arXiv subjects

Daisuke Takahashi

Publications and source records attributed to Daisuke Takahashi.

At least 19 recordsLinked to original sources

arXiv2026

Improved Scaling for Fast Mode of Ozaki Scheme II

Ozaki scheme II emulates high-precision matrix multiplication using low-precision integer matrix operations based on the Chinese remainder theorem (CRT). It first scales the high-precision matrices to convert them into integer matrices. For this scaling step, Ozaki scheme II provides two modes: accurate mode, which uses INT8 matrix multiplication to estimate scaling factors, and fast mode, which applies the Cauchy--Schwarz inequality at lower computational cost. We show that the existing formula lacks scale invariance; multiplying the input matrices by a constant changes the effective bit width of the integer matrices in the scaling step, causing accuracy degradation or CRT recovery failure. To address this, we propose a revised scaling formula derived from the CRT uniqueness condition via the Cauchy--Schwarz inequality. The proposed formula is scale-invariant by construction, guarantees that the CRT uniqueness condition is always satisfied, and introduces no additional overhead over the original fast mode. Experiments on an NVIDIA GH200 GPU show that the proposed method achieves accuracy comparable to that of accurate mode while maintaining throughput comparable to that of fast mode. In the accuracy--throughput trade-off, the proposed method overcomes the accuracy limitation of fast mode and the throughput constraint of accurate mode, offering a superior accuracy and performance.

cs.MS↗

arXiv2026

Three-dimensional Fundamental Diagrams of Five-neighbor Particle Cellular Automata

We analyze five-neighbor particle cellular automata whose conventional two-dimensional fundamental diagrams are multivalued, but whose mean flow is uniquely determined by introducing a second density. We first consider binary rules for which the second density is conserved, and then examine rules for which the second density is not conserved but converges asymptotically. These examples give three-dimensional fundamental diagrams in which the mean flow is determined by the particle density and the second density. We then investigate whether this single-valued structure is preserved under real-valued max-plus extensions. There are some rules where two different max-plus extensions are introduced, and numerical simulations show that both extensions preserve the same single-valued three-dimensional fundamental diagram. These observations imply that, in constructing real-valued max-plus extensions, it is important to choose the flux function and the second density consistently.

physics.soc-ph↗

arXiv2026

Computing FFTs at Target Precision Using Lower-Precision FFTs

Modern processors deliver higher throughput for lower-precision arithmetic than for higher-precision arithmetic. For matrix multiplication, the Ozaki scheme exploits this performance gap by splitting the inputs into lower-precision components and delegating the computation to optimized lower-precision routines. However, no similar approach exists for the fast Fourier transform (FFT). Here, we propose a method that computes target-precision FFTs using lower-precision FFTs by applying the Ozaki scheme to the cyclic convolution in the Bluestein FFT. The split component convolutions are computed exactly using the number theoretic transform (NTT), an FFT over a finite field, instead of floating-point FFTs, combined with the Chinese remainder theorem. We introduce an upper bound on the number of splits and an NTT-domain accumulation strategy to reduce the NTT call count. As a concrete implementation, we implement a double-precision FFT using 32-bit NTTs and confirm reduced relative error compared with those for FFTs based on FFTW and Triple-Single precision arithmetic, with stable error across FFT lengths, at most 96 NTT calls, or 64 NTT calls with NTT-domain accumulation. On an Intel Xeon Platinum 8468 for lengths $n=2^{10}$-$2^{18}$, the execution time is approximately 107-1315$\times$ that of FFTW's double-precision FFT, with NTTs accounting for approximately 80% of the total time.

cs.MS↗

arXiv2024

Counterfactual Explanations of Black-box Machine Learning Models using Causal Discovery with Applications to Credit Rating

Explainable artificial intelligence (XAI) has helped elucidate the internal mechanisms of machine learning algorithms, bolstering their reliability by demonstrating the basis of their predictions. Several XAI models consider causal relationships to explain models by examining the input-output relationships of prediction models and the dependencies between features. The majority of these models have been based their explanations on counterfactual probabilities, assuming that the causal graph is known. However, this assumption complicates the application of such models to real data, given that the causal relationships between features are unknown in most cases. Thus, this study proposed a novel XAI framework that relaxed the constraint that the causal graph is known. This framework leveraged counterfactual probabilities and additional prior information on causal structure, facilitating the integration of a causal graph estimated through causal discovery methods and a black-box classification model. Furthermore, explanatory scores were estimated based on counterfactual probabilities. Numerical experiments conducted employing artificial data confirmed the possibility of estimating the explanatory score more accurately than in the absence of a causal graph. Finally, as an application to real data, we constructed a classification model of credit ratings assigned by Shiga Bank, Shiga prefecture, Japan. We demonstrated the effectiveness of the proposed method in cases where the causal graph is unknown.

cs.LG↗

arXiv2023

Asymptotic solution to a fuzzy elementary cellular automaton of rule number 38

Fuzzy cellular automaton is a dynamical system with a continuous state value embedding a cellular automaton with a discrete state value. We investigate a fuzzy cellular automaton obtained from an elementary cellular automaton of rule number 38. Its asymptotic solutions are classified into two types. One is a solution where stable propagating waves exist, and the other is a static uniform solution of constant value.

nlin.CG↗

arXiv2021

Three-dimensional fundamental diagram of particle system of 5 neighbors with two conserved densities

We discuss a particle system of 5 neighbors with two independent conserved densities. The mean momentum uniquely depends on a pair of the densities and a three-dimensional fundamental diagram is obtained. It shows the phase transition of behavior of asymptotic solution to the system. Moreover, we propose two other systems which have the similar unique dependency obtained numerically.

math-ph↗

arXiv2021

Lattice equations and their solutions with complexity of polynomial class

We discuss initial value problems for time evolution equations in one dimensional space which are expressed by the lattice operators and propose some new equations to which complexity of solutions is of polynomial class. Novel type of expressions using shift operators and binary trees are applied for the derivation of solution.

nlin.SI↗

arXiv2019

Max-Plus Generalization of Conway's Game of Life

We propose a max-plus equation which includes Conway's Game of Life (GoL) as a special case. There are some special solutions to the equation which include and unify those to GoL. Moreover, the multi-value extension of GoL is derived from the equation and the behavior of solutions is discussed.

nlin.CG↗

arXiv2017

Effect of Rotation on Elastic Moduli of Solid $^4$He

We report measurements of elastic moduli of hcp solid $^4$He down to 15 mK when the samples are rotated unidirectionally. Recent investigations have revealed that the elastic behavior of solid $^4$He is dominated by gliding of dislocations and pinning of them by $^3$He impurities, which move in the solid like Bloch waves (impuritons). Motivated by the recent controversy of torsional oscillator studies, we have preformed direct measurements of shear and Young's moduli of annular solid $^4$He using pairs of quarter-circle shape piezoelectric transducers (PZTs) while the whole apparatus is rotated with angular velocity $Ω$ up to 4 rad/s. We have found that shear modulus $μ$ is suppressed by rotation below 80 mK, when shear strain applied by PZT exceeds a critical value, above which $μ$ decreases because the shear strain unbinds dislocations from $^3$He impurities. The rotation - induced decrement of $μ$ at $Ω= 4$ rad/s is about 14.7 (12.3) % of the total change of temperature dependent $μ$ for solid samples of pressure 3.6 (5.4) MPa. The decrements indicate that the probability of pinning of $^3$He on dislocation segment, $G$, decreases by several orders of magnitude. We propose that the motion of $^3$He impuritons under rotation becomes strongly anisotropic by the Coriolis force, resulting a decrease in $G$ for dislocation lines aligning parallel to the rotation axis.

cond-mat.other↗

arXiv2017

Superfluid-like TO Responses in Rotating Solid Helium

The irrotational nature of superfluid helium was discovered through its decoupling from the container under rotation. Similarly, the resonant period drop of a torsional oscillator (TO) containing solid helium was first interpreted as the decoupling of solid from the TO and appearance of supersolid. However, the resonant period can be changed by mechanisms other than supersolid, such as the elastic stiffening of solid helium that is widely accepted as the reason for the TO response. To demonstrate the irrotational nature more directly, the previous experiments superimposed the dc rotation onto the TO and revealed strong suppression on the TO response without affecting the shear modulus. This result is inconsistent with the simple temperature-dependent elasticity model and supports the supersolid scenario. Here, we re-examine the rotational effect on solid helium with a two-frequency rigid TO to clarify the conflicting observations. Surprisingly, most of the result of previous rotation experiments were not reproduced. Instead, we found a very interesting superfluid-like irrotational response that cannot be explained by elastic models.

cond-mat.other↗

arXiv2013

Max-min-plus expressions for one-dimensional particle cellular automata obtained from a fundamental diagram

We study one-dimensional neighborhood-five conservative cellular automata (CA), referred to as particle cellular automata five (particle CA5). We show that evolution equations for particle CA5s that belong to certain types can be obtained in the form of max-min-plus expressions from a fundamental diagram. The obtained equations are transformed into other max-min-plus expressions by ultradiscrete Cole-Hopf transformation, which enable us to analyze the asymptotic behaviors of general solutions. The equations in the Lagrange representation, which describe particle motion, are also presented, which too can be obtained from a fundamental diagram. Finally, we discuss the generalization to a one-dimensional conservative neighborhood-$n$ CA, i.e., particle CA$n$.

nlin.CG↗

arXiv2013

Initial value problem of evolution equations defined by lattice operators

We propose dynamical systems defined on algebra of lattices, which we call `lattice equations'. We give exact general solutions of initial value problems for a class of lattice equations, and evaluate the complexity of the solutions. Moreover we discuss the relationship between those equations and binary cellular automata.

nlin.SI↗

arXiv2012

Constraints on a class of classical solutions in open string field theory

We calculate boundary states for general string fields in the KBc subalgebra under some regularity conditions based on the construction by Kiermaier, Okawa, and Zwiebach. The resulting boundary states are always proportional to that for the perturbative vacuum |B>. In this framework, the equation of motion implies that boundary states are independent of the auxiliary parameter s associated with the length of the boundary. By requiring the s-independence, we show that the boundary states for classical solutions in our class are restricted to \pm|B> and 0. In particular, there exist no string fields which reproduce boundary states for multiple D-brane backgrounds. While we know that the boundary states |B> and 0 are reproduced by solutions for the perturbative vacuum and the tachyon vacuum, respectively, no solutions reproducing -|B> have been constructed. In this paper we also propose a candidate for such a solution, which may describe the ghost D-brane.

hep-th↗

arXiv2011

The boundary state for a class of analytic solutions in open string field theory

We construct a boundary state for a class of analytic solutions in the Witten's open string field theory. The result is consistent with the property of the zero limit of a propagator's length, which was claimed in [19]. And we show that our boundary state becomes expected one for the perturbative vacuum solution and the tachyon vacuum solution. We also comment on possible presence of multi-brane solutions and ghost brane solutions from our boundary state.

hep-th↗

arXiv2010

Max-plus analysis on some binary particle systems

We concern with a special class of binary cellular automata, i.e., the so-called particle cellular automata (PCA) in the present paper. We first propose max-plus expressions to PCA of 4 neighbors. Then, by utilizing basic operations of the max-plus algebra and appropriate transformations, PCA4-1, 4-2 and 4-3 are solved exactly and their general solutions are found in terms of max-plus expressions. Finally, we analyze the asymptotic behaviors of general solutions and prove the fundamental diagrams exactly.

nlin.SI↗

arXiv2010

Ultradiscrete Plücker Relation Specialized for Soliton Solutions

We propose an ultradiscrete analogue of Plücker relation specialized for soliton solutions. It is expressed by an ultradiscrete permanent which is obtained by ultradiscretizing the permanent, that is, the signature-free determinant. Using this relation, we also show soliton solutions to the ultradiscrete KP equation and the ultradiscrete two-dimensional Toda lattice equation respectively.

nlin.SI↗

arXiv2010

Bilinear Equations and Bäcklund Transformation for Generalized Ultradiscrete Soliton Solution

Ultradiscrete soliton equations and Bäcklund transformation for a generalized soliton solution are presented. The equations include the ultradiscrete KdV equation or the ultradiscrete Toda equation in a special case. We also express the solution by the ultradiscrete permanent, which is defined by ultradiscretizing the signature-free determinant, that is, the permanent. Moreover, we discuss a relation between Bäcklund transformations for discrete and ultradiscrete KdV equations.

nlin.SI↗

arXiv2009

Exact Results for Tunneling Problems of Bogoliubov Excitations in the Critical Supercurrent State

We show the exact solution of Bogoliubov equations at zero-energy in the critical supercurrent state for arbitrary shape of potential barrier. With use of this solution, we prove the absence of perfect transmission of excitations in the low-energy limit by giving the explicit expression of transmission coefficient. The origin of disappearance of perfect transmission is the emergence of zero-energy density fluctuation near the potential barrier.

cond-mat.quant-gas↗