Symmetrically Reciprocal Matrix

• Symmetrically reciprocal matrices are symmetric matrices whose entries are defined as reciprocals of underlying quantities such as distances or combinatorial values.
• They preserve structural and spectral properties under both standard and Hadamard inversion, enabling explicit factorizations and spectral interpolations in diverse applications.
• These matrices arise naturally in combinatorics, graph theory, and algebra, presenting challenges in spectral decomposition and positive semidefiniteness that drive current research.

A symmetrically reciprocal matrix is a real symmetric matrix whose entries are defined as the reciprocals of underlying symmetric quantities—typically, distances or combinatorial numbers—and which preserves structural and spectral properties under matrix inversion or Hadamard inversion. Such matrices arise naturally in combinatorics, algebra, and graph theory, providing a unifying framework for the spectral study of reciprocal distance matrices, reciprocal Pascal matrices, and more generally, symmetric matrices whose reciprocal varieties and associated linear algebraic structures are tractable or admit explicit characterization.

1. Definitions and Key Examples

The prototypical symmetrically reciprocal matrix appears in three main forms:

• Reciprocal Distance Matrices (Harary matrices): For a simple, undirected, connected graph $G$ of order $n$, the Harary matrix $RD(G)$ is defined by $(RD(G))_{ij} = 0$ if $i = j$ and $(RD(G))_{ij} = 1/d_G(i, j)$ for $i \neq j$, where $d_G(i, j)$ is the shortest path distance between vertices $i$ and $j$ (Tian et al., 2022).
• Reciprocal Pascal Matrix: For $0 \leq i,j \leq n$, the reciprocal Pascal matrix $R$ (sometimes denoted $M$) has entries $R_{i,j} = 1/\binom{i+j}{i}$; this is the Hadamard (elementwise) inverse of the symmetric Pascal matrix $P$ with $P_{i,j} = \binom{i+j}{i}$ (Richardson, 2014, Prodinger, 2015).
• Jordan Subalgebras of Symmetric Matrices: More generally, a linear subspace $\mathcal{L} \subset S^n$ of symmetric matrices is symmetrically reciprocal if its reciprocal variety (all inverses $X^{-1}$, $X \in \mathcal{L}$, invertible) is also a linear space, a property characterized precisely by the Jordan algebra structure (Bik et al., 2020).

The essential unifying property is symmetry: $R_{i,j} = R_{j,i}$ holds by definition or construction.

2. Structural Properties

Symmetrically reciprocal matrices are always real symmetric matrices, which implies the following:

• Diagonalizability: All such matrices are diagonalizable over $\mathbb{R}$ with real spectra.
• Spectral Interpolation: In the graph-theoretic setting, the generalized reciprocal distance matrix $RD_\alpha(G) = \alpha RT(G) + (1-\alpha) RD(G)$ interpolates continuously between $RD(G)$ (Harary matrix) and $RT(G)$ (diagonal transmission matrix) via the reciprocal distance signless Laplacian $RQ(G)$, itself symmetric: $RD_0(G) = RD(G)$, $RD_{1/2}(G) = RQ(G)$, $RD_1(G) = RT(G)$ (Tian et al., 2022).
• Hadamard and Ordinary Inverse: For some classes (e.g., reciprocal Pascal matrices), both the Hadamard inverse and the ordinary inverse preserve symmetry, and explicit integer formulas are available for the inverse matrix entries (Richardson, 2014).

3. Spectral Theory and Inequalities

The spectral theory of symmetrically reciprocal matrices is highly developed in several contexts:

• Eigenvalues for Special Graphs: For complete graphs $K_n$, $RD_\alpha(K_n)$ has spectrum $\{n-1,\,(\alpha n - 1)^{n-1}\}$; for $r$-regular graphs of diameter $2$ with adjacency spectrum $\{r, \lambda_2, ..., \lambda_n\}$,

$$\text{Eigenvalues of } RD_\alpha(G) = (\alpha(n+r)-1) \text{ and } (\alpha(n+r)-1 + (1-\alpha)\lambda_i).$$

Similar explicit formulas exist for complete bipartite, wheel, and multipartite graphs (Tian et al., 2022).

• Spectral Monotonicity: The spectral radius $p(RD_\alpha(G))$ is monotonic in both edge addition and $\alpha$. For $0 < \alpha < 1,$ adding an edge increases all eigenvalues; increasing $\alpha$ increases all eigenvalues, with equality for transmission-regular graphs (Tian et al., 2022).
• Inequalities and Bounds: Lower and upper bounds for the largest eigenvalue (spectral radius) $p(RD_\alpha(G))$ can be expressed in terms of the minimal and maximal reciprocal transmissions, as well as the minimal and maximal eigenvalues of $RD(G)$. Global bounds include

$$\frac{\sum_i RT(i)^2}{n} \leq p(RD_\alpha(G)) \leq \max_{i,j} \{ (\alpha RT(i) + (1-\alpha)/d_G(i,j)) RT(j) \}$$

and

$p(RD_\alpha(G)) \geq \frac{2H(G)}{n}$

where $H(G)$ is the Harary index (Tian et al., 2022).

• Spectral Structure of Reciprocal Pascal Matrix: The reciprocal Pascal matrix $M$ (Hadamard inverse of the Pascal matrix) is symmetric and positive definite; all its inverses are integer matrices, though a closed-form spectral decomposition remains open (Richardson, 2014, Prodinger, 2015).

4. Factorizations and Algebraic Structure

Explicit matrix factorizations play a central role in the study and application of symmetrically reciprocal matrices:

• Pascal Matrix Factorizations: The super–Catalan matrix $S$ admits two independent factorizations: $S = G R G$ (diagonal) and $S = L D L^T$ (triangular with diagonal correction), leading to an explicit formula for $R^{-1}$ in terms of matrices with integer entries (Richardson, 2014):

$$R = G^{-1} L D L^T G^{-1},\qquad R^{-1} = G (L^T)^{-1} D^{-1} L^{-1} G$$

where $G_{ii} = \binom{2i}{i}$, $L_{i,k} = \binom{2i}{i+k}$, $D$ diagonal.

• LU Decomposition: For the reciprocal Pascal matrix $M_{i,j} = \binom{i+j}{j}^{-1}$, lower- and upper-triangular factorizations with explicit closed forms exist and yield integer inverses (Prodinger, 2015).
• Jordan Algebra Characterization: The only linear spaces $\mathcal{L} \subset S^n$ whose set of inverses again spans a linear space are precisely the finite-dimensional Jordan subalgebras of $(S^n, \bullet_U)$. Here, the Jordan product is $X \bullet_U Y = \frac{1}{2}(XU^{-1}Y + YU^{-1}X)$. This equivalence gives an algebraic-geometric understanding of reciprocal symmetry (Bik et al., 2020).

5. Extremal and Combinatorial Properties

• Extremal Spectral Graphs: Within the family of $RD_\alpha(G)$ matrices, extremal graphs maximizing the spectral radius under constraints such as fixed vertex- or edge-connectivity, chromatic number, or independence number can be identified (e.g., Turán graphs, complete multipartite graphs) (Tian et al., 2022).
• Combinatorial Interpretation: The entries and determinants of symmetrically reciprocal matrices often admit combinatorial interpretations. For reciprocal Pascal matrices, their inverses have integer-valued expressions corresponding to lattice path enumeration and signed super–ballot numbers (Richardson, 2014, Prodinger, 2015).
• q-Analogues: All factorizations and formulas for the reciprocal Pascal matrix generalize to $q$-deformed settings, yielding corresponding $q$-reciprocal Pascal matrices with analogous structural and spectral properties (Prodinger, 2015).

6. Algebraic Geometry and Classification

• Reciprocal Varieties and Jordan Loci: The set of $m$-dimensional subspaces $\mathcal{L} \subset S^n$ for which the reciprocal variety $\mathcal{L}^{-1}$ is linear corresponds exactly to the Jordan locus $\Jo(m, S^n)$, stratified by Segre symbols and codimension. For low dimensions, these loci can be classified explicitly, providing a geometric stratification of the Grassmannian according to the closure under inversion (Bik et al., 2020).
• Explicit and Worked Examples: The classification yields explicit basis representations for pencils and nets of symmetric matrices with reciprocal closure, as well as polynomial (quadratic and determinantal) equations necessary for closure—allowing constructive verification whether a pencil or net is symmetrically reciprocal in the Jordan sense (Bik et al., 2020).

7. Open Problems and Future Directions

Several avenues remain open for investigation and generalization:

• Spectral Analysis: Determination of spectral decomposition for reciprocal Pascal matrices remains open, as does the explicit description of eigenstructures for broader reciprocal matrix classes (Richardson, 2014, Prodinger, 2015).
• Positive Semidefiniteness: For reciprocal distance matrices $RD_\alpha(G)$ there exists a graph-dependent threshold $\alpha_0(G)$ such that $RD_\alpha(G)$ is positive semidefinite if and only if $\alpha \geq \alpha_0$, with general characterization still an open problem (Tian et al., 2022).
• Energetic and Extremal Questions: The definition and maximization of generalized reciprocal energy, as well as spectral extremal problems for trees, unicyclic graphs, and graphs constrained by matching or cut-edge numbers, are open within the $RD_\alpha$ framework (Tian et al., 2022).
• Algebraic-Geometric Boundary Cases: Classifying higher-dimensional Jordan subalgebras, understanding the structure of non-diagonalizable Jordan loci, and potential applications to semidefinite optimization and algebraic statistics remain active directions (Bik et al., 2020).

Symmetrically reciprocal matrices offer a rich and interconnected field of study at the junction of algebraic combinatorics, spectral graph theory, and algebraic geometry, with numerous exact results and a significant landscape of open problems.