Got nan value in output

[njzjz]

Discussed in #871

^{Originally posted by jtyang-chem July 20, 2021}
I made all data by my scripts and give them into training, then I got a lcurve.out file with nan values.

• To make sure it's not a problem of version or input file, I also tested an example file of water box, everything goes well.
• my deepmd-kit version is v1.3.3 cpu.

I can't see the differences of which I made from the example data, I've already upload them to my github, could someone help me?

[njzjz]

I can reproduce using the latest development version.

[njzjz]

The reason is that some atoms are overlapping. Maybe we can add a check to throw an error for this situation.