Zulip Chat Archive

Stream: mathlib4

Topic: condexp as a map between L¹s


Yury G. Kudryashov (Sep 04 2023 at 08:36):

Do we have conditional expectation as a map between spaces w.r.t. different σ-algebras? Or as a self-map with a property that the result is a.e.-measurable w.r.t. the other σ-algebra?

Rémy Degenne (Sep 04 2023 at 09:43):

There is this: docs#MeasureTheory.condexpL1Clm , with docs#MeasureTheory.aestronglyMeasurable'_condexpL1Clm

Yury G. Kudryashov (Sep 04 2023 at 12:31):

Thank you!

Yury G. Kudryashov (Sep 04 2023 at 19:32):

What are the right assumptions to have this in L^p?

Yury G. Kudryashov (Sep 04 2023 at 19:32):

(p > 1)

Rémy Degenne (Sep 05 2023 at 07:18):

For p = 2 we have docs#MeasureTheory.condexpL2 (we build it first in L2, then extend to L1). We don't have anything already made for other p.

If your measure is finite then for any p > 1 Lp is a subset of L1, which means that the conditional expectation is well defined on Lp. The image of a Lp function is also in Lp by Jensen's inequality for the function ||.||^p (we don't have Jensen's inequality for condexp).

In "Analysis in Banach spaces (volume 1)", lemma 2.6.30, they have it for a sigma-finite measure instead of finite, but their conditional expectation is defined slightly more generally than ours: they use sigma-integrable functions, while ours is meaningful only for integrable functions. I did not add the notion of sigma-integrability to mathlib.

Yury G. Kudryashov (Sep 05 2023 at 12:51):

Thank you for the explanation!


Last updated: Dec 20 2023 at 11:08 UTC