Plenary Lecture
Results Accuracy versus Computational Costs: Indications on the Performance of Different Levels of Theory from an Extensive Computational Study of Acylphloroglucinols
Professor Liliana Mammino
Department of Chemistry
University of Venda
South Africa
E-mail: sasdestria@yahoo.com
Abstract: Finding an optimal balance between results accuracy and computational costs is a crucial issue in the study of molecules, above all when the molecules are not small. The identification of patterns in the performance of different calculation methods may thus be of use when planning calculations of new molecules.
An extensive computational study of acylphloroglucinols’ molecules – ranging from 17 to more than 100 atoms – has provided ample opportunity for the comparison of the performance of different levels of theory (HF, DFT/B3LYP and MP2), also because the structural characteristics of these molecules enable the consideration of a variety of different features (including intramolecular hydrogen bonding), which increases the comparison significance. Moreover, calculations were performed in four different media (in vacuo and in three solvents), which adds the effects of the medium to the comparison domain. The results obtained for other classes of molecules that were interesting to compare with acylphloroglucinols (e.g., several hydroxybenzenes) further expand the comparison ground.
The lecture analyses the obtained results specifically from the point of view of the calculation methods’ performance, compares the ensuing inferences with those derivable from studies of other types of molecular systems (from literature), proposes some patterns that can be useful on selecting calculation methods for middle-size molecules, and outlines further research questions. The performance of HF – proving particularly good for the studied classes of molecules – is given specific attention, as the issue of HF performance is not yet completely clarified. The performance varies remarkably for different molecular systems and different types of investigation focuses (from the good one observed in this study to very poor ones), suggesting that deeper insight is needed to identify reliable patterns; on the other hand, its comparatively low cost makes HF convenient – at least in preliminary searches – when reasonably good performance can be expected, which enhances the importance of reliable patterns. Several of the other analysed issues also suggest the relevance of patterns-identification. Furthermore, what appeared to be a systematic overestimation of hydrogen bond strength by DFT/B3LYP suggests also something beyond pattern identification, as the observed systematicity appears to recommend the opportunity of the identification of scaling factors that could reliably relate the DFT/B3LYP estimations to actual/experimental values.
Brief biography of the speaker: Liliana Mammino was born in Pisa (Italy). She obtained a degree in chemistry at the University of Pisa in 1973 and a PhD in chemistry at Moscow State University in 1982. She has worked mostly in African institutions: National University of Somalia (1974-1975), University of Zambia (1988-1992), National University of Lesotho (1993-1996) and, since 1997, at the University of Venda (South Africa) where she is currently a professor in the department of chemistry. Her research interests comprise theoretical/computational chemistry (her field of specialization), with specific interest in the computational study of biologically active molecules, and chemical education, with specific interest in conceptual understanding and the roles of language and visualization. She has published articles in both areas, for a total of more than 80 articles in journals and more that 120 conference presentations. She is also author of a chemistry textbook, of chemistry resource materials and of a book on the language of science.