Next: 26.6 Additional options available Up: 26 LOCAL CORRELATION TREATMENTS Previous: 26.4.6 Split Coulomb operator


26.5 Options

Various options can be specified using key/value pairs qualifying the LOCAL or MULTP command. For all options appropriate default values are set, and so these options must usually be modified only for special purposes. For convenience and historical reasons, alias names are available for various options, which correspond to the variable name used in the program. Table 9 summarizes the keys, aliases and default values. In the following, the parameters are described in more detail.


Table 9: Summary of local (multp) options and their default values
Parameter Alias Default value Meaning
       
General Parameters:
LOCAL   4 determines which program to use
SAVE SAVDOM 0 specifies record for saving domain info
START RESTDOM 0 specifies record for reading domain info
PIPEK LOCORB 0 activates or deactivates PM localization
SAVORB SAVLOC 0 specifies record for saving local orbitals
DOMONLY   0 if set, only domains are made.
      if 2, only orbital domains are made.
       
Parameters to select weak and distant pairs:
WEAKPAIR WEAKP 1 criterion for selection of weak pairs
DISTPAIR DISTP 0 (8) criterion for selection of distant pairs
VERYDIST VERYD 0 (15) criterion for selection of very distant pairs
       
Parameters to define domains:
DOMSEL CHGFRAC 0.98 selection criterion for orbital domains
DELCOR IDLCOR 2 delete projected core AOs up to certain shell
DELBAS IBASO 0 determines how to remove redundancies
       
Parameters for redundancy check using DOMSEL=1
TYPECHECK TYPECHK 1 activates basis function type restrictions
DELSHL IDLSHL 1 determines if whole shells are to be deleted
DELEIG IDLEIG 1 determines how to select redundant functions
DELCMIN CDELMIN 0.1 parameter for use with DELEIG=1
       
Parameter for multipole treatment of exchange operators:
DSTMLT   0 (3) expansion level for distant pairs
       
Parameters for energy partitioning:
IEPART   0 If nonzero: do energy partitioning
EPART   3.0 cutoff parameter for determining individual monomers
       
Miscellaneous options:
SKIPDIST SKIPD 3 determines at which stage weak and distant pairs are eliminated
ASYDOM JITERM 0 parameter for use of asymmetric domains
LOCSING LOCSNG 0 determines virtual space used for singles
MAXANG MAXL 99 restriction for Boughton-Pulay domain selection
CHGMIN   0.01 minimum Mulliken charge for BP domain selection
CHGMINH   0.05 minimum Mulliken charge of H-atoms for BP domain selection
CHGMAX   0.40 If charge larger than this value, atom is always included
MAXBP MAXBP 0 determines how to rank atoms in Boughton-Pulay domain selection
MULLIKEN LOCMUL 0 determines how to rank atoms for domains
PIPEKAO LOCAO 0 activates AO localization criterion
NONORM   2 determines whether projected functions are normalized
LMP2ALGO MP2ALGO 1 if nonzero, use low-order scaling method in LMP22 iterations
OLDDEF   0 allows to revert to older defaults
       
Thresholds:
THRPIP   1.d-12 Threshold for Pipek-Mezey localization.
THRORB   1.d-6 Threshold for eliminating projected orbitals with small norm.
THRLOC   1.d-6 Threshold for eliminating redundant projected orbitals.
THRCOR   1.d-1 Threshold for eliminating projected core orbitals.
THRMP2   1.d-8 Threshold for neglecting small fock matrix elements in the
      LMP2 iteration.


General Parameters:

LOCAL=local
Determines which method is used:
LOCAL=0: Conventional (non-local) calculation.
LOCAL=1: Local method is simulated using canonical MOs. The local basis is used only at an intermediate stage to update the amplitudes in each iteration (only for testing).
LOCAL=2: Calculation is done in local basis, but without using local blocking (i.e. full matrices are used). This is the most expensive method and only for testing.
LOCAL=3: Fully local calculation. This is the fastest method for local calculations with no weak pairs.
LOCAL=4: Fully local calculation (Default). This is the fastest method for large molecules with many weak pairs and requires minimum memory.

SAVE=record
Allows the domain information to be saved on record=name.ifil for later restart using START. This can be used to freeze the domains as function of geometry. Note that the domain information is automatically stored if a SAVE directive is given (see above), and in this case the record given on the SAVE card will overwrite any record given as SAVE option.

START=record
Retrieves domain information previously saved using SAVE. Note that the domain information is automatically restored if a START directive is given (see above), and in this case the record given on the START card will overwrite any record given as START option.

PIPEK=option
If this option is given and option$\gt 0$, the orbitals are localized using the Pipek-Mezey technique. If this option is not given or option=0 (default), the orbitals are localized unless localized orbitals are found in the orbital record (cf. ORBITAL and LOCALIZE directives). In the latter case, the most recent localized orbitals are used. Setting option=-1 switches the localization off. If option$\gt 1$ the localized orbitals are printed. Note: Boys localization can only be performed using the LOCALIZE command. The program will use the Boys orbitals if they are found in the orbital record and the PIPEK option is absent or option$\le 0$.

SAVORB=record
Allows the localized and projected orbitals to be saved on record=name.ifil for later use (e.g. plotting). The two orbital sets are stored in the same dump record and can be restored at later stages using
ORBITAL,record,[TYPE=]LOCAL or
ORBITAL,record,[TYPE=]PROJECTED, respectively.

DOMONLY=value
If value$\gt 0$ only domains are made, but no energy is computed. This can be used to check and save the domains for later use.

Parameters for selection of weak and distant pairs:

WEAKPAIR=distance
If all atoms of orbital domain [i] are separated by at least distance [a.u.] from any atom of orbital domain [j], pair (ij) is treated by MP2. The default is distance=1, which means that all pairs for which [i] and [j] have no atom in common are treated as weak pairs. Setting distance=0 eliminates weak pairs, i.e. all pairs are fully included in the calculation. This option has no effect for local MP2 calculations.

DISTPAIR=distance
If all atoms of orbital domain [i] are separated by at least distance [a.u.] from any atom of orbital domain [j], pair (ij) is treated approximately by MP2, provided the multipole approximation is activated. Setting distance=0 eliminates distant pairs, i.e. no pairs are treated approximately. Default is 0 (MULTP card: 8).

VERYDIST=distance
If all atoms of orbital domain [i] are separated by at least distance [a.u.] from any atom of orbital domain [j], pair (ij) is neglected. Setting distance=0 (default) eliminates very distant pairs, i.e. no pairs are neglected. Reasonable values for distance would be 12-15 [a.u.] Default is 0 (MULTP card: 15).

Parameters to define domains:

DOMSEL=value
Threshold for selecting the atoms contributing to orbital domains using the method of Boughton and Pulay. The default is value=0.98. value=1.0 would include all atoms into each orbital domain. The criterion is somewhat basis dependent. The larger the basis, the fewer functions will be selected with a given threshold. The default value usually works well for double-zeta basis sets. For larger basis sets (e.g., cc-pVTZ) it is recommended to use value=0.985. In most cases, the domain selection is uncritical for saturated molecules. However, for delocalized systems it is recommended to check the printed orbital domains! In cases of doubt, compare the domains you get with a smaller basis (e.g., cc-pVDZ). See also the MAXANG option below.

DELCOR=nshell
Activates elimination of basis functions corresponding to core orbitals. If nshell=1, only $1s$-functions are eliminated from projected space. If nshell=2 (default) $1s$ functions on first-row atoms, and $1s$, $2s$, and $2p$-functions are eliminated on second-row atoms. Nothing is eliminated on H or He atoms. If effective core potentials are used, nothing is deleted at the corresponding atom. Also, functions are only deleted if the norm of the projected function is below THRCOR (default 0.1)

DELBAS=ibaso
This parameter determines the method for eliminating redundant functions of pair domains.
ibaso=0: The space of normalized eigenvectors of ${\bf\tilde S}^{ij}$, which correspond to small eigenvalues, is eliminated (default if no gradients are computed).
ibaso $\gt$ 0: individual basis functions are eliminated. The value of ibaso affects details of the method to determine redundant functions.
ibaso=1: Redundant functions eliminated from pair domains, using Jacobi method for diagonalization of overlap matrices. This is the default if properties or gradients are computed.
ibaso=2: Redundant functions are eliminated from pair domains, using Householder method for diagonalization of overlap matrices.
ibaso=3: Redundant functions are eliminated from orbital and pair domains, using Jacobi method for diagonalization of overlap matrices.
ibaso=4: Redundant functions are eliminated from orbital and pair domains, using Householder method for diagonalization of overlap matrices.
The diagonalization method has only an effect for DELEIG=1 if degenerate eigenvalues are present. ibaso$\gt 2$ has only an effect if NONORM=0.

Parameters for selection of redundant functions if DELBAS$\gt 0$:

DELSHL=idlshl
This parameter determines if whole shells of basis functions (i.e., all $p$-functions for a given exponent at one atom) should be simultaneously eliminated. This may be useful in order to guarantee rotational invariance in geometry optimizations and frequency calculations.
idlshl=1: eliminate as many functions of a shell simultaneously as possible, but never more than determined by small eigenvalue of the overlap matrix (default).
idlshl=2: as idlshl=1, but also eliminate functions with identical norm simultaneously.
idlshl=3: eliminate all functions of a shell simultaneously, even if a larger number of functions is deleted than determined by small eigenvalues of the overlap matrix. This must be used with care, since very poor energies may sometimes result!
idlshl=4: as idlshl=3, but also eliminate functions with identical norm simultaneously.
idlshl$\gt 4$: as idlshl=4, but equivalent functions centred at all symmetry equivalent atoms are considered to form a shell (not recommended!).

TYPECHECK=typechk
If nonzero, activates basis function type restrictions in redundancy check. For a given atom, only basis functions corresponding to occupied atomic orbitals are allowed to be deleted. For instance, on first row atoms at most two $s$-functions and one $p$-shell will be deleted. No functions are deleted from hydrogen or He atoms.

DELEIG=idleig
This option determines how redundant basis functions are selected.
idleig=0: functions corresponding to the smallest diagonal elements of projected orbital matrix are eliminated.
idleig=1: Functions corresponding to the largest coefficients in the eigenvectors of ${\bf\tilde S}^{ij}$ are deleted (default). Since degenerate eigenvectors can arbitrarily mix, the selection may not be unique and depend on the diagonalization method (see DELBAS).

DELCMIN=value
Only effective with DELEIG=1. Only basis functions with coefficients larger than value in the eigenvectors of small eigenvalues can be deleted (default 0.1).

Parameters for multipole treatment of exchange operators

MULTMETHOD=option
Used internally by the MULTP card - don't mess with it.

DSTMLT=level
Determines the expansion level of the multipole expansion of distant pairs (e.g. 1 means dipole approximation, 2 quadrupole approximation and so on). Default is 0 (MULTP card: 3).

Parameters for energy partitioning:

IEPART=value
enables/disables energy partitioning.
iepart=0: energy partitioning is disabled.
iepart=1: energy partitioning is enabled.
iepart=2: energy partitioning is enabled. Additionally, a list of all pair energies and their components is printed.
EPART=cutoff
cutoff parameter to determine individual monomers in a cluster (i.e. centre groups). Should be somewhat larger than the largest intramolecular bond length (given in a.u.).

Miscellaneous options:

SKIPDIST=skipdist
Test-parameter. Its value should only affect the efficiency but not influence the results.
skipdist=-1: weak and distant pairs are set to zero after MP2 but are not eliminated from the pair list and not skipped in any loop.
skipdist=0: No pairs are deleted from pair list, but weak and distant pairs are skipped in the loops were appropriate.
skipdist=1: Very distant pairs are neglected from the beginning. Distant pairs are eliminated after MP2.
skipdist=2: As skipdist=1, but also weak pairs are eliminated after MP2.
skipdist=3: As skipdist=2, but distant pairs are eliminated from the operator list in case of LMP2 with multipole approximations for distant pairs. This is the default.

ASYDOM=jiterm
Enables the use of asymmetric domains for distant pairs. The asymmetric domain approximation supplements the multipole approximation for distant pairs, as it suppresses the treatment of configurations for which no integrals can be computed by multipole expansion. This leads to computational savings and improved numerical stability.
jiterm=0: Disable asymmetric domains.
jiterm=-1: Enable asymmetric domains (default).
jiterm=-2: Enable a variation of the asymmetric domain formalism: Exchange operators will initially be projected to the asymmetric domain instead of simply packed.

LOCSING=locsing
If locsing.ne.0, the single excitations use the full space, i.e., they are not treated locally. This is only works for LOCAL=1.

MAXANG=lmax
The purpose of this experimental option is to reduce the basis set sensitivity of the Boughton-Pulay (BP) method for domain selection. Only basis functions with angular momentum up to lmax-1 are included when computing the overlap of the approximate and exact orbitals. For example, MAXANG=2 means to omit all contributions of $d$, $f$ and higher angular momentum functions. To obtain reasonable domains, the value of DOMSEL must often be reduced (to 0.97 or so). This option should only be used with care!

MAXBP=maxbp
If maxbp=1, the atoms are ranked according to their contribution to the Boughton-Pulay overlap (default); this should normally give the smallest and best orbital domains. If maxbp=0, the atoms are ranked according to Mulliken charges. In both cases atoms with Mulliken charges greater than 0.6 are always included, and atoms with the same Mulliken charges are added as groups.

MULLIKEN=option
Determines method to determine atomic charges. option=0: diagonal elements of ${\bf S}^{\frac{1}{2}} {\bf C}$ are used. option=1: Mulliken gross

PIPEKAO=option
If option$ \ge 0$, the orbitals are localized my maximizing the coefficients of basis functions of a given type at a given atom. Normally, this is only useful to uniquely define degenerate orbitals in atoms. For instance, when this option is used to localize the orbitals for a dimer like (Ar)$_2$ at a very long distance, clean $s$, $p_x$, $p_y$, and $p_z$ atomic orbitals will be obtained. It is not recommended to use this option for molecular calculations!

NONORM=value
Determines if projected functions are normalized (not recommended). value=-1: projected orbitals are normalized before redundancy check.
value=0: projected orbitals are normalized after redundancy check (default).
value=1: projected orbitals are normalized in redundancy check, afterwards unnormalized.
value=2: projected orbitals are never normalized.

LMP2ALGO=value
If nonzero, use low-order scaling method in LMP2 iterations. This may require more CPU time in calculations for smaller molecules.

OLDDEF=value
For compatibility with older versions: if nonzero, revert to old defaults. Options set before this may be overwritten.

Thresholds:

THRPIP=thresh
Threshold for Pipek-Mezey localization. The localization is assumed to be converged if all $2\times 2$ rotation angles are smaller then thresh. The default is $1.d-12$. It can also be modified globally using GTHRESH, LOCALI=thresh. Note that GTHRESH is not an input command of the local program and must be given before the METHOD card.

THRORB=thresh
Threshold for eliminating functions from pair domains whose norm is smaller then thresh after projecting out the occupied space. The default is throrb=1.d-6.

THRLOC=thresh
Threshold for eliminating redundant basis functions from pair domains. For each eigenvalue of ${\bf\tilde S}^{ij} \lt thresh$ one function is deleted. The default is 1.d-6. The method used for deleting functions depends on the parameters IDLEIG and IBASO.

THRMP2=thresh
Threshold for neglecting small fock matrix couplings in the LMP2 iterations (default 1.d-8). Specifying a larger threshold speeds up the iterations but may lead to small errors in the energy. In the initial iterations, a larger threshold is chosen automatically. It is gradually reduced to the specified final value during the iterations.

THRCOR=thresh
Threshold for deleting projected core orbitals. The functions are only deleted if their norm is smaller than thresh (default 0.1)

The thresholds can also be specified on the THRESH card.



Next: 26.6 Additional options available Up: 26 LOCAL CORRELATION TREATMENTS Previous: 26.4.6 Split Coulomb operator

P.J. Knowles and H.-J. Werner
molpro@tc.bham.ac.uk
Jan 15, 2002