ecboghiu / inflation

Implementations of the Inflation Technique for Causal Inference.
GNU General Public License v3.0
22 stars 3 forks source link

Physical for hybrid networks #106

Closed ecboghiu closed 1 year ago

ecboghiu commented 1 year ago

This enables the generation of physical monomials for hybrid networks for linear programming relaxations of inflation.

A simple example where this is useful is that of full-network nonlocality. This PR includes a test where the critical visibility of the EJM distribution is reproduced for a certain parameter $\theta$ of the EJM family.

Given an inflation scenario, with certain number of copies for each source, a combination of events is said to be compatible if there can exist a joint probability distribution for both events. Consider a party $A$ that has access to two sources. Let $A^{ij}$ represent the event of the party using copy $i$ of the first source and copy $j$ of the second source to produce correlations. Then when both sources are unclonable, well defined probabilities $P[A^{ij} A^{i'j'}]$ exist only when $i\neq i'$ and $j\neq j'$. However, if the first source is clonable, then probabilities $P[A^{ij} A^{i'j'}]$ are well defined for all $i, i'$ as long as $j \neq j'$.

In the future this functionality, which now is implemented in InflationProblem, will be used also for InflationSDP.