Prism: Structural Symmetry Scanning via Duality-Constrained Laplacian Projection
J. Xie · 2026
Read Prism: Structural Symmetry Scanning via Duality-Constrained Laplacian Projection by J. Xie (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-20245.txt.
Introduction
Jiatong Xie introduces Prism, a framework for symmetry diagnosis in networks. Given a graph Laplacian and a duality operator, Prism computes a duality defect, a scalar measuring self-consistency loss. With the true symmetry operator, the defect starts near zero and rises steadily as structure degrades. The best symmetric operator comes from a closed-form projection, and is learned from the graph's Fiedler vector. On live S and P 500 data, Prism spots rising stress, and in a backtest over five stress events, the defect rises before each crisis spike.
Catalog record
| CREATOR | J. Xie |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26719 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.20245
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Prism: Structural Symmetry Scanning via Duality-Constrained Laplacian Projection J. Xie Abstract We introduce Prism , a framework for structural symmetry diagnosis in complex networks. Given a graph Laplacian L L and a duality operator P P (a symmetric involution), Prism computes the duality defect δ ( L , P ) = ‖ L P − P L ‖ F / ‖ L ‖ F \delta(L,P)=\|LP-PL\|_{F}/\|L\|_{F} —a scalar measuring how far the network deviates from structural self-consistency. When P P encodes the network’s true symmetry, δ \delta starts near zero and rises monotonically as structure degrades; an arbitrary P P gives noise. We prove that the optimal L ′ L^{\prime} satisfying [ L ′ , P ] = 0 [L^{\prime},P]=0 is given by a closed-form block-diagonal projection, and provide an unsupervised alternating optimization that learns P P from the graph’s own Fiedler vector. Experiments on synthetic networks show the true- P P defect is 3.38 × 3.38\times more sensitive to structural degradation than an index-reversal baseline and more sensitive than modularity. On Zachary’s Karate Club with edge noise, Prism achieves 94.5 % 94.5\% community detection accuracy at 5 % 5\% noise versus 76.6 % 76.6\% for the raw Laplacian baseline. Applied to live S&P 500 data (2026-05-17), Prism detects rising structural stress (defect 0.43 → 0.73 0.43\to 0.73 over 90 days) while surface correlations remain low—a signal invisible to correlation-based methods. In a historical backtest spanning five major stress events (2011–2020), the duality defect exhibits a consistent pattern: it reaches elevated levels before the correlation spike that accompanies each crisis, and sustains high readings during periods of structural fragility that conventional metrics classify as calm. The duality defect is a first-principles structural admissibility condition, requiring no training data and computable in milliseconds. 1 Introduction Network analysis faces a fundamental tension: real-world graphs are noisy, yet the structural properties we care about—community membership, anomaly status, regime transitions—are discrete and sharp. Spectral methods address this by working with the graph Laplacian L , whose eigenvectors encode global structure. But eigendecomposition is sensitive to noise, and the resulting community assignments can degrade rapidly as edges are perturbed. We propose a different framing. Rather than asking what communities does this graph have? , we ask how far is this graph from satisfying a natural structural symmetry? This reframing turns a classification problem into a measurement problem, and yields a scalar diagnostic—the duality defect —that is continuous, interpretable, and provably more sensitive to structural degradation than standard spectral metrics. The duality constraint. Given a graph Laplacian L\in\mathbb{R}^{n\times n} and a symmetric involution P (a duality operator satisfying P^{2}=I , P=P^{\top} ), the duality constraint is [L,\,P]\;=\;LP-PL\;=\;0. (1) When ( 1 ) holds, L and P share an eigenbasis: the network decomposes cleanly into two mirror sectors. The duality defect measures the violation: \delta(L,P)\;=\;\frac{\|LP-PL\|_{F}}{\|L\|_{F}}. (2) A network with \delta=0 is structurally self-consistent under P . Rising \delta signals symmetry breaking. Prism. We introduce Prism , a framework that: 1. Given a fixed P (supervised mode), finds the closest Laplacian L^{\prime} satisfying ( 1 ) via closed-form block-diagonal projection. 2. Without a fixed P (unsupervised mode), jointly optimizes L^{\prime} and P via alternating minimization, initializing P from the graph’s own Fiedler vector. 3. Uses \delta(L,P) as a structural health metric: low defect means the network satisfies its natural symmetry; rising defect signals structural stress. Contributions. • A closed-form projection onto the commutant of P (Section 3 ), exact and parameter-free. • An unsupervised alternating optimization that learns P from data without external reference (Section 3 ). • Empirical demonstration that the duality defect with a meaningful P is 3.38\times more sensitive to structural degradation than with an arbitrary P , and more sensitive than modularity (Section 4 ). • A noise-robustness benchmark showing Prism achieves 94.5\% community detection accuracy at 5\% edge noise versus 76.6\% for the raw Laplacian baseline (Section 4 ). • A live financial application: on 2026-05-17, Prism detected rising structural stress (defect 0.43\to 0.73 over 90 days) in S&P 500 data while surface correlations remained low—a signal invisible to correlation-based methods (Section 5 ). • A historical backtest over five major stress events (2011–2020) showing that the duality defect reaches elevated levels before correlation spikes, and sustains high readings during structurally fragile periods that conventional metrics classify as calm (Section 5 ). 2 Background and Related Work Spectral graph theory. The graph Laplacian L=D-A (where D is the degree matrix and A the adjacency matrix) encodes global network structure in its spectrum. Spectral clustering [ 11 ] partitions nodes by the sign of the Fiedler vector (second eigenvector of L ). This approach is sensitive to noise: small edge perturbations can flip Fiedler vector signs and change community assignments entirely. Random Matrix Theory denoising. A standard approach to noise robustness is to threshold eigenvalues below the Marchenko–Pastur upper edge [ 8 ] , discarding components attributed to noise. This improves robustness but discards structural information indiscriminately and provides no interpretable diagnostic. In our benchmark (Section 4 ), RMT achieves only 61.4\% accuracy at 5\% noise, below both baseline and Prism. Graph neural networks. GNN-based methods [ 6 , 5 ] learn node representations from local neighborhoods. They achieve strong performance on in-distribution graphs but require labeled training data and generalize poorly to structural regime changes not seen during training. Critically, they learn statistical correlations between features and labels—they cannot detect structural anomalies that have no historical precedent. Symmetry in graphs. Graph automorphisms and equivariant networks [ 9 ] exploit known symmetries to reduce parameter count. Our approach is complementary: rather than assuming symmetry and building it into the architecture, we measure the deviation from symmetry as a diagnostic signal. Commutant projection. The projection of a matrix onto the commutant of a given operator is a classical construction in operator theory [ 4 ] . Its application to graph Laplacians as a structural constraint—and its use as a diagnostic metric—is, to our knowledge, new. Financial network analysis. Correlation-based network methods for financial risk detection [ 7 , 10 ] construct graphs from return correlations and apply spectral or community-detection methods. These methods are blind to structural stress that accumulates while surface correlations remain low—precisely the regime Prism targets. 3 Method 3.1 Setup Let G=(V,E) be an undirected weighted graph with n=|V| nodes. Let A\in\mathbb{R}^{n\times n} be the adjacency matrix, D=\mathrm{diag}(A\mathbf{1}) the degree matrix, and L=D-A the graph Laplacian. A duality operator is a matrix P\in\mathbb{R}^{n\times n} satisfying P^{2}=I and P=P^{\top} (symmetric involution). The eigenvalues of P are \pm 1 ; let V_{+},V_{-} denote the corresponding eigenspaces. 3.2 Supervised Prism: Projection onto the Commutant Definition 3.1 . The commutant of P is the subspace of matrices commuting with P : \mathcal{C}(P)=\{M\in\mathbb{R}^{n\times n}:MP=PM\} . Proposition 3.2 . Let P have eigendecomposition P=U\Lambda U^{\top} with \Lambda=\mathrm{diag}(\lambda_{1},\ldots,\lambda_{n}) , \lambda_{i}\in\{+1,-1\} . The orthogonal projection of L onto \mathcal{C}(P) (in Frobenius norm) is L^{\prime}=U\,\Pi(U^{\top}LU)\,U^{\top}, (3) where \Pi(M)_{ij}=M_{ij} if \lambda_{i}=\lambda_{j} , and 0 otherwise (zeroing out off-block entries in the P -eigenbasis). Proof. M\in\mathcal{C}(P) iff M is block-diagonal in the P -eigenbasis. The nearest block-diagonal matrix to U^{\top}LU in Frobenius norm is obtained by zeroing off-diagonal blocks, giving ( 3 ). ∎ The projection ( 3 ) is exact