Manual:KR-CI
From Diracwiki
2- and 4-component relativistic GASCI module written by Timo Fleig based on LUCIA by Jeppe Olsen, parallelization by Stefan Knecht
The KR-CI module is a string-based Hamiltonian-direct configuration interaction (CI) program [1] [2], based on the LUCIA code [3]. It is capable of doing efficient CI computations at arbitrary excitation level, e.g. FCI, SDCI, RASCI, and MRCI using general active spaces. The code is interfaced to molecular integrals obtained in the relativistic 2c- and 4-component framework and uses double-point group symmetry. It is implemented as a full parallel version [4].
A central feature of the program is the Generalized Active Space (GAS) concept, in which the underlying total orbital space is subdivided into a basically arbitrary number (save for an upper limit) of subspaces with arbitrary occupation constraints. This is the most general approach to orbital space subdivisions. The program uses DIRAC Kramers pairs from either a closed- or an open-shell calculation in a relativistic two- and four-component formalism (see **HAMILTONIAN).
The technical limitations are roughly set by several 100 million determinants in the CI expansion on PCs and common computing clusters and several billions of determinants on supercomputers with ample memory. For calculations involving more than 500.000 determinants and in particular higher than double excitations it is strongly recommended to use the KR-CI module in 2c- and 4c applications.
If desired, the program also computes 1-particle densities from optimized CI wave functions from which the natural orbital occupations are printed.
The program is also used for the computation of various one-electron properties at the GASCI level [5] . Please refer to the KR-CI property module page for more information.
Contents |
Mandatory keywords
NOTE: the following keywords must be placed under the input deck
*KRCICALC
.CI PROGRAM
LUCIAREL
Default: None.
.MK2REF
DELTA_MK2 reference value (integer value). Should be set to zero. Change only if you know what you are doing.
Default: None.
.MK2DEL
DELTA_MK2 value (integer value). Should be set to the number of explicitly correlated electrons in the calculation. Change only if you know what you are doing.
Default: None.
.CIROOTS
number of states in symmetry X to optimize on. This keyword may be repeated several times for multi-root multi-symmetry calculations step-by-step.
Default: None.
example for 5 roots in (boson/fermion) irrep 1
.CIROOTS 1 5
If you run the calculation in linear symmetry (only available in the development version) you have to specify the 2 x OMEGA value of the state to optimize on. The doubling (2 x) stems from the fact that we want to avoid non-integer input, e.g. in case of an odd number of electrons we might have OMEGA=1/2, 3/2, 5/2, etc. values and the corresponding input for one root with the OMEGA=1/2 would then read as
.CIROOTS 1 1
If we have a system with an even number of electrons and inversion symmetry the input for two OMEGA=2g states read as
.CIROOTS 4g 2
.GASSH
Number and specification of GAS orbitals. Line with the number of GA spaces used (1-7), followed by one line per GAS with number of orbitals per fermion correp (either one (no inversion symmetry) or two (inversion symmetry: g u) entries per line). The design of GAS is non-trivial and should be motivated by the electronic structure of the system (e.g. inner core, outer core, valence, virtual space). Sometimes it is useful to subdivide the valence space, for scientific reasons, or/and the virtual space, for technical reasons (save core memory). See reference [6], pp. 27 for more details.
Default:
None.
.GASSPC
specification of CI calculation. One line per GAS with 2 entries each: The first entry gives the minimal number of accumulated (!) electrons after consideration of this GAS, the second the corresponding maximum number, separated by blanks. The minimum and maximum accumulated occupations allow for a very flexible parameterization of the wave function. All determinants fulfilling the occupation constraints will be constructed.
Default: None.
optional keywords
.MAX CI
Number of CI iterations.
Default:
.MAX CI 5
.INACTIVE
Inactive orbitals per fermion correp.
Default: All orbitals active.
.MXCIVE
size of Davidson subspace.
example:
.MXCIVE 24
Default: 3 times the number of eigenstates to optimize on.
.RSTRCI
Default: No restart. Optional restart from CI vector on file KRCI_CVECS.x where x is determined by the symmetry of the wave function (symmetry == 1 --> x = 1; symmetry == 2 --> x = 2; etc.). NOTE: in the DIRAC10 release version the x <==> symmetry relation is: symmetry == 1 --> x = a; symmetry == 2 --> x = b; etc.
.RSTRCI 1
Default: No restart.
.NOOCCN
compute natural orbital occupation numbers for each electronic state.
Default:
do not compute natural orbital occupation numbers.
.WRTFCK
write Fock matrix to file KRMC_FOCK where it can be re-read for a possible restart of a calculation (see the following keyword).
Default:
do not write Fock matrix to file.
.RDFOCK
read Fock matrix from file KRMC_FOCK.
Default:
do not read Fock matrix from file.
.CHECKP
enables a check point write of the current solution vectors to the file KRCI_CVECS.x (see above in .CIROOTS for an explanation of how x is supposed to be replaced) during the Davidson iterations. A checkpoint file will be written roughly every 6th iteration.
Default:
do not write check points.
.ANALYZ
analyze the final CI wave function printing the coefficients for each determinant above a given threshold.
Default:
do not analyze the final CI wave function.
raise default print level to the given integer value. Please use with care as you may get millions of output lines if you choose a too high value.
Default:
print level is 0.
Parallel keywords
.PARINT
Specifies the availability of the two-electron integral file on each co-workers scratch disk in parallel calculation.
Default: integral broadcast from master to co-workers
.PARINT 0
 |
Advanced options |
|
.IJKLRO
enables the storage of the resorted two-electron integrals on file IJKL_REOD which can then be read-in in a restart. This avoids a multiple reading from the 4IND* files and subsequent resorting. The keyword has to be present also in the restart to enable the potential read-in procedure. Hint: This keyword may be combined with .MAX CI == 0 in a precedent step in order to save memory for the actual production run. This is due to the fact that the resorting step itself requires the in-core storage of the unsorted and sorted integrals.
Default:
do not write the resorted integrals to file IJKL_REOD.
.IJKLSP
enables the splitting of the resorted integrals among the co-workers according to their needs in the computation of the sigma vector. It can only be used in combination with.IJKLRO.
Default:
do no split the resorted integrals among the co-workers.
[KR-CI_properties]
see the KR-CI property module page for more information.
References
- ↑ T. Fleig, J. Olsen, and L. Visscher, The generalized active space concept for the relativistic treatment of electron correlation. II: Large-scale configuration interaction implementation based on relativistic 2- and 4-spinors and its application, J. Chem. Phys. 119, 2963 (2003) .
- ↑ T. Fleig, J. Olsen, H. J. Aa. Jensen, and L. Visscher, The generalized active space concept for the relativistic treatment of electron correlation. III: Large-scale configuration interaction and multi-configuration self-consistent-field four-component methods with application to UO_2, J. Chem. Phys. 124, 104106 (2006) .
- ↑ J. Olsen and P. Joergensen and J. Simons, Passing the one-billion limit in Full CI calculations, Chem. Phys. Lett. 169, 463 (1990) .
- ↑ S. Knecht and H. J. Aa. Jensen and T. Fleig, Large-Scale Parallel Configuration Interaction. II. Two- and 4-Component Double-Group General Active Space Implementation with Application to BiH., J. Chem. Phys. 128, 014108 (2010) electronic version.
- ↑ S. R. Knecht, PhD thesis, University of Düsseldorf (Germany) ', ' (2009) .
- ↑ T. Fleig, Habilitation thesis, University of Düsseldorf (Germany) ', ' (2006) .
