Tatyana Khachkova
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Tatyana Khachkova has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Vadim Lisitsa
Sobolev Institute of Mathematics SB RAS
9 shared publications - Dmitriy Kolyukhin
Trofimuk Institute of Petroleum Geology and Geophysics SB RAS , Koptug ave. 3, 630090 Novosibirsk , Russia
2 shared publications - Dmitry Prokhorov
Trofimuk Institute of Petroleum Geology and Geophysics SB RAS
2 shared publications - Aleksei A. Manaev
Sobolev Institute of Mathematics SB RAS
1 shared publication Alexander Avdonin
Gazpromneft NTC
1 shared publication- Anita Torabi
Department of Geosciences, University of Oslo, Postboks 1047, Blindern 0316, Oslo, Norway
1 shared publication
Latest Publications
- Open Access
- 30 June 2026
- Published by National Research Mordovia State University MRSU in Zhurnal Srednevolzhskogo Matematicheskogo Obshchestva
- Vol. 28 (2) , 83-98
Machine learning-based preconditioner to solve Poisson equation
In this paper, we present a method of constructing a preconditioner based on machine learning methods for use in the numerical solution of the Poisson equation in porous media modelling. The iterative conjugate gradient method is used to solve the problem. To precondition a system of linear algebraic equations, we propose to approximate the inverse Laplace operator using a convolutional neural network of the U-net architecture. We consider two alternative approaches to the formation of a training dataset for a neural network. The first method is based on the use of pairs of vectors and on the results of applying the Laplace operator to them. In the second method, the training pairs are residual vectors obtained by implementing the conjugate gradient method and results of applying the Laplace operator to them. The neural network learning process is based on minimizing the $L^2$-norm relative error. We show that while using each of the presented learning methods, the U-net neural network with five convolutional layers approximating the inverse Laplace operator provides insufficient accuracy to reduce the number of iterations in the conjugate gradient method. Therefore, the modified conjugate gradient method is stabilized and has an irreducible residual.
- 30 June 2026
- Published by National Research Mordovia State University MRSU in Zhurnal Srednevolzhskogo Matematicheskogo Obshchestva
- Vol. 28 (2) , 83-98
Open AccessMachine learning-based preconditioner to solve Poisson equation
In this paper, we present a method of constructing a preconditioner based on machine learning methods for use in the numerical solution of the Poisson equation in porous media modelling. The iterative conjugate gradient method is used to solve the problem. To precondition a system of linear algebraic equations, we propose to approximate the inverse Laplace operator using a convolutional neural network of the U-net architecture. We consider two alternative approaches to the formation of a training dataset for a neural network. The first method is based on the use of pairs of vectors and on the results of applying the Laplace operator to them. In the second method, the training pairs are residual vectors obtained by implementing the conjugate gradient method and results of applying the Laplace operator to them. The neural network learning process is based on minimizing the $L^2$-norm relative error. We show that while using each of the presented learning methods, the U-net neural network with five convolutional layers approximating the inverse Laplace operator provides insufficient accuracy to reduce the number of iterations in the conjugate gradient method. Therefore, the modified conjugate gradient method is stabilized and has an irreducible residual.
- Open Access
- 6 October 2025
- Published by Research Computing Center Lomonosov Moscow State University in Numerical Methods and Programming (Vychislitel'nye Metody i Programmirovanie)
- Vol. 26 (4) , 479-501
Numerical algorithm to estimate formation factor of digital rocks
The paper presents a numerical algorithm of estimating the formation factor for digital rock samples constructed using microtomographic images. The algorithm is based on the numerical solution of the three-dimensional Poisson equation with rapidly changing high-contrast coefficients. The arising system of linear algebraic equations turns out to be ill-conditioned, so to speed up the convergence it is necessary to use a preconditioner. The conjugate gradient method is applied to solve the system and the preconditioner is constructed as the inverse Laplace operator corresponding to the homogeneous model. In its turn, the spectral decomposition of two tridiagonal matrices appearing in the approximation of one-dimensional derivatives is used for inversion. The resulting series of one dimensional problems is solved by the Thomas algorithm. The used preconditioner can be effectively applied both to the original problem with rapidly changing high-contrast coefficients and to the problem, in which the solution is calculated only in the pore space. In the latter case, the method provides the same accuracy of results as the original one for inhomogeneous models, but it converges almost twice as fast. Implementation using graphics processors allows solving problems up to 109 unknowns with a single GPU. В работе представлен численный алгоритм оценки параметра пористости для цифровых образцов горных пород, построенных с использованием микротомографических изображений. В основе алгоритма лежит численное решение трехмерного уравнения Пуассо на с быстро изменяющимися высококонтрастными коэффициентами. Возникающая при этом система линейных алгебраических уравнений оказывается плохо обусловленной, поэтому для ускорения сходимости требуется использование предобуславливателя. Для решения применяется метод сопряженных градиентов, а предобуславливатель строится как обратный оператор Лапласа, соответствующий однородной модели. В свою очередь, для обращения используется спектральное разложение двух трехдиагональных матриц, соответствующих аппроксимации одномерных производных. Полученная серия одномерных задач решается методом прогонки. Используемый предобуславливатель может быть эффективно применен как к исходной задаче с быстро изменяющимися высококонтрастными коэффициентами, так и к задаче, в которой решение вычисляется только в поровом пространстве. В этом случае данный метод обеспечивает такую же точность результатов, как и исходный подход для неоднородных моделей, однако сходится почти вдвое быстрее исходного. Реализация с использованием графических процессоров позволяет решать задачи размером до 109 неизвестных с помощью одного GPU.
- 6 October 2025
- Published by Research Computing Center Lomonosov Moscow State University in Numerical Methods and Programming (Vychislitel'nye Metody i Programmirovanie)
- Vol. 26 (4) , 479-501
Open AccessNumerical algorithm to estimate formation factor of digital rocks
The paper presents a numerical algorithm of estimating the formation factor for digital rock samples constructed using microtomographic images. The algorithm is based on the numerical solution of the three-dimensional Poisson equation with rapidly changing high-contrast coefficients. The arising system of linear algebraic equations turns out to be ill-conditioned, so to speed up the convergence it is necessary to use a preconditioner. The conjugate gradient method is applied to solve the system and the preconditioner is constructed as the inverse Laplace operator corresponding to the homogeneous model. In its turn, the spectral decomposition of two tridiagonal matrices appearing in the approximation of one-dimensional derivatives is used for inversion. The resulting series of one dimensional problems is solved by the Thomas algorithm. The used preconditioner can be effectively applied both to the original problem with rapidly changing high-contrast coefficients and to the problem, in which the solution is calculated only in the pore space. In the latter case, the method provides the same accuracy of results as the original one for inhomogeneous models, but it converges almost twice as fast. Implementation using graphics processors allows solving problems up to 109 unknowns with a single GPU. В работе представлен численный алгоритм оценки параметра пористости для цифровых образцов горных пород, построенных с использованием микротомографических изображений. В основе алгоритма лежит численное решение трехмерного уравнения Пуассо на с быстро изменяющимися высококонтрастными коэффициентами. Возникающая при этом система линейных алгебраических уравнений оказывается плохо обусловленной, поэтому для ускорения сходимости требуется использование предобуславливателя. Для решения применяется метод сопряженных градиентов, а предобуславливатель строится как обратный оператор Лапласа, соответствующий однородной модели. В свою очередь, для обращения используется спектральное разложение двух трехдиагональных матриц, соответствующих аппроксимации одномерных производных. Полученная серия одномерных задач решается методом прогонки. Используемый предобуславливатель может быть эффективно применен как к исходной задаче с быстро изменяющимися высококонтрастными коэффициентами, так и к задаче, в которой решение вычисляется только в поровом пространстве. В этом случае данный метод обеспечивает такую же точность результатов, как и исходный подход для неоднородных моделей, однако сходится почти вдвое быстрее исходного. Реализация с использованием графических процессоров позволяет решать задачи размером до 109 неизвестных с помощью одного GPU.
- 26 July 2024
- Published by Springer Nature
- p. 250-262
Construction of Relative Permeability Curves by Numerical Simulation of Two-Phase Flow in 3D CT-Models
- 26 July 2024
- Published by Springer Nature
- p. 250-262
Construction of Relative Permeability Curves by Numerical Simulation of Two-Phase Flow in 3D CT-Models
- Open Access
- 17 July 2024
- Published by Trofimuk Institute of Petroleum Geology and Geophysics (SB RAS) in Russian Journal of geophysical technologies
- No. 1,p. 60-71
Numerical simulation of twophase flows on the base of phase-field method
Численное моделирование двухфазных потоков методом фазового поля
- 17 July 2024
- Published by Trofimuk Institute of Petroleum Geology and Geophysics (SB RAS) in Russian Journal of geophysical technologies
- No. 1,p. 60-71
Open AccessNumerical simulation of twophase flows on the base of phase-field method
Численное моделирование двухфазных потоков методом фазового поля
- 29 June 2023
- Published by Springer Nature
- p. 91-102
Numerical Evaluating the Permeability of Rocks Based on Correlation Dependence on Geometry
- 29 June 2023
- Published by Springer Nature
- p. 91-102
Numerical Evaluating the Permeability of Rocks Based on Correlation Dependence on Geometry
- 4 August 2022
- Published by Springer Nature
- p. 328-339
Effect of the Interface Roughness on the Elastic Moduli
- 4 August 2022
- Published by Springer Nature
- p. 328-339
Effect of the Interface Roughness on the Elastic Moduli
- 1 February 2022
- Published by Elsevier in Journal of Computational Science
- Vol. 58
Topology-based characterization of chemically-induced pore space changes using reduction of 3D digital images
- 1 February 2022
- Published by Elsevier in Journal of Computational Science
- Vol. 58
Topology-based characterization of chemically-induced pore space changes using reduction of 3D digital images
- 9 September 2021
- Published by Springer Nature
- p. 532-545
Poisson Solver for Upscaling the Physical Properties of Porous Materials
- 9 September 2021
- Published by Springer Nature
- p. 532-545
Poisson Solver for Upscaling the Physical Properties of Porous Materials
- 10 December 2019
- Published by Springer Nature
- p. 290-301
Numerical Experiments with Digital Twins of Core Samples for Estimating Effective Elastic Parameters
- 10 December 2019
- Published by Springer Nature
- p. 290-301
Numerical Experiments with Digital Twins of Core Samples for Estimating Effective Elastic Parameters