linear water calculated to have no imaginary frequencies

[ptrab]

Describe the bug I constructed a linear water molecule as

3

H  0.0  0.0  1.0
O  0.0  0.0  0.0
H  0.0  0.0 -1.0    

and used xtb 6.5.1 to optimize it. I played around with some parameters as with or without gbsa, --acc and other parameters; I also looked if --ohess versus separate --opt and --hess calculations have an influence. They don't.

It results in a linear water molecule as expected but it calculates all positive frequencies, which surprised me.

To Reproduce Steps to reproduce the behaviour:

1. create water molecule input h2o.xyz as above
2. load xtb 6.5.1 module

module show xtb/6.5.1
-------------------------------------------------------------------
/data/software/modules/xtb/6.5.1:

setenv           XTBHOME /scratch/software/xtb-6.5.1 
setenv           XTBPATH /scratch/software/xtb-6.5.1/share/xtb 
prepend-path     MANPATH /scratch/software/xtb-6.5.1/share/man 
prepend-path     PATH /scratch/software/xtb-6.5.1/bin 
append-path      PKG_CONFIG_PATH /scratch/software/xtb-6.5.1/lib64/pkgconfig 
setenv           OMP_NUM_THREADS 9,1 
setenv           OMP_MAX_ACTIVE_LEVELS 1 
setenv           MKL_NUM_THREADS 9 
setenv           OMP_STACKSIZE 100G 
-------------------------------------------------------------------

4. run xtb with your options and the --verbose flag xtb h2o.xyz --acc 0.000001 --ohess extreme --alpb water --verbose &> xtb_opt.log
5. output showing the error xtb_opt.log

Expected behaviour Linear water is supposed to have two imaginary/negative frequencies.

[haneug]

This seems to be a specific problem for linear molecules with the way we calculate the numerical hessian.

[eljost]

I don't think it is related to the numerical Hessian. I can recreate this calculation in my own code pysisyphus, that just reads in the numerical Hessian produced by XTB and I obtain the two imaginary modes. So the issue is probably somewhere else. Maybe in the part where the translation/rotations is projected out.