Plans for the week of March 2-6

Welcome back to FYS4411/9411. The plans for this week involve

1) a repetition from last week (see the notebook from week 6)? about the implementation of gradient descent and the equations we need to encode.? Here we will mainly focus on the elements we need to have in our codes.

2) Then we will discuss in more detail various quasi-newton methods. The plan for the lecture is to discuss methods beyond the so-called brute force gradient descent and discuss their suitability for project 1 and project 2 eventually. We will thus discuss

Gradient methods:

Quasi-Newton methods (Broyden's algorithm, Broyden-Farberg-Goldberg-Shanno algorithm, and Powell's algorithm)

Steepest descent and conjugate gradient descent

Stochastic reconfiguration if we get time

The slides/jupyter-notebook are at https://github.com/CompPhysics/ComputationalPhysics2/tree/gh-pages/doc/pub/week7

The lecture this week is thus more technical, with a more in-depth discussion of the various quasi-newton methods. Hopefully these lectures will then give you a better overview of different algorithms for optimizing functions. The lab runs as usual from 11.15am to 2pm. The lecture will be recorded.

Best wishes,

Aleksander and Morten