Gaussian 03 Release Notes

Revision D.01/D.02

Note: Revs D.01 and D.02 differ only in that the latter supports building a few additional platforms from source code. There are no feature differences between the two revisions.

This document lists changes to and additional information about Gaussian 03 functionality since the printed documentation was prepared. The manual refers to the first printing of the Gaussian 03 User's Reference; the second edition of this book has these changes incorporated.

List of Topics

New Features and Functional Changes in Revision D.01

  • It is now possible to do NMR spin-spin coupling calculations in two steps. NMR=Mixed requests this combination and causes two job steps to be run. In the first, the basis set specified by the user is modified to be appropriate for the Fermi Contact term, by uncontracting the basis and adding tight polarization functions for the core. In the second step, the other three terms in the spin-spin coupling are done with the standard basis. The final results reported at the end of the second job step include the Fermi Contact contribution from the first step. This significantly improves the accuracy of spin-spin coupling constants, especially when done with typical valence-oriented basis sets such as 6-31G** or Aug-CC-pVTZ.

  • NMR=Susceptibility requests that the magnetic susceptibility be calculated along with the shielding tensors.

  • The External facility has been enhanced to provide much more flexible control of the scripts to be run. It also supports use of analytic second derivatives in the external program. See for details.

  • There is a new link 0 command to control use of Linda workers:


    This lists the TCP node name for each node to use. By default, one Linda worker is started on each node, but the optional number allows this to be varied. For example, if the job is to be run on one 4-processor SMP box and on one 8-processor SMP box called then one should specify


    This would run one Linda worker on machine1, which would run 4-way SMP parallel on that machine, and two Linda workers on machine2, each of which would run 4-way SMP parallel.

    Usage notes:

    • Do not use the %NProcLinda command with %LindaWorkers. G03 will compute the total number of Linda workers based on the %LindaWorkers input.

    • The g03l script is no longer needed

    • Linda always starts a worker on the master node: the one on which the main g03 process is started. So if this node is not named in %LindaWorkers then it will be added by G03. Thus if G03 is started on machine0 with input


    Then the job would run 6 Linda workers, 3 on machine2, and one each on machine1, machine3, and machine0.

  • Stable=(Opt,QCOnly) can be used to suppress use of the regular SCF (L502) during later SCF calculations of the Stable=Opt iterations.

  • The TPSSTPSS and BMK functionals have been added.

  • In frequency calculations, the zero-point energy is included in the summary block (the archive entry) at the end of the output.

  • The route keyword Pseudo without options now defaults to Pseudo=Read.

New Hardware Support in Rev. D.01

  • The Intel EM64T processor is supported with this release. This processor and the AMD Opteron use the same version of Gaussian 03.

  • Red Hat Enterprise Linux Version 4 Update 2 is supported for the IA32, AMD Opteron, EM64T and IA64 environments (although the latter in binary form only: builds from source code fail due to bugs in the loader supplied by RedHat.). Note that only the unmodified version of the operating system as delivered in the downloadable ISO images is supported.

  • Parallel calculations are supported on Windows systems. Multiprocessors can excute calculations in parallel across CPUs within shared memory multiprocessor (SMP) systems (such as commodity dual processors). In addition, jobs can also be executed in parallel across discrete computers (nodes) connected via a local area network. The latter case requires the Linda product as well as the Gaussian 03W Multiprocessor version. The two types of parallel execution can also be combined within a single job.

    Install notes for the Gaussian 03W Multiprocessor version with Linda can be found here.

The parallel (multiprocessor) versions of Gaussian 03W are licensed and priced separately from the Single CPU version already available. See the web pages:
where x is a for academic pricing, c for commercial pricing, or g for US government pricing.

Bugs Fixed in Revision D.01

  • Problems involving setup of the dummy basis set done during MM calculations have been resolved.

  • The BMO input option to CubeGen incorrectly caused alpha rather than beta MOs to be selected.

  • A problem with the localization which affected CBS calculations on some atoms was fixed. Results for molecules are unchanged.

  • A problem with post-SCF counterpoise calculations when some of the fragments had no correlation energy (e.g., Li+ with MP2(FC)) was fixed.

  • Problems with the ATLAS libraries for IBM AIX systems which caused some post-SCF jobs using large amounts of memory to fail have been corrected.

Updated Manual Documentation

This section refers to errors in the printed manual, rather than the online version.

  • The Opt=CheckCoordinate option is not implemented.

  • On page 22 of the manual, the correct keywords to compute Hyperfine spectra tensors are: Freq=(VCD,VibRot[,Anharmonic]). Note that g tensors are also available from NMR calculations.

  • In the discussion of the Pseudo keyword, both the default values for La in the SDD table should be MWB28. Also, the column header for the first default column should be IOp(3/6=6).

  • The ONIOM example on page 151 of the manual should be used for illustrative purposes only. We thank Prof. K Nishimoto for pointing out several scientific problems with running this calculation.

  • ADMP is available for semi-empirical, HF and DFT methods only (page 37). The Int=AM1, Int=MNDO or Int=PM3 keyword is required for ADMP jobs using semi-empirical methods. If you require a spin-unrestricted wavefunction, include the UHF keyword as well.

  • When using a density fitting basis set as on page 28, slashes must be placed between the method and basis set and between the basis set and density fitting basis set.

  • The reference for the APT charges included in the output for some Gaussian calculations is:  J. Cioslowski, JACS 111 (1989) 8333.

  • The official AMBER website noted on page 133 is now

  • The mPBE functional on page 74 is not implemented.

  • The default SCRF method is IEFPCM. It is erroneously given as SCIPCM on page 205.

  • The manual mentions on page 178 an option related to MP2 with PBC (NCellE2=N), but this is not in the released Gaussian 03.

  • The manual implies that CASSCF (SpinOrbit) computes the coupling between two consecutive states on page 49. This is incorrect; a line of input is read giving the two states to couple. The location of this input line is given correctly in the table on page 19.

  • The coefficients for the states in state-averaged CASSCF on page 49 are in fixed format (10 characters per coefficient: nF10.8), not free format.

  • The manual refers to ZINDO/1 on page 223, but the parametrization actually used in Gaussian 03 (and Gaussian 98) is the spectroscopic one, ZINDO/S.

  • On page 206 under "Availability and Restrictions", it should be stated that frequencies are possible with SCRF=IEFPCM (the default SCRF method) as well as with SCRF=CPCM; the manual does not include frequencies in the list of allowed SCRF jobs.

  • The manual gives an example route on page 65 containing “CIS(D)=Read”, but one must use the syntax CIS(D,Read). The manual also implies that densities are available with CIS(D), which they are not.

  • The following refers to the discussion of DFT (page 73):  There is no keyword for the G1LYP functional. There is no OPTX keyword, since this exchange functional is intended to be combined with a correlation functional.

  • The option “ONIOM=MKS” on page 149 should read ONIOM=MK. It is the default, not Mulliken.

  • BOMD is available with CIS in addition to the other methods listed in the manual (page 45).

  • SCRF=IPCM jobs can be restarted from the read-write file by using the Restart keyword, not from the checkpoint file as stated in the manual (page 207).

  • Contrary to what is stated in the manual on page 179, Polar=EnOnly is required to compute hyperpolarizabilities for DFT methods. Also, MP2 Polar=Analytic will compute only polarizabilities. Finally, the example job on page 180 should use the HF method rather than B3LYP.

  • The MaxStep=n option with ADMP (page 36) and BOMD (page 44) is incorrect. The correct option is StepSize=n.

  • The second IRCMax example on page 129 should be changed to:

    # IRCMax (MP2/6-31G(d):HF/3-21G*,Zero,Stepsize=10)

    All references to the 3-21G(d) basis set should be to 3-21G* and ones to the CBS-4 method should be to MP2/6-31G(d) instead.

  • Freq=ReadIsotopes (page 88) now uses the temperature, pressure, and scale factor specified on the route card, and reads this line with the isotope input only if none of the three were given on the route card.

  • PCM solvation can now be used with the MNDO, AM1, and PM3 semi-empirical methods. The route card must specify Int=AM1 rather than just AM1 if SCRF is also specified.

  • The original implementation of the mPW exchange functional was not consistent with the equations in the paper: the local scaling factor was applied in computing the non-local correction. The version of mPW in Gaussian 03 corrects this error, but since the parameters in the original paper were optimized with the incorrect functional, the correct one cannot reproduce the original results. We have added OmPW ("old mPW") as a functional which includes the error and is thus consistent with Gaussian 98 and the results (but not the equations) in the original paper. OmPWPW91 is thus equivalent to mPWPW91 in Gaussian 98, etc. The hybrid functionals on OmPW3PBE, OmPW1LYP and OmPW1PW91 are also defined.

  • On page 77, the parameters used to set the coefficients for general hybrid functionals are now set by options to overlay 3 rather than overlay 5, and some now provide one more decimal digit in the specification:

    Gaussian 98
    Gaussian 03
    IOp(5/42=N) IOp(3/74=N)
    IOp(5/45=K) IOp(3/76=K) where K<0

Updated Input Ordering Table

The table on page 19 has been updated:

Section Keywords Final blank line?
Link 0 commands % commands no
Route Section (# lines) all yes
Extra Overlays ExtraOverlays yes
Title section all yes
Molecule specification all yes
Connectivity specifications Geom=Connect or ModConnect yes
Modifications to coordinates Opt=ModRedundant yes
2nd title and molecule specification Opt=QST2 or QST3 yes
Connectivity specifications for 2nd
set of coordinates
Geom=Connect or ModConnect and
and QST2 or QST3
Modifications to 2nd set of coordinates Opt=ModRedun and QST2 or QST3 yes
3rd title and initial TS structure Opt=QST3 yes for both
Connectivity specifications for 3rd
set of coordinates
Geom=Connect or ModConnect
=(ModRedun, QST3)
Modifications to 3rd set of coordinates Opt=(ModRedun, QST3) yes
Atomic masses IRC=ReadIsotopes yes
Frequency of interest CPHF=RdFreq yes
Molecular Mechanics parameters HardFirst, SoftFirst, SoftOnly, Modify yes
Initial force constants (Cartesian) Opt=FCCards yes
Accuracy of energy & forces Opt=ReadError no
BOMD/ADMP input (1 or more sections) ADMP and BOMD yes
Basis set specification Gen, GenECP, ExtraBasis yes
Basis set alterations Massage yes
Finite field coefficients Field=Read yes
ECP specification ExtraBasis, Pseudo=Cards, GenECP yes
Density fitting basis set specification Extra Density Basis yes
Background charge distribution Charge yes
Symmetry types to combine Guess=LowSymm no
Orbital specifications (separate α & β) Guess=Cards yes
Orbital alterations (separate α & β) Guess=Alter yes
Orbital reordering (separate α & β) Guess=Permute yes
PCM solvation model input SCRF=Read yes
Weights for CAS state averaging CASSCF=StateAverage no
States of interest for spin orbit coupling CASSCF=Spin no
# Orbitals/GVB pair GVB no
Atoms list for spin-spin coupling constants NMR=ReadAtoms yes
Alternate atomic radii Pop=ReadRadii or ReadAtRadii yes
Data for electrostatic properties Prop=Read or Opt yes
Cube filename (& Cards input) Cube yes
NBO input Pop=NBORead no
Orbital freezing information ReadWindow options yes
OVGF orbitals to refine OVGF=ReadOrbitals yes
Temperature, pressure, atomic masses Freq=ReadIsotopes no
PROAIMS/Pickett output filename Output=WFN or Pickett no

New and Updated References

The following reference citations have been updated/corrected:

495  W. H. Miller, R. Hernandez, N. C. Handy, D. Jayatilaka, and A. Willets, Chem. Phys. Lett. 172, 62 (1990).

472  J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865-3868 (1996).

The following new references should be cited when using the anharmonic frequency features in Gaussian 03 (Freq=Anharmonic):

577     V. Barone, J. Chem. Phys. 122, 014108 (2005).

578     V. Barone, J. Chem. Phys. 120, 3059 (2004).

New Features and Functional Changes in Revision C.01

  • Raman Optical Activity can now be calculated at the HF and DFT levels of theory. By default, this is done for the static limit, but the CPHF=RdFreq keyword indicates that one or more incident light frequencies should be read in and used in the electromagnetic perturbations. Request this calculation with Freq=ROA.

  • The O3LYP and TPSS functionals are included.

  • DFT XC quadrature now uses a dynamic allocation of work in SMP-parallel jobs. Coulomb and exact exchange also use dynamic allocation for PBC calculations and for calculations on large molecules (when FMM is used, by default for more than 60 atoms).

  • The triples calculation in MP4, QCISD(T), CCSD(T), and BD(T) now uses dynamic work allocation for better SMP-parallelism.

  • MP2 energy and gradient calculations can now use both SMP and Linda parallelism in the same job.

  • NMR spin-spin coupling constants can now be calculated for selected atoms rather than all at once.  Use IOp(10/48=10) to request this. The input is a free format list terminated with a blank line. Ranges of atoms (e.g. 7-12) are allowed.

  • The program can now address single scratch files up to 16 GB under IA32/Linux and Windows, so there is no need to split the RWF file into separate files each < 2 GB.  The limitation to 16 GB total scratch space inherent in 32-bit integers remains.

  • The threshold for turning on use of linear scaling algorithms for Coulomb and exact exchange can now be controlled on the route line or in the Default.Route file.  Int=FMMNAtoms=N specifies that calculations on molecules having no symmetry and at least N atoms will go through the linear scaling code. The default is 60 atoms. Molecules with symmetry have higher crossover points and the threshold is increased accordingly, to 120 atoms for C2 and Cs point groups and 240 atoms for higher symmetry.

    Note that the linear scaling code is SMP parallelized but not Linda parallel, so that for calculations on medium-sized molecules (60-150 atoms) on clusters it may be preferable to use the non-linear-scaling code in order to run in parallel.

  • NewZMat now converts between Cartesian and fractional coordinates for 1D and 2D periodic systems.

  • Jobs which fail due to insufficient memory now give a more informative error message.

  • PBC jobs now check whether the same atom has been specified in more than one unit cell.  PBC jobs now print out the distance matrix with all atoms translated to the central cell.

  • The efficiency of large HF and DFT jobs on vector machines such as the NEC SX was improved.

  • Jobs which do numerical differentiation of second-order properties, such as Freq=Anharmonic, Freq=Raman CPHF=RdFreq, or Freq=ROA, now default to a larger DFT integration grid in the CPKS step (SG1 instead of CoarseGrid).

  • More data is stored in one area on the checkpoint file as a result of implementing ROA. C8603 has been updated to convert G03 Revision B checkpoint files to the new allocation.  This update is only required before running the Revision C formchk or freqchk using old checkpoint files.  Revision B checkpoint files will work as is when used as input to new Gaussian 03 calculations, i.e. with Geom=Checkpoint, Guess=Read, etc.

Functional Changes in Revision B.05

  • Counterpoise calculations now check whether the fragments conform to the point group of the complex and turn off symmetry if necessary (page 68).

  • Limitations on the maximum length of file names to utilities such as c8603, freqchk, formchk, and unfchk have been increased to 1024 characters. See Chapter 5.

  • The printing of g-tensors was made more self-explanatory.

  • The route generator now checks for illegal combinations of CIS with various SCF methods, such as CASSCF or DFT.

  • PBC jobs (page 177) now print information about the energies near the Fermi level and the direct and indirect band gaps.

Functional Changes in Revision B.04

  • Polar=OptRot and Freq=Anharmonic now default to CPHF=SG1 rather than the usual CPHF default of CoarseGrid. This gives somewhat better accuracy for the numerical derivatives (page 69).

  • PBC calculations with external point charges (via the Charge keyword) do not work and now produce an error message (page 60).

  • The generation of the cell list (page 177) was changed slightly to make it identical on different types of hardware.

  • The default iterative diagonalization method for large CASSCF active spaces was changed from Lanczos to Davidson, which is more reliable (page 50).

Bugs Fixed in Revisions C.01 and C.02

The following bugs were fixed in Rev. C.02:

  • A problem with PBC forces on the Sun and SGI/MIPS was fixed.

  • A problem with GUESS=READ for jobs using ONIOM and a semi-empirical low-level calculation on open-shell systems was fixed.

The following bugs were fixed in Rev. C.01:

  • A wrong coefficient in the B1B95 hybrid functional.  A new value IOp(3/74=-17) specifies the correct coefficients for this functional.
  • A bug which caused Opt Freq jobs to fail in the frequency step when the B971 and B972 functionals were specified.

  • A problem in using integers instead of real numbers for input to freqchk when the input was provided on the command line.

  • The internal coordinates were numbered correctly but labelled incorrectly in the output from Freq=Internal.

  • A bug which caused PCM second derivatives with Linda parallelism to fail.

  • Cubman would erroneously print one title line from the first file instead of that from the second file.

  • NewZMat unnecessarily reordered the atoms when -ocart was specified.

  • Freq=EnOnly and Polar=EnOnly with ONIOM now save and restore MOs for Guess=Read properly.

  • The PBC=NCellMin=N option did not always increase the number of cells considered to at least N.

Bugs Fixed in Revision B.05

  • A problem with ZINDO calculations on Cu and Zn has been fixed.

  • Problems preventing UFF calculations with atoms heavier than Xe and on single atoms has been corrected.

  • An error in Linda-parallel DFT frequency calculations was fixed.

  • A problem which could cause consistency failures in Linda-parallel MP2 has been fixed.

  • An allocation failure for PBC calculations having only one or two basis functions in the unit cell was corrected.

  • Termination during basis set initialization with the EPR-III basis set was fixed.

  • Checkpoint files converted with c8603 could be used by G03 and freqchk but not by formchk. This has been corrected.

  • An error in the D95 basis set for Be was fixed. This caused jobs to fail on machines which check for a divide by zero. The D95V basis for Be was already correct.

  • Freq=(ReadFC, ReadIso) now works properly, as does CBS-QB3 (Restart, ReadIso) and restarts after job completion of other compound jobs. Note:

  • These jobs ignore temperature, pressure, and scale factors on the route card and always read these with the isotopes.

  • Restarts of compound jobs which stopped during the calculation ignore ReadIso; this option only works when restarting from a completed calculation to recompute the final quantities.

  • A problem with using Opt=QST2 and Opt=QST3 with PBC was fixed.

  • ONIOM data was not updated by c8603. This has been corrected.

  • chkchk sometimes did not report that force constants were present when they were.

  • When a list of points was provided for cubegen to use in evaluating the electrostatic potential, the coordinates echoed in the output file were incorrect. The potential values (and electric field, etc.) were correct.

  • The dummy basis set used during molecular mechanics calculations was not defined for some elements, preventing their use with force fields. All elements should now work with molecular mechanics.

  • The option PBC=NCellK=MMM had no effect when it should have set the number of cells used for exact exchange to MMM. This has now been corrected.

  • ECPs with unusual numbers of electrons (usually odd numbers) in the core caused problems in generating the initial guess. These problems have been fixed.

  • PCM optimizations using the options to add spheres on one or more hydrogens (which are sometimes needed when the cavity changes during a reaction) did not preserve the added spheres at later points of the optimization, and has been corrected.

  • A bug in the random number generator under AIX used for the messages at the end of the job has been corrected.

  • A bug in NMR and TDDFT calculations when using HF or hybrid functionals (but not pure functionals) along with FMM was fixed. With default options, this only affected jobs having sixty or more atoms.

  • An allocation bug which caused calculations involving derivatives of I and higher basis functions (e.g., with the cc-pV5Z basis) to fail on Intel IA32 machines has been fixed.

  • Various compiler bugs which caused problems with newzmat, formchk, and freqchk under IA32/Linux and Windows were fixed.

  • Two routines were changed to avoid a bug in version 5 of the Portland Group compiler. Several script files and the IA32 makefile were updated to account for the change in directory structure of this compiler.

Bugs Fixed in Revision B.04

  • A problem with giving users access to the Gaussian 03 directories via a secondary group has been fixed. Using primary or secondary group access is the recommended way to provide access. Making the Gaussian 03 directories world accessible (i.e. by all users) is a security problem, is prohibited by the license, and the program will now refuse to run if there is world access.

  • A problem which caused some NMR calculations with CSGT to fail was fixed.

  • A problem which caused some PBC calculations involving three-dimensional PBC and translation vectors having angles greater than 90 degrees to oscillate wildly in the SCF was corrected.

  • ONIOM Freq=Numer and Polar=Numer calculations did not leave the correct information on the chk file for a subsequent ONIOM calculation with Geom=Check. This was corrected. Non-ONIOM calculations always had the correct information.

  • A problem with setup of CASSCF at later points of geometry optimizations for large active spaces was corrected.

  • A memory allocation problem which caused some calculations using pure DFT functionals and highly contracted f functions to stop on a consistency failure was fixed.

Bugs Fixed in Revision B.03

  • A problem with NMR shielding calculations using pure DFT functionals (not HF or hybrid DFT) on open-shell molecules was fixed.

  • A problem with anharmonic frequency calculations on molecules with many atoms was fixed.

  • ONIOM calculations in which the model system did not have the point-group symmetry of the real system would fail in link 401. While these conditions are almost always the result of a poorly chosen model system, Gaussian 03 now recognizes this condition and turns off symmetry, so that the calculation proceeds as requested.

  • OVGF calculations had been incorrectly defaulting to a different approximation. This was corrected.