報告人:黃榮 教授
報告題目:Threshold-Free Eaxct Rank Decay and Accurate Nested Range Subspace Tracking for Rank-Deficient Matrix Powers
報告時間:2025年12月20日(周六)下午4:00
報告地點:云龍校區6號樓304報告廳
主辦單位:數學與統計學院、數學研究院、科學技術研究院
報告人簡介:
黃榮,教授、博士生導師、湖南省杰出青年基金獲得者、湖南省芙蓉學者獎勵計劃獲得者、湖南省普通高校學科帶頭人、湖南省新世紀121人才工程人選、湖南省普通高校青年骨干教師。主要從事數值計算方面的研究工作,已(獨立)主持獲得2022-2023年度湖南省自然科學獎二等獎1項,以及主持獲得湖南省高等教育教學成果獎二等獎1項,擔任國際學術期刊《Numerical Algebra, Control and Optimization》編委等,已主持國家自然科學基金面上項目、國家自然科學基金青年項目、教育部博士點基金、湖南省杰出青年科學基金、中國博士后基金、湖南省教育廳重點項目、湖南省科技計劃項目等,研究成果全部以獨著或第一作者方式發表在Math. Comp.、SIAM. J. Matrix Anal. Appl.、J. Sci. Comput.、Adv. Comp. Math.、Appl. Numer. Math.、BIT、Numer. Linear Algebra Appl.等。
報告摘要:
This talk presents a threshold-free LU iteration method that employs Neville-type representations (NRs) to compute the generalized null space decomposition (GNSD). At each LU iteration step, the NR parametrization is updated via specialized updating/downdating algorithms with adaptive rank adjustments. One advantage of our approach is its avoidance of numerical thresholds and its reliance on subtraction-free arithmetic operations. This guarantees exact determination of the rank decay and stabilization index, and accurate computations of nested range subspaces. Consequently, the complete GNSD structure is accurately recovered: (i) the basis transformation matrices are accumulated in a subtraction-free manner, (ii) all zero Jordan blocks are exactly identified, and (iii) all nonzero eigenvalues are computed to high relative accuracy. Numerical experiments validate the high relative accuracy of our proposed method in handling structured and rank-deficient matrices, where conventional threshold-based schemes fail.