Hi,
Recently, I tried to use Forcebalance to optimize some parameters. But I got an error information:
File "/***/python/python2.7/lib/python2.7/site-packages/forcebalance/abinitio.py", line 949, in get_energy_force
Ci = 1. / C
FloatingPointError: divide by zero encountered in divide
I have same "0.0" force data in the target file, but it seems that the error does not caused by these zeros. I am not sure how to fix it. Could you help me?
Thanks!
Xu CF
a FloatingPointError
-
Lee-Ping Wang
- Posts: 102
- Joined: Sun Jun 19, 2011 5:14 pm
Re: a FloatingPointError
I'm so sorry that I missed this message!
What's happening is that you are encountering a zero stdev for the reference forces, and ForceBalance is attempting to use this to normalize your force errors (to obtain a fractional error).
Are you actually trying to fit your parameters to zero forces, or are you zeroing them out because you're not doing force matching? Or is it something else?
What's happening is that you are encountering a zero stdev for the reference forces, and ForceBalance is attempting to use this to normalize your force errors (to obtain a fractional error).
Are you actually trying to fit your parameters to zero forces, or are you zeroing them out because you're not doing force matching? Or is it something else?
Re: a FloatingPointError
Thanks so much!
Although still do not know what caused the problem, I have run ForceBalance successfully.
Now there is another problem.
I wanna optimize some van der Waals parameters, but some optimized VDWTs (the epsilon item) of some atoms are 0.0e+00. I have changed some complexes of the system which I wanna studied, but the zero result always exists.
Is it normal?
I have read the manual and followed the tutorial example. It seems that no related problem is discussed.
Thanks a lot!
XU
Although still do not know what caused the problem, I have run ForceBalance successfully.
Now there is another problem.
I wanna optimize some van der Waals parameters, but some optimized VDWTs (the epsilon item) of some atoms are 0.0e+00. I have changed some complexes of the system which I wanna studied, but the zero result always exists.
Is it normal?
I have read the manual and followed the tutorial example. It seems that no related problem is discussed.
Thanks a lot!
XU
-
Lee-Ping Wang
- Posts: 102
- Joined: Sun Jun 19, 2011 5:14 pm
Re: a FloatingPointError
Hi there,
It's fairly common for VDWT to become zero because the force field is attempting to fit some highly repulsive structure. For example, imagine a repulsion interaction of +30 kcal/mol (from the QM calculation); the MM calculation would typically be much higher, at perhaps +3000 kcal/mol (from the LJ interaction). In this case the optimization would attempt to make epsilon negative, but since that would result in a crash it instead sets the value to zero.
This is definitely a problem and not one with a perfect solution. I would suggest filtering your QM data so that nothing above a limit - say 20 or 30 kcal/mol above the minimum - is fitted. The way to do this in the $target section is:
attenuate
energy_denom 20
energy_upper 20
This is an advanced feature that I'm still developing, so please check out the latest code from GitHub before you do it, and let me know if it works.
Thanks,
- Lee-Ping
It's fairly common for VDWT to become zero because the force field is attempting to fit some highly repulsive structure. For example, imagine a repulsion interaction of +30 kcal/mol (from the QM calculation); the MM calculation would typically be much higher, at perhaps +3000 kcal/mol (from the LJ interaction). In this case the optimization would attempt to make epsilon negative, but since that would result in a crash it instead sets the value to zero.
This is definitely a problem and not one with a perfect solution. I would suggest filtering your QM data so that nothing above a limit - say 20 or 30 kcal/mol above the minimum - is fitted. The way to do this in the $target section is:
attenuate
energy_denom 20
energy_upper 20
This is an advanced feature that I'm still developing, so please check out the latest code from GitHub before you do it, and let me know if it works.
Thanks,
- Lee-Ping