Документ взят из кэша поисковой машины. Адрес оригинального документа : http://classic.chem.msu.su/gran/gamess/options.html
Дата изменения: Mon Mar 15 15:10:10 2010
Дата индексирования: Mon Oct 1 19:43:56 2012
Кодировка:
List of the PC GAMESS specific options up to PC GAMESS version 5.2

The PC GAMESS specific options up to PC GAMESS version 5.2

See the PC GAMESS/Firefly readme files for the most recent options.



Group name Option Value Description Comments
$CONTRL WIDE .True. Wide format for vectors in $VEC groups for both INPUT and PUNCH files Default is .False.
.False. Normal format (which is standard for PC GAMESS).
IREST 2 Now allows one to restart with the AOINTS file only, without DICTNRY. 1-e integrals will be recomputed if the old DICTNRY file does not exist.
MPLEVL 3,4 Requests calculation of Moller-Plesset energy corrections in the third (MP3) and fourth (MP4) orders of perturbation theory. See also the list of available options in the description of the relevant $MP3, $MP4 groups below, as well as MP3/MP4-related hints. This option is not supported by the PC GAMESS versions earlier than v. 5.1
MP3 and MP4 are currently implemented only for the RHF case.
$SYSTEM FASTF .False. Normal operation
.True. Turns on the usage of fast (non-fortran) sequential file access routines. This increases the overall performance, and allows PC GAMESS to handle large (> 2Gb) sequential access files under Windows NT. See also a more detailed discussion on the large files support in the PC GAMESS. Fast file handling is supported under NT, Win95/98, and OS/2.
Large files are supported under WinNT only.
This option is not supported by the PC GAMESS v. 4.0
Default is .True. under NT, Win95/98, and OS/2; and .False. under DOS
SAFMEM integer The size of the safety memory pool (in double precision words), which is used by several parts of PC GAMESS in the case when there is not enough dynamic memory to perform current operation. In some cases, the PC GAMESS tries to use slightly more memory than the amount available. Then, increasing SAFMEM can help. This option is not supported by the PC GAMESS v. 4.0
Default is 16384
LDAR integer The default value of LDAR parameter (the size of direct access file record) for $INTGRL, $GUGDM2, $TRFDM2, $CISORT, and $MCSCF groups. This option is not supported by the PC GAMESS versions earlier than v. 4.3
Default is 2045
TRUNCF .False. The existing files of non-zero length are not truncated when being (re)opened for writing. This may be useful in some cases, for (almost) all the needed disk space is permanently (pre)allocated at the beginning of your PC GAMESS run, which prevents other disk-intensive programs from causing PC GAMESS to die in future if the free disk space is exhausted. This option is not supported by the PC GAMESS versions earlier than v. 4.3
It has an effect only if FASTF is set to True.
.True. The files are truncated while being (re)opened for writing. This is the default behavior, which results in the faster write operations. Default is .True.
WSCTL .False. Normal operation. The operating system controls the amount of physical memory (the size of the working set) used by the PC GAMESS.
.True. The PC GAMESS tries to allocate the working set as large as the amount of memory needed for the current operation. Roughly speaking, the process working set is the amount of physical (not virtual!) memory that is allotted to the process. Thus, this strategy usually considerably reduces paging in the case of memory-demanded jobs. On the other hand, this can leave less physical memory available to other processes and to the operating system itself. See also FAQ for information concerning the possible problems (and how to solve them). This option is not supported by the PC GAMESS versions earlier than v. 4.4
It has an effect when running under WinNT only.
Default is .True.
MAXWS integer The upper limit for the size of the PC GAMESS working set, in DP words. The PC GAMESS will never try to allocate working set that is larger than this amount. The value of 0 means the automatic selection of an appropriate limit. Namely, the limit will be set to the value smaller by 16 MB than the amount of the physical memory installed (or to the half of the physical memory, if the latter value is larger). This is rather an aggressive strategy. This option is not supported by the PC GAMESS versions earlier than v. 4.4
It has an effect only when running under WinNT, and WSCTL is set to True.
Default is 0.
BLAS3 sleep Additional threads that were created by BLAS level 3 routines, are suspended when not in use. This option is not supported by the PC GAMESS versions earlier than v. 5.0
nosleep Additional threads are permanently active. This is exactly the behavior of the PC GAMESS versions earlier than v. 5.0.
Use this option only if you encounter some unexpected SMP-related problems with the default settings.
It is supported only by SMP-enabled PC GAMESS version running in SMP environment.
Default is sleep.
DECOMM .False. Old-style memory manager behavior. This is exactly the behavior of the PC GAMESS versions earlier than v. 5.1.
Use this option only if you encounter some unexpected problems with the default settings.
This option is not supported by the PC GAMESS versions earlier than v. 5.1
.True. Turns on the advanced memory management feature. The advanced memory management is enabled by default.
It is supported under NT, Win95/98, and OS/2.
FLUSH .False. Disables the usage of file cache flush operation. It is recommended to disable flushing only in the two following situations:
  • if the total size of all PC GAMESS working files is less than the available size of the OS file cache;
  • if the hardware RAID volume is used as the scratch storage.
  • This option is not supported by the PC GAMESS versions earlier than v. 5.1
    .True. Enables the file cache flush operation. PC GAMESS will flush the cache buffers of its working files into disk, when necessary. File flushing is enabled by default.
    It is supported under NT, Win95/98, and OS/2.
    L2SIZE integer The size of fast L2 cache per CPU, in Kbytes. This information is used by several parts of the PC GAMESS (e.g., by MP3/MP4 code) in order to select the computational strategy optimal for the specific hardware the PC GAMESS is running on.
    At present, the default value is 0 for all Pentium-optimized PC GAMESS versions, and 512 for Win32-specific Pentium Pro/II/III-optimized version. Change the defaults appropriately for your hardware configuration.
    This option is not supported by the PC GAMESS versions earlier than v. 5.1
    $INTGRL IREST 0 Normal operation
    1 Only 1-e integrals and integrals needed for Schwarz screening are calculated, 2-e integrals are read from the existing AOINTS file.
    2 Only the integrals needed for Schwarz screening are calculated, 2-e integrals are read from the existing AOINTS file. Default is 0
    -1,-2 Similar to 1 and 2, but 2-e integrals are never recalculated, even if the molecular geometry is changed.
    PACKAO .False. Normal operation
    .True. Turns on the additional packing of the AO integral file. Both indices and integrals are packed. This option reduces the size of AOINTS file by a factor of up to 2.5-3, thus saving the disk space and the CPU time. See also a more detailed discussion on the packing-related problems. Relevant only for NOPK=1 case (conventional integral storage mode, not in a supermatrix form) Default is .False. for the PC GAMESS v. 4.0,
    .True. otherwise
    PKTHR double precision A threshold to activate packing of 2-e integrals. Default is 0.01 for the PC GAMESS v. 4.0,
    107-ICUT otherwise
    $TRANS AOINTS DELETE Means that AOINTS file is deleted when the integral transformation is done By default, AOINTS is not deleted.
    IREST 0 Normal operation
    1 Means that only 1-e integrals are transformed, while 2-e integrals are read from the existing MOINTS file Default is 0
    2 Both 1-e and 2-e integrals are read from the existing MOINTS file
    -1,-2 Similar to 1 and 2, but the effect is permanent for the whole calculation.
    $CPHF MXCPIT integer When solving CPHF equations iteratively, MXCPIT controls the maximum allowed number of iterations. This option is not supported by the PC GAMESS versions earlier than v. 5.2
    The default value is 50.
    CPTOL double precision When solving CPHF equations iteratively, CPTOL controls the maximum allowed relative error of the approximate solution. This option is not supported by the PC GAMESS versions earlier than v. 5.2
    The default value is 10-5.
    $CIINP IREST see PC GAMESS documentation Now works at any stage if the files needed for the restart with the given IREST already exist.
    SKIP2E .False. Normal operation
    .True. The MOINTS file from the previous run is used. Only 1-e integrals are transformed, while 2-e integrals are read from the disk file MOINTS. ATTENTION: Be sure that MO's are EXACTLY the same as those used in the previous run! (i.e. the SAME vectors in GUESS=MOREAD, SCFTYP=NONE, etc...) This option is now obsolete, IREST in $TRANS should be used instead.
    Default is .False.
    $CISORT MOINTS DELETE MOINTS file is deleted after integral sorting. By default, MOINTS is not deleted.
    $GUGDRT NWORD integer Controls the memory usage during generation of DRT. It may be necessary to increase the default value of 180018 DP words when the active space used in CI/MCSCF calculation is too large.
    $GUGEM FASTCI .False. Normal operation
    .True. Turns on the packing of GUGA CI Hamiltonian file. Both indices and matrix elements are packed. This option reduces the size of WORK16 file by a factor of up to 2.7-3, thus saving the disk space and the CPU time. See also a more detailed discussion on the packing-related problems. This option is not supported by the PC GAMESS v. 4.0
    Default is .True.
    PKTHR double precision A threshold to activate packing of the matrix elements of GUGA CI Hamiltonians. This option is not supported by the PC GAMESS v. 4.0
    Default is
    107*CUTOFF
    PACK2 .False. Normal operation
    .True. Turns on the additional (secondary) packing of GUGA CI Hamiltonian file, which works in conjunction with the FASTCI method of compression. This is a lossless method, which generally reduces the size of WORK16 file by a factor of up to 3-6 additionally to the FASTCI squeezing. Thus, the total degree of compression can reach the factor of 10-20 and even more, especially for the non-FORS type CI. Hence, this option is very useful in the case of relatively large CI calculations (i.e. 50000 CSFs or more, up to several millions), saving the disk space and reducing the real execution time. See also additional hints on this topic. This option is not supported by the PC GAMESS versions earlier than v. 4.4
    Relevant only when FASTCI=.True.
    Default is .False.
    LCHAIN integer A parameter that controls the degree of compression achieved by the PACK2 packing method. The greater the LCHAIN value, the better the compression and the higher the CPU overhead. Note also that the most efficient packing is always achieved when LCHAIN=0, because this special value is simply an alias for the infinitely large LCHAIN. This option is not supported by the PC GAMESS versions earlier than v. 4.4
    Relevant only when PACK2=.True.
    Default is 7, which seems to be quite reasonable.
    DIRCI .True. Performs the direct GUGA CI calculation, in which the Hamiltonian matrix is not stored on disk, and the matrix elements are recalculated during each Davidson diagonalization step. This may consume considerably more time than normal GUGA CI calculation, but allows one to perform it in the case when the Hamiltonian matrix is too large to fit on disk. Starting from the PC GAMESS v. 5.0, this option is moved here from its previous location in the $CONTRL group.
    .False. Performs conventional GUGA CI calculation Default is .False.
    $MCSCF OPTACT .False. Normal operation
    .True. Activates minor changes in the MCSCF FULLNR converger. This may somehow help in the case of a slow convergence of FULLNR MCSCF for the non FORS-type wavefunction. This option is not supported by the PC GAMESS versions earlier than v. 4.3
    Default is .False.
    $MP2 GRDMET 1 Selects old style MP2 gradient calculations. This option is not supported by the PC GAMESS versions earlier than v. 4.3
    2 Activates the new MP2 gradient code, which is generally much faster. Default is 2
    DM2MET -3,-2,-1,
    0,1,2,3
    Selects one of seven! possible methods of MP2 DM2 calculations during MP2 gradient run.
         Methods -3, -2, and -1 are precisely the same as methods 3, 2, and 1, respectively, but they use an asynchronous disk I/O to eliminate I/O latency and to reduce the total execution time; for this reason, these methods are not supported under DOS. Method 0 is definitely the slowest one; which one is the fastest depends on the amount of the physical memory installed. Methods 2(-2) require the least CPU time, but probably not the least total time. Methods 3(-3) need the least memory. By default, method 1 is chosen, which seems to be a quite reasonable compromise. You can experiment with this option to find optimal settings for your particular task and environment.
    This option is not supported by the PC GAMESS versions earlier than v. 4.3
    Relevant only when GRDMET=2.
    Default is 1.
    MEMGRD integer The maximum number of words of the memory to be used in the MP2 gradient calculation. It controls only the last stage of calculations, namely, the evaluation of the gradient integrals themselves. Zero means the usage of all available memory. This option is not supported by the PC GAMESS versions earlier than v. 4.3
    Relevant only when GRDMET=2. The default value is 0.
    MXCPIT integer Controls the maximum number of AO CPHF iterations during the MP2 gradient runs. This option is not supported by the PC GAMESS versions earlier than v. 4.3
    The default value is 100.
    CPTOL double precision Controls the maximum allowed RMS error of the approximate solution of CPHF equations during MP2 gradient runs. This option is not supported by the PC GAMESS versions earlier than v. 5.2
    The default value is 10-10.
    MXRECL integer The maximum record size for direct access file used during the MP2 energy (method=3) and MP2 gradient runs. This option is not supported by the PC GAMESS versions earlier than v. 4.3
    The default maximum size is 65536 DP words.
    IOPARS integer Specifies the set of specific I/O optimization flags for the direct access files (DAF) used during MP2/MP2 GRADIENT runs. Actually, this is the 2-digit octal value. The least significant digit controls the general DAF writing operations, while the most significant one controls the behavior of 'READ-THEN-WRITE-TO-THE-SAME-LOCATION' operation. Each digit can have one of the following values:
    • 0 - no particular I/O optimization;
    • 1 - no particular I/O optimization, file cache write through mode;
    • 2 - optimizes file I/O for the sequential access;
    • 3 - optimizes file I/O for the sequential access, file cache write through mode;
    • 4 - optimizes file I/O for the random access;
    • 5 - optimizes file I/O for the random access, file cache write through mode;
    • 6 - optimizes file I/O for the random access with some locality;
    • 7 - optimizes file I/O for the random access with some locality, file cache write through mode;
         The default value for this option is 74 in the case of non-negative DM2MET, and 75 otherwise. If you have a plenty of RAM, or fast RAID, or your job creates DAF of a (very) small size, try to set IOPARS to 64 and see if the I/O speed changes. The optimal value may also depend on the OS used.
    This option is not supported by the PC GAMESS versions earlier than v. 4.3
    METHOD 2 Selects old style sequential UHF/ROHF MP2 energy integral transformation.
    -2 Activates a new sequential UHF/ROHF MP2 energy integral transformation code that requires approximately a half of memory used in method 2. This option is not supported by the PC GAMESS versions earlier than v. 4.3
    Default is -2.
    METHOD 1 Selects the usage of the new RHF/ROHF/UHF MP2 energy program. This program is intended to handle large systems (e.g., 500 AOs or more). It is direct, very fast, and requires much less memory as compared to other MP2 methods. It is definitely the method that is to be chosen for the large jobs. On the other hand, it requires the 2-electron AO integrals to be reevaluated four times. Thus, there can be a considerable overhead for the small jobs. This is the only reason why this method is not turned on by default. See also additional hints on this topic. This option is not supported by the PC GAMESS versions earlier than v. 4.4
    Default is 2 for RHF, and -2 otherwise (i.e. to use the old MP2 program).
    DIRECT .False. Selects the semidirect evaluation of the AO integrals during METHOD=1 MP2 calculations.
    .True. Selects the fully direct evaluation of AO integrals during METHOD=1 MP2 calculations. This somewhat reduces the amount of disk space needed, by the cost of the additional CPU time (which may be considerable). This option is not supported by the PC GAMESS versions earlier than v. 4.4
    Relevant only for the METHOD=1 case.
    Default is .False.
    PACKAO .False. Normal operation.
    .True. Activates packing of the AO integrals during METHOD=1 semidirect MP2 calculations, thus (slightly) reducing the amount of disk space needed. This option is not supported by the PC GAMESS versions earlier than v. 4.4
    Relevant only for the METHOD=1 semidirect case.
    Default is .True.
    MNRECL integer The minimum allowed record size (in terms of 12-byte elements) for the direct access file used during METHOD=1 MP2 calculations. Reducing the MNRECL value increases the job execution time to some degree, but decreases the amount of memory needed for calculations. It is not recommended to set MNRECL below 500. This option is not supported by the PC GAMESS versions earlier than v. 4.4
    Relevant only for the METHOD=1 case.
    Default is 20000.
    SVDISK .False. This allows the METHOD=1 ROHF MP2 program to use (in some rare cases) additional disk space in order to reduce the job CPU time. This option is not supported by the PC GAMESS versions earlier than v. 4.4
    .True. Disables the extra disk space usage. Relevant only for the METHOD=1 ROHF MP2 case.
    Default is .False.
    SPARSE .False. This options controls how the matrix-matrix multiplication is performed during the MP2 METHOD=1 runs. False means the usage of standard BLAS level 3 routines (the only strategy, which was implemented in the PC GAMESS 4.4 and earlier). This is the best choice when the list of two-electron integrals is dense enough, i.e., when the total number of nonzero integrals is less than the maximum of N4/8 only several times (up to 3-5). This option is not supported by the PC GAMESS versions earlier than v. 5.0
    Relevant only for the METHOD=1 MP2 case.
    .True. Selects the usage of special matrix-matrix multiplication routines. This strategy is optimal for sparse lists of two-electron integrals. By default, the PC GAMESS tries to choose the most appropriate strategy automatically.
    Note, the program choice is in some situations not the best one. This is especially true for SMP systems, or for the systems with the slow access to the main memory
    $MP3,
    $MP4
    NCORE integer Omits first NCORE occupied orbitals from the calculation. The default for NCORE is the number of chemical core orbitals. This option is not supported by the PC GAMESS versions earlier than v. 5.1
    NWORD integer Controls the memory usage during MP3/MP4 calculations.
    The default value for NWORD is 0 which means that all available memory is used.
    This option is not supported by the PC GAMESS versions earlier than v. 5.1
    CUTOFF double precision This parameter controls the precision, to which the contributions to the MP3 and MP4(D) energies due to the external exchange operator, are calculated. The default value for CUTOFF is 10-9. For MP3 energy, it is guarantied that the absolute value of the cumulative error will be less than the value of CUTOFF. For the MP4(D) part, the situation is usually the same, but, unlike MP3 energy, this is not always true, especially for the basis sets with partial linear dependence.
    Due to this reason, it is recommended to use stronger CUTOFF (e.g., 10-15), if the high-precision MP4 energy is required. This will increase the precision by the cost of additional CPU time.
    This option is not supported by the PC GAMESS versions earlier than v. 5.1
    SMPMET 1,2,3 Selects one of the three SMP-based strategies available for the calculation of the external exchange operator contribution to the MP4 energy (this option has no effect in the case of MP3 calculations).
       Method 1 is probably the best one, as it has very good SMP scaling properties. On the other hand, its memory requirements are proportional to the number of CPUs used.
       Method 2 does not require extra memory, but can be used only if the number of CPUs is even. Furtermore, its SMP scaling is generally worse than that of Method 1.
       Method 3 also does not require extra memory. It usually has the worst SMP scaling properties.
    This option is not supported by the PC GAMESS versions earlier than v. 5.1

    It is supported only by SMP-enabled PC GAMESS version running in SMP environment.

    Default is Method 1.
    SDTQ .False. Performs MP4-SDQ calculation. This is the only type of MP4 calculations that was available in the PC GAMESS version 5.1.
    This option is not supported by the PC GAMESS versions earlier than v. 5.2
    .True. Performs MP4-SDTQ (i.e., full MP4) calculation. Default is False, i.e., only MP4-SDQ energy is calculated.