🔢 mathematics

From an Elementary Proof of Error Representation for Hermite Quadrature to a Rediscovery of Legendre Polynomials and Rodrigues Formula

This paper presents an elementary derivation of an exact error representation for generalized Hermite quadrature using reverse integration by parts, which not only requires milder regularity conditions than classical methods but also elegantly reveals Legendre polynomials as the error kernels and rediscoveries the Rodrigues formula.

Original authors: Tan Bui-Thanh, Giancarlo Villatoro, C. G. Krishnanunni

Published 2026-02-24

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Tan Bui-Thanh, Giancarlo Villatoro, C. G. Krishnanunni

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 trying to measure the area of a weirdly shaped lake. You can't just use a ruler; you need to estimate the total space inside the shoreline. In the world of math, this is called numerical integration.

For a long time, mathematicians had a standard way to do this called the Newton-Cotes formulas (like the Trapezoidal Rule). Think of this like trying to guess the area of the lake by stretching a straight rope between two points on the shore and assuming the water is flat underneath. It's a quick guess, but if the lake has deep valleys or high hills, you'll be way off.

This paper introduces a smarter, more "informed" way to guess the area, called Hermite Quadrature. But the real magic of this paper isn't just the new method; it's how the authors proved it works and what they accidentally discovered along the way.

Here is the story of the paper, broken down into simple concepts:

1. The "Super-Map" vs. The "Rough Sketch"

The Old Way (Newton-Cotes): Imagine you only know the water level at the very start and very end of the lake. You draw a straight line between them. This is a "rough sketch." It's easy, but if the lake is bumpy, your sketch is wrong.
The New Way (Hermite Quadrature): Imagine you have a drone that doesn't just tell you the water level at the start and end, but also tells you how steep the slope is (the derivative) at those points.
- Analogy: Instead of just knowing "the water is 5 feet deep here," you know "the water is 5 feet deep here, and it's rising at a rate of 2 feet per second."
- With this extra info, you can draw a smooth, curvy line that hugs the actual shape of the lake much better. This gives you a much more accurate area calculation.

2. The "Reverse Flashlight" Trick (The Main Discovery)

The authors wanted to prove exactly how much error remains when using this new "Super-Map" method.

Usually, to prove how accurate a math formula is, you need to know a lot about the function's "smoothness." The old way required knowing the 20th derivative (a very high-level, complex measure of smoothness) to prove the error. It's like needing a microscope to see a grain of sand.

The authors used a clever trick they call "Reverse Integration by Parts."

Analogy: Imagine you are walking backward through a dark forest. Instead of shining a flashlight forward to see the path, you shine it backward to see where you've already been. By peeling back the layers of the math problem step-by-step, they found a way to calculate the error using only the 10th derivative (or $n$ -th derivative).
Why this matters: This is a huge deal. It means their formula works for "rougher" functions—functions that are a bit jagged or bumpy—where the old, stricter formulas would fail completely. They made the math accessible to a much wider variety of real-world problems.

3. The "Ghost" in the Machine (Rediscovering Legendre Polynomials)

Here is the most surprising part of the paper. As the authors were doing their "Reverse Flashlight" trick, they kept finding a specific mathematical shape popping up in their error calculations.

They realized this shape was actually a famous family of curves called Legendre Polynomials.

Analogy: Imagine you are trying to fix a leaky boat. As you patch the hole, you accidentally discover a hidden treasure map inside the wood of the boat. You didn't go looking for the map; you found it while doing your job.
The Connection: The authors found that these Legendre Polynomials are the "Error Kernels." This means they are the mathematical fingerprints of the mistake. If you know the shape of the Legendre Polynomial, you know exactly how your approximation is missing the mark.

4. The "Magic Recipe" (Rodrigues Formula)

Because they found these Legendre Polynomials hiding in the error, they also rediscovered a famous "recipe" for making them, called the Rodrigues Formula.

Analogy: It's like inventing a new way to bake a cake, and in the process, you accidentally figure out the original, secret recipe that the master bakers have been using for centuries.
The paper shows that this famous formula isn't just a random rule; it naturally emerges when you try to calculate the error of this specific type of area measurement.

5. The "Patchwork Quilt" (Composite Rule)

Finally, the paper explains how to use this method for a massive lake that is too big to measure all at once.

Analogy: Instead of trying to measure the whole lake in one go, you cut it into small, manageable patches (like a quilt). You use your "Super-Map" method on each small patch and stitch the results together. This is called a Composite Hermite Quadrature Rule, and it makes the method practical for computers to use on huge, complex data sets.

Summary: Why Should You Care?

This paper is a celebration of simplicity and connection.

Simplicity: They took a complex problem (calculating error) and solved it with a basic, elementary tool (reverse integration) that anyone with high school calculus could understand, rather than needing advanced, obscure machinery.
Robustness: Their method works on "rougher" data than before, making it more useful for real-world engineering and science.
Connection: They showed that two seemingly different worlds of math—measuring areas (Quadrature) and special curves (Legendre Polynomials)—are actually deeply linked. The error in one is the shape of the other.

In short, the authors didn't just build a better ruler; they discovered that the ruler itself is made of a famous, beautiful mathematical pattern that had been hiding in plain sight all along.

1. Problem Statement

The paper addresses the numerical integration (quadrature) of functions where not only the function values but also the first $n-1$ derivatives are available at the endpoints of an interval $[a, b]$ .

Context: Standard Newton–Cotes formulas (like Trapezoidal or Simpson's rules) use only function values. Hermite quadrature extends this by utilizing derivative information to construct a higher-order interpolating polynomial ( $H_n$ ) and integrating it.
The Gap: Classical error analysis for Hermite quadrature relies on the Peano kernel theorem or standard interpolation error formulas. These classical approaches require the existence of the $2n$ -th derivative ( $f^{(2n)}$ ) of the integrand to provide an exact error representation.
The Challenge: Many practical functions possess only $n$ $n$ continuous derivatives (or $f^{(n)} \in L^2$ $f^{(n)} \in L^{2}$ ) but lack the higher-order regularity ( $2n$ $2 n$ ) required by classical methods. The authors seek an elementary derivation of the error representation that:
1. Requires only the existence of the $n$ -th derivative.
2. Provides an exact error formula without relying on complex theorems like Peano's.
3. Reveals deeper connections to classical orthogonal polynomials.

2. Methodology

The authors employ a novel, elementary approach based on Reverse Integration by Parts (RIBP) and coefficient matching.

Step 1: Hermite Interpolation Formulation:
They define the two-point Hermite interpolating polynomial $H_n(f; x)$ using values $f^{(k)}(a)$ and $f^{(k)}(b)$ for $k=0, \dots, n-1$ . The quadrature rule approximates $\int_a^b f(x) dx$ by $\int_a^b H_n(f; x) dx$ .
Step 2: Reverse Integration by Parts (RIBP):
Instead of standard integration by parts, they apply RIBP $n$ times to the integral $\int_a^b f(x) dx$ . This yields an expansion involving boundary terms (derivatives at $a$ and $b$ ) and a residual integral involving the $n$ -th derivative:
$\int_a^b f(x) dx = \sum \text{Boundary Terms} + \int_a^b (-1)^n f^{(n)}(x) K_n(x, \theta) dx$
Here, $K_n(x, \theta)$ is a polynomial kernel containing free parameters $\theta = \{c, \delta_0, \dots, \delta_{n-2}\}$ .
Step 3: Coefficient Matching:
The authors explicitly calculate the weights of the Hermite quadrature rule (derived from integrating $H_n$ ) and match them against the boundary terms generated by the RIBP expansion.
- By solving a system of equations, they uniquely determine the free parameters $\theta$ such that the boundary terms in the RIBP expansion exactly match the Hermite quadrature weights.
- This process eliminates the boundary terms, leaving the error $E_n$ as the residual integral:
  $E_n = \int_a^b (-1)^n f^{(n)}(x) K_n(x, \theta^*) dx$
Step 4: Analysis of the Kernel:
With the parameters fixed, the authors analyze the properties of the resulting kernel $K_n(x, \theta^*)$ . They utilize orthogonality conditions and polynomial properties to identify the kernel's structure.

3. Key Contributions

A. Elementary Error Representation with Milder Regularity

The primary contribution is an exact error formula for Hermite quadrature that only requires the existence of the $n$ -th derivative ( $f^{(n)}$ ), whereas classical methods require $f^{(2n)}$ .

Significance: This extends the validity of Hermite quadrature error bounds to a broader class of functions (e.g., functions with singularities in higher derivatives or those only in Sobolev spaces $W^{n, \infty}$ ).
Derivation: The proof relies solely on elementary calculus (integration by parts and the General Leibniz Rule), avoiding the heavy machinery of the Peano kernel theorem for the initial derivation.

B. Rediscovery of Legendre Polynomials

A surprising by-product of the coefficient matching is the identification of the error kernel.

Result: The kernel $K_n(x, \theta^*)$ is proven to be the shifted and unnormalized Legendre polynomial of degree $n$ on the interval $[a, b]$ .
Implication: This establishes an elegant, previously unexplored relationship: Legendre polynomials are precisely the error kernels for interpolatory Hermite quadrature rules.

C. Rediscovery of the Rodrigues Formula

By integrating the identified kernel $K_n$ $n$ times and utilizing the classical error form (where $f^{(2n)}$ exists), the authors derive the Rodrigues formula for Legendre polynomials as a natural consequence of their error analysis:
$K_n(x) = \frac{1}{(2n)!} \frac{d^n}{dx^n} \left[ (x-a)^n (x-b)^n \right]$
This provides a new, intuitive pathway to deriving this fundamental formula in approximation theory.

D. New Error Bounds

The paper provides explicit error bounds for the general $n$ case:

Uniform Norm: Bounds based on the deviation of $f^{(n)}$ from its midrange.
$L^2$ Norm: Bounds based on the deviation of $f^{(n)}$ from its mean, utilizing the $L^2$ norm of the Legendre kernel.
Higher Derivatives: If $f^{(n+k)}$ exists, tighter bounds are derived using polynomial approximation theory (Chebyshev interpolation).

E. Alternative Proofs and Extensions

Peano Kernel Verification: The authors provide a secondary proof using the Peano kernel theorem (requiring $f^{(2n)}$ ) to verify that their elementary RIBP approach yields the same result when higher regularity is present.
Composite Rule: They extend the single-interval rule to a Composite Hermite Quadrature Rule for practical applications over multiple subintervals.

4. Key Results

Exact Error Formula:
$E_n = \int_a^b (-1)^n f^{(n)}(x) P_n^*(x) dx$
where $P_n^*(x)$ is the shifted Legendre polynomial.
Parameter Uniqueness: The system of equations used to match the RIBP coefficients to the Hermite weights has a unique solution, which can be solved recursively.
Kernel Properties: The kernel $K_n$ satisfies orthogonality conditions against all polynomials of degree $< n$ and is symmetric/antisymmetric about the interval midpoint depending on $n$ .
Numerical Example: The paper demonstrates via $f(x) = x^2 \sin(x)$ that Hermite quadrature significantly outperforms the Trapezoidal rule, and the derived error bounds are tight.

5. Significance

Theoretical Insight: The paper bridges numerical integration and orthogonal polynomial theory, showing that Legendre polynomials arise naturally from the error structure of Hermite interpolation, not just from orthogonal projection.
Practical Utility: By lowering the regularity requirement from $2n$ to $n$ , the method allows for rigorous error estimation in scenarios where classical methods fail (e.g., functions with limited smoothness).
Pedagogical Value: The derivation is "elementary," making the complex relationship between Hermite quadrature and Legendre polynomials accessible without advanced functional analysis.
Unified Framework: It unifies the derivation of the Rodrigues formula, the properties of Legendre polynomials, and the error analysis of Hermite quadrature into a single coherent framework.

In summary, this work transforms the error analysis of Hermite quadrature from a high-regularity requirement problem into a more accessible, elementary derivation that simultaneously uncovers deep connections to classical orthogonal polynomials.