Beruflich Dokumente
Kultur Dokumente
Molecular Simulation
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/gmos20
To cite this article: Reinier L.C. Akkermans, Neil A. Spenley & Struan H. Robertson (2013) Monte Carlo methods in Materials
Studio, Molecular Simulation, 39:14-15, 1153-1164, DOI: 10.1080/08927022.2013.843775
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Molecular Simulation, 2013
Vol. 39, Nos. 14 – 15, 1153–1164, http://dx.doi.org/10.1080/08927022.2013.843775
We survey the use of the Monte Carlo method within the Materials Studio application, which integrates a large number of
modules for molecular simulation. Several of these modules work by generating configurations of a system at random,
which can then be used to calculate averages of interest – for instance, interaction energies of contacting pairs of molecules
(Blends module) and properties of a flexible polymer chain (Conformers). A different technique is used to sample an
Downloaded by [Moskow State Univ Bibliote] at 12:26 14 November 2013
appropriate physical distribution (which in practice is that for the canonical ensemble) using the Metropolis or
configurational bias method. This is done by the Sorption module (which calculates the thermodynamic properties of small
molecules in a matrix) and Amorphous Cell (which constructs periodic simulation cells). Lastly, certain other modules use
simulated annealing and related methods to optimise a function, with application to crystal structure prediction from
molecular structure (Polymorph Predictor), to crystal structure prediction from X-ray powder diffraction data (Powder
Solve) and to find preferential sites for adsorption (Adsorption Locator).
Keywords: Monte Carlo; Materials Studio; simulated annealing; Sorption; morphology
systematic scan of all the angles. However, this can be weight, rðGÞ ¼ 0. Straightforward MC sampling is then
expensive because if each angle range is sampled on m very inefficient since the majority of random configur-
points this will generate m N configurations, a potentially ations will have negligible weight, and not contribute to
huge number of configurations which would require further the average. It would be much more efficient if it were
analysis to extract quantities of interest. Of more interest possible to draw independent configurations straight from
are the properties of typical conformations, and the the distribution rðGÞ (akin to the Box – Muller transform
generation of these configurations can often be effected for normally distributed random variables), but this is
more economically by simply generating sets of random difficult in general. Fortunately, non-uniform sampling
angles and using these to generate structures. can be realised straightforwardly using Markov chains at
With G a point in configuration space, the average of the expense of introducing correlation between the
property A can be written as an integral over the same space, samples. This is discussed in the next section.
ð
kAl ¼ dG AðGÞrðGÞ; ð1Þ
2.2 MC as importance sampling method
The MC method used as importance sampling aims to
where rðGÞ is the probability density of the configuration G,
Downloaded by [Moskow State Univ Bibliote] at 12:26 14 November 2013
temperature. Kinetic energy is not included in this definition, density of states visited to be concentrated on the lower
since integrating momentum space can be carried out energy regions, so these are now sampled thoroughly.
analytically and sampling is not required. The acceptance The simulated annealing method is used in the
probability follows from Equation (6) as follows: modules Adsorption Locator (Section 3.4) and Polymorph
( Predictor (Section 3.7) of Materials Studio. The technique
expð2bðE0 2 EÞÞ; E0 . E;
0 is quite generic, and can be used to obtain the minimum of
TðG; G Þ ¼ ð8Þ
1; E0 # E; any function. It is often used for functions that have
multiple minima and/or for which the phase space is
where E0 ¼ EðG0 Þ. simply too large to sample systematically. In the Powder
Changing a state into another is usually implemented as Solve module (Section 3.6) of Materials Studio, this is
a two-stage process. First, a trial step is proposed, which is exploited by defining a figure of merit; the negative of this
then accepted or rejected. Let vðG; G0 Þ be the attempt corresponds to the energy.
probability and aðG; G0 Þ be the acceptance probability, then
In the traditional MC method, trials are made with equal 3.1 Conformers
probability in either direction, vðG; G0 Þ ¼ vðG0 ; GÞ. For The Conformers module in Materials Studio can be used to
example, as many attempts are made to change the x explore the conformer space of molecules, which exhibit
coordinate of an atom from 5 to 5.5 Å, as moving it from 5.5 internal rotation about a bond, in particular polymers. The
to 5 Å. The acceptance probability thus reads aðG; G0 Þ ¼ preparation for a conformer search requires the torsion
minð1; rðG0 Þ=rðGÞÞ. This approach is implemented in the degrees of freedom that will be the basis of the search to be
Sorption module in Materials Studio as the Metropolis MC specified. These can be defined in the visualiser using
method (Section 3.3). Torsion Twister objects as illustrated in Figure 1. Con-
It is sometimes useful to bias the attempts to achieve a formers offer automated tools to find rotatable bonds in a
higher acceptance rate. In that case, vðG; G0 Þ – vðG0 ; GÞ.
polymer, which can be modified manually when needed.
Typically, more attempts are made to step to configur-
The module offers three algorithms with which to
ations of higher density (lower energy) to avoid
explore configuration space: systematic grid scan, random
unnecessary evaluations of high energy. By tracking the
sampling and Boltzmann jump. The systematic grid scan
bias v, the original ensemble can still be obtained by
explores the configuration space on a regular grid. As
accepting with the modified probability aðG; G0 Þ ¼
indicated in Section 2.1, generating configurations on a
minð1; ðvðG0 ; GÞ=vðG; G0 ÞÞðrðG0 Þ=rðGÞÞÞ. This approach is
implemented as the configurational bias MC method in the regular grid can be expensive, mainly because of the large
Sorption module of Materials Studio (Section 3.3.1). A number of configurations generated. In order to reduce the
similar biased insertion method (but without evolution) is number, various parameters can be set to restrict the search.
used in Amorphous Cell (Section 3.5).
For example, certain configurations can be eliminated on based on an off-lattice generalisation of the Flory–
the grounds that there is significant overlap of atomic radii Huggins theory.[8,9]
– such configurations are likely to have a very high energy The main calculation in Blends consists of estimating
and are, therefore, very unlikely to occur. The radii used the average energy between two molecules in contact at a
are specified as a percentage of van der Waals radii. given temperature. Two molecules are said to be in contact
The random sampling search generates configurations if their van der Waals surfaces touch (are tangent). The
by altering torsion angles by a random amount. The sampling is a two-stage process. First, a large number of
amount by which a torsion angle is altered, du, is generated random pair configurations are drawn. In each configur-
from a uniform distribution of width w, i.e. from the range ation, two atoms are required to be in contact, but
of 2w=2 # du # þw=2, the maximum value of w being otherwise no temperature is imposed on the MC sampling.
1808. The number of configurations generated is a In this ensemble, the energy distribution (or ‘density of
specified parameter. states’) is evaluated, based on the user-specified force
The Boltzmann jump search also generates configur- field. The histogram is then reweighted to obtain the
ations through random sampling, the main difference distribution at a given temperature, from which finally the
being that it is the Boltzmann distribution from which average energies are obtained.
Downloaded by [Moskow State Univ Bibliote] at 12:26 14 November 2013
samples are drawn. The sampling is generated as a Markov The binding energy calculation is repeated for all
chain using the Metropolis algorithm as described in unique combinations of base and screen molecules. A
Section 2.2, and as a consequence, one of the input cluster size calculation for all four combinations is also
parameters for this sampling is temperature. The specified made to estimate the average number of interactions in the
temperature will determine the width of the distribution – condensed state. The cluster size is analogous to the
the higher the temperature the broader the width. coordination number in lattice theories, but can vary
depending on the combination of species. A cluster
consists of a molecule of one species, surrounded by as
3.1.1 Applications of Conformers many molecules of another species as will fit within the
A polysaccharide molecule (pullulan) was studied by Liu contact condition. This means that the van der Waals
et al. [5]. They used molecular mechanics methods, surface of the central molecule is in contact with the
including a conformer search based on the torsion angles surfaces of all the coordinating molecules, approximating
of the glycosidic linkage, to construct rotational isomeric the first solvation shell of the molecule in solution. From
state models. Peters [6] examined tricyclic terpanes, and the binding energies and cluster sizes, the mixing energy
did a conformer search calculation to identify low energy can be obtained, which leads to the Flory– Huggins
structures, which were then optimised using a semi- interaction parameter, x.
empirical method. Surin et al. [7] used conformer search to By retaining the density of state data, the calculation of
find the stable configuration of a photoluminescent the x-parameters can be repeated at different temperatures
fluorene-based monomer, followed by energy minimis- without additional MC sampling. This temperature-
ation calculations on a stack of oligomers. dependent interaction parameter can be used in the
Flory – Huggins theory of miscibility to predict non-ideal
phase diagrams.
3.2 Blends
The Blends module in Materials Studio is designed to
estimate the miscibility behaviour of binary mixtures, such 3.2.1 Applications of Blends
as solvent – solvent, polymer – solvent and polymer – Blends is often used to screen a number of molecules for
polymer mixtures. Blends predicts the thermodynamics binding capacity on one or more base materials. Another
of mixing directly from the chemical structures of the two application of Blends is aimed at calculating x-parameters
components and, therefore, requires only their molecular to be used in the parameterisation of mesoscale models
structures and a force field as input. such as dissipative particle dynamics and dynamic density
Blends provides three tasks to the user for calculating functional theory.
binding energy, cluster size and mixing energy. The output An example of the latter is found in the mesoscale
consists of a study table (the spreadsheet document in simulation studies by Wu et al. [10] on predicting the
Materials Studio) containing the average energies of morphology of perfluorosulphonic acid fuel cell mem-
molecular pairs at the specified temperature, the cluster branes. The input parameters in this case were determined
sizes and mixing energies. The study table can be further from Blends calculations in conjunction with the COM-
analysed to obtain mixing energies at other temperatures, PASS force field. A similar workflow was used by Lin
as well as abstract thermodynamic data such as free energy et al. [11] to generate input parameters for a mesoscale
isotherms, the binodal and spinodal, and critical point, simulation of unilamellar vesicles.
Molecular Simulation 1157
Pajula et al. [12] exploited Blends as a screening tool In the grand canonical ensemble, the system has
to find potential stabilisers needed to produce a stable additional degrees of freedom N ¼ ðN 1 ; . . . ; N M Þ, speci-
amorphous binary mixture. They applied the module to a fying the loading of each component. The density is
test set of 39 drug molecules, and found a good correlation
between the predicted Flory – Huggins parameter and the rðG; NÞ ¼ CFðNÞexpð2bEðGÞÞ; ð10Þ
crystallisation tendency as measured experimentally with
differential scanning calorimetry and polarised light where E is the potential energy and FðNÞ is the product of
microscopy. Ren et al. [13] have used Blends to
investigate the solubility of a series of cyclohexanone ðbf i VÞN i
FðN i Þ ¼ exp 2bN i mintra;i ; ð11Þ
formaldehyde resins in a variety of solvents (tetrachlor- Ni!
omethane, alcohol, toluene and ethyl ether).
Modifications to the Blends algorithm to include where f i is the fugacity of component i and V is the
entropy of mixing and other correlation effects were constant volume of the system. The term mintra;i is the
discussed by Akkermans [9]. Here, it was shown that the intramolecular chemical potential. The latter is defined as
inclusion of excess entropy term in the Flory– Huggins 2kB T lnkexpð2buintra;i Þl, where uintra;i is the intramole-
Downloaded by [Moskow State Univ Bibliote] at 12:26 14 November 2013
parameter can provide a better description of the critical cular potential of a molecule of component i, obtained at
point of the nitrobenzene –hexane mixture, as well as constant temperature in vacuum. Without torsional degrees
predict more accurately the interaction parameters for of freedom this energy is constant, in which case it can be
polymer blends. subtracted from the total energy E in Equation (10),
leaving the intermolecular energy.
Substituting the ensemble density in the acceptance
3.3 Sorption rule, Equation (6) gives
The Sorption module in Materials Studio allows
0 0
F 0
simulation of a framework – sorbate system using MC a G; N; G ; N ¼ min 1; 0 expð2bðE 2 EÞÞ : ð12Þ
F
sampling. The input consists of a framework structure,
typically a unit cell (or super cell) of a microporous
The Fixed Loading task supports the following step
crystal, and the structures of one or more sorbate
types: Translation, Rotation, Regrowth and Conformer. All
molecules. The framework is always considered rigid, moves are applied to a random sorbate molecule of a
thus providing a static external field to the sorbate random component in the system. The relative probability
molecules. Sorbate molecules can also be treated as rigid of each move can be specified in the input.
bodies, in which case the degrees of freedom of the system The Translation move corresponds to moving the
are specified by the centre-of-mass position (R i ) and centre-of-mass of the selected sorbate molecule over a
orientation (V i ) of the molecules, G ¼ ðR 1 ; V 1 ; . . . ; R N ; distance dr along axis A. The distance is drawn from a
V N Þ. It is also possible to specify intramolecular torsional uniform distribution between 0 and Dt . Axis A is the vector
degrees of freedom, as described in more detail below. from a random point on a sphere to the origin of the sphere.
Bond lengths and valence angles are always held fixed. Care is taken to achieve a uniform distribution on the
Two sampling methods are supported: the Metropolis sphere surface. The maximum displacement Dt can be
and the configurational bias MC methods. Each method can specified in the input, or selected automatically by the
be used to sample two ensembles: the Fixed Loading task algorithm to achieve an acceptance rate of 50%.
samples the canonical ensemble, whereas the Fixed The Rotation step rotates the molecule about the
Pressure task samples the grand canonical ensemble. The centre-of-mass, by an angle du about axis A. The angle is
former task is described by the loading of each component drawn from a uniform distribution between 2Dr and þDr .
N ¼ ðN 1 ; . . . ; N M Þ, whereas the latter requires the Axis A is again the vector from a random point on a sphere
fugacities f ¼ ðf 1 ; . . . ; f M Þ of each component. The to the origin of the sphere. As for Translation, the
fugacity is related to the chemical potential m by maximum displacement can be fixed in the input, or
f ¼ f + expðbðm 2 m + ÞÞ, where + denotes a reference state, optimised by the algorithm to achieve an acceptance rate
for instance the ideal gas. The fugacity reduces to the of 50%.
partial pressure for an ideal gas. In the Regrowth move, a sorbate molecule is removed
The density of the canonical ensemble is given by from the system, and reintroduced with a random position
Equation (7). All terms that are independent of G, such as and orientation. Such moves can be used to transfer
the energy of the fixed framework and the intramolecular sorbate molecules between pores in the framework.
energy of the sorbate molecules (when treated as rigid), The Conformer move is used when the input of a
may be excluded and incorporated in a constant sorbate consists of multiple conformations for a collection
multiplication factor. of trans and gauche conformations. The move will attempt
1158 R.L.C. Akkermans et al.
to replace the current conformation by another randomly To determine the acceptance probability, the attempt
chosen from the trajectory. rate of the reverse step, vðG0 n ; GÞ, is also required. This is
The Fixed Pressure task supports, in addition, the the attempt rate for transforming a configuration G0 n into
Exchange step type, comprising creation of a new sorbate the known configuration G. It is calculated as before, with
and deletion of an existing molecule. Care is taken to the selected configuration G being given; hence, only K 2
always attempt creation and deletion with equal 1 trial configurations have to be generated to determine
probability, even if the system is empty. If the system is this attempt rate.
empty, deletion steps are simply rejected, adding the The above procedure applies to one degree of freedom.
empty state as an additional sample to the ensemble The configurational bias method is typically applied to
average. This can be important when calculating the chain molecules with multiple rotational degrees of
adsorption isotherm at low fugacities. freedom. In this case, the procedure is simply repeated
The output of Sorption consists of reported values for for each degree of freedom. For each degree of freedom, K
the average loading, isosteric heat per component, as well states are generated and their weights are determined. One
as average total energy of the system. The distribution of state is drawn from the weighted distribution, and the
the insertion energy can be obtained as a histogram and as attempt rate for this selection is calculated as above. After
Downloaded by [Moskow State Univ Bibliote] at 12:26 14 November 2013
a three-dimensional (3D) field. A density field may also be all degrees of freedom have been assigned a value, the
obtained, as illustrated in Figure 2. resulting configuration is the candidate for the acceptance
step. The total attempt rate of this configuration is the
product of the attempt rates for each degree of freedom.
3.3.1 Configuration bias MC The attempt rate of the reverse step is calculated in the
In the configurational bias MC method of Sorption, the same way.
attempt rates vðG; G0 Þ are constructed as follows. Instead In Materials Studio, rotational degrees of freedom are
of generating one trial configuration G0 , a total of K trial defined, and visualised, using torsion monitors, as
configurations G0 1 ; . . . ; G0 K are generated. Each configur- illustrated in Figure 1. Torsion monitors can be defined
ation G0 k is given a weight wk . One configuration is then manually on the model, or automatically, taking into
selected with a probability proportional to its weight. The account constraints due to rings, double bonds or terminal
attempt rate of a configuration n is thus given by hydrogens. The Sorption module processes the torsion
monitors on the input structure, building up a graph of
wn atom groups, or segments, mutually connected by
v G; G0 n ¼ PK : ð13Þ
1=K k¼1 wk rotatable bonds. The segments are treated as rigid bodies,
equivalent to Motion Groups in Materials Studio. As such,
Consequently, configurations with higher than average the state of a chain molecule is completely defined by
weight are more likely to be attempted, that is v . 1. Of the centre-of-mass position and orientation of the first
course, it is still possible, but less likely, to select a trial (head) segment, and the values of the torsions,
configuration with a lower than average weight, v , 1. G ¼ ðR; V; f1 ; :::; fN Þ. By assigning fewer, or more,
torsions, the size of the segments can be controlled.
To define the weights used in the selection of trial
configurations, Sorption calculates the segment energy V i ,
defined as the sum of the (non-electrostatic) intermole-
cular energy between segment i and the framework and all
other sorbate molecules,
V i ¼ V Fi þ V Si : ð14Þ
values are drawn, and the weights Sorption has also been applied to metal – organic
frameworks, focusing typically on adsorption and storage
wT;t ¼ expð2bðV tþ1 þ V tþ2 þ uintra ÞÞ ð16Þ of small molecules, such as carbon dioxide and methane.
Greathouse et al. [19] also considered more complicated
are evaluated for each resulting configuration. In this case, organic molecules relevant to chemical sensing and
the intermolecular energy of the two segments succeeding
detection, such as aromatics, explosives and certain
segment t is included. If a torsion has only one successive
chemical warfare agents. They calculated the isosteric
segment, the term V tþ2 is excluded.
heats of these materials on zinc, chromium and copper
The step types in the configurational bias method are
organic frameworks, and found that the chromium-
similar to those described above with Translation and
containing framework CrMIL-53lp had the highest
Rotation applying to the head segment. In addition, a Twist
step type is available, which will change the angle of a adsorption energy for all analytes, suggesting that this
randomly chosen torsion by an amount df, chosen material may be suitable for detection of low-level
uniformly between 2Dr and þDr . organics. Their results demonstrated that Sorption can be
used as an efficient first step in the screening of metal –
organic frameworks for detection of large molecules.
Downloaded by [Moskow State Univ Bibliote] at 12:26 14 November 2013
3.5.1 Applications of Amorphous Cell addition to the diffraction pattern, are the atoms in the unit
Amorphous Cell is typically used as the start of a cell along with the space group and the cell parameters.
modelling study to generate the input structures, to be (Materials Studio has other tools to allow these to be
further simulated using molecular dynamics. estimated from the diffraction pattern.) A crystal structure
Liu et al. [25] studied the spreading of water is defined by the positions of the atoms (or rigid groups of
nanodroplets on amorphous cellulose and polypropylene atoms) in the unit cell. The diffraction pattern of a
surfaces. They used the Confined Layer option in candidate structure is calculated and compared with the
Amorphous Cell to build initial structures of the polymers, experimental data, and the space of possible structures is
and further compacted and optimised those using a layered searched for good matches.
structure with xenon crystals. This procedure gave a degree The basic measure of similarity (figure of merit)
of molecular surface roughness that corresponds well to between a simulated and an experimental diffraction pattern
experimental values. In addition, calculated properties such is the weighted profile R-factor Rwp, a weighted sum of
as density, cohesive energy density, coefficient of thermal squares of the difference between the intensities of the two
expansion and the surface energy agreed with experimental patterns. In addition to this, it is possible to make use of
values. It was found that a water nanodroplet spreads on the some chemical information. Under some circumstances, for
Downloaded by [Moskow State Univ Bibliote] at 12:26 14 November 2013
amorphous cellulose surfaces, but there was no significant instance if the quality of the experimental data is poor, or if
change in the dimension of the droplet on the polypropylene the structure is of high symmetry and has few peaks, then
surface; the resulting water contact angles on polypropylene the Powder Solve procedure is likely to identify a number of
and amorphous cellulose surfaces were determined to be possible structures. In these cases, it is useful to be able to
1068 and 338, respectively. eliminate structures that contain overlaps between atoms.
Peng et al. [26] used Amorphous Cell to prepare This is done by calculating an energetic figure of merit
nanocomposite membranes by incorporating chitosan- Renergy which penalises close contacts between atoms. The
wrapped multiwalled carbon nanotubes into poly(vinyl search is done using a combined figure of merit, which is
alcohol). They used these structures to explore the intrinsic calculated as a linear combination of Rwp and Renergy .
correlation between pervaporation performance and free The search can be done either by a simulated annealing
volume characteristics. They found that including 1% or by a parallel tempering procedure. Simulated annealing
nanotube material gave a threefold increase in the uses a series of MC moves. Each move is an attempted
permeation flux. They compared with positron annihil- change of one degree of freedom of the structure (either a
ation lifetime spectroscopy. translation or rotation of a group of atoms) after which the
Lin et al. [27] investigated moisture diffusion in epoxy new combined figure of merit is calculated. (The size of
resins. They constructed epoxy-water boxes using Amor- the attempted move, i.e. the distance or the angle, is
phous Cell, and calculated the mean-square displacement of regulated adaptively so that the acceptance probability
water in those structures as a function of temperature. They remains around 50%.) The move is accepted or rejected
compared the desorption activation energies with those according to the Metropolis criterion (with the figure of
abstracted from moisture uptake experiments. merit playing the role of the energy). The temperature is
Pan et al. [28] studied the effects of graphite particles progressively lowered from a high to a low value.
on the diffusion behaviour of benzene and cyclohexane in Parallel tempering uses several different structures at
poly(vinyl alcohol)-graphite hybrid membranes. Amor- different temperatures, which evolve simultaneously using
phous Cell construction was used to create the structures of the same MC procedure. However, there is an additional
polymer and graphite particles with different amounts of move type which swaps a pair of structures between the
hydroxyl and carboxyl groups. Polymer chain mobility temperatures. The high temperature allows rapid movement
was analysed by mean-square displacement and glass across the configuration space, while a low temperature
transition temperature, and it was found that incorporation allows the vicinity of a local minimum to be explored.
of graphite particles into poly(vinyl alcohol) increased the In each case, the final output is a set of structures with
polymer chain mobility. their associated figures of merit. These can be inspected
An example in which the Packing task was used is and further analysed. Figure 4 shows an example of a
found in the work of Hölck et al. [29]. They developed predicted structure; also shown is the expected X-ray
procedures to construct 3D networked epoxy moulding powder diffraction result together with the experimental
compounds using a cross-linking scheme. data used as input to the calculation.
Figure 4. (Colour online) Predicted structure for (E)-2-(4,6-difluoroindan-1-ylidene)acetamide; comparison between experimental
(points) and predicted (line) X-ray powder diffraction data.
Downloaded by [Moskow State Univ Bibliote] at 12:26 14 November 2013
tion on it and analysed the data using simulated annealing space of possible crystal structures is searched for
in Powder Solve. This generated a structure which they candidates of low energy. A crystal structure is specified
then further optimised using density functional theory. by its crystal lattice parameters (cell lengths and angles)
Neumann et al. [31] were able to identify the hitherto and the position and orientation of each molecule inside
unknown structure of phase III of solid methane. Neutron it. This yields a rugged energy landscape with many
diffraction was needed for this difficult material, and was local minima, so a simulated annealing process is used.
again analysed by Powder Solve using simulated The user provides the structure of the molecule of
annealing. It had previously been thought that the crystal interest (or a structure containing multiple molecules),
had a tetragonal space group; they were able to show that together with a force field (any of the usual force fields
this is not the case (and found that the correct space group available in Materials Studio can be used). In addition,
is C m c a). the space groups of interest must be provided (usually a
Corma et al. [32] examined a zeolite, ITQ-40, with an
small subset is required, not the full set of 230 space
unusually open structure. They used several approaches
groups).
(using electron diffraction, X-ray powder and X-ray single
A trial MC move consists of reorienting and moving
crystal data) to determine the structure. Powder Solve was
each molecule as a rigid body. Flexible molecules can also
able to provide a partial structure from the powder data.
be handled: the user can specify which torsion angles are
Luneau et al. [33] synthesised various compounds
active, and these angles will also be changed during the
based on manganese, all with layered structures, most of
which had ferromagnetic behaviour. Again, Powder Solve MC procedure. Then, the size of the lattice cell is adjusted
was used for structure solution. so that the molecules are just in contact but not
Dova et al. [34] worked on an iron-based compound overlapping. The energy is calculated, and the move is
with complex spin crossover behaviour. Powder diffrac- accepted or rejected according to the usual Metropolis
tion data were obtained using synchrotron radiation and criterion. The simulated annealing regime involves an
were analysed using parallel tempering in Powder Solve, initial heating phase, followed by cooling, and the whole
as well as a rival technique (a genetic algorithm). process is repeated for each space group.
The set of structures generated by the Packing
procedure can then be subjected to a Cluster analysis,
3.7 Polymorph Predictor which reduces it by removing similar structures. The
The Polymorph Predictor module of Materials Studio criterion for similarity is based on the radial distribution
attempts to predict the most favourable structures for a function; a similarity measure is calculated from the
molecular crystal, starting only from the molecule itself. difference between the radial distribution functions of the
The basic assumption is that the structures that are most two structures, and if this lies within a specified tolerance,
likely to occur tend to be those of lowest energy – entropy the structures are deemed to be similar, and the one of
is neglected. Whilst the energy of a crystal can be readily highest energy is eliminated.
computed using force field methods, the entropy is much The candidate structures are usually optimised at this
more difficult to obtain, and is therefore ignored. point and again passed through the Cluster analysis.
A workflow diagram is given in Figure 5. The main Finally, they are ranked in order of energy. This is the
part of the procedure is the Packing phase, in which the output from Polymorph Predictor.
Molecular Simulation 1163
Downloaded by [Moskow State Univ Bibliote] at 12:26 14 November 2013
Figure 5. (Colour online) Polymorph Predictor workflow starting with a MC simulated annealing (MC-SA) step.
3.7.1 Applications of Polymorph Predictor based, with kinetic information implicit in the experimen-
Neumann and Perrin [35] studied the crystal structure of tal results.
several small molecules (ethane, ethylene, acetylene,
methanol, acetic acid and urea). Candidate crystal packings
for these molecules were produced using Polymorph 4. Conclusions
Predictor, then optimised using energies calculated by a Materials Studio contains a comprehensive set of modules
combination of density functional theory and an empirical for molecular simulation, and many of these are based on
correction for the van der Waals interaction. the MC method. This technique is exploited in a variety of
Day et al. [36] examined a set of 50 small organic ways, ranging from traditional configuration sampling to
molecules and compared predicted with known exper- structure determination from experimental diffraction
imental crystal structures; they found that about half of the data, through to simulated annealing. Where energy
known structures had an energy that was very close in calculations are required, all modules support a choice of
energy to the calculated minimum (and about a third of the force fields, including COMPASS, pcff, cvff, Universal
known crystal structures in fact were the calculated and Dreiding; in addition, new force fields can be created
minimum energy structure).
using the force field editing facilities, and used in place of
Cross et al. [37] did a combined computational and
those supplied with the application.
experimental study of diflunisal, which is a fluorinated
As all modules share a common infrastructure, they
aromatic carboxylic acid. They did structure predictions
can be brought together to support complex workflows.
using Polymorph Predictor (on five space groups) and used
For instance, crystallisation, quantum and atomistic
the results to guide their choice of crystallisation solvents.
Leusen [38] did structure prediction of a crystal of two modelling can be combined in a single simulation study.
different molecules, a cyclic phosphoric acid and an These workflows can be scripted using the Perl-based
enantiomer of ephedrine. This system is complicated by MaterialsScript API, which also allows access to the entire
both molecular flexibility and chirality. data model. In addition, workflow protocols can be created
Sarma and Desiraju [39] augmented Polymorph from Materials Studio components in the Pipeline Pilot
Predictor results using experimental data. The compu- product, and combined with any other component from the
tational procedure generated lists of low energy structures, vast selection offered. Materials Studio has now been used
and then they reordered these using known structures of effectively over more than a decade for numerous different
similar molecules. This has the effect of supplementing the applications across a range of physical and chemical
enthalpic considerations on which Polymorph Predictor is sciences and almost every industrial sector.
1164 R.L.C. Akkermans et al.
K, Brdas JL, Lazzaroni R, Leclre P. Correlation between the [27] Lin Y, Chen X. Investigation of moisture diffusion in epoxy system:
microscopic morphology and the solid-state photoluminescence experiments and molecular dynamics simulations. Chem Phys Lett.
properties in fluorene-based polymers and copolymers. Chem 2005;412:322– 326.
Mater. 2004;16:994 –1001. [28] Pan F, Peng F, Jiang Z. Diffusion behavior of benzene/cyclohexane
[8] Blanco M. Molecular silverware. I. General solutions to excluded molecules in poly(vinyl alcohol)–graphite hybrid membranes by
volume constrained problems. J Comput Chem. 1991;12:237 –247. molecular dynamics simulation. Chem Eng Sci. 2007;62:703– 710.
[9] Akkermans RLC. Mesoscale model parameters from molecular [29] Hölck O, Dermitzaki E, Wunderle B, Bauer J, Michel B. Basic
cluster calculations. J Chem Phys. 2008;128:244904. thermo-mechanical property estimation of a 3D-crosslinked epoxy/
[10] Wu D, Paddison SJ, Elliott JA. Effect of molecular weight on SiO2 interface using molecular modelling. Microelectron Reliab.
hydrated morphologies of the short-side-chain perfluorosulfonic 2011;51:1027– 1034.
acid membrane. Macromolecules. 2009;42:3358–3367. [30] Dinnebier RE, Ding L, Ma K, Neumann MA, Tanpipat N, Leusen
[11] Lin CM, Li CS, Sheng YJ, Wu DT, Tsao HK. Size-dependent FJJ, Stephens PW, Wagner M. Crystal structure of a rigid ferrocene-
properties of small unilamellar vesicles formed by model lipids. based macrocycle from high-resolution X-ray powder diffraction.
Langmuir. 2012;28:689 –700. Organometallics. 2001;20:5642 –5647.
[12] Pajula K, Taskinen M, Lehto VP, Ketolainen J, Korhonen O. [31] Neumann MA, Press W, Nöldeke C, Asmussen B, Prager M,
Predicting the formation and stability of amorphous small molecule Ibberson RM. The crystal structure of methane phase III. J Chem
binary mixtures from computationally determined Flory– Huggins Phys. 2003;119:1586–1589.
interaction parameter and phase diagram. Mol Pharmaceut. [32] Corma A, D’iaz-Cabanas MJ, Jiang J, Afeworki M, Dorset DL,
2010;7:795– 804. Soled SL, Strohmaier KG. Extra-large pore zeolite (ITQ-40) with
[13] Ren H, Zhang Q, Chen X, Zhao W, Zhang J, Zhang H, Zeng R, Xu the lowest framework density containing double four- and double
S. A molecular simulation study of a series of cyclohexanone three-rings. Proc Natl Acad Sci. 2010;107:13997–14002.
formaldehyde resins: properties and applications in plastic printing. [33] Luneau D, Borta A, Chumakov Y, Jacquot JF, Jeanneau E, Lescop
Polymer. 2007;48:887 –893.
C, Rey P. Molecular magnets based on two-dimensional Mn(II)-
[14] Wood CD, Tan B, Trewin A, Niu H, Bradshaw D, Rosseinsky MJ,
nitronyl nitroxide frameworks in layered structures. Inorg Chim
Khimyak YZ, Campbell NL, Kirk R, Stöckel E, Cooper AI.
Acta. 2008;361:3669– 3676.
Hydrogen storage in microporous hypercrosslinked organic polymer
[34] Dova E, Peschar R, Sakata M, Kato K, Schenk H. High-spin- and
networks. Chem Mater. 2007;19:2034–2048.
low-spin-state structures of [Fe(chloroethyltetrazole)6](ClO4)2 from
[15] Wood CD, Tan B, Trewin A, Su F, Rosseinsky MJ, Bradshaw D,
Sun Y, Zhou L, Cooper AI. Microporous organic polymers for synchrotron powder diffraction data. Chemistry – Eur J.
methane storage. Adv Mater. 2008;20:1916 –1921. 2006;12:5043– 5052.
[16] Li A, Lu RF, Wang Y, Wang X, Han KL, Deng WQ. Lithium-doped [35] Neumann MA, Perrin MA. energy ranking of molecular crystals
conjugated microporous polymers for reversible hydrogen storage. using density functional theory calculations and an empirical van
Angew Chem Int Ed. 2010;49:3330–3333. der Waals correction. J Phys Chem B. 2005;109:15531–15541.
[17] Yang JZ, Liu QL, Wang T. Analyzing adsorption and diffusion [36] Day GM, Chisholm J, Shan N, Motherwell WDS, Jones W. An
behaviors of ethanol/water through silicalite membranes by Assessment of lattice energy minimization for the prediction of
molecular simulation. J Membr Sci. 2007;291:1–9. molecular organic crystal structures. Cryst Growth Des.
[18] Huiyong C, Hongxia X, Xianying C, Yu Q. Experimental and 2004;4:1327–1340.
molecular simulation studies of a ZSM-5 – MCM-41 micro- [37] Cross WI, Blagden N, Davey RJ, Pritchard RG, Neumann MA,
mesoporous molecular sieve. Micropor Mesopor Mater. Roberts RJ, Rowe RC. A whole output strategy for polymorph
2009;118:396 –402. screening: combining crystal structure prediction, graph set
[19] Greathouse JA, Ockwig NW, Criscenti LJ, Guilinger TR, Pohl P, analysis, and targeted crystallization experiments in the case of
Allendorf MD. Computational screening of metal–organic frame- diflunisal. Cryst Growth Des. 2003;3:151–158.
works for large-molecule chemical sensing. Phys Chem Chem Phys. [38] Leusen FJJ. Crystal structure prediction of diastereomeric salts: a
2010;12:12621–12629. step toward rationalization of racemate resolution. Cryst Growth
[20] Khaled K. Electrochemical behavior of nickel in nitric acid and its Des. 2003;3:189–192.
corrosion inhibition using some thiosemicarbazone derivatives. [39] Sarma JARP, Desiraju GR. The supramolecular synthon approach to
Electrochim Acta. 2010;55:5375–5383. crystal structure prediction. Cryst Growth Des. 2002;2:93–100.