Currently job artifacts in CI/CD pipelines on LRZ GitLab never expire. Starting from Wed 26.1.2022 the default expiration time will be 30 days (GitLab default). Currently existing artifacts in already completed jobs will not be affected by the change. The latest artifacts for all jobs in the latest successful pipelines will be kept. More information: https://gitlab.lrz.de/help/user/admin_area/settings/continuous_integration.html#default-artifacts-expiration

Commit b4776a0e authored by Dominic Etienne Charrier's avatar Dominic Etienne Charrier
Browse files

12: Remove chapter on limiter tuning parameters.

parent 7188d612
......@@ -253,78 +253,6 @@ with each cell being divided in a set of $2N+1$ subcells. Taken from~\cite{ADERC
\caption[Sketch (borrowed): DG-FV DOF mapping.]{Projection to and recovery from the finite volume solver with $2N+1$ subcells for an polynomial order $N$ ADER-DG polynomial. The figure shows one cell with $N=8$. Taken from~\cite{Dumbser15}.}\label{fig:limiter-P7bis}
\end{figure}
\section{Optional tuning parameters}
\begin{warning}
This section is not up to date anymore.
\end{warning}
The \texttt{Limiting-ADER-DG} solver offers an optional tuning parameter
\texttt{helper-layers}. This parameter can be used for
enabling the solver to anticipate the propagation of troubled cells.
In the default configuration with \texttt{helper-layers}=1,
the \texttt{Limiting-ADER-DG} solver builds up two layers
of helper cells around a troubled cell.
In the one layer (FV helper cell domain), we compute with the FV solver and
project the result onto the DG space.
In the other layer (DG helper cell domain),
we compute with the ADER-DG method and project the result onto
a FV subgrid.
Everytime the configuration of troubled cells in the grid
changes, we need to step out of the time stepping loop
and rebuild these helper domains.
If a value \texttt{helper-layers}>1 is chosen, more than one layer for
constructing the FV and DG helper cell domains is used.
This means if a cell happens to be flagged as troubled within
the inner layers of the FV helper cell domain, it already
computes its solution using the FV solver and is further
surrounded by FV helper cells. We then do not need to step out of the time
stepping loop in order update the ADER-DG and FV domain.
The ADER-DG and FV helper domains are then rebuild on-the-fly during the
next time steps.
Using this feature comes with the price of a higher cost per time step
as well as the introduction of additional numerical diffusion.
However it might reduce the total runtime by skipping
multiple helper domain rebuilds.
\begin{code}
[..]
solver Limiting-ADER-DG MySolver
variables const = rho:1,j:3,E:1
order const = 3
maximum-mesh-size = 0.03704
time-stepping = global
type const = nonlinear
terms const = flux
optimisation const = generic
language const = C
limiter-type const = musclhancock
limiter-optimisation const = generic
limiter-language const = C
dmp-observables = 5
dmp-relaxation-parameter = 1e-4
dmp-difference-scaling = 1e-3
steps-till-cured = 0
helper-layers = 1
[..]
end solver
[..]
\end{code}
We have recognised that a seemingly cured troubled cell becomes
troubled again after few iterations.
The \texttt{Limiting-ADER-DG} offers to specify a
time window \texttt{steps-till-cured} which
determines how long a troubled cell should still be regarded
as troubled after it was marked as cured.
This optional parameter defaults to 0.
Using this feature comes with the price of a potentially
higher cost per time step as well as the introduction of additional
numerical diffusion.
However it might reduce the total runtime by skipping
multiple helper domain rebuilds.
\section{Hands on: Implementing a limited solver}
This section shall give a few tips how to solve a PDE with the LimitedADERDG framework of ExaHyPE. We refer to the FAQ chapter \ref{sec:chapter-faq} in the appendix for general tips.
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment