search

issp-center-dev / mVMC

A numerical solver package for a wide range of quantum lattice models based on many-variable Variational Monte Carlo method
http://www.pasums.issp.u-tokyo.ac.jp/mvmc/en/
GNU General Public License v3.0
59 stars 19 forks source link

StcOpt Error #48

Closed xfzhang-phys closed 12 months ago

xfzhang-phys commented 12 months ago

Dear mVMC developers,

I'm trying to use mVMC-1.2.0 code. In some of my calculations, mVMC raised Error: StcOpt info=1 step=XXX after warning: VMCMainCal rank:0 sample:XXXX e=-nan. (PS: There are also some calculations of different interaction parameters that are finished normally) These calculations are performed on a 6x6x3 Hubbard system, and typically NVMCSample=1600, NVMCWarmUp=100 are used with 64 cores. Could you give me some advice on how to prevent this error? (BTW, I find sometimes this error can be prevented by using more warm-up steps) Thanks !

tmisawa commented 12 months ago

This kind of error sometimes occurs when the initial state of the wave functions is not good. If the initial state |phi_0> has no overlap between the most real space configurations, i.e., Pf(x)=<x|phi_0> = 0, this kind of error occurs. So, please check the initial state used in your calculations.

xfzhang-phys commented 12 months ago

Dear Misawa,

Thank you for your reply! In my calculations, the above error does not raise at the begining of my calculations. It happens after several optimization steps, and I do not specify the initial file besides the RndSeed. Due to this reason, I guess if my warm-up steps are too small to generate a good enough | phi_0 > ? In the CPC paper of mVMC, you said each mVMC sampler generates $N{MC} / N{sampler}$ real-space configurations. I wonder if it's also right for NVMCWarmUp parameter?

tmisawa commented 12 months ago

Dear xfzhang-phys,

In my calculations, the above error does not raise at the begining of my calculations. It happens after several optimization steps, and I do not specify the initial file besides the RndSeed.

Ok. I understand the situation. I think that the origin of the error may be related to the optimization. By increasing the Monte Carlo sampling (NVMCSample), the optimization may become stable and it is expected that the error does not occur.

I wonder if it's also right for NVMCWarmUp parameter?

Yes. The total Mone Carlo sampling for warmup is given by NVMCWarmUp*Nsampler . But, in this case, I think that NVMCWarnUp is not directly related to this error.

xfzhang-phys commented 12 months ago

Thanks a lot for your advice ！