Interesting. Reminds me of the “snake oil” technique useful when evaluating complex sums in combinatorics. By treating the sum in question as the coefficient of x^n in another sum, you can often obtain an expression where finding the coefficient of x^n — which is the value you’re after — is simple.

There are some important mathematical subtleties that this article glosses over. More details can be found here:

http://fy.chalmers.se/~tfkhj/FeynmanIntegration.pdf

I think this is the same trick I eventually (too late) found to make my model tractable assuming a bivariate Normal. It certainly hadn't been my first instinct to consider rate of change with respect to correlation.

Highly relevant XKCD: https://xkcd.com/2117/

I wonder how many downvoters of this relevant XKCD reference actually read the article and understand that Feynman's Integral Trick involves differentiating under the integral sign. The relevant XKCD is about how much simpler it is to differentiate than to integrate (which can frequently be impossible using standard college calculus techniques).

Great article! Given we are all working from home now, some of the advice here is helpful even when just a regular dev, as opposed to DevRel.