Usage

Run the executable like

RAYON_NUM_THREADS=16 RUST_BACKTRACE=1 $bin [input file] & disown -h

The environment variables RAYON_NUM_THREADS and RUST_BACKTRACE are not required, but the first will prevent rayon from using too many resources, and the latter will help debug any issues. $bin should be the path of the executable. If you plan to run many of these or are testing the program, you may want to group this commands into a run script like

   #!/bin/sh

   d=$(date +%y-%m-%d_%H-%M-%S)

   if ! [[ -d archive ]] ; then
	mkdir archive
   fi

   mv semp.out archive/$d.out
   mv semp.log archive/$d.log
   mv version archive/$d.ver

   ls -lth $bin > version

   RAYON_NUM_THREADS=16 RUST_BACKTRACE=1 $bin & disown -h

   tail -F semp.out

which will help organize previous runs without overwriting them. Output is written to semp.out. Progress information and errors are written to semp.log.

If input file argument to the binary is omitted, a default of semp.toml is assumed. The form of this input file is given below with all of the options annotated:

# global config
job_limit = 10000 # the maximum number of jobs to submit at once
chunk_size = 128 # the number of jobs to include in one queue submission
sleep_int = 5 # the interval to sleep before polling complete jobs again
max_iter = 5 # maximum levmar iterations
# the initial parameters to optimize in MOPAC EXTERNAL format
params = """
USS        H    -11.246958000000
ZS         H      1.268641000000
BETAS      H     -8.352984000000
GSS        H     14.448686000000
USS        C    -51.089653000000
UPP        C    -39.937920000000
ZS         C      2.047558000000
ZP         C      1.702841000000
BETAS      C    -15.385236000000
BETAP      C     -7.471929000000
GSS        C     13.335519000000
GPP        C     10.778326000000
GSP        C     11.528134000000
GP2        C      9.486212000000
HSP        C      0.717322000000
FN11       C      0.046302000000
"""
optimize = "energy" # whether to optimize on "energy" or "frequency"
broyden = false # use broyden's method for approximate jacobian updates
broyd_int = 10 # interval at which to run full jacobians when broyden = true
reset_lambda = false # reset the levmar lambda parameter if the algorithm fails
queue = "slurm" # queue type to use
delta = 1e-4 # step size in the parameters for the jacobian

# this section holds configuration options for individual molecules. more than
# one [[molecule]] section is allowed in an input file, and the optimization
# will operate over all of them
[[molecule]]
atom_names = ["C", "C", "C", "H", "H"] # atom names in the order of the geometry
charge = 0 # molecular charge
dummies = [] # dummy atoms (unused?)
# the geometry as a Z-matrix or XYZ geometry, following MOPAC formatting rules
geometry = """
C
C 1 CC
C 1 CC 2 CCC
H 2 CH 1 HCC 3 180.0
H 3 CH 1 HCC 2 180.0

CC =                  1.42101898
CCC =                55.60133141
CH =                  1.07692776
HCC =               147.81488230
"""
# the coordinate type for the QFF. if sic, a corresponding intder input file is
# required. other options are "norm" and "cart"
coord_type = { sic = "test_files/intder.in" }
# the "true" values of the anharmonic frequencies to train on
true_freqs = [3139.8, 3108.7, 1595.1, 1275.8, 1056.9, 1007.9, 876.8, 876.5, 772.7]
# the symmetries of the frequencies in true_freqs. the order must match true_freqs
irreps     = [  "A1",   "B2",   "A1",   "A1",   "B2",   "A2",  "A1",  "B2",  "B1"]
# the MOPAC template for the computations
template = "scfcrt=1.D-21 aux(precision=14) PM6 threads=1"

ntBre/rust-semp

Usage