User Tools

Site Tools


Compiling Gaussion09 with the Intel compiler

Compiler and linker settings need to be done in several files in g09/bsd:

The compiliation procedure discussed in this blog has been followed.

Note: In order to use Gaussian the user has to be member of the Gaussian user group. Please, contact us for adding you to the user group.

VSC-3: submission of Gaussian jobs

Sample submit script on VSC-3 for slurm,

#SBATCH --tasks-per-node=16
#SBATCH --ntasks-per-core=1

module load gaussian/RevD.01


Submit this script to slurm via:

$ sbatch

Benchmarks for Gaussian 09 on VSC-2

Gaussian 09 Revision C.01 Release Notes

Changes between Gaussian 09 Revisions B.01 and C.01:

  - Changes to optimization algorithms and options:
    * The selection of modes to include when stepping down from a region of wrong curvature during an optimization has been improved. This can also now be controlled by route options:
    * Opt=NoDownHill ... don't try to go downhill, just take RFO-like step.
    * Opt=NGoDown=M ... Mix at most M eigenvectors in taking a downhill step.  The default is 3.
    * Linear bends are handled more reliably, and included in internal coordinate more frequently, than before. 
      This avoids many optimization problems involving nearly linear angles becoming exactly linear.

    c.  The connectivities of reactant and product are now merged in
        generating the internal coordinates for the TS during QST2 and
        QST3 optimizations.

    d.  The maximum number of steps allowed ever in an optimization (i.e.,
        include later restarts) can be reduced.  This is sometimes useful
        for very large systems in order to reduce memory and disk usage.

    e.  The program now checks if the standard orientation of a molecule
        has flipped by 180 degrees during an optimization and avoids the
        flip.  This avoids jumps when animating optimizations, IRCs, etc.
        in GaussView and improves SCF convergence.

    f.  The memory allocation for generation of internal coordinates is
        now proportional to the amount of memory provided by %mem.  This
        allows jobs with very large numbers of atoms or internal
        coordinates which previously failed to run if enough memory is

    g.  By default, internal coordinates for potential hydrogen bonds are
        not generated automatically.  Bond coordinates are still added to
        connect otherwise disjoint fragments, so coordinates for hydrogen
        bonds which connect fragments will still be included.

 2. Single-point BD calculations now default to frozen-core, with the
    core orbitals uncorrelated but updated using the BD Fock matrix.
    The previous default was to leave the core orbitals unchanged from
    the HF values, or from the orbitals read in with BD=Read.  The new
    default produces energies which are independent of the choice of
    starting orbitals.  Gradients with BD still require and defaults to
    full rather than frozen-core.  The OldFCBD keyword requests the
    old-style frozen-core.

 3  The memory required by very large ONIOM(MO:MM) and pure MM frequency
    calculations has been reduced.

 4. On some machines fully direct integral transformation and fully direct
    MP2 are chosen if there is a large amount of memory, but the semidirect
    algorithms are faster.  Tran=SemiDirect in the Default.Route file now
    forces the SemiDirect algorithm for MP2 as well as the transformation
    in higher level post-SCF calculations.  (All method keywords such as 
    MP2 are ignored in the Default.Route file, because otherwise they
    would force that model in all calculations.)
 5. Output=Wfn and Output=WfX with post-SCF methods now defaults to
    Density=Current and Pop=NOAB, both of which are necessary for
    the post-SCF density to be stored in the .wfn/.wfx file.  Problems
    with the orientation of the forces in these files and in generating
    them with ROHF wavefunctions and/or linearly dependent basis sets
    have also been fixed.  Core densities are stored in the wfx file
    for calculations which use ECPs, so that AIM and other analysis can
    be done correctly for these cases.

 6. Several customers have used the file generated for COSMORS as input
    to their local simulation programs, so this capability has been put
    back into G09.

 7. Polar=Gamma has been added as more descriptive option for requesting
    second hyperpolarizabilities.  It is synonym for Polar=(DCSHG,Cubic).

 8. The definition of improper torsions in the Amber force field is dependent
    on the ordering of atoms in the molecule.  Calculations in the Amber 
    program on typical proteins are consistent because of the standard
    ordering of atoms withing residues and residues within a PDB file, but
    for general molecules produced with GaussView the results depend on
    the order of atoms in an arbitrary way.  G09 has been changed to average
    over the 6 possible orders of atoms in an Amber improper torsion, making
    the results slightly different than the standard Amber force field, but
    making the energy independent of permuations of atoms in the molecule.

 9. The RevTPSS exchange and correlation functionals have been added.

10. SDD now defaults to more recent basis sets for Actinides; OldSDD requests
    the previous default.

11. Printing during Pop=MK with IOp33 increased has been restored to include
    the data required for RESP charge fitting.  However, G09 can now generate
    the data file for AnteChamber directly, by setting IOp(6/50=1) in the
    Pop=MK job, and this is the recommended method for generating input for

12. A bug in CIS frequencies with PCM solvation was fixed.

13. MaxDisk in a Default.Route file now applies to all steps of a compound
    job; previously, only the first step was defaulted properly.

14. A bug which prevented reading AlpB parameters for AM1 was fixed.

15. Convergence during SCVS calculations is now checked more carefully.
    Refer to the input files for tests 935-939 and 945 for examples of
    using SCVS.

16. TB and TW can now be used to specify memory and disk allocations in
    units of terabytes and terawords, respectively.

17. Pop=SaveBio in Stable=Opt jobs caused the stability calculation to
    be wrong or fail.  This now works properly, saving the biorthogonal
    orbitals only after the wavefunction has been made stable.

18. External point charges now work with symmetry turned on.

19. A bug in TD-DFT gradients with frozen core was fixed.

20. Print statements for NMR shielding were fixed to work with
    more than 999 atoms.

21. A bug in DFTB using interpolated (not analytic) parameters with
    d functions was fixed.

22. A rare problem with uncompleted write statements on slow file
    systems was fixed.

23. Problems with some combinations of charge and multiplicity in
    fragments during Guess=Fragment calculations were fixed.

24. Printing of Coriolis terms during Freq=VibRot was restored.

25. Some memory allocation problems for PBC calculations with large
    unit cells were fixed.

26. Inconsistencies in how the geometry was modified in some cases by
    Symm=Loose were fixed.

27. A bug in the ROMP4 triples energy when NoSymm was specified was fixed.

28. A warning "The extrapolated energy is higher than the direct energy"
    is no long printed unnecessarily by the CBS extrapolation. 

29. ONIOM(MO:MM) jobs which do microiterations and which fail to finish
    are now restartable.

30. A bug in reading ECPs with ONIOM when the same ECP was placed on
    multiple centers was fixed.

31. The combination of IRC and Freq, which did the frequency calculation
    at the last point of the IRC rather than the TS, is now rejected.

32. Several unsupported combinations of Douglas-Kroll-Hess with properties
    now generate an error message rather than incorrect answers.

33. A bug in generating the default (Harris) initial guess when using ECPs
    on charged species was fixed.  The quality of the initial guess when
    using ECPs has also been improved.

34. Several defaults for whether to use FMM and other integral options
    have been updated for better performance on current models of CPU.

35. FormChk now writes -1 rather than ****** to the formatted checkpoint
    file if the value exceeds 10^13-1.  This allows unfchk and other
    utilities to process the resulting fchk file.

36. A bug affecting geom=check after numerical frequencies when using
    ONIOM was fixed.

37. The Direct option is available for SAC-CI.  This requests an
    integral-direct algorithm suitable for larger molecules.

38. A %oldchk link0 command has been added.  The contents of the
    checkpoint file specified by %oldchk are copied to the checkpoint
    file of the current job step at the start of the job step.
    This allows data to be picked up from a previous calculation 
    without destroying anything on the chk file from it.

39. The combination of BD or W1BD with SCRF, which does not work
    correctly, is now rejected by the route generator.

40. A new version of the ATLAS blas library is used on most platforms.
    This fixes several problems when using very large amounts of memory.
    In the event of such problems, IOp1=NoAssem can now be specified on
    the route line to turn off use of the ATLAS matrix multiplication

41. A problem in reported transition moments between excited states
    computed in SAC-CI jobs was fixed.

42. Empirical dispersion with DFT and ghost atoms now runs.  Empirical
    dispersion and PBC now produces an error message, since it is not

43. "Opt Freq" with ROHF/RODFT now works correctly, doing Freq=Numer
    with the restricted open-shell wavefunction in the second job step.

44. Franck-Condon calculations now function correctly for forbidden

45. The route generator rejects the combination of TD and double-hybrid
    DFT methods, which never worked (previously, TD was done based on
    only the SCF part of the double-hybrid).

46. IRC=(RCFC,GradientOnly) calculations now correctly use the Hessian
    from the chk file.

47. Diffuse (aug-) functions were added for cc-pVDZ for the first
    transition row.

48. A memory allocation bug for very large systems which could cause
    a failure with the message "NIJ > Max2 in MMCore" was fixed.

Changes specific to IBM AIX/Power machines:

1.  When building from source, the default is to build for the
    current (Power7) processors.  To compile a version for the
    Power5 or Power6 machines, instead of just "bldg09" you must
    give the command "bldg09 all ibmp5".

Changes specific to the Windows versions:

1.  A Windows64 version is now available.

2.  The external keyword functions correctly.  Look at
    g09\tests\com\test726.gjf for an example of using it.

3.  A problem with the G09W front-end writing out multi-step
    jobs when the --Link1-- lines were truncated was fixed.

Corrections to deprecated features:

1.  Problems with setting non-integer nuclear charges in Massage input have
    been corrected.  There is now a ZNuc function in Massage input which
    changes the nuclear charge but not the atomic number.

2.  A problem with AIM analysis on Windows only was fixed.

Changes between Gaussian 09 Revisions A.02 and B.01:

1.  A bug in MP2 frequencies with PCM was fixed.

2.  An updated version of the SAC-CI code is included.  This includes
    a new integral-direct algorithm -- SACCI=(Direct,...) -- which is
    much faster for large systems.

3.  The ExtraOverlay route keyword did not function in A.02; this has
    been corrected.

4.  "Opt Freq" with TD now runs both job steps properly.

5.  NewZMat now writes out secondary structure information, if present,
    with -opdb.

6.  NewZMat can now merge data from two input files.  Either two text
    files or an input and a checkpoint file can be merged.  See the
    website for examples.
7.  Problems with the dummy basis set used with molecular mechanics
    when the system was highly charged or very high spin were fixed. 

8.  Polar=(Cubic,DCSHG) can now be used to numerically differentiate
    frequency-dependent hyperpolarizabilities (betas) to produce second
    hyperpolarizabilities (gammas).  These polarizabilities are now
    printed in the standard coordinate systems (i.e., with components
    of beta along and perpendicular to the dipole moment).

9.  WfnX files, used by the newer versions of AIMPAC, can now be written
    via Output=WfnX.

10. Performance for very large MM systems (>20K atoms) has been improved,
    especially when range limits are applied to the Coulomb and 
    Van der Waals terms.  There is a new route option, Geom=Huge, which
    turns off various actions, useful in QM calculations but
    unnecessary and expensive with enormous MM runs.

11. MaxDisk can now be specified in the Default.Route file.

12. The free-format input routines have been generalized in order to make
    data from newer DFTB parameter files acceptable.  These files still
    require some modification to be used with G09; refer to the web site
    for details.

13. The full tensors for ECD using TDDFT (including the quadrupole component)
    are now printed.

14. The Hu, Lu, and Wang charge fitting model (JCTC 3, 1004-1013, 2007)
    is now available via Pop=HLY.  The authors only parametrized the
    atomic densities required for the model for the first 18 elements.
    An alternative version, Pop=HLYGAt, uses the HLY fitting scheme but
    with Gaussian's standard atomic densities, which are available for
    the entire periodic table.  For systems which can be done either way,
    the difference in atomic charges is usually between 1% and 5%.

15. The SCVS method of Todd Keith, which scales the molecule in order to
    make the virial condition satisfied exactly, has been added.

16. The use of IOp's to specify user-selected ranges for integrals has
    been updated in order to make it more general.

17. The default algorithm for optimizations when minimizing in a region
    of incorrect curvature has been improved.

18. The initial guess for AM1 and PM6 has been improved.

19. More analysis of input ONIOM and MM parameters with respect to secondary
    structure is now done (by default for systems with <10,000 atoms when
    secondary structure information is available).  The net MM charges on
    residues and average distances between residues are reported.

20. Various performance improvements, including ones for larger numbers
    of SMP processors and for SCF frequency calculations.

Usage Notes:

1.  If CIS frequencies are to be used with the Herzberg-Teller or
    Franck-Condon-Herzberg-Teller analysis, the CIS frequencies must
    be done numerically (Freq=Numer rather than Freq).  This is because
    the transition dipole derivatives are not computed during the
    analytic force constant evaluation.  The corresponding HF frequency
    calculation on the ground state, which is also required, can be done
    analytically as usual.

2.  CIS and CASSCF frequencies with PCM solvation must also be done numerically
    using Freq=Numer.

3.  The linear scaling (FMM-based) algorithms are now Linda-parallel, so Linda
    parallel jobs on large molecules do not need to specify NoFMM, and will run
    faster with the default algorithms chosen by the program.

4.  Opt=GDIIS is still present but deprecated; the new default optimization
    algorithm (Opt=GEDIIS) is better than GDIIS for the few cases where GDIIS was
    better than the G03 default (Opt=RFO).

5.  Optimizations of large molecules which have many very low frequency vibrational
    modes with DFT will often proceed more reliably when a larger DFT integration
    grid is requested (Int=UltraFine).

6.  Density fitting can be made the default for jobs using pure DFT functionals
    by adding the DenFit keyword to the route (-#-) line in the Default.Route file.
    Fitting is faster than doing the Coulomb term exactly for systems up to several
    hundred atoms (depending on basis set), but is slower than exact Coulomb using
    linear scaling techniques (which are turned on automatically with exact Coulomb)
    for very large systems.

7.  The default IRC algorithm has changed; refer to the User's Guide for details.
    The default is to report only the energies and reaction coordinate at each
    point on the path; if geometrical parameters along the path are desired, these
    should be defined as redundant internal coordinates via Geom=ModRedundant or
    as input to the IRC code via IRC=Report=Read.

Changes in Usage and Defaults between Gaussian 03 and Gaussian 09:

1.  There are many changes in the PCM algorithms:

    a.  The default surface integration is new and gives continuous
        potential energy surfaces.  It is strongly recommended for all
        new studies.  The route option SCRF=G03Defaults restores most
        of the defaults to those in G03, but should be used only for
        comparison with previous calculations done using G03.

    b.  When using the default IEFPCM solvation method or SCRF=CPCM, 
        Gaussian 03 computed and reported non-electrostatic contributions
        to the solvation energy but did not include these in the energies
        and they were not included in the energies used for geometry
        optimizations, frequencies, etc.  By default Gaussian 09 does not
        compute these values at all.  

    c.  The new SMD solvation model is recommended for absolute solvation
        energies and other properties for which the non-electrostatic
        solvation terms are significant.  When SCRF=SMD is specified, the
        SMD non-electrostatic terms are included in the basic energies
        (the SCF energy reported in the "SCF Done" line, correlated energies,
        etc.) and are included in the geometry optimization and frequency
        calculations.  The non-electrostatic energy is also reported

        Absolute solvation energies should be computed by doing a gas-phase
        optimization and frequency calculation on the system, followed by
        the same calculations with SCRF=SMD or SCRF=(SMD,Solvent=...).

    d.  The SCFVac PCM input option has been removed.  If a preliminary
        gas-phase energy is desired, do this in a separate job step before
        the solvated calculation.

    e.  If the quality of the PCM integration grid is to be changed, the
        keyword PTSDens should be used in the PCM input section.  This specifies
        the density of integration points in points per Angstrom^2.  The
        old keyword TSAre specifies the area per tessera and hence implies use
        of the old integration scheme involving tesserae, which is inferior.
        The default is approximately 5 points per A^2, so PTSDens=10.0 will
        cause about twice as many points to be used compared to the default.

2.  MP and CC calculations now default to a partial transformation (Tran=IABC).
    This is faster on most systems, especially when several processors are used.
    A full transformation can be requested using Tran=Full.

3.  The default SCF convergence is 10^-8 on the density for all calculations,
    including single points.

4.  The physical constants used by default are those from the 2006 CODATA tables;
    those used in Gaussian 03 can be requested via Constants=1998.

5.  AM1, PM3, and PM3MM by default use the new semi-empirical code, which has
    proper analytic first and second derivatives but which gives slightly different
    total energies because it computes the overlap integrals via 6-Gaussian expansions
    rather than over Slater functions.  AM1=Old and Use=L402 both request use of the
    old (MOPAC 6) code, through the regular links or through link 402, respectively.
    The new code is strongly recommended except when comparison with results from
    Gaussian 03 is required.

6.  Stable=Opt defaults to the usual (L502) SCF procedure for the initial SCF but
    then uses SCF=QC for additional SCF calculations, if they are required.

Changes between Gaussian 09 Revision A.01 and A.02:

1.  The logic for handling extra negative eigenvalues of the Hessian
    during minimizations has been improved.

2.  The combination of DFT and General SCF, automatically turned on if
    DFT is requested along with Int=DKHSO, does not work and is now
    rejected by the route generator.

3.  ONIOM input is now checked for divalent link atoms.  The position
    of these atoms is ill-defined unless the distance scale factors
    are set to 1, and the model is usually poor if the scale factor
    is forced to be 1.  Since this input is normally an error, it is
    now rejected by link 101.  IOp 132 can be used to force acceptance
    of this input, but this is strongly discouraged.

4.  Semidirect integral transformation is the the default.  This
    code parallelizes better than the fully direct or in-core
    algorithms and is similar in speed on a single processor.

5.  A bug which caused ONIOM=InputFiles to fail when PDB secondary
    structure information was included in the input has been fixed.

Building from source code:

1.  Building Gaussian 09 with Linda requires Linda version 8.2; the
    executables will not build with previous version of Linda.

2.  Building on Intel Macs requires a case-sensitive file system.
    In order to build the ia32 version you must specify:
    bsd/bldg09 all mac32

    as there is no way for the build script to determine that it is
    running on a x86 rather than x86-64 machine and the default is to
    build for x86-64.

3.  When building from source on IBM Power systems, the default is to
    build for the current (Power7) processors.  To compile a version
    for the Power5 or Power6 machines, instead of just "bldg09" you must
    give the command "bldg09 all ibmp5".

Missing Reference:

The G09 User's Guide omitted the paper Clemente08 from the
bibliography.  It is:

F. Clemente, T. Vreven, and M. J. Frisch, in Quantum Biochemistry,
Ed. C. Matta (Wiley VCH, 2008).
doku/gaussian09.txt · Last modified: 2022/01/26 10:22 by goldenberg