Differentiability of superposition operators on bounded continuous functions #
Let G : X → Y be a Lipschitz map between normed spaces over ℝ or ℂ. Postcomposition with
G is the superposition (or Nemytskii) operator f ↦ G ∘ f on bounded continuous functions,
BoundedContinuousFunction.comp G hG : (α →ᵇ X) → (α →ᵇ Y), for the sup norm. This file shows
that it is continuously differentiable wherever G is, in the following uniform sense.
Suppose that G has derivative G' x at every point x of a set s, that
G' : X → (X →L[𝕜] Y) is uniformly continuous on s, and that the values of f₀ stay a fixed
distance δ > 0 inside s. Nothing is assumed about G' off s. Then the superposition
operator is differentiable at f₀, and its derivative is pointwise application of the family of
derivatives, h ↦ (t ↦ G' (f₀ t) (h t)), that is applyCLM Φ for the bounded continuous family
Φ t = G' (f₀ t); the Lipschitz constant of G bounds G' on s, so this family is bounded.
The derivative is moreover strict, and the operator is C¹ near f₀.
Uniform continuity of G' is what makes the remainder estimate hold with one constant at all
the points f₀ t at once; nothing makes the range of f₀ compact, for instance for curves on
[0, ∞). The distance δ allows G to be differentiable only on part of its domain, as when a
smooth map is cut off outside a ball by a merely Lipschitz retraction.
This is the differentiability input to the Lyapunov--Perron construction of stable manifolds.
Its integral operator acts on bounded continuous curves through the superposition by the
nonlinearity of the differential equation, and the implicit function theorem turns a C¹
operator into C¹ dependence of its fixed points on parameters.
Main declarations #
BoundedContinuousFunction.exists_eq_comp: along a function with values ins, the derivativest ↦ G' (f t)form a bounded continuous family.BoundedContinuousFunction.hasFDerivAt_comp: the superposition operator is differentiable atf₀, with derivativeh ↦ (t ↦ G' (f₀ t) (h t)).BoundedContinuousFunction.hasStrictFDerivAt_comp: the derivative is strict.BoundedContinuousFunction.contDiffAt_comp: the superposition operator isC¹nearf₀.BoundedContinuousFunction.contDiff_comp: whenGis differentiable everywhere with a derivative uniformly continuous on bounded sets, the superposition operator isC¹.
References #
- J. Appell and P. P. Zabrejko, Nonlinear Superposition Operators, Cambridge Tracts in Mathematics 95, Cambridge University Press, 1990.
Along a function f with values in s, the derivatives t ↦ G' (f t) form a bounded
continuous family: they are bounded by the Lipschitz constant of G. This supplies the family Φ
in BoundedContinuousFunction.hasFDerivAt_comp and
BoundedContinuousFunction.hasStrictFDerivAt_comp.
The derivative of a superposition operator. Let G be Lipschitz, with derivative G' x
at every point x of s, where G' is uniformly continuous on s. If the values of f₀ stay a
distance δ > 0 inside s, then f ↦ G ∘ f is differentiable at f₀ for the sup norm, with
derivative h ↦ (t ↦ G' (f₀ t) (h t)), that is applyCLM Φ for the bounded continuous family
Φ t = G' (f₀ t).
Strict differentiability of a superposition operator. Under the hypotheses of
BoundedContinuousFunction.hasFDerivAt_comp, the derivative
h ↦ (t ↦ G' (f₀ t) (h t)) of f ↦ G ∘ f at f₀ is strict.
A superposition operator is C¹. Let G be Lipschitz, with derivative G' x at every
point x of s, where G' is uniformly continuous on s. If the values of f₀ stay a distance
δ > 0 inside s, then the map f ↦ G ∘ f is continuously differentiable at f₀.
A superposition operator is C¹, global form. If G is Lipschitz and differentiable
everywhere, with a derivative G' that is uniformly continuous on every ball about 0, then
f ↦ G ∘ f is continuously differentiable on the bounded continuous functions. Since each
bounded continuous f₀ takes values in a ball, uniform continuity of G' on bounded sets
suffices; it holds for instance when X is finite-dimensional and G' is continuous.