Reply To: Minimization question

[Anonymous]

“Now I realize that I’d need to restrict attention to those randomized structures in which any domain interpenetration distance is less than an atomic radius, which I’d hope would merely be those structures in which the energy is less than some value.”

This is likely to be true. It sounds like you really need a proper sampling protocol for the linker, not just minimization.

I agree with you that the minimizer should be returning the input structure when it gets confused in this case. I don’t know why it isn’t. With severe clashes, perhaps there is enough rounding noise moving between processor’s longer registers and cache/RAM that the minimizer isn’t aware of the error. The minimizer isn’t doing a Monte Carlo protocol so under normal circumstances there’s no need to check that the output energy is lower than the input. Maybe there’s just a bug somewhere…