Interpolation inequalities for derivatives #
This file proves the supremum-norm interpolation inequalities that control an intermediate derivative by a small multiple of a higher one plus a large multiple of the function. They are used to absorb lower-order terms in Schauder estimates, elliptic and parabolic.
The basic estimate is a first-order Taylor bound: if g has derivative g' and g' is
α-Hölder with constant H on a segment from x to x + w, then
‖g (x + w) - g x - g' x w‖ ≤ H ‖w‖^(1 + α). Hence
‖g' x w‖ ≤ ‖g (x + w)‖ + ‖g x‖ + H ‖w‖^(1 + α).
Choosing ‖w‖ = ε in every direction gives, on the whole space,
‖D^(k+1) f‖_∞ ≤ 2 ε⁻¹ ‖D^k f‖_∞ + ε^α [D^(k+1) f]_α
for every ε > 0. For α = 1 this is the Landau-type inequality
‖D^(k+1) f‖_∞ ≤ 2 ε⁻¹ ‖D^k f‖_∞ + ε ‖D^(k+2) f‖_∞. Combining the two and absorbing the
supremum of D^(k+2) f, which must be finite a priori, gives
‖D^(k+2) f‖_∞ ≤ 32 ε⁻² ‖D^k f‖_∞ + 2 ε^α [D^(k+2) f]_α.
All constants are absolute: they depend neither on the dimension nor on α.
Interior estimates need local bounds, which use the second derivative. If g'' is α-Hölder with
constant H on s, the second-order Taylor remainder of g along a segment in s is at most
H ‖w‖^(2 + α). In the symmetric difference g (x + w) - g (x - w) the second-order terms cancel,
so at the centre x of a closed ball of radius ε in s, with ‖g‖ ≤ M there,
‖g' x‖ ≤ M / ε + H ε^(1 + α),
and at the centre of a closed ball of radius 2 ε in s, ‖g'' x‖ ≤ 2 M / ε² + 3 H ε^α. Only
the values of g and the Hölder constant of g'' on s enter. For a function which is C² on
an open set containing s, with D²u α-Hölder on s, these are bounds on ‖Du‖ and ‖D²u‖.
On a closed interval [a, b] the same argument applies to one-sided derivatives within
[a, b], for steps h with 2 h ≤ b - a, since from each point of [a, b] a step of length h
in one of the two directions stays in the interval. With temporal Hölder exponent α / 2 and
h = ε², this is the estimate for time derivatives on a slab used in parabolic Hölder spaces.
For a function u (x, t) of space and time, the same Taylor bound at the spatial scale
h = √|t - s|, applied to the difference u (·, t) - u (·, s), gives
‖D_x u (x, t) - D_x u (x, s)‖ ≤ (C₂ + Cₜ) |t - s|^((1 + α) / 2),
where C₂ is the α / 2-Hölder constant in time of D_x² u and Cₜ is the α-Hölder constant
in space of ∂ₜ u. The increment of the difference over a spatial step of length h is
controlled by the mean value inequality in time and the spatial Hölder bound on ∂ₜ u, and its
linear approximation error by the difference of the Hessians.
Main declarations #
HolderOnWith.norm_image_sub_sub_le_of_hasFDerivWithinAt: the first-order Taylor bound for a Hölder derivative.HolderOnWith.norm_sub_fderiv_le_of_mem_ball: the Taylor remainder on a ball for the derivative of a differentiable function.HolderOnWith.norm_apply_le_of_hasFDerivWithinAt: the resulting bound on a directional derivative.HolderOnWith.norm_le_of_hasFDerivWithinAt_ball: the local interpolation inequality on a ball, also applicable to differences of two functions.HolderOnWith.norm_fderiv_sub_fderiv_le_of_norm_sub_le_ball: comparison of two derivatives from uniform function error and possibly different Hölder constants.HolderWith.norm_le_of_hasFDerivAt,HolderWith.norm_fderiv_le_of_norm_leandHolderWith.norm_iteratedFDeriv_succ_le_of_norm_le: the interpolation inequality on the whole space.ContDiff.norm_iteratedFDeriv_succ_le_of_norm_iteratedFDeriv_le: the Landau-type inequality.HolderWith.norm_iteratedFDeriv_add_two_le_of_norm_leandHolderWith.norm_iteratedFDeriv_succ_le_of_norm_le_of_bddAbove: the absorbed bounds on the two derivatives aboveD^k f, andHolderWith.norm_iteratedFDeriv_succ_le_and_norm_iteratedFDeriv_add_two_le: both bounds with one constant, for scalesε ≤ 1, andHolderWith.norm_fderiv_le_and_norm_iteratedFDeriv_two_le: its specialisation tok = 0, stated for a function bounded in norm.HolderOnWith.norm_image_sub_sub_sub_le_of_hasFDerivWithinAt: the second-order Taylor bound for a Hölder second derivative.HolderOnWith.norm_le_of_hasFDerivWithinAt_of_closedBall_subsetandHolderOnWith.norm_le_of_hasFDerivWithinAt_of_closedBall_two_mul_subset: the local interpolation inequalities for the first and second derivatives on a ball.HolderOnWith.norm_fderiv_le_of_contDiffOnandHolderOnWith.norm_iteratedFDeriv_two_le_of_contDiffOn: the same for aC²function.HolderOnWith.norm_le_of_hasDerivWithinAt_Icc: the interpolation inequality for derivatives within a closed interval.TauCeti.holderOnWith_Icc_of_hasFDerivAt_of_hasDerivWithinAt: the temporal(1 + α) / 2-Hölder continuity of the spatial derivative.
References #
- D. Gilbarg, N. S. Trudinger, Elliptic Partial Differential Equations of Second Order, Section 6.8 (interpolation inequalities).
- A. Lunardi, Analytic Semigroups and Optimal Regularity in Parabolic Problems, Chapter 0.
- O. A. Ladyzhenskaya, V. A. Solonnikov, N. N. Ural'tseva, Linear and Quasi-linear Equations of Parabolic Type, Chapter II (parabolic Hölder spaces).
First-order Taylor bound for a Hölder derivative. If g has derivative g' within s
and g' is α-Hölder with constant H on s, then for a segment from x to x + w contained
in s, the error of the linear approximation is at most H ‖w‖^(1 + α).
Taylor remainder on a ball. A differentiable function whose derivative is (H, α)-Hölder
on a ball satisfies ‖g y - g x - Dg(x)(y - x)‖ ≤ H ‖y - x‖^α ‖y - x‖ at its centre x.
Neither finite dimensionality, completeness nor a positive exponent is required.
Interpolation along a segment. If g has derivative g' within s and g' is
α-Hölder with constant H on s, then for a segment from x to x + w contained in s,
the derivative in the direction w is bounded by the values of g at the endpoints and the
Hölder constant.
Local interpolation inequality. On a ball of radius ε, a function bounded by M
whose derivative is (H, α)-Hölder has derivative norm at the centre at most
4 M / ε + H ε^α. Applying the estimate to a difference compares derivatives using only
uniform function errors and Hölder derivative bounds.
Derivative comparison on a ball. If two differentiable functions differ by at most M
on a ball of radius ε, and their derivatives have Hölder constants H and K with a common
exponent α, their derivatives at the centre differ by at most 4 M / ε + (H + K) ε^α.
Neither completeness nor a positive exponent is required.
Interpolation inequality on an interval. Let f have derivative f' within [a, b],
with f' β-Hölder with constant H on [a, b] and ‖f‖ ≤ M there. Then for every step
h > 0 with 2 h ≤ b - a, the derivative satisfies ‖f' t‖ ≤ 2 M / h + H h^β on [a, b].
With β = α / 2 and h = ε² this reads ‖f'‖ ≤ 2 ε⁻² M + ε^α H.
Second-order Taylor bound for a Hölder second derivative. If g has derivative g'
within s, g' has derivative g'' within s, and g'' is α-Hölder with constant H on
s, then for a segment from x to x + w contained in s, the error of the second-order Taylor
polynomial is at most H ‖w‖^(2 + α).
Interpolation of the first derivative on a ball. If g has derivative g' within s,
g' has derivative g'' within s, g'' is α-Hölder with constant H on s and ‖g‖ ≤ M
on s, then ‖g' x‖ ≤ M / ε + H ε^(1 + α) at every centre x of a closed ball of radius
ε > 0 contained in s. The symmetric difference g (x + w) - g (x - w) approximates 2 g' x w
up to 2 H ‖w‖^(2 + α), since the second-order terms cancel.
Interpolation of the second derivative on a ball. If g has derivative g' within s,
g' has derivative g'' within s, g'' is α-Hölder with constant H on s and ‖g‖ ≤ M
on s, then ‖g'' x‖ ≤ 2 M / ε² + 3 H ε^α at every centre x of a closed ball of radius
2 ε > 0 contained in s. Only g and the Hölder constant of g'' enter, so this is the local
form of the interpolation inequality used to absorb lower-order terms in interior estimates.
Interpolation of the gradient of a C² function on a ball. Let u be C² on an open set
U ⊇ s, with D²u α-Hölder with constant H on s and ‖u‖ ≤ M on s. Then
‖Du(x)‖ ≤ M / ε + H ε^(1 + α) at every centre x of a closed ball of radius ε > 0 in s.
Interpolation of the Hessian of a C² function on a ball. Let u be C² on an open set
U ⊇ s, with D²u α-Hölder with constant H on s and ‖u‖ ≤ M on s. Then
‖D²u(x)‖ ≤ 2 M / ε² + 3 H ε^α at every centre x of a closed ball of radius 2 ε > 0 in
s.
Interpolation inequality. If g has derivative g' everywhere, ‖g‖ ≤ M, and g' is
globally α-Hölder with constant H, then ‖g' x‖ ≤ 2 M / ε + H ε^α for every ε > 0. With
α = 1 this is the Landau-type inequality ‖g' x‖ ≤ 2 M / ε + L ε for a derivative that is
Lipschitz with constant L.
Interpolation inequality. If g is differentiable with ‖g‖ ≤ M and its derivative is
globally α-Hölder with constant H, then ‖fderiv ℝ g x‖ ≤ 2 M / ε + H ε^α for every
ε > 0.
Interpolation inequality for iterated derivatives. If ‖D^k f‖ ≤ M and D^(k+1) f is
globally α-Hölder with constant H, then ‖D^(k+1) f x‖ ≤ 2 M / ε + H ε^α for every
ε > 0.
Landau-type interpolation inequality. If f is C^(k+2) with ‖D^k f‖ ≤ M₀ and
‖D^(k+2) f‖ ≤ M₂, then ‖D^(k+1) f x‖ ≤ 2 M₀ / ε + M₂ ε for every ε > 0.
Absorbed interpolation inequality for the top derivative. If f is C^(k+2) with
‖D^k f‖ ≤ M, D^(k+2) f is bounded and globally α-Hölder with constant H, then
‖D^(k+2) f x‖ ≤ 32 M / ε² + 2 H ε^α for every ε > 0.
Absorbed interpolation inequality for the intermediate derivative. If f is C^(k+2)
with ‖D^k f‖ ≤ M, D^(k+2) f is bounded and globally α-Hölder with constant H, then
‖D^(k+1) f x‖ ≤ 34 M / ε + 2 H ε^(1 + α) for every ε > 0.
Absorbed interpolation inequalities with one constant. If f is C^(k+2) with
‖D^k f‖ ≤ M, D^(k+2) f is bounded and globally α-Hölder with constant H, then for every
scale 0 < ε ≤ 1 both ‖D^(k+1) f x‖ and ‖D^(k+2) f x‖ are at most
(34 / ε + 32 / ε²) M + 2 H ε^α.
Absorbed interpolation inequalities for the first two derivatives. If f is C² with
‖f‖ ≤ M, and D²f is bounded and globally α-Hölder with constant H, then for every scale
0 < ε ≤ 1 both ‖Df x‖ and ‖D²f x‖ are at most (34 / ε + 32 / ε²) M + 2 H ε^α.
Temporal Hölder continuity of the spatial derivative. Let u (x, t) be twice
differentiable in x ∈ E and differentiable in t ∈ [a, b] within [a, b], with spatial
derivatives u', u'' and time derivative uₜ. If the Hessian u'' (x, ·) is α / 2-Hölder in
time with constant C₂ and the time derivative uₜ (·, t) is α-Hölder in space with constant
Cₜ, then the gradient u' (x, ·) is (1 + α) / 2-Hölder in time with constant C₂ + Cₜ.
This controls the temporal (1 + α) / 2 seminorm of the spatial gradient, which appears in the
norm of the parabolic Hölder space C^{2+α, 1+α/2}, by the parabolic seminorms of the Hessian
and of the time derivative.