Molecular Modeling in
Chemical Industry R&D
Brian Peterson May 7, 2004
Outline
 What Molecular Modeling is and is not
 Examples
 How MM relates to Chemical Engineering
 Thoughts on Curriculum
2
(Chemical) Industry Drivers
 Societal Needs & Wants
 Potential to make or lose $
 Competition- increasing and global
 Efficient Management of
– Time
– Capital
– Knowledge
– Creativity
 Compression of R&D timescales
 Rapid & Accurate Evaluation of Complete Solutions
– Rapid & Parallel Development (e.g. Materials,
Process, Environmental, & Economic)
– Go/NoGo decisions, “Stage Gate”
– Modeling & Optimization
3
Hierarchy of Models
Time
year
Quantitative Structure/Property
Relationships (QSPR) &
Theory
min.
Macroscopic
Plant
Process
Fab
ms
Continuum
ns
Mesoscale
10-12 s
CFD
Mechanical
Kinetic
Emag
Molecular
Molecular Dynamics
Monte Carlo
10-15 s
Quantum
1A
4
10A
100A
1mm
1cm
Distance
km
Size
Easy
Structure
Energy
Molecules
Small
Organic
Equilibrium
Gases
Medium
Inorganic
Fast (t < ns)
Polymers
Binding Energy
IR Spectra
Transition States
Crystal Defects
Amorphous
Solids
Dipole Moments
Polarizability
Perfect Crystals
Liquids
Enthalpy
Large
Hybrid
Intermediate
Activation Energy
NMR Spectra
Elastic Modulus
uv Spectra
5
Hard
Free Energy
Enthalpy of Formation via QM
Molecule
ethylene
cyclopropane
bicyclobutane
ammonia
phosgene Cl2C=O
methyl fluoride
aminotriazole
tetraHpyrimidine
acrylonitrile
G2 Theory
Literature
(kcal/mol)
(kcal/mol)
12.9
13.8
55.2
-10.8
-54.9
-58.3
45.2
18.9
46.2
12.5
12.7
51.9
-11.0
-51.9, -52.4
-59
47.6
13.2
43.2
D. Frurip et al., ACS SS 677, 1998
•Heats of reaction can be more accurate than heats of formation
•Can use QM to calculate group contributions for new groups
•Outliers are not always predictable a priori
•Many outliers are the result of experimental problems
6
Example: Molecular Sieving
Desired Product + Impurities
Difficult Separation
Calculate Characteristic Sizes
Calculate Polarizability,
Dipole, Quadrupole Moments
no
Dprod – Dimp > ~ 0.3 A ?
yes
“Easy” Separation
Find Zeolite such that
Dprod > Dzeo > Dimp
7
Significant Difference ?
Detailed
Modeling
no
“Real Difficult”
Separation
yes
Find Appropriate Material
Experiment
Other
Separation
Methods
Grand Canonical Monte Carlo
A classical force-field method where molecules interact and are ...
MOVED
INSERTED
X
DESTROYED
... such that the proper distribution of positions, energies, and
numbers of molecules is achieved for a system at fixed chemical
potential (or fugacity or pressure) and temperature. GCMC is often
used to study the adsorption of small molecules in inorganic materials
such as zeolites.
m,V,T
8
N,E
Molecular Size via Computational Sieving
Use a Force-Field and Grand Canonical Monte Carlo to adsorb
a gas molecule into a confined system at a standard T & P. If
significant numbers of the molecule fit into the system, the
characteristic size of the system is related to the size of the molecule.
An unbiased method which uses the information inherent in the FF.
Dslit
Number Adsorbed
20
16
Nthreshold
12
8
4
0
5.6
5.7
5.8
5.9
Dslit (A)
Dtube
D = Dslit - d
9
6
6.1
6.2
10
I8
Pr I
od 9
uc
t
I6
I7
Zeolites
I4
I5
9
I2
I3
Na
X
Ca
ni X
te
LP
Be
Ch C ta
ab aA
a
Er zit
io e
M
ni
or
t
de N e
a
n
K ite A
Er S
io P
ni
te
Li
A
KA
ZS B
M
ZS -5 eta
ZS M-5 min
M
ZS -1 ma
M 1mx
ZS -11 in
M
ZS -2 ma
M 3 x
M -2 mi
C 3 n
M M-2 ma
C
x
Fe M- 2 m
2
i
r
n
Fe rie 2 m
r
rr ite ax
ie
rit mi
e n
m
ax
I1
de
or
M
Size (A)
Zeolite Pore Diameter + Molecular Sizes
Molecules
8
7
6
5
4
3
2
1
0
Engineer had already tried 5A and did not get a good separation.
After seeing this analysis, they repeated the experiment, found good
separation, and commercialized the process.
Molecular Size via QM Density Contours
1. Geometry optimize molecule
2. Calculate Electron Density Contours
3. Measure molecular width on computer screen
11
Example: Hydrogen Storage
ab initio MD of H2 in SWNT
 SWNT highly fluxional; large
C-C-C bond angle
deformations are observed.
 Adsorption energies much
higher than previously
published calculations using
classical simulation methods:
enhanced potential from
curved carbon surface.
 Improved, curvaturedependent potentials were
created.
H. Cheng, G. P. Pez, A. C. Cooper,
J. Am. Chem. Soc. 123, 5845 (2001).
DHExpt (kcal/mol) DHSim (kcal/mol)
(95% swnt, 7-14Å) (7.8Å, 11.8Å)
4 – 4.8
4.8, 3.3
M. K. Kostov, H. Cheng, A. C. Cooper, G. P. Pez
Phys. Rev. Lett. 89, 6105 (2002)
12
Xenon binding to Proteins
•Xenon/protein interactions are important...
•Xenon is an anesthetic
•Xenon is a neuroprotectant
•Xenon is used to prepare “heavy atom” derivatives for X-ray diffraction
• 129Xe is used in NMR studies of cavities
•Predictive methods for binding of xenon would enable better
understanding of ...
•the mechanisms of physiological activity
•the behavior of xenon in NMR and XRD experiments
•binding sites not visible by XRD (due to resolution, occupancy, disorder)
•The goal of this work was to show how grand canonical Monte Carlo
Simulations (GCMC) coupled with a clustering algorithm can determine the
positions, occupancy, and free energies of binding of small molecules.
13
Mass Clouds and Clusters in COMP
Xenon (blue) & Water (red)
Minimum Extent DRmin DRavg
Energy
(Å)
(Å)
(Å)
(kcal/mol)
*
2
0.97
-7.30
2.3
0.20
0.26
*
3
0.98
-6.89
3.7
0.71
0.39
*
26
0.94
-5.91
3.8
0.54
0.29
*
11
1.00
-6.30
3.9
0.31
0.31
*
15
0.92
-6.14
3.9
0.26
0.43
*
10
0.96
-6.37
4.0
0.40
0.24
*
24
0.94
-5.93
4.3
1.91
0.38
4
0.97
-6.68
4.5
6
0.98
-6.56
5.4
8
0.91
-6.46
5.5
1
1.03
-7.31
5.8
*
45
1.03
-5.63
5.9
1.50
0.67
Clusters with Occupancy > 0.9 sorted by extent. Clusters
assigned with Rmax = 3Å, Rtest = 1Å, Etest = 0 kcal/mole. fXe = 10
bar. * Cluster corresponds closely to a site found in experiment.
Cluster
Number
Blue = Simulation, Black = X-Ray Diffraction
Occupancy
•GCMC + Force Field Reproduces
Experimental Binding Locations
•Occupancy + Input Fugacity 
Equilibrium Constant  Free Energy
14
Thinking Like a Chemical Engineer
Processes
Process Analysis Optimization Control
Unit Operations
time
Petroleum
Surface Science
Biological Systems
Semiconductors
Production
Specific Problem
Energy
Food
Pharmaceuticals
Materials Design Environment
Polymers
Transport
Thermodynamics
Kinetics
Quantum, Classical & Statistical Mechanics
Particles
15
application
Summary: Mol. Modeling & ChE
 Molecular Modeling (Computational Chemistry,
Computational Materials Science, “Theory”, “Modeling”)
is the natural extension and limit of the reductionist
approach to chemical engineering.
 MM is broadly applicable because “everything” is made
of atoms and molecules.
 The power of MM is growing rapidly with the continuing
development of computer power, new algorithms, and
the availability of software.
 Today MM can sometimes provide useful estimates of
the properties and behavior of materials- even before
they have been synthesized. (materials design)
 Today MM can sometimes provide useful estimates of
the parameters and behavior needed to do traditional
chemical engineering process development & design.
 Today MM is sometimes the most efficient way to obtain
these estimates.
 MM works best in partnership with experiments and with
traditional estimation and design approaches.
16
Molecular Modeling and the UG
Chemical Engineering Curriculum
 Minimum: UG Chemical Engineers should be aware of the
possibility that useful estimates of some material
properties can be calculated for some systems and that
the number of such properties and systems is continually
increasing.
 Optimum (?) : UG Chemical Engineers should have some
familiarity with the techniques of MM and should be able
to make informed guesses as to whether any given
properties and materials are amenable to MM.
 “Win/Win” MM techniques are wonderful pedagogical
tools for understanding the fundamental physical
processes which underlie thermodynamics, transport
phenomena, and chemical kinetics. Much of the requisite
familiarity could be obtained via the permeation,
throughout the undergraduate curriculum, of computation
and simulation as methods of understanding
complementary to experiment and theory.
17
Thank you
A black box is not a silver bullet
 Molecular Modeling will not replace all other scientists and
engineers. Among other reasons, the techniques of
computational chemistry and molecular modeling employ
approximations. The validity of these approximations
varies with the method and with the system considered.
Therefore, one cannot blindly apply a given method to all
systems and rationally expect useful answers.
 Molecular Modeling can and does replace some unnecesary
experimentation and it can lead to insights which initiate
new experiments. Some approximations are quite valid for
some systems and one can expect useful results when a
suitable method is used to predict some subset of
properties for those systems.
 A given method will typically supply only some of the
properties and information needed to solve a given
problem. MM techniques are most useful when used in
combination with each other and with experiment.
19
Binding Equilibrium and Free Energy
Occupancy + Input Fugacity  Equilibrium Constant  Free Energy
Source
a
b
a
c
Equilibrium Constants
T (K)
273
273
298
300
K1
5.0
K2
8.4
K12
1.7
2.0
0.31
1.3
0.56d
0.65
Binding Free Energies
(kcal/mol)
DG1
DG2
DG12
-0.87
-1.15
-0.28
-0.41
0.70
-0.17
0.35
0.24
0.80e
-0.35f
Cavity
Occupancy
@ 8 bar
1.95
1.4
1.82
1.88g
a. GCMC @ 8 bar xenon in T4 lysozyme. Ki from slopes pi/po vs f/Po. b. Quillen et al. XRD.
c. Mann & Hermans MD simulations.
d. Corrected.
e. DG12 from independent simulation.
f. DG12 calculated self-consistently from DG1 and DG2.
g. Recalculated from correct K1, K2.
20