search

grimme-lab / xtb

Semiempirical Extended Tight-Binding Program Package
https://xtb-docs.readthedocs.io/
GNU Lesser General Public License v3.0
581 stars 144 forks source link

Geometry optimization failure for the system with Ti atom #409

Open amdmaniac opened 3 years ago

amdmaniac commented 3 years ago

Describe the bug

Geometry optimization for C(34)H(36) system with Ti atom introduced in it goes nuts using GFN2-xTB (6.3.3). Please see the attached xtb2.trj.xyz file. The same is also true for GFN1-xTB (6.3.3).

To Reproduce

../xtb input.xyz --gfn 2 --opt tight --uhf 0 --chrg 0 > output.out

The input file input.xyz is attached.

Expected behaviour

The system under consideration is physically relevant as the analogous geometry optimizations with PBE/def2-SVP or PBEh-3c (as implemented in ORCA 4.2.1) are perfectly fine. Please see the attached pbeh-3c.trj.xyz file.

Additional context

In my opinion, there is something wrong with the description of Ti atom. If Ti atom is interchanged with Ni atom (of a similar kind), GFN2-xTB causes no issues.

Files.zip

awvwgk commented 3 years ago

The SCC can somehow arrive at a non-physical solution for early transition metals. You could try another guess with

$scc
  guess=sad

Sometimes this helps to converge the SCC to a physically correct solution for early transition metals.

amdmaniac commented 3 years ago

I have tried

$scc guess=sad $end

and also

$scc guess=gasteiger $end

but with no success. GFN2-xTB as well as GFN1-xTB still behave totally wrong.

awvwgk commented 3 years ago

If the electronic structure is completely off with the xTB Hamiltonian, it might be worth trying the GFN-FF instead.

amdmaniac commented 3 years ago

GFN-FF and GFN0-xTB work fine. However, I have to check the accuracy of their geometries with respect to PBEh-3c for the physical properties that I am interested in.