Advanced Drug Delivery Reviews 117 (2017) 47–70

Contents lists available at ScienceDirect

Advanced Drug Delivery Reviews

journal homepage: www.elsevier.com/locate/addr

Polymorphic phase transitions: Macroscopic theory and

molecular simulation
Jamshed Anwar a,⁎, Dirk Zahn b,⁎
a
Chemical Theory and Computation, Department of Chemistry, Lancaster University, Lancaster LA1 4YB, United Kingdom
b
Lehrstuhl für Theoretische Chemie/Computer-Chemie-Centrum, Friedrich-Alexander-Universität Erlangen-Nürnberg, Nägelsbachstr. 25, 91052 Erlangen, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Transformations in the solid state are of considerable interest, both for fundamental reasons and because they
Received 2 June 2017 underpin important technological applications. The interest spans a wide spectrum of disciplines and application
Received in revised form 27 August 2017 domains. For pharmaceuticals, a common issue is unexpected polymorphic transformation of the drug or excip-
Accepted 7 September 2017
ient during processing or on storage, which can result in product failure. A more ambitious goal is that of
Available online 20 September 2017
exploiting the advantages of metastable polymorphs (e.g. higher solubility and dissolution rate) while ensuring
Keywords:
their stability with respect to solid state transformation. To address these issues and to advance technology, there
Polymorphic phase transformation is an urgent need for significant insights that can only come from a detailed molecular level understanding of the
Solid state phase transformation involved processes. Whilst experimental approaches at best yield time- and space-averaged structural informa-
Transformation kinetics tion, molecular simulation offers unprecedented, time-resolved molecular-level resolution of the processes tak-
Phase stability ing place. This review aims to provide a comprehensive and critical account of state-of-the-art methods for
Crystal engineering modelling polymorph stability and transitions between solid phases. This is flanked by revisiting the associated
Molecular simulation macroscopic theoretical framework for phase transitions, including their classification, proposed molecular
Molecular dynamics simulation
mechanisms, and kinetics. The simulation methods are presented in tutorial form, focusing on their application
to phase transition phenomena. We describe molecular simulation studies for crystal structure prediction and
polymorph screening, phase coexistence and phase diagrams, simulations of crystal-crystal transitions of various
types (displacive/martensitic, reconstructive and diffusive), effects of defects, and phase stability and transitions
at the nanoscale. Our selection of literature is intended to illustrate significant insights, concepts and understand-
ing, as well as the current scope of using molecular simulations for understanding polymorphic transitions in an
accessible way, rather than claiming completeness. With exciting prospects in both simulation methods develop-
ment and enhancements in computer hardware, we are on the verge of accessing an unprecedented capability
for designing and developing dosage forms and drug delivery systems in silico, including tackling challenges in
polymorph control on a rational basis.
© 2017 Elsevier B.V. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2. Crystal polymorphism and phase stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3. Polymorphic phase transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.1. Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2. Molecular mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3. Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4. Molecular simulation of polymorphic phase transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.1. Molecular simulation methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2. Simulating phase transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5. Molecular simulations of phase stability and transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.1. Crystal structure prediction and polymorph screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

⁎ Corresponding authors.
E-mail addresses: [email protected] (J. Anwar), [email protected] (D. Zahn).

https://doi.org/10.1016/j.addr.2017.09.017
0169-409X/© 2017 Elsevier B.V. All rights reserved.
48 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

5.2. Coexistence and phase diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

5.3. Displacive/martensitic transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.4. Reconstructive transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.5. Diffusive transitions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.6. Effects of defects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.7. Phase stability and transitions at the nanoscale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.8. Kinetics from simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6. Future perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

1. Introduction phase stability and phase transitions, including their classification,

mechanisms, and kinetics. The literature on phase transitions is vast,
Crystal to crystal phase transitions are of considerable scientific inter- disparate, and continually expanding, and insights and developments
est and industrial importance. The interest spans numerous fields and ap- in one (sub)discipline rarely carry over into another. This is not surpris-
plication domains that include Earth sciences [1,2], materials science [3– ing given that phase transitions in solids are a ubiquitous phenomenon,
5], biomineralisation [6–8], explosives [9], and pharmaceuticals [10,11]. exhibited by a spectrum of materials. Here we identify and develop uni-
For pharmaceuticals, drug substances and formulation excipients can fying concepts with a view to providing a better understanding of the
exist in different solid forms, with particular form(s) having advantages nature of phase transitions. The discussion is focused on first-order
over others in terms of efficacy, ease of manufacture, or stability on stor- phase transitions given their applicability to polymorphic phase
age [12–17]. Such a selected form within a formulation may transform to transitions.
another on storage or during manufacture with potentially disastrous We should add that whilst we favour the term phase transition rather
consequences including loss of efficacy [15–17]. An infamous example is than phase transformation, we believe the current consensus assumes
that of the protease inhibitor Norvir® (retonovir) developed for the treat- the two terms to be synonymous and consequently we employ as
ment of HIV-1, for which the second, a previously unobserved polymorph such. We do not review the literature in any comprehensive way, but
precipitated in the soft gelatin capsule resulting in a product recall [18]. rather use the literature to discuss or highlight molecular insights, or il-
Indeed, the issue is recognised by the major product licensing authorities, lustrate a particular issue or behavior. Notable reviews or monographs
whose guidelines recommend that the polymorphism of all new drug en- on phase transitions in solids include: the classic text of Christian [24]
tities is thoroughly investigated [19,20]. though focused on metals and alloys; the more general monographs
Phase transitions in crystals of drugs and excipients have been stud- of Rao and Rao [25], Toledano and Dmitriev [26], Mnyukh [27], and
ied extensively [10,11], with the focus being largely on causal factors, Fultz [5]; pharmaceutically-focused review by Morris et al. [10]; and re-
characterization of the phases, and to a lesser extent kinetics. However, views with an emphasis on transformation mechanisms by Herbstein
despite this interest, our understanding of the mechanisms by which [28], James and Hane [29] on martensitic transformations, and that by
such transitions occur and the factors that govern the kinetics is still Dove [30] on displacive transitions in minerals. Particularly insightful
relatively rudimentary. In particular, we do not have a handle on the are the review papers by Ubbelohde [31–35], despite being rather
molecular level processes occurring during the phase transitions. Exper- dated.
imental approaches at best (e.g. X-ray or neutron diffraction) yield time
and space-averaged structural information, which for polymorphic 2. Crystal polymorphism and phase stability
forms implies primarily the coordinates of the starting (parent) and
the final (daughter) crystalline form. The molecular processes occurring Crystal polymorphism is defined as the ability of a substance, of a
during nucleation or at the transition interface are generally inaccessi- constant chemical composition, to exist in more than one crystalline
ble. This lack of molecular level understanding is limiting the develop- structure. In its broadest definition, the concept includes all possible dif-
ment of strategy and rational approaches that could potentially lead to ferences in the crystalline structure with the exception of homogeneous
significant technological impact e.g. reliable stabilization of metastable deformation, but excludes solvates and amorphous modifications.
forms that offer benefits such as enhanced bioavailability or ease of Structurally, the different modifications, or polymorphs, differ in the ar-
processing. rangement of the atoms or atomic groups in a space lattice. These differ-
Computer simulation, more precisely, molecular simulation, offers ences may be slight, involving only reorientation of identical atomic
the required molecular resolution to make the critical molecular pro- groups on the same lattice, or total with little or no relationship be-
cesses in phase transitions accessible [21–23]. The basis for molecular tween either the connectivity of atomic groups (which may have entire-
simulation are the forces between atoms, which are now relatively ly different coordination numbers) or the lattices. Polymorphism is
well characterized. These forces enable the simulation of the molecular particularly abundant for molecular compounds which, in addition to
trajectories to yield effectively a molecular-resolution microscope, albe- differences in the packing of the molecules, may also exhibit conforma-
it based on simulation. Molecular simulation is now transforming our tional differences that include different resonance structures and rota-
understanding of phase transition phenomena as we illustrate below. tion of moieties around a single bond [36].
The simulations promise molecular level insights that will rationalize Polymorphs of a given compound, whilst having identical chemical
experimental observations, predict phase stability and transition kinet- properties (though they may differ in terms of solid state reactivity)
ics, and enable the development of a robust theoretical framework that can differ markedly in their physical properties. Properties affected in-
could drive advances in pharmaceutical technology. clude solubility, melting point, hardness, density, chemical stability,
The review is concerned primarily with the understanding of poly- crystal morphology and various optical and electrical characteristics
morph phase stability and phase transitions in the solid state. The em- [12,13,15,37], and references in therein.
phasis is on molecular crystals, given that drug molecules are mostly The phenomenon of polymorphism is ubiquitous [38] and its impor-
organic. We also, however, refer to simulations on ionic or inorganic tance in pharmaceutical development is well recognised. Use of an appro-
systems wherever they offer generic inferences and studies on molecu- priate polymorph can enhance the rate and extent of absorption of the
lar systems are lacking. In particular, the interest lies in the molecular drug into the bloodstream, improve the chemical or physical stability of
level processes that characterize how transitions are initiated and prop- the product, or ease the necessary processing involved in the manufacture
agate. We set out to develop a coherent understanding of polymorph of the pharmaceutical product [12,13,15, and references therein].
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 49

employed in tabletting can also induce phase transformations in drugs

[41–43], as can milling [44].
Classical thermodynamics provides a general and sound basis for un-
derstanding structural phase transformations, such as those exhibited
by polymorphic forms. The Clausius-Clapeyron equation, Gibbs phase
rule [45] and the Gibbs free energy function G all satisfactorily predict
the important features exhibited by such transformations. In thermody-
namics, a spontaneous reaction is defined as one that has a natural ten-
dency to proceed without the assistance of any external potential, and is
therefore feasible. Note, however, that nothing is implied in this defini-
tion about the mechanism and any associated energy barriers, and thus
the rate at which the reaction proceeds. Spontaneous reactions are
characterised by a negative change in the Gibbs free energy function
which is given by

ΔG ¼ ΔU þ pΔV−TΔS ð1Þ

where T is the absolute temperature, S is the entropy, U is the internal

energy, p the pressure and V volume.
It follows that under any given set of conditions of temperature and
pressure, of the many possible phases, only the one with the lowest
value of G will be thermodynamically stable. All other phases, under
the same set of conditions, will be unstable and would tend to transform
to the stable phase.
The Gibbs free energy surface G(p,T), therefore, provides a convenient
means for examining the respective stability of different phases. Such a
free energy profile but as a function only of temperature (with pressure
fixed at ambient) is shown for a hypothetical dimorphic system in Fig.
1. Points of intersection of the G surfaces of the phases represent condi-
tions at which the phases are in equilibrium with each other and can
co-exist. At these points ΔG = 0 for a transition between the phases.
The conditions at the intersection points, that is, temperature and pres-
sure, are referred to as the transition conditions, since any departure
from them results in one of the phases becoming thermodynamically sta-
ble, provoking the transformation of the other phases to this one. For the
system in Fig. 1 the transition temperature is TI–II. Below this temperature,
Form II has the lower free energy and is thus the stable phase. Above TI–II,
Form I becomes the thermodynamically stable phase.
Fig. 1. Top: Free-energy surface of a hypothetical enantiotropic dimorphic system as a In general, polymorphic transformations do not occur at exactly the
function of temperature. Bottom: Hysteresis of the I ↔II transition when crossing the transition conditions but show hysteresis, that is, a difference in the
transition temperature TI–II. The II–I and I–II transitions observed upon heating (red
transition temperature or pressure depending on whether the transfor-
curve) and cooling (blue) respectively, are subject to nucleation barriers and thus
experience kinetic trapping. Note that the centre of the hysteresis loop, commonly taken mation is in the forward or the reverse direction [46,32,47]. Hysteresis
as the experimental transition temperature, needs not to coincide with the occurs because of kinetic barriers to nucleation of the new phase [24,
thermodynamic definition of the transition temperature i.e. at which GI = GII. 25]. At conditions close to the transition point, the energy barriers to nu-
cleation may be considerable. The rate of transformation within the
time period of the experiment is then too low as to be imperceptible
A significant proportion of problems of product instability occur as to measurement. It is only when the conditions depart further away
a result of polymorphic phase transitions. On storage or processing, from the transition point that the transformation becomes sufficiently
the selected polymorph transforms to a more stable form. Examples fast so as to be measurable. The extent of the transition barrier varies
of problems encountered include reduction in syringeability of inject- from substance to substance depending on the changes involved in
able products, caking of suspensions, grittiness in creams and the crystal structure on transformation. The extent of hysteresis in
ointments (which may cause these products to be not only cosmeti- turn depends exponentially on the barrier and the rate at which the
cally unacceptable but also painful on application), crystal growth in thermodynamic conditions (temperature, pressure etc.) are changed
products intended to be solutions, and depression of the melting within a given experiment. For barriers less than kβT (where kβ is the
point of suppositories such that their use in the summer becomes im- Boltzmann constant and T the absolute temperature), any hysteresis
practical [12,13,39,40]. The new phase may exhibit a different solubil- may be experimentally indiscernible.
ity and hence dissolution kinetics, which in turn may adversely affect Polymorphic transitions are often also observed to be irreversible [12,
the bioavailability. Clearly, to avoid such issues, the appropriate choice 25]. Throughout the conditions studied the transformation might be ob-
of polymorph would be that which is thermodynamically stable under served to proceed only in one direction. This may be because the reverse
the conditions of storage experienced by the product. However, this is rate is too slow, or because the G surfaces of the phases concerned do not
at odds with the often-desired need to increase the solubility of the intersect and the one and same phase remains thermodynamically stable
drug with a view to enhancing its bioavailability, which can be met throughout, right up to its melting point. If the latter is the cause, then the
by a thermodynamically metastable form. Moreover, the conflicting transformation is referred to as monotropic, in contrast to thermodynam-
storage versus bioavailability issues may be complicated by process- ically reversible transformations which are called enantiotropic [48,49].
ing, for instance it has been demonstrated that the pressures normally Monotropic transformations are not subject to a transition temperature,
50 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

but are controlled by kinetic factors. These transformations accordingly capacity, at the transition point. The transitions are classed in terms of
occur as a function of both temperature and time. ‘nth order’ where n is an integer given by the lowest derivative of
Forms that are thermodynamically unstable but can be isolated are Gibbs free energy G with respect to temperature T and pressure p,
termed metastable. Metastable forms once isolated can be quite stable; which shows a discontinuous change at the transition. Thus, a first
an extreme example being the diamond phase of carbon, since the dia- order transition is defined as one in which a discontinuity occurs in
mond → graphite transition occurs at a rate of practically zero. The de- the first derivative of the free energy. These derivatives correspond to
gree of metastability depends on the energy barriers to phase entropy S (or latent heat of transition) and volume V respectively, viz.
transition, which depend on the changes involved in bonding to form
a nucleus of the stable form within the parent lattice. G ¼ minfGI ðT; pÞ; GII ðT; pÞg ð2Þ

3. Polymorphic phase transformations ΔG ¼ GII −GI ¼ ΔH−TΔS ¼ ΔU þ pΔV−TΔS ¼ 0 at transition point




3.1. Classification ∂GI ∂GII ∂ΔG ΔH
¼ SI ≠ SII ¼ ; ¼ −ΔS ¼ − ≠0 ð3Þ
∂T p ∂T p ∂T p T
Classification schemes which have been found to be particularly use-
ful are those proposed by Ehrenfest [50], Buerger [51–53], and


∂GI ∂GII
Ubbelohde [32,35]. Ehrenfest's classification is based on the behaviour ¼ V I ≠V II ¼ ð4Þ
∂p T ∂p T
of thermodynamic quantities, such as entropy, volume and heat

Second order transitions do not involve entropy or volume changes,

but are characterised by discontinuities in the second derivatives of free
energy, that is heat capacity Cp, thermal expansivity κ, and compressibil-
ity β, whilst having continuous first derivatives.
2
!
 2
!
∂ GI ∂V I ∂ GII
¼ ¼ −V I βI ≠−V II βII ¼ ð5Þ
∂p2 ∂p T ∂p2
T T

2
!
 2
!
∂ GI ∂V I ∂ GII
¼ ¼ V I κ I ≠V II κ II ¼ ð6Þ
∂p∂T ∂T p ∂p∂T
!
 !
2
∂ GI ∂SI C Ip C IIp 2
∂ GII
¼ ¼− ≠− ¼ ð7Þ
∂T 2 ∂T p T T ∂T 2
p p

Third and higher order transitions can be defined in principle by dif-

ferentiating further.
The changes in the thermodynamic properties (as a function of tem-
perature) at the transition point for first and second order phase transi-
tions are shown in Fig. 2. First order transitions show discontinuities in
volume, enthalpy and entropy, and have infinite heat capacity at the
transition. The heat capacity is infinite at the transition because it is
given by the derivative of enthalpy with respect to temperature, that
is, Cp = ∂H/∂T, and the enthalpy shows a discontinuous change (Eq.
(3)). Physically, the energy added to the system is utilised in driving
the transition rather than on raising the temperature, and hence it rep-
resents the latent energy of the transition.
For second order transitions the gradient of Gibbs free energy is con-
tinuous, and the entropy and volume of the systems do not change.
Since there is no entropy of transition ΔS, neither is there any enthalpy
of transition ΔH. The important discontinuous thermodynamic function
for second order transitions is that of heat capacity (see Fig. 2).

Table 1
Classification of phase transformations according to Buerger [51].

Type of transition Transition Example

velocity

First-coordination transformations
(a) Reconstructive Sluggish Calcite-aragonite (CaCO3)
(b) Dilatational Rapid Caesium chloride (CsCl)

Second-coordination transformations
(a) Reconstructive Sluggish Quartz-cristobalite-tridymite (SiO2)
(b) Displacive Rapid High-low in SiO2

Transformations involving disorder

(a) Orientational Rapid Ferroelectric-paraelectric NH4H2PO4
Fig. 2. Variation of thermodynamic properties at the transition temperature Tt for first and
(b) Substitutiona Sluggish
second order phase transitions. The blue and red curves reflect the free energy of the low-
Transformations of bond type Sluggish Grey-white tin
and high-temperature phases II and I, respectively.
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 51

It is difficult to visualise the nature of the free energy surface of sec- associated with this class, there appears to be no clear definition. Char-
ond order transitions. Since the gradients (the first derivatives) of the acteristics typically associated with martensitic transitions include ve-
free energy are continuous, the gradients of the free energy surfaces of locity of transition being of the order of propagation of sound and
the phases concerned are equal at the transition point. Hence, the sur- independent of temperature (athermal), diffusionless and cooperative
faces cross at a sharp transition point as with first order transitions, movement of atoms/molecules, definite orientational relationship be-
but instead partially overlap at a transition regime (see Fig. 2). tween the lattices of the phases concerned, extent of transformation
The scheme of Buerger is based on structural relationships between being dependent on the degree of cooling below the critical tempera-
the crystal structures of the phases concerned. Transformations are classi- ture, and transformation being affected by shear stresses. Another key
fied on the basis of structural changes involving primary (nearest neigh- stated feature is macroscopic change of shape of the transformed region.
bours) or higher (next nearest neighbours) coordination, changes in Some of these characteristics, however, are now being abandoned [24],
type of bonding, and whether disorder is involved. An integral part of and others do not offer any clear distinction. The reduced criteria appear
the scheme is the relationship between the structural changes involved to be that martensitic transitions (i) are first order, (ii) exhibit orienta-
and energy barriers affecting the kinetics of the transformations. The clas- tional relationships between the lattices, and (iii) and displacive shear
sification is given with examples in Table 1. Note the focus on inorganics. of the lattice upon transformation gives rise to shape-change in the ma-
If the crystal structures are not similar, and the change from one struc- terial. Martensitic transformations are observed in metals, alloys, and
ture to the other involves significant re-organisation of the atomic (mo- ceramics, and have significant technological applications that include
lecular) species, requiring (inter-species) bonds to be broken and new manufacture of transformation-toughened materials, smart materials
ones to be formed, the transformation is considered to be reconstructive. utilizing shape memory effects, and self-healing ceramics [29,54,55].
Since bond breaking is involved, Buerger expects the energy barriers for The term martensitic is derived from the transformation of the austenite
such transformations to be high and the transformation rate to be form of iron (containing a small amount of carbon) by rapid quenching
sluggish. to yield the hard form of steel called martensite. A number of molecular
Transformations involving subtler structural change in the first coor- crystals are known to exhibit martensitic-type characteristics e.g.
dination are termed dilatation. Here no bonds are broken and the trans- hexamethylbenzene and DL-norleucine [56]. The transformation-in-
formation is thought to occur by mere differential dilatation of the duced shape change of martensitic materials implies the conversion of
whole structure. Transformations of CsCl, NH4Cl and NH4Br, which are chemical energy to mechanical activity i.e. work. On this basis,
all relatively simple structures, are considered to occur by this mecha- thermosalient molecular crystals [15, and references therein]; [57] (col-
nism. According to Buerger, as no bond breaking is involved, the energy loquially known as ‘jumping crystals’) in the act of jumping from a hot
barriers in dilatational transformations are considered to be relatively plate on transformation do mechanical work, and hence these transfor-
small and the transformation rate (at a given driving force) to be rapid. mations would be considered to be martensitic.
Transformations involving only the second coordination and where So, how do polymorphic phase transitions fit into these classification
the crystal structures of the phases concerned are similar, such that schemes? The transitions are invariably first order, and therefore local
going from one phase to another does not require breaking or making in nature occurring by nucleation and growth. They may or may not ex-
of bonds but rather distortion of bond angles and distances, are classi- hibit an orientational relationship between the lattices of the phases
fied as displacive. Again, since only minor structural changes are in- concerned. We do find the terms displacive and reconstructive useful,
volved, the energy barriers are considered by Buerger to be small. reflecting the nature of the lattice re-arrangements between the initial
Buerger's final category is that in which the crystal forms concerned and final lattices and the potential link with kinetics. We should add
differ greatly in the nature of bonding. Examples are polymorphs of tin that until quite recently, structural concepts developed in the early
(grey and white), where the character changes from semiconducting to last century combined with intuition of atomic bonding was the only
metallic, and carbon (diamond-graphite), where the change is from an way to estimate whether the energy barriers to nucleation were high
insulator to a semiconductor. Note that both of these transformations or low. Using molecular simulations, as described below, we are now
involve major structural changes and, therefore, could also be classified able to characterise phase transitions by computing both atomic path-
as reconstructive having large energy barriers. ways and energy profiles.
Although Buerger's classification has become well established, it does
appear to have some significant drawbacks. Firstly, the classification ap- 3.2. Molecular mechanisms
pears to have been developed with ionic and framework crystals in
mind and mapping onto molecular crystals is not straightforward. For in- How do structural phase transitions occur at the molecular level?
stance, molecular crystals rarely exhibit meaningful second coordination What actually happens to the molecules in time and space? This implies
shells. Nevertheless, the terms displacive and reconstructive are now the elucidation of the ‘reaction coordinate’ – the mapping of the ener-
being used for phase transitions in molecular crystals, the former when getics and geometrical changes that occur in the course of a reaction, a
the structural differences between the phases are minor, and the latter seminal development by Eyring in 1935 [58]. However, whilst the im-
when the two phases differ in their hydrogen bonding networks. The plementation of this concept for solid-solid transformations is in princi-
use of the term `mechanism’, to describe the observed geometrical rela- ple straight-forward, in practice it is seriously complicated by the large
tionships (for example, distortion and dilatation between the static crystal number of molecules (atoms) involved. This particularly holds for mo-
structures, is particularly unfortunate, being the cause of much confusion. lecular crystals which not only show displacements of molecular cen-
Its use has implied that the geometrical relationships actually describe the tres of masses, but also rotation and deformation of the molecular
structural changes taking place at the atomic/molecular level during a entities.
transition which we now know not to be the case. Mechanistic studies based on Buerger's ideas involve comparisons of
Ubbelohde [32–35] divided phase transitions into continuous and dis- the crystal structures of the polymorphs, and then a search for particular
continuous. In continuous transitions the crystal structure is expected to lattice transformations and molecular translations and rotations which
change smoothly and continuously from one form to another. In discon- link the structures. The molecular translations and rotations are then
tinuous transitions the structural change involved is not smooth. Al- proposed as mechanisms for the transitions. A significant issue
though the classification is not based on any theoretical considerations, with this approach is that it suggests that the transitions occur
discontinuous transitions could correspond to Ehrenfest's first order tran- homogenously via the proposed molecular translations and rotations
sitions and the continuous transitions to order-disorder type transitions. throughout the bulk crystal. Such smooth transformation suggests that
Another class of transitions often discussed separately is that of mar- the free energy surfaces of the two phases overlap within an extended
tensitic transformations [24]. Although various observations are regime, i.e. refers to second order transitions (Fig. 2) [59]. In contrast
52 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

to this, first order transitions occur via nucleation and phase growth, In real crystals, nucleation occurs at preferred sites such as edges,
and molecular organisation occurring at the reactive interface for surfaces, grain boundaries, stacking faults, dislocations and point de-
these transitions is unlikely to follow the structural changes proposed fects. Such deviations from the ideal lattice can considerably lower the
by Buerger-type analysis (see below in Section 5). This is also reflected necessary activation energy for the nucleation step. Indeed, these crystal
by the final state found after a first order phase transition, as multiple imperfections appear to be a necessary condition reducing the hystere-
nucleation events generally yield polycrystalline or domain structures sis of polymorphic transitions. An extreme example illustrating this are
with single crystal to single crystal transitions being exceptions and typ- high quality single crystals that only transform after defects are intro-
ically limited to small or nanocrystals [28,60]. duced by mechanical means, for example, by pin pricks [65].
There is now overwhelming evidence that all solid-state transforma- At the phase front (the region where active rearrangement is taking
tions of the first order kind occur heterogeneously by way of nucleation place) it is thought (e.g. [63,25]) that the process taking place is just
and growth [24]. Further, elegant experiments by Mnyukh and co- simple relocation of the atoms/molecules from the initial to the product
workers [56,61–66], on single crystals of molecular compounds, have phase, and that there is no intermediate amorphous layer, as had been
shown that there is much similarity between the growth of a new proposed by Hartshorne and co-workers [70,71] and Bradley [72], but
phase in a solid-state transition and the growth of a crystal from a liquid only a small gap (due to the mismatch between the lattices of the initial
or a gaseous phase. The studies involved direct microscopic observation, and the product phase) which is on average approximately half a molec-
under maximum optical resolution, of single-crystal to single-crystal ular layer. The process is illustrated in Fig. 3. The idea of the amorphous
transitions, coupled with Laue X-ray diffraction on a variety of molecu- interface was discounted because the observed rates of transformations
lar crystals. For each compound, the crystals of the new phase (daugh- could never be reconciled with this postulate. In addition, results of ex-
ter), which grew within the parent crystals, exhibited facets with low perimental measurements and theoretical calculations (using the atom-
crystallographic indices that were, in general, irrational relative to the atom potential method) of the tensile strength of the interface [65] have
parent lattice. The developing facets (representing the interface) also been found to be consistent with the above proposed mechanism, rath-
showed a series of steps such as those normally observed in crystals er than with the existence of an amorphous layer. However, recent mo-
growing from the liquid or the gaseous phase. In general, no preferred lecular simulations reveal that the molecular processes at the
orientation of the daughter crystals with respect to the parent lattice transformation interface do not always fall so neatly into this scheme.
was observed, and a certain degree of superheating and supercooling Depending on the nature of bonding, interfaces can be sharp, extended,
was always found to be necessary to induce the transitions. This all even diffusive [59,73].
strongly suggests that the mechanics of polymorphic phase transitions
are, in essence, similar to that of crystallisation but with the difference 3.3. Kinetics
that the bulk medium for the phase transitions is a crystalline lattice.
For glutaric acid, the habit faces of the daughter crystals have indeed The rates at which polymorphic transitions occur vary enormously.
been observed to be the same as those of crystals grown from the Depending on the nucleation barrier, transitions may be extremely
melt [62]. rare as, for example, the diamond to graphite transformation that occurs
The theory of nucleation in solids developed by Turnbull [67] is es- over geological timescales at ambient conditions. At the other extreme,
sentially an extension, to include effects of factors specific to condensed many compounds show transformations that are so spontaneous that
phases, of the theory proposed by Volmer and Weber [68] and Becker time-resolved experiments become challenging [74]. In developing
and Doring [69] for homogeneous crystal nucleation from the melt. At the theoretical framework for kinetics, it is necessary to make a clear
conditions just above the transition point, local fluctuations due to ther- distinction between the two stages of transformation, i.e. nucleation
mal agitation are thought to cause some atoms (or molecules) of the ini- and phase propagation, each of which is subject to a distinct energy bar-
tial phase to take up structural positions corresponding to the product rier. Typically, the energy barrier for propagation is much smaller than
phase. The majority of these fluctuations result in the emerging periodic that for nucleation. Indeed, the barrier to nucleation constitutes a con-
structure being below a certain critical size, and hence show a net in- ceptual maximum for the propagation energy barrier. As a consequence,
crease in Gibbs free energy and are unstable. Only those fluctuations phase front propagation can be very fast, even at the speed of sound -
which lead to phase domains that exceed this critical size are capable which is that of (elastic) density wave propagation in the crystal. An ex-
of continued existence and become the nuclei of the product phase ample of fast phase-front propagation in molecular crystals is that ex-
(the full thermodynamic account is given in Section 3.3). Growth of hibited by DL-norleucine, where the rate of advance of the interface
the new phase then reduces the net Gibbs free energy and proceeds can be of the order of 10 cm/s even at very low superheating or
by relocation of the molecules at the interface from the initial phase supercooling [56].
onto the nuclei. At the phase front, the molecular arrangements match The phase transition kinetics (for both nucleation and interface ad-
neither the initial nor the final crystal structure and are thus energetical- vance) have a certain dependence on the extent of superheating or
ly unfavorable. This local instability hence propagates such that the supercooling (see Fig. 4). The implication is that standard Arrhenius ki-
transition process continues until the entire crystal has transformed to netics (that assume temperature-independent free energy barriers) are
the new phase. poorly applicable. At temperatures very close to the transition

Fig. 3. Proposed mechanism for the molecular rearrangement at the interface of a polymorphic transformation. Molecules detach one by from the parent phase and attach onto the surface
of the emergent daughter phase. Energy considerations suggest that the process of detachment and attachment would occur on a molecule per molecule basis. This is akin to shifting a
carpet by inducing a kink which is then propagated rather than wholesale shift of the carpet which is much more challenging.
(Reproduced with permission from reference [127] J. Anwar, S. C. Tuble and J. Kendrick, J. Am. Chem. Soc., 2007, 129, 2542–2547; copyright 2007 American Chemical Society).
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 53

Fig. 4. Temperature dependence of the transformation rate for the transformations I → II

and II → I as observed below or above Tt, respectively. Note that the overall rate is
mainly determined by that of nucleation, whilst interface advance for polymorphic
phase transformations is typically relatively fast. Fig. 5. Variation of the free energy change for nucleation ΔGN as a function of radius r at
various temperatures. ΔGN⁎ is the critical free energy characterizing a nucleus with a
critical radius rc.

temperature Tt, the net rate of transition is negligible (be it because of

the structure being inert or because of balancing of the phase-front The corresponding expression for a cubic nucleus of dimension a
propagation in either direction). This temperature range over which would read:
the transformation rate is low corresponds to the hysteresis in the ob-
served transition temperature. As the temperature is further removed ΔGN ¼ a3 Δg 0V −6a2 γ0 ð9bÞ
from Tt the rate increases. For the transition driven by supercooling,
the rate attains a maximum and then begins to decrease. For the where ΔgV is the bulk free energy per unit volume and γ is the surface
superheating-induced transition, the rate continues to increase rapidly free energy per unit area.
without any maximum. This behaviour is attributed to two causes: the
8
dependence of the rate of nucleation and of interface advance on the ex- >
> GII −GI
>
< II I→II ðTbT t Þ
tent of superheating or supercooling, that is, ΔT = (T−Tt), and on the ΔGbulk V nucleus
absolute temperature T. Δg V ¼ ¼ ð10Þ
V nucleus >
> GI −GII
First order phase transitions are initiated by local nucleation events, >
: I II→I ðTNT t Þ
V nucleus
resulting in budding of new phase domains, the nuclei, which then grow
until the full crystal is transformed. The formation of nuclei results in the
From the expression 9a, it is clear that for small radii r the second
creation of new surfaces and, if volume change is involved (which it in-
term (that due to the interface) will dominate, and since its contribution
variably is), stresses and strains. These make a positive contribution to
is positive, ΔGN will be positive. Consequently, any embryo that might
the net free energy change involved in the process of nucleation. The
form with a radius below some critical size will be thermodynamically
net free energy ΔGN, therefore, does not simply equate to the free ener-
unstable and hence will disperse. With increasing r, ΔGN goes through
gy change due to the formation of the new phase within the nuclei
a maximum at r = rc, the critical radius, and then begins to decrease
ΔGbulk, but is the combination of this and the free energy change due
as the bulk free energy term begins to dominate. The variation of ΔGN
to the generation of the new interface ΔGinterface, and associated strains
as a function of the radius at a number of different temperatures is
ΔGstrain [67]. For nucleation to occur, ΔGN must be negative.
shown in Fig. 5. At r = rd, ΔGN goes through zero and for r N rd it is neg-
ative. Nuclei with r N rd become stable crystals. Embryos with r between
ΔGN ¼ ΔGbulk þ ΔGþinterface ΔG ð8Þ rc and rd are metastable nuclei – they are thermodynamically unstable
since ΔGN is positive but would require a re-crossing of the nucleation
barrier to dissolve. Consequently, the down-hill move to larger and
The first term is negative while the second and third terms are pos- thus more stable nuclei is preferred.
itive. To a first approximation the strain term may be ignored (or, more Expressions for the critical radius rc and the critical excess free ener-
precisely, implicitly considered as part of the bulk and interface terms). gy ΔGN⁎ can be obtained by using the necessary condition of ∂ΔGN/∂r =
Indeed, it is intuitive to assume that a forming nucleus would adopt a 0 and rearranging for rc:
shape such that shear stresses at its surfaces cancel out. The actual
shape of a forming phase domain depends on the relation of parent 2γ
rc ¼ ð11Þ
and daughter lattices and may take the form of a cube, prism, or a com- Δg V
plex polyhedra. For the sake of simplicity, the nucleation theory
discussed in the following is elaborated for spherical nuclei. While nu- and substituting this into Eq. (9a)
clei in solid-solid transformation are typically not spherical, this is still
16πγ3
the most commonly used model. The general physics discussed below ΔGN  ¼ ð12Þ
may be transferred to all shapes as analogous expressions can be devel- 3Δg V 2
oped for any regular shape. Then, for a spherical nucleus of radius r the
explicit form of ΔGN is To a first approximation the dependence of γ on temperature may
be ignored. ΔgV, as it may be seen below, whilst zero at Tt, is however
proportional to (T-Tt) at other temperatures. At temperature Tt
4 3
ΔGN ¼ πr Δg V −4πr 2 γ ð9aÞ
3 ΔG ¼ GII −GI ¼ ΔH−T t ΔS ¼ 0 or; if written per volume unit;
54 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

ΔG of the critical-sized embryos that may be present and the probability

Δg V ¼ ¼ Δh−T t Δs ¼ 0
V nucleus (frequency) of their conversion to the new phase. If the number of em-
bryos present in the parent phase are N0, then from Boltzmann statistics
the number of these which are of critical size (Nc) can be approximated
So, by


−ΔGN 
ΔH Δh Nc ¼ N0 exp ð17Þ
ΔS ¼ and Δs ¼ ð13Þ kB T
Tt Tt
where kB is the Boltzmann constant.
The conversion of critical-sized embryos to stable nuclei requires the
Therefore, at any other temperature T (substituting into the Gibbs transfer of at least one atom or molecule across the interface onto the
free energy function i.e. Eq. (1) and assuming ΔH and ΔS do not change embryo. Such transfers or jumps are subject to the activation energy
with temperature – hence the reason for the non-exact term) G⁎ of phase-front propagation. For the transformation of phase II to I, if
the embryos of emerging phase I are surrounded by n atoms (or mole-
ΔH ΔH cules) at the interface, the frequency νgrowth step with which the atoms
ΔG≅ΔH−T ¼ ðT t −T Þ
Tt Tt ð14Þ (or molecules) cross the interface and attach to the embryo is given by
Δh
Δg V ≅ ðT t −T Þ

Tt G
νgrowth step ¼ ν exp − II→I ð18Þ
kB T
Substituting this into Eqs. (11) and (12) gives
where ν is the lattice vibration frequency, and G⁎II → I is the free energy
2γT t of activation for the transfer of atoms (or molecules) of phase II to the
rc ¼ − ð15Þ
ðT t −T ÞΔh emerging nuclei of phase I (see Fig. 6).
The rate of post-critical nuclei formation (per unit volume per unit
time) is the product of the population of critical-sized embryos and
16πγ 3 T t 2
ΔGN  ¼ 2
ð16Þ their rate of conversion to stable nuclei and is therefore given by
3Δh ðT t −T Þ2


dNCþ ΔGN  þ GII→I
where Δh is the enthalpy change per unit volume of the phase ¼ N0 ν exp − ð19Þ
dt kB T
transition.
Hence, the critical radius rc is inversely proportional to ΔT = Tt-T. where G⁎II → I reflects the kinetic barrier to phase propagation in the II
At temperatures close to Tt, ΔgV → 0 and rc becomes infinitely large. → I direction, whilst ΔGN⁎ describes the thermodynamic disfavoring of
Consequently, nucleation cannot occur at T t. On increasing ΔT the forming a nucleus of critical size of the new phase.
magnitude of rc and the free energy barrier ΔGN⁎ decrease, resulting The nucleation rate predicted by this expression as a function of
in an increase, within a given time, in the population of embryos temperature takes the form shown in Fig. 4. At Tt the rate is zero. In
which go on to become stable nuclei. The rate of nucleation, there- the temperature region below Tt it passes through a maximum and
fore, increases with increase in the degree of superheating or then tends to zero at 0 K. Above Tt the rate increases rapidly with tem-
supercooling. perature. The cause of this characteristic variation is the temperature
Consider now the effect of absolute temperature on the nucleation dependency of ΔGN⁎ described earlier (Eq. (16)).
rate. The rate of nucleation is given by the product of the population Once a post-critical nucleus is formed, the transformation kinetics
then depends on the rate of phase propagation (interface advance).
The effect of temperature on the rate of phase propagation can also ob-
tained from transition state theory. Here we follow and build on the for-
malism developed by Young [75] and Rao and Rao [25]. In full analogy to
the above discussion, the advance of the interface is considered to be via
an atom-by-atom (or molecule-by-molecule) transfer across the inter-
face between the two phases, with the transfers or jumps being restrict-
ed by a free energy barrier. The latter is G⁎II → I for the forward II → I
direction (phase propagation in favor of the more stable phase) and
G⁎I → II for the reverse I → II process (see Fig. 6). Note that G⁎II→ I +
(−ΔGI–II) = G⁎I → II and when ΔGI–II b 0 i.e. T N Tt, the barrier for back-
propagation (I → II) is larger than for the forward direction i.e. G⁎II → I
b G⁎I → II.
Consider the transition of phase II into phase I (net growth of phase
I). The number of atoms (or molecules) transferred from phase II to I via
the interface area A per unit time may be written as


dNII→I GII→I 
¼ A  nII pII→I ν II exp − ð20Þ
dt kB T

where nII is the number of atoms (or molecules) per unit area of phase II
at the interface, νII is the frequency of atom (molecule) vibration normal
Fig. 6. Free energy barriers to transfer of molecules across an interface. Note that G⁎II → I + to the interface and pII → I is the probability that this vibration is directed
(−ΔGI–II) = G⁎I → II. towards a lattice site of phase I.
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 55

Similarly, the number of atoms (or molecules) leaving phase I is Eq. (25) predicts that at Tt, since ΔGI–II = 0 and exp.(ΔGI–II/kBT) = 1,
given by the net speed of phase front propagation is zero. At moderate favoring of

 the phase transition, ΔG is negative but small and exp.(ΔGI–II/kBT) b 1,
dNI→II G hence the rate will also be low. With increasing thermodynamic
¼ A  nI pI→II νI exp − I→II
dt k T driving of the II → I transition, ΔGI–II gets increasingly negative and

 B  ð21Þ
G −ΔGI−II G⁎II → I bb G⁎I → II, thus leading to a diminishing rate of back-propagation.
¼ A  nI pI→II νI exp − II→I
kB T For ΔGI–II → −∞, exp.(ΔGI–II/kBT) = 0, and Eq. (25) formally predicts a
speed limit – which may be associated with the speed of sound within
The net transfer of atoms (molecules) into phase I is the parent phase II, i.e. the maximum speed of any mechanical action

 applied to the solid.
dNI GII→I  If the transition is induced by supercooling, fast phase-front propaga-
¼ A  nII pII→I νII exp − −A
dt
kB T  tion driven by maximum ΔG potential is counteracted by a decrease in

GII→I −ΔGI−II
 nI pI→II νI exp − ð22Þ the temperature-dependent rate constant k(T) due to the drop in tem-
kB T
perature. The equation, therefore, predicts that with an increase in
supercooling, there is a maximum in the phase transformation rate
At phase coexistence, phase fronts may be subject to fluctuations, followed by a decrease. The location of this maximum depends on
but no net growth of either phase occurs. As ΔGI–II = 0 the two expo- both, the thermodynamic favoring of the transition ΔG and the barrier
nential terms become identical and the coexistence condition reads: to phase front propagation G⁎, which in principle can be deduced from
experiment using Eqs. (26) and (27). On the other hand, on superheating
dNI the rate of the transformation increases rapidly with no maximum pre-
0¼ at T ¼ Tt → nII pII→I ν II ¼ nI pI→II ν I ð23Þ
dt dicted, with both the thermodynamic potential ΔG and the tempera-
ture-dependent rate constant k(T) acting in concert. An illustrative
example of this kinetic behavior of interface advance is exhibited in sin-
In other terms, the differences in density n and vibrational frequency gle crystals of p-dichlorobenzene as a function of ΔT [61].
υ in phases I and II is compensated by the different probabilities pII → I In typical kinetic studies of phase transitions, see e.g. [76], it is nor-
and pI → II that a spontaneous vibration locks into a lattice site of phase mally assumed that the effect of temperature on the rate of interface ad-
I and II, respectively. vance is solely due to its effect on the rate constant as given by the
To rationalize this issue, consider for example the case of nI b nII. For a Arrhenius equation. From the foregoing discussion, it is clear that this
successful jump from phase I to phase II, atoms (molecules) will need to is only true at temperatures well away from Tt. This conclusion is indeed
leave a comparably large volume and lock into a smaller one. For an in- borne out in experimental studies, see e.g. [77,78].
dividual jump across the interface, chances to reach a suitable lattice site The characterisation of bulk kinetics of phase transitions in the solid
are thus pI → II b pII → I. However, the denser phase II offers more candi- state is fundamentally different from that in the liquid or the gaseous
dates for such jumps, effectively balancing the net transformation rate phase. In the solid state the important conventional concepts of concen-
at Δ G = 0. It is intuitive to assume nIIpII→ IνII ≈ nIpI → IIνI to hold as a tration, order of reaction or molecularity have little or no application. In-
good approximation beyond the phase coexistence line, i.e. for ΔG ≠ 0, stead, because of the relative immobility of the constituent atoms or
and what follows will be based on this consideration. molecules, the kinetics are governed by topochemical factors (n, p and
The net speed of phase interface propagation may be deduced from ν in Eq. (20)).
the increase of a nucleus volume as a function of time Ideally, in common with interface-controlled solid state reactions,
the fundamental parameters in solid phase transitions are the rate of


 nucleation characterized by the rate constant kN, the spatial distribution
dx d V nucleus 1 d NI
¼ ¼  ð24Þ of the formed nuclei, that is, whether the nuclei are confined to the sur-
dt dt A ρI dt A face or occur throughout the bulk crystal, and the rate of subsequent ad-
vance of the formed interface, characterized by the rate constant kG. The
rate of interface advance may be anisotropic, being characteristic for the
where ρI is the number density of phase I. Using the above consider-
different crystallographic directions in the crystal k(hkl)
G . Coupled to the
ations, we get
temperature-dependent rate constants are the temperature-indepen-


 dent activation energies EN⁎, EG⁎ (hkl). Nucleation may be instantaneous,
dx 1 GII→I  GII→I  −ΔGI−II
¼ nII pII→I ν II exp − −nI pI→II νI exp − or its rate may follow a linear, exponential, or another power law. The
dt ρI
kB T 
 kB T
1 GII→I  ΔGI−II progress of interface advance is, to a first approximation, usually linear
¼ nII pII→I ν II exp −  1− exp with time, governed by a maximum speed given by that of sound
ρI kB T kB T
along the corresponding cystallographic direction [79]. Further, finite
ð25Þ
size effects may be crucial for materials that either exhibit domain struc-
ture or occur as discrete crystallites.
which is commonly written as
In practice, the kinetics are often further complicated and not always


dx ΔGI−II quantitatively reproducible [66]. Irreproducibility, in the main, stems
¼ kðT Þ 1− exp − ð26Þ from the variation in crystal perfection. Both nucleation and the rate
dt kB T
of interface advance are markedly dependent on defects in the crystal.
where k(T) is the temperature-dependent rate constant for the atomic In a good quality crystal, that is, one with few defects, nucleation is dif-
jump across the interface triggered by the activation energy GII → I* for ficult if not impossible. If, however, certain defects are introduced artifi-
forward propagation of the phase front. cially, for example, by means of pin pricks, then nucleation at these

 defects is almost instantaneous even at very low superheating or
GII→I  supercooling. Likewise, the rate of interface advance can also show con-
kðT Þ ≈ const  exp − ð27Þ
kB T siderable variation depending on the quality of the crystal, and may not
always be linear with time: it may increase or decrease with time or on
This rate constant relationship is similar to the classical Arrhenius recycling, or show stop-start or rapid burst behavior. Such behaviour is
equation. again attributed to defects. Acceleration of the rate of interface advance
56 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

is thought to occur as a result of defects accumulating ahead of the inter- distribution of the nuclei; (iii) appropriate nucleation rate law and cor-
face, whilst rapid-bursts, deceleration or temporary stoppage are con- responding rate constant; (iv) rate of interface advance, which usually
sidered to be caused by the release of strain built up at the interface; can be assumed to be linear with time but may be anisotropic, being
the rapid bursts occurring due to generation of defects ahead of the in- characteristic for the individual crystal facets of the new phase; and
terface possibly initiating additional nucleation sites, and stoppages be- (v) correction for the ingestion of potential nucleation sites and the con-
cause of disturbance of the proper contact between the two forms at the tact/coalescence of the growing nuclei, which results in reduction/loss
interface – an example of the latter being stacking fault formation [74]. of reactive interface as the reaction proceeds to completion.
Work on single crystals of p-dichlorobenzene [61] has shown that The development of a generalised kinetic expression, therefore, is
the velocity of interface advance can also be dependent on the number not a trivial task. Consequently, a number of equations for specific
of cycles of transition that a crystal has undergone. With an increasing cases have been developed [80–84]. Of these, probably the most suc-
number of transitions the velocity was observed to decrease. This was cessful is that proposed by Avrami [80–82]. This takes the form
explained as being due to depletion of defects with each cycle, the de-  n
fects being consumed by the interface and then released at the crystal α ¼ 1− exp −kt ð27Þ
surface. The velocity was also influenced by the presence of other inter-
faces nearby; it increased greatly as the approach distance between the where k is the overall transformation rate constant, and 1 ≤ n ≤ 4 de-
interfaces became less than about 1.5 mm. Buildup of defects at the in- pending on the spatial distribution and the rate of nucleation.
terface and their transport into the region between was proposed as the When dealing with powdered samples comprising very small parti-
explanation. cle size, the Avrami model may not apply. The Avrami model assumes
Experimental determination of the kinetic parameters is best carried that potential nuclei of the new phase are randomly distributed
out on single crystals using direct microscopic observation. This, howev- throughout the bulk of the crystal, and that these nuclei, once activated,
er, may be either impossible because of a lack of single crystals, or inap- grow throughout the old phase until the transformation is complete.
propriate when, for example, kinetic parameters are required for a When a crystal is progressively subdivided (i.e. comminuted or pow-
polycrystalline or a powdered sample. When single crystals are not dered), there comes a stage when there will be only a few potential nu-
available, the fundamental kinetic parameters cannot be determined di- clei within each crystallite. As the rate of nucleation is considered to be
rectly. Experimentally, only the overall rate in terms of the fraction of proportional to the number of potential nuclei present, the transforma-
material transformed α(t,T) as a function of time and/or temperature tion will, therefore, become nucleation controlled. Consequently, the
is accessible. This α − time curve is typically sigmoid in appearance. overall kinetics may not comply with the Avrami model, since the trans-
The problem, therefore, is that of extraction of the fundamental param- formation due to any nucleation event will be constrained to the partic-
eters for a given sample from the overall rate data (α−time curve). The ular crystallite within the powder.
task is complicated further because the overall rate is strongly depen- Further, for powdered samples, the phase transformation often does
dent on the crystal morphology, crystal size and size distribution as not go to completion, with the maximum fraction (αmax) transformed
well as crystal perfection. Moreover, changes in shape, the degree of depending on the extent of superheating or supercooling [85,86]. The
poly-crystallinity and even fragmentation may result from polymorphic cause here is that some of the crystallites do not have potential nuclei
transitions themselves, thus complicating experimental reproducibility that can be activated at the particular temperature of study, and hence
even further. the transformation does not proceed to completion. Higher tempera-
The modelling of solid state kinetic data to a large degree is a prob- tures are able to activate the higher activation energy nuclei, bringing
lem in solid geometry. In developing an expression that describes the additional crystallites into the transformation. Consequently, αmax de-
overall rate of a polymorphic transition one needs to take into account: pends on the temperature, with higher temperatures leading to higher
(i) crystal morphology, size and size distribution; (ii) spatial values of αmax [86,87]. Clearly, for powder samples there is a need to

Fig. 7. Ising model simulation of magnetization. The volume elements or domains can be either spin up (red) or spin down (blue). The outcome of a particular domain is influenced by the
spins of the nearest neighbor domains and any external potentials such an external magnetic field or temperature. The simulation image is from Wolfgang Christian's (Davidson College)
3D checkerboard Decomposition Model code [89], and is reproduced here with permission.
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 57

incorporate a distribution of activation energies in the kinetic model function of time. Molecular dynamics simulation therefore serves as
[86,88]. an atomic microscope revealing the temporal changes in molecular or-
In summary, characterisation of the kinetics of polymorphic phase ganisation. Molecular simulations can be carried out at constant tem-
transformations, particularly transformations in powders, is complex. perature and pressure (NpT ensemble), hence mimicking conditions in
For powders, usually only a single overall activation energy can be esti- the laboratory. The implication is that the simulations (in principle)
mated, and this by itself, because it contains contributions from nucle- converge to lowest free energy state, akin to real systems. The temper-
ation events and phase growth, conveys no rigorous mechanistic ature and pressure, being the primary potentials of interest driving
insight. In addition, because of thermodynamic considerations the phase transitions, can be set as required.
data will invariably depart from ideal Arrhenius behaviour, especially Molecular simulations offer a number of methods and approaches to
at temperatures close to phase coexistence. Finally, the data may be ir- predicting and rationalizing phase stability and polymorphic transi-
reproducible from sample to sample, because of variation in crystallite tions. For a given molecular arrangement (configuration), the best accu-
size, shape, quality or history. racy in terms of structure, energies and interaction forces is achieved by
quantum chemistry. The downside of using such a high level of theory is
4. Molecular simulation of polymorphic phase transitions the immense computational demand, which means that quantum
chemistry calculations are restricted to small model systems and only
4.1. Molecular simulation methodology limited sampling of the manifold of different configurations. To assess
larger systems and extended statistics, it is necessary to employ the
Possibly the simplest simulation approach to studying phase transi- computationally inexpensive, lower-resolution approximation termed
tion phenomena is the Ising model and its generalization, the Potts molecular mechanics. This comprises empirical interaction potentials,
model [23]. The Ising model consists of discrete elements on a lattice commonly referred to as a force field [90,91], to describe the interaction
that represent magnetic dipole moments of spins, with each element forces. The most common source for defining force field parameters is
for instance representing a crystallite/domain (Fig. 7). The spin in each quantum mechanical calculations performed for a single molecule in
element can be either up or down and would be influenced by the the gas phase to characterize molecular flexibility and for dimers (or
spin in its neighbouring (interacting) elements and any external field. oligomers) of molecules to characterise non-bonded intermolecular
Phase transitions can be induced and their development tracked by forces, respectively. Replacing quantum mechanics by molecular me-
monitoring the spins as they undergo realignment, to yield the new chanics i.e. methodology based on force fields, enable simulations com-
phase consistent with the applied field. These models have contributed prising millions of molecules to be carried out for up to microseconds of
extensively to our generic understanding of phase transitions. simulation time.
Molecular simulation may be perceived to be a significant step up in The definition and parameterisation of a force-field reflects a coars-
complexity from the Ising/Potts models. The interacting volume ele- ening of quantum mechanical models, such as approximating diffuse
ments of these simple models are replaced by molecules which are electron density distribution within molecules by point charges placed
free to translate, rotate and flex, making molecular simulation an ‘off- on the atoms. While this reflects a reasonable simplification of molecu-
lattice’ technique. Further, the element-element interactions give way lar electrostatics, further potential energy terms are added to mimic
to intermolecular forces between the atoms, which determine their dy- electronic polarization. In many cases, the interplay of electrostatics,
namics and simulations track the trajectories of the molecules as a van-der-Waals interactions and atomic repulsion stemming from

Fig. 8. The molecular mechanics potential energy function comprising the van der Waals (term 1, Lennard-Jones) and coulombic (term 2) interactions, and the three valence terms, bond,
angle bending, and dihedral energy. The summations for van der Waals and coulombic terms indicate all pairwise interactions between atoms that are not either bonding or linked via a
bond angle. The Lennard Jones parameters εij and σij, partial charges qi and qj, and the force constants kb, ka, and kϕ are all atom-specific parameters that comprise the force field and are
inputs to the simulation.
58 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

Fig. 9. Molecular dynamics simulation. (1) The initial configuration comprising a set of atomic coordinates is specified and the atoms are assigned a set of random velocities that are
consistent with the temperature of the simulation. The total force resulting from the interaction with all neighbouring atoms is calculated. (2) The new position of each atom at a very
short time interval in the future (Δt ~ 10−15 s) is calculated using Newtonian mechanics (Eq. (29)) from a knowledge of the force on each atom, its current velocity and its mass. The
particles are then displaced to the new set coordinates. This process of evolving the system one step at a time into the future is then iterated millions of time to yield a trajectory of the
molecular behavour as a function of time.

overlapping electron density, is mimicked by three simple functions, sampling techniques that are able to simulate phase formation and tran-
using qi, qj, Aij and Bij as adjustable parameters. sitions that are beyond such time scales (see Section 4.3).
Other than the time scale limitation, the other two primary limita-
D E X  
^
H ≅ V r ij molecular mechanics tions are the number of molecules in the system, and the accuracy of
quantum mechanics
i; j the force field parameters that characterise the intermolecular interac-
X qi  q j Aij Bij tions. The molecular system size is typically of the order of 100,000 par-
¼ þ − ð28Þ
4π r ij r ij 12 r ij 6 ticles in a volume element of about a 10 × 10 × 10 nm3, which is well
i; j
away from cm-sized samples in the laboratory and Avagadro's number.
The simulations do however employ periodic boundaries, which give
where rij represents the separation distance between atoms i and j, and
periodicity to the system as benefits a crystal. The force fields are im-
qi are qj partial charges on atoms i and j. Similarly, molecular flexibility is
proving continually; the current level of accuracy has yielded some suc-
mimicked by assigning harmonic springs to characterise bonds and va-
cess in crystal structure prediction (given a 2-D molecular structure),
lence angles, and by employing periodic potential energy terms (often
which is a remarkable feat (see Section 5.1) [92–94].
based on cosine functions) to describe the energy profile of torsional de-
An alternative to molecular dynamics simulations is the method of
grees of freedom (rotations about bonds). The typical full potential en-
Monte Carlo (MC) simulation [23]. Instead of tracking a trajectory as a
ergy function is given in Fig. 8.
function of time, MC is implemented as a set of random moves, typically
In molecular dynamics simulations the time evolution of a given mo-
displacements of atoms and deformations of the simulation cell. New
lecular system is simulated by solving Newton's equations of motion for
molecular configurations are accepted/rejected on the basis of their dif-
the atomic interaction forces as derived from the potential energy rep-
ference in potential energy ΔE. The imposed probability for accepting a
resentation (Fig. 8). As the continual interaction of molecules prohibits
trail displacement or cell deformation move is:
an analytical solution, the molecular dynamics simulation technique
uses an approximation valid only for short time intervals Δt.
 
ΔE
pMC move
accept ¼ min 1; exp − ð30Þ
kB T
d  2 * * *
* d * r i ðt þ Δt Þ−2 r i ðt Þ þ r i ðt−Δt Þ
− * hH iqm or V mm ¼ F i ¼ mi  ri ≅ mi
dri dt
2
Δt 2 Simulations can be carried out in a variety of ensembles including
*
* * * Fi 2 NpT. The method yields equilibrium structures (hence phase stability)
→ r i ðt þ Δt Þ ≅ 2 r i ðt Þ− r i ðt−Δt Þ þ Δt ðΔt ¼ 0:1−2:0 fsÞ
mi and thermodynamic quantities. A particular advantage of MC is that
ð29Þ one can utilize non-physical moves, which can be useful for enhancing
ergodicity (overcoming barriers) within the system. Clearly, if there is
where t and mi represent time and atomic mass, respectively. The time no reference to time, MC contains no information on dynamical process-
evolution of the atomic positions is calculated from the second line of es. Kinetic Monte-Carlo however reintroduces time by attributing a
Eq. (29). Depending on the model system and the applied temperature, characteristic time scale to each type of Monte-Carlo move. It usually re-
the underlying approximation is limited to time step Δt of about 2 fem- quires experimental data or molecular dynamics simulations to assess
toseconds (10−15 s) or lower. Thus for assessing longer trajectories, Eq. the average time scales of the underlying processes (diffusion moves,
(29) is employed iteratively with updates of the positions and forces rotation, etc.)
being calculated after each time step to evolve the system forward in
time (Fig. 9). Given that each force calculation uses a certain amount 4.2. Simulating phase transitions
of cpu and evolves the system by only ~ 10−15 s, a typical simulation
on a high-performance computing facility can take days to carry out Among the first breakthroughs in simulating solid-solid phase trans-
the billions or trillions of force calculations to evolve the molecular sys- formations using molecular dynamics simulations was the pioneering
tem to nano- or micro-second time scale. This limited time scale (of the work of Parrinello and Rahman [95,96]. As indicated above, molecular
order of microseconds) is a major limitation of standard (brute force) simulations of crystals invariably employ periodic boundaries to
molecular dynamics simulation. However, there are now enhanced model a bulk system without surfaces. These periodic boundaries can
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 59

Rahman algorithm allows the simulation cell to change volume and

shape in response to the difference in the internal stress of the system
and the set pressure. The change in shape facilitates the transformation
of a lattice of the initial phase with a particular symmetry and space
group into another.
Where the energy barriers to transformation are much larger than
kBT, setting up the model system in a standard (brute force) MD simula-
tion and waiting for the transition to happen spontaneously during the
limited time scales (nano- to micro-seconds) of the simulation is unlike-
ly to yield success. An intuitive way to promote the kinetics of phase
transformations, with a view to observing the transformations in an
MD simulation, is to apply enhanced pressure, superheating or
supercooling. In the laboratory, moderate increases in any such driving
force can be employed to enable phase transformations to within time
scales of seconds to hours. In molecular dynamics simulations, the in-
crease of thermodynamic driving however needs to be dramatically
larger. First, the accessible time scales (nano- to micro-seconds) are at
least six orders of magnitudes lower. An additional issue is that the typ-
ical system size is considerably smaller. Nucleation events occurring at
some location within cm-sized samples in the laboratory, need to man-
ifest within nm-sized systems in the computer simulation. For this rea-
son, the state-of-the-art in molecular simulations for investigating
phase transformations refers to technical strategies that – in one way
or the other – coerce or direct the system to transform and then try to
correct for the biasing. Whilst the system is forced to transform from
one phase to another, the actual pathway and mechanism by which
the transformation proceeds is (largely – see discussion below) unbi-
ased, which represents the new information. Additionally, these tech-
niques yield the free energy barriers to transformation – the key
quantities that define the kinetics.
Directed simulations implement artificial forces or steered molecular
displacements to induce the process of interest. The simplest imple-
mentation is to define (on the basis of intuition or insight) a model re-
action coordinate RC and then constrain the system to move the along
this coordinate whilst allowing all other degrees of freedom to relax.
For a phase transition, the reaction coordinate would be the transforma-
tion pathway or mechanism by which the parent phase transforms to
the new phase, which may for instance take the form of an order param-
eter for investigating the nucleation step. A series of simulations are car-
ried out with the model structures constrained at particular points of
the reaction coordinates (particular values of the order parameter for
the case of nucleation) whilst monitoring the constraint forces. From
the constraint forces one can calculate the free energy profile, often
termed the potential of mean force, as a function of the model reaction
coordinate (Fig. 10) including the free energy barriers characterizing
the phase transformation process. A variation on this approach is the
method of umbrella sampling. Rather than simulating the system
constrained at various points on the reaction coordinate, the system is
instead restrained at the particular points on the reaction coordinate
using forces derived from artificial harmonic potentials U(RC) (Fig. 10).
The tendency of the system to depart from the restrained positions is
monitored yielding a probability histogram of occurrence h(RC). A
strong tendency for the system to depart from a particular restrained
position reflects an unfavourable (high energy) position on the free en-
ergy profile characterising the reaction coordinate. Using Boltzmann
statistics, the occurrence histogram may related to the corresponding
free energy profile, the potential-of-mean-force pmf(RC)
Fig. 10. Illustration of energy profile calculations from molecular simulation. Top: Potential
energy is scanned as a function of the model reaction coordinate. Middle: Boosting free
 
energy calculations from enhanced sampling of occurrence profiles using attractive hðRC Þ
potentials. Bottom: Escaping energy minima from metadynamics using locally repulsive pmf ðRC Þ ¼ kB T  ln −U ðRC Þ ð31Þ
potentials.
h0

In the case of low energy barriers, straight-forward molecular dy-

restrain and limit solid-solid phase transitions in crystals that involve namics or Monte-Carlo simulations can provide the full occurrence his-
significant changes in the lattice. The boundary effects become less sig- togram h(RC) without the use of artificial potentials U(RC). The quality of
nificant as the simulation cell (system size) is increased. The Parrinello- directed simulation methods depends critically on the choice of the
60 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

model reaction coordinate. For complex processes, defining an appro- The effectiveness of the metadynamics for simulating phase transi-
priate reaction coordinate may be challenging. tions is illustrated by a study from Parrinello's laboratory where they in-
Clearly, un-directed simulation methods are preferable, being free duced changes to the cell vectors of the crystals [99]. On this basis, free
from errors resulting from either a biased or an ill- pre-defined reaction energy (and free enthalpy) landscapes were sampled as a function of
coordinate. The increasingly popular metadynamics approach [97,98] the length of the cell vectors a, b, c and the respective angles α, β, γ –
accelerates the phase transformation process by filling-in and a total of 6 collective coordinates. Screening different unit cell shapes
swamping (on the fly) the full spectrum of local minima in the free en- and sizes in this way provided a route to polymorph screening. While
ergy surface, thus raising and leveling the energy surface and enabling computationally demanding, metadynamics simulations include entro-
the system to negotiate the energy barriers more readily, undirected. pic effects and convey significant further insights. First, metadynamics
The minima swamping is tracked, from which the free energy profile can explicitly be performed as a function of temperature thus providing
i.e. the barriers to transformation, may be reconstructed. Metadynamics temperature and pressure dependent polymorph search. Further, phase
thus utilises bias potentials akin to umbrella sampling, but relies on re- stability can be directly elucidated from respective minima in the free
pulsive potentials G(rC) to disfavor configurations that occur particular- energy landscape [100]. Finally, mechanistic insights into the transfor-
ly frequently. This has the great advantage in that the system moves mation process can be derived from the minimum energy path
away from configurations of the reaction coordinate that have already connecting the stable domains of the energy landscape. When moving
been explored, whilst allowing it to sample alternative configurations the system in 6-dimensional coordinate space (a,b,c and α, β, γ),
without directional forces (other than disfavoring the already known metadynamics shows the optimal molecular rearrangements for pro-
structures). A simulation started from a relaxed structure may thus es- viding such unit cell distortion [101]. It is important that the simulation
cape local energy minima by iteratively adding repulsive potentials cell is a supercell comprising multiples of the unit cell (which then has
until the effective barrier to reaching a new energy minimum can be periodic boundaries), which is more realistic as it can offer insight into
surmounted (Fig. 10). Convergence is reached after filling of all local en- phase interface formation, a prominent example being the observation
ergy minima, thus resulting in a flat histogram for the sampling of the of stacking faults as intermediates in the high-pressure transformation
reaction coordinate. The potential of mean force that reflects the free of MgSiO3 minerals [102]. A simulation comprising a single unit cell is
energy barriers can be reconstructed from the tracked, added repulsive akin to implementing a second order phase transition, that is collective
potentials. molecular displacements throughout the bulk crystal – whatever hap-
The appealing efficiency of metadynamics allows scanning of free pens in the single unit-cell simulation box is simultaneously reproduced
energy profiles as functions of several descriptors, named collective co- in all the periodic images.
ordinates. The significance is that the reaction coordinate i.e. the molec- Biasing is pretty much eliminated in transition path sampling (TPS)
ular pathway by which the transformation is observed to proceed, simulations [103,104]. Here we sample an ensemble of system trajecto-
becomes less biased by our choice of collective variable(s) as the num- ries between the starting state (in the present context, the parent crys-
ber of collective variables increases. The reaction coordinate is then de- tal) and the final state (the new daughter phase), and identify the low-
duced from the minimum energy path sampled for this set of variables. energy barrier (hence the most probable) pathways. This method can
The current state-of-the-art is to sample coordinate space in up to 6-di- be seen as a Monte-Carlo iteration of sampling dynamic transition path-
mensions; however, in many cases mechanistic insights of high quality ways. While the starting molecular configurations may be strongly bi-
can already be obtained from 2 to 3 adequately chosen descriptor ased (selected perhaps from a biased simulation), the nature of
variables. Monte-Carlo moves in trajectory space results in evolution to low-ener-
gy pathways, i.e. unbiased convergence to the most favorable transition
route. Given an initial configuration lying close to the transition state re-
gime (that is, the manifold of all possible transition states), molecular
dynamics simulations are performed in both directions of time i.e. for-
wards to generate the new phase, and backwards to yield the parent
crystal. This constitutes a single trajectory linking the two states. To ex-
plore further routes, configurations of the previous trajectory are slight-
ly modified and the process repeated. To confine trajectory space
sampling to transition routes only, pathways that do not connect the de-
sired endpoints (starting and final phase) of the (transformation) pro-
cess are discarded. Within a few tens of such iterations, the observed
pathways typically converge to the favorable transition mechanism
(Fig. 11). In other terms, the arbitrarily chosen transition state of the ini-
tial pathway is optimized in favor of the energetically most preferred
transition state. Moreover, by comparing the likelihood of trajectory
propagation to the desired endpoints it is possible to calculate forward
(r→) and backwards (r←) reaction rates. This can be done for a series
of intervals i of the descriptor variable, thus providing a reaction flux
profile. The free energy difference between these intervals may then
be estimated from the ratio of forward and backward rates using
Boltzmann statistics.
 
r→ ði→i þ 1Þ
Δpmf ði→i þ 1Þ ¼ −kB T  ln ð32Þ
r←ði þ 1→iÞ
Fig. 11. Illustration of transition path sampling between two phases A and B. The initial
route is derived from simple geometric interpolation (red dashed line). Subsequent Transition path sampling was first applied some 15 years ago to sim-
Monte-Carlo moves (indicated by numbers 0,1,2) in trajectory space lead to new ulate pressure-induced phase transitions [105]. While typically starting
pathways (blue curves). Upon sufficient iterations the pathways converge to the favored
mechanistic route, represented by a bundle of similar trajectories (green curves). For
from artificial transformation pathways (such as collective unit cell dis-
rate calculations, it is useful to devise pathways in intervals of milestones, each tortion), the iterative sampling of transformation pathways in TPS was
reflecting a transition interface (i) in trajectory space. found to quickly evolve in favor of low-cost routes. Upon convergence,
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 61

these pathways all reflected the same overall mechanism, but differed were predicted by one or more of the submissions. In terms of method-
in terms of where the nucleation started. ology, the success stories are the methodologies of Neumann, Leusen
The performance of either of these (un-)directed methods, and Kendrick [92], and of Price et al. [109]. The Neumann methodology
metadynamics or transition path sampling, depends on the system is implemented in the code Grace®, which is marketed by Avant-garde
and the actual process of interest. Based on our personal experience, Materials Simulation. Grace is being used in the pharmaceutical industry
metadynamics is suggested if the key descriptor variables appear safe for polymorph prediction. A recent contribution of Nyman and Day
to guess. The (more expensive) transition path sampling approach [110] indicates that entropy contribution (which was calculated using
(and its derivatives like transition interface sampling and aimless shoot- lattice dynamics) is important in ranking of polymorph stability.
ing) is appropriate in case of complex processes with larger likelihood of Clearly, in simulating phases and phase transitions, the accuracy of
showing unexpected mechanistic routes. Examples for both cases are the force field parameters is key. An accurate forcefield should be able
discussed in the following section. to preserve the crystalline phase structure at a given temperature and
pressure. A significant deviation (N5%) in the crystal structure and lat-
5. Molecular simulations of phase stability and transitions tice parameters implies an inadequate force field. Should the latter be
the case, one should consider an alternative set of forcefield parameters
5.1. Crystal structure prediction and polymorph screening or optimize the parameters to reproduce the known crystal structures
and their lattice energies [111–113].
Being able to predict the crystal structure of a chemical compound The blind tests and broader crystal structure prediction studies re-
given only its molecular structure has been a fundamental challenge veal the current accuracy of molecular mechanics force fields in this
in computational chemistry [106–108]. Having such an ability would context to be comparable or just beyond the small differences in lattices
enable prediction and rational design of a whole variety of solid forms energies of the polymorphic forms, circa 0.2–2 kJ/mol. Hence, the need
including polymorphs, solvates, co-crystals and salt forms. It would to resort to quantum chemistry methods to re-rank the final short list of
open the door to the more significant goal of predicting material prop- structures as obtained from force-fields.
erties such as mechanical properties, solubility, dissolution, and surface
and interfacial energies using a minimum of a-priori information. Such 5.2. Coexistence and phase diagrams
methodology could form the basis of an in-silico screen to identify mol-
ecules and associated solid forms with optimal properties without syn- Crystal structure prediction methods are based on identifying struc-
thesizing the pool of potential compounds. A particularly important tures with minimum potential energy and typically do not include en-
application is solid form screening to identify potential forms that may tropic effects. Such an approach therefore cannot predict the phase
show up either during processing or on storage - which is crucial not diagram of a material as a function of temperature. At phase coexis-
only for selecting favorable candidates, but also to avoid undesirable tence, the chemical potential of the phases concerned is identical. For
forms. a pure system, the chemical potential is given by
There has been considerable progress in both methods and proto-

cols, reflected with some successes, for crystal structure prediction of ∂G
μ¼ ð33Þ
organic molecules. Whilst there are a variety of proposed methods for ∂N p;T
predicting the crystal structure, most of them comprise 3 distinct steps:
which expresses the affinity of a molecule to be within the given system.
(i) Exploration of conformational flexibility of the molecule, usually Hence, free energy calculations are required to predict phase dia-
involving quantum chemistry codes to ascertain and characterise grams. There are three main approaches to predicting phase diagrams:
the barriers to rotation about bonds; lattice dynamics [114]; thermodynamic integration/perturbation using
(ii) Generation of tentative crystal packings for the molecule, the an Einstein crystal [115,116] coupled with Gibbs Duhem integration
common approaches being either random or pseudo-random [117]; and (iii) density of states methods [118,112]. Applications of
exploration of phase space (lattice parameters, molecule posi- the latter have so far been restricted to very simple systems.
tion and orientation and internal degrees of freedom) and simu- For solids, the number of configurations is limited and direction inte-
lated annealing. gration of the partition function becomes feasible, which forms the basis
(iii) Ranking of the trial crystal structures in terms of potential ener- for lattice dynamics. The method involves calculation of the vibrational
gy, the lowest energy structures being considered to be the and librational modes of motion of the molecules in the crystal lattice in
most plausible. The rigorous criterion, of course, is free energy the harmonic approximation using the intermolecular forces. The
but this is demanding to evaluate in terms of computing re- coupled vibrational motion of the molecules is transformed into inde-
source. The neglect of entropy implies the potential energy crite- pendent harmonic oscillators using normal mode analysis to yield a
rion to be a 0 K approximation. Initial screening of the large phonon density of states, which enables the calculation of the free ener-
number (1000 s) of trial structures is commonly carried out gy of the solid via the partition function. The harmonic approximation
using the computationally efficient (but lower accuracy) molec- appears to be valid up to about two-thirds of the melting temperature
ular mechanics forcefields to yield a short list of plausible struc- of the crystal. Applications of lattice dynamics to predict phase diagrams
tures. The latter are then ranked using quantum chemistry include MnO forms as a function of pressure [119], DL-norleucine as a
codes, typically density function theory (DFT) with dispersion in- function of temperature [111], and methane gas hydrate as a function
teraction corrections. of pressure [120].
The Einstein crystal approach coupled with Gibbs-Duhem is more el-
egant than lattice dynamics, and in principle more accurate at any tem-
The capability of the computational chemistry community to predict perature. It is however more demanding in terms of computing
crystal structures has been regularly tested by ‘blind tests’, which have resource. There are two stages: the determination of a coexistence
served to identify the truly effective approaches from the competing point on the phase diagram using the Einstein crystal approach, and
claims made in the literature. The 6th blind test was launched in 2014 (given a co-existence point) the tracing of the phase diagram by inte-
with the results being reported in 2016 [94]. Within this blind test, all grating the Classius-Clapyron equation. The Einstein crystal approach
of the experimental crystal structures of the five targets (rigid molecule; links reversibly an ideal (non-interacting) system to the real lattice.
partially flexible molecule; partially flexible salt form; multiple partially The free energy of the ideal Einstein crystal is calculated analytically
flexible molecules forming co-crystal or solvate; large flexible molecule) and the change in free energy from the ideal to the real crystal is
62 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

Fig. 12. Predicted (for TIP4P model of water, left; for SPC/E model of water, right) and experimentally determined (centre) phase diagrams of water. Only the stable phases of ice are
included in the diagrams.
(Reproduced with permission from reference [121] E. Sanz, C. Vega, J. L. F. Abascal, and L. G. MacDowell, Phys. Rev. Lett. 2004, 92, 255701; copyright 2004 American Physical Society).

determined by molecular simulation using thermodynamic integration, interesting to consider as to what exactly needs to occur in terms of
the total free energy for the crystal being given by the sum of the two. mechanism to give rise to macroscopic shape-change that can result
Notable applications include the prediction of the phase diagram of in the crystal doing external work? Uncoordinated molecular displace-
water [121] – a tour de force (see Fig. 12), and the melting point of ments lead to a cancelling-out of any net motion or net force that
NaCl [116]. could result in a specific deformation. A specific deformation of the ma-
terial must therefore result from synchronized, collective molecular dis-
5.3. Displacive/martensitic transformations placements. Collective molecular displacements imply that the energy
barrier for a particular (or a few select) transition pathways – the partic-
The seminal work of Parrinello and Rahman [95,96] – enhancement ular collective molecular displacements – is significantly lower than all
of the MD simulation method to enable the simulation cell to respond to other potential pathways and outcomes. A topotaxial relationship be-
internal shear stresses – opened the way to simulating crystal-crystal tween the initial and the final lattice would suggest that a few select
phase transitions. Their studies on simple ionic systems were immedi- mechanism pathways are favoured, and hence may represent a charac-
ately followed by what were then considered to be large scale simula- teristic feature of martensitic transitions. It is therefore clear that the na-
tions (typically 2048 molecules) on molecular crystals that included ture of the molecular displacements during a phase transition is at the
SF6 [122] and n-butane [123,124] using the massively parallel ICL Digital heart of understanding martensitic transformations.
Array Processor (DAP) computer. It is notable that these simulations Martensitic transformations, as indicated above (Section 3.1), have
were of phase transitions between plastic phases, involving relatively significant technological applications, though these are currently largely
small molecular displacements and low transition barriers i.e. displacive restricted to metals, alloys and ceramics, A particularly exciting prospect
transformations in Buerger's language. is the possibility of exploiting such transformations in molecular crys-
The primary macroscopic characteristic of martensitic transforma- tals to develop tiny machines for biomedical use, for instance, to target
tions is that they exhibit displacive shear upon transformation giving and deliver genes or drugs. An extraordinary example of one such
rise to shape-change in the material. The other often stated characteris- nature's machine is the T4 bacteriophage virus. This virus employs a
tics are a transition velocity being of the order of speed of sound, and martensitic transformation of its sheath protein to puncture the bacteri-
diffusionless, cooperative motion of the atoms/molecules. It is al membrane to enable it to deliver its DNA content [125,126].

Fig. 13. Illustration of a periodic simulation cell of DL-norleucine. Left: β (low-temperature) form characterised by hydrogen bonded layers (H-bond donors and acceptors are shown as
blue and red balls, respectively). Center-to-right: upon heating, adjacent layers that interact via comparably weak van-der-Waals forces can shift (white arrows) leading to the α' form
(right). Note that within the layers the molecules move in a quite concerted manner thus apparently skipping nucleation and growth.
(Reproduced with permission from reference [128] D. Zahn & J. Anwar, Chem Euro. J. 2011, 17, 11,186–11,192; copyright 2011 Wiley-VCH Verlag 11186 GmbH).
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 63

Converting chemical energy into mechanical activity is the essence of a the β→α' (α' being a variant of the experimentally observed α phase)
machine. A greater molecular-level understanding of thermosalient transition in DL-norleucine reveal a remarkable wholesale shifting (a
molecular crystals [15,57] (often referred to as ‘jumping crystals’) synchronized displacement of all the molecules within a bilayer) of
could set the foundation for developing tiny machines based on organic the molecular bilayers relative to each other (see Fig. 13). There ap-
molecules. peared to be no nucleation event or local hot spot at which the transfor-
For molecular crystals, mechanistic insight into martensitic-type mation is initiated. Effects of vacancy defects were also investigated but
transformations has come from simulations of phase transitions in crys- again the transformation proceeded by concerted molecular displace-
tals of DL-norleucine [111,127–130] and terephthalic acid [131]. The ments. An often quoted argument against collective motion in phase
three known forms of DL-norleucine consist of stacked, hydrogen- transitions is that the overall barrier is extensive (thermodynamic defi-
bonded bilayers, separated by a van der Waals surface. Simulations on nition), i.e. depends on the number of molecules involved in the

Fig. 14. Snapshots from the polymorphic transition of a terephalic acid crystal comprising 343 molecules shown along the [010] (left column) and [100] (right) direction, respectively.
Nucleation starts at the crystallites corner (snapshot at the top left) and the progress of the transformation is monitored by the change of b/c aspect ratio from 1.6 to 1.
(Reproduced with permission from reference [131] G. T. Beckham et al. J. Am. Chem. Soc., 2007, 129, 4714–4723; copyright 2007 American Chemical Society).
64 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

collective motion [27]. If the activation barrier for a single molecule is G‡, Terephthalic acid crystals apparently release mechanical energy
then for the collective motion of a layer of n molecules,the overall barri- during the form I to form II transformation that can cause the crystals
er would be nG‡. The implication is that the barrier will become insur- to jump, suggesting a martensitic-type transformation [132]. Beckham
mountable even for a relatively small number of molecules engaged in et al. [131] investigated this transition by molecular simulation using
collective motion. And yet the DL-norleucine simulations revealed con- the technique of aimless shooting, a variant of transition path sampling.
certed motion over hundreds of molecules, indeed along at least the full Notably, these simulations were carried out on nanocrystals of
10 nm dimension of the simulation cell. terephthalic acid rather than bulk crystals, which enabled the possibility
The first DL-norleucine simulations were carried out by standard of crystallite shape change and associated release of mechanical energy
brute force MD and employed what could be considered as rather to be explored (Fig. 14). The TPS simulations yielded the transformation
high superheating to induce the transformation. The apparent observed pathway, which was validated using committer probability analysis. In
collective motion may not reflect the mechanism closer to co-existence this case, the transformation reveals a nucleation and growth mecha-
which would be more relevant to real world applications. In view of this, nism, with nucleation being initiated at the surface. The nuclei were
the DL-norleucine transformation was further investigated using the elongated which appears to be reasonable given that the 1-d hydrogen
unbiased transition path sampling approach [129]. The TPS method binding between the carboxylic groups yields integrated chains. The
was combined with quenching of the system energy during the trajecto- transformation velocity along the predicted edge was estimated to be
ry simulations, which enabled the identification of the single low-barri- approximately 8 m/s equating to a kinetic energy of about 0.3 kcal/
er pathway. This pathway too was characterized by a wholesale shifting mol, which was considered sufficient to induce jumping of the crystal.
of the integrated bilayers, as observed in the brute force MD. These re- The study investigated two nanocrystal models of identical shape
sults are atypical in that such TPS simulations on other systems have in- but with different total number of terephtalic molecules (343 and
variably revealed a nucleation and growth mechanism. For DL- 216). For these two models, umbrella sampling simulations showed
norleucine, the process appears to involve a compression wave but free energy differences ΔG between the two polymorphs of 10 and 3
with a wavelength beyond the 10 nm scale of the simulation cell. The kcal/mol, respectively. Thus, by reducing crystal size from 343 to 216
transition energy barrier was not found to be extensive i.e. not depen- molecules, ΔG does not scale linearly, which would imply about 6 kcal/-
dent on the number of molecules involved in the collective motion. mol for the N = 216 molecules model. This discrepancy illustrates the
The argument for the energy barrier to collective motion being exten- importance of local nucleation and growth phenomena – here initiated
sive presumes that the shear at the interface occurs as a result of an ap- at the crystal surface – for characterizing polymorphic transitions.
plied stress. In this respect i.e. an externally applied stress, the argument
is expected to be valid. However, with thermally induced transforma- 5.4. Reconstructive transformations
tions, the thermal energy is distributed across all degrees of freedom,
and individual molecules would have an independent capacity to In contrast to displacive transformations, reconstructive transforma-
mount the local barrier. The analogy given in support of an extensive tions are considered to be characterised by strong interactions with lit-
barrier is the increasing difficulty one encounters in attempting to pull tle or no structural relationship between the parent and the new phase.
a carpet across a room as the carpet area increases. For a thermally-in- Whilst the strength of the interactions can affect kinetic stability and
duced transformation, staying with the carpet analogy, the thermal en- material hardness at ambient conditions, putting different materials
ergy acts as an anti-gravity potential (on each and every molecule) at on a reduced temperature scale (T/Tmelting point) reveals that classifying
every point on the carpet, enabling the carpet to glide into another pre- transformations on the basis of the strength of interactions is rather ar-
ferred location (another minimum). With the carpet, an alternative bitrary. The underlying physics of transformations in hard and soft ma-
low-cost approach to displacement is to introduce a local kink and terials may be similar.
then to push that from one end to another. This is akin to the compres- Hard and brittle compounds such as ionic crystals strongly disfavor
sion wave observed in the DL-norleucine transformation. lattice distortions and during a transformation tend to show sharp

Fig. 15. Reconstructive transition of KF involving a change in coordination number from 6 (B1) to 8 (B2). The change of coordination number is highlighted for the potassium ions using red
(6), yellow (7) and green (8), respectively. The mechanism involves the shifting of adjacent ion columns along [100, 010] or [001], resulting in the up/down shuffling of layers as highlighed
by plus and minus signs. The example shows two nuclei of the B2 phase which senses of (110) and (−110) layers mismatch. Upon contact of the two phase domains a grain boundary is
formed (here in form of a twinning plane).
(Reproduced with permission from reference [126] D. Zahn, O. Hochrein, S. Leoni, Physical Review B 2005, 72, 094106; copyright 2005 American Physical Society).
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 65

phase interfaces at which bonds are broken and reformed [59]. In other providing the basis for observing even subtle changes in the nucleation
terms, such reconstructive transformations provide comparably low mechanisms upon defect incorporation. Substituting gallium (Ga) by
strain within the phase domains at the cost of relatively high interface more polarizable indium (In) atoms leads to preferential nucleation
energy. For instance, reconstructive transitions were observed for the near the defects, whilst substitution by harder aluminium (Al) atoms
pressure-induced B1 − B2 transformation in alkali halides [60,133] was demonstrated to hinder B4 nucleation. Using N5% Ga → Al substitu-
(Fig. 15). For these compounds, transition-path sampling molecular dy- tion even altered phase stability in favor of the high-pressure form.
namics simulations revealed that nucleation starts with the displace- The second example is that of the pressure-induced B1(Fm3m)–
ment of a single ion, followed by shifting of an entire column. Phase B2(Pm3m) transformation in KCl. This transformation is rapid and revers-
growth was identified as the parallel shifting of adjacent columns in ible, occurring at 1.9 GPa but exhibits hysteresis that depends on purity
one direction, whilst anti-parallel shuffling of layers occurred in the per- and history of the sample. The effects of vacancy defects (both quantity
pendicular direction. The local nature of such nucleation events allowed and distribution) were investigated using molecular dynamics simulation
the observation of independent nucleation events separated by dis- at high pressures [137]. Nucleation always occurred at the vacancy defect
tances within the nm scale. Thus, sufficiently large simulation models sites and hysteresis was reduced with increase in density of defects. The
can capture coexisting nuclei and allow the investigation of phase do- transition pressure observed in simulations of perfect crystals was about
main coalescence, competing growth or merging by grain boundary for- 7 GPa which dropped to about 2.5 GPa on introduction of the vacancies,
mation. For a 14 × 14 × 14 supercell model of RbBr, the different senses just 0.6 GPa above the experimental transition pressure.
of column-wise shuffling along the (initially) equivalent a, b and c direc-
tion of the parent lattice lead to ‘mismatching’ nuclei that, upon contact 5.7. Phase stability and transitions at the nanoscale
of the phase fronts, yielded a poly-crystalline structure comprising
grains, even though the original structure was a coherent single crystal Going down the length scales gives rise to some new physics. As a
[60]. crystal decreases in size, its surface or interfacial free energy (the choice
It is interesting to compare the transformation of alkali halides with depends on whether the crystal is stand alone or surrounded by some
that of the somewhat softer and less brittle semiconductor CdSe [59]. medium) becomes significant relative to its bulk free energy and the
For the latter material, more extended interface regions separating the thermodynamics of the crystal are now determined by the interplay of
parent and high-pressure phases were observed. In other terms, more these two energies. For polymorphs, the molecular organisation at the
elastic compounds may accommodate phase interfaces in a structurally crystal surfaces is likely to be different and so would the surface ener-
more continuous manner and thus avoid harsh interface energy by gies. Therefore, depending on the crystal size, the surface (interfacial)
some degree of elastic smoothening. Such elastic deformations in turn energy can favour a particular phase that otherwise may be unstable
imply longer-ranged stress emitted by the phase fronts, and thus pre- in the bulk or even an entirely new phase, hence enabling phase stabil-
vent the formation of coexisting nuclei at short distance. Indeed, crystal ity to be modulated at the nanoscale.
fragmentation into grains could not be observed in the CdSe study and is Particle-size dependent phase stability was first recognized for inor-
instead believed to occur at by far longer length scales compared to the ganics, specifically metal oxides [138]. Of the various polymorphs of
alkali halides [134]. TiO2, rutile was found to be the stable phase in coarse materials, whilst
Another notable transformation is that of graphite to diamond as a the anatase and brookite were commonly observed in nanosized samples.
function of pressure. This has been reproduced by simulation using ab These observations were rationalized by surface energy measurements to
initio (brute force) molecular dynamics, although the required pressure yield phase diagrams comprising enthalpy as a function surface area (par-
was 90 GPa, which is 6 times larger than the experimental estimate of ticle size), which revealed cross overs in stability below about 200 nm
15 GPa [135]. This large excess pressure is necessary to make the trans- [139,140]. The oxides ZrO2 [141] and Al2O3 [142] also exhibit similar par-
formation occur within the relatively short timescales of simulation. In ticle size-dependent phase stability, as does carbon for which
contrast, a metadynamics simulation using a tight-binding potential nanodiamonds are more stable than graphite [143]. Molecular simula-
supplemented with a 2-body van der Waals interaction was able to in- tions have played a key role in predicting phase stability switch-overs
duce the transformation close to the experimental transformation pres- for these inorganic oxides and in rationalizing observed experimental
sure [101]. data, for example, see surface energy calculations on Al2O3 [144].
In contrast to inorganic oxides, there appears to be little or no exper-
5.5. Diffusive transitions imental data on phase stability of stand-alone organic nanocrystals. This
is not surprising since it is pretty much impossible to generate stable
Diffusive transitions do not show a rigorous mapping of atomic dis- stand-alone nanocrystals of organics, given that they are relatively soft
placements from one lattice to another, but instead involve a temporary compared with metal oxides. Simulation studies are also sparse. A nota-
degree of disorder. There are examples of polymorphic transitions in ble contribution is that of Hammond and colleagues [145], who investi-
molecular crystals induced by high temperature that follow the scheme gated the potential energy of the two known forms of L-glutamic acid
phase I → melting → phase II type transformations or are solvent-medi- nanocrystals as function of crystal size. The calculations suggest that
ated phase transitions, i.e. dissolution and re-crystallization. These ex- the metastable α-form is more stable than that β-form (which is the
ceed the scope of the present review. Interestingly, the temporary stable form in the bulk) for small particle sizes. Using potential energy
disorder may be limited to a specific constituent of a multinary solid, implies a 0 K model with the entropic effects being ignored. An en-
as for example observed for ion conductors [73]. hancement would be to carry out free energy calculations, which are
now feasible. As yet there is no experimental verification of the predict-
5.6. Effects of defects ed switch-over in phase stability for L-Glutamic acid polymorphs.
Whilst preparation of stand-alone, organic nanocrystals is challeng-
While mechanisms of polymorphic transitions are often discussed ing, nanocrystals can be crystallized under nanoscale confinement in
on the basis of perfect single crystals, we should bear in mind that nanoporous matrices [146, and references therein]. As would be expect-
such models ignore possibly quite important features of real crystals. ed, there are many examples where the confined nanocrystals exhibit
It is intuitive to expect defects to locally affect the energy barrier to nu- striking departure in phase stability from that in bulk crystals. Notable
cleation (and phase propagation). Simulation work exploring this issue examples include glycine [147], paracetomol [148], and 5-methyl-2-
is still rare. An illustrative example is the B1–B4 pressure-induced tran- [(2-nitrophenyl)- amino]-3-thiophenecarbonitrile (often referred to as
sition of GaN [136]. Transition path sampling simulations allowed the ROY due to the red, orange and yellow colors of its various polymorphs)
transition to be simulated without drastic over-pressurization, [149]. In confined systems, the important quantity is the interfacial
66 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

Fig. 16. Polymorph phase stability and solid-solid phase transformations in bulk (mm) and nm-sized crystals. While surface effects are negligible in large crystals, at the nanoscale the
surface energy can be significant and differences in surface structure between polymorphs can change the relative phase stability. Further, for nm-sized crystals, any local and/or
global lattice expansion due to nucleation or interface development and advance could give rise to significant increase in the surface (and associated surface energy) of the particle,
serving to increase the transformation barrier and leading to kinetic hindering.
(Reproduced with permission from reference [128] D. Zahn & J. Anwar, Chem Euro. J. 2011, 17, 11,186–11,192; copyright 2011 Wiley-VCH Verlag 11186 GmbH).

energy rather than surface energy, which would depend on the nature that phase transitions from one phase to another will invariably involve
of the confining surface and hence in principle could lend to being mod- some local or global lattice expansion, the free energy barriers associat-
ulated by design. ed with any required increase in surface could be significant. Conse-
A recent study by Belenguer et al. [150] investigated solvent effects quently, kinetics may have a stronger role in determining phase
on phase stability at the nanoscale on grinding, complimented by inter- stability of nanocrystals than previously recognised.
facial energy calculations using density functional theory. They were The role of the surface in modulating the barriers to phase transition
able to demonstrate phase stability switch-over as a function of size was recognised in simulations carried out on both bulk crystals
for the molecular systems studied. and nanocrystals of DL-norleucine [128]. In the simulations, the β → α'
At the nanoscale, other than the potential for switch-over of phase transformation in bulk crystals (with periodic boundaries) is observed
stability (thermodynamics), the barriers to phase transitions (kinetics) at about 390 K. For the nanocrystal (dimensions of approximately 12
can also be modulated. The transition state theory indicates that, of × 10 × 10 nm), the β-phase remained stable until 470 K, above which
the ensemble of phase transition pathways that are available to a sys- the nanocrystal transformed to another phase altogether. From the
tem, the ones explored by the system are those with the lowest free-en- above discussion, the lack of transformation could be either thermody-
ergy barrier. For nanocrystals (nanosystems in general), as the overall namics phase-stability switch-over or kinetic hindering. Thermodynam-
free energy is composed of both a surface (or interfacial) and a bulk ics could be discounted on the basis of the calculated potential energies
component, the important pathways will be those with the lowest sur- of the nanocrystals of the two phases (the difference in entropy for the
face free-energy barrier. The significance is that for nanocrystals, given surfaces was considered to be minimal), implying kinetic hindering

Fig. 17. Illustration of size-induced polymorphic transitions as observed from molecular simulations of DL-norleucine aggregation in a-polar solution. The least cost route to molecular
crystal formation involves several structural transformations leading from micelles to bilayers to staggered bilayers and further ordering to reach the final crystal structure.
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 67

behind the lack of transformation. The role of the surface in enhancing development and advances in diffraction methods. In the recent com-
barriers to transformation is illustrated schematically in Fig. 16. puting era, we have seen computer modelling employing simple phys-
Phase transformations at the nanoscale are also relevant at earliest ics-type models ‘synthesising’ a generic understanding of phase
stages of crystallization [128,151], where for example a system follows transitions. Subsequently, molecular simulations have drilled down to
Ostwalds rule of stages [152,153] with the phase with the lowest energy the molecular resolution and we are beginning to see the generic phys-
barrier nucleating first. Size-dependent transformations were recently ics perspective of phase transitions being modulated by the rich behav-
observed in simulations of nucleation in DL-norleucine [151]. The simu- iour of specific chemical species.
lations revealed alternative forms, micelles and bilayer segments, at the As we have shown, the macroscopic theoretical framework for the
initial stages, presumably because these forms provide more favorable kinetics of crystal-crystal transformations is now pretty much resolved.
interfacial energy. These structures become unstable at large size in fa- Consequently, the expectation is to see (experimental) thermal kinetics
vour of a structure that begins to resemble the bulk periodic structure. studies of both nucleation and interface advance where the analysis in-
The corresponding free energy profiles, illustrated by green and blue cludes the driving potential i.e. the extent of supercooling/superheating,
curves in Fig. 17, hence represent local sketches of the minimum energy rather than the use of the Arrhenius equation whose application may be
pathway to crystal formation. In-between, polymorphic transitions are inappropriate close to phase transition conditions. It is now feasible to
induced by size-dependent phase stability, but also confined by size-de- go beyond the use of a generalized kinetic equation such as the Avrami
pendent barriers. To contrast such structural evolution from crystal pre- equation [80–82]. We can, in principle, numerically model crystal-crys-
cipitation from solution, the latter is referred to as primary nucleation tal phase transformation kinetics using a coarse-grained, volume ele-
whilst the former are secondary, ternary etc. nucleation processes. In ment representation (not molecular) wherein the sample crystal or
principle, unlike the example of DL-norleucine shown here, the barrier crystals are sub-divided into small volume elements. The transforma-
to secondary nucleation (the solid-solid transformation) may be sub- tion of each of the volume elements and the overall progression of the
stantial, thus arresting the precipitate in a metastable form. transformation is evolved using transition rates derived from experi-
A common route to the production of relatively inert metastable mental nucleation rates and anisotropic rates of interface advance. The
polymorphs is to use alternative solvents. Whilst the basis for this is modelling will include specification of the crystal morphology and
not well expressed in the literature, molecular simulations suggest size, and for polycrystalline samples a crystallite size distribution, for
that choice of solvent selects the polymorph on the basis of favorable the particular material studied experimentally, thus matching its char-
surface-solvent interactions. However, this can only occur at the nucle- acteristics closely. For polycrystalline samples, this will enable the inclu-
ation stage as surface-driven thermodynamic stabilization ceases with sion of a distribution of activation energies for nucleation [86], rather
increasing crystal size. On the other hand, the barrier to secondary nu- than a minimum or mean activation energy for nucleation that forms
cleation increases with crystal size. Hence, successful arresting of meta- the basis of generalized kinetics equations. This numerical simulation
stable polymorphs depends on an interplay of thermodynamic and approach would be more accurate, being more specific to the particular
kinetic aspects for which rationalization molecular scale insights are of system and underpinned by experimentally-determined parameters.
crucial importance [154]. Moving onto molecular simulations, as the review illustrates, this
methodology has in the last two decades uncovered incomparable mo-
5.8. Kinetics from simulations lecular insights, enhancing considerably our mechanistic understanding
of phase transitions in crystals. Coupled to the insights is its powerful
The primary kinetic parameters are the rate of nucleation and rate of predictive capability e.g. crystal structure prediction and associated
interface advance at the selected temperature and pressure. Kinetics of phase stability, and prediction of thermodynamics quantities such as
nucleation are accessible from molecular simulation but studies to date lattice energies. How might the insights and the molecular-level mech-
have been on either nucleation from the melt or from solution. These anistic understanding contribute to chemical and pharmaceutical prod-
simulations employ directed or biased methods e.g. umbrella sampling uct development? The primary issue confronting industry is solid-state
or metadynamics, wherein nucleation is coerced to occur whilst track- phase stability, in particular being able to identify the stable polymorph,
ing the free energy [155–158]. This yields the free energy barrier to nu- and then, if chosen, to ensure its stability during manufacture and pro-
cleation ΔGN⁎, which is the key quantity in the nucleation rate equation, cessing, e.g. granulation and compaction of tablets, and on storage. De-
Eq. (19). Extending these methods to nucleation in a solid matrix (as in ciding on the stable polymorph need not solve all stability issues: in
a solid-solid phase transformation) is feasible. principle, tableting could induce a phase transformation to a high-pres-
In contrast to the rate of nucleation, the rate of interface advance is sure form, which may or may not revert back to the stable form, and/or
readily obtainable from molecular simulation but requires the use of lead to undesired product characteristics or behavior such as tablet lam-
large systems. The initial step would be to induce the transformation ination. Beyond identifying the stable polymorph is the even more am-
using a biased/directed method to yield a molecular configuration com- bitious goal of exploiting the advantages of metastable polymorphs,
prising the emerging new phase embedded in the original phase with such as higher solubility and dissolution rate, while ensuring their sta-
the phases separated by an interface. This configuration is then used bility. The mechanistic insights from molecular simulations should en-
as the initial state of a standard (brute force) molecular dynamics simu- able the development for rational strategies for kinetically stabilizing
lation. The rate of interface advance is tracked as a function of time as metastable forms. A common approach for kinetic stabilization of meta-
new layers deposit onto the growing new phase. Equating the deposi- stable forms in ceramics, metals and alloys is the use of dopants i.e. the
tion of the consecutive layers to length scale would yield the rate in inclusion of a trace impurities into the lattice [163]. These third-party
unit length per unit time. There are many examples of estimates of molecules inhibit the transformation. A similar approach in principle
rate of interface advance from molecular simulation, though most should work in molecular crystals. The knowledge of the key molecular
refer to crystal growth from either the melt or from solution [159– interactions characterizing nucleation and interface advance could help
162]. Examples for solid-state phase transformations include DL- to rationally design dopant molecules that inhibit nucleation and/or in-
norleucine [127], and terephthalic acid [131]. terface advance. Indeed, molecular simulation methods could be devel-
oped to screen such candidate molecules before experimentation in the
6. Future perspectives laboratory. The insightful design rules uncovered for crystal nucleation
[164] and growth [165] could possibly map on to solid-state transfor-
Our understanding of phase transitions in solids has developed via a mations. For instance, a potential dopant molecule should have both
number of distinct stages, beginning with empirical observations which an affinity for the lattice and also be small enough to be sterically ac-
were later enhanced with crystallographic structural data as a result of commodated within the lattice.
68 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

Finally, we comment more broadly on the application of molecular friendliness, and faster turnaround of simulation results. For all of these
simulation in industry, which is still very much in its infancy. We are aspects, we feel that current progress and future perspectives are indeed
acutely aware that whilst elegant simulation methods and approaches appealing: modelling packages are increasingly becoming more accurate,
abound, many are developed in the context of simple models and/or faster and easier to use. The software advances are flanked by the ongoing
exist in the form of bespoke computer codes that are not widely available improvement of CPUs and the increasing application of GPUs (graphics
or supported. This lack of capability for molecular systems, in particular processors) which have drastically reduced hardware costs whilst in-
pharmaceutical molecules, and the inaccessibility of the computer codes creasing process capacity. The latter issue is not a purely economic
has limited wider uptake of molecular simulation. However, as evidenced point, but also allows the scope of model complexity to increase continu-
by some of the literature reviewed here, a steadily increasing number of ously. Thus, we are able to simulate increasingly larger molecular systems,
the molecular systems simulated are directly connected to application- both increase in the complexity of individual molecules and also the num-
driven research in the chemical and pharmaceutical industry. Further, ber of molecules in the system. Moreover, large-scale computation
there are important advances occurring in methodology including its im- parallelization offers screening of a variety of compositions and process
plementation into robust codes. There is a focus on realistic models of conditions within reasonable time scales.
pharmaceutical and biological molecules (in contrast to simplistic generic All this makes molecular simulations an increasingly indispensible
models) and an interest into modelling systems and processes that ad- approach to pharmaceutical development. Within the next decade, we
dress challenges encountered not only in the wet lab, but also in pharma- feel that the role of molecular simulation in drug formulation will be-
ceutical industry. This review is intended to provide such bridging - so, come very similar to the role of computational modelling in drug design.
how can application-driven problems such as polymorph control and stabil- While fast and approximate models provide quick screening of candi-
ity in drug formulation benefit from molecular simulation? dates, more dedicated molecular simulations will elaborate in-depth
Our vision for the future is the routine use of molecular simulation in understanding to drive innovation beyond trial-and-error. Along these
the workflow of the chemical and pharmaceutical industry in real time, lines, the gaps between experimental and computational characteriza-
informing fewer and better experiments, facilitating more effective tion are decreasing and modern innovation will surely rely on both.
products made efficiently, and time reduction to market. Indeed, molec-
ular simulation is now an integral part of the predictive science strategy
of a number of major pharmaceutical and chemical companies. BASF, for References
example, have recently committed to installing a high-performance
[1] M. Murakami, K. Hirose, K. Kawamura, N. Sata, Y. Ohishi, Post-perovskite phase
supercomputing facility with stated applications being a better under- transition in MgSiO3, Science 304 (2004) 855–858.
standing of catalyst surfaces and faster design of new polymers. [2] F.J. Manjon, D. Errandonea, Pressure-induced structural phase transitions in mate-
rials and earth sciences, Phys. Status Solidi B Basic Solid State 246 (2009) 9–31.
The predictive capability of molecular simulation will drive screening
[3] R.E. Newnham, Phase transformations in smart materials, Acta Crystallogr. A 54
of polymorphic forms in tandem with experimental programmes. The (1998) 729–737.
routine ability to predict the crystal structure from a 2-D molecular struc- [4] G. Kostorz (Ed.), Phase Transformations in Materials, Wiley-VCH Verlag GmbH,
ture would open the door to predicting solid state properties such as dis- 2001.
[5] B. Fultz, Phase Transitions in Materials, 1st edition Cambridge University Press,
solution rate, solubility, crystal surface energy, and mechanical properties 2014.
without having to synthesise the molecule, setting the foundations for in- [6] A.M. Belcher, X.H. Wu, R.J. Christensen, P.K. Hansma, G.D. Stucky, D.E. Morse, Con-
silico drug development. This will also give rise to commercial opportuni- trol of crystal phase switching and orientation by soluble mollusc-shell proteins,
Nature 381 (1996) 56.
ty by providing a list of possible (energetically reasonably favorable) can- [7] S.V. Dorozhkin, M. Epple, Biological and medical significance of calcium phos-
didates, hence expanding the realm of targetable polymorph structures. phates, Angew. Chem. Int. Ed. 41 (2002) 3130.
Beyond crystalline phase stability and polymorphic transitions, molecular [8] Y.H. Tseng, C.Y. Mou, J.C.C. Chan, Solid-state NMR study of the transformation of
octacalcium phosphate to hydroxyapatite: a mechanistic model for central dark
simulation can be used to investigate a wide spectrum of systems in all line formation, J. Am. Chem. Soc. 128 (2006) 6909.
the main states of matter: solid, soft matter, liquid and solution, and gas [9] M. Ghosh, S. Banerjee, M.A.S. Khan, N. Sikdera, A.K. Sikder, Understanding metasta-
phase. Solution and liquid state systems are wholly accessible to standard ble phase transformation during crystallization of RDX, HMX and CL-20: experi-
mental and DFT studies, Phys. Chem. Chem. Phys. 18 (2016) 23554.
molecular simulation, as are soft matter systems such as simple lipid
[10] K.R. Morris, U.J. Griesser, C.J. Eckhardt, J.G. Stowell, Theoretical approaches to phys-
membranes (see for instance [166] that can enable the prediction of per- ical transformations of active pharmaceutical ingredients during manufacturing
meation across membranes [167]. Thus, how drugs interact with cyclo- processes, Adv. Drug Deliv. Rev. 48 (2001) 91–114.
[11] G.G.Z. Zhang, D. Law, E.A. Schmitt, Y.H. Qiu, Phase transformation considerations
dextrins or polymer nanoparticles in solution, micellar systems [168],
during process development and manufacture of solid oral dosage forms, Adv.
nanoemulsions, and membrane permeation can be investigated in a rou- Drug Deliv. Rev. 56 (2004) 371–390.
tine manner. There are still some technical challenges with respect to sim- [12] J.K. Haleblian, W. McCrone, Pharmaceutical applications of polymorphism, J.
ulating the more condensed phases (e.g. structural organisation in the Pharm. Sci. 58 (1969) 911–929.
[13] J.K. Haleblian, Characterization of habits and crystalline modification of solids and
essentially crystalline lipid phases of the skin), simulation of longer time their pharmaceutical applications, J. Pharm. Sci. 64 (1975) 1269–1288.
scale phenomena (e.g. precipitation of solids from solution), and the sim- [14] S.R. Vippagunta, H.G. Brittain, D.J.W. Grant, Crystalline solids, Adv. Drug Deliv. Rev.
ulation of large scale systems e.g. monoclonal antibodies. However, 48 (2001) 3–26.
[15] J. Bernstein, Polymorphism in Molecular Crystals, Oxford University Press, Oxford,
methods for resolving these issues are in place though need to be extend- U.K., 2002
ed to realistic systems and/or implemented in accessible and robust com- [16] C.R. Gardner, C.T. Walsh, O. Almarsson, Drugs as materials: valuing physical form in
puter codes. Indeed, it is now timely to consider applications of molecular drug discovery, Nat. Rev. Drug Discov. 3 (2004) 926–934.
[17] J. Bernstein, Polymorphism — a perspective, Cryst. Growth Des. 11 (2011)
simulation within the workflow of the industry. Molecular simulation ex- 632–650.
pertise needs to be integrated into the industrial project teams, where it [18] J. Bauer, S. Spanton, R. Henry, J. Quick, W. Dziki, W. Porter, J. Morris, Ritonavir crys-
can add value in real time (that is on scales of hours to weeks, rather tal forms — an extraordinary example of conformational polymorphism, Pharm.
Res. 18 (2001) 859–866.
than months and years).
[19] ANDAs: Pharmaceutical Solid Polymorphism, Chemistry, Manufacturing, and Con-
While the broad capability of molecular simulation demonstrates its trols Information. , FDA Guidance for Industry, July 2007.
readiness to not only supporting drug design but also its formulation, [20] International Conference on Harmonisation (ICH) Q6A, Specifications: Test Proce-
dures and Acceptance Criteria for new Drug Substances and new Drug Products:
we are still at the beginning of exploiting the full scope of this emerging
Chemical Substances, Step 4, October 1999.
technique. Simulation methodology not only needs further development [21] M.P. Allen, D.J. Tildesley, Computer Simulation Of Liquids, Oxford Science Publica-
(we attempted to describe the status-quo, but are well aware that meth- tions, 2006.
od development will probably never actually end), but also requires ma- [22] D. Frenkel, B. Smit, Understanding Molecular Simulation: From Algorithms to Ap-
plications, 2nd edition Academic Press, 2001.
turing in terms of ready modelling of realistic molecular systems, industry [23] D.P. Landau, K. Binder, A Guide to Monte Carlo Simulations in Statistical Physics,
standard computer codes implementing advanced methods, user- 3rd edition Cambridge University Press, 2009.
 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70 69

[24] J.W. Christian, The Theory of Transformations in Metals and Alloys, Part I, Equilib- [65] Y.V. Mnyukh, N.A. Panfilova, Polymorphic transitions in molecular crystals. 2.
rium and General Kinetic Theory, 2nd ed Pergamon, Oxford, 1975. Mechanism of molecular re-arrangement at the contact surface, J. Phys. Chem.
[25] C.N.R. Rao, K.J. Rao, Phase Transitions in Solids, McGraw Hill, N.Y, 1978. Solids 34 (1973) 159–170.
[26] P. Toledano, V. Dmitriev, Reconstructive Phase Transitions (in Crystals and Quasi- [66] Y.V. Mnyukh, Polymorphic transitions in crystals – kinetics, Mol. Cryst. Liq. Cryst.
crystals), World Scientific, Singapore, 1996. 52 (1979) 505–521.
[27] Y.V. Mnyukh, Fundamentals of Solid-State Phase Transitions, Ferromagnetism and [67] D. Turnbull, in: F. Seitz, D. Turnbull (Eds.), Solid State Physics, vol. 3, Academic
Ferroelectricity, AuthorHouse, Bloomington, IN, 2001. Press, N.Y, 1956.
[28] F.H. Herbstein, On the mechanism of some first-order enantiotropic solid-state [68] M. Volmer, A. Weber, Z. Phys. Chem. 119 (1925) 277.
phase transitions: from Simon through Ubbelohde to Mnyukh, Acta Crystallogr. [69] R. Becker, W. Doring, Kinetische behandlung der keimbildung in übersättigten
B62 (2006) 341–383. dämpfen, Ann. Phys. 24 (1935) 719.
[29] R.D. James, K.F. Hane, Martensitic transformations and shape memory materials, [70] N.H. Hartshorne, P.M. Swift, Studies in polymorphism. Part VII. The linear rate of
Acta Mater. 48 (2000) 197–222. polymorphic transformation of cubic to monoclinic carbon tetrabromide, J.
[30] M.T. Dove, Theory of displacive phase transitions in minerals, Am. Mineral. 82 Chem. Soc. (1955) 3705–3720.
(1997) 213–244. [71] N.H. Hartshorne, M. Thackray, Studies in polymorphism. Part VIII. The linear rate of
[31] A.R. Ubbelohde, Crystallography and the phase rule, Br. J. Appl. Phys. 7 (1956) transformation of β- into α-sulphur at low temperatures and at temperatures just
313–321. below the transition point, J. Chem. Soc. (1957) 2122.
[32] A.R. Ubbelohde, Thermal transformation in solids, Quart. Rev. Lond. 11 (1957) [72] R.S. Bradley, The energetics and statistical mechanics of the kinetics of solid–Nsolid
246–272. reactions, J. Phys. Chem. 60 (1956) 1347–1354.
[33] A.R. Ubbelohde, Transitions between condensed phases of matter (second Van't [73] S.E. Boulfelfel, D. Zahn, O. Hochrein, Y. Grin, S. Leoni, Low-dimensional sublattice
Hoff lecture), Kon. Ned. Akad. Wetens. Proc. 65 (B) (1962) 459–471. melting by pressure: superionic conduction in the phase interfaces of the fluo-
[34] A.R. Ubbelohde, Premonitory phenomena in phase transitions in solids, Z. Phys. rite-to-cotunnite transition of CaF2, Phys. Rev. B 74 (2006) 94106.
Chem. N. F. 37 (1963) 183–195. [74] J. Anwar, P. Barnes, S.M. Clark, E. Dooryhee, D. Hausermann, S.E. Tarling, The use of
[35] A.R. Ubbelohde, Molecular movements and phase transitions in solids. 1 – Thermo- fast powder diffraction methods to study transformations, J. Mater. Sci. Lett. 9
dynamic and structural aspects of phase transitions that are wholly or partly con- (1990) 436–439.
tinuous, J. Chim. Phys. 62 (1966) 33–42. [75] D.A. Young, Decomposition of Solids, Pergamon Press, Oxford, 1966.
[36] A.J. Cruz-Cabeza, J. Bernstein, Conformational polymorphism, Chem. Rev. 114 [76] Y. Kishi, M. Matsuoka, Phenomena and kinetics of solid state polymorphic transi-
(2014) 2170–2191. tion of caffeine, Cryst. Growth Des. 10 (2010) 2916–2920.
[37] A.Y. Lee, D. Erdemir, A.S. Myerson, Crystal polymorphism in chemical process de- [77] M.E. Villafuerte-Castrajon, A.R. West, Kinetics of polymorphic transitions in tetra-
velopment, Annu. Rev. Chem. Biomol. Eng. 2 (2011) 259–280. hedral structures. Part 1. - Experimental methods and the transition γ → β
[38] A.J. Cruz-Cabeza, S.M. Reutzel-Edens, Joel Bernstein, Facts and fictions about poly- Li2ZnSiO4, J. Chem. Soc. Faraday 1 75 (1979) 374–384.
morphism, Chem. Soc. Rev. 44 (2015) 8619. [78] M.E. Villafuerte-Castrajon, A.R. West, Kinetics of polymorphic transitions in tetra-
[39] S.R. Byrn, R.R. Pfeiffer, J.G. Stowell, Solid-State Chemistry of Drugs, SSCI, West Lafa- hedral structures. Part 2. - Temperature dependence of the transition β⇌γ
yette, IN, 1999. Li2ZnSiO4, J. Chem. Soc. Faraday 1 77 (1981) 2297–2307.
[40] H.G. Brittain, E.F. Fiese, Effects of Pharmaceutical Processing on Drug Polymorphs [79] D. Zahn, Nucleation mechanism and kinetics of the perovskite to post-perov-
and Solvates, in: H.G. Brittain (Ed.) Polymorphism in Pharmaceutical Solids, vol. skite transition of MgSiO 3 under extreme conditions, Chem. Phys. Lett. 573
95, Dekker, New York 1999, pp. 331–361. (2013) 5–7.
[41] H.G. Ibrahim, F. Pisano, A. Bruno, Polymorphism of phenylbutazone: properties and [80] M.J. Avrami, Kinetics of phase change. I General theory, Chem. Phys. 7 (1939) 1103.
compressional behavior of crystals, J. Pharm. Sci. 66 (1977) 669–673. [81] M.J. Avrami, Kinetics of phase change. II Transformation-time relations for random
[42] H.K. Chan, E. Doelker, Polymorphic transformation of some drugs under compres- distribution of nuclei, Chem. Phys. 8 (1940) 212.
sion, Drug Dev. Ind. Pharm. 11 (1985) 315–332. [82] M.J. Avrami, Granulation, phase change, and microstructure kinetics of phase
[43] F.P.A. Fabbiani, D.R. Allan, W.I.F. David, A.J. Davidson, A.R. Lennie, S. Parsons, C.R. change. III, Chem. Phys. 9 (1941) 177.
Pulham, J.E. Warren, High-pressure studies of pharmaceuticals: an exploration of [83] W.A. Johnson, R.F. Mehl, Reaction kinetics in processes of nucleation and growth,
the behavior of piracetam, Cryst. Growth Des. 7 (2007) 1115–1124. Metall. Mater. Trans. B AIME 135 (1939) 416–458.
[44] H.G. Brittain, Effects of mechanical processing on phase composition, J. Pharm. Sci. [84] B.V. Erofeyev, A generalized equation of chemical kinetics and its application in re-
91 (2002) 1573–1580. actions involving solids, C. R. Dokl. Akad. Sci. URSS 52 (1946) 511–514.
[45] A. Findlay, The Phase Rule and its Applications, 9th ed Dover, N.Y, 1951. [85] R.E. Cech, D. Turnbull, Hetrogeneous nucleation of the martensite transformation,
[46] P.W. Bridgman, A transition of silver oxide under pressure, Rec. Trav. Chim. 51 AIME Trans. 206 (1956) 124–132.
(1932) 627–632. [86] A.K. Sheridan, J. Anwar, Kinetics of the solid-state phase transformation of form β
[47] A.J. Dornell, W.A. McCollum, High Temp. Sci. 2 (1970) 331. to γ of sulfanilamide using time-resolved energy-dispersive X-ray diffraction,
[48] O. Lehmann, Ueber physikalische untersuchungen, Z. Krist. 1 (1877) 97–131. Chem. Mater. 8 (1996) 1042.
[49] O. Lehmann, in Molekularphysik, Ed. W. Engelmann, Leipzig, 1888. [87] P.T. Cardew, R.J. Davey, A.J. Ruddick, Kinetics of polymorphic solid-state transfor-
[50] A.R. Ehrenfest, Phase Transitions in the Usual and Generalized Sense, Classified Ac- mations, J. Chem. Soc. Faraday Trans. 2 Mol. Chem. Phys. 80 (1984) 659–668.
cording to the Singularities of the Thermodynamic Potential, Leiden Comm, Vol. [88] C.L. Magee, The kinetics of martensite formation in small particles, Metall. Trans. 2
Suppl. No. 75b, 1933. (1971) 2419–2430.
[51] M.J. Buerger, in: R. Smoluchowski, J.E. Mayer, W.A. Weyl (Eds.), Phase Transforma- [89] W. Christian, Computer program: Ising 3D checkerboard decomposition model,
tions in Solids, John Wiley and Son, N.Y, 1951. version 1.0WWW Document http://www.compadre.org/Repository/document/
[52] M.J. Buerger, Polymorphism and phase transformations, Fortschr. Mineral. 39 ServeFile.cfm?ID=13033&DocID=3602 2013.
(1961) 9–24. [90] L. Monticelli, D.P. Tieleman, Force fields for classical molecular dynamics, Methods
[53] M.J. Buerger, Phase transformations, Sov. Phys. Crystallogr. USSR 16 (1972) 959. in Molecular Biology, Springer Nature, Clifton, N.J. 2013, pp. 197–213.
[54] K. Otsuka, C.M. Wayman (Eds.), Shape Memory Materials, Cambridge University [91] A.D. Mackerell Jr., Empirical force fields for biological macromolecules: overview
Press, Cambridge, 1998. and issues, J. Comput. Chem. 25 (2004) 1584–1604.
[55] P.M. Kelly, L.R.F. Rose, The martensitic transformation in ceramics — its role in [92] M.A. Neumann, F.J.J. Leusen, J. Kendrick, A major advance in crystal structure pre-
transformation toughening, Prog. Mater. Sci. 47 (2002) 463–557. diction, Angew. Chem. Int. Ed. 47 (2008) 2427–2430.
[56] Y.V. Mnyukh, N.A. Panfilova, N.N. Petropavlov, N.S. Uchvatova, Polymorphic transi- [93] M.A. Neumann, Tailor-made force fields for crystal-structure prediction, J. Phys.
tions in molecular crystals. 3. Transitions exhibiting unusual behaviour, J. Phys. Chem. B 112 (2008) 9810–9829.
Chem. Solids 36 (1975) 127–144. [94] A.M. Reilly, et al., Report on the sixth blind test of organic crystal structure predic-
[57] P. Naumov, S. Chizhik, M.K. Panda, N.K. Nath, E. Boldyreva, Mechanically respon- tion methods, Acta Crystallogr. B Struct. Sci. Cryst. Eng. Mater. 72 (2016) 439–459.
sive molecular crystals, Chem. Rev. 115 (2015) 12440–12490. [95] M. Parrinello, A. Rahman, Crystal structure and pair potentials: a molecular-dy-
[58] Eyring, et al., J. Chem. Phys. 3 (1935) 786. namics study, Phys. Rev. Lett. 45 (1980) 1196–1199.
[59] D. Zahn, Modeling martensitic transformations in crystalline solids: validity and re- [96] M. Parrinello, A. Rahman, Polymorphic transitions in single crystals: a new molec-
design of geometric approaches, Z. Krist. 226 (2011) 568–575. ular dynamics method, J. Appl. Phys. 52 (1981) 7182.
[60] D. Zahn, H. Tlatlik, Atomistic in-situ investigation of the morphogenesis of grains [97] A. Laio, M. Parrinello, Escaping free energy minima, Proc. Natl. Acad. Sci. U. S. A. 99
during pressure-induced phase transitions: molecular dynamics simulations of (2002) 12562–12566.
the B1–B2 transformation of RbCl, Chem. Eur. J. 16 (2010) 13385–13389. [98] A. Barducci, M. Bonomi, M. Parrinello, Metadynamics, Wires Comput. Mol. Sci. 1
[61] Y.V. Mnyukh, A.I. Kitaigorodskiy, Y.G. Asadov, A study of the polymorphic transi- (2011) 826–843.
tion in monocrystalline para-dicholorobenzene, Zh. Eksp. Teor. Fiz. 48 (1965) 19; [99] R. Martonak, A. Liao, M. Parrinello, Predicting crystal structures: the Parrinello-
Sov. Phys. JETP 21 (1965) 12. Rahman method revisited, Phys. Rev. Lett. 90 (2003), 075503.
[62] Y.V. Mnyukh, N.I. Musaev, A.I. Kitaigorodskiy, Crystal growth upon polymorphous [100] A.B. Belonoshko, S. Arapan, R. Martonak, A. Rosengren, MgO phase diagram from
transformation in glutaric acid and hexachlorotane, Akad. Nauk. SSSR 174 (1967) first principles in a wide pressure-temperature range, Phys. Rev. B 81 (2010),
345; 054110.
Sov. Phys. Dokl. 12 (1967) 409. [101] R. Martonak, A. Laio, M. Bernasconi, C. Ceriani, P. Raiteri, F. Zipoli, M. Parrinello,
[63] Y.V. Mnyukh, Molecular mechanism of polymorphic transitions, Dokl. Akad. Nauk Simulation of structural phase transitions by metadynamics, Z. Kristallogr. Cryst.
SSSR 201 (1971) 573–576 (Soviet Physics -Doklady 16, 977–980). Mater. 220 (2005) 489–498.
[64] Y.V. Mnyukh, N.N. Petropavlov, Polymorphic transitions in molecular crystals. 1. [102] A.R. Oganov, R. Martonak, A. Laio, P. Raiteri, M. Parrinello, Anisotropy of Earth's D
Orientations of lattices and interfaces, J. Phys. Chem. Solids 33 (1972) layer and stacking faults in the MgSiO3 post-perovskite phase, Nature 438
2079–2087. (2005) 1142–1144.
70 J. Anwar, D. Zahn / Advanced Drug Delivery Reviews 117 (2017) 47–70

[103] P.G. Bolhuis, D. Chandler, C. Dellago, P.L. Geissler, Transition path sampling: throw- [135] S. Scandolo, M. Bernasconi, G.L. Chiarotti, P. Focher, E. Tosatti, Pressure-induced
ing ropes over rough mountain passes, in the dark, Annu. Rev. Phys. Chem. 53 transformation path of graphite to diamond, Phys. Rev. Lett. 74 (1995) 4015–4018.
(2002) 291–318. [136] S.E. Boulfelfel, D. Zahn, Y. Grin, S. Leoni, Walking the path from B4 to B1 type struc-
[104] B. Peters, Recent advances in transition path sampling: accurate reaction coordi- tures in GaN, Phys. Rev. Lett. 99 (2007) 125505.
nates, likelihood maximisation and diffusive barrier-crossing dynamics, Mol. [137] S. Devani, J. Anwar, A molecular dynamics simulation study of the effects of defects
Simul. 36 (2010) 1265–1281. on the transformation pressure of polymorphic phase transformations, J. Chem.
[105] D. Zahn, S. Leoni, Nucleation and growth in pressure-induced phase transitions Phys. 105 (1996) 3215–3218.
from molecular dynamics simulations: mechanism of the reconstructive transfor- [138] H.Z. Zhang, J.F. Banfield, Thermodynamic analysis of phase stability of nanocrystal-
mation of NaCl to the CsCl-type structure, Phys. Rev. Lett. 92 (2004) line titania, J. Mater. Chem. 8 (1998) 2073–2076.
250201–250204. [139] M.R. Ranade, A. Navrotsky, H.Z. Zhang, J.F. Banfield, S.H. Elder, A. Zaban, P.H. Borse,
[106] S.M. Woodley, R. Catlow, Crystal structure prediction from first principles, Nat. S.K. Kulkarni, G.S. Doran, H.J. Whitfield, Energetics of nanocrystalline TiO2, Proc.
Mater. 7 (2008) (2008) 937–946. Natl. Acad. Sci. U. S. A. 99 (2002) 6476–6481.
[107] A. Gavezzotti, Are crystal structures predictable? Acc. Chem. Res. 27 (1994) [140] A.A. Levchenko, G. Li, J. Boerio-Goates, B.F. Woodfield, A. Navrotsky, TiO2 stability
309–314. landscape: polymorphism, surface energy, and bound water energetics, Chem.
[108] T.S. Thakur, R. Dubey, G.R. Desiraju, Crystal structure prediction, Annu. Rev. Phys. Mater. 18 (2006) 6324–6332.
Chem. 66 (2015) 21–42. [141] M.W. Pitcher, S.V. Ushakov, A. Navrotsky, B.F. Woodfield, G. Li, J. Boerio-Goates,
[109] S.L. Price, M. Leslie, G.W.A. Welch, M. Habgood, L.S. Price, P.G. Karamertzanis, G.M. B.M. Tissue, Energy crossovers in nanocrystalline zirconia, J. Am. Ceram. Soc. 88
Day, Modelling organic crystal structures using distributed multipole and polariz- (2005) 160–167.
ability-based model intermolecular potentials, Phys. Chem. Chem. Phys. 12 (2010) [142] J.M. McHale, A. Auroux, A.J. Perrotta, A. Navrotsky, Surface energies and thermody-
8478–8490. namic phase stability in nanocrystalline aluminas, Science 277 (1997) 788–791.
[110] J. Nyman, G.M. Day, Static and lattice vibrational energy differences between poly- [143] C. Wang, J. Chen, G. Yang, N. Xu, Thermodynamic stability and ultrasmall-size effect
morphs, CrystEngComm 17 (2015) 5154–5165. of nanodiamonds, Angew. Chem. Int. Ed. 44 (2005) 7414–7418.
[111] S.C. Tuble, J. Anwar, J.D. Gale, An approach to developing a force field for molecular [144] S. Blonski, S.H. Garofalini, Molecular dynamics simulations of α-alumina and γ-alu-
simulation of martensitic phase transitions between phases with subtle differences mina surfaces, Surf. Sci. 295 (1993) 263.
in energy and structure, J. Am. Chem. Soc. 126 (2004) 396–405. [145] R.B. Hammond, K. Pencheva, K.J. Roberts, Simulation of energetic stability of facet-
[112] J. Chatchawalsaisin, J. Kendrick, S.C. Tuble, J. Anwar, An optimized force field for ted L-glutamic acid nanocrystalline clusters in relation to their polymorphic phase
crystalline phases of resorcinol, CrystEngComm 10 (2008) 437–445. stability as a function of crystal size, J. Phys. Chem. B 109 (2005) 19550–19552.
[113] H. de Waard, A. Amani, J. Kendrick, W.L.J. Hinrichs, H.W. Frijlink, J. Anwar, Evalua- [146] Q. Jiang, W.D. Ward, Crystallisation under nanoscale confinement, Chem. Soc. Rev.
tion and optimization of a force field for crystalline forms of mannitol and sorbitol, 43 (2014) 2066–2079.
J. Phys. Chem. B 114 (2010) 429–436. [147] B.D. Hamilton, M.A. Hillmyer, M.D. Ward, Glycine polymorphism in nanoscale crys-
[114] N.L. Allan, D. Gustavo, D. Barrera, J.A. Purton, C.E. Simsa, M.B. Taylor, Ionic solids at tallization chambers, Cryst. Growth Des. 8 (2008) 3368–3375.
elevated temperatures and/or high pressures: lattice dynamics, molecular dynam- [148] G.T. Rengarajan, D. Enke, M. Steinhart, M. Beiner, Size-dependent growth of poly-
ics, Monte Carlo and ab initio studies, Phys. Chem. Chem. Phys. 2 (2000) morphs in nanopores and Ostwald's step rule of stages, Phys. Chem. Chem. Phys.
1099–1111. 13 (2011) 21367–21374.
[115] D. Frenkel, A.J.C. Ladd, New Monte Carlo method to compute the free energy of ar- [149] J.M. Ha, J.H. Wolf, M.A. Hillmyer, M.D. Ward, Polymorph selectivity under
bitrary solids. Application to the fcc and hcp phases of hard spheres, J. Chem. Phys. nanoscopic confinement, J. Am. Chem. Soc. 126 (2004) 3382–3383.
81 (1984) 3188–3193. [150] A.M. Belenguer, G.I. Lampronti, A.J. Cruz-Cabeza, C.A. Hunter, J.K.M. Sanders, Solva-
[116] J. Anwar, D. Frenkel, M.G. Noro, Calculation of the melting point of NaCl by molec- tion and surface effects on polymorph stabilities at the nanoscale, Chem. Sci. 7
ular simulation, J. Chem. Phys. 118 (2003) 728–735. (2016) 6617.
[117] D.A. Kofke, Gibbs-Duhem integration: a new method for direct evaluation of phase [151] P. Ectors, P. Duchstein, D. Zahn, From oligomers towards a racemic crystal: molec-
coexistence by molecular simulation, Mol. Phys. 78 (1993) 1331–1336. ular simulation of DL-norleucine crystal nucleation from solution, CrystEngComm
[118] S. Singh, M. Chopra, J.J. de Pablo, Density of states–based molecular simulations, 17 (2015) 6884–6889.
Annu. Rev. Chem. Biomol. Eng. 3 (2012) 369–394. [152] W. Ostwald, Studien über die Bildung und Umwandlung fester Körper, Z. Phys.
[119] W.C. Mackrodt, E.-A. Williamson, D. Williams, N.L. Allan, A first-principles Chem. 22 (1897) 289–330.
Hartree-Fock description of MnO at high pressures, Philos. Mag. B 77 (1998) [153] N.I. Stranski, D. Totomanow, Rate of formation of (crystal) nuclei and the Ostwald
1063–1075. step rule, Z. Phys. Chem. 163 (1933) 399–408.
[120] R.E. Westacott, P.M. Rodger, Full-coordinate free-energy minimisation for complex [154] D. Zahn, Thermodynamics and kinetics of prenucleation clusters, classical and non-
molecular crystals: type I hydrates, Chem. Phys. Lett. 262 (1996) 47–51. classical nucleation, ChemPhysChem 16 (2015) 2069–2075.
[121] E. Sanz, C. Vega, J.L.F. Abascal, L.G. MacDowell, Phase diagram of water from com- [155] C. Valeriani, E. Sanz, D. Frenkel, Rate of homogeneous crystal nucleation in molten
puter simulation, Phys. Rev. Lett. 92 (2004) 255701. NaCl, J. Chem. Phys. 122 (2005) 194501.
[122] G.S. Pawley, G.W. Thomas, Computer simulation of the plastic-to-crystalline phase [156] A. Brukhno, J. Anwar, R. Davidchack, R. Handel, Challenges in molecular simulation
transition in SF6, Phys. Rev. Lett. 48 (1982) 410–413. of homogeneous ice nucleation, J. Phys. Condens. Matter 20 (2008) 1–17.
[123] K. Refson, S. Pawley, Molecular dynamics studies of the condensed phases of n-bu- [157] M. Salvalaglio, C. Perego, F. Giberti, M. Mazzottia, M. Parrinello, Molecular-dynam-
tane and their transitions I. Techniques and model results, Mol. Phys. 61 (1986) ics simulations of urea nucleation from aqueous solution, Proc. Natl. Acad. Sci. 112
669–692. (2014) E6–E14.
[124] K. Refson, S. Pawley, Molecular dynamics studies of the condensed phases of n-bu- [158] J. Anwar, D. Zahn, Uncovering molecular processes in crystal nucleation and
tane and their transitions II. The transition to the true plastic phase, Mol. Phys. 61 growth by using molecular simulation, Angew. Chem. Int. Ed. 50 (2011)
(1986) 693–709. 1996–2013.
[125] G.B. Olson, H. Hartman, Martensite and life: displacive transformation as biological [159] P. Ectors, W. Sae-Tang, J. Chatchawalsaisin, D. Zahn, J. Anwar, The molecular mech-
processes, J. Phys. Paris Colloq. 43 (1982) (C4-855 – C4-865). anism of α-resorcinol's asymmetric crystal growth from the melt, Cryst. Growth
[126] W. Falk, R.D. James, Elasticity theory for self-assembled protein lattices with appli- Des. 15 (2015) 4026–4031.
cation to the martensitic phase transition in bacteriophage T4 tail sheath, Phys. [160] P. Ectors, J. Anwar, D. Zahn, Two-step nucleation rather than self-poisoning: an un-
Rev. E 73 (2006), 011917. expected mechanism of asymmetrical molecular crystal growth, Cryst. Growth
[127] J. Anwar, S.C. Tuble, J. Kendrick, Concerted molecular displacements in a thermally- Des. 15 (2015) 5118–5123.
induced solid-state transformation in crystals of DL-norleucine, J. Am. Chem. Soc. [161] S. Piana, M. Reyhani, J.D. Gale, Simulating micrometre-scale crystal growth from
129 (2007) 2542–2547. solution, Nature 438 (2005) 70–73.
[128] D. Zahn, J. Anwar, Size-dependent phase stability of a molecular nanocrystal: a [162] M. Salvalaglio, T. Vetter, F. Giberti, M. Mazzotti, M. Parrinello, Uncovering molecu-
proxy for investigating the early stages of crystallization, Chem. Eur. J. 17 (2011) lar details of urea crystal growth in the presence of additives, J. Am. Chem. Soc. 134
11186–11192. (2012) 17221–17233.
[129] D. Zahn, J. Anwar, Collective displacements in a molecular crystal polymorphic [163] B. Djuricic, S. Pickering, P. Glaude, D. McGarry, P. Tambuyser, Thermal stability of
transformation, RSC Adv. 3 (2013) 12810–12815. transition phases in zirconia-doped alumina, J. Mater. Sci. 21 (1997) 589–601.
[130] J. van den Ende, B. Ensing, H. Cuppen, Energy barriers and mechanisms in solid- [164] J. Anwar, P.K. Boateng, R. Tamaki, S. Odedra, Mode of action and design rules for ad-
solid polymorphic transitions exhibiting cooperative motion, CrystEngComm 18 ditives that modulate crystal nucleation, Angew. Chem. Int. Ed. 48 (2009)
(2016) 4420–4430. 1596–1600.
[131] G.T. Beckham, B. Peters, C. Starbuck, N. Variankaval, B.L. Trout, Surface-mediated [165] M. Lahav, L. Leiserowitz, Tailor-made auxiliaries for the control of nucleation,
nucleation in the solid-state polymorph transformation of terephthalic acid, J. growth and dissolution of two- and three-dimensional crystals, J. Phys. D. Appl.
Am. Chem. Soc. 129 (2007) 4714–4723. Phys. 26 (1993) B22–B31.
[132] R.J. Davey, S.J. Maginn, S.J. Andrews, A.M. Buckley, D. Cottier, P. Dempsey, J.E. Rout, [166] R. Notman, J. Anwar, Breaching the skin barrier — insights from molecular simula-
D.R. Stanley, A. Taylor, Morphology and polymorphism in molecular crystals: tion of model membranes, Adv. Drug Deliv. Rev. 65 (2013) 237–250.
terephthalic acid, J. Chem. Soc. Faraday Trans. 90 (1994) 1003–1009. [167] D. Bemporad, J.W. Essex, C. Luttmann, Permeation of small molecules through a
[133] D. Zahn, O. Hochrein, S. Leoni, Multicenter multidomain B1–B2 pressure-induced lipid bilayer: a computer simulation study, J. Phys. Chem. B 108 (2004) 4875–4884.
reconstructive phase transition in potassium fluoride, Phys. Rev. B 72 (2005), [168] A. Amani, P. York, H. de Waard, J. Anwar, Molecular dynamics simulation of a poly-
094106. sorbate 80 micelle in water, Soft Matter 7 (2011) 2900–2908.
[134] D. Zahn, Y. Grin, S. Leoni, The mechanism of the pressure-induced wurtzite to
rocksalt transition of cadmium selenide, Phys. Rev. B 72 (2005) 064110–064117.