Hi, all
I am using the fortran library of Reax in lammps.
Now I am wondering, how forces are calculated in reax? Is it analytical(like in charmm) or numerical. And to what extent I can trust the forces. 10e-2? 10e-4?
Thank you!
Rui DONG
Hi, all
I am using the fortran library of Reax in lammps.
Now I am wondering, how forces are calculated in reax? Is it analytical(like
in charmm) or numerical. And to what extent I can trust the forces. 10e-2?
10e-4?
how much can you trust *any* force field that you don't know how it works?
...and what difference does it make if you compute the forces
analytically or numerically, if the parameterization and overall
choice of model are bad?
axel.
Axel, could you please be more specific? What exactly is bad: reaxff
parameters or the reaxff approach itself? I am asking because I want to
use it in my work, and I thought I should ask people who are more
educated than me before I start.
Cheers,
Bartek
you’re really asking a ReaxFF question, not a LAMMPS
question. I suggest you read some of the ReaxFF papers
or look at Adri van Duin’s web site. They will explain
how ReaxFF is formulated and what kind of accuracies
are inherent to it.
LAMMPS just has an implementation of ReaxFF.
Steve
> Hi, all
>
> I am using the fortran library of Reax in lammps.
> Now I am wondering, how forces are calculated in reax? Is it analytical(like
> in charmm) or numerical. And to what extent I can trust the forces. 10e-2?
> 10e-4?how much can you trust *any* force field that you don't know how it works?
...and what difference does it make if you compute the forces
analytically or numerically, if the parameterization and overall
choice of model are bad?Axel, could you please be more specific? What exactly is bad: reaxff
parameters or the reaxff approach itself? I am asking because I want to
use it in my work, and I thought I should ask people who are more
educated than me before I start.
i didn't say that reaxff is bad or the (unknown) parameters.
what i am saying is:
a) if you want to use an empirical model, you first have to learn how
it works in principle. and the question of the original poster
indicated that this had not happened
and:
b) accuracy is not accessible on an absolute scale for any empirical
model. you always have to test for each parameter set and for the
given circumstances whether a particular parameterization is suitable
or not.
there are cases where reaxff is better than other models and there are
cases where reaxff is worse. there are parameter sets that work well
for some usage and not so well for some other usage of the same model.
so the real problem is that the original poster is asking the wrong
questions and has to do some more homework.
axel.
Steve, I've read a lot on reaxff including the most of Adri's papers,
and I am pretty much convinced this is the way to go for me (or at
least: to try). I misinterpreted Axel's response that there is something
fundamentally wrong with the approach itself, but now everything is
clear to me.
Thanks to you both,
Bartek