🔢 mathematics

A note on the Goldberg-Thorp example in light of the classification of linear ill-posed problems in Banach spaces

This note analyzes the 1963 Goldberg-Thorp example of a strictly singular mapping from $\ell^1$ onto $\ell^2$ as a hybrid-type operator with a non-complemented null-space, summarizing its structural properties and discussing the implications for well-posedness definitions and regularization strategies in the classification of ill-posed linear problems in Banach spaces.

Original authors: Bernd Hofmann, Jens Flemming

Published 2026-03-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Bernd Hofmann, Jens Flemming

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

The Big Picture: A Broken Map and a Lost City

Imagine you are trying to navigate a massive, infinite city (the space $\ell_1$ ) to reach a specific destination in a different, infinite city (the space $\ell_2$ ). You have a map (a mathematical operator called $B$ ) that tells you how to get from one city to the other.

In a perfect world, this map would be well-posed. That means:

Reliable: If you give the map a destination, it always gives you a route.
Stable: If you make a tiny mistake in your destination (like a typo in the address), the route it suggests changes only a tiny bit.
Clear: There is only one best route to take.

This paper is about a specific, famous map created by Goldberg and Thorp in 1963. The authors, Hofmann and Flemming, are saying: "This map is broken in a very tricky way. It looks like it works, but it's actually a trap for anyone trying to find a stable solution."

Part 1: The Two Rules of a Good Map

To understand why this map is broken, the authors explain that a "good" mathematical problem needs two things to be stable:

The Destination Must Be Reachable (Closed Range): The map must actually lead to the places you are trying to go. If the map leads to a foggy area where you can get close but never quite land, that's bad.
The Starting Point Must Be Unique (Complemented Null-Space): This is the tricky part. Imagine you are at a train station. If the station has many different platforms that all lead to the exact same train, you have a "null-space."
- Good Scenario: The platforms are clearly separated by walls (complemented). You can easily pick the right one.
- Bad Scenario (The Goldberg-Thorp Trap): The platforms are a giant, tangled maze with no walls (uncomplemented). If you try to pick a starting point, you might be standing on a platform that is infinitely close to another one, but they lead to different outcomes. A tiny error in where you stand sends you to a completely different part of the city.

The Paper's Main Point: The Goldberg-Thorp map ( $B$ ) has a destination that is reachable (it covers the whole city), but the starting point is a tangled, infinite maze. Because of this, the map is ill-posed.

Part 2: The "Hybrid" Monster

The authors classify this map as a "Hybrid-Type" operator. Think of it like a chameleon that is half one thing and half another, making it impossible to handle with standard tools.

It's not "Compact": Usually, if a map squishes things down (like a compact operator), it's easier to handle. This map doesn't squish things enough.
It's "Strictly Singular": It's so weird that if you try to look at any large, infinite chunk of the starting city, the map breaks down and stops working like a normal machine.

The Metaphor: Imagine trying to pour water (the data) from a bucket with a hole in the bottom (the starting space) into a cup (the destination).

In a normal problem, the hole is small and predictable.
In this "Hybrid" problem, the hole is a swirling vortex. You can pour water in, and it reaches the cup, but if you try to reverse the process (figure out exactly how much water you poured based on what's in the cup), the math explodes. The "vortex" (the uncomplemented null-space) makes it impossible to find a stable answer.

Part 3: Why Standard Fixes Fail

When mathematicians have a broken map, they usually try to fix it with Regularization.

The Analogy: Imagine you are trying to find a lost hiker in a foggy forest. You can't see them, so you guess where they might be and add a "penalty" if your guess is too far from the center of the forest. This is called Tikhonov regularization.

The Problem: The authors show that for this specific Goldberg-Thorp map, standard regularization fails.

Why? Because the "vortex" (the uncomplemented null-space) is so chaotic that no matter how you try to smooth out the data, the answer jumps wildly.
Even worse, for many destinations, there is no single "best" starting point (no minimum-norm solution). It's like asking, "What is the shortest path to a place that has infinite paths of the exact same length?" The math says: "There is no answer."

Part 4: Is There Any Hope?

The paper ends with a glimmer of hope, but it comes with a catch.

The authors found a way to "slice" the starting city. If you ignore the tangled, messy part of the city and only look at a specific, clean, infinite corridor (a subspace $U$ ), the map works perfectly again!

On this specific corridor, the map becomes stable.
You can use standard regularization techniques here.

The Catch: To do this, you need to know the exact structure of that clean corridor. But in the real world, we often don't know the exact structure of the "vortex." So, while the math proves a solution exists in theory, it's very hard to build a practical tool to find it.

Summary in One Sentence

This paper warns us that the famous Goldberg-Thorp example is a mathematical "hybrid monster" that looks like it works but hides a chaotic, infinite maze at its core, making it impossible to solve stably with standard methods unless you can magically isolate a specific, clean path through the chaos.

1. Problem Statement

The paper addresses the classification of linear operator equations $Ax=y$ in infinite-dimensional Banach spaces, specifically focusing on the definition of well-posedness and the characterization of ill-posedness.

The Core Issue: Traditional definitions of well-posedness (often attributed to Nashed) rely primarily on the operator having a closed range. However, the authors argue that this is insufficient for non-injective operators in Banach spaces. A critical missing component is the complementedness of the null-space ( $N(A)$ ).
The Specific Case: The paper analyzes a specific counter-example from 1963 by Goldberg and Thorp: a bounded linear operator $B: \ell_1 \to \ell_2$ that is surjective but not injective.
The Challenge: For the operator $B$ , the null-space is uncomplemented in $\ell_1$ . This structural defect leads to instability in regularization methods (specifically Tikhonov regularization) and the non-existence of minimum-norm solutions for general data, challenging standard assumptions in inverse problem theory.

2. Methodology

The authors employ functional analysis techniques within the framework of Banach space theory to re-evaluate the stability and solvability of operator equations.

Re-definition of Well-Posedness: The authors propose a refined definition requiring two conditions for well-posedness:
1. The range $R(A)$ must be closed.
2. The null-space $N(A)$ must be topologically complemented in $X$ (i.e., $X = N(A) \oplus U$ for a closed subspace $U$ ).
Classification of Ill-Posedness: They distinguish between:
- Type I: The range contains an infinite-dimensional closed subspace (includes "hybrid-type" operators).
- Type II: The range does not contain an infinite-dimensional closed subspace (includes compact operators with infinite-dimensional range).
Analysis of the Goldberg-Thorp Operator ( $B$ ):
- They represent $B$ as a Mazur-type operator, defined by $Bx = \sum x_k \zeta(k)$ , where $(\zeta(k))$ is a dense sequence in the unit sphere of $\ell_2$ .
- They utilize the Bartle-Graves Theorem (existence of a continuous nonlinear right inverse) and the Open Mapping Theorem.
- They analyze the adjoint operator $B^*: \ell_2 \to \ell_\infty$ to determine properties of the range and null-space.
Regularization Analysis: The paper investigates the failure of standard Tikhonov regularization ( $\min \|Bx - y^\delta\|^2 + \alpha \|x\|_1$ ) due to the lack of weak*-to-weak continuity of $B$ and the uncomplemented null-space.

3. Key Contributions

A. Refinement of Well-Posedness Definitions

The paper argues that a closed range alone does not guarantee stability for non-injective operators. If $N(A)$ is uncomplemented, the pseudo-inverse $A^\dagger$ is unbounded, leading to a situation where arbitrarily small noise in $y$ can cause arbitrarily large errors in the reconstructed solution $x$ . Therefore, complementedness of the null-space is a necessary condition for a stable, well-posed problem.

B. Characterization of Hybrid-Type Operators

The authors formally define Hybrid-type operators as those that are:

Strictly singular (not an isomorphism on any infinite-dimensional closed subspace).
Have a range containing an infinite-dimensional closed subspace.
They prove that for such operators, the null-space is always uncomplemented.

C. Analysis of the Goldberg-Thorp Example

The paper provides a comprehensive theorem (Theorem 3.4) regarding the operator $B: \ell_1 \to \ell_2$ :

$B$ is a Hybrid-type operator (Strictly singular + Surjective).
The equation $Bx=y$ is Ill-posed of Type I.
The null-space $N(B)$ is uncomplemented in $\ell_1$ .
The adjoint operator $B^*: \ell_2 \to \ell_\infty$ is injective and has a closed range, making the adjoint equation $B^*\eta = z$ well-posed.

D. Regularization Limitations and Opportunities

Failure of Standard Methods: The authors prove that for the Goldberg-Thorp operator, minimum-norm solutions do not exist for general data (Proposition 4.1), and Tikhonov regularization fails to produce stable solutions for arbitrary noisy data because the operator lacks weak*-to-weak continuity.
Theoretical Workaround: They demonstrate that by restricting the operator $B$ to a specific closed infinite-dimensional subspace $U \subset \ell_1$ (constructed via a Riesz basis subsequence), the restricted operator becomes injective and weak*-to-weak continuous. On this subspace, Tikhonov regularization becomes viable and convergent (Corollary 4.5).

4. Key Results

Proposition 2.4: For any hybrid-type operator, the null-space is uncomplemented. This explains the separation between hybrid operators and compact operators in the classification diagram.
Theorem 3.4: The surjective operator $B: \ell_1 \to \ell_2$ is strictly singular, has an uncomplemented null-space, and is ill-posed of Type I. Its adjoint is well-posed.
Proposition 4.1: For the Goldberg-Thorp operator, a minimum-norm solution $x_{mn}$ exists only if the data $y$ is a scalar multiple of one of the basis vectors $\zeta(k)$ . For general data, no such solution exists.
Proposition 4.3 & Corollary 4.5: It is possible to construct a subspace $U \subset \ell_1$ such that the restriction $B|_U$ is injective, has a dense range, and is weak*-to-weak continuous. This allows for the application of Tikhonov regularization with convergence guarantees, provided the problem is restricted to this subspace.

5. Significance

Theoretical Clarification: The paper corrects a common misconception in inverse problems literature by highlighting that closed range is not sufficient for well-posedness in Banach spaces; complemented null-spaces are equally critical for stability.
Understanding Instability: It provides a concrete mathematical explanation for why certain regularization methods (like $\ell_1$ -minimization) fail for specific classes of operators (strictly singular, surjective maps between $\ell_1$ and $\ell_2$ ).
Practical Implications: While the Goldberg-Thorp operator is a theoretical construct, the results warn practitioners that in Banach spaces (like $\ell_1$ used in sparse recovery), the geometry of the null-space dictates the feasibility of stable reconstruction.
New Classification Framework: The distinction between Type I and Type II ill-posedness, refined to include the "Hybrid" case, offers a more nuanced tool for analyzing the solvability of linear operator equations in non-Hilbert settings.

In summary, the paper uses the Goldberg-Thorp example to demonstrate that the topological structure of the null-space is a decisive factor in the stability of linear inverse problems, necessitating a broader definition of well-posedness that goes beyond the mere closedness of the range.