Curious about a specific algorithm in the energy minimization section

GROMACS version: 2020.2
GROMACS modification: Yes
Here post your question

I was having a look through some of the Gromacs C++ source code and found an algorithm I thought could be modified to produce a more accurate and faster search algorithm, and I want to know what you all think. The algorithm is in minimize.cpp under /src/gromacs/mdrun/ and is on line 1297. Its purpose is to determine a maximum amount of higher potential energy the search algorithm will tolerate as it is searching for new energy states, with the equation being: tmp = (std::sqrt(GMX_REAL_EPS)) * fabs(s_a->epot);

While accepting higher energy states during energy minimization may seem counterproductive, it does seem to give lower results at the end.

However, I think that the use of the square root function in the equation seems rather arbitrary and could be changed. Anyone got a second opinion they could spare?