You can subscribe to this list here.
2002 
_{Jan}

_{Feb}
(1) 
_{Mar}

_{Apr}

_{May}

_{Jun}
(10) 
_{Jul}
(4) 
_{Aug}
(1) 
_{Sep}
(3) 
_{Oct}

_{Nov}

_{Dec}
(2) 

2003 
_{Jan}
(30) 
_{Feb}
(20) 
_{Mar}
(2) 
_{Apr}
(19) 
_{May}
(21) 
_{Jun}
(10) 
_{Jul}
(13) 
_{Aug}
(9) 
_{Sep}
(15) 
_{Oct}
(9) 
_{Nov}
(2) 
_{Dec}
(4) 
2004 
_{Jan}
(1) 
_{Feb}
(4) 
_{Mar}

_{Apr}
(6) 
_{May}
(12) 
_{Jun}
(60) 
_{Jul}
(14) 
_{Aug}
(12) 
_{Sep}
(3) 
_{Oct}
(24) 
_{Nov}
(5) 
_{Dec}
(2) 
2005 
_{Jan}
(26) 
_{Feb}
(6) 
_{Mar}
(6) 
_{Apr}
(9) 
_{May}
(21) 
_{Jun}
(44) 
_{Jul}
(32) 
_{Aug}
(20) 
_{Sep}
(28) 
_{Oct}
(13) 
_{Nov}
(30) 
_{Dec}
(12) 
2006 
_{Jan}
(21) 
_{Feb}
(19) 
_{Mar}
(32) 
_{Apr}
(9) 
_{May}
(13) 
_{Jun}
(11) 
_{Jul}
(37) 
_{Aug}
(35) 
_{Sep}
(33) 
_{Oct}
(93) 
_{Nov}
(41) 
_{Dec}
(12) 
2007 
_{Jan}
(10) 
_{Feb}
(41) 
_{Mar}
(22) 
_{Apr}
(21) 
_{May}
(32) 
_{Jun}
(27) 
_{Jul}
(66) 
_{Aug}
(101) 
_{Sep}
(58) 
_{Oct}
(55) 
_{Nov}
(62) 
_{Dec}
(29) 
2008 
_{Jan}
(63) 
_{Feb}
(143) 
_{Mar}
(56) 
_{Apr}
(66) 
_{May}
(79) 
_{Jun}
(68) 
_{Jul}
(115) 
_{Aug}
(72) 
_{Sep}
(63) 
_{Oct}
(96) 
_{Nov}
(31) 
_{Dec}
(14) 
2009 
_{Jan}
(24) 
_{Feb}
(34) 
_{Mar}
(73) 
_{Apr}
(86) 
_{May}
(58) 
_{Jun}
(62) 
_{Jul}
(40) 
_{Aug}
(48) 
_{Sep}
(48) 
_{Oct}
(23) 
_{Nov}
(69) 
_{Dec}
(35) 
2010 
_{Jan}
(54) 
_{Feb}
(42) 
_{Mar}
(163) 
_{Apr}
(52) 
_{May}
(54) 
_{Jun}
(57) 
_{Jul}
(110) 
_{Aug}
(45) 
_{Sep}
(17) 
_{Oct}
(79) 
_{Nov}
(66) 
_{Dec}
(59) 
2011 
_{Jan}
(128) 
_{Feb}
(111) 
_{Mar}
(140) 
_{Apr}
(86) 
_{May}
(66) 
_{Jun}
(67) 
_{Jul}
(59) 
_{Aug}
(40) 
_{Sep}
(138) 
_{Oct}
(106) 
_{Nov}
(102) 
_{Dec}
(60) 
2012 
_{Jan}
(71) 
_{Feb}
(65) 
_{Mar}
(90) 
_{Apr}
(93) 
_{May}
(125) 
_{Jun}
(109) 
_{Jul}
(139) 
_{Aug}
(84) 
_{Sep}
(73) 
_{Oct}
(139) 
_{Nov}
(162) 
_{Dec}
(65) 
2013 
_{Jan}
(98) 
_{Feb}
(43) 
_{Mar}
(58) 
_{Apr}
(57) 
_{May}
(73) 
_{Jun}
(67) 
_{Jul}
(101) 
_{Aug}
(27) 
_{Sep}
(46) 
_{Oct}
(80) 
_{Nov}
(55) 
_{Dec}
(41) 
2014 
_{Jan}
(60) 
_{Feb}
(75) 
_{Mar}
(122) 
_{Apr}
(55) 
_{May}
(93) 
_{Jun}
(51) 
_{Jul}
(50) 
_{Aug}
(32) 
_{Sep}
(36) 
_{Oct}
(83) 
_{Nov}
(53) 
_{Dec}
(14) 
2015 
_{Jan}
(31) 
_{Feb}
(27) 
_{Mar}
(44) 
_{Apr}
(34) 
_{May}
(15) 
_{Jun}
(14) 
_{Jul}
(21) 
_{Aug}
(69) 
_{Sep}
(15) 
_{Oct}
(55) 
_{Nov}
(47) 
_{Dec}
(45) 
2016 
_{Jan}
(24) 
_{Feb}
(21) 
_{Mar}
(28) 
_{Apr}
(15) 
_{May}
(74) 
_{Jun}
(64) 
_{Jul}
(9) 
_{Aug}
(13) 
_{Sep}
(11) 
_{Oct}
(18) 
_{Nov}
(8) 
_{Dec}
(7) 
2017 
_{Jan}
(1) 
_{Feb}

_{Mar}

_{Apr}

_{May}

_{Jun}

_{Jul}

_{Aug}

_{Sep}

_{Oct}

_{Nov}

_{Dec}

S  M  T  W  T  F  S 



1

2

3
(2) 
4

5
(7) 
6

7

8
(1) 
9

10
(4) 
11
(3) 
12

13

14
(1) 
15
(9) 
16
(7) 
17
(1) 
18
(2) 
19

20
(1) 
21
(10) 
22
(3) 
23
(5) 
24

25

26

27

28
(4) 
29
(3) 
30
(3) 



From: Stephane Popinet <s.popinet@ni...>  20080430 20:55:05

Hi Thomas, > unfortunately i cannot access the gerrissFAQ internet side. > http://gfs.sourceforge.net/ > "connection timout" for the last two days. Hmm, it has been working OK for me recently. Slow but no connection timeout errors. I guess that could be different for people in other timezones. I have been thinking about changing hosting arrangements for Gerris (i.e. dump sourceforge) but the migration would require some work. Please try accessing the page again (as a last resort did you try the google cache?). cheers Stephane 
From: Markus Hitter <mah@ju...>  20080430 15:09:12

Am 30.04.2008 um 13:39 schrieb Thomas Zauner: > unfortunately i cannot access the gerrissFAQ internet side. > http://gfs.sourceforge.net/ > "connection timout" for the last two days. Hmm. I've covered this in my Tutorial for Engineers, which works here, albeit slowly: <http://gfs.sourceforge.net/wiki/index.php/An_engineer%27s_pipe_flow>; I'll send you a PDF copy of this and of the FAQ. Markus                    Dipl. Ing. Markus Hitter http://www.jumping.de/ 
From: Thomas Zauner <thomas.zauner@ic...>  20080430 14:41:32

Hi, unfortunately i cannot access the gerrissFAQ internet side. http://gfs.sourceforge.net/ "connection timout" for the last two days. I have simulation data i need to validate and don't know how to convert the results to SI units. velocity U,V,W and pressure P the total lenght of one GfsBox (to get the pressure gradient) and viscosity (dyn/kin ? ) as given in the input (.gfs) file. if anybody knows how to do this without the FAQ page, PLEASE let me know. thanks in advance thomas  Universität Stuttgart Institut für Computerphysik Pfaffenwaldring 27 70569 Stuttgart Room: 211 Phone: +49 711 68560458 Fax: +49 711 68563658 EMail: Thomas.Zauner@... ThomasZauner@... Homepage: http://www.icp.unistuttgart.de/~zauner 
From: Stephane Popinet <s.popinet@ni...>  20080429 20:45:15

> Can Gerris simulate two phase flows, e.g. water and air, with surface > tension effect in smallscales? What I mean small scale is that the > cell size reaches to > 1e4 m or 1e5 m . Yes. Stephane 
From: berry <lua.byhh@gm...>  20080429 16:02:18

Hi, Can Gerris simulate two phase flows, e.g. water and air, with surface tension effect in smallscales? What I mean small scale is that the cell size reaches to 1e4 m or 1e5 m .  Pang Shengyong Solidification Simulation Lab, State Key Lab of Mould & Die Technology, Huazhong Univ. of Sci. & Tech. China 
From: Haavard Holm <haavard.holm@nt...>  20080429 07:46:25

Hello, It is very appealing that you can implement the model in the parameter file. The shearimproved model, is it the model that was first tested by E. Leveque et. al, JFM/2007/vol 570/491502 ? Seems like we want to use the same case for validation. Can you release your parameter file with the turbulence model included ? Best regards Håvard Holm Stephane Popinet wrote: > Hi Haavard, > > >> I do not know the code very well. Is the diffusive terms calculated by >> using the second >> derivative of the velocity, or the first derivative of the stress tensor >> ? Answer to such >> questions may give a hint of how much work it will be to implement the >> model. >> > > We have been working quite a bit on turbulence models here at NIWA in > the past few months. We have mainly focused on LES Smagorinsky and > shearimproved Smagorinsky. The implementation is relatively simple and > can be done entirely within the parameter files. > > The difficulty is more one of model validation. This is not specific to > Gerris however. We have been testing the model on turbulent channel flow > and this seems to give reasonable answers but is very computationally > expensive: long runtimes are needed to gather meaningful turbulence > statistics etc... > > As with any turbulence model, you will need to consider exactly which > type of flow you are interested in. You cannot expect that a given model > working well in a given situation will work well in another etc... i.e. > we really do not know how to model turbulence... > > Depending on your area of application, LES may also not be what you are > after e.g. RANS with ke model may be. > > hope this helps somewhat > > Stephane > > > >  > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > Don't miss this year's exciting event. There's still time to save $100. > Use priority code J8TL2D2. > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone > _______________________________________________ > Gfsusers mailing list > Gfsusers@... > https://lists.sourceforge.net/lists/listinfo/gfsusers > 
From: Stephane Popinet <s.popinet@ni...>  20080428 23:12:14

Hi all, I just found (by chance) this nice conference paper: http://scholar.google.co.nz/scholar?hl=en&lr=&cluster=5706433537943292818 using Gerris to study "freefountains". On a general note, if you use Gerris for you research and publish papers as a result, I would like to know about it. I can then add the corresponding references in the bibliography page of the Gerris site. Of course, you don't _have to_ let me know but this is useful to show that Gerris has a wide range of uses which helps me secure continuing funding for Gerris development. In the end all Gerris users (I included) benefit from this. cheers Stephane 
From: Stephane Popinet <s.popinet@ni...>  20080428 22:06:14

Hi Haavard, > I do not know the code very well. Is the diffusive terms calculated by > using the second > derivative of the velocity, or the first derivative of the stress tensor > ? Answer to such > questions may give a hint of how much work it will be to implement the > model. We have been working quite a bit on turbulence models here at NIWA in the past few months. We have mainly focused on LES Smagorinsky and shearimproved Smagorinsky. The implementation is relatively simple and can be done entirely within the parameter files. The difficulty is more one of model validation. This is not specific to Gerris however. We have been testing the model on turbulent channel flow and this seems to give reasonable answers but is very computationally expensive: long runtimes are needed to gather meaningful turbulence statistics etc... As with any turbulence model, you will need to consider exactly which type of flow you are interested in. You cannot expect that a given model working well in a given situation will work well in another etc... i.e. we really do not know how to model turbulence... Depending on your area of application, LES may also not be what you are after e.g. RANS with ke model may be. hope this helps somewhat Stephane 
From: Stephane Popinet <s.popinet@ni...>  20080428 21:23:50

Bonjour Guillaume, Nice to hear from you. Sorry for the late answer. > What I am looking for is a kind of implicit RANS solver ... candidate > is/was OpenFOAM but knowing GTS for a long time I thought to Gerris > which documentation > sounds to me much better ;) but before going further I would like to > check if it can solve my problem properly. There isn't any implicit RANS solver in Gerris for the moment. Gerris is really focused on timedependent flows at this time, although extending the multigrid solver to timedependent flows would be possible. I would have thought that for acoustic simulations the timedependence of the base flow would be important (e.g. vortex shedding frequency etc...). From your message this does not seem to be the case, why? cheers Stephane 
From: Ruhollah Tavakoli <rohtav@gm...>  20080428 16:19:03

Hi Geordie and others, It was a permission problem for external access. I just fix it and so please try again. Best ... ruholla. On Mon, Apr 28, 2008 at 9:25 AM, Geordie McBain <gdmcbain@...> wrote: > Dear Ruhollah: Is the address > http://mehr.sharif.ir/~tav/gerris_stuff.htm right? I get 404 Not > Found, though http://mehr.sharif.ir/~tav/ is fine. > >  Original Message  > Subject: Re: [Gfsdevel] Turbulence modelling > From: "Ruhollah Tavakoli" <rohtav@...> > Date: Sun, April 27, 2008 3:41 am > To: "GFS developper discussion list" < > gfsdevel@...> > >  > > Hi Haavard, > > > > > > > I do not know the code very well. Is the diffusive terms calculated by > > using the second derivative of the velocity, or the first derivative of > the > > stress tensor? Answer to such > > questions may give a hint of how much work it will be to implement the > > model. > > > > diffusitive terms are treated with FVM method in gerris, this issue was > exactly disscused in S. Popinet JCP paper (there: treatment of laplacian > operator of poisson eq. was disscused), for convenience let me to > clarify, > > e.g. Diff_term = div (k \nabla u), in FVM treatment, we use divergence > theorem to convert this term into summation of flux throuth cell > boundaries. > Gerris just do this. > > I'm not familiar with LES models, but as a comment if you will need > mixed > derivatives, it would not be not straigntforward to implement in Gerris > and > needs more programming (I did some job for this to solve elasticity > problems > by FVM, not completed yet), if LES needs (i do not know again) some > special > treatments near solid boundaries (like two eq. closure) this could be the > most difficult part of your work (just consider that at the moment, > treatment of solid boundaries in gerris are in question, in my opinion > just > homogeneous neumann bc works well). > > To be familiar with Gerris programming, S. Zaleski note maight helps: > > http://gfs.sourceforge.net/wiki/index.php/Gerris_Flow_Solver_Programming_Course_for_Dummies > or a note by mine: http://mehr.sharif.ir/~tav/gerris_stuff.htm > > > Hope this helps, > > > Best ... ruholla. > > > > > > > > > I may give expressions / algorithm's of how the turbulence should be > > implemented, but > > need someone highly skilled to do the implementation. I think channel > > flow with > > periodic boundary conditions should be the first validating case. > > > > Best regards > > > > Håvard Holm > > > > > > > > >  > > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > > Don't miss this year's exciting event. There's still time to save $100. > > Use priority code J8TL2D2. > > > > > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone > > _______________________________________________ > > Gfsdevel mailing list > > Gfsdevel@... > > https://lists.sourceforge.net/lists/listinfo/gfsdevel > > >  > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > Don't miss this year's exciting event. There's still time to save $100. > Use priority code J8TL2D2. > > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone_______________________________________________ > Gfsdevel mailing list > Gfsdevel@... > https://lists.sourceforge.net/lists/listinfo/gfsdevel > 
From: Markus Hitter <mah@ju...>  20080423 17:02:15

Am 23.04.2008 um 15:48 schrieb Thomas Zauner: > if i define one GfsBox and use a GfsRefine 8. > then use s 4 and p 5 ...... > with what kind of system do i end up? 256x256x256 split in 2^5 > subcubes each > of size 8x8x8 ? or is the "Refine 8" applied after the s p ? Two totally nonscientific ways to learn this:  Caluclate a few steps without, then with parallel processing. If the computing wall clock time stays about the same, the total number of cubes is increased.  Look at the results. GfsView can show you the grid. Markus                    Dipl. Ing. Markus Hitter http://www.jumping.de/ 
From: Thomas Zauner <thomas.zauner@ic...>  20080423 13:48:11

Hello, does anybody know how the GfsRefine 8 work together mit the s and p switch og gerris3Dß if i define one GfsBox and use a GfsRefine 8. then use s 4 and p 5 ...... with what kind of system do i end up? 256x256x256 split in 2^5 subcubes each of size 8x8x8 ? or is the "Refine 8" applied after the s p ? Yes i am confused. Help would be much appreciated Thomas  Universität Stuttgart Institut für Computerphysik Pfaffenwaldring 27 70569 Stuttgart Room: 211 Phone: +49 711 68560458 Fax: +49 711 68563658 EMail: Thomas.Zauner@... ThomasZauner@... Homepage: http://www.icp.unistuttgart.de/~zauner 
From: Wolfgang Betz <betz.wolfgang@go...>  20080423 11:43:29

Hi, have a look at http://gfs.sourceforge.net/wiki/index.php/GfsOutputSimulation! cheers, Wolfgang On Wed, Apr 23, 2008 at 10:25 AM, Thomas Zauner <thomas.zauner@...> wrote: > Morning, > > I use vtk output at the beginning and every XXX steps but also do a text > output at the end to process the date with other tools and scripts > > GfsOutputSimulation{ start=end } finalv.txt{ > variables = U > binary = 0 > solid = 1 > format = text > } > > > > thomas > > > > On Wed April 23 08 09:53, princedue@... wrote: > > Hi all, > > > > My simulations are running fine and I am able to make movies and pictures > > with great ease using Gfsview3D.? However, I was wondering if there was a > > way for Gerris to output data in a format that I am able to read instead of > > binary.? It would be nice to have values of some sort so I could calculate > > average velocities and/or vorticities.? Any suggestions? > > > > Eric > >  > Universität Stuttgart > Institut für Computerphysik > Pfaffenwaldring 27 > 70569 Stuttgart > > Room: 211 > Phone: +49 711 68560458 > Fax: +49 711 68563658 > EMail: Thomas.Zauner@... > ThomasZauner@... > Homepage: http://www.icp.unistuttgart.de/~zauner > >  > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > Don't miss this year's exciting event. There's still time to save $100. > Use priority code J8TL2D2. > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone > _______________________________________________ > Gfsusers mailing list > Gfsusers@... > https://lists.sourceforge.net/lists/listinfo/gfsusers > 
From: Thomas Zauner <thomas.zauner@ic...>  20080423 08:25:32

Morning, I use vtk output at the beginning and every XXX steps but also do a text output at the end to process the date with other tools and scripts GfsOutputSimulation{ start=end } finalv.txt{ variables = U binary = 0 solid = 1 format = text } thomas On Wed April 23 08 09:53, princedue@... wrote: > Hi all, > > My simulations are running fine and I am able to make movies and pictures > with great ease using Gfsview3D.? However, I was wondering if there was a > way for Gerris to output data in a format that I am able to read instead of > binary.? It would be nice to have values of some sort so I could calculate > average velocities and/or vorticities.? Any suggestions? > > Eric  Universität Stuttgart Institut für Computerphysik Pfaffenwaldring 27 70569 Stuttgart Room: 211 Phone: +49 711 68560458 Fax: +49 711 68563658 EMail: Thomas.Zauner@... ThomasZauner@... Homepage: http://www.icp.unistuttgart.de/~zauner 
From: <princedue@ao...>  20080423 07:53:11

Hi all, My simulations are running fine and I am able to make movies and pictures with great ease using Gfsview3D.? However, I was wondering if there was a way for Gerris to output data in a format that I am able to read instead of binary.? It would be nice to have values of some sort so I could calculate average velocities and/or vorticities.? Any suggestions? Eric 
From: Ruhollah Tavakoli <rohtav@gm...>  20080422 18:54:48

/* This small patch includes all required modification to include nonhomogeneous Dirichlet and Neumann BC on solid obstacles to Gerris Poisson solver either pure Poisson or navier stokes eqs. Please follows given comments exactly step by step. Note that this modification is consistent with previous version and does not disturb previous stuff, i.e., you could produce your previous results without problem, this modification plays rule when you define D/N BC for a Poisson variable, i.e., pressure (in formal Gerris) which is detected by "P" in input file e.g., SurfaceBc P Neumann 1. or SurfaceBc P Dirichlet 1. Note that based on my test, Neumann condition is OK but regarding to Dirichlet BC I think that there is an internal problem in Gerris (not modification of this patch), in clear word, i think that function "gfs_cell_dirichlet_gradient_flux" does not correctly compute flux and we have leak of flux, anyway it needs further attention, so take care for Dirichlet BC In the following sections required modification are listed */ /* note that my included modification are flaged by "rt@..." go to poisson.c: from top of file to bottom 1. replace "relax2D" and "relax" functions by the following ones: */ static void relax (FttCell * cell, RelaxParams * p) { GfsGradient g; FttCellNeighbors neighbor; FttCellFace f; GfsGradient ng; g.a = GFS_VARIABLE (cell, p>dia); g.b = 0.; //rt@... if (GFS_IS_MIXED (cell)) if (((cell)>flags & GFS_FLAG_DIRICHLET) != 0) g.b = gfs_cell_dirichlet_gradient_flux (cell, p>u, p>maxlevel, 0.); f.cell = cell; ftt_cell_neighbors (cell, &neighbor); for (f.d = 0; f.d < FTT_NEIGHBORS; f.d++) { f.neighbor = neighbor.c[f.d]; if (f.neighbor) { gfs_face_weighted_gradient (&f, &ng, p>u, p>maxlevel); g.a += ng.a; g.b += ng.b; } } if (g.a > 0.) GFS_VARIABLE (cell, p>u) = (g.b  GFS_VARIABLE (cell, p>rhs))/g.a; else GFS_VARIABLE (cell, p>u) = 0.; } static void relax2D (FttCell * cell, RelaxParams * p) { GfsGradient g; FttCellNeighbors neighbor; FttCellFace f; GfsGradient ng; g.a = GFS_VARIABLE (cell, p>dia); g.b = 0.; //rt@... if (GFS_IS_MIXED (cell)) if (((cell)>flags & GFS_FLAG_DIRICHLET) != 0) g.b = gfs_cell_dirichlet_gradient_flux (cell, p>u, p>maxlevel, 0.); f.cell = cell; ftt_cell_neighbors (cell, &neighbor); for (f.d = 0; f.d < FTT_NEIGHBORS_2D; f.d++) { f.neighbor = neighbor.c[f.d]; if (f.neighbor) { gfs_face_weighted_gradient (&f, &ng, p>u, p>maxlevel); g.a += ng.a; g.b += ng.b; } } if (g.a > 0.) GFS_VARIABLE (cell, p>u) = (g.b  GFS_VARIABLE (cell, p>rhs))/g.a; else GFS_VARIABLE (cell, p>u) = 0.; } /* 2. repalce "residual_set" and "residual_set2D" functions by the following ones */ static void residual_set (FttCell * cell, RelaxParams * p) { GfsGradient g; FttCellNeighbors neighbor; FttCellFace f; GfsGradient ng; g.a = GFS_VARIABLE (cell, p>dia); g.b = 0.; // rt@... if (GFS_IS_MIXED (cell)) { if (((cell)>flags & GFS_FLAG_DIRICHLET) != 0) g.b = gfs_cell_dirichlet_gradient_flux (cell, p>u, p>maxlevel, GFS_STATE (cell)>solid>fv); else g.b = GFS_STATE (cell)>solid>fv; } f.cell = cell; ftt_cell_neighbors (cell, &neighbor); for (f.d = 0; f.d < FTT_NEIGHBORS; f.d++) { f.neighbor = neighbor.c[f.d]; if (f.neighbor) { gfs_face_weighted_gradient (&f, &ng, p>u, p>maxlevel); g.a += ng.a; g.b += ng.b; } } GFS_VARIABLE (cell, p>res) = GFS_VARIABLE (cell, p>rhs)  (g.b  GFS_VARIABLE (cell, p>u)*g.a); } static void residual_set2D (FttCell * cell, RelaxParams * p) { GfsGradient g; FttCellNeighbors neighbor; FttCellFace f; GfsGradient ng; g.a = GFS_VARIABLE (cell, p>dia); g.b = 0.; // rt@... if (GFS_IS_MIXED (cell)) { if (((cell)>flags & GFS_FLAG_DIRICHLET) != 0) g.b = gfs_cell_dirichlet_gradient_flux (cell, p>u, p>maxlevel, GFS_STATE (cell)>solid>fv); else g.b = GFS_STATE (cell)>solid>fv; } f.cell = cell; ftt_cell_neighbors (cell, &neighbor); for (f.d = 0; f.d < FTT_NEIGHBORS_2D; f.d++) { f.neighbor = neighbor.c[f.d]; if (f.neighbor) { gfs_face_weighted_gradient (&f, &ng, p>u, p>maxlevel); g.a += ng.a; g.b += ng.b; } } GFS_VARIABLE (cell, p>res) = GFS_VARIABLE (cell, p>rhs)  (g.b  GFS_VARIABLE (cell, p>u)*g.a); } /* 3. add this new function somewhere upper than "gfs_poisson_coefficients" */ // rt@... static void poisson_mixed_cell (FttCell * cell, gpointer * data) { gdouble * lambda2 = data[0]; gdouble alpha = (data[1]) ? gfs_function_value (data[1], cell) : 1.; gdouble h = ftt_cell_size (cell); if (((cell)>flags & GFS_FLAG_DIRICHLET) != 0) { GFS_STATE (cell)>solid>v = h*alpha; } else { guint i; FttVector n; gfs_solid_normal (cell, &n); for (i = 0; i < FTT_DIMENSION; i++) (&n.x)[i] *= lambda2[i]; GFS_STATE (cell)>solid>fv *= h*ftt_vector_norm (&n); } } /* 4. replace "gfs_poisson_coefficients" by the following one */ void gfs_poisson_coefficients (GfsDomain * domain, GfsFunction * alpha) { gdouble lambda2[FTT_DIMENSION]; FttComponent i; gpointer data[2]; g_return_if_fail (domain != NULL); for (i = 0; i < FTT_DIMENSION; i++) { gdouble lambda = (&domain>lambda.x)[i]; lambda2[i] = lambda*lambda; } // rt@... data[0] = lambda2; data[1] = alpha; gfs_domain_traverse_mixed (domain, FTT_PRE_ORDER, FTT_TRAVERSE_LEAFS, (FttCellTraverseFunc) poisson_mixed_cell, data); if (alpha == NULL) gfs_domain_cell_traverse (domain, FTT_PRE_ORDER, FTT_TRAVERSE_LEAFS, 1, (FttCellTraverseFunc) poisson_coeff, lambda2); else { gfs_domain_cell_traverse (domain, FTT_PRE_ORDER, FTT_TRAVERSE_LEAFS, 1, (FttCellTraverseFunc) reset_coeff, NULL); data[0] = lambda2; data[1] = alpha; gfs_domain_face_traverse (domain, FTT_XYZ, FTT_PRE_ORDER, FTT_TRAVERSE_LEAFS, 1, (FttFaceTraverseFunc) poisson_alpha_coeff, data); } gfs_domain_cell_traverse (domain, FTT_POST_ORDER, FTT_TRAVERSE_NON_LEAFS, 1, (FttCellTraverseFunc) face_coeff_from_below, NULL); } /* 5. now there is just one remained task out of "poisson.c" file in fact you sould add call for functin "gfs_domain_surface_bc" for your Poisson variable befor construction of poisson coefficient e.g. for pure poisson solver goto "simulation.c" in function "poisson_run" add this //rt@... gfs_domain_surface_bc (domain, p); before this line: "gfs_poisson_coefficients (domain, NULL);" or for NS solver you should do same job in functions "gfs_mac_projection", "gfs_approximate_projection" for Poisson variable (take care) */ /* Finally: feel free to contact me in the case of problem, any comment is wellcome ! r. tavakoli (rohtav@...) 22 april 08 */ 
From: Ruhollah Tavakoli <rohtav@gm...>  20080422 05:03:21

Hi, Geordie, Thanks for the link, it has the same structure but there are sill some remained problems if we implement Robin bc in this manner. The main issue is related to stability of convdiff. solver, the consistance implementation should consider explicit/implicit issues (diff. solver) and include the term contained field, implicitely in the diagonal entries of system of linear equation. Also stability issue should be added to time step routine too. Note that the mentioned BC by you could encounter stability issue too (depend on lambda value). i.e., when we have , v = v_0 + lambda v_n > vn (flux across boundary face) = v/lambda  v_0/lambda assume v_0 is rezo (or small) for small lambda, face flux will be large ... stability!!! Best ... ruholla. On Tue, Apr 22, 2008 at 7:15 AM, Geordie McBain <gdmcbain@...> wrote: > On Tue, Apr 22, 2008 at 8:10 AM, Stephane Popinet <s.popinet@...> > wrote: > > > > > > > I got my answer. But main remained issue for me is related to third > > > kind BC as you said (of course depend on field solution too). > > > Unfortunately Gerris does not support such BCs while they are > > > important in real world applications. They are not difficult to > > > include in Gerris. I do some early effort but, indeed, needs your > > > assistance to complete this job in a consistence manner. Anytime you > > > decide for this I will be happy to assist. I think that It will take > > > at maximum couple of hours. I will be very grateful if you attend to > > > this problem. > > > > It's on my list of things to do. You could help by reformatting the > test > > case you sent (on the wiki page) following the guidelines in the > > previous message. We could then include this test case in the default > > test suite of Gerris and then implement the required changes to make it > > work. > > Hi Ruhollah. I stumbled across > http://gfs.sourceforge.net/wiki/index.php/GfsBcNavier which I see is > for specifying `third kind' (a.k.a. Robin or radiation) boundary > conditions for velocity components. Can GfsBcNavier be applied to > diffusing scalars? If not, could the code behind it be used to > implement a GfsBcRobin command? > >  > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > Don't miss this year's exciting event. There's still time to save $100. > Use priority code J8TL2D2. > > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone > _______________________________________________ > Gfsusers mailing list > Gfsusers@... > https://lists.sourceforge.net/lists/listinfo/gfsusers > 
From: Geordie McBain <gdmcbain@fr...>  20080422 03:45:26

On Tue, Apr 22, 2008 at 8:10 AM, Stephane Popinet <s.popinet@...> wrote: > > > > I got my answer. But main remained issue for me is related to third > > kind BC as you said (of course depend on field solution too). > > Unfortunately Gerris does not support such BCs while they are > > important in real world applications. They are not difficult to > > include in Gerris. I do some early effort but, indeed, needs your > > assistance to complete this job in a consistence manner. Anytime you > > decide for this I will be happy to assist. I think that It will take > > at maximum couple of hours. I will be very grateful if you attend to > > this problem. > > It's on my list of things to do. You could help by reformatting the test > case you sent (on the wiki page) following the guidelines in the > previous message. We could then include this test case in the default > test suite of Gerris and then implement the required changes to make it > work. Hi Ruhollah. I stumbled across http://gfs.sourceforge.net/wiki/index.php/GfsBcNavier which I see is for specifying `third kind' (a.k.a. Robin or radiation) boundary conditions for velocity components. Can GfsBcNavier be applied to diffusing scalars? If not, could the code behind it be used to implement a GfsBcRobin command? 
From: Stephane Popinet <s.popinet@ni...>  20080421 22:12:32

> Considering this, what method do you recommend: conservative > differencing or nonconservative? (or is essentially preference for > any method in this case?) Try both. My feeling is that it should not make much difference but you will only know for sure by measuring the errors for both methods on specific test cases. cheers Stephane 
From: Stephane Popinet <s.popinet@ni...>  20080421 22:10:36

> I got my answer. But main remained issue for me is related to third > kind BC as you said (of course depend on field solution too). > Unfortunately Gerris does not support such BCs while they are > important in real world applications. They are not difficult to > include in Gerris. I do some early effort but, indeed, needs your > assistance to complete this job in a consistence manner. Anytime you > decide for this I will be happy to assist. I think that It will take > at maximum couple of hours. I will be very grateful if you attend to > this problem. It's on my list of things to do. You could help by reformatting the test case you sent (on the wiki page) following the guidelines in the previous message. We could then include this test case in the default test suite of Gerris and then implement the required changes to make it work. cheers Stephane 
From: Ruhollah Tavakoli <rohtav@gm...>  20080421 22:01:52

Hi Vladimir, > Can you use a Dirichlet boundary condition (say, you want to specify the > value of potential) at the solid surface in the new solver ? > > The current Poisson solver can only use the Neumann boundary condition > (df/dn = 0). No, What i mean is improving current Poisson solver not including additional BC support. In fact I use conjugate gradient with MG preconditioner. This solver is more rubost in constrast to MG, e.g. if you play a free surface case with large density ratio, there is always danger for solver failur, or MG solver can not converge to specified treshold (it reach to ossilation). In these cases MGCG behaves more robust (both theoritically and numerically proved). Regarding to BC: you are right, I am interested to improve range of supported BCs in gerris too. Anyway if you just need a BC like df/dn = constant on solid boundaries, you are lucky and this knid of BC is easy to include to current Gerris. If you need assistance contact me out of this forum. I used such BC in previous. Maybe couple of request regarding to BC motivate Stephane to attend to this issue. Best ... ruholla. > > > Thank you, > Vladimir. > > Ruhollah Tavakoli wrote: > > Hi all, > > > > > > Implementation of MGCG Poisson solver under Gerris API is available > > (compatible with Gerris parallel stuff too). In my experience, a few > > test cases to solve just Poisson equation with large coefficient jump > > (upto 100000:1), it is very promising in contrast to formal Gerris MG > > solver. > > > > > > Anyway I am looking for some tough test cases to better evaluation of > > this solver; hopefully I will share it after (probably) required > revision. > > > > > > Please forward to me any test case that you think it could be > > challenging to solve for Gerris, 2D cases and just Poisson equation > > (GfsPoisson) are more desirable. > > > > > > Thanks for your assistance. > > > > > > Best … RT. > > > >  > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > Don't miss this year's exciting event. There's still time to save $100. > Use priority code J8TL2D2. > > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone > _______________________________________________ > Gfsusers mailing list > Gfsusers@... > https://lists.sourceforge.net/lists/listinfo/gfsusers > 
From: Vladimir Kolobov <vik@cf...>  20080421 21:28:36

Dear RT, Looks very interesting and promising ! Can you use a Dirichlet boundary condition (say, you want to specify the value of potential) at the solid surface in the new solver ? The current Poisson solver can only use the Neumann boundary condition (df/dn = 0). Thank you, Vladimir. Ruhollah Tavakoli wrote: > Hi all, > > > Implementation of MGCG Poisson solver under Gerris API is available > (compatible with Gerris parallel stuff too). In my experience, a few > test cases to solve just Poisson equation with large coefficient jump > (upto 100000:1), it is very promising in contrast to formal Gerris MG > solver. > > > Anyway I am looking for some tough test cases to better evaluation of > this solver; hopefully I will share it after (probably) required revision. > > > Please forward to me any test case that you think it could be > challenging to solve for Gerris, 2D cases and just Poisson equation > (GfsPoisson) are more desirable. > > > Thanks for your assistance. > > > Best … RT. 
From: Ruhollah Tavakoli <rohtav@gm...>  20080421 21:28:09

Hi Stephane, Thanks for the given information, my question arises form my application where solid/liquid interface (phase change problem) is represented by a normalized level set, I have control on curvature of interface to optimal control of solidification process and so computation of curvature is just important for me (without attention to surface tension etc.), Considering this, what method do you recommend: conservative differencing or nonconservative? (or is essentially preference for any method in this case?) Best ... RT. Gerris does not use this approach in the case of an interface defined > using VOF. It uses an "HeightFunction" (HF) method. > > Gerris has an implementation of this method for a preliminary "levelset" > implementation (GfsVariableDistance object). This implementation has > been superseded by the VOFHF method above. > > Conservation issues for surface tension are more complex than just > diferencing the normal "conservatively" or "nonconservatively". The > only conservative formulations I am aware of are the > ContinuumSurfaceStress (CSS) approach of Gueyffier et al. (1998) and > the marker technique of Popinet & Zaleski (1999). > > cheers > > Stephane > > > >  > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > Don't miss this year's exciting event. There's still time to save $100. > Use priority code J8TL2D2. > > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone > _______________________________________________ > Gfsusers mailing list > Gfsusers@... > https://lists.sourceforge.net/lists/listinfo/gfsusers > 
From: Ruhollah Tavakoli <rohtav@gm...>  20080421 21:16:45

Hi Stephane, Thanks for your response. > Hmm not sure what you mean... Do you mean that the BCs depend on the > values of the solution itself? (i.e. similar to the problem you reported > a while ago?) > I got my answer. But main remained issue for me is related to third kind BC as you said (of course depend on field solution too). Unfortunately Gerris does not support such BCs while they are important in real world applications. They are not difficult to include in Gerris. I do some early effort but, indeed, needs your assistance to complete this job in a consistence manner. Anytime you decide for this I will be happy to assist. I think that It will take at maximum couple of hours. I will be very grateful if you attend to this problem. Best ... RT. PS: third kind BC is in fact generalization of Neumann BC, it is usually expressed as du/dn = a u + b where "u" is field solution, "n" is outward normal to boundary, "a" and "b" are some functions. Stephan: coefficient "a" could be included in system of equations in a manner like "dia" in Helmholtz equation (of course only for boundary cells). > > > cheers > > Stephane > > > > >  > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > Don't miss this year's exciting event. There's still time to save $100. > Use priority code J8TL2D2. > > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone > _______________________________________________ > Gfsusers mailing list > Gfsusers@... > https://lists.sourceforge.net/lists/listinfo/gfsusers > 
From: Stephane Popinet <s.popinet@ni...>  20080421 21:15:13

Hi Ruholla, > Curvature field is equal to divergence of normalized normal field, so > to compute curvature from a volumetric distance field, there are two > strategies: > 1. Conservative: converting divergence to summation of flux > across cell faces (using gauss theorem) > 2. Nonconservative: applying cell center gradient (dot product) > on cell centered normalized normal field. > Gerris uses the second approach; I would like to know its benefit in > contrast, any commnt? Gerris does not use this approach in the case of an interface defined using VOF. It uses an "HeightFunction" (HF) method. Gerris has an implementation of this method for a preliminary "levelset" implementation (GfsVariableDistance object). This implementation has been superseded by the VOFHF method above. Conservation issues for surface tension are more complex than just diferencing the normal "conservatively" or "nonconservatively". The only conservative formulations I am aware of are the ContinuumSurfaceStress (CSS) approach of Gueyffier et al. (1998) and the marker technique of Popinet & Zaleski (1999). cheers Stephane 