Abstract
For a $m \times n$ matrix $B=(b_{ij})_{m \times n}$ with nonnegative entries $b_{ij}$ and any $k \times l$-submatrix $B_{ij}$ of $B$, let $a_{B_{ij}}$ and $g_{B_{ij}}$ denote the arithmetic mean and geometric mean of elements of $B_{ij}$ respectively. It is proved that if $k$ is an integer in $(\frac{m}{2}, m]$ and $l$ is an integer in $(\frac{n}{2}, n]$, then $$\Big(\prod_{i=k,j=l\atop B_{ij}\subset B}a_{B_{ij}}\Big)^{\frac{1}{C_m^k\cdot C_n^l}} \geq\frac{1}{C_m^k\cdot C_n^l}\Big(\sum_{i=k,j=l\atop B_{ij}\subset B}g_{B_{ij}}\Big),$$ with equality if and only if $b_{ij}$ is a constant for every $i,j$.
Introduction
The classical Arithmetic-Geometric Mean (AM-GM) inequality is a cornerstone of mathematical analysis, establishing that for any set of non-negative numbers, the arithmetic mean is always greater than or equal to the geometric mean. In recent decades, mathematical research has expanded this concept into "mixed-mean" inequalities. These inequalities explore nested applications of means—for instance, comparing the average of products against the product of averages across various subsets of a data sequence.
While previous work by mathematicians such as Carlson and Kedlaya successfully established these relationships for one-dimensional sequences (vectors), a gap remained regarding two-dimensional structures. This paper addresses that gap by extending mixed-mean theory to the sub-blocks, or submatrices, of a larger matrix.
The Mathematical Framework
The authors approach the problem through the lens of generalized power means ($m_r$). By treating the arithmetic mean as the limit of the power mean as $r \to 1$ and the geometric mean as the limit as $r \to 0$, they create a continuous bridge between the two measures.
The core of the proof relies on a sophisticated "Intersection Lemma." This lemma demonstrates that the power mean of any given submatrix can be decomposed and expressed through the power means of its intersections with all other possible submatrices of the same dimensions. To ensure this decomposition is valid, the submatrices must possess a high degree of overlap. This requirement leads to the paper's most significant constraint: the submatrices must cover more than half of the original matrix's rows ($k > m/2$) and columns ($l > n/2$).
The Mixed-Mean Theorem
The central result is a formal proof that for an $m \times n$ matrix $B$, the geometric mean of the arithmetic means of its $k \times l$ submatrices is consistently greater than or equal to the arithmetic mean of their geometric means.
Formally, the inequality is expressed as: $$\left( \prod_{B_{ij} \subset B} a_{B_{ij}} \right)^{\frac{1}{C_m^k \cdot C_n^l}} \geq \frac{1}{C_m^k \cdot C_n^l} \left( \sum_{B_{ij} \subset B} g_{B_{ij}} \right)$$
This theorem represents a "strong" generalization of existing work. If the matrix dimensions are reduced to $m=1$ or $n=1$, the formula collapses into previously established vector-based inequalities. Furthermore, if the submatrix size is set to equal the parent matrix ($k=m, l=n$), the expression simplifies to the classical AM-GM inequality. The authors also demonstrate that equality only occurs in the trivial case where the matrix is constant (all entries are identical).
Key Insights and Implications
The Half-Matrix Threshold
The necessity of the condition $k > m/2$ and $l > n/2$ is a critical insight. If submatrices are too small, they do not share enough common elements to maintain the structured relationship required for the inequality. The authors provide counterexamples using matrices of ones and zeros to show that when submatrices cover half or less of the parent matrix, the inequality can indeed fail.
Analytical Tools
Beyond the Intersection Lemma, the proof utilizes Discrete Hölder’s Inequality. This tool is essential for managing the transition between the product of sums and the sum of products, which typically represents the primary hurdle in mixed-mean proofs.
Conclusion and Future Directions
This research successfully extends mixed-mean theory into the second dimension, providing a robust theoretical framework for matrix sub-blocks. However, several avenues for future research remain open:
- Small Submatrices: The current proof is strictly bound by the $k > m/2, l > n/2$ requirement. Future work could investigate if alternative relationships exist for smaller sub-blocks.
- Alternative Means: This study focuses on Arithmetic and Geometric means. The behavior of Harmonic means or Stolarsky means in a similar 2D context remains unexplored.
- Higher-Order Tensors: As data science increasingly relies on multi-dimensional arrays (tensors), extending this 2D matrix inequality into $N$-dimensions would be a significant contribution to the field.
Summary updated