[gmx-users] More details on the "Steepest Descents"

[ZHANG Cheng]

Dear Gromacs friends,


I am trying to understand more on the "Steepest Descents" so that I can decide if I need to better minimise the system by reducing the "emtol" or increasing the "nsteps", or other ways.


The prompt is shown in the end. My understanding is, the algorithm randomly chooses a starting atom in the system (in my case the 41651th atom), it tries to change its conformation so as to yield lower "Epot" and "Fmax". Then it moves to the second atom, with "Dmax" distance from the first atom, and conducts the conformation change. It keeps moving around until the "Fmax" is lower than the "emtol", or the maximum steps has finished, or there is no more improvement.


Please correct my understanding if possible. Basically, can we optimise the minimisation parameters simply based on the prompt (e.g. Dmax, Epot, Fmax, atom)?


Cheng




Steepest Descents:
   Tolerance (Fmax)   =  1.00000e+01
   Number of steps    =        10000
Step=    0, Dmax= 1.0e-02 nm, Epot= -1.92831e+06 Fmax= 1.63594e+03, atom= 41651
Step=    1, Dmax= 1.0e-02 nm, Epot= -1.95056e+06 Fmax= 8.94326e+02, atom= 2170
Step=    2, Dmax= 1.2e-02 nm, Epot= -1.98293e+06 Fmax= 4.67586e+02, atom= 73626
... ...

Step= 9998, Dmax= 2.3e-02 nm, Epot= -2.38236e+06 Fmax= 5.07179e+02, atom= 33278
Step= 9999, Dmax= 1.1e-02 nm, Epot= -2.38236e+06 Fmax= 1.74071e+02, atom= 33278
Step=10000, Dmax= 1.4e-02 nm, Epot= -2.38236e+06 Fmax= 2.22063e+02, atom= 33278
Energy minimization reached the maximum number of steps before the forces
reached the requested precision Fmax < 10.
writing lowest energy coordinates.
Steepest Descents did not converge to Fmax < 10 in 10001 steps.
Potential Energy  = -2.3823640e+06
Maximum force     =  2.2206276e+02 on atom 33278
Norm of force     =  1.2667693e+00