Include η-reduction in interaction improvement

Determine whether η-reduction is included in the interaction improvement preorder Eint-imp on ordinary untyped lambda-terms (defined via black lifting into the checkers calculus), i.e., prove that for all terms t and u with t →η u, it holds that t Eint-imp u. This would establish that the inclusion Eint ⊆ Eint-imp is strict.

Background

The paper introduces interaction improvement (Eint-imp) as a preorder refining contextual reasoning by counting interaction steps while ignoring silent ones, in the checkers calculus and then transferring it to ordinary λ-terms via black lifting. While the interaction preorder Eint is shown to be an inequational λ-theory, its relationship with η-reduction remains unresolved.

The authors argue that if η-reduction were included in Eint-imp, it would strictly extend Eint, reflecting that η-reduction can decrease the number of interactions in head evaluations (as illustrated by their examples). However, standard rewriting techniques and their semantic tools do not presently handle η within the checkers framework.

References

Interaction Improvement and n. Note that the corollary does not say anything about Cint imp and n. We conjecture that n-reduction is included in Eint-imp, which would imply that the obvious inclusion gint c gint.imp is strict, since it is not included in Eint (Ex. 3.6) (beware that strictness of the inclusion does not follows from Cox CEct as strictness in that case relies on black and white terms). At present, however, it is only a conjecture.

Interaction Equivalence  (2409.18709 - Accattoli et al., 2024) in Section 3, Interaction Improvement and n