Bounded difference quotients give weak derivatives #
For a locally integrable u : E → ℝ on an open set Ω, a direction v and a real t ≠ 0, the
difference quotient of u is
t⁻¹ * (u (x + t • v) - u x).
This file proves the L² form of the difference-quotient criterion for weak differentiability:
if the difference quotients of u in the direction v are bounded by C in L²(K) for
arbitrarily small t, on every compact K ⊆ Ω, then u has a weak derivative in the
direction v which lies in L²(Ω) with norm at most C. This is the step of the
difference-quotient method that turns the uniform L² bounds on difference quotients of a weak
solution into membership of its derivatives in L², as in the proof of interior H² regularity
for elliptic equations.
The criterion combines two results of independent use.
- The difference quotients of
uconverge to its distributional derivative: paired with a test functionφ, they tend to-∫ ∂_v φ * uast → 0. - A distributional derivative that is bounded against the
L²norm of test functions is anL²function: this isTauCeti.exists_norm_le_hasWeakLineDerivOn_of_abs_integral_lineDeriv_mul_le, proved inTauCeti.Analysis.Sobolev.TestFunctionLp.
The criterion is stated for the exponent 2, where every bounded functional on L²(Ω) is
represented by an element of L²(Ω).
Main declarations #
TauCeti.integral_inv_mul_sub_mul_tendsto_neg_integral_lineDeriv_mul: the difference quotients of a locally integrable function converge to its distributional derivative when paired with a test function.TauCeti.exists_norm_le_hasWeakLineDerivOn_of_frequently_eLpNorm_inv_mul_sub_le: the difference-quotient criterion.TauCeti.exists_integrable_eqOn_tsupport_add: a locally integrable function agrees with a globally integrable one on the support of a test function and on its small translates.
References #
- L. C. Evans, Partial Differential Equations, §5.8.2, Theorem 3.
- D. Gilbarg, N. S. Trudinger, Elliptic Partial Differential Equations of Second Order, Lemma 7.24.
Localization of a locally integrable function at a test function. A function locally
integrable on Ω agrees, on the support of a test function φ and on all small translates of
that support in a direction v, with a function that is integrable on all of E: its truncation
to a compact neighbourhood of tsupport φ inside Ω.
This replaces local integrability by genuine integrability in any argument that only sees u
through φ and its small translates, so that the global integral theorems apply.
Difference quotients converge to the distributional derivative. If u is locally
integrable on Ω and φ is a test function on Ω, then as t → 0 with t ≠ 0,
∫ t⁻¹ * (u (x + t • v) - u x) * φ x → -∫ ∂_v φ * u.
No differentiability of u is assumed: when u has a weak derivative u' in the direction v,
the limit is ∫ φ * u'.
The difference-quotient criterion for weak differentiability. Let u be locally integrable
on Ω and C ≥ 0. Suppose that on every compact K ⊆ Ω the difference quotients of u in the
direction v satisfy
‖t⁻¹ * (u (· + t • v) - u)‖_{L²(K)} ≤ C
for arbitrarily small t ≠ 0. Then u has a weak derivative in the direction v on Ω that
lies in L²(Ω) and has norm at most C.
The hypothesis only asks for the bound frequently as t → 0; the usual hypothesis that it holds
for all sufficiently small t ≠ 0 implies it by Filter.Eventually.frequently.