Incorrect treatment of perturbed constraint length for the Verlet buffer estimate
The Verlet buffer estimate always used the A-state constraint length. This leads to an incorrect buffer estimate with perturbed constraints. In practice this is never an issue as nearly always very few constraints are perturbed, so the effect on the buffer is much smaller than the effect of other approximations that always give a large overestimate of the buffer size.
Another effect was that a division by zero (constraint length) occurs in the verlet buffer calculation when the A-state constraint length is zero with a dynamic FEP lambda. But in the 2020 and 2021 releases this setup is disallowed by grompp, so only the 2022-RC was affected.