GBTOlib: a high-performance library for evaluation of molecular integrals

This library is used by the UKRmol+ codes but has a general applicability and has been interfaced with other codes. It follows the Fortran 2003 standard and has been written as a highly flexible object-oriented code.

Capabilities

• Evaluation of various types of molecular integrals in the basis of atom-centered Gaussian orbitals and center-of-mass centered B-splines and/or Gaussians.

• The center-of-mass (i.e. continuum) basis can be built from either Gaussians, B-splines or a combination of the two.

• An arbitrary angular momentum for the continuum basis.

• Transformation of the atomic integrals into basis of molecular orbitals.

• Orbital orthogonalization using Gramm-Schmidt and/or Symmetric orthogonalization.

• Flexible configuration of the library allowing it to be ran on various machines ranging from single-node workstations to massively-parallel HPC architectures.

• Possibility to configure the library to use quadruple precision arithmetics and achieve higher than double precision of the actual integrals calculated.

• Input of molecular geometry, Gaussian basis and orbitals via the standard MOLDEN file.

General information for version 4.0.0

• The new release comes with the option to run the code immediately without the need to compile it manually by using one of the two prepared computing environments: (1) Docker container (docker_image_gbtolib-4.0.0.zip) and (2) Conda environment. Detailed instructions for using and setting up these environments are provided in the conda/README.md file. Note that the Docker container is known to run without problems in MPI mode on a single computing node. For MPI runs across nodes we strongly advise you to compile the code yourself on your platform. Follow the instructions in the main README.md file.
• The current code implements a number of significant performance improvements. The performance of the new version with respect to the previous ones is illustrated in the figure provided (performace_improvements_illustration.zip). The benchmark was performed for a representative calculation of molecular integrals for the CO2 molecule with a large B-spline type basis for the continuum.
• A detailed description of the new features and fixes is below.

New features

• Implementation of Effective Core Potentials

• Enable MPI build without MPI compiler wrappers

• Interface for loading GBTO MO basis information and 2-el integrals in ADC codes

• Implementation of BGGG semi-analytic integrals and 2p continuum integrals

• Implemented triple overlap integrals over an arbitrary combination of GTO and BTO functions

• Poisson method for 2-electron integrals implemented

• Enable disk mapping for two-electron integrals over atomic orbitals

• Allow selection of Molden orbitals by occupation

• Compact indexing of AO integrals

• DALTON Molden file support

• Enable scratch use in sparse integral transformation

• A new program “orbitals_to_cube” for sampling of orbitals into Gaussian CUBE format

• New build options: Conda, Docker

Bug fixes

• A combined ALLOCATED(X) check and SIZE(X) comparison has been split into two separate IF conditions to avoid calling SIZE(X) with a possibly non-allocated array X.

• Use double precision walltime in omp_get_wtime

• Support 32-bit MPI in large allgatherv

• Correct CGTO shell order in written Molden files + allow writing in spherical basis

• Use TYPE instead of CLASS for non-polymorphic argument

• Fix dimension of angular_integrals_tmp

• Reset dummy INTENT(OUT) arguments in MPI module

• Replace a hard-coded value by a predefined MPI constant

• Fixing r2 grid in legendre_grid_r1_r2_obj

• Do not finalize uninitialized integral storage

• Avoid IFX deadlock in sort_intermediate_integrals

• Fix CAS in C1 test Psi4 inputs

• Synchronize orthogonalized MOs in MPI mode

• Avoid knot interpolation at endpoints

• Fix two calls to gl_expand_A_B

Optimizations of speed and memory use

• Improve load balancing in CGTO pw expansion precomputing

• Evaluate molecular orbital in parallel

• Distribute 1e transformation of integrals

• Finalize AO 1e integrals earlier

• Remove w1_w2 array

• Optimize Ylm calculation by factorization

• Avoid a few OpenMP critical sections in 2p integrals

• Threaded evaluation of BB 1el integrals

• Parallelize complex Gaunt precalculation

• Add threading also to CMS routines for BGBG integrals

• Reduce intermediate storage in sparse transformation

• Jointly evaluate BGGG and BBGG ints also for mixed method 1

• Add threading to the last part of omp_calculate_CGTO_pw_coefficients_analytic

• Sort also the final integrals in parallel

Other improvements

• Use ISO/IEC 80000 symbol for memory size

• add PCCHF tests for electron scattering

• Increase verbosity of basis_read

• Remove parameter max_contr_len

• Simplify install path handling in CMake scripts

• Store LOGICAL "use_symmetry" in moints as integer

• Construct B-spline grid as linear interpolation

• Do not require C++ in CMake scripts

• Avoid passing global variables to program routines

• Use 3.5 as minimum version of CMake

• Improve error message on failure of MPI_File_set_view

• Define local masters comm in mpi_mod

• Avoid CMake 3.10 deprecation warning

• Report BLAS integer size

• Replace legacy Cilk directive with OpenMP counterpart

• Print more details when "Sym=" not present in molden

• Polish memory mapping of files in Windows

• Changes to support addition of enhancement factors for positron scattering