Abstract
In this note, we provide a formal proof for the trace inequality $\text{Tr}(MN + PQ) \geq 0$ under specific structural conditions. We define $M = A - B$, $N = B^{-1} - A^{-1}$, $P = C - D$, and $Q = (B + D)^{-1} - (A + C)^{-1}$, where $A$ and $B$ are positive definite matrices and $C$ and $D$ are positive semidefinite matrices. This result extends a known scalar inequality to the matrix domain, providing a necessary mathematical foundation for establishing the uniqueness of Nash equilibria in concave communication games.
Introduction
In the fields of control theory and wireless communications—particularly in the study of Multiple-Input Multiple-Output (MIMO) systems—researchers frequently encounter the need to prove system stability or the uniqueness of an equilibrium. A recurring challenge in these proofs is demonstrating that certain mathematical expressions involving the traces of matrix products are consistently non-negative.
While a simplified version of the inequality (where $C = D = 0$) has been previously established, modern applications in game theory require a more generalized form. Specifically, to prove that a game (such as multiple users competing for bandwidth) has a single, stable outcome—a pure Nash equilibrium—one must often satisfy "diagonally strict concavity" conditions. This paper provides the formal matrix counterpart to the scalar inequalities used in these proofs, filling a gap in the existing literature.
Mathematical Framework and Methodology
The paper presents a purely theoretical proof leveraging the algebraic properties of Positive Definite (PD) and Positive Semidefinite (PSD) matrices.
Core Properties
The proof relies on several fundamental matrix identities and properties:
- Monotonicity of the Inverse: For PD matrices, if $A \geq B$ (in the Loewner order), then $B^{-1} \leq A^{-1}$.
- Cyclic Property of Traces: The identity $\text{Tr}(AB) = \text{Tr}(BA)$ allows for the strategic rearrangement of terms within the trace operator to simplify complex products.
The Auxiliary Result
A pivotal step in the derivation is the construction of Lemma 2.3. This technical lemma proves a specific relationship between traces: $$\text{Tr}\{XA^{-1}XB^{-1}\} - \text{Tr}\{X(A+C)^{-1}X(B+D)^{-1}\} \geq 0$$ This result allows the authors to compare the behavior of the trace under the perturbations introduced by matrices $C$ and $D$.
Decomposition Strategy
The authors decompose the target inequality into two distinct components, $\mathcal{T}_1$ and $\mathcal{T}_2$. The complexity of the proof arises from the fact that $\mathcal{T}_2$ is not guaranteed to be non-negative on its own. The core of the mathematical contribution is demonstrating that whenever $\mathcal{T}_2$ is negative, the magnitude of the positive component $\mathcal{T}_1$ is sufficiently large to compensate, ensuring that the total sum $\mathcal{T}_1 + \mathcal{T}_2$ remains $\geq 0$.
Key Insights
The Scalar-to-Matrix Bridge
The intuition for the theorem stems from a basic scalar inequality: $$(\alpha - \beta)(\frac{1}{\beta} - \frac{1}{\alpha}) + (\gamma - \delta)(\frac{1}{\beta+\delta} - \frac{1}{\alpha+\gamma}) \geq 0$$ The authors successfully bridge the gap between this scalar logic and complex matrix structures, showing that the non-negativity holds even when the variables are replaced by PD and PSD matrices. This is non-trivial due to the non-commutative nature of matrix multiplication.
Generalizing Rosen’s Condition
By establishing Theorem 1.1, the authors provide the mathematical machinery required to apply J.B. Rosen’s 1965 "diagonally strict concavity" condition to modern communication games. This condition is the standard tool for ensuring uniqueness in game theory. This paper allows that tool to be applied to multi-antenna settings where strategies are represented by matrices rather than simple scalars.
Conclusion
The primary result, Theorem 1.1, confirms that for PD matrices $A, B$ and PSD matrices $C, D$: $$\text{Tr} \{(A - B)(B^{-1} - A^{-1}) + (C - D)[(B + D)^{-1} - (A + C)^{-1}]\} \geq 0$$
While the inequality is highly specific to the defined structures of $M, N, P,$ and $Q$, it serves as a robust tool for optimization and game theory. Future research may involve applying this theorem to specific physical layer communication models to provide numerical validation of equilibrium uniqueness in complex network environments.
Summary updated