🔢 mathematics

A divided difference identity for a class of multiple integrals

This paper derives an identity linking multiple integrals of Vandermonde polynomials to divided differences, while also establishing that specific sums of their pure and mixed partial derivatives vanish.

Original authors: Michael S. Floater

Published 2026-03-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Michael S. Floater

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a chef trying to bake a very specific, complex cake. The recipe involves mixing ingredients in a specific order, but the math behind how the flavors blend is incredibly complicated. Usually, to figure out the final taste, you'd have to do a massive amount of calculations, integrating (summing up) tiny bits of flavor across the entire cake.

This paper, written by Michael Floater, discovers a magic shortcut. It proves that for a specific type of mathematical "cake" (involving something called a Vandermonde polynomial), you don't need to do all that heavy lifting. Instead, the final result is simply related to a concept called a divided difference, which is like a "rate of change" or a "slope" between specific points.

Here is the breakdown of the paper's journey, using everyday analogies:

1. The Setting: A Staircase of Intervals

The paper starts by defining a specific shape called a "rectangle," but not just any rectangle. Imagine a staircase.

You have a list of numbers: $x_1, x_2, \dots, x_{n+1}$ .
The "rectangle" is actually a multi-dimensional box where the first variable goes from $x_1$ to $x_2$ , the second from $x_2$ to $x_3$ , and so on.
Think of this as a conveyor belt where each step depends on the previous one. You can't step on the 3rd step unless you've already passed the 2nd.

2. The Star Ingredient: The Vandermonde Polynomial

Inside this box, there is a special function called the Vandermonde polynomial.

The Analogy: Imagine a group of people standing in a line. The Vandermonde polynomial measures the "total distance" between every single pair of people. If two people stand on top of each other (same number), the distance is zero, and the whole measurement collapses.
In the paper, this polynomial acts like a filter or a weight. It gives more importance to certain parts of the calculation and less to others, based on how spread out the numbers are.

3. The Big Discovery: The Shortcut

The main goal of the paper is to calculate the total volume under a curve inside this box. Usually, this requires a massive, multi-step integration (summing up infinite tiny slices).

The Theorem says:
Instead of doing the hard work of summing everything up, you can just look at the divided difference of the function at the edges of your staircase.

The Metaphor: Imagine you want to know the total change in temperature across a long, winding road.
- The Hard Way: Measure the temperature at every single inch of the road, add them all up, and do the math.
- The Paper's Way: The paper proves that for this specific type of road and temperature function, you only need to look at the temperature at the start, the middle, and the end points, and calculate a specific "average slope" between them. The messy middle part cancels itself out perfectly.

4. The Secret Sauce: Why Does It Work?

How did the author prove this? He looked at the derivatives (rates of change) of that Vandermonde polynomial.

The "Ghost" Effect: He discovered that if you add up all the ways this polynomial changes (its derivatives), they cancel each other out to zero.
The Analogy: Imagine a tug-of-war team where every person pulling to the left is perfectly balanced by a person pulling to the right. The net movement is zero.
Because these "tug-of-war" forces cancel out, the complex integral simplifies dramatically. The "noise" of the middle disappears, leaving only the clean signal from the boundaries.

5. The Result: A New Formula

The paper concludes with a beautiful formula:
$\text{The Big Integral} = (\text{Vandermonde of the inputs}) \times (\text{Divided Difference of the function})$

This is useful because:

Divided differences are easy to calculate (they are just basic algebra).
Integrals are hard to calculate (they often require computers or advanced calculus).
This formula turns a hard calculus problem into an easy algebra problem.

Summary for the General Audience

Think of this paper as finding a secret tunnel through a mountain.

The Mountain: A complex, multi-dimensional math problem involving sums and products of numbers.
The Tunnel: A mathematical identity that shows you don't need to climb the whole mountain. You can just look at the entrance and the exit, and the math guarantees the path inside is perfectly straight.
The Use: This helps mathematicians and computer scientists calculate things faster and more accurately, especially when dealing with exponential matrices (which are used in physics, engineering, and computer graphics).

In short, Floater found that a very messy, high-dimensional sum of changes is actually just a simple, elegant relationship between the start and end points of a sequence. It's a reminder that even in the most complex mathematical landscapes, there is often a simple, hidden symmetry waiting to be found.

1. Problem Statement

The paper addresses the evaluation of a specific class of multiple integrals over a sequential rectangular domain $R(x) \subset \mathbb{R}^n$ . The integrand involves the Vandermonde polynomial $V(t)$ multiplied by the $n$ -th derivative of a function $f$ composed with the sum of the variables.

Specifically, given an increasing sequence of real numbers $x = (x_1, \dots, x_{n+1})$ , the domain is defined as:
$R(x) := [x_1, x_2] \times [x_2, x_3] \times \dots \times [x_n, x_{n+1}]$
The goal is to evaluate the integral:
$I = \int_{R(x)} V(t) f^{(n)}(t_1 + t_2 + \dots + t_n) \, dt$
where $V(t) = \prod_{1 \le i < j \le n} (t_j - t_i)$ .

The author aims to derive a closed-form identity relating this integral to divided differences of the function $f$ , providing an alternative to existing formulas like the Genocchi-Hermite formula or the B-spline integral formula.

2. Methodology

The proof strategy relies on transforming the integral of a product into a boundary evaluation using the Fundamental Theorem of Calculus in multiple dimensions. The methodology proceeds through the following logical steps:

A. Analysis of Vandermonde Derivatives

A crucial part of the derivation involves analyzing the partial derivatives of the Vandermonde polynomial $V(t)$ . The author establishes two key properties regarding the sums of these derivatives:

Pure Derivatives: The sum of pure $k$ -th order partial derivatives with respect to any single variable is zero.
Mixed Derivatives: The sum of mixed partial derivatives of any fixed order $k$ $k$ (denoted by the operator $E_k$ $E_{k}$ ) is zero.
- Key Lemma: $E_k V(t) = 0$ for all $k = 1, \dots, n$ .
- Proof Technique: This is proven by relating the differential operators to elementary symmetric polynomials and power sums, utilizing Newton's identities. Since the power sum operators annihilate $V(t)$ , the symmetric operators do as well.

B. The Product Rule for Mixed Derivatives

The author defines a sum function $s(t) = \sum t_i$ and considers the product $\phi(t) = V(t)f(s(t))$ .
Using the vanishing property of the mixed derivatives of $V(t)$ , the author proves (Lemma 7) that the highest-order mixed derivative of the product simplifies significantly:
$\frac{\partial^n}{\partial t_1 \dots \partial t_n} \left( V(t) f(s(t)) \right) = V(t) f^{(n)}(s(t))$
This identity allows the original integrand to be replaced by a total mixed derivative.

C. Integration over Sequential Rectangles

The paper applies a generalized version of the Fundamental Theorem of Calculus (Lemma 8) to integrate the mixed derivative over the rectangle $R(x)$ .

Normally, the integral of a mixed derivative over a hyper-rectangle equals a sum over all $2^n$ vertices.
However, due to the specific geometry of $R(x)$ (where adjacent intervals share endpoints) and the property that $V(t)$ vanishes whenever $t_i = t_{i+1}$ , the sum collapses.
Lemma 9 shows that for functions satisfying the zero property on the boundaries $t_i = t_{i+1}$ , the sum reduces from $2^n$ terms to exactly $n+1$ terms corresponding to specific vertices $v_i$ .

3. Key Contributions and Results

Main Theorem (Theorem 1)

The central result establishes the following identity for $f \in C^n[y_1, y_{n+1}]$ :
$\int_{R(x)} V(t) f^{(n)}(t_1 + \dots + t_n) \, dt = V(x) [y_1, y_2, \dots, y_{n+1}]f$
Where:

$[y_1, \dots, y_{n+1}]f$ is the divided difference of $f$ at the points $y_1, \dots, y_{n+1}$ .
The points $y_i$ are defined as the sum of all $x_j$ excluding $x_{n+2-i}$ :
$y_i = \sum_{j=1}^{n+1} x_j - x_{n+2-i}$
$V(x)$ is the Vandermonde determinant of the sequence $x$ .

Corollary 1

By setting $f(\alpha) = \alpha^n / n!$ , the author derives a formula for the integral of the Vandermonde polynomial itself:
$\int_{R(x)} V(t) \, dt = \frac{1}{n!} V(x)$

Alternative Viewpoint

The paper demonstrates that the identity can be inverted to provide an integral formula for divided differences:
$[y_1, \dots, y_{n+1}]f = \frac{1}{V(y)} \int_{R(x)} V(t) f^{(n)}(s(t)) \, dt$
where $x$ is recovered from $y$ via a linear transformation involving a specific matrix $M$ .

4. Significance and Applications

Theoretical Insight: The paper provides a novel connection between multivariate integration, Vandermonde determinants, and the theory of divided differences. It offers a distinct integral representation for divided differences that differs from classical results (Genocchi-Hermite, Curry-Schoenberg).
Independent Interest: The proof that sums of mixed partial derivatives of Vandermonde polynomials vanish (Lemma 5) is highlighted as a result of independent mathematical interest, potentially useful in other areas of approximation theory or algebraic combinatorics.
Practical Application: The author notes that this identity was originally motivated by the need to bound the determinant of an exponential matrix. The formula serves as a foundational tool for these bounds, which are detailed in a companion paper [5].
Simplicity: The resulting formula is described as "surprisingly simple," converting a complex $n$ -dimensional integral into a standard divided difference multiplied by a Vandermonde determinant.

Conclusion

Floater's paper successfully derives a compact identity linking a multiple integral of a Vandermonde-weighted derivative to a divided difference. The proof is elegant, relying on the structural properties of the Vandermonde polynomial (specifically the vanishing of its symmetric derivative sums) to reduce a high-dimensional integral to a boundary sum. This result enriches the toolkit of numerical analysis and interpolation theory.