arXiv subjects

Tatsuya Kaneko

Publications and source records attributed to Tatsuya Kaneko.

At least 19 recordsLinked to original sources

arXiv2026

Total variation bounds for Schrödinger--Föllmer sampling of boundary-vanishing densities

We derive total variation bounds for the Schrödinger--Föllmer sampler when the target density vanishes at the boundary of its support. This boundary behavior produces a terminal-time singularity in the drift near the boundary. We first prove a base estimate under a single quantitative hypothesis, finiteness of the relative Fisher information, without assuming boundedness or Lipschitz continuity of the drift. We then combine this estimate with a mixture regularization, thereby reducing the problem to determining the growth rate of the regularized Fisher information. For densities with an exact polynomial vanishing order on the boundary of a smooth bounded domain, we determine this growth rate and its leading constant. Orders above one yield the square-root rate in the number of steps, with a logarithmic loss at order one. For orders between one half and one, the rate exponent decreases continuously from one half toward one third. Finite mixtures of triangular kernel densities fall into the linear case.

math.PR↗

arXiv2026

Superconductivity in doped spin multimer systems

Binding energy, which quantifies pair formation, is a key factor in the emergence of superconductivity. Here, we show that even when multiple spins are complexly coupled, hole-doped systems can be mapped, in the strong-binding-energy limit, onto an effective hardcore boson model equivalent to the XXZ model, and exhibit promising signatures of superconductivity. We demonstrate this theory analytically and numerically in the double Kondo lattice model. Using the density-matrix renormalization group method, we show that a pairing state is maintained via a crossover even for parameters away from the strong-coupling regime. Additionally, we find that once binding energies are sufficiently generated, pair correlations develop similarly regardless of the details of local spin correlations. Our findings provide useful insights into the mechanism of superconductivity.

cond-mat.supr-con↗

arXiv2026

Interaction-enhanced antiferromagnetic correlations in a Kondo ladder with a gapped band structure

We investigate magnetic correlations in the two-leg Kondo-Hubbard ladder using the density-matrix renormalization group method. We focus on a parameter regime in which the noninteracting electronic system has a narrow indirect band gap. In this regime, there is no Fermi surface, and the conventional Fermi-surface-based Ruderman-Kittel-Kasuya-Yosida picture is not directly applicable; magnetic correlations between localized spins become weaker as the noninteracting band gap increases. We find that a moderate on-site Coulomb interaction enhances the antiferromagnetic correlations between localized spins. This enhancement, which is also seen in the staggered magnetization induced by a pinning field at the edges, occurs for both ferromagnetic and antiferromagnetic Kondo couplings. Our results show that electron-electron interactions can enhance magnetic correlations between localized spins even when the underlying noninteracting electronic structure is gapped, suggesting an interaction-driven route to magnetic instabilities in narrow-gap Kondo-lattice systems.

cond-mat.str-el↗

arXiv2026

TRIP-Shapley: Propagation-Aware Contribution Attribution Framework for Peer-to-Peer Federated Learning

Contribution attribution is a fundamental problem in federated learning (FL), yet it remains challenging in peer-to-peer FL (P2PFL), where participant influence propagates through repeated model exchanges. Recent work, DICE, addresses this issue by directly tracking multi-round propagation paths. However, its practical deployment is hindered by two challenges: (i) communication explosion caused by multi-hop propagation path tracking, and (ii) vulnerability to misreporting attacks. To overcome these challenges, we consider contribution attribution in P2PFL with a model-free coordinator used solely for auditing. Based on this setting, we propose TRIP-Shapley, a propagation-aware Shapley-inspired surrogate for contribution attribution. Instead of explicitly tracking multi-round paths, TRIP-Shapley computes round-wise contribution scores independently and propagates only scalar contribution values. This enables attribution of influence accumulated throughout the entire training process while maintaining practical communication costs. In addition, we introduce lightweight mechanisms to detect and mitigate misreporting attacks. Rather than theoretically approximating Shapley values, TRIP-Shapley is empirically evaluated against Shapley values as a reference. Experiments demonstrate communication savings of up to $45.8\times$ relative to the compared DICE configurations, with favorable cost--reference-alignment trade-offs overall, although alignment is weaker in some FMNIST and 25-client settings. The proposed defenses reduce contribution inflation under the two evaluated attacks.

cs.LG↗

arXiv2026

CARVY-FL: Client Anticlustering for Robust Voting in Provably Secure Federated Learning

Federated learning (FL) enables collaborative training without directly sharing raw data, but remains vulnerable to malicious clients. Voting-based FL improves robustness by partitioning clients into groups, training one model per group, and aggregating predictions by plurality voting. However, under class-disjoint non-IID data, distribution-oblivious grouping can yield highly variable certified accuracy (CA). We propose CARVY-FL, which estimates client distribution types from one-epoch model updates and uses anticlustering to increase within-group distributional diversity. Under a fixed grouping, CARVY-FL retains the voting-based CA guarantee while increasing vote margins. Experiments on MNIST and Fashion-MNIST show higher CA than FLCert. Under BadNets with model replacement, CARVY-FL improves the AUC of 100-ASR by 11.1% and 14.9%, respectively.

cs.CR↗

arXiv2026

ActionCache: Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinement

Vision-Language-Action (VLA) models have emerged as a promising approach for generalizable robotic manipulations. In particular, flow-matching-based VLA models have shown remarkable success due to their capability to generate precise and smooth action sequences and capture multimodal distributions. However, the iterative denoising process in the action head acts as a major computational bottleneck, posing a critical challenge for real-time deployment. To address this challenge, we propose ActionCache, a plug-and-play external cache that opportunistically reuses past intermediate actions to warm-start generations from the vicinity of target actions, drastically reducing the inference latency. Specifically, ActionCache stores the intermediate actions with compact multimodal keys, which enables retrieval from similar past contexts across different episodes or even different tasks. Experimental results in simulation and real-world environments demonstrate that ActionCache maintains high task success rates in a low-latency regime, achieving action head inference acceleration of up to $10.44\times$ and $40.17\times$ for representative flow-based VLA, $π_{0.5}$ and GR00T-N1.6, respectively.

cs.RO↗

arXiv2026

Pairing properties of correlated three-leg ladders with strong interchain couplings near 1/3 filling

We investigate the ground-state properties of correlated three-leg ladders near 1/3 filling. We apply the density-matrix renormalization group method to the three-leg t-J ladder with strong interchain couplings and evaluate its pairing nature. When holes are doped into the spin-gapped state at 1/3 filling, we find that pair correlations develop with power-law decays while spin correlations decay exponentially. On the other hand, doping of electrons into the 1/3-filled state does not give rise to substantial pair correlations. We also discuss the hole-doped state in the three-leg Hubbard model to compare it with the pairing state in the t-J model. Our numerical demonstrations provide insights into the electronic properties of trilayer nickelate superconductors.

cond-mat.supr-con↗

arXiv2026

Correlation-driven branch in doped excitonic insulators

We investigate the spectral properties of doped one-dimensional excitonic insulators. Employing matrix-product-state-based methods, we compute the single-particle spectrum and optical conductivity in a correlated two-band model. Our numerical calculation reveals the emergence of a correlation-driven in-gap branch in the doped state. The origin of the in-gap branch is examined by decomposing the propagation dynamics of a single particle, elucidating that the doping-induced branch is associated with excitonic correlations. Our demonstrations suggest that the doping-induced branch can serve as an indicator of electron-hole correlations.

cond-mat.str-el↗

arXiv2026

Doublon-Holon Pairing State in Photodoped Mott Insulators

We demonstrate the existence of an unconventional pairing state in photodoped Mott insulators on ladder and quasi-two-dimensional geometries, characterized by quasi-long-range doublon-holon correlations that signal Mott exciton condensation. The doublon-holon pairing exhibits correlations of $d$-wave-like symmetry, reminiscent of superconducting pairing in chemically doped Mott insulators. By constructing the phase diagram, using density matrix renormalization group, we reveal that the doublon-holon pairing state in the photodoped ladder emerges between the spin-singlet, charge-density-wave, and $η$-pairing phases. Our study suggests that the interplay of charge, spin, and $η$-spin degrees of freedom can give rise to exotic quantum many-body states in photodoped Mott insulators.

cond-mat.str-el↗

arXiv2026

Nonlinear phononics in LaFeAsO: Optical control of the crystal structure toward possible enhancement of superconductivity

Nonlinear phononics provides a route to control crystal structures through light-induced phonon excitation. In this study, we apply nonlinear phononics to an iron-based superconductor, LaFeAsO, with the aim of tuning its crystal structure toward the ideal one to enhance superconductivity. We simulate light-induced phonon dynamics on the anharmonic lattice potential determined by first-principles calculations. We find that the anion height $h$, a key structural parameter in iron-based superconductors, approaches its ideal value when an appropriate infrared-active phonon mode is selectively excited. This result suggests the possibility of controlling crystal structures and enhancing superconductivity in iron-based superconductors based on the concept of nonlinear phononics.

cond-mat.supr-con↗

arXiv2026

Optical control of the crystal structure in the bilayer nickelate superconductor La3Ni2O7 via nonlinear phononics

Superconductivity in the bilayer nickelate La$_3$Ni$_2$O$_7$ occurs when the interlayer Ni-O-Ni bond angle becomes straight under pressure, suggesting a strong relationship between the crystal structure and the emergence of superconductivity. In this study, we theoretically propose a way to control the crystal structure of La$_3$Ni$_2$O$_7$ toward the tetragonal symmetry via light irradiation instead of pressure using the idea of nonlinear phononics. Here, resonant optical excitation of an infrared-active (IR) lattice vibration induces a nonlinear Raman-mode displacement through the anharmonic phonon-phonon coupling. We calculate the light-induced phonon dynamics on the anharmonic lattice potential determined by first-principles calculation. We find that the interlayer Ni-O-Ni bond angle gets slightly closer to straight when an appropriate IR mode is selectively excited. Our study suggests that light irradiation can be a promising way for structural control of La$_3$Ni$_2$O$_7$.

cond-mat.supr-con↗

arXiv2026

Electronic band structure, phonon dispersion, and magnetic triple-q state in GdGaI

We theoretically investigate the physical properties of the magnetic van der Waals material GdGaI. Using first-principles calculations, we compute the phonon dispersion of GdGaI and show no imaginary phonons, suggesting that phonon-driven phase transitions are unlikely to occur in GdGaI. Our band calculation reveals that the electronic bands near the Fermi energy are composed of Gd 5d and Ga 4p orbitals. We construct a tight-binding model that incorporates the Gd 5d and Ga 4p orbitals to investigate the magnetic structure. We introduce Kondo coupling between electrons in Gd 5d orbitals and localized spins in Gd 4f orbitals and present the modified band structure when localized spins form a magnetic order characterized by three q vectors that connect the valence and conduction bands. We discuss the origin of the spin order based on the Ruderman-Kittel-Kasuya-Yosida mechanism and suggest that Coulomb interactions acting on electrons near the Fermi level can contribute to the ordering of localized spins.

cond-mat.str-el↗

arXiv2025

$t$-$J$ model for strongly correlated two-orbital systems: Application to bilayer nickelate superconductors

We derive a $t$-$J$ model applicable to strongly correlated two-orbital systems including bilayer nickelate superconductors. Using the Schrieffer-Wolff transformation, we exclude the doubly occupied states, which raise the onsite Coulomb energy, and derive the resulting spin interactions from the two-orbital Hubbard model. We also introduce effective interactions attributed to the interorbital Coulomb interaction. To adapt the effective model to bilayer nickelates that exhibit high-temperature superconductivity, we quantitatively evaluate the strengths of the spin interactions based on the hopping parameters in La$_3$Ni$_2$O$_7$. Considering the evaluated effective interactions, we propose a simplified $t$-$J$ model for bilayer nickelate superconductors.

cond-mat.str-el↗

arXiv2025

PRIOT: Pruning-Based Integer-Only Transfer Learning for Embedded Systems

On-device transfer learning is crucial for adapting a common backbone model to the unique environment of each edge device. Tiny microcontrollers, such as the Raspberry Pi Pico, are key targets for on-device learning but often lack floating-point units, necessitating integer-only training. Dynamic computation of quantization scale factors, which is adopted in former studies, incurs high computational costs. Therefore, this study focuses on integer-only training with static scale factors, which is challenging with existing training methods. We propose a new training method named PRIOT, which optimizes the network by pruning selected edges rather than updating weights, allowing effective training with static scale factors. The pruning pattern is determined by the edge-popup algorithm, which trains a parameter named score assigned to each edge instead of the original parameters and prunes the edges with low scores before inference. Additionally, we introduce a memory-efficient variant, PRIOT-S, which only assigns scores to a small fraction of edges. We implement PRIOT and PRIOT-S on the Raspberry Pi Pico and evaluate their accuracy and computational costs using a tiny CNN model on the rotated MNIST dataset and the VGG11 model on the rotated CIFAR-10 dataset. Our results demonstrate that PRIOT improves accuracy by 8.08 to 33.75 percentage points over existing methods, while PRIOT-S reduces memory footprint with minimal accuracy loss.

cs.LG↗

arXiv2025

Floquet engineering of effective pairing interactions in a doped band insulator

We investigate the pairing state in a doped band insulator under a periodic driving field. We employ a correlated fermionic model on a honeycomb lattice, in which pairing glue is obtainable via repulsive interactions, and derive an effective model under circularly polarized light. We demonstrate that the effective pairing interaction for doped fermions obtained with the second-order perturbation theory is tunable by the frequency and amplitude of the driving field. We find the optimal frequency range to enhance the pairing interaction and show that the modified effective system can strengthen the two-body bound state. Our study suggests that external driving light can reinforce superconducting pair states in strongly correlated electron systems.

cond-mat.str-el↗

arXiv2024

A New Era of Excitonic Insulators

The fundamental idea of the excitonic insulator (EI) driven by electron-hole correlations in narrow-gap semiconductors or semimetals was originally proposed in the 1960s, and only theoretical studies had been advanced for a long time. However, the rise of new candidate materials and recent developments in measurement techniques have enabled us to discuss the possibilities of EI states in real materials experimentally. In this article, we review recent progress in the research of EIs. We start with an introduction to the theoretical background of the EI and the mechanism of the order formation including its relation to the physics of the Bardeen-Cooper-Schrieffer (BCS) - Bose-Einstein condensation (BEC) crossover. We also review the EI states studied in the context of strongly correlated electron systems. Then, we introduce the candidate materials for the EI and the issues raised by recent experiments. For example, the phase transitions in several candidate materials are accompanied by lattice distortions, and the contributions from electron-lattice coupling hinder the identification of the excitonic entity. We also review the profiles of the collective modes to discuss the dynamical signatures of the EI.

cond-mat.str-el↗

arXiv2024

Federated Learning with Relative Fairness

This paper proposes a federated learning framework designed to achieve \textit{relative fairness} for clients. Traditional federated learning frameworks typically ensure absolute fairness by guaranteeing minimum performance across all client subgroups. However, this approach overlooks disparities in model performance between subgroups. The proposed framework uses a minimax problem approach to minimize relative unfairness, extending previous methods in distributionally robust optimization (DRO). A novel fairness index, based on the ratio between large and small losses among clients, is introduced, allowing the framework to assess and improve the relative fairness of trained models. Theoretical guarantees demonstrate that the framework consistently reduces unfairness. We also develop an algorithm, named \textsc{Scaff-PD-IA}, which balances communication and computational efficiency while maintaining minimax-optimal convergence rates. Empirical evaluations on real-world datasets confirm its effectiveness in maintaining model performance while reducing disparity.

stat.ML↗

arXiv2024

Optical signatures of dynamical excitonic condensates

We theoretically study dynamical excitonic condensates occurring in bilayers with an imposed chemical potential difference and in photodoped semiconductors. We show that optical spectroscopy can experimentally identify phase-trapped and phase-delocalized dynamical regimes of condensation. In the weak-bias regime, the trapped dynamics of the order parameter's phase lead to an in-gap absorption line at a frequency almost independent of the bias voltage, while for larger biases, the frequency of the spectral feature increases approximately linearly with bias. In both cases there is a pronounced second harmonic response. Close to the transition between the trapped and freely oscillating states, we find a strong response upon application of a weak electric probe field and compare the results to those found in a minimal model description for the dynamics of the order parameter's phase and analyze the limitations of the latter.

cond-mat.str-el↗