🔢 mathematics

Adaptive integration of 5-convex and 5-concave functions

This paper introduces and investigates an adaptive numerical integration method for 5-convex functions that combines 3-point Gauss and 4-point Lobatto quadrature rules.

Original authors: Szymon Wąsowicz

Published 2026-01-26

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Szymon Wąsowicz

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 exact amount of water in a strangely shaped swimming pool. You can't just use a ruler because the bottom curves up and down in complex ways. Instead, you have to take a guess based on a few sample points.

This paper is about a new, smarter way to take those guesses to measure the area under a curve (which is the mathematical equivalent of that pool's volume). The author, Szymon Wąsowicz, has created a "super-estimator" specifically for a special type of curve called a 5-convex or 5-concave function.

Here is the breakdown of how it works, using simple analogies:

1. The Problem: Guessing is Hard

In math, when we want to find the area under a curve, we often use standard "rulers" called quadrature rules.

The Old Way (Simpson's Rule): Imagine trying to guess the pool's volume by measuring the depth at just three spots: the left edge, the middle, and the right edge. It's a good guess, but if the pool has weird bumps, you might be off by a lot.
The Adaptive Way: To get a better guess, you chop the pool into smaller and smaller sections and measure each one. But how do you know when to stop chopping? If you chop too little, your answer is wrong. If you chop too much, you waste time.

2. The New Tool: Two Different Rulers

The author introduces two specific "rulers" (mathematical formulas) to measure the curve:

The Gauss Ruler (G): This ruler looks at points inside the interval but ignores the very edges. It's very precise for smooth curves.
The Lobatto Ruler (L): This ruler looks at the edges and points inside. It's also very precise but behaves differently.

For these specific "5-convex" curves (which are like a specific type of smooth, curvy bowl), the author proves a fascinating fact: The true answer always lies somewhere between the result of the Gauss ruler and the Lobatto ruler.

3. The Magic Mix: The "3-to-1" Smoothie

The paper's big idea is to mix these two rulers together to create a new, super-accurate ruler called Q.

The formula is: Q = ¾ (Gauss) + ¼ (Lobatto).
Think of this like making a smoothie. You take three parts of the "Gauss" flavor and one part of the "Lobatto" flavor.
The author proves that for these specific curves, this new "Q" smoothie is incredibly close to the real answer.

4. The "Stop Sign" (The Stopping Criterion)

This is the most practical part of the paper. Usually, to know if your math is good enough, you need to know how "wiggly" the curve is (its sixth derivative). But calculating that wiggliness is often impossible or very hard.

The author found a clever shortcut:

The Rule: You don't need to know the "wiggliness" of the curve. You just need to compare your two rulers (G and L).
The Logic: If the Gauss ruler and the Lobatto ruler give you results that are very close to each other, then your new "Q" ruler is definitely accurate enough.
The Safety Net: The paper proves mathematically that if the difference between G and L is small, the error in your final answer is guaranteed to be even smaller (specifically, 1/4th of that difference).

So, the computer just keeps chopping the pool into smaller pieces until the Gauss and Lobatto rulers agree with each other. Once they agree, the computer hits the "Stop" button, knowing the answer is perfect.

5. The Results: Faster and Smarter

The author tested this method on two famous functions:

The Reciprocal Function (1/x): A curve that gets steep quickly.
The Exponential Function (e^x): A curve that grows very fast.

The Outcome:
When compared to an older method (which used a different mix of rulers for simpler curves), this new method was a massive winner.

To get the same level of precision (up to 16 decimal places!), the new method needed far fewer sub-intervals.
For example, to measure the area under 1/x with extreme precision, the old method needed to chop the pool into 1,572 pieces. The new method only needed 84 pieces.

Summary

Think of this paper as inventing a new, ultra-efficient GPS for measuring curved areas. Instead of driving every single street to check the distance (which takes forever), this new method uses two different map apps (Gauss and Lobatto). As long as the two apps agree on the route, the GPS knows it has found the perfect path and stops driving, saving you a tremendous amount of time and fuel.

Technical Summary: Adaptive Integration of 5-Convex and 5-Concave Functions

Problem Statement
The paper addresses the challenge of adaptive numerical integration for functions $f: [a, b] \to \mathbb{R}$ that belong to the class $C^6[a, b]$ and possess a sixth derivative of constant sign. Such functions are characterized as either 5-convex or 5-concave. While standard adaptive methods often rely on error bounds involving the maximum of higher-order derivatives (e.g., $|f^{(4)}|$ for Simpson's rule), these approaches can be difficult to implement if the derivative bounds are hard to estimate or if the function lacks sufficient regularity. Existing adaptive techniques for lower-order convexity (specifically 3-convex functions) utilize combinations of Simpson and Chebyshev quadratures, but the author seeks a more efficient approach tailored to the higher-order convexity of 5-convex/5-concave functions.

Methodology
The proposed method is built upon a refined inequality relating the exact integral to two specific quadrature rules:

Three-point Gauss quadrature ( $G$ ): A standard Gaussian rule.
Four-point Lobatto quadrature ( $L$ ): A rule utilizing the interval endpoints.

For a 5-convex function $f \in C^6[a, b]$ , the paper utilizes a known inequality established by Bessenyei and Páles, which states that the Gauss quadrature underestimates the integral while the Lobatto quadrature overestimates it:
$G[f; a, b] \leq \int_a^b f(x)dx \leq L[f; a, b]$

The core of the methodology involves a specific linear combination of these two rules:
$Q[f; a, b] = \frac{3}{4}G[f; a, b] + \frac{1}{4}L[f; a, b]$

The author demonstrates that for functions in $C^6[a, b]$ with a constant-sign sixth derivative, the error of this combined quadrature $Q$ is bounded by the difference between the Lobatto and Gauss approximations. Specifically, the paper proves that:
$\left| \int_a^b f(x)dx - Q[f; a, b] \right| \leq \frac{1}{4} \left( L[f; a, b] - G[f; a, b] \right)$

This theoretical result is extended to composite rules ( $G_n, L_n, Q_n$ ) over $n$ subintervals. The adaptive algorithm proceeds by starting with $n=1$ and incrementally increasing the number of subintervals until the stopping criterion is met:
$|L_n[f; a, b] - G_n[f; a, b]| \leq 4\varepsilon$
where $\varepsilon$ is the user-defined tolerance. The existence of such an $n$ is guaranteed by the convergence properties of the composite quadratures.

Key Contributions

Theoretical Justification: The paper provides a rigorous proof (Theorem 3) that the difference between the Lobatto and Gauss composite quadratures serves as a valid, computable upper bound for the error of the specific linear combination $Q_n$ , provided the function is 5-convex or 5-concave.
Refined Stopping Criterion: Unlike methods requiring the estimation of $\sup |f^{(6)}|$ , this approach derives a stopping criterion solely from the computed quadrature values, making it robust for functions where derivative bounds are difficult to determine.
Algorithmic Efficiency: The paper introduces a specific adaptive algorithm (Corollary 4) that leverages the $3/4$ and $1/4$ weighting to achieve high precision with fewer subintervals compared to existing methods designed for lower-order convexity.

Results
Numerical experiments were conducted using SageMath to compare the proposed method against an analogous adaptive method for 3-convex functions (based on Simpson and Chebyshev quadratures) described in prior work [12].

Experiment 1 (Reciprocal Function): For $f(x) = 1/x$ on $[1, 2]$ , the proposed method required significantly fewer subdivisions to achieve accuracies up to $10^{-16}$ . For instance, at $\varepsilon = 10^{-16}$ , the proposed method used 84 subdivisions, whereas the 3-convex method required 1,572.
Experiment 2 (Exponential Function): For $f(x) = e^x$ on intervals $[0, b]$ for $b \in \{1, \dots, 10\}$ with $\varepsilon = 10^{-8}$ , the proposed method consistently required fewer subdivisions. For the interval $[0, 10]$ , the proposed method used 93 subdivisions compared to 1,244 for the 3-convex method.

Significance and Claims
The paper claims that the proposed adaptive integration method "significantly outperforms existing adaptive techniques designed for lower-order convexity." The primary advantage highlighted is the substantial reduction in the number of required subintervals to achieve high precision (up to $10^{-16}$ ). The author notes that while the method is specifically designed for 5-convex and 5-concave functions (a subset of functions with constant-sign sixth derivatives), the efficiency gains are evident even when applied to functions that are also 3-convex. The work establishes a rigorous foundation for using the difference between Gauss and Lobatto quadratures as a reliable error estimator for this specific class of functions.