A randomized operator splitting scheme inspired by stochastic optimization methods

被引:3
作者
Eisenmann, Monika [1 ]
Tony, Stillfjord [1 ]
机构
[1] Lund Univ, Ctr Math Sci, POB 118, S-22100 Lund, Sweden
基金
瑞典研究理事会;
关键词
65C99; 65M12; 90C15; 65M55; SCALE-SPACE METHODS; ITERATIVE REGULARIZATION; THRESHOLDING ALGORITHM; BREGMAN ITERATIONS; INVERSE; CONVERGENCE; SPARSITY; DESCENT; INEXACT;
D O I
10.1007/s00211-024-01396-w
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
In this paper, we combine the operator splitting methodology for abstract evolution equations with that of stochastic methods for large-scale optimization problems. The combination results in a randomized splitting scheme, which in a given time step does not necessarily use all the parts of the split operator. This is in contrast to deterministic splitting schemes which always use every part at least once, and often several times. As a result, the computational cost can be significantly decreased in comparison to such methods. We rigorously define a randomized operator splitting scheme in an abstract setting and provide an error analysis where we prove that the temporal convergence order of the scheme is at least 1/2. We illustrate the theory by numerical experiments on both linear and quasilinear diffusion problems, using a randomized domain decomposition approach. We conclude that choosing the randomization in certain ways may improve the order to 1. This is as accurate as applying e.g. backward (implicit) Euler to the full problem, without splitting.
引用
收藏
页码:435 / 461
页数:27
相关论文
共 38 条
[1]   Fast/slow diffusion and growing sandpiles [J].
Aronsson, G ;
Evans, LC ;
Wu, Y .
JOURNAL OF DIFFERENTIAL EQUATIONS, 1996, 131 (02) :304-335
[2]   Randomized Runge-Kutta method - Stability and convergence under inexact information [J].
Bochacik, Tomasz ;
Gocwin, Maciej ;
Morkisz, Pawel M. ;
Przybylowicz, Pawel .
JOURNAL OF COMPLEXITY, 2021, 65
[3]   Optimization Methods for Large-Scale Machine Learning [J].
Bottou, Leon ;
Curtis, Frank E. ;
Nocedal, Jorge .
SIAM REVIEW, 2018, 60 (02) :223-311
[4]   On the randomized solution of initial value problems [J].
Daun, Thomas .
JOURNAL OF COMPLEXITY, 2011, 27 (3-4) :300-311
[5]  
Eisenmann M., 2019, Methods for the temporal approximation of nonlinear, nonautonomous evolution equations
[6]   Sub-linear convergence of a stochastic proximal iteration method in Hilbert space [J].
Eisenmann, Monika ;
Stillfjord, Tony ;
Williamson, Mans .
COMPUTATIONAL OPTIMIZATION AND APPLICATIONS, 2022, 83 (01) :181-210
[7]   A variational approach to the sum splitting scheme [J].
Eisenmann, Monika ;
Hansen, Eskil .
IMA JOURNAL OF NUMERICAL ANALYSIS, 2022, 42 (01) :923-950
[8]   On a Randomized Backward Euler Method for Nonlinear Evolution Equations with Time-Irregular Coefficients [J].
Eisenmann, Monika ;
Kovacs, Mihaly ;
Kruse, Raphael ;
Larsson, Stig .
FOUNDATIONS OF COMPUTATIONAL MATHEMATICS, 2019, 19 (06) :1387-1430
[9]   Convergence analysis of domain decomposition based time integrators for degenerate parabolic equations [J].
Eisenmann, Monika ;
Hansen, Eskil .
NUMERISCHE MATHEMATIK, 2018, 140 (04) :913-938
[10]  
Emmrich E., 2009, COMPUT METHODS APPL, V9, P37, DOI DOI 10.2478/CMAM-2009-0003