Friday, March 28, 2014

Why do we say prediction when we really mean postdiction?

I have heard it claimed that some string theorists say that the theory does make experimentally testable predictions because it predicts gravity! I think this is silly because it is a postdiction or retrodiction, i.e., one is explaining already known phenomena not suggesting new experiments.

However, we should not just make fun of string theorists, because most of us actually do it to! For example,
"our LDA calculations predict a bond length of 2.145 Angstroms, compared to the experimental value of 2.15 Angstroms" 
"our numerical simulations for this Hubbard model predict a superconducting transition temperature of 150 K."
I do it too! In my latest paper I just counted 11 times where I wrote things like "our model predicts….". Yet in every case it concerns quantities that have already been observed.

Why do we use "predict" like this?
My guess is that it is because once we write down some model Hamiltonian and make some sort of approximation to calculate some observable beforehand we don't know exactly what the outcome of the calculation is going to be.
I welcome suggestions of other justifications.

Why should we stop this practise?
I think using "postdiction" would highlight just how feeble some of our theoretical efforts are. It is very rare in condensed matter and chemistry that theorists predict new phenomena or can make accurate quantitative predictions about unmeasured quantities. This is the challenge of emergence. Furthermore, the choice of model Hamiltonians and approximations is usually based on an intuition as to what is going to give us the answer [both qualitatively and sometimes even quantitatively] that we want.

I think this also highlights the value of the method of multiple alternative hypotheses including comparing the results of different models, methods, and approximations.

This practise seems to be so ingrained it is hard to see it changing. Should it? I welcome comments.

No comments:

Post a Comment