Quantum algorithm for the gradient of a logarithm-determinant

Jan 16, 2025
20 pages
e-Print:

Citations per year

0 Citations
Abstract: (arXiv)
The logarithm-determinant is a common quantity in many areas of physics and computer science. Derivatives of the logarithm-determinant compute physically relevant quantities in statistical physics models, quantum field theories, as well as the inverses of matrices. A multi-variable version of the quantum gradient algorithm is developed here to evaluate the derivative of the logarithm-determinant. From this, the inverse of a sparse-rank input operator may be determined efficiently. Measuring an expectation value of the quantum state--instead of all N2N^2 elements of the input operator--can be accomplished in O(kσ)O(k\sigma) time in the idealized case for kk relevant eigenvectors of the input matrix. A factor σ=1ε3\sigma=\frac1{\varepsilon^3} or 1ε2log2ε-\frac1{\varepsilon^2}\log_2\varepsilon depends on the version of the quantum Fourier transform used for a precision ε\varepsilon. Practical implementation of the required operator will likely need log2N\log_2N overhead, giving an overall complexity of O(kσlog2N)O(k\sigma\log_2 N). The method applies widely and converges super-linearly in kk when the condition number is high. For non-sparse-rank inputs, the algorithm can be evaluated provided that an equal superposition of eigenstates is provided. The output is given in O(1)O(1) queries of oracle, which is given explicitly here and only relies on time-evolution operators that can be implemented with arbitrarily small error. The algorithm is envisioned for fully error-corrected quantum computers but may be implementable on near-term machines. We discuss how this algorithm can be used for kernel-based quantum machine-learning.
Note:
  • 20 pages, 3 figures, 2 circuit diagrams, 1 table
  • [1]
    Statistical physics of fields (Cambridge University Press,)
    • M. Kardar
  • [2]
    Superlinear convergence of randomized block lanczos algorithm, in
    • Q. Yuan
      ,
    • M. Gu
      ,
    • B. Li
  • [2]
    International Conference on Data Mining (ICDM)
    • Q. Yuan
      ,
    • M. Gu
      ,
    • B. Li
  • [3]
    Convergence Analysis of Randomized Block Lanczos Algorithms for Low-Rank Matrix Approximations (University of California, Berkeley,)
    • Q. Yuan
  • [4]
    Randomized block krylov space methods for trace and log-determinant estimators, arXiv preprint
    • H. Li
      ,
    • Y. Zhu
  • [5]
    Bayesian inference of log determinants, arXiv preprint
    • J. Fitzsimons
      ,
    • K. Cutajar
      ,
    • M. Osborne
      ,
    • S. Roberts
      ,
    • M. Filippone
  • [6]
    Understanding neural networks with logarithm determinant entropy estimator, arXiv preprint
    • Z. Zhouyin
      ,
    • D. Liu
  • [7]
    Gravitation and cosmology: principles and applications of the general theory of relativity
    • S. Weinberg
  • [8]
    Quantum field theory (The Advanced Book Program, Perseus Books Reading, Massachusetts,)
    • M.E. Peskin
      ,
    • D.V. Schroeder
  • [10]
    A bayesian beta kernel model for binary classification and online learning problems, Statistical Analysis and Data Mining: The ASA
    • C.A. MacKenzie
      ,
    • T.B. Trafalis
      ,
    • K. Barker
      • Data Sci.J. 7 (2014) 434
  • [11]
    Fundamentals of statistical and thermal physics (Waveland Press,)
    • F. Reif
  • [12]
    Matrix inversion algorithms by means of automatic differentiation
    • K. Kubota
      • Appl.Math.Lett. 7 (1994) 19
  • [13]
    Numerical recipes in C++
    • W.H. Press
      ,
    • S.A. Teukolsky
      ,
    • W.T. Vetterling
      ,
    • B.P. Flannery
  • [14]
    An iteration method for the solution of the eigenvalue problem of linear differential and integral operators (United States Governm. Press Office Los Angeles, CA,)
    • C. Lanczos
  • [15]
    Lanczos recursion on a quantum computer for the Green’s function and ground state
    • T.E. Baker
      • Phys.Rev.A 103 (2021) 032404
  • [17]
    Improved analyses of the randomized power method and block lanczos method, arXiv preprint
    • S. Wang
      ,
    • Z. Zhang
      ,
    • T. Zhang
  • [18]
    On randomized lanczos algorithms, in Proceedings of theinternational symposium on Symbolic and algebraic computation pp. 176-183
    • W. Eberly
      ,
    • E. Kaltofen
  • [19]
    A randomized algorithm for principal component analysis
    • V. Rokhlin
      ,
    • A. Szlam
      ,
    • M. Tygert
      • Anal.Appl. 31 (2010) 1100
  • [20]
    An algorithm for the principal component analysis of large data sets
    • N. Halko
      ,
    • P.-G. Martinsson
      ,
    • Y. Shkolnisky
      ,
    • M. Tygert
      • SIAM J.Sci.Comput. 33 (2011) 2580
  • [21]
    Subspace iteration randomization and singular value problems
    • M. Gu
      • SIAM J.Sci.Comput. 37 (2015) A1139
  • [22]
    Randomized block krylov methods for approximating extreme eigenvalues
    • J.A. Tropp
      • Numer.Math. 150 (2022) 217
  • [23]
    Restarted block lanczos bidiagonalization methods
    • J. Baglama
      ,
    • L. Reichel
      • Num.Algorithms 43 (2006) 251
  • [24]
    Structural dynamics analysis using an unsymmetric block lanczos algorithm
    • H.M. Kim
      ,
    • R.R. Craig, Jr.
      • Int.J.Numer.Meth.Eng. 26 (1988) 2305