The expression used for delta in the soft-core potential docs is wrong. The expression used should be detla = shift_coulomb^2 * alpha as in the comments within the code. The OpenMM expression was updated to be consistent with the website docs, so is also incorrect.
