Oblate model runs very slowly -- February 06, 2017 - 19:32

Dear Edward,

I have run the model free analysis based on the Dauvergne protocol on our
university cluster. Everything runs smoothly. The local tm model runs
first, then I run the sphere, prolate, oblate, ellipsoid models separately.
Each model converges for about 12 hours as I use 16 processes by employing
the "mpirun" protocol. Ellipsoid finishes with 284 rounds, sphere ends in
10 rounds, prolate ends in 14 round. However, the oblate seems freeze in
round 10 after about 10 days. I open the log file and see it "freezes" at
the minimization part.

I searched the mailing lists and then follow "sam mahdi" tutorial to add
the following lines in the protocol for oblate model:
# Change some opt params.

# COMMENT THESE OUT FOR REAL CALCULATION. THIS IS ONLY TO SPEED UP THE
PROCESS IN ERROR SEARCHING.
dAuvergne_protocol.opt_func_tol = 1e-5
dAuvergne_protocol.opt_max_iterations = 1000

I submit the new oblate yesterday and now it is at round 14. Is that

normal the oblate model running so slowly?


For the tolerance, setting at which value is reasonable for publication?

I know the default tolerance for relax is 1e-25, but I do not know when

the oblate model will finish at this point.


Thanks,


Shen