Vol 63, No 9 (2023)

A constructive algorithm to compute elimination L and duplication D matrices for the operation of P ⊗ P vectorization when P = PT is proposed. The matrix L, obtained according to such algorithm, allows one to form a vector that contains only unique elements of the mentioned Kronecker product. In its turn, the matrix D is for the inverse transformation. A software implementation of the procedure to compute the matrices L and D is developed. On the basis of the mentioned results, a new operation vecu(.) is defined for P ⊗ P in case P = PT and its properties are studied. The difference and advantages of the developed operation in comparison with the known ones vec(.) and vech(.) vecd(.)) in case of vectorization of P ⊗ P when P = PT are demonstrated. Using parameterization of the algebraic Riccati equation as an example, the efficiency of the operation vecu (.) to reduce overparameterization of the unknown parameter identification problem is shown.

В настоящей работе разработан численно-аналитический алгоритм оценки ошибок округления в равномерной метрике. Установлена их ограниченность на всем интервале вычисления вольт-амперных характеристик длинных джозефсоновских переходов при использовании предлагаемой схемы второго порядка точности. На примере системы двух разностных уравнений показано, как можно исследовать численно скорость роста ошибок округления в равномерной метрике в случае степенной неустойчивости. Кроме того, получены оценки скорости роста ошибок округления в раномерной метрике для схемы Русанова третьего порядка точности. Расчеты проводились на суперкомпьютере “Говорун” с использованием системы REDUCE. Библ. 9. Фиг. 6.

This paper studies non-smooth problems of convex stochastic optimization. Using the smoothing technique based on the replacement of the function value at the considered point by the averaged function value over a ball (in l1-norm or l2-norm) of a small radius centered at this point, and then the original problem is reduced to a smooth problem (whose Lipschitz constant of the gradient is inversely proportional to the radius of the ball). An essential property of the smoothing used is the possibility of calculating an unbiased estimation of the gradient of a smoothed function based only on realizations of the original function. The obtained smooth stochastic optimization problem is proposed to be solved in a distributed federated learning architecture (the problem is solved in parallel: nodes make local steps, e.g. stochastic gradient descent, then communicate—all with all, then all this is repeated). The goal of the article is to build on the basis of modern achievements in the field of gradient–free non-smooth optimization and in the field of federated learning gradient-free methods for solving problems of non-smooth stochastic optimization in the architecture of federated learning.

The paper continues the construction of the Lp-theory of elliptic Dirichlet and Neumann boundary value problems with discontinuous piecewise constant coefficients in divergent form for an unbounded domain  R2 with a piecewise 
 smooth noncompact Lipschitz boundary and C1 smooth discontinuity lines of the coefficients. An earlier constructed Lp-theory is generalized to the case of different smallest eigenvalues corresponding to a finite and an infinite singular point, and the effect of their interaction is further studied in the class of functions with first derivatives from Lp( ) in the entire range of the exponent p (1, ).
.

Tikhonov’s result uses values of a function on an interval; i.e., restoring the medium requires infinitely many values of the function. In this paper, the question is posed: what information about the environment can be obtained if only several values of this function are known? The answer turned out to be most favorable. If the data array contains k function values, then the environment can be characterized by the same number of parameters.

The aim of this work is to investigate a three-dimensional second-order monotone finite-difference scheme for the transport equation. The investigation is conducted for model three-dimensional transport equations of an incompressible medium. The properties of the three-dimensional extension of the Z-scheme with nonlinear correction are studied in this work. This study is an extension of the author’s previous works, where a nonlinear correction of one-dimensional transport equations was constructed. The proposed scheme uses “skew differences” containing values from different time layers instead of fluxes for the correction. The monotonicity of the obtained nonlinear scheme is numerically studied for a family of limiter functions for both smooth and non-smooth continuous solutions. The constructed scheme is absolutely stable but loses the monotonicity property when the Courant step is exceeded. The distinctive feature of the proposed finite-difference scheme is the minimalism of its template. The constructed numerical scheme is designed for models of plasma instabilities of various scales in the low-latitude ionospheric plasma of the Earth. One of the real problems that arise in solving such equations is the numerical modeling of strongly non-stationary medium- and small-scale processes in the low-latitude ionosphere of the Earth under conditions of the occurrence of the Rayleigh–Taylor instability and other types of instabilities, leading to the phenomenon of F‑scattering. Due to the fact that transport processes in the ionospheric plasma are controlled by the magnetic field, it is assumed that the plasma is incompressible in the direction perpendicular to the magnetic field.

The aggregation (consolidated, simplified representation) of systems of partial differential equations and control systems with distributed parameters is considered. Decomposition conditions based on aggregation are obtained.