Wednesday, November 11, 2009

Computational modeling of complex chemical systems: the state of the art

If you were going to an isolated island to do computational chemistry and you could take this year’s computers and 10-year-old algorithms or this year’s algorithms and 10-year-old computers, which would choose to take?

This is a question that Donald Truhlar asks in a JACS Editorial for a Select issue of 23 papers on Molecular Modeling of Complex Chemical Systems.

How would you answer the question?
You can look in the article to see how most computational chemists would answer the question.

The article is a very nice read to a physicist because it provides a very helpful and concise summary of historical landmarks in the computational modeling of large chemical systems.

However, I disagree and am concerned with one of the opening statements in the article:
Almost all modern theoretical chemistry is computational chemistry, because most of the progress that can be made with pencil and paper without a computer has been already made. Computations on complex systems are, in my opinion, the current frontier of theoretical chemistry.
I fear this does reflect the view of most in the theoretical chemistry community. However, (as a physicist) I think a much greater emphasis needs to be placed on gaining insights, finding organisation principles, and developing analytical models that complement simulations. But, that is what I am trying to do (and excited about!).


  1. The problem is that the models used in simulations of complex systems are themselves complex, so many people who use them do not try to understand how to interpret them. It has become sufficient to write papers that say "I used method Y and got number X", without discussing what state or ensemble of states is actually being modelled by method Y in the first place. Quantum mechanics tells us that the answer we get depends on the question we ask, so it is VERY important to understand exactly what question one is asking if one wants to make sense of the answer!

  2. I would obviously this year computers and ten-years-old algorithms. Moore's law is much faster than the ingenuity of coders.

    By the way, I am also a physicist doing computational chemistry and I believe that there are a few or maybe no organizing principles and/or useful analytical models for complex chemical or biochemical systems. The fundamental laws are known and we only need to look for good enough approximations that are not too computationally demanding. This depends on the problem and on the question asked.

    Useful biochemical systems, the ones I work with, do not have symmetries or few particles or anything that can help simplify the analytical pen-and-paper analysis.

    Maybe physicists moving into new complex-systems grounds need to ask themselves if the approach mentioned in the post, which has been so useful in, say, particle physics has also to be useful in other fields. I believe that the answer is "not necessarily". Different branches of knowledge are approachable by in principle different ways of thinking. What has worked for particle physics does not have to work for complex systems, and in fact I believe it is not working.

    Nice blog!