Add documentation on the current state #18

davidscn · 2022-03-29T17:51:03Z

Please rectify in case anything is wrong or not enough.

davidscn · 2022-03-29T17:53:11Z

README.md

+
+In a first step, the model order reduction was applied to the deal.II (Neumann) participant. Since deal.II is written in `C++` and the model order reduction through pyMOR is carried out through the python programming language, we compile the `C++` functions of deal.II into a python compatible library using `pybin11`, which is already included as a submodule within this project. Therefore, the deal.II source code can be found in the `lib` directory and the function calls for the deal.II heat problem as well as the pyMOR model order reduction are located in `example/neumann-reduced/heat_equation_reduced.py`. Although model order reduction with FEniCS is supported by pyMOR, we don't apply any model order reduction on the FEniCS side, i.e., the example code `example/dirichlet-fenics/heat.py` solves always the full order model.
+
+In order to apply the model order reduction on the Neumann side, we parametrize the diffusion coefficient within this participant: we split the squared domain on the right side once more into a square and an L-shaped remainder. In the offline phase, we perform multiple coupled simulations between the Dirichlet and Neumann participant using a different diffusion coefficient in the sub-square on the Neumann side. The FEniCS side computes during the offline phase always the same computational setup. By default, the sub-square with the varying diffusion coefficient is part of the coupling interface. The motivation for such a setup is the runtime reduction for the reduced order model during the online phase: Building the deal.II code in `Release` mode results in a speedup factor of around 10 for 8 global refinements, which corresponds to 66.049 degrees of freedom.


@sdrave The very last sentence #17 (comment)

That should be 66,049 degrees, no?

yes, my bad

lib/CMakeLists.txt

uekerman

Good summary 👍
A few suggestions below how we could still improve the documentation.

uekerman · 2022-04-28T07:09:30Z

README.md

+
+## Description
+
+The repository contains a ready-to-run example in the `example` directory. It consists of a simplified and adapted version of the [partitioned-heat tutorial](https://precice.org/tutorials-partitioned-heat-conduction.html), where we split a domain artificially into two parts, solve in each subdomain the same problem and carry out a Dirichlet-Neumann coupling across a common coupling interface in order to recover a global solution. However, instead of the time-dependent heat equation, we solve here a stationary Laplace problem with Dirichlet boundary conditions on the left side of the domain `u_D = 3` and homogeneous Dirichlet boundary conditions on `u_D = 0` on the right side of the domain. The left side ('Dirichlet participant') is computed using FeniCS and the right side ('Neumann participant') is computed using deal.II.


Suggested change

The repository contains a ready-to-run example in the `example` directory. It consists of a simplified and adapted version of the [partitioned-heat tutorial](https://precice.org/tutorials-partitioned-heat-conduction.html), where we split a domain artificially into two parts, solve in each subdomain the same problem and carry out a Dirichlet-Neumann coupling across a common coupling interface in order to recover a global solution. However, instead of the time-dependent heat equation, we solve here a stationary Laplace problem with Dirichlet boundary conditions on the left side of the domain `u_D = 3` and homogeneous Dirichlet boundary conditions on `u_D = 0` on the right side of the domain. The left side ('Dirichlet participant') is computed using FeniCS and the right side ('Neumann participant') is computed using deal.II.

The repository contains a ready-to-run example in the `example` directory. It consists of a simplified and adapted version of the [partitioned-heat tutorial](https://precice.org/tutorials-partitioned-heat-conduction.html), where we split a domain artificially into two parts, solve in each subdomain the same problem and carry out a Dirichlet-Neumann coupling across a common coupling interface in order to recover a global solution. However, instead of the time-dependent heat equation, we solve here a stationary Laplace problem with Dirichlet boundary conditions on the left side of the domain `u_D = 3` and homogeneous Dirichlet boundary conditions `u_D = 0` on the right side of the domain. The left side ('Dirichlet participant') is computed using FeniCS and the right side ('Neumann participant') is computed using deal.II.

uekerman · 2022-04-28T07:13:46Z

README.md

+
+In a first step, the model order reduction was applied to the deal.II (Neumann) participant. Since deal.II is written in `C++` and the model order reduction through pyMOR is carried out through the python programming language, we compile the `C++` functions of deal.II into a python compatible library using `pybin11`, which is already included as a submodule within this project. Therefore, the deal.II source code can be found in the `lib` directory and the function calls for the deal.II heat problem as well as the pyMOR model order reduction are located in `example/neumann-reduced/heat_equation_reduced.py`. Although model order reduction with FEniCS is supported by pyMOR, we don't apply any model order reduction on the FEniCS side, i.e., the example code `example/dirichlet-fenics/heat.py` solves always the full order model.
+
+In order to apply the model order reduction on the Neumann side, we parametrize the diffusion coefficient within this participant: we split the squared domain on the right side once more into a square and an L-shaped remainder. In the offline phase, we perform multiple coupled simulations between the Dirichlet and Neumann participant using a different diffusion coefficient in the sub-square on the Neumann side. The FEniCS side computes during the offline phase always the same computational setup. By default, the sub-square with the varying diffusion coefficient is part of the coupling interface. The motivation for such a setup is the runtime reduction for the reduced order model during the online phase: Building the deal.II code in `Release` mode results in a speedup factor of around 10 for 8 global refinements, which corresponds to 66.049 degrees of freedom.


Could we add a rough sketch here? domains and subdomains.

What could also be a good is a flow chart: what happens in the offline phase, what happens in the online phase.

uekerman · 2022-04-28T07:15:15Z

README.md

+```bash
+./run.sh -n 15
+```


What is the n here standing for?

uekerman · 2022-04-28T07:15:49Z

README.md

+
+## Running a simulation
+
+The example setup is located in the `example` directory. In a first step, the offline phase, multiple coupled simulations need to be performed in order to generate the parametrized reduced basis later on. By default the coupled simulation is performed `5` times using uniform samples of the diffusion coefficient. Afterwards, `5` random diffusion coefficients are used in order to compare the full order model and the reduced order model. In order to execute all simulations, execute


Uniform samples from a fixed interval?

uekerman · 2022-04-28T07:17:34Z

README.md

+python3 heat_equation_reduced.py
+```
+
+from the `neumann-reduced` directory. The Neumann participant prints out statistics regarding the error and speedup of the reduced basis.


Running these two scripts triggers the offline phase and afterwards the online phase?
Maybe we could make this more explicit.

README.md

davidscn added 2 commits March 29, 2022 16:59

Adjust runner script and require preCICE version 2.4

6d8352c

Add documentation on the current state

5367f41

davidscn requested a review from sdrave March 29, 2022 17:51

Rectify speedup sentence

7cf1db9

davidscn commented Mar 29, 2022

View reviewed changes

davidscn added the documentation Improvements or additions to documentation label Mar 29, 2022

davidscn commented Mar 29, 2022

View reviewed changes

lib/CMakeLists.txt Show resolved Hide resolved

uekerman reviewed Apr 28, 2022

View reviewed changes

davidscn commented Jun 13, 2024

View reviewed changes

README.md Outdated Show resolved Hide resolved

Update README.md

df7b047

davidscn merged commit ad623bc into main Jun 13, 2024
1 of 5 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add documentation on the current state #18

Add documentation on the current state #18

davidscn commented Mar 29, 2022 •

edited

Loading

davidscn Mar 29, 2022

renefritze Mar 30, 2022

davidscn Mar 30, 2022

uekerman left a comment

uekerman Apr 28, 2022

uekerman Apr 28, 2022

uekerman Apr 28, 2022

uekerman Apr 28, 2022

uekerman Apr 28, 2022


		In a first step, the model order reduction was applied to the deal.II (Neumann) participant. Since deal.II is written in `C++` and the model order reduction through pyMOR is carried out through the python programming language, we compile the `C++` functions of deal.II into a python compatible library using `pybin11`, which is already included as a submodule within this project. Therefore, the deal.II source code can be found in the `lib` directory and the function calls for the deal.II heat problem as well as the pyMOR model order reduction are located in `example/neumann-reduced/heat_equation_reduced.py`. Although model order reduction with FEniCS is supported by pyMOR, we don't apply any model order reduction on the FEniCS side, i.e., the example code `example/dirichlet-fenics/heat.py` solves always the full order model.

		In order to apply the model order reduction on the Neumann side, we parametrize the diffusion coefficient within this participant: we split the squared domain on the right side once more into a square and an L-shaped remainder. In the offline phase, we perform multiple coupled simulations between the Dirichlet and Neumann participant using a different diffusion coefficient in the sub-square on the Neumann side. The FEniCS side computes during the offline phase always the same computational setup. By default, the sub-square with the varying diffusion coefficient is part of the coupling interface. The motivation for such a setup is the runtime reduction for the reduced order model during the online phase: Building the deal.II code in `Release` mode results in a speedup factor of around 10 for 8 global refinements, which corresponds to 66.049 degrees of freedom.


		## Description

		The repository contains a ready-to-run example in the `example` directory. It consists of a simplified and adapted version of the [partitioned-heat tutorial](https://precice.org/tutorials-partitioned-heat-conduction.html), where we split a domain artificially into two parts, solve in each subdomain the same problem and carry out a Dirichlet-Neumann coupling across a common coupling interface in order to recover a global solution. However, instead of the time-dependent heat equation, we solve here a stationary Laplace problem with Dirichlet boundary conditions on the left side of the domain `u_D = 3` and homogeneous Dirichlet boundary conditions on `u_D = 0` on the right side of the domain. The left side ('Dirichlet participant') is computed using FeniCS and the right side ('Neumann participant') is computed using deal.II.


		## Running a simulation

		The example setup is located in the `example` directory. In a first step, the offline phase, multiple coupled simulations need to be performed in order to generate the parametrized reduced basis later on. By default the coupled simulation is performed `5` times using uniform samples of the diffusion coefficient. Afterwards, `5` random diffusion coefficients are used in order to compare the full order model and the reduced order model. In order to execute all simulations, execute

Add documentation on the current state #18

Add documentation on the current state #18

Conversation

davidscn commented Mar 29, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

uekerman left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

davidscn commented Mar 29, 2022 •

edited

Loading