Oliver Stueker
February 18th, 2020
%MEM=30GB {Link0 Commands}
%CHK=mycoolmolecule.chk
# B3LYP/6-31G(d) Opt Freq MaxDisk=400GB {Keyword lines}
{EMPTY LINE}
This is a very cool molecule {Title section}
{EMPTY LINE}
0 1 {Charge, Multiplicity}
H 0.0000000000 0.0000000000 0.0000000000 {Molecule spec}
O 1.0480000000 0.0000000000 0.0000000000
C 1.5064000000 1.3312900000 0.0000000000
H 2.5540780000 1.3053060000 -0.0000000000
H 1.1371360000 1.7931990000 0.8652100000
H 1.1371360000 1.7931990000 -0.8652100000
{EMPTY LINE}
{optional sections}Important: Don’t forget to include the following empty lines:
Molecules (elements and coordinates) can be specified in a number of different ways, using:
Further, it is possible to add more information such as:
MolSpec reference at: http://gaussian.com/molspec/
Z-Matrices are an alternative way to specify molecule geometries, which can be more intuitive to use for a chemist. They specify a number of distances (e.g. bond lengths), angles and dihedral angles between the atoms.
Indeed atom 3 can also be defined distance 3-1 and angle 3-1-2 and atom 4 via any permutation of the already defined atoms 1, 2 & 3.
Z-Matrix reference http://gaussian.com/zmat/
Z-Matrix of H2O2 with values within the coordinates (compact):
Z-Matrix of H2O2 using symbolic values:
The keyword line starts with a #-symbol either on it’s own or followed by a letter T, N or P. This controls the amount of information that printed to the Gaussian output file:
#N: “Normal” amount of output (same as #)#T: “Terse” (reduced) output#P: Print additional outputIn most cases the Model chemistry is defined by a combination of “method” and “basis set”. Examples of commonly used methods are:
HF (Hartree Fock),B3LYP (a common DFT method) andMP2 (second order Møller-Plesset expansion).Most methods can be prefixed with R (closed-shell restricted wavefunctions), U (unrestricted open-shell wavefunctions) or RO (restricted open-shell wavefunctions).
Links:
A basis set defines a set of basis functions, which are used to calculate the molecular orbitals. Common basis sets are:
STO-3G (minimal basis using “Slater Type Orbitals”),6-31G (a Pople-type split-valence basis set).6-31G(d) (same but with d-type polarization functions)cc-pVDZ (correlation-consisted polarized Double-zeta basis set)For some methods however, no basis sets can be specified, e.g.: PM7 (semi-empirical “Parametric Model number 7”) or G3 (“Gaussian-3” composite method).
Links:
The job type defines what kind of calculation should be performed:
SP – Single Point – just calculate the energy for the given coordinatesOPT – Geometry Optimization – typically an energy minimizationOpt=TS – Transition State searchOPT is the same as Opt or opt.Optimization, but as no other keyword begins with Opt, it can be abbreviated.Opt=Z-Matrix or or Opt=TS. These options can also be combined e.g. Opt=(TS,Z-Matrix).List of Gaussian Keywords: https://gaussian.com/keywords/
%MEM=30GB – request 30GB of memory
%NProcShared=4 – (or %NProc=4) request a job for 4 CPUs
%CHK=/scratch/username/mycoolmolecule.chk – set checkpoint filename
Gaussian Link0 commands: http://gaussian.com/link0/
Order of precedence: Link0 > command line > environment variable
| Link0 (%-lines) | command line | environment variable | Note |
|---|---|---|---|
%Mem=30GB |
-m=30GB |
export GAUSS_MDEF="30GB" |
* |
%NProc=4 |
export GAUSS_PDEF="4" |
† | |
%CPU=0-31 |
-c="0-31" |
export GAUSS_CDEF="0-31" |
‡ |
See Equivalency-tab on http://gaussian.com/options/ or http://gaussian.com/link0/.
*: Unit is indicated as MB or GB (not M or G as for SLURM).
†: The command line option -p does not work for Gaussian 16.c01.
‡: The Gaussian manual suggests using %CPU instead of %NProc to make sure that Gaussian processes are “pinned” to the processors, which can avoid cache misses. On Compute Canada clusters, processes are pinned automatically by the scheduler and at the time of job-submission it cannot be determined which specific CPU cores will be assigned to the job. If at all, the %CPU notation (or its counterparts) should only be used when requesting whole nodes.
Gaussian is only available at Graham and Cedar.
In order to use Gaussian, the PI (supervisor) and each user needs to accept Gaussian’s terms of use by sending an email with the following content to support@computecanada.ca:
I am not a member of a research group developing software competitive to Gaussian.
I will not copy the Gaussian software, nor make it available to anyone else.
I will properly acknowledge Gaussian Inc. and Compute Canada in publications.
I will notify Compute Canada of any change in the above acknowledgement.
small_gaussian_job.sh:
#!/bin/bash
#SBATCH --mem=2G # memory per node
#SBATCH --time=0-00:30 # expected runtime
#SBATCH --cpus-per-task=1 # cpus as defined by %NProcShared
module load gaussian/g16.c01
g16 < h2o2_b3lyp.com > h2o2_b3lyp.logh2o2_b3lyp.com:
big_gaussian_job.sh:
#!/bin/bash
#SBATCH --mem-per-cpu=4000M # memory-per-cpu
#SBATCH --cpus-per-task=8 # cpus as defined by %NProcShared
#SBATCH --time=03-00:00 # expect run time (DD-HH:MM)
module load gaussian/g16.c01
# use localscatch:
export GAUSS_SCRDIR=$SLURM_TMPDIR
g16 < big_molecule.com > big_molecule.logbig_molecule.com:
big_gaussian_job.sh:
#!/bin/bash
#SBATCH --mem=125G # memory per node
#SBATCH --cpus-per-task=32 # cpus as defined by GAUSS_PDEF
#SBATCH --time=03-00:00 # expect run time (DD-HH:MM)
module load gaussian/g16.c01
# use localscatch:
export GAUSS_SCRDIR=$SLURM_TMPDIR
export GAUSS_MDEF="124GB" # A bit lower than --mem=
export GAUSS_PDEF="32"
g16 < big_molecule.com > big_molecule.logbig_molecule.com:
%CHK=/scratch/<USERNAME>/big_molecule.chk
#N MP2/6-311++G(3df,2p) OPT FREQ MaxDisk=400GB
My Big Molecule with a large basis set
... Note that the unit for --mem is G, while for GAUSS_MDEF it is GB!
Gaussian recommends allocating 4GB or more per CPU-core for calculations with 50+ atoms and/or 500+ basis functions. http://gaussian.com/relnotes/?tabid=3
Many, but not all stages of Gaussian calculations can be carried out utilizing multiple CPU cores and to varying degrees. Hartree-Fock and DFT energies, gradients and frequencies and MP2 energies and gradients seem to be well supported but comprehensive documentation is lacking.
Number of CPUs that can be efficiently used depends on:
Remember:
Your Mileage May Vary.
%MEM=120GB
%NProcShared=16
#N B3LYP/6-311++G(3df,2p) Opt FREQ$ seff 26948781
Job ID: 26948781
Cluster: graham
User/Group: stuekero/stuekero
State: COMPLETED (exit code 0)
Cores per node: 16 <===
CPU Utilized: 13-06:23:40
CPU Efficiency: 96.40% of 13-18:17:52 core-walltime
Job Wall-clock time: 20:38:37 <===
Memory Utilized: 37.02 GB
Memory Efficiency: 29.61% of 125.00 GB <=== !!!$ seff 26948862
Job ID: 26948862
Cluster: graham
User/Group: stuekero/stuekero
State: COMPLETED (exit code 0)
Cores per node: 32 <===
CPU Utilized: 13-07:42:35
CPU Efficiency: 99.85% of 13-08:11:12 core-walltime
Job Wall-clock time: 10:00:21 <===
Memory Utilized: 73.00 GB
Memory Efficiency: 58.40% of 125.00 GB <===The freqmem utility helps to estimate the required memory for a frequency calculation.
It uses the following syntax:
freqmem natoms nbasis r|u functionExample (C20O4H28 at B3LYP/6-311++G(3df,2p)):
$ grep --max 1 Stoichiometry *.log
Stoichiometry C20H28O4
$ grep --max 1 "basis functions," *.log
1216 basis functions, 1728 primitive gaussians, 1360 cartesian basis functions
$ freqmem 52 1216 r spd
NAtoms=52 NBasis=1216 closed-shell frequencies with up to SPD functions:
Minimum of 231.68 megawords per thread.1 word = 64 bit = 8 bytes
1 megaword = 64 Mbit = 8 MB
232 megawords = 1856 MB
Gaussian freqmem utility: http://gaussian.com/freqmem/