Approximating a lower semicontinuous function from below #
On a pseudometric space, an ℝ≥0∞-valued lower semicontinuous function is the pointwise supremum
of an increasing sequence of bounded Lipschitz functions. This file builds that sequence
explicitly, as the truncated inf-convolution
lscApproxAux f n x = ⨅ y, ((f y ⊓ n).toReal + n * dist x y),
and packages it as a sequence in Ω →ᵇ ℝ≥0.
Monotone approximation from below by bounded continuous functions is what turns the monotone
convergence theorem into a lower semicontinuity statement for integrals against a varying measure,
which is why the target of the approximation is Ω →ᵇ ℝ≥0 rather than merely a continuous
function: the weak topology on measures is defined by pairing with exactly these test functions.
Main declarations #
TauCeti.lscApproxAux: the truncated inf-convolution, a real-valued function bounded bynand Lipschitz with constantn.TauCeti.iSup_ofReal_lscApproxAux: at a point wherefis lower semicontinuous, the approximations increase to the value off.TauCeti.lscApprox: the same sequence bundled as bounded continuousℝ≥0-valued functions.LowerSemicontinuous.exists_boundedContinuous_monotone_iSup_eq: the resulting interface, stating only that some monotone sequence of bounded continuous functions has pointwise supremumf.
Implementation notes #
The inf-convolution is taken of the truncation f ⊓ n rather than of f itself, so that the
n-th approximation is bounded by n; truncating is what allows infinite values of f to be
reached in the limit. The infimum is formed in ℝ (rather than in ℝ≥0∞) because the Lipschitz
estimate is then Mathlib's LipschitzWith.of_le_add_mul, and because bounded continuous functions
are the intended target. The construction is the one used for LipschitzOnWith.extend_real.
The n-th truncated inf-convolution of f : Ω → ℝ≥0∞ on a pseudometric space: the infimum
over y of (f y ⊓ n).toReal + n * dist x y. It is nonnegative, bounded by n, Lipschitz with
constant n, increasing in n, and, when f is lower semicontinuous, increases pointwise to
f.
Instances For
The defining upper bound for TauCeti.lscApproxAux, one for each competitor y.
A lower bound valid for every competitor y is a lower bound for
TauCeti.lscApproxAux.
The approximations are nonnegative.
Taking y = x in the defining infimum bounds the approximation by the truncation of f
at x.
The n-th approximation is bounded by n.
Every approximation lies below f.
The n-th approximation is Lipschitz with constant n.
The approximations are continuous.
The approximations increase with n, both the truncation level and the Lipschitz constant
growing.
At a point where f : Ω → ℝ≥0∞ is lower semicontinuous, its truncated inf-convolutions
increase to the value of f.
The truncated inf-convolution of f : Ω → ℝ≥0∞, bundled as a bounded continuous
ℝ≥0-valued function: the nonnegative part of TauCeti.lscApproxAux f n, which is bounded
by n.
Instances For
The value of the bundled approximation, as a nonnegative real. This is deliberately not
@[simp]: the ℝ≥0∞ coercion TauCeti.coe_lscApprox_apply is the simp-normal form, and marking
this lemma as well would pre-simplify that lemma's left-hand side (the simpNF linter rejects
the pair).
The value of the bundled approximation, as an element of ℝ≥0∞.
The bundled approximations increase pointwise with n.
At a point where f is lower semicontinuous, the bundled approximations increase to the
value of f.
On a pseudometric space, a lower semicontinuous f : Ω → ℝ≥0∞ is the pointwise supremum of
an increasing sequence of bounded continuous ℝ≥0-valued functions.