Integral transformation module.
This module transforms integrals from the scalar AO basis to integrals in the molecular spinor basis. The algorithm is not yet described in detail but it may be helpful to read Ref. [Visscher2002] to find information about the double quaternion formalism that is used. This module is usually invoked by one of the correlation modules but can also be run as a stand-alone module by specifying .4INDEX in the **DIRAC section of input. The input serves to change the default active space and to specify special wishes (e.g modification of .INTFLG).
Three separate transformations are considered:
- Transformation of a core Fock matrix, giving effective one-electron matrices.
- Transformation of two-electron Coulomb integrals.
- Transformation of integrals over general (one-electron) operators.
The output is written to the files MDCINT, MRCONEE, and MDPROP, respectively, and can be read in by the **RELCC and DIRRCI – Direct CI module and programs to perform correlated calculations and by the RELADC, FANOADC and LANCZOS program to perform propagator calculations. Some common input needs to be specified both in this section and in the DIRRCI – Direct CI module sections.
Specify the active set of spinors.
Arguments: One line with an Specification of orbital strings for each fermion irrep.
energy -20.0 10.0 1.0
Threshold value for writing transformed 2-electron integrals to file. A higher value will decrease the size of the MDCINT files but will decrease the accuracy of the calculation.
Perform transformation of property integrals. The type(s) of integrals that are to be transformed can be specified with the .OPERATOR keyword.
Default: Do not transform property integrals.
Selects the integral transformation algorithm (“scheme”). Several schemes have been implemented, but users are to consider only “scheme 4” and “scheme 6” for production calculations. Other schemes refer to outdated to special-purpose algorithms, and should be used with care. The most suitable scheme to use depends to a significant extent on the type of calculation in question (size of the active space etc), and on the computational resources used (compilers, amount of memory/scratch disk space, whether or not disks are shared to local to compute nodes etc).
Scheme 6 (Default):
Scheme 6 is designed to decrease I/O operations at the expense of more communication between processes, something which is more often than not advantageous in today’s architectures. It should be noted, however, that the significant amount of communication between processes happens for the so-called “1HT” step (“1HT” standing for first half-transformation, since it is where the first two indexes of the electron repulsion integrals are transformed to MO basis and these intermediate, half-transformed entities are stored on disk), with no communication on the “2HT” step (which will yield the final integrals in MO basis). Due to the large amount of half-transformed entities, this scheme may demand large scratch spaces (e.g. a Dirac-Coulomb calculation on water with the aug-cc-pV5Z basis, where the full spinor space is active and LL and SL integrals are transformed can reach nearly 1Tb of disk use) but with the advantage that the full size of the problem is exactly divided between processes - so the larger the number of MPI processes, the less disk space is used per process.
In connection to scheme 6 the user can also choose whether or perform an intermediate sorting step prior to the second half-transformation step, via the keyword .HTSORT. In this step the 1HT entities on disk are rearranged from their “natural” order (arising from the two-electron integral algorithm) to an order which will minimize the amount of I/O operations taking place at the 2HT step.
HT sort disabled (Default):
.SCHEME 6 .HTSORT 0
Sorting of HT intermediates is disabled by default. Without such sorting, scheme 6’s performance can be negatively affected on computer systems where reading data is relatively expensive (one example is for IBM p4/p5/p6 systems and the XLF compiler) even for relatively modest calculations Dirac-Coulomb calculations (where LL and SL integrals are considered). However, in two-component calculations (where only LL integrals are transformed and therefore there are much less 1HT entities on disk) of similar or larger size, in terms of active space, good performance can still be achieved in the same architecture.
HT sort (strategy #1) enabled:
.SCHEME 6 .HTSORT 1
This sort strategy minimizes the amount of read/write operations by taking up as much 1HT data in memory and sorting it before writing it to disk again. This provides efficiency for systems (such as the IBM systems mentioned above) where reading is expensive, at the expense of larger memory usage (rule of thumb: the extra memory required *per MPI process* will be about 1-2% of the *total disk* space taken up by the 1HT quantities - so if that is 1Tb, the extra memory used will be of of about 1-2Gb; see the calculation output for more details).
Scheme 4 is designed to reduce communication between different processors at the expense of increasing I/O operations. This increase in I/O operations stems from the integrals being replicated for all MPI processes, so that doubling the number of MPI processes doubles the total disk usage but not the usage per MPI process.
Specify which classes of integrals should contribute to the transformed two-electron integrals and the effective (core) Hamiltonian. Default is to follow the definitions given in **HAMILTONIAN. If you want different integral classes in the 2- and 4-index transformation, then use the .INTFL2 and .INTFL4 keywords below (default: .INTFLG from **HAMILTONIAN).
Specify the frozen core spinors,
For each fermion irrep, give an Specification of orbital strings of active orbitals.
Default: All occupied orbitals that are not active.
Allow negative energy (“positronic”) spinors in active set. If this keyword is not specified then only electronic spinors will be included and any positronic spinors specified with .ACTIVE are ignored.
Skip the 2-index transformation of the effective Fock matrix.
Skip the 4-index transformation.
Screening threshold in 4-index transformation (a negative value disables screening).
Write integrals to the ASCII file MO_integrals.txt to facilitate interfaces with correlation codes. Warning: only meant for initial interfacing as this format is very inefficient.
Default: Do not write the ASCII file.
Recanonize orbitals before transforming.
Default: Do not recanonize orbitals before transforming.
Ranges of active orbitals in 2-index transformation module specified for index 1 and 2 and fermion irrep 1 and 2.
Default: Ranges set by .ACTIVE.
Ranges of active orbitals in 4-index transformation module specified for index 1 to 4 and fermion irrep 1 and 2.
Default: Ranges set by .ACTIVE.
Write 4-index transformed integrals to file (default).
Skip writing 4-index transformed integrals to file. Instead, write file MDINFO, which contains infos used in the MCSCF program.
Scatter 4-index transformed integrals to all nodes (default).
Do not scatter 4-index transformed integrals to all nodes
Specify number of batches for each node in 4-index transformation. Only working for parallel distributions.
Property integrals transformation¶
This section defines the property operators that should to be transformed. Such integrals are only used in an experimental section of **RELCC, and this subsection can thus usually be omitted.