Can't reproduce NCI mode example (CREGEN only leaves one structure)

[juius]

Describe the issue
CREST 3.0.2 only generates one cluster instead of 69 different clusters of (H2O)6 from the websites example.

CREGEN only leaves one structure in the energy window, regardless of the threshold set with --ewin:

CREGEN> running RMSDs ... done.
CREGEN> E lowest :   -30.49613
 1 structures remain within    12.00 kcal/mol window
 This is less than 5% of the initial 294 structures.
 Increasing energy window to include more...
CREGEN> running RMSDs ... done.
CREGEN> E lowest :   -30.49613
 1 structures remain within    18.00 kcal/mol window

To Reproduce
Follow the example on the website:

crest struc.xyz --nci

with struc.xyz:

   18

 O    -1.469470     0.057971     1.425525
 H    -0.584285     0.519331     1.484152
 H    -1.949531     0.297378     2.218248
 O    -0.414388    -2.005083    -0.137355
 H    -0.827334    -1.527314     0.598606
 H    -0.829507    -1.642295    -0.927593
 O    -1.740922     0.293387    -1.291566
 H    -2.429374     0.885495    -1.595601
 H    -1.784260     0.283934    -0.315678
 O     2.037846    -0.900550    -0.123510
 H     2.719638    -1.557770    -0.261948
 H     1.180681    -1.385350    -0.122605
 O     0.833144     1.402993    -1.296134
 H     1.432120     0.648938    -1.204997
 H    -0.044502     1.022356    -1.465271
 O     0.879327     1.214234     1.389828
 H     0.853406     1.590056     0.483434
 H     1.511729     0.488538     1.308001

Additional context

CREST output:
crest302-h20-cluster.log

[juius]

seems to be a problem specific to CREST on Mac (installed from conda-forge) as it's working fine on Linux

crude pre-optimization
 ----------------------
 Optimization engine: ANCOPT
 Hessian update type: BFGS
 E/G convergence criteria:  0.500E-03 Eh, 0.100E-01 Eh/a0
 maximum optimization steps: 200
 |>0.3% |>10.2% |>20.1% |>30.3% |>40.1% |>50.0% |>60.2% |>70.1% |>80.3% |>90.1% |>100.0%
 done.
> 294 of 294 structures successfully optimized (100.0% success)
> Total runtime for 294 optimizations:
 * wall-time:     0 d,  0 h,  1 min, 16.150 sec
 *  cpu-time:     0 d,  0 h,  1 min, 15.745 sec
 * ratio c/w:     0.995 speedup
> Corresponding to approximately 0.259 sec per processed structure

CREGEN> running RMSDs ... done.
CREGEN> E lowest :   -30.49617
 283 structures remain within    12.00 kcal/mol window

[Physopholy]

I am also having this issue on Linux. It isn't specific to NCI mode for me.

[matsumuraf]

#371 is also on linux, i guess not really specific to the mac OS

[pprcht]

I'm not having the issue and the example works as intended, even with the plain crest struc.xyz --nci command.

I'm aware this won't be much help for macOS, but could (any of) you try with the prebuilt binaries from GtiHub or a self-built version rather than the conda build? My best guess right now is that it's some issue due to different build types.

[Physopholy]

Hello Dr. Pracht,

I'm using the prebuilt binary of CREST 3.0.2 already, along with xTB 6.7.1. I will try compiling it myself soon.