Dear all,
I was calculating bulk modulus of Al2O3 with ReaxFF. I got an order of small B than it was supposed. I varied three body cutoff in control file from default 0.001Å to 2.5Å. At value of 0.1Å I got reasonable value for B. But when I went further and increased, B converged to too high value. Is it normal that three body cutoff affects that much? Or what does this mean, is the correct cutoff which gives converged values (that would be more logical) or the one which gives more correct B?
Thanks for any comment.
Best wishes,
Manana
Unlike simpler potentials, ReaxFF is not fit to elastic constants, so I
expect it to be somewhat less accurate than those.
The threebody cutoff has units of bond order, not distance. A value
greater than 1 will turn off almost all valence angles, so it is not
surprising that you see large changes.
You should use the value corresponding to the particular ReaxFF potential
that was fit to Al2O3. That usually means 0.001.
In general, for convergence parameters like bond-order and distance
cut-offs, the fitting is done using values that are accurate, but not
fully converged (too expensive). This means that running a simulation with
a higher-accuracy value will not necessarily give you more accurate
results, because you are not using the value used in the fitting. You are
correct to say that you definitely should not use a less accurate value
that happens to reproduce some desired property.
Aidan
Hello Aidan,
Thanks for the comments.
Unlike simpler potentials, ReaxFF is not fit to elastic constants, so I
expect it to be somewhat less accurate than those.
I was checking based on their paper, where energy vs volume curve is compared to QM and gives a good match:
http://authors.library.caltech.edu/2825/1/ZHAprb04.pdf
The threebody cutoff has units of bond order, not distance. A value
greater than 1 will turn off almost all valence angles, so it is not
surprising that you see large changes.
You should use the value corresponding to the particular ReaxFF potential
that was fit to Al2O3. That usually means 0.001.
In general, for convergence parameters like bond-order and distance
cut-offs, the fitting is done using values that are accurate, but not
fully converged (too expensive). This means that running a simulation with
a higher-accuracy value will not necessarily give you more accurate
results, because you are not using the value used in the fitting. You are
correct to say that you definitely should not use a less accurate value
that happens to reproduce some desired property.
Yeah, that’s right. By the way, I got very nice result with bulk Al with default 0.001. It’s just a problem with the oxide.