Dear all,
I’m a new MD learner and you can say me a beginner. For very basic study I’m following the refereed books by MD community. I’ve some confusion regarding the very basics of MD. I know this is not related to lammps, but as a huge number of MD experts all over the world are following this forum so it’s a opportunity for me to know from them. I think admin would allow me to put my Qs. here.
A. I’ve read statistical molecular thermodynamics. I’ve learned how to get macroscopic properties from microscopic variables via partition function of different ensembles. But during my MD study I can not get the correlation how I’m invoking this ensemble idea and Ergodisity in MD calculation ? Why most of the MD study says that my system is assumed to be Ergodic ? After reading the theory behind MD simulation, it seems to me that I’m doing integration of EOM and applying some constraints (like fixed T and/or P) and finally getting updated coordinates then applying different formula (like, 1/2 m v^2 = 3/2 NKT to get T) for property calculation. I’m apologetic if I’m wrong, but it’s really unclear to me how we are applying ensemble idea also bypassing partition function in MD ?
B. Whats the actual difference b/w equilibration and minimization, when in this both way we are searching for minima of PES ?
Please help me to learn. Thanks in advance for your reply.
Regards,
R D K R
Dear all,
I'm a new MD learner and you can say me a beginner. For very basic study I'm
following the refereed books by MD community. I've some confusion regarding
the very basics of MD. I know this is not related to lammps, but as a huge
number of MD experts all over the world are following this forum so it's a
opportunity for me to know from them. I think admin would allow me to put my
Qs. here.
A. I've read statistical molecular thermodynamics. I've learned how to get
macroscopic properties from microscopic variables via partition function of
different ensembles. But during my MD study I can not get the correlation
how I'm invoking this ensemble idea and Ergodisity in MD calculation ? Why
most of the MD study says that my system is assumed to be Ergodic ? After
reading the theory behind MD simulation, it seems to me that I'm doing
integration of EOM and applying some constraints (like fixed T and/or P) and
finally getting updated coordinates then applying different formula (like,
1/2 m v^2 = 3/2 NKT to get T) for property calculation. I'm apologetic if
I'm wrong, but it's really unclear to me how we are applying ensemble idea
also bypassing partition function in MD ?
B. Whats the actual difference b/w equilibration and minimization, when in
this both way we are searching for minima of PES ?
Please help me to learn. Thanks in advance for your reply.
please find a local teacher. there are plenty of them around. these
kind of discussions are very time consuming to do over e-mails and
thus this is not the right way to learn MD. you don't need an expert,
just somebody that knows enough.
axel.
Dear Sir,
From your discussions in lammps mail list I learned a lot regarding many lammps issues. I approaches to my local MD teachers but my confusions are remain unclear obviously I’m not so sharp to get them. Please can I get a piece of advice from you regarding this ?
i already *gave* you advice. this is a mailing list for users and
developers, not a classroom and the fact that you struggle
understanding your teachers doesn't entitle you to personalized
tutoring from people on this list. if you want a personal
teacher/trainer/consultant, you would have to hire somebody to do
that. after all, it *is* the job of your teachers and advisers to
teach you.
axel.
Another suggestion, if for some reason you find it difficult to learn from people, it might be good to get a good book specifically on how computer experiments actually work. I’m not sure if there is a list of good textbooks on the LAMMPS website or not, I personally find “Understanding Molecular Simulation” by Frenkel and Smit to be quite good, but there might be others as well.
Another suggestion, if for some reason you find it difficult to learn from
people, it might be good to get a good book specifically on how computer
experiments actually work. I'm not sure if there is a list of good textbooks
on the LAMMPS website or not, I personally find "Understanding Molecular
Simulation" by Frenkel and Smit to be quite good, but there might be others
as well.
MD is not a field that is easy to learn from books. But while the
physics behind computing MD trajectories is easy, there are a lot of
practical issues to consider and they often require a lot of practical
experience that can only be obtained from mentoring. That is my
explanation for why we have to few good books on MD. The Frenkel and
Smit book is probably the best around. There also is quite a bit of
practical knowledge in the Allen and Tildesley book, but a lot of the
technical details in that book and what is considered practical are
way outdated compared to current best practices.
One of the recurring themes we see in people struggling with
understanding MD is that they are rushing ahead to fast and skipping
too much of the basics. In that regard, MD is more a craft than a
science and it just requires some time and banging your head against
the wall. In the same way, e.g. chemists have to do basic experiments
in the wet lab (that nobody otherwise do, but rather buy the
compounds), or mechanics have to work with simple tools to produce
workpieces that you would normally let a CNC machine do and so on. One
has to realize that there is a difference between "locating the
answer" and "understanding a problem". sadly, the current education
culture favors the former instead of the latter (as it is often very
effective until you reach the point, where google doesn't immediately
point you to the solution).
axel.
Hi Dines,
I’m also in learning phase of MD and playing around lammps. Unfortunately i don’t have such MD mentor who can help me to understand rather every time when I go to them, they instruct me to use this this and that fixes in lammps to do my simulations. The scenario is quite frustrating here. Anyway, I’m trying to answer your Qs as per my understanding.
A. how we are invoking ensemble idea in MD ?
As we know, to get macroscopic properties from microscopic variables statistical mechanics is the right approach. But due to the multidimensional nature of partition function we can not solve it. So we don’t do ensemble average in MD directly. but the another way is doing dynamic averaging/ time averaging. Here Ergodisity comes into play i.e. if we do our averaging for sufficient time and assuming that every micro-state is equally probable in a macro-state then those two average property should be equal. Here we are taking this central cell of our MD system as a micro-state of that particular system. And to invoke ensemble idea we are applying constraints in the equation of motion (EOM), e.g- if we like to sample our trajectories by NVT then to do so we are just switching off those d.o.f and controlling velocities (in different way) to maintain desired temperature. So to apply ensemble essence to MD calculation means we are restricts and/or giving constraints to the EOM of the particles. And we are doing averaging for enough time assuming that Ergodisity would valid.
B. Difference b/w equilibration and minimization.
In every MD calculation you need initial structure of the system. It may be experimental and/or guess structure at 0 K. But when the provided initial structure is much different than the actual, that time we need to minimize our initial coordinates to be approached to the actual. in that case we do minimization at 0 K. But eqilibration means we are trying to get the equlibriated geometry at non zero desired temperature. In this process of equilibration Liouville’s function (density of phase point) should be constant throught the phase points.
It is quite obvious that i have some misconception and know wrong. But i know that somebody having much MD knowledge would correct me and help us, the leaner. I know that our teachers here (surely my local teachers) are also following this mail to learn something.