SimLib

This package contains all the ingredients to run simulations of disordered spin systems.

Basic concept

Each step of the simulation (with the exception of ED) has two types associated with it: A Descriptor holding all relevant parameters and a Data type which contains the descriptor and the computed values. Usually the workflows look like this: You initialize a Descriptor with the parameters of interest and either create or load the corresponding Data (depending on whether you ran the simulation already or not). When creating new results, the code will try to load required data for the computation (f.e. the ensemble computation need eigenstate occupation, operator diagonal and eigenenergy data) if you did not pass in data explicitely.

Note: It's not really possible to pass additional data to create. Right now there are (sometimes) specific methods (i.e. ensemble_predictions) for the given computation, but create never accepts additional arguments. This should probably be changed to unify the interface of the computational tasks.

Exact diagonalization

There is an exception with computations that require Exact Diagonalization. It is a lot more efficient to bundle computations s.t. you only perform the diagonalization once and reuse it for different quantities of interest. To facilitate this, you can specify the parameters once (using EDDataDescriptor) and then call run_ed with a list of Tasks to fulfill. This results in a list of Data structs - one for each task.

Saving/Loading

Every descriptor has a SaveLocation to control where the data should be saved to/loaded from and knows how to generate a good file from it's values. The SaveLocation(prefix, suffix) denotes the path prefix/<descriptor-specific-name>_suffix.jld2. Derived (or otherwise related) data also keep that information.

All methods dealing with file operations take also prefix, suffix keywords which take precendence over the SaveLocation of the Descriptor at hand.

Tests

There are also some very basic tests around, running all simulation steps for some small parameters, and some notebooks to visualize raw data (useful for finding bugs).

Example

For a small example showcasing a simulation script, see notebooks/generate_test_data.jl (which actually not a Pluto.jl notebook and probably should be in a different place).

Implemented quantities

• Energies (= levels)
• eigenstate occupation
• operator diagonal (= eigenstate expectation values)
• Half-chain entropy (arb. subsystem lengths, only for zblock basis)
• eigenstate locality
• inverse participation ratio (basic)
• Level spacing ratio
• Ensemble prediction

Every quantity (except the last 2) need to be computed as an EDTask! LSR can be computed via a Task.

Sparse diagonalization

New Feature: Construct EDDataDescriptor with diagtype=Sparse(σ, count) to compute only count many levels around σ!