tldr: Manually or script dihedral angle changes, then use a GFN or another DFTB method to find any/all conformers
Without seeing the compound, it's really hard to know the problem. Most programs, including those you mentioned, have a set of rules to identify "rotatable bonds" and suggest likely dihedral angles.
As you might expect, such rules are derived for neutral, ground-state organic molecules.
There are a few options:
- As suggested by Jan Jensen's comments above, you can use RDKit, Open Babel or another tool to drive your own dihedral angle search.
- As suggested by Martin's comments above, there are some tools in Grimme's
xtb program to sample conformations. These will use the GFN1 or GFN2 approximate density functional (pretty much DFTB + dispersion correction) to evaluate energies.
I'd probably suggest a combination of both. From what I understand of Grimme's conformer search scripts, it calculates the vibrational modes of the molecule and samples geometries along vibrational modes. This is a good idea, although in some molecules it may not sample all minima (e.g., if you have many conformations in a large molecule, you might never "reach" the whole potential energy surface from any given initial geometry).
So you can change the dihedral angles using your program of choice, generate a few initial structures, then use
xtb to optimize the geometries and
confscript to sample a few more geometries.
I feel pretty confident that in small to medium geometry molecules, this will do a good job.
A few notes:
- If you're sampling a radical, you definitely want to use a good DFT or quantum method to evaluate the energies. I doubt any current force field will do what you want.
- We haven't finished our evaluation, but right now, GFN2 is performing better than any current method on conformer energetics for neutral organic molecules. We haven't checked radicals, but I would trust it more than any other current "fast" method.
- If you don't have the
xtb package, the method and conformer sampling scripts are now available as part of
ORCA 4.1 too. There's a description of this kind of conformer sampling in the manual when discussing prediction of NMR spectra.