Saturday, July 20, 2013

Are large atomistic quantum dynamical simulations falsifiable?

There are a whole range of dynamical processes found in biomolecular systems that one would like to simulate and understand. Examples include:
  • charge separation at the photosynthetic reaction centre
  • proton transfer in an enzyme
  • photo-isomerisation of a fluorescent protein
  • exciton migration in photosynthetic systems.
These are particularly interesting and challenging because they lie at the quantum-classical boundary. There is a subtle interaction between a quantum subsystem [e.g. the ground and excited electronic states of a chromophore] and the environment [the surrounding protein and water] whose dynamics are largely classical.

[image from here]

Due to recent advances in computational power and new algorithms [some based on conceptual advances] it is now possible to perform simulations with considerable atomistic detail. This is an example of "multi-scale" modeling.

However, it is important to bear in mind the many ingredients and many approximations employed, at each level of the simulation. These can include

  • uncertainty in the actual protein structure
  • the force fields used for the classical dynamics of the surrounding water 
  • the atomic basis sets used in "on the fly" quantum chemistry calculations 
  • density functionals used [if DFT is a component] 
  • surface hopping for non-adiabatic dynamics [e.g. non-radiative decay of an excited state].
These approximations can be quite severe. For example, a widely used water force field used in biomolecular simulations, TIP3P, predicts that water freezes as -130 degrees Centigrade!

Then there is the whole question of how one decides on how to divide the quantum and classical parts of the simulation.

If there are 10 approximations involving errors of 20 per cent each, what is the likely error in the whole simulation? Obviously, it is actually much more complicated than this. Will the errors be additive or independent of one another? Or is one hoping for some sort of cancellation of errors?

What is one really hoping to achieve with these simulations?
In what sense are they falsifiable? 
When they disagree with experiment what does one conclude?

An earlier post highlighted the critical assessment  of one expert, Daan Frenkel of the role of classical simulations. I hope we will see a similar assessment of quantum simulations in molecular biophysics.

1 comment:

  1. I think the answer is conceptually obvious: do they give non-obvious predictions relating to an experiment that hasn't been done yet?

    My feeling is that usually this kind of simulation is more useful qualitatively than quantitatively, e.g. it can provide the shape of the phase diagram, or similar equivalent for the system, but not the values of transitions. But if this is to be the case, instead of trying to pack in all the physics 90% right, we're much better off distilling the system down and finding the 10% of the physics that actually determines the behaviour, and using that to guide our experiments to determine, probably phenomenologically, the more detailed information.