Condensing and then strongly extracting from flat mixtures #
A condenser may approximate a mixture of flat distributions whose supports vary with its retained seed. A strong extractor applied with a fresh, independent seed extracts from every component. The two errors add, and the conclusion retains both seeds. Witness injections describe the ideal conditional sources; no injectivity of the actual condenser is assumed.
This is a finite mixture formulation of the condense-then-extract argument in Guruswami--Umans--Vadhan, Proposition 4.5 and Remark 5.15 (https://people.seas.harvard.edu/~salil/research/PVcondenser-jacm.pdf), and Chattopadhyay--Goodman--Liao, Lemma 4.9 (https://eccc.weizmann.ac.il/report/2021/075/download/). The hypotheses supply the condenser witnesses and the extractor guarantee. They do not construct an extractor, prove seed-length bounds, or decompose arbitrary weighted sources into flat distributions.
Extraction after a supplied normalized mixture of seedwise flat witnesses.
Each component has at least K points, and its support may depend on the
condenser seed. Tests retain both independent seeds; their errors add.
The subset-mixture witness contract of a lossless condenser composes with a strong extractor. The resulting extractor retains both independent seeds.