tetsu Narumi. Cutoff radius effect of

isotropic periodic sum method for transport coefficients of Lennard-Jones liquid.

The Journal of Chemical Physics, 127(11):114511, 2007.

[48] M. P. Allen and D. J. Tildesley. Computer Simulation of Liquids. ”Oxford Univer-

sity Press”, 1990.

[49] Phil Attard. Thermodynamics and Statistical Mechanics: Equilibrium by Entropy

Maximisation. Academic Press, 2002.

[50] Manuel J. Louwerse and Evert Jan Baerends. Calculation of pressure in case of

periodic boundary conditions. Chem. Phys. Lett., 421(1-3):138–141, 2006.


136

[51] H.J.C. Berendsen. Simulating the physical world: hierarchical modeling from quan-

tum mechanics to fluid dynamics, chapter Pressure and virial. Cambridge Univ Pr,

2007.

[52] Glenn J. Martyna, Adam Hughes, and Mark E. Tuckerman. Molecular dynam-

ics algorithms for path integrals at constant pressure. The Journal of Chemical

Physics, 110(7):3275–3290, 1999.

[53] Karsten Meier and Stephan Kabelac. Pressure derivatives in the classical

molecular-dynamics ensemble. The Journal of Chemical Physics, 124(6):064104,

2006.

[54] Jhih-Wei Chu, Bernhardt L. Trout, and Bernard R. Brooks. A super-linear min-

imization scheme for the nudged elastic band method. The Journal of Chemical

Physics, 119(24):12708–12717, 2003.

[55] David J. Keffer, Chunggi Baig, Parag Adhangale, and Brian J. Edwards. A gen-

eralized Hamiltonian-based algorithm for rigorous equilibrium molecular dynamics

simulation in the canonical ensemble. Journal of Non-Newtonian Fluid Mechanics,

152(1-3):129–139, 2008.

[56] H. Flyvbjerg and H. G. Petersen. Error estimates on averages of correlated data.

The Journal of Chemical Physics, 91(1):461–466, 1989.

[57] Theodor Grotthuss. On the decomposition of water and their solution species using

galvanic electricity. Annales de Chimie, 58:54–74, 1806.

[58] I. Ruff and V. J. Friedrich. Transfer diffusion. IV. Numerical test of the correlation

between prototrope mobility and proton exchange rate of H3O+ and OH- ions with

water. The Journal of Physical Chemistry, 76(21):2954–2957, 1972.


137

[59] Wikipedia. Grotthuss mechanism, 2010. [Online; accessed 2-April-2010].

[60] Omer Markovitch, Hanning Chen, Sergei Izvekov, Francesco Paesani, Gregory A.

Voth, and Noam Agmon. Special pair dance and partner selection: Elementary

steps in proton transport in liquid water. The Journal of Physical Chemistry B,

112(31):9456–9466, 2008.

[61] Dominik Marx. Proton Transfer 200 Years after von Grotthuss: Insights from Ab

Initio Simulations. ChemPhysChem, 7(9):1848–1870, 2006. An addendum to this

article appears in Ref. [100].

[62] Dominik Marx, Mark E. Tuckerman, Jurg Hutter, and Michele Parrinello. The

nature of the hydrated excess proton in water. Nature, 397(6720):601–604, 1999.

[63] Edward F. Valeev and Henry F. Schaefer III. The protonated water dimer: Brueck-

ner methods remove the spurious C1 symmetry minimum. The Journal of Chemical

Physics, 108(17):7197–7201, 1998.

[64] The American Institute of Chemical Engineers. Study of Proton Transport Using

Reactive Molecular Dynamics, Computational Molecular Science and Engineering

Forum, 2009. See also http://utkstair.org/clausius/docs/atoms/RMD/index.html.

[65] Arieh Warshel and Robert M. Weiss. An empirical valence bond approach for

comparing reactions in solutions and in enzymes. Journal of the American Chemical

Society, 102(20):6218–6226, 1998.

[66] Udo W. Schmitt and Gregory A. Voth. Multistate Empirical Valence Bond

Model for Proton Transport in Water. The Journal of Physical Chemistry B,

102(29):5547–5551, 1999.
138

[67] Giuseppe Brancato and Mark E. Tuckerman. A polarizable multistate empirical

valence bond model for proton transport in aqueous solution. The Journal of

Chemical Physics, 122(22):224507, 2005.

[68] Feng Wang and Gregory A. Voth. A linear-scaling self-consistent generalization

of the multistate empirical valence bond method for multiple excess protons in

aqueous systems. The Journal of Chemical Physics, 122(14):144105, 2005.

[69] Diane E. Sagnella and Mark E. Tuckerman. An empirical valence bond model

for proton transfer in water. The Journal of Chemical Physics, 108(5):2073–2083,

1998.

[70] Udo W. Schmitt and Gregory A. Voth. The computer simulation of proton trans-

port in water. The Journal of Chemical Physics, 111(20):9361–9381, 1999.

[71] Yongho Kim, José C. Corchado, Jordi Villà, Jianhua Xing, and Donald G. Truhlar.

Multiconfiguration molecular mechanics algorithm for potential energy surfaces of

chemical reactions. The Journal of Chemical Physics, 112(6):2718–2735, 2000.

[72] Eikerling M., Paddison S.J., Pratt L.R., and Zawodzinski T.A. Defect structure

for proton transport in a triflic acid monohydrate solid. Chemical Physics Letters,

368(1):108–114, 2003.

[73] Pathumwadee Intharathep, Anan Tongraar, and Kritsana Sagarik. Ab initio

QM/MM dynamics of H3 O+ in water. Journal of Computational Chemistry,

27(14):1723–1732, 2006.

[74] Solvey Solexis. Hyflon® Ion E83 Membranes Data Sheet, 2006. [Online; accessed

9-March-2010].
139

[75] H. L. Morgan. The Generation of a Unique Machine Description for Chemical

Structures-A Technique Developed at Chemical Abstracts Service. Journal of

Chemical Documentation, 5:107–113, 1965.

[76] David Weininger. SMILES, a chemical language and information system. 1. Intro-

duction to methodology and encoding rules. Journal of Chemical Information and

Computer Sciences, 28:31–36, 1988.

[77] CADD Group Chemoinformatics Tools and User Services. Chemical Identifier Re-

solver. [Online; accessed 9-March-2010].

[78] CADD Group Chemoinformatics Tools and User Services. Online SMILES Trans-

lator and Structure File Generator. [Online; accessed 9-March-2010].

[79] N. H. Weiderman and B. M. Rawson. Flowcharting loops without cycles. SIG-

PLAN Not., 10(4):37–46, 1975.

[80] Open Babel: The Open Source Chemistry Toolbox. [Online; accessed 9-March-

2010].

[81] E. Spohr, P. Commer, and A. A. Kornyshev. Enhancing Proton Mobility in Poly-

mer Electrolyte Membranes: Lessons from Molecular Dynamics Simulations. The

Journal of Physical Chemistry B, 106:10560–10569, 2002.

[82] Yushu Matsushita. Precise Molecular Design of Complex Polymers and Morphology

Control of Their Hierarchical Multiphase Structures. Polymer Journal, 40(3):177–

183, 2008.

[83] Reginald Paul and Stephen J. Paddison. Effects of dielectric saturation and ionic

screening on the proton self-diffusion coefficients in perfluorosulfonic acid mem-

branes. The Journal of Chemical Physics, 123(22):224704, 2005.


140

[84] E. Spohr. Molecular Dynamics Simulations of Proton Transfer in a Model Nafion

Pore. Molecular Simulation, 30(2/3):107–115, 2004.

[85] P. Commer, C. Hartnig, D. Seeliger, and E. Spohr. Modeling of Proton Transfer in

Polymer Electrolyte Membranes on Different Time and Length Scales. Molecular

Simulation, 30(11/12):755–763, 2004.

[86] J. M. Haile. Molecular Dynamics Simulation: Elementary Methods. ”John Wiley

& Sons”, 1997.

[87] David J. Keffer, Chunggi Baig, Parag Adhangale, and Brian J. Edwards. A gen-

eralized Hamiltonian-based algorithm for rigorous equilibrium molecular dynamics

simulation in the isobaric-isothermal ensemble. Molecular Simulation, 32(5):345–

356, 2006.

[88] David J. Keffer, Carrie Y. Gao, and Brian J. Edwards. On the Relationship be-

tween Fickian Diffusivities at the Continuum and Molecular Levels. The Journal

of Physical Chemistry B, 109(11):5279–5288, 2005.

[89] Michael Levitt, Miriam Hirshberg, Ruth Sharon, Keith E. Laidig, and Valerie

Daggett. Calibration and Testing of a Water Model for Simulation of the Molecu-

lar Dynamics of Proteins and Nucleic Acids in Solution. The Journal of Physical

Chemistry B, 101(25):5051–5061, 1997.

[90] K.D. Kreuer, M. Schuster, B. Obliers, O. Diat, U. Traub, A. Fuchs, U. Klock, S.J.

Paddison, and J. Maier. Short-side-chain proton conducting perfluorosulfonic acid

ionomers: Why they perform better in PEM fuel cells. Journal of Power Sources,

178(2):499 – 509, 2008.

[91] Robin L. Hayes, Stephen J. Paddison, and Mark E. Tuckerman. Proton Trans-

port in Triflic Acid Hydrates Studied via Path Integral Car-Parrinello Molecular
141

Dynamics. The Journal of Physical Chemistry B, 113(52):16574 – 16589, 2009.

[92] Sergei Izvekov and Gregory A. Voth. Ab initio molecular-dynamics simulation of

aqueous proton solvation and transport revisited. The Journal of Chemical Physics,

123(4):044505, 2005.

[93] H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, A. DiNola, and J. R.

Haak. Molecular dynamics with coupling to an external bath. The Journal of

Chemical Physics, 81(8):3684–3690, 1984.

[94] In-Chul Yeh and Gerhard Hummer. System-Size Dependence of Diffusion Coeffi-

cients and Viscosities from Molecular Dynamics Simulations with Periodic Bound-

ary Conditions. The Journal of Physical Chemistry B, 108(40):15873–15879, 2004.

[95] Biswajit Santra, Angelos Michaelides, and Matthias Scheffler. On the accuracy

of density-functional theory exchange-correlation functionals for H bonds in small

water clusters: Benchmarks approaching the complete basis set limit. The Journal

of Chemical Physics, 127(18):184104, 2007.

[96] Eric W. Weisstein. “Circle-Circle Intersection.” From MathWorld – A Wolfram

Web Resource. [Online; accessed 9-March-2010].

[97] S. R. Bahn and K. W. Jacobsen. An object-oriented scripting interface to a

legacy electronic structure code. COMPUTING IN SCIENCE & ENGINEERING,

4(3):56–66, 2002.

[98] J. J. Mortensen, L. B. Hansen, and K. W. Jacobsen. Real-space grid imple-

mentation of the projector augmented wave method. PHYSICAL REVIEW B,

71(3):035109, 2005.

[99] Wikipedia. Particle swarm optimization, 2010. [Online; accessed 9-March-2010].


142

[100] Dominik Marx. Proton Transfer 200 Years after von Grotthuss: Insights from Ab

Initio Simulations. ChemPhysChem, 8(2):209–210, 2007.

Anda mungkin juga menyukai