Gaussian 03 Online Manual
Last update: 5 May 2008

Integral

The Integral keyword modifies the method of computation and use of two-electron integrals and their derivatives.

INTEGRATION GRID SELECTION OPTION

Grid=grid
Specifies the integration grid to be used for numerical integrations. Note that it is very important to use the same grid for all calculations where you intend to compare energies (e.g., computing energy differences, heats of formation, and so on). The parameter to this option is either a grid name keyword or a specific grid specification. If a keyword is chosen, then the option name itself may be optionally omitted (i.e, Integral(Grid=FineGrid) and Integral(FineGrid) are equivalent).

"Pruned" grids are grids that have been optimized to use the minimal number of points required to achieve a given level of accuracy. Pruned grids are used by default when available (currently defined for H through Kr).

The default grid is a pruned (75,302) grid, having 75 radial shells and 302 angular points per shell, resulting in about 7000 points per atom; the value FineGrid is used to specify this grid. Other grids may be selected by giving an integer value N as the argument to Grid.

Grid=UltraFine requests a pruned (99,590) grid. It is recommended for molecules containing lots of tetrahedral centers and for computing very low frequency modes of systems.

Other special values for this parameter are CoarseGrid, which requests a pruned version of the (35,110) grid, and SG1Grid, a pruned version of (50,194). Note, however, that the FineGrid has considerably better numerical accuracy and rotational invariance than these grids, and they are not recommended for production calculations [511]. Pass0Grid requests the obsolete pruned (35,110) grid once intended for pass 0 of a tight SCF calculation.

Specific grids may be selected by giving an integer value N as the argument to Grid. N may have one of these forms:

  • A large positive integer of the form mmmnnn, which requests a grid with mmm radial shells around each atom, and nnn angular points in each shell. The total number of integration points per atom is thus mmm*nnn. For example, to specify the (99,302) grid, use Int(Grid=99302). The valid numbers of angular points are 38, 50 [512], 72 [513], 86, 110 [512], 146, 194, 302 [514], 434 [515], 590, 770, and 974 [516]. If a larger number of angular points is desired, a spherical product grid can be used.

  • A large negative integer of the form -mmmnnn, which requests mmm radial shells around each atom, and a spherical product grid having nnn θ points and 2*nnn φ points in each shell. The total number of integration points per atom is therefore 2*mmm*nnn2. This form is used to specify the (96,32,64) grid commonly cited in benchmark calculations: Int(Grid=-96032).

  • Note, that any value for nnn is permitted, although small values are silly (values of nnn < 15 produce grids of similar size and inferior performance to the special angular grids requested by the second format above). Large values are expensive. For example, a value of 200100 would use 2*200*100*100 or 4 million points per atom!

RELATIVISTIC CALCULATIONS

DKH
Requests a Douglas-Kroll-Hess 2nd order scalar relativistic calculation [517,518,519,520] (see [521,522] for an overview). This method uses a Gaussian nuclear model [523]. DKH2 and DouglasKrollHess are synonyms.

NoDKH and NonRelativistic request a non-relativistic core Hamiltonian, which is the default.

DKH0
Requests a Douglas-Kroll-Hess 0th order scalar relativistic calculation

RESC
Requests a RESC scalar relativistic calculation

INTEGRAL FORMAT OPTION

Raff
Raff requests that the Raffenetti format for the two-electron integrals be used. This is the default. NoRaff demands that the regular integral format be used. It also suppresses the use of Raffenetti integrals during direct CPHF. This affects conventional SCF and both conventional and direct frequency calculations.

CNDO
Do calculation in main code using CNDO/2 ints.

INDO
Do calculation in main code using INDO/2 ints.

ZINDO1
Do calculation in main code using ZINDO/1 ints.

ZINDOS
Do calculation in main code using ZINDO/S ints.

ALGORITHM SELECTION OPTIONS

SSWeights
Use the weighting scheme of Scuseria and Stratmann [524] for the numerical integration for DFT calculations. This is the default.

BWeights
Use the weighting scheme of Becke for numerical integration.

NoSComp
Turn off symmetry blocking of MO 2-electron integrals. NoSymmComp is a synonym for NoSComp.

FMMNAtoms=N
Set the threshhold size for turning on FMM by default to N. The default is 60 atoms. Molecules with symmetry have higher crossover points and the threshold is increased accordingly, to 120 atoms for the C2 and Cs point groups and 240 atoms for higher symmetry.

DPRISM
Use the PRISM algorithm [27] for spdf integral derivatives. This is the default.

Rys1E
Evaluate one-electron integrals using the Rys method [525,526,527], instead of the default method. This is necessary on machines with very limited memory.

Rys2E
If writing two-electron integrals, use Rys method (L314) [192,525,526,527]. This is slower than the default method, but may be needed for small memory machines and is chosen by default if regular (non-Rafenetti) integrals are requested (by the NoRaff option).

Berny
Use Berny sp integral derivative and second derivative code (L702).

Pass
Pass specifies that the integrals be stored in memory via disk, and NoPass disables this. Synonymous with SCF=[No]Pass, which is the recommended usage.

Symm
NoSymm disables and Symm enables the use of symmetry in the evaluation and storage of integrals (Symm is the default). Synonymous with the keywords Symm=[No]Int, which is the recommended usage.

NoSP
Do not use the special sp integral program (L311) when writing integrals to disk.

RevDagSam
Reverse choice of diagonal sampling in Prism.

CPKS1Mat
Don't use CPKS multiple-matrices code.

SquareLoops
Forces square loops. SqLoops is a synonym for this option.

NoJEngine
Forbid use of special Coulomb code.

FofCou
Use FoFCou even when it would not otherwise be used. NoFoFCou forbid uses of FoFCou.

RevRepFock
Reverse choice of Scat20 vs. replicated Fock matrices.

Schwartz
Turn on Schwartz cutoffs in FMM/NFx. The default is off.

NoMPCut
Turn off MP-based cutoffs in FMM/NFx.

NoDFTCut
Turn off extra DFT cutoffs.

LTrace
Trace Linda transactions.

SplitSP
Split AO S=P shells into separate S and P shells. NoSplitSP is the default.

SplitSPDF
Split AO S=P=D and S=P=D=F shells into S=P, D, and F. NoSplitSPDF is the default.

SplitDBFSP
Split density S=P shells into separate S and P shells. NoSplitDBFSP is the default.

SplitDBFSPDF
Split density S=P=D and S=P=D=F into S=P, D, and F. NoSplitDBFSPDF is the default.

NoGather
Forbid use of gather/scatter digestion, even when processing small numbers of density matrices. Splatter is a synonym for this option.

ForceNuc
Do nuclear-electron Coulomb with electron-electron.

ECPAcc=N
Set ECP accuracy parameter to N.

NoSqrtP
Turn off use of Sqrt(P) in density-based cutoffs.

SepJK
Do J and K in HF/hybrid DFT separately for testing.

UnconAOBasis
Uncontract all the primitives in the AO basis. UncontractAOBasis is a synonym for this option.

UnconDBF
Uncontract all the primitives in the density fitting basis. UncontractDensityBasis is a synonym for this option.

NoPCXC
Do not precomputed grid information for DFT XC quadrature. NoPreComputeXC is a synonym for this option.

PCXCP
Precompute XC quadrature parameters (number of significant functions, etc.) used for allocation, but do not store information about individual grid points. PreComputeXCParameters is a synonym for this option.

PCXCWt
Precompute XC quadrature parameters and store weights for each point, to save the work of recalculating the weights. PreComputeXCWeights is a synonym for this option.

PCXCGrid
Precompute XC quadrature parameters and store both the weight and coordinates for each grid point. PreComputeXCGridPoints is a synonym for this option.

Seq2E
Set up for parallel 2 electron integral evaluation but then do not run in parallel (for debugging).

SeqXC
Set up for parallel 2 electron integral evaluation but then do not run in parallel (for debugging).

BigAtoms
Make all atom sizes large in XC quadrature.

BigShells
Make all shell sizes large in XC quadrature.

NoSymAtGrid
Do not use (Abelian) symmetry to reduce grid points on symmetry-unique atoms.

LinMIO
Convert to linear storage in FoFCou for testing.

RevDistanceMatrix
Reverse choice of whether to precompute distance matrix during numerical quadrature. The default is to precompute for molecules but not for PBC.

NoXCTest
Skip tests of numerical accuracy of XC quadrature.

NoDynParallel
Turn off dynamic work allocation.

INTEGRAL FILE-RELATED OPTIONS

ReUse
Use an existing integral file. Both the integral file and checkpoint file must have been preserved from a previous calculation. Only allowed for single point calculations and Polar=Restart.

WriteD2E
Forces the integral derivative file to be written in HF frequency calculations. Useful only in debugging new derivative code.

BUFFER SIZE OPTIONS

IntBufSize=N
Sets the integral buffer size to N integer words. The default value (which is machine-dependant) is generally adequate.

D2EBufSize=N
Sets the integral derivative buffer size to N words.

SCF