Here’s a supercool controversy for ya

Raghu Parthasarathy writes:

You might like this very good article by Ashley Smart on a recent fight about the statistical mechanics of water, and a feud that was made worse by a lack of sharing code

Condensed matter theory! That’s what I worked on, back when I was a physicist. We did an experiment that demonstrated superheating of silicon. It turns out it can get really hot inside and still not liquefy, as long as the wafer is defect-free so that there’s no place for the melting to start.

9 thoughts on “Here’s a supercool controversy for ya

  1. “We’ll never know if the results were enough to sway Chandler; that April, the 72-year-old professor emeritus succumbed to prostate cancer.”

    Cue Max Planck…

  2. So from the article I got the impression that the fault was in using “incorrect” initial values for the MCMC sampling. Was that all? Scary that that could have such effects. Or was it improper priors?

    Anyone understands this and could explain it like I’m not at condensed matter theorist (ELINCMT)?

    • My impression was it was a problem with the transitions. To propose a new state they would perturb the existing state and then run molecular Dynamics until the system equilibrated to a new value. The way they were doing this induced some kind of unintended change in temperature.

      If you think about it in terms of stats style MCMC imagine you create a proposal by uniformly perturbing every dimension in your problem and then running a Gibbs sampler for a few thousand steps and then accept the final state… if the uniform perturbation was too big, a few thousand steps of Gibbs sampling won’t be enough to give you a sample that’s in equilibrium with the target distribution. I think this is kinda what was going on

    • There is a detailed explanation in http://ftp.aip.org/epaps/journ_chem_phys/E-JCPSA6-148-015813/760767_1_attach_15_7469_convrt.pdf which is the supplementary material of from https://aip.scitation.org/doi/full/10.1063/1.5029463

      In summary, the problem is that the Monte Carlo simulation proceeds by doing a manipulation on the current state to generate a proposal for the next state and they were doing it wrong: “exchanging space-frame site velocities between molecules with different space-frame orientations will generally lead to violations of the rigidity constraint.”

Leave a Reply to jim Cancel reply

Your email address will not be published. Required fields are marked *