FINDING COMPLEX-VALUED SOLUTIONS TO THE BRENT EQUATIONS BY REDUCING THEM TO A NONLINEAR LEAST SQUARES PROBLEM

封面

如何引用文章

全文:

开放存取 开放存取
受限制的访问 ##reader.subscriptionAccessGranted##
受限制的访问 订阅或者付费存取

详细

Finding nontrivial solutions to the trilinear Brent equations corresponds to the construction of asymptotically fast matrix multiplication algorithms is an important, but in general a very difficult computational task. Methods of parameterization of the Brent equations based on the use of symmetries of the matrix product tensor are proposed, which make it possible to repeatedly reduce the dimension of the problem. The numerical solution of the obtained trilinear or cubic systems of nonlinear equations is carried out by reducing to a nonlinear least squares problem and applying to it a specially developed iterative method that does not require calculation of derivatives. The found solutions of the parameterized Brent equations, as a rule, have a rank no higher (and sometimes even lower) than the known results. Thus, an algorithm for multiplying two 4th-order matrices in 48 active multiplications is obtained.

作者简介

I. Kaporin

Federal Research Center Computer Science and Control of the Russian Academy of Sciences

Email: igorkaporin@mail.ru
Moscow, Russia

参考

  1. Brent R. P. Algorithms for matrix multiplication. (Report No. STAN-CS-70-157). Stanford Univ. CA Dept. of Computer Science, 1970, 58 p.
  2. Strassen V. Gaussian elimination is not optimal //Numer. Math. 13, 354—356 (1969).
  3. Laderman J. D. A noncommutative algorithm for multiplying 3x3 matrices using 23 multiplications. Bull. Amer. Math. Soc. 82, 126-128 (1976).
  4. Pan V. How we can speed up matrix multiplication? // SIAM Review, 26(3), 393-415 (1984).
  5. Smirnov A. V. The bilinear complexity and practical algorithms for matrix multiplication // Comp. Math. Mathem. Phys. 53, 1781-1785 (2013).
  6. Karstadt E., Schwartz O. Matrix multiplication, a little faster // J. of the ACM (JACM). 2020, 67(1), 1-31.
  7. Kaporin I. A Derivative-Free Nonlinear Least Squares Solver. In: Olenev N.N., Evtushenko Y.G., Jacimovic M., Khachay M., Malkova V. (eds.) Optimization and Applications. OPTIMA 2021. Lecture Notes in Computer Science, V. 13078. P. 217-230. Springer, Cham. (2021). https://doi.org/10.1007/978-3-030-91059-4_16
  8. Oseledets I. V., Savostyanov D. V. Minimization methods for approximating tensors and their comparison // Computational Mathematics and Mathematical Physics, 2006, 46(10), 1641-1650
  9. Kaporin I. E., Axelsson O. On a class of nonlinear equation solvers based on the residual norm reduction over a sequence of affine subspaces // SIAM J. Sci. Comput. 16(1), 228-249 (1994)
  10. Kaporin I. Preconditioned Subspace Descent Method for Nonlinear Systems of Equations // Open Computer Science, 2020, 10(1), 71-81
  11. Kozak D., Molinari C., Rosasco L., Tenorio L., Villa S. Zeroth order optimization with orthogonal random directions. // Mathematical Programming. 2022. V. 199. P. 1-41.
  12. Armijo L. Minimization of functions having Lipschitz continuous first partial derivatives // Pacific Journal of mathematics 16(1), 1-3 (1966).
  13. Ballard G., Ikenmeyer C., Landsberg J. M., Ryder N. The geometry of rank decompositions of matrix multiplication II: 3x3 matrices. // Journal of Pure and Applied Algebra, 2019, 223(8), 3205-3224.
  14. Berger G. O., Absil P. A., De Lathauwer L., Jungers R. M., Van Barel M. Equivalent polyadic decompositions of matrix multiplication tensors // J. of Computational and Applied Mathematics, 2022, 406, 113941. https://doi.org/10.1016/j.cam.2021.113941.
  15. Hopcroft J. E., Kerr L. R. On minimizing the number of multiplications necessary for matrix multiplication // SIAM Journal on Applied Mathematics, 1971, 20(1):30—36.
  16. Fawzi A. et al. Discovering faster matrix multiplication algorithms with reinforcement learning. Nature, 2022, 610(7930): 47-53.

补充文件

附件文件
动作
1. JATS XML
下载

版权所有 © Russian Academy of Sciences, 2024