method keywords request a Hartree-Fock calculation (RHF for singlets, UHF for
higher multiplicities) followed by a Møller-Plesset correlation energy
correction , truncated at second-order for MP2
third order for MP3 [61,66],
fourth-order for MP4 , and fifth-order
for MP5 . Analytic gradients are available
for MP2 [22,23,139,140],
MP3 and MP4(SDQ) [141,142],
and analytic frequencies are available for MP2 .
ALGORITHMS FOR MP2
There are four basic algorithms for MP2 calculations
and for producing transformed (MO) integrals on disk:
which uses both main memory and external (disk) storage as available .
This is the default algorithm.
Direct, which uses no external
storage by recomputing the integrals as needed during the
Conventional, which stores the transformed
integrals on disk. This was the only method available
in Gaussian 88, and the only method for generating MO integrals
on disk in Gaussian 90. It is seldom a good choice on any but the
smallest computer systems.
In-core, in which all the AO
integrals are generated and stored in main memory, then used without
storing them externally.
The default is to decide between the
in-core, direct, and semi-direct algorithms based on available memory and disk.
The available disk can be specified via the MaxDisk
keyword, either in the route section or (preferably) in the Default.Route
Note that selection of the direct or semi-direct MP2 and transformation
algorithms is separate from selecting direct SCF (which is the default SCF algorithm
in Gaussian 03). The E(2) calculation or transformation then
recomputes integrals as needed in the form required for vectorization.
MP4(DQ) is specified to use only the space of double and
quadruple substitutions, MP4(SDQ) for single, double and quadruple substitutions,
or MP4(SDTQ) for full MP4 with single, double, triple and quadruple
Just specifying MP4 defaults to MP4(SDTQ).
The MP5 code has been written for the open shell case only, and so
specifying MP5 defaults to a UMP5 calculation. This method requires
O3V3 disk storage and scales as O4V4
in cpu time.
FROZEN-CORE OPTIONS (POST-SCF METHODS)
The frozen-core options for defining inner-shells to be excluded from the correlation
calculation are valid with these keywords. See the discussion here
ALGORITHM SELECTION OPTIONS (MP2 METHODS)
appropriate algorithm for MP2 will be selected automatically based on the
settings of %Mem and MaxDisk.
Thus, these options are almost never needed.
the "fully direct" algorithm, which requires no external storage beyond that for
the SCF. Requires a minimum of 2OVN words of main memory (O=number of occupied
orbitals, V=number of virtual orbitals, N=number of basis functions).
This is seldom a good choice, except for machines with very large main memory
and limited disk.
Forces the semi-direct algorithm.
Requests some sort of direct algorithm. The choice between in-core, fully direct
and semidirect is made by the program based on memory and disk limits and the
dimensions of the problem.
Forces the in-memory algorithm.
This is very fast when it can be used, but requires N4/4 words of memory.
It is normally used in conjunction with SCF=InCore. NoInCore prevents
the use of the in-core algorithm.
Energies, analytic gradients, and analytic frequencies. ROMP2 is available
for energies only.
MP3, MP4(DQ) and MP4(SDQ):
Energies, analytic gradients, and numerical frequencies.
and MP5: Analytic energies, numerical gradients, and numerical frequencies.
The MP2 energy appears in the output as follows, labeled as EUMP2:
E2= -.3906492545D-01 EUMP2= -.75003727493390D+02
higher-order Møller-Plesset methods follow. Here is the output from an
Time for triples= .04 seconds.
E3= -.10847902D-01 EUMP3= -.75014575395D+02
E4(DQ)= -.32068082D-02 UMP4(DQ)= -.75017782203D+02
E4(SDQ)= -.33238377D-02 UMP4(SDQ)= -.75017899233D+02
E4(SDTQ)= -.33794389D-02 UMP4(SDTQ)= -.75017954834D+02
The energy labelled EUMP3 is the
MP3 energy, and the various MP4-level corrections appear after it, with the MP4(SDTQ)
output coming in the final line (labeled UMP4(SDTQ)).