🔢 mathematics

A negative exponent range for Audenaert's complementary McCarthy trace inequality

This paper proves Audenaert's conjectured complementary McCarthy trace inequality for all negative exponents $q < 0$ by utilizing matrix means and log-majorization techniques, while also extending the known positive exponent range to $q \ge 2$ and characterizing the equality condition as $A=B$ .

Original authors: Xing Li, Bin Zhou

Published 2026-08-24

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Xing Li, Bin Zhou

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

In the world of mathematics, there is a branch called linear algebra that deals with grids of numbers known as matrices. These grids are not just abstract lists; they are the tools scientists use to describe how systems change, from the flow of electricity in a circuit to the vibration of atoms in a crystal. When these grids represent physical quantities that are always positive, like energy or probability, they are called positive matrices. Mathematicians have long been interested in how these positive matrices behave when you add them together or raise them to a power. A famous rule, known as McCarthy's inequality, tells us that for certain types of powers, the total sum of the combined grid is always larger than the sum of the individual parts. However, when the power becomes negative, the rules flip, and the relationship becomes much harder to predict. This creates a puzzle: if you know how the parts behave on their own, can you accurately predict how they behave when combined, especially when dealing with these tricky negative powers?

For years, researchers have been trying to map out exactly where these rules hold true and where they break down. A specific question, posed by mathematicians Audenaert and Kittaneh, focused on a gap in the map. They knew the answer for very large negative numbers and for small positive numbers, but there was a mysterious middle range of negative numbers where the answer was unknown. It was like having a map of a coastline that was complete at the northern and southern extremes, but with a foggy, uncharted stretch in the middle. The question was whether a specific, more complex formula could accurately describe the behavior of these combined matrices in that foggy zone.

Two mathematicians, Xing Li and Bin Zhou, have now cleared that fog. They proved that the formula works for every negative number, filling in the entire missing gap. Their work confirms that the relationship between the combined matrix and its parts follows a strict, predictable pattern, even in the most difficult negative range. They did not just guess or simulate the answer; they constructed a rigorous mathematical proof that leaves no room for doubt. Furthermore, they discovered that the formula is perfectly precise only when the two original matrices are identical. If the matrices are different, the formula gives a strict inequality, meaning the combined result is always strictly less than the sum of the parts plus a correction term.

To reach this conclusion, the researchers had to change their perspective. Instead of looking at the matrices directly, they turned the problem upside down by using the inverse of the numbers involved. This transformation turned the difficult problem into a question about a specific way of combining matrices known as the parallel sum, a concept familiar to engineers who study how electrical resistors work when connected side-by-side. By analyzing this parallel sum, they were able to use a chain of established mathematical tools to show that the inequality holds true. They relied on a method that compares the "size" of matrices in a very specific way, ensuring that the relationship holds even when the numbers involved are fractions or decimals.

The team also revisited the positive side of the problem, where the powers are greater than two. They showed that this part of the puzzle was already solved by a different, earlier discovery involving block matrices, which are grids made of smaller grids. By connecting their new findings with this older result, they completed the full picture. The work demonstrates that the mathematical landscape is more orderly than previously suspected, with a single, unified rule governing the behavior of these matrices across all negative powers. The only time the rule becomes an exact equality is when the two starting matrices are exactly the same, a condition that acts as a unique fingerprint for the formula's precision.

This resolution is significant because it closes a chapter that had been open for some time. It removes a conjecture, replacing a question mark with a definitive answer. The proof is built on solid ground, using finite-dimensional matrices, which are the kind used in practical computing and physics. The researchers did not rely on approximations or computer simulations that might miss a rare exception. Instead, they used logical deduction to show that the inequality must hold for every possible case within the defined range. Their work ensures that anyone using these mathematical tools in the future can rely on this specific relationship without worrying about the unknown middle ground.

The journey to this answer involved a careful examination of how different mathematical means, or ways of averaging numbers, interact with each other. The researchers used a specific type of average that sits between the harmonic and geometric means, which are standard ways of calculating averages. By tracing the path through these means, they were able to bound the error and prove that the formula never fails. They also checked the boundaries of their proof to ensure that the numbers they used were the best possible, meaning the formula cannot be made any tighter or more precise without breaking it. This attention to detail confirms that the result is not just a lucky guess but a fundamental property of how these positive matrices interact.

In the end, the paper provides a complete and verified map of the behavior of these matrices for negative powers. It resolves the uncertainty that lingered in the range between negative two and zero, showing that the rules are consistent and predictable. The finding is a testament to the power of mathematical logic to illuminate dark corners of theory, turning a conjectural gap into a known territory. For anyone working with these systems, the result offers a reliable guide, ensuring that calculations involving combined positive matrices and negative powers can be performed with confidence. The mystery is solved, and the path forward is clear.

Technical Summary: A Negative Exponent Range for Audenaert's Complementary McCarthy Trace Inequality

Problem Statement
The paper addresses a specific gap in the theory of trace inequalities involving positive matrices $A$ and $B$ . Audenaert previously established complementary McCarthy-type inequalities involving the mixed term $\text{Tr}(A^{1/2}BA^{1/2})^{q/2}$ . The known results covered specific ranges of the exponent $q$ :

Reverse inequality ( $\ge$ ): Valid for $0 < q \le 1$ and $2 \le q \le 3$ .
Direct inequality ( $\le$ ): Valid for $q \le -2$ .

The ranges $-2 < q < 0$ and $q \ge 3$ were left as conjectural cases in the problem list of Audenaert and Kittaneh. The primary objective of this work is to resolve the conjecture for the negative exponent range $-2 < q < 0$ , and to clarify the status of the positive range $q \ge 3$ .

Methodology
The authors employ a combination of operator theory techniques, specifically focusing on matrix means and log-majorization:

Inversion and Parallel Sum: For the negative exponent case ( $q < 0$ ), the authors set $q = -r$ (where $r > 0$ ) and perform the inversion $X = A^{-1}$ and $Y = B^{-1}$ . This transforms the problem into establishing a trace inequality for the parallel sum $X : Y = (X^{-1} + Y^{-1})^{-1}$ .
Kubo–Ando Means: The proof utilizes the chain of Kubo–Ando means, specifically the relationship between the harmonic mean ( $X!Y$ ), the geometric mean ( $X\#Y$ ), and the arithmetic mean. The inequality $X!Y \le X\#Y \le (X+Y)/2$ is central to the derivation.
Log-Majorization: The authors apply an Ando–Hiai type log-majorization result (Lemma 2.1) which relates the eigenvalues of the geometric mean $X\#Y$ to the singular values of the product $X^{1/2}Y^{1/2}$ . This allows the bounding of trace terms involving geometric means by Schatten quasi-norms.
Schatten Hölder Inequality: A finite-dimensional Schatten Hölder inequality is used to bound the mixed term $\|X^{1/2}Y^{1/2}\|_r^r$ in terms of $\text{Tr}(X^r)$ and $\text{Tr}(Y^r)$ .
Norm-Compression Inequality: For the positive exponent case, the authors utilize Audenaert's existing norm-compression inequality for positive semidefinite $2 \times 2$ block matrices to derive the result directly, rather than developing a new matrix inequality.

Key Contributions and Results

Resolution of the Negative Exponent Conjecture: The paper proves that the inequality
$\text{Tr}(A+B)^q \le \text{Tr}(A^q) + \text{Tr}(B^q) + (2q-2)\text{Tr}(A^{1/2}BA^{1/2})^{q/2}$
holds for all $q < 0$ , not just $q \le -2$ . This settles the conjectural range $-2 < q < 0$ .
- Intermediate Result: The authors first establish a one-parameter asymmetric trace inequality for the parallel sum (Theorem 3.1) and then specialize it to the symmetric case (Theorem 3.3), which yields the optimal coefficient $2 - 2^{-r}$ (equivalent to $2q-2$ when $q=-r$ ).
Clarification of the Positive Exponent Range: The paper demonstrates that the inequality
$\text{Tr}(A+B)^q \ge \text{Tr}(A^q) + \text{Tr}(B^q) + (2q-2)\text{Tr}(A^{1/2}BA^{1/2})^{q/2}$
holds for all $q \ge 2$ . This covers the conjectural range $q \ge 3$ and strengthens the known bound from $q \ge 3$ down to $q \ge 2$ . The authors note this is a bibliographical observation derived from Audenaert's norm-compression inequality, not a new block matrix inequality.
Equality Conditions: The paper provides a rigorous characterization of the equality case for the negative exponent range. It is proven that equality holds if and only if $A = B$ . The proof involves analyzing the equality conditions in the Kubo–Ando mean chain and the log-majorization steps, showing that $X:Y = \frac{1}{2}(X\#Y)$ implies $X=Y$ .

Significance and Claims
The paper claims to settle the specific open problem regarding the negative exponent range $-2 < q < 0$ in the Audenaert–Kittaneh problem list. The authors emphasize that the negative exponent part is the "new contribution" of the work, relying on a novel application of the parallel sum trace inequality and log-majorization techniques.

Regarding the positive exponent range, the authors modestly frame their contribution as a "bibliographical observation" that clarifies the natural endpoint $q=2$ and confirms that the $q \ge 3$ case follows directly from previously established norm-compression results. The paper does not propose new applications or future experimental directions, focusing strictly on the theoretical resolution of these specific trace inequality conjectures.

Technical Summary: A Negative Exponent Range for Audenaert's Complementary McCarthy Trace Inequality

More like this