[Paper Review] On Montel's theorem in several variables

J. M. Almira, Kh. F. Abu-Helaiel|arXiv (Cornell University)|Oct 12, 2013

Functional Equations Stability Results7 references7 citations

TL;DR

This paper extends Montel's theorem to several complex and real variables using invariant subspaces of translation operators. It proves that continuous or distributional solutions to higher-order difference equations Δₕ^{m+1}f = 0 with independent periods must be polynomials, generalizing classical results to d-dimensional settings and establishing a version for measurable functions that fails due to pathological counterexamples.

ABSTRACT

Recently, the first author of this paper, used the structure of finite dimensional translation invariant subspaces of C(R,C) to give a new proof of classical Montel's theorem, about continuous solutions of Fréchet's functional equation $Δ_h^mf=0$, for real functions (and complex functions) of one real variable. In this paper we use similar ideas to prove a Montel's type theorem for the case of complex valued functions defined over the discrete group Z^d. Furthermore, we also state and demonstrate an improved version of Montel's Theorem for complex functions of several real variables and complex functions of several complex variables.

Motivation & Objective

• To generalize Montel’s classical theorem on polynomial solutions of Fréchet’s functional equation to functions of several real and complex variables.
• To establish a version of Montel’s theorem for distributions on ℝᵈ using translation-invariant finite-dimensional subspaces.
• To investigate whether Montel’s theorem holds for measurable functions, leading to a negative result with a constructed counterexample.
• To unify the functional equation approach with the theory of invariant subspaces in harmonic analysis and functional equations.
• To extend earlier results on one-variable solutions to higher-dimensional settings using exponential monomial bases and density arguments.

Proposed method

• Uses the structure of finite-dimensional translation-invariant subspaces in C(ℝᵈ, ℂ) and spaces of distributions to characterize solutions of Δₕ^{m+1}f = 0.
• Applies Anselone-Korevaar’s characterization of translation-invariant subspaces via exponential monomials x^α e^{⟨x,λ⟩} to build a basis for solution spaces.
• Employs the higher-order difference operator Δₕ^{m+1}f(x) = ∑_{k=0}^{m+1} (-1)^{m+1-k} C(m+1,k) f(x + kh) to define the functional equation.
• Leverages density of the group h₁ℤ + ⋯ + h_ℓℤ in ℝᵈ to infer invariance under dense translation sets and hence full translation invariance.
• Uses Djoković’s identity to relate mixed difference operators Δ_{h₁⋯h_{m+1}}f to iterated differences Δₕ^{m+1}f.
• Constructs a counterexample using a measurable, non-polynomial function on ℝ² with Δₕ₁ⁿf = Δₕ₂ⁿf = 0 for n ≥ 2, showing failure of Montel’s theorem in the measurable category.

Experimental results

Research questions

• RQ1 Can Montel’s theorem on polynomial solutions of Δₕ^{m+1}f = 0 be extended to functions of several real variables?
• RQ2 Does the same conclusion hold for complex-valued functions on ℂᵈ, and can it be generalized to distributions?
• RQ3 Is Montel’s theorem valid for measurable functions on ℝᵈ, or do pathological solutions exist?
• RQ4 What is the structure of finite-dimensional translation-invariant subspaces of C(ℝᵈ, ℂ) or spaces of distributions under higher-order difference operators?
• RQ5 Can the basis of solutions be explicitly described using exponential monomials when the translation group is dense?

Key findings

• If f ∈ C(ℝᵈ, ℂ) satisfies Δₕₖ^{m+1}f = 0 for ℓ ≥ d independent vectors h₁,…,h_ℓ with h₁ℤ + ⋯ + h_ℓℤ dense in ℝᵈ, then f is a polynomial of degree ≤ m.
• For distributions f ∈ X_d (C(ℝᵈ, ℂ) or distributions on ℝᵈ), the same condition implies f is a polynomial of degree ≤ m.
• The result holds with f(x) = ∑_{|α|<N} a_α x^α for some N ∈ ℕ and complex coefficients a_α.
• A counterexample shows Montel’s theorem fails for measurable functions: a measurable f on ℝ² with Δₕ₁ⁿf = Δₕ₂ⁿf = 0 for n ≥ 2, but f is not a polynomial.
• The failure arises because the span of f is not invariant under all translations, even though it is invariant under a dense subgroup.
• The paper confirms that regularity (continuity or distributional) is essential—measurability alone is insufficient to force polynomial solutions.