[Paper Review] Difference Problems and Differential Problems

Wolfgang Bertram|ArXiv.org|Dec 3, 2007

Mathematics and Applications9 references3 citations

TL;DR

This paper investigates the foundational relationship between difference calculus and differential calculus, proposing that higher-order difference quotients and their continuous limits (differentials) form a unified framework for understanding calculus. It identifies unresolved problems in explicit formulae, integration, and pointwise differentiability, emphasizing that a deeper grasp of difference calculus is essential for solving the anti-derivative problem in generalized settings.

ABSTRACT

We state some elementary problems concerning the relation between difference calculus and differential calculus, and we try to convince the reader that, in spite of the simplicity of the statements, a solution of these problems would be a significant contribution to the understanding of the foundations of differential and integral calculus.

Motivation & Objective

• To clarify the structural relationship between difference calculus and differential calculus as a continuous limit of discrete quotients.
• To identify and formulate open foundational problems in higher-order difference and differential calculus.
• To explore the possibility of a generalized anti-derivative theory based on difference calculus.
• To establish pointwise notions of differentiability that are stronger than classical Fréchet differentiability and compatible with global C^k theory.
• To investigate integral representations of higher-order difference quotients in the real case, linking them to Peano kernels and multivariate integration.

Proposed method

• Defines the first-order difference quotient map $ f^{]1[}(x,v,t) = \frac{f(x+tv) - f(x)}{t} $ for $ t \in \mathbb{K}^\times $, with domain $ U^{]1[} \subset U \times V \times \mathbb{K}^\times $.
• Introduces the extended difference quotient map $ \Delta^{]1[}f(x,v,t) = \bigl(f(x), f^{]1[}(x,v,t), t\bigr) $, which satisfies functorial properties $ \Delta^{]1[}(g \circ f) = \Delta^{]1[}g \circ \Delta^{]1[}f $.
• Defines the differential as the continuous extension of $ f^{]1[} $ to $ t = 0 $, yielding $ f^{[1]}(x,v,0) = df(x)v $, with $ f $ of class $ \mathcal{C}^1 $ iff this extension exists and is continuous.
• Iterates the construction to define higher-order difference and differential maps $ f^{]k[} $, $ f^{[k]} $, and $ \Delta^{[k]}f $, with increasing complexity in their domain structures.
• Uses density of $ \mathbb{K}^\times $ in $ \mathbb{K} $ (for topological rings) to ensure uniqueness of the continuous extension to $ t = 0 $, thereby deriving the chain rule and linearity from difference calculus.
• Proposes integral representations for $ f^{[k]} $ in the real case via iterated integration of $ d^kf $, linking them to the standard simplex and Peano kernel functions.

Experimental results

Research questions

• RQ1 Can an explicit, closed-form formula be derived for the higher-order difference quotient map $ f^{]k[} $, particularly for $ k \geq 2 $, in terms of iterated function evaluations?
• RQ2 Is there a generalized anti-derivative theory for difference calculus that provides necessary and sufficient conditions for the existence of a function whose difference quotient yields a given map?
• RQ3 Can a pointwise $ \mathcal{C}^k $-class be rigorously defined such that global $ \mathcal{C}^k $-smoothness is equivalent to pointwise $ \mathcal{C}^k $-smoothness at every point?
• RQ4 Can higher-order differential quotients $ f^{[k]} $ be expressed via integral formulas involving the $ k $-th derivative and a kernel function, analogous to the Peano kernel formula for divided differences?
• RQ5 How do the structures of $ f^{[k]} $ and $ \Delta^{[k]}f $, which depend on $ 2^{k+1} - 1 $ variables, relate to combinatorial and geometric structures such as simplices or Carnot groups?

Key findings

• The first-order difference quotient $ f^{]1[} $ admits a continuous extension to $ t = 0 $ if and only if $ f $ is of class $ \mathcal{C}^1 $, and this extension yields the standard differential $ df(x)v $, recovering the classical chain rule via density arguments.
• For $ k = 2 $, an explicit formula for $ f^{]2[} $ is derived involving nested evaluations of $ f $ at points like $ x_1 + t_3x_2 + (t_1 + t_2t_3)(v_1 + t_3v_2) $, showing the complexity of higher-order expressions.
• The extended map $ \Delta^{]k[}f $ has $ 2^{k+1} - 1 $ components, reflecting the combinatorial explosion in the number of variables required to track all intermediate difference quotients.
• In the real case, $ f^{[k]} $ can be expressed as an iterated integral of $ d^kf $ over a $ k $-dimensional simplex, with the final expression involving the Peano kernel $ M_{k+1} $, which is $ \mathcal{C}^{k-2} $ and piecewise polynomial of degree $ \leq k-2 $ on intervals between interpolation points.
• The pointwise $ \mathcal{C}^k $-class concept is defined via extension of $ f^{]k[} $ to a neighborhood of $ t = 0 $ in a set $ U_a^{[k]} $, and this notion is equivalent to global $ \mathcal{C}^k $-smoothness when satisfied at all points.
• The theory generalizes classical calculus: for finite-dimensional real vector spaces, the $ \mathcal{C}^1 $-class defined here coincides with the standard $ \mathcal{C}^1 $-class, and strict differentiability at a point is equivalent to the pointwise $ \mathcal{C}^1 $-class.