Can't get Conformational Search Alanine Example to Run

[kshen-noble]

Describe the issue

I'm trying to follow this CREST guide on conformational sampling.

command:
crest ./alanine.xyz --gfn2 --gbsa h2o

but I run into an error very early on, even in the initial geometry optimization:

 -----------------------------
 Initial Geometry Optimization
 -----------------------------

  Initial geometry optimization failed!
  Please check your input and, if present, crestopt.log.
ERROR STOP 

Error termination. Backtrace:
#0  0x103637cf7
#1  0x103638b07
#2  0x103639f17
#3  0x10278cdfb
#4  0x1027ae3ab
#5  0x102a6791b
#6  0x1025f7397
#7  0x102c86f8f

I suspect I'm running into challenges installing the software for Mac, and would appreciate any insights on how to troubleshoot.

To Reproduce

I installed via conda:

• crest==3.0.2
• tblite==0.4.0
• xtb==6.7.1

Since my machine is a M3 Mac, I've also tried compiling CREST from source just in case, but ran into the same errors.

I've verified that the xtb and tblite programs work independently on the CLI, and that geometry optimization works with the xtb program.

Additional context

My crest opt.log is not very informative:

  20
 Etot=         -33.8649290264
 C         -2.2423217693       -0.5287748502        0.3281832641
 C         -1.1339831252        0.5138443877        0.0790053036
 N          0.1426014202        0.0020152962        0.1139376188
 C          1.3372946099        0.8069729173       -0.1324846822
 C          2.6267170075       -0.0061636567       -0.0053412725
 O          3.7614500516        0.4039758427       -0.1714283230
 N         -3.5874195683        0.0724101912        0.2901126031
 O         -1.3698068126        1.7010887734       -0.1555542767
 O          2.4112957426       -1.2994221527        0.3443843098
 H         -3.7216414427        0.6412963530        1.1266542620
 H         -3.5997567190        0.7551890933       -0.4737237184
 C         -2.1646252935       -1.5431812466       -0.7049884984
 H         -2.0890731880       -1.0044531303        1.3020611672
 H          0.3005013399       -0.9639003709        0.3860012642
 H          1.3661729845        1.6204173883        0.5981203975
 H          1.2821178923        1.2158719637       -1.1464913779
 H          3.3195884712       -1.6714184648        0.3982104550
 H         -2.9425094897       -2.2876579894       -0.5419829138
 H         -2.3046618238       -1.0794238939       -1.6803038386
 H         -1.1883522567       -2.0243141103       -0.6686276085

I've also tried using an input.toml (suggested in this thread), :

#This is a CREST input file
input = 'alanine.xyz'
runtype='ancopt'
optlev = 'vtight'

#parallelization
threads = 4

#calculation level
[[calculation.level]]
method = "xtb"
binary='xtb --alpb water --chrg 0'

This ran, however the geometry optimization still failed. Again, this is confusing because if I call xtb --opt --gfn2 alanine.xyz I can converge the minimization.

[pprcht]

I guess running on Mac is the problem, it's likely the same kind of issue as in #382. Just tested the alanine example again and on Ubuntu and CentOS everything works fine with both conda and fresh compiled builds.

xtb and tblite working on their own provides only limited insight, because the memory handling is different here (and apparently a problem for MacOS). Internally crest alanine.xyz --opt --gfn2 would be the most similar to the xtb --opt call, try if that works at least.
I will need to get access to some Mac system to do some proper error search.

[kshen-noble]

Thanks, @pprcht.

I got a virtual machine spun up and got things to run -- I'll do that for now.

Some other odd behavior I noticed:

• cregen also doesn't work well on Mac. Passing options like ewin didn't seem to change the behavior at all (e.g. ewin always ended up as 0.0, even though the default value is supposed to be 6.0)

[pprcht]

Alright, thanks for letting me know. It really looks like I need to get my hands on a MacOS for debugging. I'll add a label to this issue and the other one

[ghutchis]

@pprcht - If you use https://github.com/mxschmitt/action-tmate in your GitHub actions, you can remotely connect for debugging. It's very useful

[pprcht]

@ghutchis - oh, that might be a much better option than buying a Mac for debugging... Thanks for the tip!