Take the 2-minute tour ×
Physics Stack Exchange is a question and answer site for active researchers, academics and students of physics. It's 100% free, no registration required.

Not sure if this is a physics or chemistry question. But if the motion of atoms and it's particles can be described by quantum mechanics, then is there a software that simulate full atoms and it's boundings, in a way you can visualize them, and that can be used, for instance, to throw 2 molecules together and watch them reacting?

share|improve this question
    
Hi Dokkat This is kind of an overlap topic, but since you're asking about chemical processes in more detail than "X reacts with Y" I think it's okay here. By the way, if you're working with scientific software, you might be interested in the beta Computational Science site. –  David Z Jan 1 '12 at 6:49
    
The problem here is that there is no "one molecule slowly flows towards the other, their electronic clouds get in touch and the new molecule is now born", it's rather 'there is a 0.000xxx% probability that this system will turn into this state', and it's rather hard to visualize all possible probabilities.... –  BarsMonster Jan 1 '12 at 10:30
    
Molecular dynamics studies often use gromacs. I'm not sure whether it comes with a visualization tool, but I know people get it to work with some such tools. –  Adam Zalcman Jan 1 '12 at 11:05
    
@BarsMonster: The picture of two molecules slowly flowing, and the electrons reconfiguring, is the nearly perfect Born Oppenheimer approximation, and is not a problem at all. The issue in most day-to-day simulations is that the reconfiguring is not done dynamically for the valence electrons, leading to completely wrong forces on the nucleus position. This is fixed by Car Parrinello MD (and only by hybrid methods like this, which do a minimal amount of quantum chemistry to find the force). The CPMD will describe the process as the OP requested. –  Ron Maimon Jan 3 '12 at 7:05
    
Possible duplicates: physics.stackexchange.com/q/10311/2451 and links therein. –  Qmechanic Apr 2 at 14:15

2 Answers 2

up vote 3 down vote accepted

There are many, many algorithms and pieces of software to do this. In addition to Molecular Dynamics, there are also methods based on statistical simulations in Quantum Monte Carlo, and density functional theory as implemented in programs like Quantum Espresso. It is a simple and worthwhile exercise to program these things yourself - if you wish to study the oscillatory behavior of a molecule subject to some arbitrary external potential, you can do this quite readily using basic programming and visualization tools provided you establish the proper functions and equations to describe your system.

I will note that these algorithms all have explicit ranges of validity and underlying assumptions, and one must very carefully understand the limitations before interpreting the results. In many cases, the accuracy and precision of the algorithms will be questionable, because assumptions at some level have to be made to reduce the system size since not even the most powerful supercomputer can handle a calculation with anything approaching a macroscopic number of particles. Nevertheless, they can provide some sense of the starting point and can give insight into trends.

Edit to add: See giant list of software applications for Quantum Chemistry

share|improve this answer
    
Nothing but MD gives a time-dependent picture as far as I know. –  Ron Maimon Jan 2 '12 at 9:02
    
@RonMaimon - Car-Parinello molecular dynamics is a level of theory that combines electronic structure with MD, and is implemented in codes such as CPMD. –  Richard Terrett Jan 2 '12 at 11:57
    
@Richard Terett: That's interesting, I didn't know this method, but reading the description, if you implement it well, it should be the ideal way to deal with the main issue of electronic delocalization. But there is also the less pressing issue of nonlocal electromagnetic forces, which is important also for large molecules. The final issue is with the stochastic approximation. Thanks for the pointer. –  Ron Maimon Jan 3 '12 at 6:19
    
@Jen: It is important to say flat out that the MD codes, leaving aside CPMD, are no good, because the electronic fluid is mis-simulated, leading to much more floppy molecules than the ones you find in nature. Further, only CPMD can deal with bond-breaking and reforming. It is the only correct simulation method. Combined with a good nonlocal electrodynamic force model and a good coarse-grained in time stochastic updating, it gives the only complete answer to the question. –  Ron Maimon Jan 3 '12 at 6:49
    
@RonMaimon - Glad CPMD was of interest to you. It's certainly not the only 'ab initio MD' method but it seems the state of the art is very experimental. I saw a few lectures recently on 'quantum dynamics' which take traditional ab initio calculations into the time evolution realm but it seems fairly bleeding edge, e.g. AFAIK the Born-Oppenheimer approximation needs to be replaced with a nuclear wavefunction. Incidentally, this is just mind-boggling: petachem.com/demo.html –  Richard Terrett Jan 3 '12 at 12:07

EDIT: Car Parrinello Works!

The criticism below applies only to the type of molecular dynamics done using molecular potentials, as exemplified by CHARMM. This is the only molecular dynamics I had been exposed to, and it is total crap.

There is a second kind of Molecular dynamics which includes the valence electrons, a core potential, and an clever algorithm to update the valence electron fluid. This is the Car Parrinello method. The Car Parrinello method contains just the right amount of information to do a quantitatively correct simulation, it is the ideal computational dynamics, and I am stunned to learn about it. Thanks to Richard Terrett for pointing it out.

EDIT: Answer to Question

Use CPMD. If your molecule is in solution, use CPMD plus a brownian random force representing the effects of the solute.

Criticism of everything else

There is software which claims to do this, it is called (non CPMD) Molecular Dynamics, and Molecular Dynamics is often used in Chemistry and material science to produce simulations which the proponents claim are quantitatively accurate to predict the detailed microscopic motion of say organic molecules.

The principles behind this software are fundamentally deeply flawed, and, without huge modifications, it simply does not give quantitatively accurate results. Molecular dynamics doesn't quantitatively work for anything at all beyond the homogenous non-conducting solids and small molecules it is calibrated to match.

I consider this a big problem, since a lot of research money goes towards producing this software and calibrating it, and there are a lot of people involved in promoting this stuff, which is a total waste of time and money.

What is Molecular Dynamics?

The idea is to model each atom as a point, and to give a force-law between adjacent atoms, of each different type. So that there is a C-C force law, which models the preferred bond-angles, and a H-C force law, and and O-C and an O-O force law. Then you add up the forces. Then you correct the model for three-nucleus interactions, like a H-H-O interaction which fixes the water bond angle, and so on, until you get a reasonable match to a large swath of experimental data.

Then you take a molecule where you know the structure, and you simulate the atoms using these force laws derived from simple molecules and monatomic solids. The idea is that you are getting some approximate picture of the dynamics from the structure only. This allows you to get a dynamical picture of chemistry

Molecular Dynamics Is A Complete Fake

The reason Molecular Dynamics fails is that the relevant electronic and electromagnetic forces between atoms and molecules are not at all local, not even to a good approximation, and the fundamental approximation in Molecular dynamics is that you can model them using local forces, local pushes and pulls that don't depend on the global positions of far away atoms. These forces are assumed to be mechanical, like the forces in a tinkertoy model, and this means that forces transmitted like sound waves between atoms, while in actual molecules the relevant forces are often purely electronic and grossly nonlocal.

It is simple to illustrate this using symmetric molecules: if you have a Benzene ring, the electrons propagate around the ring, like a wire conducting a current, and these electron currents give the ring rigidity, like a microscopic stiff metal wire. The mechanical properties of Benzene cannot be approximated by a model which does not take into account the delocalized electrons in the ring. The electronic currents is disrupted by moving the carbons, and they transmit forces at the speed of electron-density-variations (the orbital speed of electrons in the Bohr model, much greater than the sound speed).

This is not an atypical situation. The molecular backbones of DNA, RNA contain delocalized electrons, and the mechanical properties of the molecules are determined by the regions where the electrons delocalize. You just can't predict the stiffness of DNA by knowing the stiffness of two-atom or three-atom sub-parts, without knowing the easy paths for electron delocalization.

This problem is nearly impossible to fix, because adding more nonlocal forces requires sophisticated calculations which are fundamentally different type than 2-body and 3-body potentials. They require a seperate fluid flow model over the atoms at the very least, which will give the delocalized electron flow.

Is it fixable?

I think that the fundamental idea of MD is wrong, so I don't hold out any hope for it being useful at any time, ever. But a different idea might do the same thing correctly.

In order for something like MD to work, it must take into account at least:

  • Delocalized electrons in molecules
  • long-range electromagnetic interactions

Most importantly, even if you know the exact forces, the idea of simulating molecular motion by accelerations and collisions is simply idiotic. The simulation should be at the very least stochastic, not deterministic, so that the momentum flows through each local region is found by Monte-Carlo, while only the slow global variables (like the overall shape of the molecules) are simulated by stochastic dynamics. The little things are always either frozen out (like electrons) or in thermal equilibrium (like jiggling nuclei or small molecules).

  • MD needs to take into account local statistical equilibrium.

CPMD fixes the main issue, of electronic delocalization. This issue is so pressing, because without it, molecular simulations are basically fraudulent.

share|improve this answer
    
Downvote away, MD is still crap. –  Ron Maimon Jan 1 '12 at 19:00
    
Didn't DV, but a few points. While this is entertaining, you're off the mark in the following: First, MD is not the only game in town when it comes to doing this stuff, and this isn't a question about how bad MD is, though I guess it is within scope. Second: You're correct that MD doesn't handle non-local electrons, but again as you point out it isn't meant to - don't get mad at the toaster because it can't cut the grass! Third: Don't reinvent the wheel again. DFT, QMC, and many others exist, with basic conditions of validity and applicability. –  Jen Jan 2 '12 at 6:35
    
@Jen: The problem with MD is that it is sold as actually being accurate for something, and people get sucked into it, and write bogus papers based on the results. DFT is too numerically intensive for real life simulations. The only thing delocalized electrons really do is create certain bending energies in delocalized chains, which can be approximated geometrically without a full simulation of the details. You can give a description of the bending of a garden hose just by using a coarse model for the water, but if you just use the rubber elasticity, and there is water flowing through... –  Ron Maimon Jan 2 '12 at 8:19
    
Quantum Monte-Carlo is no good either for real-time simulations. The molecular dynamics idea, of doing a classical simulation with a force law, is correct (if implemented stochastically with the right coarse time scale). It is just the actual implementation, which does not parametrize delocalized electrons and nonlocal electrromagnetic forces, which is off. It is not too difficult to model both phenomenologically in a way that is not computationally intensive, but which requires one-time data on how the electronic energy of delocalized electrons in different atomic clusters depends on shape. –  Ron Maimon Jan 2 '12 at 8:22

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.