Problem with NPT ensemble 64 water training set with PBC

[saeedshahi]

Hello everyone,

I'm trying to train a dataset 64 water NPT for the 5000-time frame but when I run it with DeepMD it crashed, and when I remove PBC, it runs perfectly. I used the input.json same as in the se_e3 example on the DeepMD website and the latest version of DeepMD 2.0.2.
What is the problem? also, I attached the error file.

Thanks
Saeed
job11861122.log

[saeedshahi]

I found out the NPT dataset does not work with Descriptor "se_e3". I can train my dataset with Descriptor "se_e2_a" easily.

[wanghan-iapcm]

Could you please provide way of reproducing the bug? thanks!

[saeedshahi]

I tried to send the NPT dataset that I got this error by email but it didn't deliver. I don't know why. so if you know another way for sending the data (about 40 Mb) please let me know.

Best,
Saeed

[wanghan-iapcm]

According to our experience the se_3 descriptor works fine with data of changing simulation box.

One may half the dataset and finally find the minimal dataset that can reproduce the bug.

Your error log is mixing with the error message from dp freeze. Could you please clean it and only show us the error from the dp train? Thanks!

[saeedshahi]

Sure.
train.log

[njzjz]

It seems to be a known issue #1062. Please try CUDA 10.1 instead.