• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to previous page

Fusion of Iterative Solvers for Linear Equations with Sparse Matrices and Verified Numerical Computation Methods

Research Project

Project/Area Number 20H04195
Research Category

Grant-in-Aid for Scientific Research (B)

Allocation TypeSingle-year Grants
Section一般
Review Section Basic Section 60090:High performance computing-related
Research InstitutionShibaura Institute of Technology

Principal Investigator

Ozaki Katsuhisa  芝浦工業大学, システム理工学部, 教授 (90434282)

Co-Investigator(Kenkyū-buntansha) 荻田 武史  東京女子大学, 現代教養学部, 教授 (00339615)
相原 研輔  東京都市大学, 情報工学部, 准教授 (70735498)
Project Period (FY) 2020-04-01 – 2023-03-31
Project Status Completed (Fiscal Year 2023)
Budget Amount *help
¥16,900,000 (Direct Cost: ¥13,000,000、Indirect Cost: ¥3,900,000)
Fiscal Year 2022: ¥5,980,000 (Direct Cost: ¥4,600,000、Indirect Cost: ¥1,380,000)
Fiscal Year 2021: ¥7,280,000 (Direct Cost: ¥5,600,000、Indirect Cost: ¥1,680,000)
Fiscal Year 2020: ¥3,640,000 (Direct Cost: ¥2,800,000、Indirect Cost: ¥840,000)
Keywords精度保証付き数値計算 / 連立一次方程式 / 反復解法 / 数値線形代数 / 高性能計算
Outline of Research at the Start

数値線形代数は,物理現象,社会現象のシミュレーションを含む非常に多くの分野に応用があり,科学技術計算に必要不可欠である.特に,疎行列を係数行列とする連立一次方程式は多くの問題に現れ,主に反復解法を用いて近似解を得る.従来の反復解法では,相対残差を基準として収束判定を行うことが多く,「収束しない」,または「不正確な結果を得る」など,精度面に問題がある事例がある.また精度の検証に関する第二の選択肢として,複数の近似解法において得た近似解を比較する方法がある.本研究では,精度の検証に関する第三の選択肢として「誤差」に着目した精度保証付き数値計算を展開し,高速・高信頼数値計算の基盤研究を行う.

Outline of Final Research Achievements

To enable verified numerical computations for iterative solvers of systems of linear equations with sparse coefficient matrices, we precomputed and published the maximum norm of the inverse matrices from the SuiteSparse Matrix Collection. This allows verified numerical computations to obtain tight error bounds for many problems within approximately twice the time of approximate computations (including the time for approximate computations). Additionally, we developed a mixed-precision iterative solver for systems of linear equations that does not require complex procedures to obtain very good approximate solutions from an error perspective, and demonstrated its usefulness through numerical experiments.

Academic Significance and Societal Importance of the Research Achievements

SuiteSparse Matrix Collectionにある行列は、科学技術計算において具体的に現れる行列であり、反復解法の有用性を評価するためによく使用されてきた。通常、相対残差ノルムを用いて近似解の良し悪しを議論してきたが、相対誤差の観点から精度を議論できるようになり、新たな視点で反復解法を評価し、より深い精度の議論が可能となった。反復解法は多くの分野のシミュレーションにおいて必須であり、この研究で可能としたこと、提案した反復解法は将来のシミュレーション技術の発展に寄与するものである。

Report

(4 results)
  • 2023 Final Research Report ( PDF )
  • 2022 Annual Research Report
  • 2021 Annual Research Report
  • 2020 Annual Research Report
  • Research Products

    (47 results)

All 2024 2023 2022 2021 2020 Other

All Journal Article (10 results) (of which Int'l Joint Research: 1 results,  Peer Reviewed: 10 results,  Open Access: 7 results) Presentation (35 results) (of which Int'l Joint Research: 10 results,  Invited: 1 results) Remarks (2 results)

  • [Journal Article] Mixed-precision conjugate gradient algorithm using the groupwise update strategy2024

    • Author(s)
      Kensuke Aihara, Katsuhisa Ozaki, Daichi Mukunoki
    • Journal Title

      Jpn. J. Ind. Appl. Math.

      Volume: 41 Issue: 2 Pages: 837-855

    • DOI

      10.1007/s13160-024-00644-8

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] DGEMM on integer matrix multiplication unit2024

    • Author(s)
      Hiroyuki Ootomo, Katsuhisa Ozaki, Rio Yokota
    • Journal Title

      The International Journal of High Performance Computing Applications

      Volume: to appear Issue: 4 Pages: 297-313

    • DOI

      10.1177/10943420241239588

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Inclusion Methods for Multiplication of Three Point Matrices2023

    • Author(s)
      Yuki Uchino, Katsuhisa Ozaki, Takeshi Terao
    • Journal Title

      Journal of Advanced Simulation in Science and Engineering

      Volume: 10 Issue: 1 Pages: 83-101

    • DOI

      10.15748/jasse.10.83

    • ISSN
      2188-5303
    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Acceleration of iterative refinement for singular value decomposition2023

    • Author(s)
      Uchino Yuki、Terao Takeshi、Ozaki Katsuhisa
    • Journal Title

      Numerical Algorithms

      Volume: 95 Issue: 2 Pages: 979-1009

    • DOI

      10.1007/s11075-023-01596-9

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Generation of test matrices with specified eigenvalues using floating-point arithmetic2022

    • Author(s)
      Ozaki Katsuhisa、Ogita Takeshi
    • Journal Title

      Numerical Algorithms

      Volume: 90 Issue: 1 Pages: 241-262

    • DOI

      10.1007/s11075-021-01186-7

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] 実対称固有値分解に対する反復改良法の高速化2022

    • Author(s)
      内野 佑基,尾崎 克久,荻田 武史
    • Journal Title

      情報処理学会論文誌,コンピューティングシステム

      Volume: 15 Pages: 1-12

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Verified Numerical Computations for Searching for Vectors with the Maximum Sum2021

    • Author(s)
      Uchino Yuki、Ozaki Katsuhisa
    • Journal Title

      Journal of Advanced Simulation in Science and Engineering

      Volume: 8 Issue: 1 Pages: 53-72

    • DOI

      10.15748/jasse.8.53

    • NAID

      130007995882

    • ISSN
      2188-5303
    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Performance and energy consumption of accurate and mixed-precision linear algebra kernels on GPUs2020

    • Author(s)
      Mukunoki, D., Ogita, T.
    • Journal Title

      Journal of Computational and Applied Mathematics

      Volume: 372 Pages: 1-12

    • DOI

      10.1016/j.cam.2019.112701

    • Related Report
      2020 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Modified error bounds for approximate solutions of dense linear systems2020

    • Author(s)
      Minamihata Atsushi、Ogita Takeshi、Rump Siegfried M.、Oishi Shin’ichi
    • Journal Title

      Journal of Computational and Applied Mathematics

      Volume: 369 Pages: 112546-112546

    • DOI

      10.1016/j.cam.2019.112546

    • Related Report
      2020 Annual Research Report
    • Peer Reviewed / Int'l Joint Research
  • [Journal Article] Iterative refinement for singular value decomposition based on matrix multiplication2020

    • Author(s)
      T. Ogita, K. Aishima
    • Journal Title

      Journal of Computational and Applied Mathematics

      Volume: 369 Pages: 112512-112512

    • DOI

      10.1016/j.cam.2019.112512

    • Related Report
      2020 Annual Research Report
    • Peer Reviewed
  • [Presentation] Groupwise更新戦略を用いた混合精度CG法の有効性2024

    • Author(s)
      相原 研輔,尾崎 克久,椋木 大地
    • Organizer
      日本応用数理学会2024年研究部会連合発表会
    • Related Report
      2022 Annual Research Report
  • [Presentation] Krylov部分空間法に対する右辺ベクトルの追加による収束性の改善2023

    • Author(s)
      海老澤 勇一郎,相原 研輔
    • Organizer
      第21回計算数学研究会
    • Related Report
      2022 Annual Research Report
  • [Presentation] 連立一次方程式に対する可変的前処理付きglobal CG法の収束性2023

    • Author(s)
      山本 実央,相原 研輔
    • Organizer
      第21回計算数学研究会
    • Related Report
      2022 Annual Research Report
  • [Presentation] Level 3 BLASに対する尾崎スキームの設計2023

    • Author(s)
      内野 佑基, 尾崎 克久
    • Organizer
      日本応用数理学会第19回研究部会連合発表会
    • Related Report
      2022 Annual Research Report
  • [Presentation] 浮動小数点数で表現される領域の境界上の節点の計算法2023

    • Author(s)
      尾崎 克久, 劉 雪峰
    • Organizer
      日本応用数理学会第19回研究部会連合発表会
    • Related Report
      2022 Annual Research Report
  • [Presentation] Lanczos型反復法における安定化多項式の発展と初期シャドウ残差の選択2023

    • Author(s)
      相原 研輔
    • Organizer
      RIMS共同研究(公開型)新時代における高性能科学技術計算法の探究
    • Related Report
      2022 Annual Research Report
    • Invited
  • [Presentation] Numerical studies on preconditioned iterative solvers with minimal residual smoothing2023

    • Author(s)
      Arisa Kawase, Kensuke Aihara
    • Organizer
      The 42nd JSST Annual International Conference on Simulation Technology
    • Related Report
      2022 Annual Research Report
    • Int'l Joint Research
  • [Presentation] A mixed-precision algorithm of the CG method using the group-wise update strategy2022

    • Author(s)
      Kensuke Aihara, Katsuhisa Ozaki, Daichi Mukunoki
    • Organizer
      The 41st JSST Annual International Conference on Simulation Technology
    • Related Report
      2022 Annual Research Report
    • Int'l Joint Research
  • [Presentation] 3つの点行列の積の包含法とその解析について2022

    • Author(s)
      内野 佑基, 尾崎 克久, 寺尾 剛史
    • Organizer
      日本応用数理学会2022年度年会
    • Related Report
      2022 Annual Research Report
  • [Presentation] 数値線形代数の諸問題に対する真の解がわかるテスト行列の生成法2022

    • Author(s)
      尾崎 克久, 寺尾 剛史, 内野 佑基
    • Organizer
      日本応用数理学会2022年度年会
    • Related Report
      2022 Annual Research Report
  • [Presentation] 不等分割による行列積のエラーフリー変換の高精度計算への応用2022

    • Author(s)
      尾崎 克久, 椋木 大地, 荻田 武史
    • Organizer
      日本応用数理学会2022年度年会
    • Related Report
      2022 Annual Research Report
  • [Presentation] 固有値・特異値分解の混合精度数値計算法の性能評価2022

    • Author(s)
      内野 佑基, 寺尾 剛史, 尾崎 克久
    • Organizer
      2022年並列/分散/協調処理に関するサマー・ワークショップ
    • Related Report
      2022 Annual Research Report
  • [Presentation] Flying restart付きCG法に対する混合精度演算による近似解精度の向上2022

    • Author(s)
      相原 研輔,尾崎 克久, 椋木 大地
    • Organizer
      日本応用数理学会第18回研究部会連合発表会
    • Related Report
      2021 Annual Research Report
  • [Presentation] 特異値分解に対する反復改良法の高速化2022

    • Author(s)
      内野 佑基,尾崎 克久,寺尾 剛史
    • Organizer
      日本応用数理学会第18回研究部会連合発表会
    • Related Report
      2021 Annual Research Report
  • [Presentation] 実行列の全特異値に対する効率的な精度保証法2022

    • Author(s)
      寺尾 剛史,内野 祐基,尾崎 克久,今村 俊幸
    • Organizer
      日本応用数理学会第18回研究部会連合発表会
    • Related Report
      2021 Annual Research Report
  • [Presentation] 行列積に対する試行型エラーフリー変換に対する誤差の対処法とその応用2022

    • Author(s)
      尾崎 克久,椋木 大地,荻田 武史
    • Organizer
      日本応用数理学会第18回研究部会連合発表会
    • Related Report
      2021 Annual Research Report
  • [Presentation] Rayleigh quotient and accurate matrix multiplication for eigenvalue problems2022

    • Author(s)
      K. Ozaki, D. Mukunoki, T. Ogita
    • Organizer
      International Workshop on Reliable Computing and Computer-Assisted Proofs (ReCAP 2022)
    • Related Report
      2021 Annual Research Report
    • Int'l Joint Research
  • [Presentation] 精度保証付き数値計算の数値再現性への貢献について2021

    • Author(s)
      尾崎 克久
    • Organizer
      日本応用数理学会「行列・固有値問題の解法とその応用」研究部会 第32回研究会
    • Related Report
      2021 Annual Research Report
  • [Presentation] 行列積のエラーフリー変換に関する基礎・応用とMATLABコード生成のチュートリアル2021

    • Author(s)
      尾崎 克久
    • Organizer
      第5回 精度保証付き数値計算の実問題への応用研究集会
    • Related Report
      2021 Annual Research Report
  • [Presentation] 実対称固有値分解に対する反復改良法の高速化2021

    • Author(s)
      内野 佑基, 尾崎 克久, 荻田 武史
    • Organizer
      日本応用数理学会2021年度年会
    • Related Report
      2021 Annual Research Report
  • [Presentation] 2ステップ型反復改良法を用いた連立1次方程式の高精度計算法2021

    • Author(s)
      寺尾 剛史,尾崎 克久,今村 俊幸
    • Organizer
      日本応用数理学会2021年度年会
    • Related Report
      2021 Annual Research Report
  • [Presentation] 行列積に対するエラーフリー変換の多倍長精度計算への応用について2021

    • Author(s)
      尾崎 克久
    • Organizer
      日本応用数理学会2021年度年会
    • Related Report
      2021 Annual Research Report
  • [Presentation] 高速かつ高精度な行列積計算を用いた固有値と特異値の計算法2021

    • Author(s)
      尾崎 克久, 荻田 武史, 椋木 大地
    • Organizer
      2021年並列/分散/協調処理に関するサマー・ワークショップ (SWoPP2021)
    • Related Report
      2021 Annual Research Report
  • [Presentation] 低精度計算を活用した線形方程式に対する残差反復法の改良2021

    • Author(s)
      寺尾 剛史,尾崎 克久,今村 俊幸
    • Organizer
      2021年並列/分散/協調処理に関するサマー・ワークショップ (SWoPP2021)
    • Related Report
      2021 Annual Research Report
  • [Presentation] Error-free transformation of matrix multiplication for multi-precision computations2021

    • Author(s)
      K. Ozaki
    • Organizer
      The 19th International Symposium on Scientific Computing, Computer Arithmetic, and Verified Numerical Computations (SCAN2020)
    • Related Report
      2021 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Acceleration of Iterative Refinement for Symmetric Eigenvalue Decomposition2021

    • Author(s)
      Y. Uchino, K. Ozaki, T. Ogita
    • Organizer
      The 40th JSST Annual International Conference on Simulation Technology (JSST2021)
    • Related Report
      2021 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Verified Numerical Computations for a Standard Eigenvalue Problem Without Directed Rounding2021

    • Author(s)
      T. Terao, K. Ozaki, T. Ogita, T. Imamura
    • Organizer
      The 40th JSST Annual International Conference on Simulation Technology (JSST2021)
    • Related Report
      2021 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Acceleration of Error-Free Transformation of Matrix Multiplication using GPU Tensor Cores2021

    • Author(s)
      K. Ozaki, D. Mukunoki, T. Ogita
    • Organizer
      The 40th JSST Annual International Conference on Simulation Technology (JSST2021)
    • Related Report
      2021 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Interval Matrix Multiplication Using Fast Low-Precision Arithmetic on GPU2021

    • Author(s)
      K. Ozaki, T. Ogita, D. Mukunoki
    • Organizer
      The theme of the 9th International Workshop on Reliable Engineering Computing (REC2021)
    • Related Report
      2021 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Efficient Parallel Multigrid Methods on Manycore Clusters with Double/Single Precision Computing2021

    • Author(s)
      K. Nakajima, T. Ogita, M. Kawai
    • Organizer
      2021 IEEE International Parallel and Distributed >> Processing Symposium Workshops (IPDPSW)
    • Related Report
      2021 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Iterative Refinement and Verified Numerical Linear Algebra2021

    • Author(s)
      T. Ogita
    • Organizer
      SIAM Conference on Applied Linear Algebra (LA21)
    • Related Report
      2021 Annual Research Report
    • Int'l Joint Research
  • [Presentation] 疎行列に対する逆行列のノルムデータとその活用法2021

    • Author(s)
      尾崎 克久, 荻田 武史
    • Organizer
      日本応用数理学会第17回研究部会連合発表会
    • Related Report
      2020 Annual Research Report
  • [Presentation] エラーフリー変換を用いた悪条件行列の生成法2021

    • Author(s)
      尾崎 克久, 荻田 武史
    • Organizer
      日本応用数理学会第17回研究部会連合発表会
    • Related Report
      2020 Annual Research Report
  • [Presentation] 高精度行列積を用いて固有ベクトル行列から固有値を計算する方法について2020

    • Author(s)
      尾崎 克久,荻田 武史,椋木 大地
    • Organizer
      「行列・固有値問題の解法とその応用」研究部会 第30回研究会
    • Related Report
      2020 Annual Research Report
  • [Presentation] GPUの単精度演算・Tensorコアを用いた行列積のエラーフリー変換2020

    • Author(s)
      尾崎 克久,椋木 大地,荻田武史
    • Organizer
      日本応用数理学会年会
    • Related Report
      2020 Annual Research Report
  • [Remarks]

    • URL

      https://www.mathsci.shibaura-it.ac.jp/ozaki/smc_norminf.html

    • Related Report
      2021 Annual Research Report
  • [Remarks] 行列の逆行列の最大値ノルムの上限

    • URL

      https://www.mathsci.shibaura-it.ac.jp/ozaki/smc_norminf.html

    • Related Report
      2020 Annual Research Report

URL: 

Published: 2020-04-28   Modified: 2025-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi