Wednesday, May 22, 2013

Listen to experimentalists (sometimes)

Over the years I have benefited greatly from my interactions with experimentalists. These interactions have varied from informal discussions, listening to talks, and reading papers.
These interactions have led me to work on interesting and important problems and helped make my theoretical work sharper and more relevant to experiment.

But on reflection, I regret I have also wasted significant amounts of time, energy, and money because I have listened (too much) to experimentalists.

So is there a key to getting the benefits without the liabilities?

I think the key is to listen to the "broad brush strokes" and not get distracted or hung up on the details.

Experimentalists can teach us
  • what measurable quantities we should aim to calculate [e.g. thermopower vs. temperature, interlayer magnetoresistance vs. magnetic field direction]
  • what specific systems or materials are of particular interest
  • what is actually known about specific systems or materials [e.g. the shape of the Fermi surface]
  • orders of magnitude and reasonable parameter ranges for experimental observables
  • we should not believe every published experimental result. Things can go wrong. If you visit a lab and see how complicated and delicate some of the apparatus are you may wonder why there aren't more spurious results!
Theory papers often contain things like a graph of ground state energy vs. a variational parameter for a trial wave function. That may be of some theoretical relevance or importance. But, it is not something an experimentalist is going to be interested in. It is not something they can measure!

Theorists will sometimes be not that concerned with the actual magnitude of energy, temperature, or field scales. Instead they just work with some model parameter, e.g. t and U in a Hubbard model. But, experimentalists really want to know how this translates to temperature or magnetic field. Earlier I have posted about this issue for the case of magnetic fields in Hubbard model calculations.

So when should theorists ignore experimentalists? When do I wish I had ignored them?

I think my mistake has been to sometimes get caught up in the puzzle of specific results of individual experimentalists.

Suppose an experimentalist comes to you and says "We have these interesting/weird results on this exotic material using our fancy new fangled measuring technique. Maybe you can help us explain them." I recommend listening politely and not getting involved. Wait until another group has independently observed similar results on a different sample with a different technique.

I also think that sometimes experimentalists get hung up on small discrepancies between theory and experiment and try to get us to explain them. There are important historical cases where this has been important. But, I think often the benefits to theorists can be marginal.

I have just focussed on the scientific benefits from the fruitful interaction of theorists with experimentalists. I think there can also be some significant career benefits (and liabilities) because generally experimentalists have more money and clout than theorists and so being liked by them can be a significant career booster. But, that is another story....

I welcome comments and peoples own experiences.
I would also be interested to hear an experimentalists view on listening (not too much) to theorists!


  1. I left my thoughts on your last point on your blog some years ago...

    I wrote then...

    "I remember reading this in a back issue of PT and the opening line stayed with me.

    "The principal error I see in most current theoretical work is that of imagining that a theory is really a good model for . . . nature rather than being merely a demonstration (of possibility)—a 'don't worry' theory."

    I would say that this also the error in the use of theory by experimentalists to fit their data. In many-body physics, theories are almost always too idealized to expect that they will be an excellent fit to data. Most of the time theory should be taken as a demonstration of possibility or as a clue to how analyze a set a data, but explicit fitting to data is almost always a no-no.

    Exceptions of course exist when one is talking about matters of symmetry, or at asymptotically low energies, or when there is a gap in the system (It is no coincidence that these are often different perspectives to the same three sided coin)."

    1. Hi Peter,

      Thanks for the reminder of your helpful thoughts.


  2. I wonder if there is a simple question or set of questions that distinguishes abstract mathematical modeling from great statistical physics work like Baxter's or Onsager's. It seems to me --- perhaps I'm quite wrong --- that statistical mechanists require minimal interaction with experimenters (excluding the fact that the model to begin with might have been experimentally motivated) during the course of their work and yet produce good stuff.


  3. Hi Ross - Like I've said before, I think the best theorists know quite a bit about experimental work (what is realistic, how seriously should one take certain types of measurements), and the best experimentalists know a fair bit about theory (how seriously should one take certain calculations, when should one expect tight quantitative accuracy vs. qualitative scaling). The theory/expt relationship is much better (perhaps I should say "healthy") overall in condensed matter and AMO than in high energy, in my view.