Incremental lambda lifting: An exercise in almost-denotational semantics

We prove the correctness of incremental lambda-lifting, an optimization that attempts to reduce the closure allocation overhead of higher-order programs by changing the scope of nested procedures. This optimization is invalid in the standard denotational semantics of Scheme, because it changes the storage behavior of the program. Our method consists of giving Scheme a denotational semantics in an operationally-based term model in which interaction is the basic observable. Lambda lifting is then shown to preserve meaning in the model