Iterative Solvers for Partial Differential Equations (Iterative Löser für partielle Differentialgleichungen)

Summer Semester 2025


Lecturer: Christos Pervolianakis
Office: Ernst-Abbe-Platz 2, Room 3533
Email: christos.pervolianakis (AT) uni-jena.de
Dates: Tuesday and Thursday 12-14Uhr
Office hours: Tuesday and Thursday 15-17Uhr.


Description

Many physical and engineering problems are modeled using partial differential equations (PDEs). However, their solutions often cannot be expressed in closed form, requiring numerical approximation. Discretization methods such as finite differences and finite elements transform PDEs into large-scale linear systems that must be solved efficiently. This course delves into iterative solvers, essential for efficiently handling these systems, particularly for sparse and structured matrices. We will explore fundamental iterative methods such as Jacobi, Gauss-Seidel and multigrid methods, as well as minimization techniques like Krylov subspace methods. Additionally, we will examine their convergence properties and preconditioning techniques to accelerate computations.


Recommended knowledge

Course Evaluation


Lecture Notes:

The lecture notes will be uploaded here (.pdf or .html) after each class. To access them, you must log in to Moodle using your university account.


Theoretical Exercises


Programming Exercises

Here you will find the some numerical exercises that are optional, but interesting.

Announcements

01.04.2025  I would like to inform you that the lectures for this course will begin on April 15, 2025, as I will be attending a conference next week. The two lectures scheduled for next week will be rescheduled in the near future.

15.04.2025  During the lectures on 24.04, 06.05, 20.05, 03.06, 01.06, and 08.07, we will solve exercises.

30.04.2025 Here you will find a list of potential topics for the optional project, along with instructions.

27.05.2025 I would like to inform you that on 02.06.2025, there will be a replacement lecture in Seminar Room 3517, at Ernst-Abbe-Platz 2, from 14:00 to 16:00.

13.06.2025 The exercise session on 17.06 is rescheduled to 08.07. Additionally, during the lecture on 17.06, we will discuss the possibility of an extra exercise session. The date and time will be announced afterwards. The exercise set 4 will be announced on 19.06 after the lecture.


Lecture Calendar

15-04-2025  Two boundary value problem. Maximum principle. Finite Difference Method for solving the two boudary value problem.

17-04-2025  Finite Difference Method for solving the two boudary value problem. The discrete maximum principle.

22-04-2025  Finite Difference Method for solving the Dirichlet problem for Poisson equation. The discrete maximum principle.

24-04-2025  Solution of the exercise 4 from the Exercise set 1.

28-04-2025  Introduction to iterative methods. General iterative method with the matrix splitting \(\mathbf{A} = \mathbf{M} - \mathbf{N}\). Jordan normal form.

29-04-2025  Necessary and sufficient condition for the convergence of the general iterative method. Classical iterative methods (e.g., Jacobi and Gauẞ-Seidel methods)

06-05-2025  Solution of the exercises 1,2,5,6,7 from the Exercise set 1.

08-05-2025  No lecture.

12-05-2025  Convergence of Jacobi and Gauss-Seidel iteration.

13-05-2025  Convergence of Jacobi iteration for the linear system derived from FDM method in \(\mathbb{R}^2\) with 5 point stencil.

15-05-2025  Continue from the previous lecture. Introduction to relaxation techniques.

20-05-2025  Solution of the exercises 1,2,3 from the Exercise set 2 and the exercise 3 from Exercise set 1.

22-05-2025  Relaxation techniques. Optimal parameter for a symmetrizable iteration. Successive Over-Relaxation. Necessary criterion for the convergence. Also, a necessary and sufficient criterion for Hermite positive definite matrices.

27-05-2025  Relaxation techniques. Successive Over-Relaxation. Optimal value of SOR.

02-06-2025  Symmetric Successive Over-Relaxation (SSOR). Nececcary condition for its convergence. Sufficient and necessary condition for its convergence for \(\mathbf{A}\in\mathbb{C}^{n,n}\) Hermite and positive definite. Introduction to minimization methods. Minimization property: For an s.d.p matrix \(\mathbf{A}\in\mathbb{R}^{n,n}\), the vector \(\mathbf{x}\in\mathbb{R}^n\) solves the linear system \(\mathbf{A}\mathbf{x} = \mathbf{b}\) iff \(f(\mathbf{z}) = \min_{\mathbf{z}\in\mathbb{R}^n}f(\mathbf{z})\) where the functional is defined as \(f\,:\,\mathbb{R}^n \mapsto \mathbb{R},\,f(\mathbf{z}) = \frac{1}{2}\mathbf{z}^T\mathbf{A}\mathbf{z} - \mathbf{z}^T\mathbf{b}\).

03-06-2025  Steepest Descent Method. Error estimation. Generalized Directions Method. An interesting 3D visualization of the steepest descent method can be found in the following video https://www.youtube.com/watch?v=iudXf5n_3ro.

05-06-2025  Conjugate Directions Method. Proof that the conjugate direction method ends in at most \(n\) iterations. Further, we proved that minimizes the functional \(f(\mathbf{z}) = \frac{1}{2}\mathbf{z}^T\mathbf{A}\mathbf{z} - \mathbf{z}^T\mathbf{b}\) along the direction \(\mathbf{p}^{(\ell)}\) and at the same time in the subspace \(\mathrm{span}\{\mathbf{p}^{(1)}, \mathbf{p}^{(2)},\ldots,\mathbf{p}^{(\ell)}\}\).

10-06-2025  Solution of the exercises 4-7 from the Exercise set 2 and the exercise 1 from Exercise set 3.

12-06-2025  Proof that the conjugate direction method ends in at most \(n\) iterations. Further, we proved that minimizes the functional \(f(\mathbf{z}) = \frac{1}{2}\mathbf{z}^T\mathbf{A}\mathbf{z} - \mathbf{z}^T\mathbf{b}\) along the direction \(\mathbf{p}^{(\ell)}\) and at the same time in the subspace \(\mathrm{span}\{\mathbf{p}^{(1)}, \mathbf{p}^{(2)},\ldots,\mathbf{p}^{(\ell)}\}\). Conjugate Gradient Method. We proved also some preparatory results.

17-06-2025  Conjugate Gradient Method: Error estimation. Comparison of the convergence rate of CG with Steepest Descent method.


Literature



Last Modified: