The core mathematical observation is deceptively simple. In kernel ridge regression, the prediction for the training data can be written as a weighted sum of the eigenvectors of the kernel matrix, where each weight is the projection of the target vector onto that eigenvector, scaled by the corresponding eigenvalue divided by the eigenvalue plus the regularization parameter. This spectral decomposition means that whatever the model learns is confined to the span of the kernel’s eigenvectors, and the contribution of each direction depends jointly on how strongly the target aligns with it and how large its eigenvalue is. A direction with a huge eigenvalue but poor alignment contributes almost nothing, while a weakly aligned trailing eigenvector can quietly dominate the error.

From this decomposition, the authors derive a striking result about reconstruction, the model’s ability to reproduce its own training labels. They prove that as long as the kernel matrix has sufficient rank, the reconstruction error can be made arbitrarily small simply by shrinking the regularization parameter, an algebraic fact that holds regardless of the data distribution. In high-dimensional settings, near-zero training loss is therefore nearly always achievable. The researchers demonstrate this empirically with synthetic targets built from eigenvectors of a linear kernel: once the input dimension exceeded roughly 500, reconstruction errors collapsed toward zero across the board. The practical implication is unsettling for practitioners who treat a low training loss as evidence of a healthy model, because the training error carries almost no predictive power for test performance.

The study goes further and shows that near-zero reconstruction error is not even a reliable sign of overfitting. In experiments on MNIST images of the digits four and nine, the team constructed learning targets aligned either with the top eigenvectors of the kernel matrix or with its trailing eigenvectors. When the target was aligned with the leading directions, both training and test predictions were nearly perfect. When it was aligned with the trailing directions, the model still reproduced the training data almost exactly, yet its test performance collapsed. The same pattern held on CIFAR10, FashionMNIST, and SVHN2, confirming that the gap between training and test behavior is governed by which part of the eigenspectrum the target inhabits, not by the training loss itself.

To explain these observations, the authors turn to matrix perturbation theory, treating the kernel matrix built from any finite sample as a perturbed version of a hypothetical optimal kernel matrix of the same size. Using classical tools such as the Davis-Kahan theorem, they derive an explicit finite-sample bound on the generalization error expressed in terms of three quantities: the alignment of the learning targets with the eigenvectors, the magnitude of the eigenvalues, and the gaps between consecutive eigenvalues. Strong generalization, the theorem says, requires increasing any of these. Targets aligned with the top eigenvectors benefit from all three at once, because leading eigenvalues are large, well separated, and estimated most accurately from finite data.

This perturbation perspective yields a reinterpretation of the familiar bias-variance tradeoff. In their framing, bias arises when the top eigenvectors cannot represent the learning targets, while variance arises when the targets must be represented by trailing eigenvectors, whose sample estimates are unstable and noisy. Overfitting in kernel ridge regression, on this view, is essentially a failure of the kernel to distinguish information from noise, either because the data are fundamentally too noisy or because the kernel choice is suboptimal. The finding echoes recent work on benign overfitting, but grounds it in concrete spectral quantities that can be computed from a single training set.

Perhaps the most provocative claim concerns regularization. The ridge parameter, long regarded as the main lever for controlling model complexity in the non-parametric setting, turns out to have minimal influence on generalization for most learning targets. Because leading eigenvalues are typically large relative to any reasonable regularization strength, the parameter barely changes their weights. It matters only when targets align with trailing eigenvectors, precisely the regime where generalization is already poor. The authors summarize this bluntly: regularization is more about salvaging bad performance than improving good performance, and its primary role is mitigating numerical instability rather than shrinking the hypothesis space. Their experiments confirm the pattern, showing that stronger regularization often increases generalization error except in the trailing-eigenvector regime.

The paper also delivers a practical payoff in the form of a simple empirical risk estimator, built from the projections of the target onto each eigenvector weighted by the inverse eigenvalue gaps. When benchmarked against the Kernel Alignment Risk Estimator of Jacot and colleagues and the eigenlearning estimator of Simon and colleagues on MNIST and FashionMNIST, the new estimator tracked the true testing error most closely in the leading eigendirections, roughly up to index 250 for MNIST and 300 for FashionMNIST. Existing estimators significantly underestimated the risk in trailing eigendirections, where the new bound, though loose, at least captured the upward trend. The authors note that sharper eigenvector perturbation bounds could improve the estimator further.

Methodologically, the work distinguishes itself from prior analyses by avoiding Gaussian assumptions on the sampling operator, an assumption that recent critiques have shown to be unrealistic in practice, and by focusing on the finite-sample regime rather than asymptotic limits. The comparison with a hypothetical optimal training set, chosen to minimize misalignment with the actual sample, allows the authors to state their bounds without distributional assumptions. They also highlight a connection to empirical observations in deep learning, noting that well-generalized neural networks tend to have weight matrices with isolated eigenvalues, consistent with the emphasis on eigenvalue gaps in their theory.

The authors are candid about limitations. Their analysis assumes noiseless labels, does not handle missing data, and produces bounds that may not be tight, since they are built from a chain of inequalities. Even so, the message is likely to resonate widely: what a kernel machine learns is determined not by how well it fits its training data, but by whether the structure of the data and the kernel conspire to place the target in the stable, well-estimated part of the spectrum. For a field increasingly concerned with understanding why overparameterized models generalize at all, that reframing turns the spotlight away from the loss curve and toward the geometry of the eigenspectrum.

Subject of Research: Eigen-alignment and generalization in kernel ridge regression

Article Title: On Kernel Eigen-Alignments of KRR: Reconstruction and Generalization

Article References: Liu, Y., Fokoue, E., Lange, R., & Krutz, D. (2026). On Kernel Eigen-Alignments of KRR: Reconstruction and Generalization. Machine Learning, 115(10), Article 238. https://doi.org/10.1007/s10994-026-07177-w

Image Credits: AI Generated

DOI: 10.1007/s10994-026-07177-w

Keywords: kernel ridge regression, eigenvectors, generalization, machine learning, kernel alignment, matrix perturbation theory, regularization, overfitting, learning theory, spectral bias, eigenvalues, risk estimation

The original Sraffa model describes a multi-sector economy conceived as a circular flow, in which goods are produced by means of other goods. Each industry uses inputs, labor, and produced means of production to generate its output, and the central analytical question concerns how the social surplus is distributed between profits and wages. In the static formulation, owners of the means of production target a uniform rate of profit across all sectors, and prices adjust to make that target consistent with the given technology and the real wage. Prices, in this classical view, are not signals of marginal utility but means of distributing the social surplus, reflecting both technical conditions of production and social arrangements surrounding income distribution.

What has remained underexplored, the authors argue, is the temporal dimension of this system. Mainstream dynamic models frequently treat prices and wages as exogenous or predetermined variables, overlooking the feedback loops that arise when each is determined within the same economy it helps to describe. Earlier work by Nikaido and Kobayashi added dynamics to Sraffa’s equations using continuous time, capturing a wage-price spiral in which wage growth responds to inflation with a bargaining-power parameter. But a continuous-time formulation has a subtle drawback: without explicit lags, production delays vanish, and with them the very possibility of capital advances and a meaningful rate of profit. As the economist Luigi Boggio observed, without stocks and production lags there is no place for a rate of profit at all. Discrete time, by contrast, imposes a time-to-build structure that mirrors the sequential nature of real production cycles.

Building on a discrete-time extension developed by Brida and colleagues, Álvarez and Lorenzo modify the price equation so that the wage paid in each period is not a constant but the cost of a standard subsistence basket valued at the previous period’s prices. The dynamics then collapse into a single linear system: prices today equal a matrix applied to prices yesterday, where that matrix combines the technology coefficients scaled by the profit rate with a term capturing labor’s consumption of the subsistence basket. This compact formulation admits an exact analytical solution, allowing the researchers to trace price paths from any initial configuration of prices forward through time without approximation.

The mathematics that emerges is strikingly elegant. Because the system’s matrix contains only non-negative entries, the Perron-Frobenius theorem guarantees that its largest eigenvalue is a real number associated with an eigenvector whose coordinates are also real. When the matrix is diagonalizable, which occurs with probability one when coefficients are drawn from continuous distributions, the individual price levels may grow or shrink without bound, but the ratios between any two prices converge to the ratio of the corresponding coordinates of the dominant eigenvector. In a two-sector economy, the ratio of the second price to the first inexorably approaches the quotient of the second and first components of that eigenvector, and the same logic generalizes to economies with any number of sectors.

Numerical simulations confirm the analytical results with almost disorienting speed. In a two-sector economy with initial prices set at equal values and randomly generated coefficients constrained to keep the system economically reasonable, relative prices become essentially constant by roughly the fifth period of a thirty-period horizon, settling near 0.33, precisely the ratio predicted by the dominant eigenvector. A three-sector simulation tells the same story: relative prices converge rapidly to the eigenvector ratio of approximately 2.57, even though the dominant eigenvalue in that case exceeds one and absolute price levels are exploding. The economy, in other words, can experience unbounded inflation in all prices simultaneously while the structure of relative values locks into place within a handful of periods.

The speed of that convergence depends on a quantity economists have long recognized as decisive: the subdominant eigenvalue, the second-largest in absolute value. The smaller the subdominant eigenvalue relative to the dominant one, the faster the transitory components of the price dynamics die away and the faster the system settles onto its eigenvector trajectory. This echoes earlier work by András Bródy, who showed that for Leontief matrices the second eigenvalue is the crucial quantity for calculating convergence speed. Random-matrix reasoning suggests that in larger systems convergence should be even faster, though recent empirical studies caution that multiple subdominant eigenvalues in real input-output structures can slow the adjustment, leaving open important questions about price stickiness in high-dimensional industrial economies.

Perhaps the most philosophically significant finding concerns what determines the equilibrium relative prices. They do not depend on the profit rate at all, nor on preferences or bargaining strength, but solely on the technological coefficients of the production system. This resonates with Sraffa’s own insistence that relative prices are completely constrained by the system of production and the conditions of its reproduction, and it aligns with the interpretation, advanced by Steenge and others, that Sraffa’s famous standard commodity is nothing more than an eigenvector of the system. It also departs from comparative statics treatments and from models in which individual preferences drive the dynamics. In this framework, the properties of the whole determine the properties of the parts: uniform profit rates emerge from an interconnected production system rather than being imposed as an assumption.

Allowing wages to adjust endogenously does change one headline conclusion relative to earlier discrete-time work. There is no fixed long-run equilibrium price level; prices change from period to period, with the rate of price growth governed by technological factors. What stabilizes is not the level but the structure, the web of relative prices that classical economists identified with natural prices. The wage-price spiral, driven by workers’ purchasing power over a subsistence basket, generates persistent nominal movement while leaving the deep architecture of value untouched, a result consistent with Nikaido and Kobayashi’s continuous-time analysis and robust to the shift into discrete time.

The authors are candid about the model’s limits. Technology is held fixed, the profit rate is assumed uniform and decided in advance, and the wage adjustment rule, in which wages rise exactly in proportion to prices, leaves little room for institutional bargaining power or the nonlinearities of the wage-profit curve. The simulations, moreover, are confined to low-dimensional systems. Future extensions could allow sectoral profit rates to differ and adjust toward a global rate, let technological coefficients evolve over time, or redefine the wage basket in terms of Sraffa’s standard commodity, which would freeze distribution entirely. The policy implications, the authors suggest, point toward regulating corporate profit margins, strengthening antitrust enforcement against clustered oligopolies, and bolstering collective bargaining, so that the distributive struggle this framework is designed to illuminate can be observed and shaped in the real economies it describes.

Subject of Research: Dynamic price and wage adjustment in a discrete-time Sraffa-Leontief production system

Article Title: Dynamic price and wage setting in a discrete-time Sraffa-Leontief system

Article References: Dynamic price and wage setting in a discrete-time Sraffa-Leontief system. (n.d.). https://doi.org/10.1016/j.heliyon.2026.e45505

Image Credits: AI Generated

DOI: 10.1016/j.heliyon.2026.e45505

Keywords: Sraffa system, Leontief model, price dynamics, wage-price spiral, eigenvalues, Perron-Frobenius theorem, discrete time, income distribution, classical economics, relative prices, input-output analysis, production prices