Are DFT level calculations the answer to real-world molecular systems? Robert J. Meier * DSM Research, P.O. Box 18, Geleen 6160 MD, The Netherlands Abstract This contribution is intended to initiate discussion on the performance of density functional theory based calcula- tions for molecular systems, in particular performance in chemical applications. Compared to a decade ago, it is now possible to study molecular species of a size that is becoming relevant for both the academic as well as for the industrial chemist. But how good are the numbers we generate using such, nowadays almost exclusively, DFT calculations? Despite successes, there seem important cases where current functionals reveal serious discrepancies. This will be illustrated and discuss accuracy and reliability. In addition, throughput of calculations is a real bottleneck. What we may need are quantum computational methods providing us with reliable (relative) energies, for systems up to at least a hundred atoms. Such calculations should be possible on a routine basis. We plead for a set of benchmark systems to evaluate newly introduced functionals, and mutually compare performance of existing methods. Ó 2002 Elsevier Science B.V. All rights reserved. Keywords: Density functional theory; Ab initio; Quantum methods; Molecular simulation 1. Introduction The key problem of quantum mechanics is the lack of exact solutions to the Schr€ odinger equa- tion. Therefore approximate solutions/methods were devised, comprising the Hartree–Fock and post-HF methods, density functional theory (DFT), and quantum Monte Carlo (QMC). It may be argued some of these methods are formally exact, however, practical application always in- volves approximations. In general I think it may be stated that accuracy roughly is in the follow- ing, descending, order: QMC > post-HF > DFT > HF. It is well-known that the problem with HF is the lack of account for electron correlation, re- sulting in a problem of quality of results in many systems of interest. Still, for certain classes of problems HF performs very well, including simple organic chemical systems. The problem with post- HF calculations is more of a practical kind rather than of a deficiency in methodology: they are very expensive, at least the more decent methods, with scaling behaviour like N 57 . The latter makes these methods unfeasible with respect to real-world molecular systems, and even prevents regular Computational Materials Science 27 (2003) 219–223 www.elsevier.com/locate/commatsci * Tel.: +31-46-476-1654; fax: +31-46-476-1200. E-mail address: r.meier@wxs.nl (R.J. Meier). 0927-0256/03/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved. doi:10.1016/S0927-0256(02)00448-2