Highly parallelised relativistic Coupled Cluster Code¶
Highly parallelised relativistic Coupled Cluster Code. In this release only computations using the X2C Hamiltonian (with either .X2Cmmf or .X2C) are possible.
This code is based on the math libraries TAL-SH and ExaTENSOR by Dmitry Lyakh. The tensors are kept in working memeory, sufficent RAM needs to be available. In order to test memory requirements instructions can be found in the
exacorr_exatensor_memory tests. TAL-SH runs on a single node which has to have enough memory (.TALSH_BUFF). In ExaTENSOR the memory is distributed, so each additional node will contribute its memory to the memory pool accessible by the library. Currently, it is recommended to use enough nodes that the tensors fit, but not substantially more.
In the current release, if the library runs out of memory the code will not stop but enter a blocked state and calculations will not advance. So carefully control the advancement of your calculations, stopping them if they appear to hang.
Defines occupied orbitals. Specification of list or energy range (see Specification of orbital strings).
.OCCUPIED energy -1.0 0.0 0.00001
Defines virtual orbitals. Specification of list or energy range (see Specification of orbital strings).
Set convergence criteria (CC iterations, Lambda equations)
Maximum number of allowed CC iterations to solve the CC and LAMBDA equations.
This keyword activates the full multinode EXATENSOR library, which is designed for massively parallel supercomputers. The additional infrastructure needed for parallel communication makes this implementation inefficient when used for single node runs. For such purposes the use of only the TALSH library component is recommended, which is designed for one node but will make use of GPUs (if available and suitable).
Do not use EXATENSOR
Solve Lambda-equations, needs to be activated in order to compute the one particle density matrix and molecular properties.
This calculation generates the file CCDENS, which contains the CC ground-state density matrix in AO basis. In this release, CCDENS is used by the property module to calculate ground-state expectation values.
If saved, CCDENS can be used in a property calculation (see .RDCCDENS) without the need to invoke this module.
Lambda equations are not solved
Deactivates computation of triples energy corrections (useful for ExaTENSOR as the current implementation is not efficient)
Triples are done
Performs a CC2 calculation instead of the default CCSD. Currently supported only for energies.
CC2 is not activated
Expert option to choose another AO to MO integral transformation scheme. Change at your own risk.
In TALSH only schemes 3 (default) and 42 (using Cholesky decompostion) are available.
In ExaTensor schemes 1-4 and 42 are available with 42 using Cholesky decompostion. Scheme 4 is default for ExaTensor as it reduces the memory footprint by only keeping part of the AO integrals in memory. The other methds keep all AO integrals in memeory. Scheme 0 prints the memory requirements and attempts to allocate the memory without doing the calculation.
Can be used to specify a “high-spin” reference determinant with a different number of “barred” occupied orbitals, than “unbarred” occupied spinors. If .OCC_BETA is specified .OCCUPIED is interpreted as a list of unbarred (alpha) spinors. NB: alpha and beta are used in a loose sense in relativistic calculations to indicate the (un)barred spinors.
.OCC_BETA energy -1.0 0.0 0.00001
Can be used to specify a different number of “barred” virtual orbitals than “unbarred” occupied spinors. If .VIR_BETA is specified .VIRTUAL is interpreted as a list of unbarred (alpha) spinors. NB: alpha and beta are used in a loose sense in relativistic calculations to indicate the (un)barred spinors.
Performs a CCD calculation instead of the default CCSD (switch off the contributions of single excitations).
CCDOUBLES is not activated
Expert option: Number to tune the parallel distribution (branching) of the spinor spaces.
Maximum memory (in gigabytes) used in TALSH, aim at about 80% of available memory on your machine.
Threshold to define the accuracy of the Cholesky decomposition (MOINT scheme 42), resulting in inaccuracies of the computed energy of this order of magnitude (in Hartree units).
Expert option: Level shift of orbital energies, ignored for values smaller 0.
Perform a closed shell MP2 calculation and generate the frozen natural orbitals(FNOs) by diagonalization of the virtual-virtual block of the MP2 density matrix, outputting the original occupied orbitals and a set of truncated virtuals in AO basis (the FNOs are transformed into AO basis and saved on file MP2NOs_AO, which has the same structure as DFCOEF.)
The user must specify a threshold indicating a NO occupation number, and orbitals with occupation below that will not be retained in the truncated AO basis.
If one would like to use the new set including the FNOs to do higher-level computation like CC, the MP2NOs_AO should be retrieved from the work directory for the calculati, and used as a standard DFCOEF file in subsequent calculations (currently a run in which FNOs are generated and used in the same post-SCF calculation is not supported).
Apart from MP2NOs_AO, this option also ouputs the full, untruncated FNOs space (MO basis) in file NOs_MO. The Fock matrix (FNO basis) is also outputted to file FM_in_NatOrb.
The NOs_MO is provided so that users wishing to change the truncation thresold above don’t have to repeat the same MP2 calculation (If present in the scratch directory, the NOs_MO will make the code skip the MP2 calculation).
MP2FNO is not activated.
Calculate natural occupation numbers and orbitals in AO (quaternion) basis, from a density matrix in the same basis (such as the one generated by the first-order property code, which saved by default in file CCDENS), and store them in file DFNOSAO.
This option assumes that CCDENS is in the work directory for the calculation.
DONATORB is not activated.