marshrossney/percolation

A "computer experiment" for first-year UG labs, loosely based on pandemic modelling

Hello

🔭 Right now I'm working on ...

• my phd thesis and accompanying code

🔜 In the pipeline ...

• generative AI in physics undergraduate degrees

🔥 On the back burner ...

• pyrex: Python tools for reproducible experiments
• biodiversity metric: some quantitative analysis of the UK biodiversity metric, which I do not like

👀 In future I'm looking to collaborate on ...

• Statistical models for nature restoration and rewilding
• Generative AI in education
• Effective field theories for neural networks

percolation's Issues

Model is an awkward superposition of node-based and link-based operations

Since the model started life as a simple model of a pandemic, it made sense at the time to write everything in terms of the nodes, which played the part of 'units of population'. For example, each node has a 'neighbour list' which contains indices corresponding to other nodes from which the virus may be transmitted, and the initial 'nucleus' of infection is defined by L^2 nodes forming a square on the lattice.

With a percolation model based on a network, it makes infinitely more sense to write everything in terms of the links. The underlying object becomes an adjacency matrix representing the weights of the connections between nodes. This is way more flexible. For example, anisotropic models become asymmetric adjacency matrices.

Since #1 the model kind of mixes these two ways of operating and it's kind of confusing. It probably needs a modest but ground-up reformulation that explicitly uses the adjacency matrix everywhere, renames some things and updates the docstrings.

Some things to think about:

• Currently, we use a boolean mask to remove the inert nodes from the transmission step. We should instead represent inert nodes by columns of zeros in the adjacency matrix.
• Links can be time-dependent. So e.g. we need not have a hard boundary between infected/transmitting and recovered.