-
设A是一个m阶n维的实张量,即
其中[n]={1,2,…,n},[m]={1,2,…,m}.记R[m,n]为m阶n维实张量的全体构成的集合.
设Πm是由指标{1,2,…,m}的所有置换构成的集合.如果对Πm中的任意置换π,有
则称A为对称张量[1].
近年来,关于张量的研究受到了广泛的关注[2-10].由于张量Z-特征值在图匹配[11]、数据挖掘[12-13]和目标跟踪[14]等许多实际问题中有着重要应用,张量Z-特征值问题逐渐成为张量谱理论研究的热点问题.下面给出张量Z-特征值的定义.
定义1[1] 设A∈R[m,n]是实对称张量,如果存在λ∈ℝ和非零向量x=(x1,…,xn)T∈ℝn,使得
则称λ为A的一个Z-特征值,x为与λ相对应的Z-特征向量,称(λ,x)为A的一个Z-特征对,其中n维向量Axm-1的第i个分量为
文献[15]在研究基于张量算法的高阶图匹配时提出了包含边和超边的匹配模型,该模型可导出如下不同阶对称张量组特征值问题:
定义2 设Ai∈R[i,n]是实对称张量,i=2,3,如果存在λ∈ℝ和非零向量x∈ℝn,使得
则称λ为不同阶对称张量组{Ai}i=23的一个特征值,x为与λ相对应的特征向量,称(λ,x)是不同阶对称张量组{Ai}i=23的一个特征对.
本文讨论求解不同阶对称张量组特征值问题(1)的带位移高阶幂法,并给出其收敛性分析.
Shifted Symmetric Higher-Order Power Method for Computing Eigenvalues of Multiple Order Symmetric Tensors
-
摘要: 求解不同阶对称张量组的特征值和特征向量问题在超图匹配中具有重要的作用.首先,基于求解对称张量Z-特征值的带位移高阶幂法(SS-HOPM),利用系数张量组构造一个带位移因子的辅助函数,将求解不同阶对称张量组的特征值问题转化为求解辅助函数的极值点问题,提出了求解不同阶对称张量组特征值和特征向量的带位移高阶幂法.其次,利用凸函数的性质和单调有界原理,讨论了辅助函数的性质,确定了位移因子的取值范围,使得所给算法是收敛的.最后,通过数值算例对理论结果进行了验证,数值结果表明所提出的算法是有效的,并且该算法也能有效求出不同阶对称非半正定张量组的特征值和特征向量.Abstract: The problem of computing the eigenvalues and eigenvectors of multiple order symmetric tensors plays an important role in hyper-graph matching. Firstly, based on the shifted symmetric higher-order power method (SS-HOPM), we use the coefficient tensors to construct an auxiliary function with a shift factor, and transform the problem of solving the eigenvalues of multiple order symmetric tensors into one of solving extreme points of the auxiliary function. Thus we propose herein a shifted symmetric higher-order power method for computing eigenvalues and eigenvectors of multiple order symmetric tensors. Secondly, we utilize the properties of convex functions and the monotonic bounded principle to discuss the properties of the auxiliary function and determine the range of the shifted factor, which makes the given algorithm convergent. Finally, numerical examples are given to verify the theoretical results and show the effectiveness of the proposed algorithm, and the algorithm can also effectively compute the eigenvalues and eigenvectors of multiple order symmetric tensors that are not semi-positive definite.
-
表 1 例1不同阶对称张量组{A3,A2}的实特征对
λ xT 0.874 5 [-0.392 0 0.724 6 0.566 8] 0.431 4 [-0.718 8-0.125 8-0.683 7] 0.229 8 [-0.845 80.437 1-0.305 9] 0.019 4 [0.714 9 0.507 7-0.480 8] 0.004 3 [0.449 9 0.775 2-0.443 5] 0.002 8 [0.328 5 0.628 5-0.705 0] 0.001 4 [0.285 4 0.737 0-0.612 7] 0.000 3 [-0.296 0-0.734 9 0.610 1] -0.000 9 [-0.332 7-0.634 3 0.697 8] -0.002 2 [-0.445 5-0.772 7 0.452 2] -0.016 6 [-0.711 4-0.510 9 0.482 6] -0.229 1 [0.843 4-0.440 1 0.308 1] -0.429 8 [0.718 7 0.123 2 0.684 4] -0.871 5 [0.392 3-0.725 2-0.565 9] 表 2 例1算法2的数值结果
α Occurrences λ xT Iter 0 63 0.874 5 [-0.392 0 0.724 6 0.566 8] 16 37 0.431 4 [-0.718 8-0.125 8-0.683 7] 198 1 42 0.874 5 [-0.392 00.724 6 0.566 8] 31 28 0.431 4 [-0.718 8-0.125 8-0.683 7] 50 11 0.019 4 [0.714 9 0.507 7-0.480 8] 108 19 0.000 3 [-0.296 0-0.734 9 0.610 1] 147 9.357 7 38 0.874 5 [-0.392 0 0.724 6 0.566 8] 185 29 0.431 4 [-0.718 8-0.125 8-0.683 7] 334 21 0.019 4 [0.714 9 0.507 7-0.480 8] 953 12 0.000 3 [-0.296 0-0.734 9 0.610 1] 1 206 表 3 例2不同阶对称张量组{A3,A2}的实特征对
λ xT 2.750 1 [0.507 7-0.571 1-0.645 0] 0.785 9 [0.492 6 0.237 3 0.837 3] 0.120 1 [0.934 5-0.127 0 0.332 5] -0.853 3 [-0.448 8 0.762 8-0.465 5] -0.994 2 [-0.183 3 0.566 6-0.803 4] -1.291 8 [-0.495 8-0.506 5 0.705 4] -1.339 6 [-0.152 8-0.818 3 0.554 1] -2.653 1 [-0.689 3 0.469 2 0.552 0] 表 4 例2算法2的数值结果
α Occurrences λ xT Iter 0 65 2.750 1 [0.507 7-0.571 1-0.645 0] 28 35 - - 4 000 3 56 2.750 1 [0.507 7-0.571 1-0.645 0] 34 44 0.785 9 [0.492 6 0.237 30.837 3] 29 28.735 6 63 2.750 1 [0.507 7-0.571 1-0.645 0] 228 37 0.785 9 [0.492 6 0.237 30.837 3] 241 表 5 例3算法2的数值结果
α Occurrences λ Iter 0 100 - 4 000 2 47 1.157 9 201 53 1.157 5 202 28.803 1 55 1.157 9 1 550 45 1.157 5 1 535 -
[1] doi: https://www.polyu.edu.hk/ama/staff/new/qilq/qi-eigen-1.pdf QI L Q. Eigenvalues of a Real Supersymmetric Tensor[J]. Journal of Symbolic Computation, 2005, 40(6): 1302-1324. [2] doi: https://epubs.siam.org/doi/abs/10.1137/S0895479801387413 KOFIDIS E, REGALIA P A. On the Best Rank-1 Approximation of Higher-Order Supersymmetric Tensors[J]. SIAM Journal on Matrix Analysis and Applications, 2002, 23(3): 863-884. [3] COUR T, SRINIVASAN P, SHI J B. Balanced Graph Matching[M] //Advances in Neural Information Processing Systems. Cambridge: The MIT Press, 2007: 313-320. [4] 杨占英, 于云霞.高维张量积Meyer小波的一种新分解(英文)[J].西南大学学报(自然科学版), 2012, 34(2): 74-77. doi: http://xbgjxt.swu.edu.cn/article/id/jsunsz20120213 [5] doi: https://www.aimsciences.org/article/doi/10.3934/dcdsb.2017009 WANG G, ZHOU G L, CACCETTA L. Z-Eigenvalue Inclusion Theorems for Tensors[J]. Discrete and Continuous Dynamical Systemsseries, 2017, 22(1): 187-198. [6] doi: https://link.springer.com/article/10.1186/s13660-017-1428-6 WANG Y N, WANG G. Two S-Type Z-Eigenvalue Inclusion Sets for Tensors[J]. Journal of Inequalities and Applications, 2017, 2017: 152. [7] 刘蕊, 陈震, 刘奇龙.判定对称强H-张量的迭代算法[J].贵州师范大学学报(自然科学版), 2019, 37(3): 72-76. doi: http://www.cqvip.com/QK/96121A/20193/7001930030.html [8] 何军, 刘衍民.张量伪谱的新包含域[J].西南师范大学学报(自然科学版), 2019, 44(8): 7-10. [9] 钟琴.非负矩阵最大特征值的新界值[J].西南大学学报(自然科学版), 2018, 40(2): 40-43. doi: http://xbgjxt.swu.edu.cn/article/doi/10.13718/j.cnki.xdzk.2018.02.007 [10] 闫逸波, 陈震.一类张量方程的基于梯度的迭代方法[J].贵州师范大学学报(自然科学版), 2017, 35(3): 59-62. doi: http://www.cnki.com.cn/Article/CJFDTotal-NATR201703009.htm [11] doi: https://ui.adsabs.harvard.edu/abs/2017arXiv170200391D/abstract DUTTA A, LLADÓS J, BUNKE H, et al. Product Graph-Based Higher Order Contextual Similarities for Inexact Subgraph Matching[J]. Pattern Recognition, 2018, 76: 596-611. [12] doi: https://www.cs.cornell.edu/~arb/papers/3evec-SIMODS-2019.pdf BENSON A R. Three Hypergraph Eigenvector Centralities[J]. SIAM Journal on Mathematics of Data Science, 2019, 1(2): 293-312. [13] BENSON A R, GLEICH D F, LESKOVEC J. Tensor Spectral Clustering for Partitioning Higher-Order Network Structures[C] //Proceedings of the 2015 SIAM International Conference on Data Mining, Philadelphia, PA: Society for Industrial and Applied Mathematics, 2015: 118-126. [14] doi: https://link.springer.com/article/10.1007/s11263-018-01147-z SHI X C, LING H B, PANG Y, et al. Rank-1 Tensor Approximation for High-Order Association in Multi-Target Tracking[J]. International Journal of Computer Vision, 2019, 127(8): 1063-1083. [15] doi: https://ieeexplore.ieee.org/document/5871640 DUCHENNE O, BACH F, KWEON I S, et al. A Tensor-Based Algorithm for High-Order Graph Matching[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(12): 2383-2395. [16] doi: https://prod-ng.sandia.gov/techlib-noauth/access-control.cgi/2010/106131.pdf KOLDA T G, MAYO J R. Shifted Power Method for Computing Tensor Eigenpairs[J]. SIAM Journal on Matrix Analysis and Applications, 2011, 32(4): 1095-1124.
计量
- 文章访问数: 990
- HTML全文浏览数: 990
- PDF下载数: 186
- 施引文献: 0