[AMBER] xmin fails with error in load_lbfgs()

From: <Don.Bashford.stjude.org>
Date: Tue, 19 Oct 2010 14:53:04 -0500

I'm trying to minimize hydrogen positions by putting harmonic
restraints on non-H atoms (in sander, amber11). If I set ntmin=3, and
don't specify any of the xmin_* options I get, "ERROR in load_lbfgs():
YS=0" and the minimizer stops with a ":-(".

Upping the verbosity of xmin, I get the following:

 MIN: It= 1 E= -5154.24 ( 2.828)
  LS: i= 1 lhs_f= -2.5577449 rhs_f= -0.033863839
            lhs_g= 158815.43 rhs_g= 95005.505
  LS: i= 2 lhs_f= -155.6904 rhs_f= -0.013521306
            lhs_g= 261.0678 rhs_g= 95005.505
  LS: step= 0.0012808916 it= 2
 MIN: It= 2 E= -5309.93 ( 3.292)
  LS: i= 1 lhs_f= 92.80803 rhs_f= -0.026413056
            lhs_g= 171372.09 rhs_g= 129362.17
  LS: i= 2 lhs_f= 14.144082 rhs_f= -0.0081198419
            lhs_g= 47107.56 rhs_g= 129362.17
  LS: i= 3 lhs_f= 51.625208 rhs_f= -0.0017614175
            lhs_g= 122306.52 rhs_g= 129362.17
  LS: i= 4 lhs_f= 67.101822 rhs_f= -7.9175093e-05
            lhs_g= 142164.99 rhs_g= 129362.17
  LS: i= 5 lhs_f= 67.756085 rhs_f= -1.3206528e-05
            lhs_g= 142943.51 rhs_g= 129362.17
Line minimizer aborted: step at lower bound
  LS: step= 9.1880612e-07 it= 5
 MIN: It= 3 E= -5242.17 ( 3.283)
  LS: i= 1 lhs_f= 93.079451 rhs_f= -0.0086974762
            lhs_g= 16.120035 rhs_g= 78.277286
  LS: i= 2 lhs_f= 119.59594 rhs_f= -0.00095856763
            lhs_g= 75.757991 rhs_g= 78.277286
  LS: i= 3 lhs_f= 128.45274 rhs_f= -1.2043799e-05
            lhs_g= 87.010375 rhs_g= 78.277286
  LS: i= 4 lhs_f= 128.57331 rhs_f= -1.8805857e-09
            lhs_g= 87.153552 rhs_g= 78.277286
  LS: i= 5 lhs_f= 128.57333 rhs_f= -4.5844175e-17
            lhs_g= 87.153575 rhs_g= 78.277286
  LS: i= 6 lhs_f= 128.57333 rhs_f= -8.6974762e-23
            lhs_g= 87.153575 rhs_g= 78.277286
Line minimizer aborted: step at lower bound
  LS: step= 1e-20 it= 6

ERROR in load_lbfgs(): YS=0.
----------------------------------------------------------------
 FIN: :-( E= -5131.09 ( 3.294)


So it looks like the line search step gets very small, and the
load_lbfgs fails on a zero y vector. I can minimize my system using
another method, PRCG, without problems. Maybe this is a bug in the
xmin code?

-Don


Email Disclaimer: www.stjude.org/emaildisclaimer


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Oct 19 2010 - 13:00:03 PDT
Custom Search