WENO 5 extension to curvilinear grids

fjavi90 · November 8, 2017, 15:41

Hello everyone,

I have successfully implemented a finite-difference WENO5 2D compressible flow solver for the Euler equations in cartesian coordinates, but I'm having trouble with the generalization to curvilinear body-fitted grids.

I was given a compact finite-difference solver that works fine in curvilinear grids, so my guess is that the problem may be in:

the flux splitting procedure (currently using global Lax-Friedrichs).
the WENO5 scheme itself and has to do either with the eigenvalues, eigenvectors (Roe matrices) or the aplication of the metrics and jacobian of the transformation (although the metrics and jacobian are correctly calculated).

I read several articles on the matter such as this or this, but I cannot understand precisely the procedure they are using.

Any help about the generalization of the global LxF splitting and/or WENO5 would be appreciated.

Thank you.

t.teschner · November 11, 2017, 05:54

you'd be surprised, it is not that difficult. In-fact, if you use a curve-linear solver, you don't need to change your WENO5 subroutine, it will still be used the same way for curve-linear grids. The only difference is in the discretisation of the governing equation. Lets take the divergence of the velocity field (continuity equation), for example. we have (ignoring density), in cartesian form

$\nabla \cdot \textbf{U} = \frac{\partial u}{\partial x} + \frac{\partial v}{\partial y}$ .

Now to extend that to curve-linear coordinate systems, you simply need to do the coordinate transformation into a new, cartesian coordinate system, usually denoted by the new directions $\xi$ and $\eta$ so that we have now (using the chain rule)

$\nabla \cdot \textbf{U} = \frac{\partial u}{\partial \xi} \frac{\partial \xi}{\partial x} + \frac{\partial u}{\partial \eta} \frac{\partial \eta}{\partial x} + \frac{\partial v}{\partial \xi}\frac{\partial \xi}{\partial y} + \frac{\partial v}{\partial \eta}\frac{\partial \eta}{\partial y} = \xi_x \frac{\partial u}{\partial \xi} + \eta_x \frac{\partial u}{\partial \eta} + \xi_y \frac{\partial v}{\partial \xi} + \eta_y \frac{\partial v}{\partial \eta}$ .

Since your derivatives of u and v are now evaluate on the new, transformed (cartesian) grid with direction of $\xi$ and $\eta$ , you can use your WENO5 subroutine just like you would when using a cartesian grid. the only difficulty arises now how you get your so called metric coefficients, i.e.: $\xi_x , \xi_y , \eta_x , \eta_y$ (note here that the subscript just means that the derivative with respect to the subscript is taken, see also equation above).

The answer takes a bit longer and there are a couple of good books which explain that rather well (you actually have to take the so called inverse transformation, i.e. your metric coefficients are $x_\xi , y_\xi , x_\eta , y_\eta$ which changes the above equation slightly (you'll end up with the jacobian of the system)). Don't worry if that seems confusing at the moment. Have a look at "Anderson, Computational Fluid Dynamics, The basics with applications (Chapter 5 (theory) and Chapter 8 (example calculation using curve linear grids) )" and "Hoffman, Chiang, Computational Fluid Dynamics for Engineers, Vol. 1 (Chapter 9)". Both are excellent references and I was able to write my own curve-linear code after consulting those two references.

fjavi90 · November 12, 2017, 08:19

Thank you very much for your reply, t.teschner. I will have a look at those books.

mprinkey · November 14, 2017, 03:54

This is all true. Just be sure that your grid mapping is smooth, say from a full-domain conformal mapping. It should be at least several times differentiable. You should recognize that the WENO smoothness metrics are based on the assumption of evenly spaced samples. If the grid spacing changes in an non-smooth way, the effect of the mesh stretching will not be "felt" by the smoothness indicators, so the WENO stencils will be weighted based on incorrect data. The same is true for normal linear finite differences as well, but the potential for out-sized errors are introduced with the nonlinear behavior of WENO or limiters, etc.

t.teschner · November 14, 2017, 14:14

good point, did not consider that. thanks for adding that!

praveen · November 16, 2017, 23:28

A very recent preprint on the same topic

https://arxiv.org/abs/1711.06212

fjavi90 · November 23, 2017, 13:35

Thank you very much for all your answers. I am still putting the pieces together, but I am on the correct path.

On a related note, I hope it is not off-topic, I am interested on implementing a slightly different slip wall boundary condition.

Many authors mirror the flow values (i.e. density, tangent velocity and pressure) across the wall boundary (which is placed on an interface midway between nodes) and negate the normal velocity. Is there a way to make the wall boundary fall on a node instead of on an interface using WENO? The reasoning behind this is to be able to compare the results of a problem with a Finite Difference code having the nodes placed on the same coordinates.

November 8, 2017, 15:41	WENO 5 extension to curvilinear grids	#1
fjavi90 New Member Javier Join Date: Nov 2017 Posts: 6 Rep Power: 8	Hello everyone, I have successfully implemented a finite-difference WENO5 2D compressible flow solver for the Euler equations in cartesian coordinates, but I'm having trouble with the generalization to curvilinear body-fitted grids. I was given a compact finite-difference solver that works fine in curvilinear grids, so my guess is that the problem may be in: the flux splitting procedure (currently using global Lax-Friedrichs). the WENO5 scheme itself and has to do either with the eigenvalues, eigenvectors (Roe matrices) or the aplication of the metrics and jacobian of the transformation (although the metrics and jacobian are correctly calculated). I read several articles on the matter such as this or this, but I cannot understand precisely the procedure they are using. Any help about the generalization of the global LxF splitting and/or WENO5 would be appreciated. Thank you.

November 11, 2017, 05:54		#2
t.teschner Senior Member Tom-Robin Teschner Join Date: Dec 2011 Location: Cranfield, UK Posts: 204 Rep Power: 16	you'd be surprised, it is not that difficult. In-fact, if you use a curve-linear solver, you don't need to change your WENO5 subroutine, it will still be used the same way for curve-linear grids. The only difference is in the discretisation of the governing equation. Lets take the divergence of the velocity field (continuity equation), for example. we have (ignoring density), in cartesian form $\nabla \cdot \textbf{U} = \frac{\partial u}{\partial x} + \frac{\partial v}{\partial y}$ . Now to extend that to curve-linear coordinate systems, you simply need to do the coordinate transformation into a new, cartesian coordinate system, usually denoted by the new directions $\xi$ and $\eta$ so that we have now (using the chain rule) $\nabla \cdot \textbf{U} = \frac{\partial u}{\partial \xi} \frac{\partial \xi}{\partial x} + \frac{\partial u}{\partial \eta} \frac{\partial \eta}{\partial x} + \frac{\partial v}{\partial \xi}\frac{\partial \xi}{\partial y} + \frac{\partial v}{\partial \eta}\frac{\partial \eta}{\partial y} = \xi_x \frac{\partial u}{\partial \xi} + \eta_x \frac{\partial u}{\partial \eta} + \xi_y \frac{\partial v}{\partial \xi} + \eta_y \frac{\partial v}{\partial \eta}$ . Since your derivatives of u and v are now evaluate on the new, transformed (cartesian) grid with direction of $\xi$ and $\eta$ , you can use your WENO5 subroutine just like you would when using a cartesian grid. the only difficulty arises now how you get your so called metric coefficients, i.e.: $\xi_x , \xi_y , \eta_x , \eta_y$ (note here that the subscript just means that the derivative with respect to the subscript is taken, see also equation above). The answer takes a bit longer and there are a couple of good books which explain that rather well (you actually have to take the so called inverse transformation, i.e. your metric coefficients are $x_\xi , y_\xi , x_\eta , y_\eta$ which changes the above equation slightly (you'll end up with the jacobian of the system)). Don't worry if that seems confusing at the moment. Have a look at "Anderson, Computational Fluid Dynamics, The basics with applications (Chapter 5 (theory) and Chapter 8 (example calculation using curve linear grids) )" and "Hoffman, Chiang, Computational Fluid Dynamics for Engineers, Vol. 1 (Chapter 9)". Both are excellent references and I was able to write my own curve-linear code after consulting those two references. selig5576 and fjavi90 like this.

November 14, 2017, 03:54		#4
mprinkey Senior Member Michael Prinkey Join Date: Mar 2009 Location: Pittsburgh PA Posts: 363 Rep Power: 25	This is all true. Just be sure that your grid mapping is smooth, say from a full-domain conformal mapping. It should be at least several times differentiable. You should recognize that the WENO smoothness metrics are based on the assumption of evenly spaced samples. If the grid spacing changes in an non-smooth way, the effect of the mesh stretching will not be "felt" by the smoothness indicators, so the WENO stencils will be weighted based on incorrect data. The same is true for normal linear finite differences as well, but the potential for out-sized errors are introduced with the nonlinear behavior of WENO or limiters, etc. fjavi90 likes this.

November 16, 2017, 23:28		#6
praveen Super Moderator Praveen. C Join Date: Mar 2009 Location: Bangalore Posts: 342 Blog Entries: 6 Rep Power: 18	A very recent preprint on the same topic https://arxiv.org/abs/1711.06212 fjavi90 likes this. __________________ http://cpraveen.github.io http://twitter.com/cfdlab http://github.com/cpraveen

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[OpenFOAM] Xlib: extension "GLX" missing on display	goldbeard	ParaView	5	March 24, 2013 13:12
errors when installing openfoam2.1 on ubuntu12.o4	hewei	OpenFOAM Installation	5	May 29, 2012 07:43
[OpenFOAM] Saving ParaFoam views and case	sail	ParaView	9	November 25, 2011 15:46
why low number of grids in WENO	HaKu	Main CFD Forum	0	April 17, 2011 02:28
[OpenFOAM] Xwindows crash with paraview save	srinath	ParaView	1	October 15, 2008 09:37

November 12, 2017, 08:19		#3
fjavi90 New Member Javier Join Date: Nov 2017 Posts: 6 Rep Power: 8	Thank you very much for your reply, t.teschner. I will have a look at those books.

November 14, 2017, 14:14		#5
t.teschner Senior Member Tom-Robin Teschner Join Date: Dec 2011 Location: Cranfield, UK Posts: 204 Rep Power: 16	good point, did not consider that. thanks for adding that!

November 23, 2017, 13:35		#7
fjavi90 New Member Javier Join Date: Nov 2017 Posts: 6 Rep Power: 8	Thank you very much for all your answers. I am still putting the pieces together, but I am on the correct path. On a related note, I hope it is not off-topic, I am interested on implementing a slightly different slip wall boundary condition. Many authors mirror the flow values (i.e. density, tangent velocity and pressure) across the wall boundary (which is placed on an interface midway between nodes) and negate the normal velocity. Is there a way to make the wall boundary fall on a node instead of on an interface using WENO? The reasoning behind this is to be able to compare the results of a problem with a Finite Difference code having the nodes placed on the same coordinates.