Accurate Atomic and Molecular Calculations without
Gradient Corrections: Scaled SVWNV Density Functional
Kevin E. Riley, Edward N. Brothers, Kenneth B. Ayers, and Kenneth M. Merz*
Department of Chemistry, The PennsylVania State UniVersity, 104 Chemistry Building,
UniVersity Park, PennsylVania 16802
Received January 12, 2005
Abstract: The local spin density approximation (LSDA) approximation was one of the first density
functional theory (DFT) methods employed to calculate atomic and molecular properties. As
newer, more sophisticated methods, such as BLYP and B3LYP, were developed, the LSDA
approximation has grown less popular for molecular systems. In this paper we revisit the LSDA
method and investigate a simple way to improve the results that can be obtained using this
approximation. By scaling the contribution of the local correlation to the SVWNV functional,
improved results can be obtained for heats of formation, ionization potentials, electron affinities,
bond angles, bond distances, vibrational frequencies, conformational energies, interaction
energies, and barrier heights. The results of our studies show that scaling the SVWNV functional
yields heats of formations with average unsigned errors up to about nine times smaller than
those of the standard SVWNV functional. The decreases in the errors of other properties studied
in this work were not as dramatic as those of the heat of formation but were, in most cases,
significant. There is a notable time saving in this density only functional. For a 9-alanine system
SVWNV is 55% faster than B3LYP and 40% faster than BLYP at a 3-21G* basis set. Based on
our observations we propose an improved SVWNV density functional that is suitable for the
study of molecular systems at a fraction of the cost of more sophisticated DFT methods, which
also produces reasonable accuracy at small basis sets. One type of application for which the
improved SVWNV functional would be extremely well suited is QM/QM methods where a fairly
inexpensive method is needed for the larger part of a system that is treated at a lower level of
theory.
Introduction
As the speed of modern computers increases while their cost
decreases it becomes possible to consider the use of ab initio
methods to calculate properties of very large molecules, such
as biomolecules. These types of calculations promise to
greatly increase our understanding of the structure and
function of large molecular systems.
1
Although a few efforts
have already been made to make such large scale calcula-
tions,
2-6
it is not yet a common practice due to the great
computational expense.
1
The first step one would take when
making ab initio calculation on very large systems would
be to use a relatively inexpensive density functional theory
(DFT) method,
7,8
such as LSDA, along with a small split
valence basis set with a low degree of contraction, such as
3-21G*.
9-11
The problem with such an approach is that the
LSDA method is known to give inaccurate results when used
to calculate such properties as the heat of formation.
12,13
In
this study we use an empirical technique to scale the
correlation (VWNV)
14
part of the SVWNV functional in
order to improve the overall accuracy of several atomic and
molecular properties as calculated using the LSDA method.
This approach was initially described by Brothers and Merz,
and we have expanded on the initial observation herein.
13
* Corresponding author phone: (814)865-3623; e-mail:
merz@psu.edu.
546 J. Chem. Theory Comput. 2005, 1, 546-553
10.1021/ct050007c CCC: $30.25 © 2005 American Chemical Society
Published on Web 06/02/2005