proteneer/timemachine

fixed point madness

Closed this issue · 1 comments

Found a really subtle bug.

When we do partial linear mixing
U(lambda) = (1-lambda)U0(lambda) + lambdaU1(lambda)

When we evaluate U0 at lambda=1, we often times end up with a situation where i is a core atom, j is non-core atoms switching planes, but there is an exclusion between i and j. In the absence of linear mixing this causes huge singularities due to the way we back-subtract the exclusions from a fully dense computation.

If we turn on the bonded terms at lambda=1 then we should actually never have this issue.

No longer relevant - closing