DIESEL-MP2: A New Program to Perform Large-Scale
Multireference-MP2 Computations*
PATRICK MUSCH, BERND ENGELS
University of Wuerzburg, Institute for Organic Chemistry, Am Hubland,
97074 Wuerzburg, Germany
Received 5 October 2005; Accepted 16 January 2006
DOI 10.1002/jcc.20416
Published online in Wiley InterScience (www.interscience.wiley.com).
Abstract: This article presents a new MR-MP2 code (Multi-Reference Møller–Plesset 2nd order) suitable for the
computation MR-MP2 energies of extended systems with strong near degeneracy effects (e.g., open shell systems). It
is based on the DIESEL program package developed by Hanrath and Engels. Due to improved algorithms the new code
is able to handle systems with 400 –500 basis functions and more than 100 electrons. The code is made for parallel
computers with distributed memory, but can also be run on local machines. It possesses two integral interfaces
(MOLCAS, TURBOMOLE). The algorithms are briefly introduced and timings for the Neocarzinostatin chromophore
are presented. The efficiencies of the codes obtained with Intel or GNU compilers are compared.
© 2006 Wiley Periodicals, Inc. J Comput Chem 27: 1055–1062, 2006
Key words: DIESEL-MP2; large-scale computations
Introduction
Size and complexity of the systems in material or biological
science have imposed novel challenging questions to the field of
computational chemistry and lead to the successful development of
more efficient computational algorithms. Examples are recent de-
velopments for MP2,
1,2
coupled cluster,
3
or DFT,
4
which have
reduced the computational cost for large-scale systems to a linear-
dependency allowing the treatment of systems up to some hundred
atoms and up to a few thousand basis functions. Their application
contributes to a deeper understanding of the factors that control
material properties or enzymatic activities.
5
However, most of the methods mentioned before refer to a
single reference, and thus possess only limited accuracy if near
degeneracy effects are important.
6
Such effects are present in
many systems being involved in material or enzymatic processes,
for example, radicals, diradicals, or transition metals with open
shells.
7,8
For such problems multireference methods like MR-CI or
MR-PT are necessary to obtain reliable results or at least, to
validate the single reference methods.
The advantage of multireference methods is their great flexi-
bility, which enables these approaches to describe reaction paths or
excited states very accurately.
9
The main drawback of multirefer-
ence methods is their high computational cost compared to single-
reference approaches. Several modifications have been proposed to
improve the computational efficiency like the individually select-
ing MR-CI,
10,11
CASPT2,
12
MR-MPn,
13–16
MCCEPA,
17
or inter-
nally contracted MR-CI.
1,18
The ansatz of local correlation treat-
ments
2,19
were also developed and implemented for multireference
approaches like local MR-CI
2d,20
and local MR-PT2.
21,22
The
computational cost of the latter approaches was reduced to a
certain degree, but a reliable estimate on the errors coming along
with this approximation has not yet been assessed for reaction
paths or for systems with a difficult electronic or a bulky geomet-
rical structure.
23
For large-scale multireference computations, the computation-
ally quite efficient MR-MP2 approach seems to be a good choice.
In this work we present a new code for this purpose. It is based on
the DIESEL program package developed by Hanrath and Engels.
11
The advantage of the new code is its ability to handle several
hundred basis functions and the possibility to run it on parallel
computers with distributed memory. Additionally further subdivi-
sion of the interacting space leads to more efficient algorithms.
This article consists of a brief introduction to the used MR-MP2
approach, followed by a description of the matrix element com-
Correspondence to: B. Engels; e-mail: bernd@chemie.uni-
wuerzburg.de
*Dedicated to Professor Volker Staemmler on the occasion of his 65
th
birthday.
Contract/grant sponsor: the Deutsche Forschungsgemeinschaft (in the
framework of SFB 410)
Contract/grant sponsor: the Volkswagen foundation
© 2006 Wiley Periodicals, Inc.