🔢 mathematics

Linear preserver problems in matrix positivity theory

This survey presents the current state of research on linear preserver problems for various matrix positivity classes, offering an overview of recent developments and identifying gaps to guide future studies.

Original authors: Projesh Nath Choudhury, Shivangi Yadav

Published 2026-02-25

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Projesh Nath Choudhury, Shivangi Yadav

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 have a giant, magical factory that takes in boxes of numbers (matrices) and spits out new boxes of numbers. This factory is run by a very strict rule: it must be linear. This means if you feed it two boxes and add them together before processing, it's the same as processing them separately and adding the results later. It's a predictable, straight-line machine.

Now, imagine inside this factory, there are special "VIP Zones." These zones contain boxes with very special properties. For example:

The "Safe" Zone: Boxes that represent stable, positive energy (Positive Semidefinite matrices).
The "Optimization" Zone: Boxes that help solve complex puzzles in economics and engineering (Copositive matrices).
The "Pattern" Zone: Boxes where every little sub-group of numbers follows a specific sign pattern (Sign Regular matrices).

The Big Question:
The authors of this paper, Projesh Nath Choudhury and Shivangi Yadav, are asking: "What kind of machines (linear maps) can we build that take a box from a VIP Zone, process it, and guarantee the output box is still in that same VIP Zone?"

They call these machines "Preservers."

The Two Types of Preservers

The paper distinguishes between two types of machines:

The "Perfect Match" Machine (Onto Preserver):
This machine is a perfect translator. It takes every box from the VIP Zone, turns it into a new box, and ensures that every single box in the VIP Zone can be made by this machine. It's a one-to-one swap. If you have a "Safe" box, the machine turns it into another "Safe" box, and you can reverse the process perfectly.
- Analogy: Imagine a translator who speaks two languages perfectly. If you give them a sentence in English, they give you the exact French equivalent, and if you give them that French sentence back, they give you the exact English original. Nothing is lost or gained; it's a perfect swap.
The "One-Way" Machine (Into Preserver):
This machine is safer but messier. It guarantees that if you put a "Safe" box in, the output is also "Safe." However, it might not be able to produce every possible "Safe" box, and you might not be able to reverse the process.
- Analogy: Imagine a filter that turns all red water into blue water. If you put red water in, you get blue water out. But maybe the filter only makes some shades of blue, or maybe you can't turn the blue water back into red water easily. It's a one-way street.

The Journey Through the VIP Zones

The paper acts as a travel guide, visiting different "VIP Zones" (types of matrices) and reporting on what kind of machines work there.

1. The "Safe" Zone (Positive Semidefinite Matrices)

These are the most famous VIPs. They show up everywhere, from statistics to quantum physics.

The Finding: For the "Perfect Match" machines, we know exactly what they look like. They are basically just rotating or flipping the boxes in very specific ways (mathematically, multiplying by invertible matrices).
The Mystery: For the "One-Way" machines, we are still lost. We know some that work, but we don't have a complete list of all the machines that can keep things "Safe" without being perfect reversals. It's like knowing how to lock a door, but not knowing every possible way to pick a lock that still leaves the door locked.

2. The "Pattern" Zone (Sign Regular Matrices)

These are boxes where the numbers follow a strict rhythm of pluses and minuses.

The Finding: The authors (Choudhury and Yadav) recently solved the puzzle for these! They found that the only machines that preserve these patterns are very simple: they just multiply rows and columns by positive numbers, flip the whole box, or swap rows and columns.
The Metaphor: It's like a dance troupe. If you want to keep the dance pattern the same, you can only tell the dancers to move faster (multiply by numbers), turn around (flip), or swap partners (permute). You can't invent new dance moves.

3. The "Optimization" Zone (Copositive Matrices)

These are tricky. They are "Safe" only when you look at them from a specific angle (non-negative vectors).

The Finding: The "Perfect Match" machines here are also well understood—they are just specific types of flips and swaps.
The Mystery: The "One-Way" machines are a nightmare. The paper shows a machine that keeps things "Copositive" but doesn't fit the standard "flip and swap" mold. It's a weird, custom-built machine that breaks the rules we thought applied to everything else.

4. The "Stability" Zone (P-Matrices and D-Stable Matrices)

These are boxes that ensure systems (like bridges or economies) don't collapse.

The Finding: For "Perfect Match" machines, we have a solid list. But for "One-Way" machines, the rules get complicated.
The Twist: In 2D (2x2 boxes), the rules are weird and specific. But in 3D and larger, the rules seem to simplify again, though the authors note that proving this for real-world numbers (not just complex ones) is still a work in progress.

Why Should You Care?

You might think, "Who cares about boxes of numbers?" But these "VIP Zones" are the backbone of the modern world.

Engineering: Ensuring a bridge won't vibrate apart.
Economics: Making sure a market model doesn't crash.
AI and Data: Ensuring algorithms converge to the right answer.

The "Linear Preserver Problems" are like asking: "What are the fundamental laws of physics for these mathematical structures?"

If we know exactly which machines preserve these structures, we can:

Simplify Complex Problems: Instead of checking a million different scenarios, we just check the few "Preserver" machines.
Build Better Systems: We can design algorithms that are guaranteed to stay stable because we know exactly how they transform.
Find the Gaps: The paper highlights where we don't know the rules yet. It's like a map with "Here Be Dragons" written on the edges, inviting future mathematicians to explore and discover new laws.

The Bottom Line

This paper is a status report on a century-old quest.

What we know: For the "Perfect Match" machines (Onto preservers), we have a very clear, elegant list for almost every VIP Zone. The solutions are usually simple: flip, swap, or scale.
What we don't know: For the "One-Way" machines (Into preservers), the picture is foggy. There are weird, unexpected machines that work, and we haven't found them all yet.

The authors are essentially saying: "We've mapped the main highways, but the backroads are still full of surprises. Here is where you should go next if you want to solve the mystery."

1. Problem Statement

The paper addresses Linear Preserver Problems (LPPs), a central theme in matrix and operator theory. The core problem is to characterize all linear operators $L: V \to V$ (where $V$ is a space of matrices) that preserve specific structural properties or subsets of matrices.

Specifically, the authors focus on positivity classes of matrices. The problem is formulated in two distinct ways:

Onto Preservers: Characterize $L$ such that $L(S) = S$ (bijective preservation).
Into Preservers: Characterize $L$ such that $L(S) \subseteq S$ (inclusion preservation).

The survey aims to synthesize the current state of research for various positivity classes, highlighting complete classifications, identifying gaps in the literature (particularly for "into" preservers), and formulating open questions.

2. Methodology

The authors employ a structured survey approach, combining:

Literature Review: Systematic analysis of foundational results (e.g., Frobenius, Schneider) and recent advancements (up to 2026).
Classification via Structural Forms: Most results characterize preservers as compositions of standard transformations, such as:
- Standard Form: $L(A) = XAY$ or $L(A) = XA^T Y$ .
- Congruence: $L(A) = QAQ^T$ or $QAQ^*$ .
- Hadamard Product: $L(A) = H \circ A$ .
- Permutation/Exchange: Multiplication by permutation or exchange matrices.
Density Arguments: Utilizing the fact that certain sets (like positive definite matrices) are dense in their closures (positive semidefinite matrices) to extend results from "onto" to broader contexts (Lemma 2.2).
Counterexample Construction: Using specific matrix examples (e.g., Horn matrix, Choi map) to disprove conjectures regarding "into" preservers and to demonstrate that standard forms do not always suffice.
Graph-Theoretic Connections: Linking matrix properties (like complete positivity) to graph properties (e.g., bipartite graphs, triangle-free graphs) to determine ranks and preserver structures.

3. Key Contributions and Results by Matrix Class

The paper organizes results by specific positivity classes:

A. Positive Semidefinite (PSD) Matrices

Onto Preservers: Completely classified by Schneider (1965). Maps are of the form $A \mapsto RAR^*$ or $A \mapsto RA^T R^*$ for invertible $R$ .
Into Preservers: Open Problem. While congruences and Hadamard products with PSD matrices preserve PSD, a complete characterization of all into preservers is unknown.

B. Completely Positive (CP) Matrices & cp-rank

Context: CP matrices are entrywise non-negative PSD matrices ( $A = BB^T, B \ge 0$ ).
cp-rank Preservers:
- For $n=2,3$ : Preservers are of the form $A \mapsto S^T AS$ with $S \ge 0$ (Shaked-Monderer).
- For $n \ge 4$ : The conjecture that preservers must be monomial congruences ( $A \mapsto Q^T AQ$ where $Q$ is a monomial matrix) was proven by Shitov (2023).
- Gap: The case where $n > m \ge 4$ (maps from $S_m$ to $S_n$ ) remains open.

C. Copositive Matrices

Definition: $A$ is copositive if $x^T Ax \ge 0$ for all $x \ge 0$ .
Onto Preservers: Completely characterized as $A \mapsto QAQ^T$ where $Q$ is a non-negative monomial matrix (Gowda-Sznajder-Tao, Shitov).
Into Preservers:
- Standard forms ( $A \mapsto XAX^T$ with $X \ge 0$ ) and Hadamard products with CP matrices ( $A \mapsto H \circ A$ ) are known preservers.
- Counterexample: The authors show via Example 2.19 that not all into preservers fit the form $\sum X_i A X_i^T$ .
- Gap: A general classification of into copositive preservers is missing.

D. Sign Regular (SR) and Totally Positive (TP) Matrices

Results: Recent work by Choudhury–Yadav (2023/2024) provides complete classifications for onto preservers of SR, SSR, TP, and TN matrices for all dimensions and sign patterns.
Key Insight: The classification of preservers for general SR matrices reduces to the study of order-2 minors ( $SR_2$ ).
Forms: Preservers are compositions of diagonal scaling, transposition, exchange matrices, and sign changes.

E. M-matrices, Inverse M-matrices, and H-matrices

Onto Preservers: Characterized by Berman–Hershkowitz–Johnson and Tam–Liou.
Forms: Combinations of diagonal scaling ($FAE$), transposition ( $A^T$ ), permutation congruence ( $Q^T AQ$ ), and for H-matrices, Hadamard multiplication by unit modulus matrices ( $K \circ A$ ).
Gap: Into preservers for these classes are explicitly noted as "unexplored."

F. P-matrices, D-stable, and A-matrices

Onto Preservers: Characterized for real and complex spaces. Forms include diagonal scaling, transposition, permutation congruence, and signature matrix congruence ($SAS$).
Into Preservers:
- Hershkowitz–Johnson (1986) classified invertible into preservers under specific constraints (e.g., null space conditions).
- For $n \ge 3$ , forms include $A \mapsto A + D$ (where $D$ is a diagonal matrix derived from $A$ ).
- Gap: The classification for $n=2$ real matrices differs from the complex case, and general non-invertible into preservers are not fully characterized.

G. Semipositive (SP) and Minimally Semipositive (MSP) Matrices

Onto Preservers: Must be of standard form $L(A) = XAY$ with $X, Y$ being non-negative monomial matrices.
Into Preservers:
- Standard Form: $L(A) = XAY$ where $X$ is Row Positive (RP) and $Y$ is Inverse Non-negative (IN).
- Counterexamples: The paper highlights that not all into preservers are of standard form (Example 2.48).
- Recent Progress: Jayaraman–Mer (2021) proved that if $L$ is invertible, it must be of the standard form $XAY$ with $X$ invertible RP and $Y$ IN.
- Gap: Classification of non-invertible into preservers for MSP matrices, especially when $m=n$ , remains open.

4. Significance and Future Directions

Unification: The survey unifies over a century of research, providing a coherent framework for understanding how different positivity classes relate to linear transformations.
Clarification of Gaps: The paper explicitly identifies that while onto preservers are largely classified for most major positivity classes, into preservers remain a significant open frontier. The "into" case is harder because the maps need not be bijective, allowing for singular or non-standard transformations.
Methodological Shift: The work emphasizes the transition from ad-hoc proofs to structural characterizations using graph theory, density arguments, and specific counterexamples to refine conjectures.
Open Questions: The authors conclude by summarizing unresolved problems, particularly regarding:
- Into preservers for Copositive, M-matrices, and H-matrices.
- The $n > m \ge 4$ case for CP-rank preservers.
- Non-invertible into preservers for Semipositive and Minimally Semipositive matrices.

The paper serves as a definitive reference for researchers entering the field of linear preserver problems, offering a clear roadmap of solved problems and the specific challenges that define the current research landscape.