Question about Gradient Tolerance during minimization

From: Wolbach, Jeffrey P \(WOLBACH\) (
Date: Wed Sep 08 2004 - 13:00:15 CDT

My basic question is: What is the Gradient Tolerance quoted in the
output measuring? Is it some measure of a 2nd derivative matrix for the

The calculation I am attempting to run is a minimization of a protein
(1DKI) without any sort of boundary conditions (in vacuo). The
structure from the PDB was missing 3-4 loops, and I added those by hand.
The structure was then minimized for a small number of iterations in
HyperChem, prior to preparation for NAMD, to relieve bad contacts
between the protein residues. (I was also locally minimizing the loop
regions as I went.) I ran the minimization for 5000 iterations in NAMD.

What puzzles me is that the Gradient Tolerance on the first iteration
was ~26000 (expected due to guessed H's and atoms for terminal
residues), and quickly dropped to ~200. But then it climbed back up to
~35000 and stayed there for the balance of the run. What I find even
odder is that the system potential energy only dropped by 4.7 kcal/mol
(system potential ~ -30,000, so <0.1 % drop) over the last 300
iterations. So just considering the energies, the system is (at least)
very nearly converged, but the Gradient Tolerance seems to be indicating
a system very far from equilbrium. What does the Gradient Tolerance

I had no frozen atoms in my minimization, and have seen the same
behavior from this system while minimizing in a periodic box of water.
Very small energy changes during long runs, but very large Gradient
Tolerance values. This was true whether I froze the backbone or not.


Jeff Wolbach
Von Leibig Researcher
Juniata College

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:37:52 CST