COLUMBUS

Columbus is an ab-initio electronic structure package implementing multi-configurational self consistent field (MCSCF),multi-reference (MR) configuration interaction singles and doubles (CISD), MR averaged quadratic coupled cluster (AQCC) and MR perturbation theory(PT) methodologies based on the graphical unitary group approach (GUGA). For relativistic two-component ("spin-orbit") CI calculations relativistic spin-orbit effective core potentials (SO-RECPs) are required. They replace the relativistic core so that the smaller basis set size allows a fairly cost-effective treatment provided suitable high quality SO-RECPs are at hand. An alternative are the scalar relativistic Douglas-Kroll-Hess (DKH) expansion or the X2C approach. Both represent scalar relativistic effects via a modification of the one-electron integrals in an all-electron basis. Hence, they are compatible with any non-relativist electronic structure method. In this case the non-scalar spin-orbit contribution is accessible through the atomic mean field integrals (AMFI) which - just as SO-RECPs - represent spin-orbit interaction with an additional term in the one-electron hamiltonian.

An efficient implementation of the GUGA approach requires a structural specification of the underlying n-electron space (as opposed to selective approaches like "determinant picking") to permit an efficient vectorization of large parts of the code. Otherwise there are no constraints on the structure of the n-electron space. Analytical gradients at MCSCF/MRCISD/MRAQCC level are available allowing for structure optimizations or molecular dynamics. There are interfaces to other electronic structure codes for the provision of integrals and/or complementary electronic structure methods or tools, most notably but not restricted to Molcas/OpenMolcas and Dalton.

The code can handle multi-billion MRCI expansions at acceptable computational costs, as a rule of the thumb expansions of 10**8 CSFs converge within an hour on a single node of JUWELS (40 cores). The computational cost with respect to the size of the MRCI expansion is approximately linear scaling.

The program version maintained and developed at JSC is a development version. The current development is optionally replacing the processing of 4-index integrals by three-index approximations in terms of Cholesky decomposition (taken from the Molcas interface) or the almost equivalent RI approximation (taken from the Turbomole interface). This replacement substantially reduces I/O and memory requirements while gaining speed through a more efficient vectorization and parallelizations. The necessary substantial modification of the algorithms, removal of redundancies in GUGA related data processing and more adequate workflows furnish the basis for handling medium-sized molecules more easily.

Official links:
https://columbus-program-system.gitlab.io/columbus
www.dalton.org
www.molcas.org
www.openmolcas.org