(595a) Will Molecular Modeling Ever Become a Mainstream Chemical Engineering Tool?

Authors: 
Maginn, E. J., University of Notre Dame
Most people have heard the quip â??Everyone trusts the results of an experiment except the person who did it. Nobody trusts the results of a molecular modeling study except the person who did it.â?

While funny, this statement conveys several important truths. First of all, performing careful experiments is difficult and those who are the most careful are also the most skeptical of their own measurements. Good experimentalists know how many things can go wrong, and so when it comes to experimental research, we have come to insist on reproducibility and transparency. By and large the problems of reproducibility and transparency have been solved for many â??mainstreamâ? chemical engineering experimental tools such that complex techniques like NMR, X-ray diffraction, calorimetry, and chromatography are used routinely. Results from these techniques are believable because the instruments used in such experiments are commercially available and the databases and software analysis tools used to interpret the output from these instruments are well established and validated. Anyone with a similar instrument who follows the description of how the experiment was conducted could replicate the experiment. Such reproducibility is one of the main principles of the scientific method and gives people confidence that the results are â??rightâ?.

If we are honest, molecular modeling is not yet close to being a trusted, reliable and reproducible tool, which is the second thing the quip points out. There is still widespread distrust of the veracity of molecular modeling work among many experimentalists. Their skepticism is justified, in my view, and usually has to do with concerns about the use of classical potentials or small systems or short time scales. While these are all valid areas of concerns, molecular modeling practitioners know that there are many other things to worry about. Is there an error in the force field parameters? Was the proper cutoff used? Has the system been properly equilibrated? Was a valid analysis carried out in order to arrive at a property? Is there a bug in the code? It is these latter areas of concern that we as a community can and must address if molecular modeling is ever going to be a mainstream tool.

In this talk, I will highlight what I think some of the biggest challenges are when it comes to making molecular modeling more accurate, reliable and reproducible. I will emphasize best practices that we as a community should adopt and point out some examples of initiatives that are going on that are moving us in the right direction.