PARTY
CONCLUSION
New Models for Intermolecular Repulsion a nd their Application to van d er W aals Helen H.Y. Tsui
10 Final Conclusion and Suggestions for Further Work________________________________ 189
10 Final Conclusion and Suggestions
for Further Work
The development of realistic model potentials is a highly challenging task, since the amount of time and resources required is tremendous. We have demonstrated the need to derive new model potentials for simulation. The overlap model has been shown to be capable of producing intermolecular atom- atom repulsion models for organic molecules from the monomer wavefunctions. These models together with an accurate DMA [1,2] for the electrostatic model with appropriate dispersion terms have produced simulation results that are comparable with experiments. We have demonstrated that the overlap model can be used to derive repulsion terms that absorb other minor contributions to the short-range energies, namely penetration and charge-transfer, can represent non-spherical atoms and that various methods can be used to derive the proportionality constant. This thesis has contributed towards a better understanding of the capabilities of the overlap model and the results are encouraging for the development of non-empirical potentials for organic molecules. We have tested the model in both crystal structures and van der Waals complexes, so it could be used for a wide range of simulation.
Further work is needed in the development of the overlap model, to enable its application for larger molecules. Since the IMPT calculations are both expensive and limited to small molecules, it would speed up the construction of the repulsion model, if we could assume a transferable value for the proportionality constant, K . In this thesis, the sensitivity of the repulsion model to variations in
K has been investigated (Chapter 4 and 7), and there is considerable sensitivity particular for the
New Models for Intermolecular Repulsion and their Application to van d er W aals Helen H.Y. Tsui
10 Final Conclusion and Suggestions for Further Work__________________________________ 190 phenol-water complex. At present, the range of values used for the proportionality constant in various studies [33,40,69,101,113] including this thesis is large, varying from 4 a.u. to 10 a.u. However, part of this variation in the proportionality constant is due to the variations in the methods used. The variations in the methods include the choice of orientations used for fitting (Chapter 4, 6, 7 and 9), the choice of basis sets (Chapter 7), the inclusion of charge-transfer and penetration energies (Chapter 7), and all other minor differences (Chapter 9). It would be ideal to test the overlap model with the same method and analysis for a wide range of small organic molecules, to build up sets of proportionality constants, and investigate whether an optimal value could give useful potentials. A more automated system containing the programs GMUL [116], and CADPAC [56] for IMPT along with input and analysis methods should speed up the process. Such an automated system would be useful for a wide range of modelling studies to derive particular type of empirical parameters, which are either not available or unsuitable for use.
The overlap model is the core of this thesis, but there are other scientific findings. The well studied phenol-water complex [5-23,25,181] provided a good validation for the overlap model, since the amount of experimental and theoretical published results allow comparison and checking. Our high-level supermolecule calculations provided a detailed study of the system, using minima obtained from simple model potentials as starting points. Although Fang's recent ab initio studies [10] use a higher level of theory, he did not report the existence of the 7i-bonded minimum in the ab initio
surface. Benoit reported that this rt-bonding is more significant for larger clusters o f water molecules interacting with phenol molecule [6,25]. This minimum appears to be too shallow to be a stable minimum once the zero-point energy is included. Our non-empirical model potential is fairly good, as demonstrated in its ability to produce energy and structure of the global minimum that are comparable with experiments in the DMC simulations. The one-particle density plots provided good visualisation of the possible hydrogen bonding, and would be of benefit to show the hydrogen bonding system for larger cluster [27].
Although our prediction for the blind test was unsuccessful [32], our model potential is still sufficiently realistic to produce a crystal structure that is very close to the experimental structure, should the correct conformation be used [33]. This study demonstrated that smaller systems that have similar bonding environments could be used for deriving the repulsion terms using the overlap model, by assuming transferability between the small system and the target molecule. This shows one of the alternative ways for deriving the repulsion model when the system is too large for IMPT calculations. This thesis has also contributed to the development of crystal structure prediction methods, in which the flexible 2-(2-phenylethenyl)-1,3,2-benzodioxaborole molecule emphasise that there is a balance between the inter- and intra- molecular forces needed for some systems.
Other problems that are encountered in crystal structure predictions challenges are problems of polymorphism and the need for anisotropic repulsion. For chlorothalonil, the C1...N and Cl...Cl interactions required anisotropic repulsion parameters, and the overlap model has been applied. The repulsion model constructed has reproduced the experimental crystal structure with only minor deviations. This study predicted the most stable polymorph to accuracy suitable for refinement. This has demonstrated that the overlap model could be applied to large systems. The new development of
New Models for intermolecular Repulsion and their Application to van der W aals Helen H.Y. Tsui
10 Final Conclusion and Suggestions for Further Work__________________________________191 the overlap model to derive anisotropic repulsion has been applied in the recent blind test (2001) [161]. The target molecule II has a bromine atom and anisotropy could be important. Therefore similar methods of applying anisotropy to the bromine atom were carried out as shown in our study of chlorothalonil. The global minimum found was a correct prediction. One of the major problems in crystal structure prediction is the generations of many low energy minima. Our model potential for chlorothalonil has produced many low energy predicted structures in our search, and by taking their mechanical stabilities into account [41-43], it helped to eliminate some of the possible structures. Further elimination was carried out after morphology calculation [43], where the growth rate and shape for the crystals are determined. By considering the kinetics of crystals, this would help to reduce the number of structures for experimentalists to study and search for the other polymorphic forms.
Besides the need of an accurate model potential, we have discovered that there is a need to develop new techniques for crystal structure prediction to improve the thermodynamics, such as estimating the harmonic zero-point energy for the crystals. This may be able to provide greater distinction between the predicted structures by comparison with their lattice energy. The method is currently being implemented into DMAREL [175,236], and in the recent blind test (2001), our group has attempted to use such a method to decide on the predicted structures chosen for submission. Unfortunately, on this occasion, it did not contribute any changes that would have influenced the order of choice. Nevertheless, it could be important for other systems, as demonstrated computationally in the gas phase DMC simulations [6,25,28-30] with considerations of zero-point energy. Perhaps Maddox's comment [237] was correct; 'One o f the continuing scandals in the physical sciences is that it remains in general impossible to predict the structure o f even the simplest crystalline solids from a knowledge o f their chemical composition'. Even though Maddox's statement was based on silicon crystals, it seems that this statement could also apply to organic crystal structures as well. Advances outlined above show progress in prediction using kinetic and thermodynamic models, with nucléation kinetics available from the CSD [145]. Hence improvements can be made on the procedures and approach of crystal structure predictions. Information can be extracted from the database, in the hope of finding a trend, a relationship, and knowledge to provide new theory to aid the development of crystal structure prediction [145,238].
There is still considerable scope for improvements to intermolecular potentials. One of the best approaches is to construct model potentials, where the different contributions from the intermolecular interactions can be identified and each contribution can be systematically improved. Non-empirical model potentials would be ideal; they can provide accurate description on the intermolecular forces far better than empirical models that assume transferability. We have demonstrated in this thesis that it would be possible to derive non-empirical repulsion models using the overlap model, and this should be studied further for the development of non-empirical repulsion models. Improving systematic potentials requires analogous studies to improve the dispersion model, and consideration of missing terms such as induction. It is true that the construction of non-empirical model potentials require vast amount of effort. However, it would be beneficial to obtain information from the study of smaller systems that could be applied to larger systems. Ultimately, the next step that needs to be taken in order to progress further in this field is to form collaborations between theoreticians and
New Models for Intermolecular Repulsion and their Application to van d er W aals Helen H.Y. Tsui
10 Final Conclusion and Suggestions for Further Work__________________________________192 experimentalists as shown in Chapter 9. Testing theory against theory (e.g. IMPT vs overlap model) is useful for very direct comparisons, but the true test is whether the total intermolecular potentials are useful in understanding experiments.
New Models for Intermolecular Repulsion and their Application to van der W aals Helen H.Y. Tsui