Transporting antichains across a measurable embedding #
A measurable embedding g : X → Y between Polish spaces that preserves and reflects two
equivalence relations (r' (g x) (g y) ↔ r x y) transports Cantor antichains in both
directions (hasCantorAntichainOn_image_iff), hence thinness
(isThinOn_image_iff). The ambient class A need not be Borel, and neither relation need be
Borel: only the witnesses are Borel.
The argument transports a Borel witness and then extracts a new Cantor copy; it does not claim
that any composed map is continuous. Forward: the image under g of the range of a Cantor
antichain is an uncountable Borel set (Lusin–Souslin, through the measurable embedding) of
pairwise inequivalent points, so it contains a Cantor copy
(MeasurableSet.exists_nat_bool_injection_of_not_countable), which is a Cantor antichain in the
image class. Backward: the preimage under g of the range of a Cantor antichain in the image
is an uncountable Borel subset of A of pairwise inequivalent points, and the same extraction
applies.
Ambient Polish spaces and Cantor-copy extraction are used; no logic topology and no spectrum characterization.
Antichain transport, both directions. For a measurable embedding preserving and reflecting the relations, the image class carries a Cantor antichain iff the class does.
Thinness transport. In Polish ambient spaces perfect and Cantor antichains coincide, so the image class is thin iff the class is.