b347eb4459c127ec.tex
1: \begin{abstract}
2: 
3: Molecule-optimized basis sets, based on approximate natural
4: orbitals, are developed for accelerating the convergence of quantum
5: calculations with strongly correlated (multi-referenced) electrons.
6: We use a low-cost approximate solution of the anti-Hermitian
7: contracted Schr{\"o}dinger equation (ACSE) for the one- and
8: two-electron reduced density matrices (RDMs) to generate an
9: approximate set of natural orbitals for strongly correlated quantum
10: systems.  The natural-orbital basis set is truncated to generate a
11: molecule-optimized basis set whose rank matches that of a standard
12: correlation-consistent basis set optimized for the atoms. We show
13: that basis-set truncation by approximate natural orbitals can be
14: viewed as a one-electron unitary transformation of the Hamiltonian
15: operator and suggest an extension of approximate natural-orbital
16: truncations through two-electron unitary transformations of the
17: Hamiltonian operator, such as those employed in the solution of the
18: ACSE.  The molecule-optimized basis set from the ACSE improves the
19: accuracy of the equivalent standard atom-optimized basis set at
20: little additional computational cost. We illustrate the method with
21: the potential energy curves of hydrogen fluoride and diatomic
22: nitrogen.   Relative to the hydrogen fluoride potential energy curve
23: from the ACSE in a polarized triple-zeta basis set, the ACSE curve
24: in a molecule-optimized basis set, equivalent in size to a polarized
25: double-zeta basis, has a nonparallelity error of 0.0154~a.u. which
26: is significantly better than the nonparallelity error of 0.0252~a.u.
27: from the polarized double-zeta basis set. \\
28: 
29: {\bf Keywords}: multi-reference electron correlation, natural
30: orbitals, reduced density matrices, anti-Hermitian contracted
31: Schr{\"o}dinger equation, unitary transformations \\
32: \indent {\bf Contact}: damazz@uchicago.edu (e-mail); 1-773-834-1762
33: (phone)
34: 
35: \end{abstract}