Polymorphic Phase Transitions in Carbamazepine and 10,11‐Dihydrocarbamazepine

Abstract Temperature‐induced phase transitions in carbamazepine (CBZ) and 10,11‐dihydrocarbamazepine (DHC) were studied by simultaneous differential scanning calorimetry–X‐ray diffraction in this work. The transitions generally involve a transitional melt phase which is quickly followed by recrystallisation. The expansions of the unit cell as a function of temperature could be quantified and allow us to determine a directional order of stability in relation to the lattice constants. Dihydrocarbamazepine form II undergoes a conversion to form I by a localised melt phase. Carbamazepine (CBZ) form IV converts to form I at 182 °C, again by a localised intermediate melt phase. CBZ form II converted to form I at 119 °C by a pathway that appears to have included some melting, and form III underwent a part melt‐recrystallisation and a part sublimation‐recrystallisation to form I.


Introduction
Pharmaceutical materials may exist in many different physical forms, each of whichw ill have unique physicochemical properties such as solubility,d issolution rate, stability,h ygroscopicity, mechanical strength, flowability and compressibility. [1] All of these properties will have as ignificant influence on the utility of the compound in am edicine. It is therefore essential that as much as possible is known about the physicalf orm of an active ingredient and how it will behaveu nder different conditions before it can be used in the clinic. However,s ystemic approaches to understanding polymorphic diversityi no rganic solids remain elusive.
The standard approach to studying polymorphic transitions is differential scanning calorimetry (DSC).T his provides information on the melting point and heat of fusion of polymorphs, and so relative thermodynamic stabilities, but does not give structurali nformation. For the latter,X -ray diffraction( XRD) is required. Variable-temperature XRD approaches have led to useful insights into physical form transitions in pharmaceutical materials, but standardl ab instruments take ac onsiderable time to record ad iffractionp attern (ca. 30 mins) and so it is difficultt os tudy transient or short-lived phases. Differences in sample sizes between XRD and DSC instruments can also affect the natureo ft he eventsobserved. [2] We recently demonstrated that asimple modification to astandardl ab DSCinstrument permits it to be mounted on as ynchrotron X-ray source, such that diffraction patterns can be obtained in as little as 2s, as the instrumentr ecordst hermal data in real-time. [3] Thisa llowed us to perform simultaneous DSC-XRD, and collectl arge amountso fX RD data of aq uality suitable for Rietveld refinement during heatinga ts tandard DSC rates (10 8Cmin À1 ). As a result, we could accurately explore and obtain new insights into phase transformationso fs everal pharmaceutical active ingredients, including sulfathiazole, glutaric acid and paracetamol. [3,4] Carbamazepine (CBZ) (Figure 1) is an anticonvulsant commonly used to treat epilepsy and trigeminal neuralgia. [5] It has extremelyp oors olubility in water [6] and is therefore an ideal candidate to screen for metastable polymorphs,s ince these will have higher solubility and more rapid dissolution rates than the thermodynamically most stable form. To date there have been five CBZ polymorphs reported. [7][8][9][10][11] CBZ is enantiotropic, and the most stable form at room temperature is form III, [10] while that at higher temperatures is form I. [9] The majority of the knownp olymorphs of CBZ (modifications I-IV) pack as dimers with the two molecules connected anti to each other by H-bonds between the oxygen and nitrogen atoms of the carbamoyl group. [9,11,12] The differences in structure stem mainly from the way in which these dimers are packed relative [a] Dr.A .E. Clout CBZ II undergoes as olid-solid phase transition  to form Ib etween 140 and 160 8C, and form III undergoes a  melt-recrystallisation to form Io ver the range 168-175 8C. [9] At the same heating rate form IV melts at 188 8Ca nd partially converts to form I, but slower heating results in am ore complete conversion. In all cases form Iw as shown to melt at 192-194 8C. [9] The discovery of CBZ Vw as achieved by templating crystal growth on the surface of ac rystal of form II of 10,11-dihydrocarbamazepine (DHC) (Figure 1). The latter is as tructural analogueo fC BZ. The difference between the two molecules lies in the presence or absence of ad ouble bond opposite the nitrogen of the azepine ring ( Figure 1). 10,11-dihydrocarbamazepine has four knownp olymorphs, [13][14][15][16] three of which display ac atemericH -bonded motif similart ot hat of CBZ V. In contrast,t he most recently discoveredf orm (IV) has ad imer motif similar to that seen in CBZ forms I-IV.T here has been little research into DHC, with no studies reported into the phase transitions between polymorphsord escribing their behaviour upon heating.
Here we report, for the first time, detailed studies into the polymorphism of DHC and CBZ, describing an umber of new insights into the transformations between the variousf orms of each.

Results and Discussion
Dihydrocarbamazepine Combined DSC-XRD data for as ample of DHC are shown in Figure 2. TheX RD data clearly showt hree distinct regions. In two of these there are numerous Bragg reflections, while the third is conspicuousb yitsl ack of reflections. TheD SC thermogram shows no events until the sample reached ca. 190 8C, where there is as mall endotherm. At the same temperature there is also ac hange in the positions of Bragg reflections in the XRD data. Followingt his, at around2 07 8C, there is am uch larger endothermic event coinciding with the total loss of Bragg reflectionsint he contour plot.
Rietveld refinement of an umber of DHC crystal structures from the Cambridge StructuralD atabase (CSD) against the pattern recorded for the initial starting materiala t5 6.3 8C( see Supporting Information Figure S1 and Ta ble S1) confirms that it was mostly polymorph II, with av ery small amount of polymorph Ia lso present. The refinementsf it with a R wp of 0.0349, and all subsequent refinements were carried out using the startingm odelsV ACTAU01 (form I) and VACTAU02 (form II) from the CSD.
From 56 to 190 8Ct here are no events in the DSC thermogram, ands ot here are no phase transitions occurring. However,t here are subtle changes in the diffraction data. Some, but not all, of the reflections relating to form II gradually shift to lower 2q angles. This is ac onsequence of the expansion of the unit cell as it is heated. Figure3 gives plots of the lattice constants as af unctiono ft emperature. The expansion in b is larger per degree increase in temperature than that in a and c, by factorso f9and 4, respectively.T his is ac onsequence of the alignment of the molecules relative to each other in the crystal and so the distances between them. The dominant intermolec-  ular force in the structure of polymorph II is hydrogenb onding between one hydrogen from the amide group and the oxygen of an adjacent molecule [14] ( Figure S2 andF igure 4a). This is in fact the only hydrogen bonding throughout the lattice, and while relativelyw eak for an H-bond with an H···O distance of 2.206 [14,17] is the strongest interactionh oldingt he crystal together.A na mide protono ft he second molecule forms as imilar H-bondw ith the oxygen of at hird adjacent DHC molecule, and so strings of molecules are formed in one direction. This corresponds exactly to the lattice constant a.
The second hydrogen of each NH 2 group interacts with the benzene ring centroid of the adjacent molecule ( Figure 4a). This enables interaction between the proton and the p-electrons, which appears to stabiliset he one-dimensional chains. [14] When viewed in the ac plane ( Figure 4a)i ti sc lear that these two interactions stabilise the crystal in the a direction.
When viewed in the bc plane ( Figure 4b)i tc an be seen that the chains are arranged in pseudo-sheets with the orientation of the carbamoyl group alternating with each new layer.B onding in the b direction is much weaker than in any other direction as there are no H-bonds and only interactions between p electrons of the aromatic rings.H arrisone tal. [14] have suggested that these interactionsm ust be relatively weak as the shortest centroid···centroid separation between adjacent molecules is 4.82 .B onding in the c axis is al ittle more interesting. The NÀH···O H-bonds between the carbamoyl groups and the supporting NÀH···p interactions form chains in the a dimension and offer some support in c (Figure 4a). Nevertheless, between the chains in this direction there are no hydrogen bonds and only Vand er Waals interactions similar to those in the b axis, but slightly stronger,w ith the shortest C···C separation being 3.651 .T his intermolecularb ondings tructure fits well with the unit cell expansion pattern observed, with the greatest expansion seen in b,followed by c and finally a.
At 190 8Ct he form II reflectionsf ade away and new ones grow in at different angles;as mall endotherm also appears in the thermogram.T hese two events indicate ap hase transition. Rietveld refinement of the diffraction pattern recorded at 202.5 8C( Ta ble S1 and Figure S3) demonstrates that the second phase is DHC I. Thus, the phase transition occurring is form II converting to form I. It is notable that the presence of crystalline materiali nt he beam was constant.T he resultso fi ntegration of the calculated patternsf or the two forms as af unction of temperature are presented in Figure 5. It is clear that the initial sample consisted almost entirely of form II. As the temperature rises the content of form II appearst oi ncrease, whilst the amount of form Ir emains relatively constant.O ne explanation for the apparent growth of form II may be that there was some amorphous material presenti nt he initial sample and that the energy supplied upon heatinga llowed sufficient molecular mobility for crystallisation. However,g lassy DHC would produce ab road shallow "hump" rather than sharpp eaks in a  There is av isible curve to the background but this persists throughout all of the data collected on all of the samples during this beamtime. More significantly,there are no exothermice vents in the DSC trace, ruling out any rapid crystallisation occurring. It could be that the increase in form II arises because of gradual crystallisation occurring during the heat (below the detection limit of the DSC) or it may be ac onsequenceo ft he expansiono fc rystals in the sample resulting in more material being lifted from the bulk into the passing beam.
At around 150 8Ct he amount of form Ib egins to increase. Following this, at 175 8C, the amount of form II beginst od ecrease and the rate of growth of form Ir ises sharply.T he growth and decay of these two crystal structures continue in an approximately linear fashion until the sample reaches 200 8Ca nd there is no more form Ip resent. The small endotherm in the calorimetric data coverst he same temperature range and the combination of these results indicates the occurrenceo faphase transition from form II to form I. The crossing of the two curvesa ta roundh alf the maximum quantity of either of the two species indicates the transition does not occur via ac omplete melt of the sample. Instead it is likely a solid-solid transition, or possibly the result of many smaller melt-recrystallisation events on ap articleb yp articleb asis. The absence of an exotherm in the thermograma nd the presence of the endotherm suggests the latter is perhaps more likely (althoughi tc ould be that exothermic and endothermic events are happening concurrently,r esulting in an et endotherm). It is likely that the presence of some form Ii nt he initial sample seededt he process. Analysis of the gradiento ft he two curves has shown the declineo ff orm II content to be À0.78 8C À1 and the growth of It ob e0 .81 8C À1 between 191 and 196 8C. The similarity betweent hese two numbers indicates that the conversion was asingle-step process.
Following the II!Ic onversion, crystalline material is only present over at emperature range of around1 0 8Cb efore a total loss of reflections is observed in the XRD data. The large endotherm in the DSC trace at 207 8Cc onfirms that the sample has melted. [18] Although polymorph Iw as not present for long, it was nevertheless possible to extract the lattice constants and cell volume from the refinementsa nd plot them as af unction of temperature ( Figure S4).
As with form II the unit cell expands in three dimensions as the temperature increases. Expansion in c per degree temperature rise is greater than that in a and b by factorso f3 .5 and 7 respectively.T he reasons for this are similart ot hose for form II. Form Ie xhibits the same molecular chains held together by weak hydrogenb ondingb etween an amide proton and the oxygen of an adjacent molecule [13,14] and stabilised by NÀ H···p interactions between the second amide proton and an adjacent benzene centroid, but instead of the a direction they propagate along b ( Figure S6). When viewedi nt he ac plane ( Figure S7) the differenceb etween the two structures is clear. In both, the chains are arranged in pseudos heets, but where in form II the orientation of the carbamoyl groupings alter-nates with each layer (Figure 4b), form Ip resents them in the same orientation. In both forms the layers are positioned so that the ring structures of each molecule are adjacent to ring structuresi na nother molecule, allowing for p···p interactions.
For both DHC Ia nd II, the smallest expansioni so bserved in the same direction as the propagation of the chains. This is to be expected as this is the direction in which the strongest intermolecular interactions are observed. However,t he greatest expansion and so the weakest interactions are seen in different relative directions. Form II expands most in b (equivalent to a in form I), effectively increasing the area of the pseudos heets, whereas form Ie xpandsm ost in c (equivalent to c in form II), increasing the space between the pseudos heets. Both of these dimensions are dominated by p···p interactions,w hich are weaker than H-bonds. [19,20] CBZ IV There has been some confusion overt he nomenclatureo fC BZ polymorphs in the literature; in this work, we use the numbering of the CSD (Table 1). DSC-XRD data for as ample of anhydrous CBZ supplied as form IV can be seen in Figure 6. The diffractiond ata are similart ot hose of DHC, with the occurrence of one crystalline to crystalline and one crystalline to liquid transition. Data collection began at 52 8Ca nd no major structural changes occurred until 182 8C, at which point there is a change in the 2q positions of the Bragg reflections. The second crystalline phase is presentu ntil the sample reaches around1 92 8C, at whichp oint all reflections disappear. The DSC trace shows as mall endotherm-exotherm event coinciding with the crystalline-crystalline transition. Immediately following this and superimposed upon it there is am uch larger endotherm, resulting from meltingo ft he material (evident from the total loss of diffractedi ntensity at the same temperature). The proximity of these eventsi ndicates that, at ah eating rate of 10 8Cmin À1 ,t he two transitions occur at very similar temperatures. Unfortunately,aconsequenceofthe overlapping thermale vents is that there can be no accurate quantification of the associated enthalpies.
Batch Rietveld refinement was carried out on all patterns recordeda nd selected patterns were examined individually.I nitially,s elected patterns were analysed to establish which of the five reported forms of CBZ [8][9][10][11]21] were present throughout the Refinement of the initial pattern recorded at 52 8Ca nd ap attern recorded at 189 8Care given in Figure S8. Evidently, the initial sample was entirely form IV,a nd the structuralr efinements fit with a R wp of 0.0441. Although CBZ is analogous to DHC with some of its polymorphsh aving similar structures to those of DHC already discussed, the structure of form IV (Table 1) is not one of these. Unsurprisingly,p lottingt he lattice parameters as af unction of temperature ( Figures S9 and S10) reveals that the unit cell of form IV expands upon heating, with expansion in b being aroundt wice that of a and c. CBZ IV packs as dimers, held by two H-bonds (1.86 )t hrough the carboxamide group with the two molecules anti to each other [10] (Figure S11). The oxygen also takes parti na nother interaction (2.28 )w ith ah ydrogen on the seven-membered ring of an adjacent molecule, and so chains of molecules are formed, which propagatea long c (Figure S12). These chains are held together along a and b by centroid-centroid interactions at ad istance of 3.809 .T he bondingp attern in a alternates between the two H-bonds forming the dimers and centroid-centroid interactions linking each dimer with the next ( Figure S13). Clearly the domination of H-bonding in c makes the interactions in this direction stronger than a,w hich has ac ombination of both H-bonds and p···p interactions. Axis a in turn has stronger intermolecular interactions than b,w hich exhibits almoste ntirely p···p interactions. As ar esult, expansion in the a axis is slightly larger (1.39 )p er 8Ct han that in the c axis and that in b is larger than a and c by factorso f2 .35 and 3.26 respectively.
Subsequentt ot his expansiono ft he unit cell the datas how as mall endotherm-exotherma nd as harp change in profile of the diffraction patterns. The initial small endotherm has an onset of 180 8Ca nd represents the meltingo fform IV.T he exotherm denotes ar ecrystallisation process.R ietveld refinement of the pattern recorded at 189 8Ca fter the profile change and the peak of the exotherm ( Figure S8) identifies the second phase as CBZ polymorph I, with ar esidual trace of IV remaining. Plotting the integratedt otal diffracted intensity for each pattern as af unction of temperature ( Figure 8) reveals that the lack of form Ib elow 170 8Ci sc onstant and the amount of form IV seems to grow as the temperature increases. This is likely due to thermale ffects. The total integrated intensity peaks at 152 8C, at which point it begins to decrease before a very sharp drop which flattenso ut at 192 8C. This signifies the total loss of all form IV content in the sample. At 172 8Cf orm I begins to grow; this is around the same point at which the decrease in form IV accelerates. The two changes considered together suggestthat form IV converts to form Iu pon heating.
The integrated data were converted to phase fractionsa nd plotteda safunction of time (Figure 8i nset). The curves cross at 0.5 and so the transition occurred without any wholesale  melt (i.e. melting and recrystallisation occur concomitantly rather than sequentially). An approximate linear fit of the integrated data has been carried out around the intersection point. It appearst hat the decline of form IV (À5.100 8C À1 ) occurs at as lightly faster rate than the evolution of form I (3.567 8C À1 ). This is unsurprising as meltingi sathermodynamic event and occurs very quickly while crystallisation is ak inetic event requiring molecularo rdering. The presence of the endotherm-exotherm in the DSC trace ( Figure 6) offers strong support to the theory of phase transformationv ia melting. The reason for their overlap is that the temperature at whicht he materialc rystallisest oIis reached by the instrument before all of form IV has melted. It appearst hat the transformationo f CBZ IV to CBZ Im ust occur by am elt-recrystallisation mechanism at these heating rates.T he possibility of as eparate melt and recrystallisation occurring at al ower heatingr ate cannot be ruled out.
Following the conversion of CBZ IV to Ii tc an be seen that the maximum form Ic ontentb arely reaches 70 %o ft he maximum form IV content prior to the conversion (Figure 8), as a result of incomplete recrystallisation to form I. At 10 8Cmin À1 the DSC reached the melting temperature of form I [22] (represented by the large endotherm,o nset 191 8Ci nF igure 6b) before the materialh ad all been able to crystallise. When the experiment was repeated at 2 8Cmin À1 (data not shown) the endotherm and exotherm were no better resolved but the resultant crystallisation exotherm had time to complete prior to the onseto fm eltinga nd the total content of form Ia fter the conversion had completed was similart ot hat of form IV.T his agrees with work by Grzesiak et al. [9] Presumably at the faster heatingr ate, the meltedf orm IV remains in the molten state until the end of the experiment.

CBZ II
DSC-XRDd ata for as ample of carbamazepine supplieda s form II are given in Figure 9. The DSC data show as mall exotherm-endotherm with an onset at 126 8C, and then al arge endotherm with two peaks and an onset at 190 8C. The former corresponds to ac omplete change in the diffraction pattern, while the latter is concurrent with the complete loss of all Bragg reflections and is in agreement with the reported melting pointo fC BZ form I. [22,23] The reasonf or the doublep eak is unclear,b ut it may be explained by the large sample size (12.1 mg). In all DSC-XRD experiments it was necessary to use as ample size significantly larger than the 5mgr ecommended, to ensure that there was alwayssample in the X-ray beam.
The initial change in diffraction pattern is accompanied by an exotherm-endotherm in the DSC trace, indicatingc rystallisation followed by melting. Following the transition some crystalline material remains, butaportion appearst oh ave melted. This suggestst hat the initial sample may have been am ixture of polymorphs rather than pure form II. Furthermore, form II is reported to undergo exothermic conversion to form Ib etween 140 and 160 8Ca taheating rate of 20 8Cmin À1 . [9] The onset of the exotherm in this experiment occurs at 119 8C, which can be accounted for by the slower heating rate (10 8Cmin À1 ), but the endotherm cannotb ea ttributed to the same conversion. Unfortunately,t he onset occurs whilst the preceding exotherm is ongoing and so cannot be accurately determined, but it must be above 119 8C. Additionally,d ue to the small enthalpy of the endotherm it is unclearw hether the signal subsequently returnst ob aselineo ri ft here is another exotherm prior to the melt.  Phase identification was carriedo ut on patterns collected at 42 and 178 8Cu sing the Rietveld method. Ar elatively poor fit was obtained at 42 8Cw hen considering only form II in the model,l eading to am ore detailed analysis in which the structures of all five reported polymorphs werei ntroduced. The conclusion of these refinementsw as that the initial sample was in fact am ixture of forms I( 3.5%), II (87.8 %) and III (8.7 %) ( Figure S14a). All further refinements were carriedo ut using startingm odelsf rom the CSD (summarised in Ta ble 1). Upon closer inspectiono ft he pattern recorded at 42 8C( Figure S14a) it appeared that all form Ic ontentw as attributable to as hift in the background and that there were no specific reflections assigned to that structure;a saresult, we determinedt hat the detection of form Iw as an artefact and it was excluded from refinementsa tl ow temperature. Refinement of the higher temperature pattern ( Figure S14b) revealed that following the phase transition almost all of the material had converted to form I( 96 %) with av ery small quantity of form II (2 %) and III (2 %) remaining. However,a sw ith form Ii nt he lowt emperature pattern, there were no characteristicr eflections of form III remainingi nt he observed data at 178 8Ca nd the 2% can again be accounted for by as light discrepancy in the background. Final unit cell data for the patterns discussed are presented in Table S3.
Batch refinements were carriedo ut and fits were obtained with R wp values between 0.0944 and 0.1401. Figures S15-S18 show plots of lattice constants as af unctiono ft emperature for the three forms of CBZp resent.A sw ith all previousm aterials, increasing the temperature of the sample causes expansion of the unit cell in all three dimensions for all three polymorphs. In all three structures the molecules pack as very similar dimers to those already discussed for CBZIV ( Figure 7a nd Figure S11), heldbyintermolecular H-bonds between the carboxamide groups.P olymorph II expands over 5times more per 8C in the a and b dimensions than it does in the c dimension. This is because the dimers in the structure arrange to form pseudo-layers in the ab plane. [8] These layers stack with translational symmetry that runs parallel to the c axis. It is clear from these data that the sum of the intermolecular bonding in the c axis is much stronger than in the other dimensions, resulting in the broadening of the layers as the sample is heated, whilst the chains of stacked dimers increaseinlength more slowly.
Expansion differences between the three lattice constantso f form III are far less pronounced than those of form II. The intermolecular bonding in this structure consists of the H-bonds forming dimers, two centroid···centroid interactions of slightly differentl engths, two CÀH···O interactions,aN ÀH···p contact and aC ÀH···p interaction. [24] This combination of intermolecular bondingi sc omplex and has clearly led to less H-bond dominance throughout the structure,h ence the more uniform expansion of the unit cell upon heating.
Intermolecular bondingi np olymorph Ii sa lso complex and has many similarities with form III.I nb oth forms the oxygen is involved in dimer formation as well as interactions with the vinylic carbonso fa djacent molecules. Am ore detailed description is given by Grzesiak and co-workers. [9] Using this description it was possible to examine the packing of the crystal using Mercury 3.8. When only the H-bonds which form the dimers are considered (Figure 10), it can be seen that three of the four pairs of bonds are orientated so that they offer support along b and only one of the pairs offers support to c. As these are the dominant intermoleculari nteractions in the crystal this explains the much larger extent of expansion in c.
As can be seen from the contourp lot (Figure 9), the presence of solid crystalline material in the path of the beam was constant from 42 to 196 8C. However,a st he initial sample was am ixture of two polymorphsi tc annotb ed etermined from this plot alone whether either of the components melted at any point. Figure1 1s hows the amounto fe ach polymorph present in the sample as af unctiono ft emperature. Below 100 8Ct he relative contributions of the three polymorphs is constant,w ith no form I, al ittle form III and the sample consisting mainly of form II. As with DHC and CBZ IV the total content of form II ap-  pears to increase with the temperature until it reaches 100 8C, at which point the first phase transition begins. This increase is probably ar esulto ft hermale xpansion. Form Ib egins to grow in at around1 00 8Ca nd the gradients of both curves become much steeper at 120 8C. This coincides with the onset of the exotherm in the DSC trace (119 8C) and suggests crystallisation from II to I. Al inear fit was carriedo ut at the straightest section of the curves around the point at which they cross (132-144 8C) and the rate of decay of form II (À3.794 8C À1 )a nd the rate of growth of form I( 3.164 8C À1 )a re similar, with form I growingalittle more slowly,a si st ob ee xpected when comparing crystallisation (kinetic) to melting( thermodynamic).
During this time form III appearst og row at an increasing rate until its content peaks at around 140 8Ca nd immediately begins to decay.I ti sd ifficult to say whether form II melts or converts directly to form I; however, the presence of the endotherm in the DSC data suggests am elt. The crossing of the phase fraction curves at approximately 0.5 meanst hat this is not aw holesale melt, but insteadl ocalised melting could be occurring. Ap lot of the sum of the calculated contributions of forms Ia nd II shows ad ip in the crystalline content in the beam between 120 and 160 8C, during the phase transition and prior to the waning of form III. This supports the hypothesis that there is some melting of form II involved in the transformation,b ut disagrees with the literature, which describes a solid-solid transition (based on ac ombination of thermomicroscopy and DSC carried out on separate samples). [9] Presumably there must be ap oint during the transition at which the structure of the materials its somewhere between those of the two crystalsa nd is disordered, unlesst he conversion is simultaneous and instantaneousf or all molecules in the crystal lattice; this may be what causest he drop in overall crystalline content at ca. 140 8C, and may not strictly be describeda sam elt.
Plottingt he sum of the contento ft he three polymorphs as af unctiono ft emperature ( Figure 11)i tc an be seen that the total crystalline content is lower following the phaset ransition. At horough investigation of the thermalr elationship between forms Ia nd III was carried out by Behme and Brooke [25] who concluded that the heating rate has as trong effect on the behaviour of form III. At 2 8Cmin À1 there is sufficient time allowed for the full conversion to form I, via as ublimation-condensation mechanism between 150 and 170 8C, with no apparent melting. However, once increased to 10 8Cmin À1 they state that "the endotherm recorded in the range 165-175 8Cr eflects severalt hermale vents", representing the combination of some conversion to form Iv ia sublimation and some melting. At all rates studied in the range 2-40 8Cmin À1 the endotherm was followed by an exothermc orresponding to the crystallisation of form If rom the melt. This melt-recrystallisation pathway is well documented in the literature. [9,22,23,[25][26][27] It is likely that the drop in overall diffracted intensity is ar esult of some form III subliming and movingo ut of the incident beam whilst in the gas phase and condensing in another area of the sample or being lost from the pan, which was left open for these experiments.Hence, it seems that there are two distinct phase transitions occurring simultaneously by three mechanisms. Form II is undergoing melt-recrystallisation to form Io namicroscopic particle by particle basisand form IIIisundergoingpartsublimation-recrystallisation andpartmelt-recrystallisation to form,I.
The conversion data presented here are consistent with the literature, except that the conversion of III occursa talower temperature than expected.T hism ay be due to the presence of form II destabilising it. Due to the overlapping events in the thermograma nd the open pan it is impossible to quantify the enthalpyo fc onversion for either the form III!Io rI I!It ransitions. Consequently,t he amount of form III lost by sublimation cannotb ed etermined via enthalpic calculations.H owever,t he diffraction data were used to calculate ar ough fraction of III lost by calculating the differenceb etween the maximum total crystal content prior to the phase transition and the equivalent immediately following it.T his was compared to the maximum area under the curve for form III and the resulting Figure suggests that only 30 %r ecrystallised to form If ollowing the melt and 70 %w as lost from the beam. This may be partially prevented in future work by using hermetically sealed DSC pans; however,t his would not prevent recrystallisation from the vapour phase in areas of the sample not interrogatedb yt he incident beam.

Conclusions
Te mperature-induced phase transitions in carbamazepine (CBZ) and 10,11-dihydrocarbamazepine (DHC) were studied by simultaneous DSC-XRD in this work. The transitions generally involve at ransitional melt phase. This melt is so quickly followed by recrystallisation that regions of the sample melt and recrystallise before other regions begin to melt. Consequently,t he presence of crystalline materiali nt he sample is continuous, leadingt ot he assumption of solid-solid transitions in the previous literature. Batch Rietveld refinements allowed the expansion of the unit cell to be quantified as af unctiono ft emperature, and these data in conjunction with structural information allow us to determine ad irectional order of stabilityi nr elation to the lattice constants. The dimensions whichc ontain stronger intermolecular bonding( e.g. H-bonds)h ave shown smaller expansion per 8Ct han those with weaker interactions (e.g. p···p). DHC II was shown to undergo ac onversion to formI by what appearst ob eal ocalised meltp hase. CBZ IV converted to form Ia t1 82 8C, again by al ocalised intermediate melt phase. As ample procured as pure CBZ II was in fact found to contain as mall amount of form III. Form II converted to form I at 119 8Cb yap athway that appearst oh ave included some melting, and form III underwent what seems to be ap art melt-recrystallisation and ap art sublimation-recrystallisation to form, I.

DSC-XRD
DSC Measurements were performed with modified TA 2010 or Q20 instruments (TAI nstruments LLC), with holes drilled in the furnace to permit the passage of the X-ray beam as detailed in our previous study. [3] Calibration was performed with ac ertified indium standard according to the manufacturer's instructions. Samples of all materials (5-20 mg) were held in Tzero aluminium pans and heated at 10 8Cmin À1 from ambient to 220 8C. Experiments were performed on Beamline I12 of the Diamond Light Source using a 0.5 0.5 mm beam of monochromated X-rays at 52.4 keV (0.236 ). AThales Pixium RF4343 detector,c alibrated with aC eO 2 standard, was located 2.4 ma way from the sample. Diffraction patterns were recorded every six seconds (data were collected for 4s with a2s pause between collections).

Data analysis
The DAWN Science Workbench was first used to convert the 2D data into 1D diffraction patterns. [28] Contour plots of the raw XRD data were then plotted using OriginPro 2016. Selected patterns were analysed using the Rietveld method implemented within the TOPAS-Academic suite of programmes, [29] in order to obtain realistic values for the unit cell parameters at elevated temperatures. Backgrounds were fitted using as hifted Chebyshev polynomial of the first kind with between 6a nd 15 terms. Lattice parameters and peak shape parameters were refined. In cases where more than one phase was present, the peak shapes for each phase were constrained to be the same and the phase fraction was refined. The models used came from the CCDC (details are given above). The atom positions were not refined. Atom displacement parameters, U iso were set to be 0.15 2 in each phase. Once starting parameters were obtained batch refinements were performed on all datasets collected. No zero point was refined as entire diffraction patterns were collected using a2Darea detector.