## Deriving natural laws from raw data

So I read this article today. Cornell researchers have successfully implemented a parallel computer system that is able to basically derive physics laws from observation data. This is extremely interesting. Before deciding to take computer science, I was split between majoring in physics or computer science (and, to much lesser extent, mathematics). To see an inter-disciplinary research like this one made me very happy and impressed. In fact, I do believe that many of the more interesting research (current and future) will involve multi-disciplinary twists. Just recently I read a paper describing translation of polynomial (or restricted-polynomial), completely-partitionable, first order differential equations into distributed protocols (that satisfy the same mathematical properties of the derivatives)^{1}.

So, in this deduction system, it basically figures out derivatives (changes in values w.r.t. to some other changes in the system, e.g. time, position, etc.) that describe the observables, and uses this derivatives to derive system invariants. Technically, invariants are not the actual physical equations. However, it encodes all the information needed to derive physical equations for the system. With this technique, Cornell researchers were able to derive conservation of energy and momentum, and also Newton’s second laws from observations taken from simple experiments (including spring-loaded acceleration, simple and double pendulums).

Read the article for more details. Also, I’m definitely waiting to read the Science article (Vol. 323, No. 5924) that was supposed to come out on April 3 (the digital version seems to not have come out though).

- Gupta, I. On the Design of Distributed Protocols from Differential Equations. PODC’04. [↩]