Causal inference in the continuous limit
2021-02-17 — 2021-02-17
Wherein graphons are examined and are found insufficient to capture covariate-driven spatial or temporal proximity for fields of continuously indexed variables, and light-cone intuitions are considered.
Where our causality is over fields of continuously indexed variables rather than discrete nodes, what can we do? Do we talk about light cones? Wave propagation? Covariance?
Unclear.
Graphons look like they should help here, except they do not capture what I would typically imagine about causality, e.g. which covariates influence causation. Something which encodes spatial proximity, or temporal ordering, or other covariates, looks more useful at a glance? Presumably, I am behind the literature.