🔢 mathematics

Galerkin Approximation of the Fractional Hardy Constant

This paper establishes sharp estimates for the discrete optimal constant of the fractional Hardy inequality in dimensions $N \geq 1$ and derives convergence rates for its Galerkin approximation using piecewise linear elements on quasi-uniform meshes within bounded, convex, smooth domains containing the origin.

Original authors: Andreea Dima, Liviu I. Ignat

Published 2026-07-30

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Andreea Dima, Liviu I. Ignat

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 "stiffness" of a rubber sheet. In the world of physics and math, there are famous rules, called inequalities, that tell us the absolute minimum amount of energy or tension required to stretch a sheet in a certain way. One of the most famous of these is called Hardy's Inequality. Think of it as a cosmic speed limit or a safety net: it says that no matter how you try to wiggle your function (a mathematical shape representing a physical quantity), you can never get the energy below a specific number without the shape blowing up or breaking. This number is called the "optimal constant." It's the sharpest, tightest possible limit nature allows.

Now, computers can't handle infinite, smooth sheets. They have to chop the world into tiny, flat pieces—like a mosaic made of triangles—to do calculations. This is called discretization. The big question for mathematicians is: "When we chop the smooth world into tiny pixels, how much of that perfect, sharp limit do we lose?" Does the computer's answer stay close to the real truth, or does it drift away? This paper dives into a specific, tricky version of this problem involving fractional calculus. While normal calculus deals with smooth slopes, fractional calculus deals with "in-between" slopes—connections that reach out over distances, like a spiderweb where every point feels the pull of every other point, not just its immediate neighbors. The authors are asking: if we use a computer to approximate this fractional limit, how close can we get, and how fast does it get there as we make the computer's pixels smaller?

The Pixelated Puzzle: Chasing a Ghost Number

In the world of math, some numbers are like ghosts. They are the perfect, theoretical limits of a system, but you can never actually reach them with a real, physical object. The Fractional Hardy Constant is one of these ghosts. It represents the absolute minimum energy required for a specific type of mathematical shape to exist without collapsing, especially when that shape is centered around a singularity—a point where things get infinitely intense, like the center of a black hole or a sharp spike in a graph.

The authors of this paper, Andreea Dima and Liviu I. Ignat, are playing a game of "how close can you get?" They are using a method called the Galerkin approximation. Imagine you are trying to draw a perfect circle on a screen. You can't draw a true circle with a pixelated grid; you can only draw a jagged polygon that looks like a circle if you use enough tiny squares. The "optimal constant" in the real world is the perfect circle. The "discrete constant" is the jagged polygon. The paper asks: as we make the pixels smaller and smaller (a process controlled by a variable called $h$ ), how quickly does our jagged polygon's energy match the perfect circle's energy?

The answer they found is surprisingly slow, but mathematically beautiful. They proved that the difference between the computer's answer and the real, perfect answer shrinks at a rate of $1 / |\log h|^2$ .

To understand what that means, let's use an analogy. Imagine you are trying to fill a bucket with water using a teaspoon. If you just keep scooping, you might think you'll fill it quickly. But in this mathematical world, the "teaspoon" gets smaller and smaller, yet the "bucket" has a weird, logarithmic shape that makes the water level rise incredibly slowly. Even if you make your pixels (the teaspoon) a million times smaller, the error doesn't vanish instantly. It vanishes, but it does so with a "logarithmic" slowness. It's like trying to hear a whisper in a storm; even if you turn up the volume (make the mesh finer), the background noise (the error) fades away very gradually, following a specific, predictable pattern involving the square of a logarithm.

The Tools of the Trade

How did they prove this? They didn't just run a simulation and guess; they built a rigorous mathematical bridge.

First, they needed a lower bound. They had to prove that the computer's answer could never be too good. They used a "logarithmic improvement" of the Hardy inequality. Think of this as adding a tiny, extra safety net under the ghost number. This safety net has a specific shape involving a logarithm (a function that grows very slowly). This proved that no matter how clever the computer is, the error must be at least as big as $1 / |\log h|^2$ . It set a floor for the performance.

Second, they needed an upper bound. They had to show that the computer could actually achieve this speed. To do this, they constructed a "competitor"—a specific, made-up mathematical shape that is almost perfect but slightly flawed. They called this a "pseudo-minimizer." It's like a runner who is almost world-record speed but trips slightly on a pebble. By carefully analyzing this runner's performance on the computer's grid, they showed that the error was at most $1 / |\log h|^2$ .

When the floor and the ceiling meet, you have the exact answer. The paper proves that the convergence rate is exactly $1 / |\log h|^2$ . This is the same rate found in the classical (non-fractional) version of the problem, which is a significant result because fractional problems are usually much messier and harder to predict.

What This Means (and What It Doesn't)

The authors are very clear about the scope of their victory. They have solved the puzzle for the case where the power $p$ is equal to 2 (which is like measuring energy in a standard, quadratic way). They have shown that for a smooth, convex domain (a nice, roundish shape) containing the origin, the piecewise linear elements (the triangular pixels) converge at this specific logarithmic rate.

However, they also explicitly point out where the map ends. They admit that for other values of $p$ (where the energy is measured differently, like $p=3$ or $p=4$ ), the story is different. The "logarithmic improvement" tool they used for $p=2$ doesn't exist yet for other values. They also note that while they used a specific type of mesh (triangles), the question remains open whether using different types of approximations, like combinations of Gaussian functions (bell curves), might change the speed of convergence.

So, this paper doesn't claim to have solved every version of the Fractional Hardy problem. Instead, it provides a precise, sharp map for one specific, important territory. It tells us that when we try to simulate these fractional, long-range interactions on a computer, we shouldn't expect miracles. The error will shrink, but it will do so with a stubborn, logarithmic slowness. It's a reminder that even with the most advanced math and the finest grids, some of nature's sharpest limits are incredibly difficult to pin down, and the path to the truth is often a slow, steady climb rather than a sudden leap.

Technical Summary: Galerkin Approximation of the Fractional Hardy Constant

Problem Statement
This paper addresses the numerical approximation of the optimal constant in the fractional Hardy inequality using finite-dimensional schemes. Specifically, it investigates the convergence rate of the discrete optimal constant, $\lambda_{N,s,h}(\Omega)$ , to the continuous sharp constant, $\lambda_{N,s}$ , as the mesh size $h \to 0$ . The study focuses on the case $p=2$ in dimension $N \geq 1$ with a fractional exponent $s \in (0, \min\{1, N/2\})$ . The domain $\Omega$ is a bounded, convex, and smooth subset of $\mathbb{R}^N$ containing the origin.

The continuous fractional Hardy inequality is given by:
$[u]_{\dot{H}^s(\mathbb{R}^N)}^2 \geq \lambda_{N,s} \int_{\mathbb{R}^N} \frac{|u(x)|^2}{|x|^{2s}} dx,$
where $[u]_{\dot{H}^s(\mathbb{R}^N)}$ denotes the Gagliardo seminorm. The constant $\lambda_{N,s}$ is known to be sharp but is not attained in the homogeneous fractional Sobolev space $\dot{H}^s(\mathbb{R}^N)$ . The infimum is approached by a "pseudo-minimizer" behaving like $|x|^{-(N-2s)/2}$ . The paper seeks to quantify how well piecewise linear finite element approximations capture this sharp constant.

Methodology
The authors employ a Galerkin approximation using piecewise linear finite elements ( $V_h$ ) on a quasi-uniform and regular triangulation of $\Omega$ . The discrete constant is defined as the infimum of the Rayleigh quotient over $V_h$ . To establish the convergence rate, the authors utilize a two-sided bounding strategy:

Lower Bound: The proof relies on a logarithmic improvement of the fractional Hardy inequality on bounded domains. By incorporating a remainder term involving a logarithmic weight, the authors derive a lower bound for the difference $\lambda_{N,s,h} - \lambda_{N,s}$ . This involves splitting the domain into regions near the origin (where the singularity lies) and away from it, utilizing inverse estimates and properties of the finite element space to control the error.
Upper Bound: The construction of a specific test function (competitor) $u_\varepsilon$ is central to the upper bound. This function is a modification of the pseudo-minimizer, cut off near the origin using a smooth cut-off function $\eta_\varepsilon$ and a logarithmic weight. The authors use a "ground state representation" (a technique adapted from previous work on the classical case) to express the energy deficit of this competitor. They then project this competitor onto the finite element space $V_h$ and estimate the interpolation error using homogeneous Gagliardo-Nirenberg inequalities and specific lemmas regarding the behavior of the cut-off functions.

Key Contributions and Results
The primary result of the paper is Theorem 1.1, which establishes the sharp order of convergence for the discrete constant:
$\lambda_{N,s,h}(\Omega) - \lambda_{N,s} \sim \frac{1}{|\log h|^2}.$
This result demonstrates that the convergence rate for the fractional case ( $s \in (0,1)$ ) is identical to the rate previously established for the classical case ( $s=1$ ) in works by the authors and others.

The technical achievements supporting this theorem include:

Logarithmic Improvement: The application of a refined Hardy inequality with a logarithmic remainder term to obtain the lower bound.
Ground State Representation: The adaptation of a ground state representation formula to the fractional setting to construct a near-optimal competitor sequence $u_\varepsilon$ that yields the matching upper bound.
Explicit Error Estimates: Detailed quantitative estimates for the energy and $L^2$ norms of the constructed competitor sequence, showing that the deficit scales as $|\log \varepsilon|^{-2}$ .

Significance and Scope
The paper extends the program of analyzing the discretization of functional inequalities, specifically following the work on Sobolev and classical Hardy inequalities. The authors note that the convergence rate of $|\log h|^{-2}$ reflects deep structural features of the extremizing sequence, particularly the concentration of mass near the origin and the non-attainment of the minimum.

The authors explicitly state that their method is tailored to the $L^2$ case ( $p=2$ ). In Section 5, they discuss the limitations of their approach regarding $L^p$ versions ( $p \neq 2$ ). They identify that extending these results to general $p$ requires:

A logarithmic improvement inequality for the $L^p$ deficit (currently unavailable in the literature).
A ground-state representation for the $L^p$ deficit to construct explicit competitors.
Resolution of whether the convergence rate remains logarithmic or degrades due to the specific structure of the $L^p$ deficit (which involves powers like $\max\{4, 2p\}$ ).

The paper concludes by suggesting that the $|\log h|^{-2}$ rate may be intrinsic to the fractional Hardy deficit itself, rather than an artifact of the piecewise linear discretization, and posits that similar rates might persist in nonlinear approximation classes (e.g., Gaussian mixtures), though this remains an open question.

The Pixelated Puzzle: Chasing a Ghost Number

The Tools of the Trade

What This Means (and What It Doesn't)

More like this