Skip to content

Latest commit

 

History

History
34 lines (29 loc) · 2.24 KB

README.md

File metadata and controls

34 lines (29 loc) · 2.24 KB
Raw

Shaken Lattice Statistical Analysis (Rust version)

This project began as a learning experience: to code the C++ version (see Github: End2End_SLI) of the SLI in Rust. However, this has since become the main code for generating momentum probabilities and doing the statistical analysis section of the paper. Main differences from C++ code:

  • Implemented multithreading using Rayon crate for iterators, so faster runs
  • Was able to do more precise runs, and hence do faster comparisons between JSD plots.
  • The C++ code library includes the Environment class that I used for training with the Libtorch DQNN implementation in the Holland Group. Here, we have a simpler struct Lattice that differs slightly from the Wavepacket class in C++.
  • The Rust style docs are available as .html under docs/ folder.
  • The Obsidian folder contains notes for project in Markdown.

Planned Updates:

  • Adding augmented state evolution to code (done)
  • Moving Bayesian updating to Rust from Python as well (mostly done)
  • Code for Jittery Lattice (done)
  • Separate module for importing shaking sequences as static variables (done. However, a lot of old files archived in Data rely on entering hard coded sequences)
  • Statistics module in Rust for calculating divergences etc.

Using Code

The code is pretty simple.

Generating Bayesian priors

The Rust code takes in a vector of shaking function amplitudes for the lattice, and generates final momentum probabilities for a range of acceleration and lattice depth values. To build the rust code, just use cargo build --release within this directory, and to run cargo run --release. The code will generate a test.txt file that contains the following info:

[acceleration index, lattice depth index, acceleration value, lattice depth value, P(p)]

The no. of values in $P(p)$ is determined by N_STATES constant.

Running Bayesian Updating and Plotting

There are three example notebooks for this:

  • 1param Bayesian
  • JSD plots for multiparameter
  • Comparing Bayesian between two sequences

While the notebooks may illustrate multiple different ways of plotting, recommend using them only as illustrative examples. Use the Sequence2param.py and Units.py files when doing your own plots for consistency.

Statistical Tools