Commit 2021-09-11 15:52 62de5912
View on Github →feat(interval_integral): generalize change of variables (#8869)
- Generalizes
interval_integral.integral_comp_mul_deriv'
. In this version: (1)f
need not be differentiable at the endpoints of[a,b]
, only continuous, (2) I removed themeasurable_at_filter
assumption (3) I assumed thatg
was continuous onf '' [a,b]
, instead of continuous at every point off '' [a,b]
(which differs in the endpoints). This was possible after @sgouezel's PR #7978. The proof was a lot longer/messier than expected. Under these assumptions we have to be careful to sometimes take one-sided derivatives. For example, we cannot take the 2-sided derivative ofλ u, ∫ x in f a..u, g x
whenu
is the maximum/minimum off
on[a, b]
. @urkud: I needed moreFTC_filter
classes, namely for closed intervals (to be precise:FTC_filter x (𝓝[[a, b]] x) (𝓝[[a, b]] x)
). Was there a conscious reason to exclude these classes? (The documentation explicitly enumerates the existing instances.)