Page 2 of 2

Re: varmin gives NaN

Posted: Fri Aug 25, 2017 12:56 pm
by Katharina Doblhoff
Dear Pablo,
Yes, actually the rest of the calculation does significantly misbehave and in a strange way too:
In the optimization it says "Bad parameter set" and does not alter the parameters.1.casl compared to the parameters.0.casl. Never the less the energy and variance change between the 0th and the 1st iteration (see output of envmc below. Later on, the linear parameters change to very small numbers (in the 3rd iteration) and then the calculation goes wild.):

Code: Select all

VMC #1: E     = -244.06(4) ; var  = 59.1(7) (parameters.0.casl)
VMC #2: E     = -239.44(8) ; var  = 84(1) (parameters.1.casl)
VMC #3: E     = -239.44(8) ; var  = 83(1) (parameters.2.casl)
VMC #4: E     = -239.48(8) ; var  = 84(2) (parameters.3.casl)
Warning: [CHECK_KE] This may indicate a problem. Continuing run, however, 
because the KE_FORGIVE flag is set to T. 
VMC #5: E     = NaN ; var  = NaN (parameters.4.casl)*
Warning: [CHECK_KE] This may indicate a problem. Continuing run, however, 
because the KE_FORGIVE flag is set to T. 
VMC #6: E     = NaN ; var  = NaN (parameters.5.casl)*
The cutoffs are set by default (I guess to the Wigner-Seitz radius) and the value is non-integer.

In principle it is not a big issue in my case, since I am pretty sure that I can get away with a non-spin-polarized e-e-n term, but the fact that the energy changes even if the parameters.casl file does not change got me kind of worried.

Re: varmin gives NaN

Posted: Fri Aug 25, 2017 1:22 pm
by Pablo_Lopez_Rios
Hi Katharina,

The difference between VMC cycles 1 and 2 has a simple explanation: if opt_method/='emin', the first cycle runs without a Jastrow factor, while the second runs with a Jastrow factor in which all free parameters are set to zero; the cusp conditions set non-free parameters to non-zero values, hence J(R) /= 0.

Try specifying the cut-offs for the e-e-n term, e.g.,

Code: Select all

    ...
    e-n cutoff:
      Type: polynomial
      Constants: [ C: 3 ]
      Parameters:
        Channel 1-n1:
          L: [ 3.80,  optimizable ]
        Channel 2-n1:
          L: [ 3.90,  optimizable ]
    ....
(replace n1 with the specific nucleus in the channel that is causing the problem.)

Best,
Pablo

Re: varmin gives NaN

Posted: Fri Aug 25, 2017 1:33 pm
by Katharina Doblhoff
Dear Pablo
if opt_method/='emin', the first cycle runs without a Jastrow factor
Oh, this was an easy explanation - considering the numbers of the energy and variance at iteration 0, I could have figured that myself...
Thanks and all the best,
Katharina