From ipatov at dtm.ciw.edu Thu Apr 12 16:42:19 2007 From: ipatov at dtm.ciw.edu (ipatov at dtm.ciw.edu) Date: Thu, 12 Apr 2007 17:42:19 -0400 (EDT) Subject: [FLASH-USERS] Problems with multigrid in cylindrical coordinates Message-ID: <40097.10.101.1.203.1176414139.squirrel@www.dtm.ciw.edu> To: flash-users Problems with multigrid in cylindrical coordinates (we consider collapse+sod: collapse of a cloud together with a moving front): (1) Our runs with multipole were OK in FLASH2.5. When we tried to make the same runs with multigrid (i.e, used only -auto), then in FLASH2.5 we got errors after gmake [e.g., external subroutine MG-INIT CHARACTER*1 MPI_ARGVS_NULL(1,1) external subroutine MG_INIT_SRC CHARACTER*1 MPI_ARGVS_NULL(1,1) mg_unk(:,:,nguard+1,img_src,lb) & ] (2) For FLASH2.3, gmake was passed for multigrid (we use 2D cylindrical), but after initial plot I got "mg_init: unsupported grid geometry": [04-11-2007 14:42.00] *** plotfile: dustcoll_hdf5_plt_cnt_0000 [04-11-2007 14:42.00] *** wrote to dustcoll_hdf5_plt_cnt_0000 [04-11-2007 14:42.00] step 1 t= 0.000000E+00 dt= 1.000000E+00 [04-11-2007 14:42.00] abort_flash(): called by PE 0; message follows: [04-11-2007 14:42.00] mg_init: unsupported grid geometry Best regards Sergei From sgu at oact.inaf.it Tue Apr 17 11:15:31 2007 From: sgu at oact.inaf.it (Salvo Guglielmino) Date: Tue, 17 Apr 2007 18:15:31 +0200 Subject: [FLASH-USERS] external force References: <40097.10.101.1.203.1176414139.squirrel@www.dtm.ciw.edu> Message-ID: <00a901c7810b$9ef52dc0$99116197@Lothlorien> Dear flash-users, I'm using MHD and hydrodinamic modules, and I'm trying to include an external force, which should mantain a velocity profile against the Lorentz force. Is there any possibility to include in FLASH2.5 such a force, different from gravity, of course? Thanks in advance Salvo ********************************************************************************** Salvo Guglielmino Ph.D. Student in Physics Dipartimento di Fisica e Astronomia - Sez. Astrofisica Universit? di Catania - Via Santa Sofia 78 I-95123 Catania (Italy) Tel: +39-095-7332-330 E-mail: sgu at oact.inaf.it From rfisher at flash.uchicago.edu Tue Apr 17 11:37:03 2007 From: rfisher at flash.uchicago.edu (Robert Fisher) Date: Tue, 17 Apr 2007 11:37:03 -0500 (CDT) Subject: [FLASH-USERS] external force In-Reply-To: <00a901c7810b$9ef52dc0$99116197@Lothlorien> References: <40097.10.101.1.203.1176414139.squirrel@www.dtm.ciw.edu> <00a901c7810b$9ef52dc0$99116197@Lothlorien> Message-ID: Ciao Salvo : Both FLASH2 anf FLASH3 have source term modules to force a turbulent fluid. These is contained in /FLASH2/source/source_terms/stirring The key routine to focus on is stir, stored in stir.F90. The remainder of the files are part of the machinery to stochastically vary the driving field. I believe should be able to use this module as a starting point from which to build a driving term of the kind you describe. Best wishes, Bob From ipatov at dtm.ciw.edu Fri Apr 20 20:03:06 2007 From: ipatov at dtm.ciw.edu (ipatov at dtm.ciw.edu) Date: Fri, 20 Apr 2007 21:03:06 -0400 (EDT) Subject: [FLASH-USERS] plots for 3D In-Reply-To: <40066.10.101.1.203.1176406019.squirrel@www.dtm.ciw.edu> References: <40066.10.101.1.203.1176406019.squirrel@www.dtm.ciw.edu> Message-ID: <33605.10.101.1.203.1177117386.squirrel@www.dtm.ciw.edu> I have made a 3D cartesian run for 2 levels in FLASH2.5 (sphere + moving front of material). Does somebody know that the below problems with making plots are real or is it possible to avoid them? It was not possible to make any plot for any files for ms_4 and other ms. It was not possible to show blocks and make contours. There were normal plots for x-y and y-z; but for the first plots for x-z, 'density' and 'pressure' were empty; they began to be present on later x-z plots, but only outside the initial sphere, which remained empty (but it was not empty for x-y and y-z). For full obtained files 'chk', I was able to plot only histograms and a plot for 'pres', but not any other plots. Plots of 'dens' were made only for 'plt' files, but I was not able to plot 'pres' for 'plt' files. For 2D, I had no such problems. Best regards Sergei PS. I tried to run ./setup sedov -auto -3d (to check plots for this problem), but it crashed before it made an initial plot. -------------- next part -------------- A non-text attachment was scrubbed... Name: dens-xy-0000.ps Type: application/postscript Size: 39807 bytes Desc: not available Url : http://flash.uchicago.edu/pipermail/flash-users/attachments/20070420/09ff1745/attachment.ps -------------- next part -------------- A non-text attachment was scrubbed... Name: dens-xz-0000.ps Type: application/postscript Size: 38034 bytes Desc: not available Url : http://flash.uchicago.edu/pipermail/flash-users/attachments/20070420/09ff1745/attachment-0001.ps -------------- next part -------------- A non-text attachment was scrubbed... Name: dens-yz-0000.ps Type: application/postscript Size: 39890 bytes Desc: not available Url : http://flash.uchicago.edu/pipermail/flash-users/attachments/20070420/09ff1745/attachment-0002.ps From duffindf at univmail.cis.mcmaster.ca Mon Apr 23 14:57:28 2007 From: duffindf at univmail.cis.mcmaster.ca (Dennis Duffin) Date: Mon, 23 Apr 2007 15:57:28 -0400 Subject: [FLASH-USERS] Resistive MHD question Message-ID: Hello all, I have a Resistive MHD question. I'm playing with a custom built module modeled after the resistive MHD stuff in FLASH2.5. The non-ideal physics I'm playing with is mainly ambipolar diffusion (which kinda looks like ohmic diffusion). Anyhow, midway through the simulation the run will crash (getting negative densities; though even if I make these positive it will crash on the same step). I have no idea why. One thing I forgot to add were the divB terms and I noticed that the ohmic diffusion module includes them (as the MHD method is not assuming divB = 0). Is there any chance this could cause my program to crash (note: I will soon find out, but there are no div B anomalies in my crashed run.)??? However, if anything else pops to mind I appreciate the input! I figure similar things may have happened in ohmic diffusion runs. thanks for any help! Dennis From rfisher at flash.uchicago.edu Mon Apr 23 15:54:40 2007 From: rfisher at flash.uchicago.edu (Robert Fisher) Date: Mon, 23 Apr 2007 15:54:40 -0500 (CDT) Subject: [FLASH-USERS] Resistive MHD question In-Reply-To: References: Message-ID: Dennis : Obtaining negative mass densitites is a classic signature of a numerical instability -- you are advecting more mass out of a cell in a timestep than it has in total. This can happen in a few different ways, but the most common problem (barring major bugs) is that you are using an timestep criterion which is not sufficient. How have you generalized CFL in this case? Also, note that CFL is itself not sufficient if the accelerations are large, or if there is sudden heating in the system. Best wishes, Bob From dongwook at flash.uchicago.edu Mon Apr 23 16:20:27 2007 From: dongwook at flash.uchicago.edu (Dongwook Lee) Date: Mon, 23 Apr 2007 16:20:27 -0500 (CDT) Subject: [FLASH-USERS] Resistive MHD question In-Reply-To: References: Message-ID: Hi Dennis, I'd like to point out several things: In FLASH2.5's 8wave solver (see mhd_interpolate.F90) tries to prevent those unphysical states by reducing interpolation gradients when constructing left and right states. But if you take a close look at line 307 in the interpolation routine, you'll see the algorithm is just exiting out the DO loop when "counter" is negative, without fixing the possible unphysical state(s). So the current mhd_interpolate.F90 does not fix the issue if any negative states were introduced during the interpolation step. If you want to fix it you might want to try modifying lines to ... if( counter < 0 ) then Ux = 0 Um(:,i) = Uc(:,i) Up(:,i) = Uc(:,i) exit else ... As for the divB terms in resistive flux implementation, the inclusion of divB meant to improve numerical behaviour of the 8wave solver, because the solver does not assume divB=0. Since the divB gets bigger near discontinuities, and if you're using 8wave solver for your resistive mhd implementation, I'd suggest you put those terms to make the scheme stable. Other choice would be to use the robust HLLE solver to provide positivity. Hope this helps. Dongwook ---------------------------------------------------- Dongwook Lee, Ph.D., Research Associate ASC Center for Astrophysical Thermonuclear Flashes University of Chicago 5640 S. Ellis, RI 484 (773) 834-6830 On Mon, 23 Apr 2007, Dennis Duffin wrote: > Hello all, > > I have a Resistive MHD question. > > I'm playing with a custom built module modeled after the resistive MHD > stuff in FLASH2.5. The non-ideal physics I'm playing with is mainly > ambipolar diffusion (which kinda looks like ohmic diffusion). > > Anyhow, midway through the simulation the run will crash (getting > negative densities; though even if I make these positive it will crash > on the same step). I have no idea why. > > One thing I forgot to add were the divB terms and I noticed that the > ohmic diffusion module includes them (as the MHD method is not assuming > divB = 0). Is there any chance this could cause my program to crash > (note: I will soon find out, but there are no div B anomalies in my > crashed run.)??? > > However, if anything else pops to mind I appreciate the input! I > figure similar things may have happened in ohmic diffusion runs. > > thanks for any help! > Dennis > From townsley at uchicago.edu Thu Apr 26 23:42:43 2007 From: townsley at uchicago.edu (Dean Townsley) Date: Thu, 26 Apr 2007 23:42:43 -0500 Subject: [FLASH-USERS] flash and absoft compiler Message-ID: Hi everyone, I'm trying to help my friend Evan get flash working. He's got an Absoft fortran compiler on a Mac, and starting from the Absoft Makefile in the Prototypes directory we got as far as linking. But it seems that this fortran compiler creates symbols in all caps but the C functions are in lowercase (like every other compiler wants them). Does anyone know off-hand the way to make this work? I tried -YEXT_NAMES=LCS which changes all the fortan symbols to lowercase, but then it has problems with the mpi functions. I guess we need to rebuild mpich with a similar flag? And it makes _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is special. I'm just unsure if we are headed down the right path here. Compiler is Absoft Pro fortran version 8.0. Thanks for any help, -Dean From dubey at flash.uchicago.edu Fri Apr 27 08:00:04 2007 From: dubey at flash.uchicago.edu (Anshu Dubey) Date: Fri, 27 Apr 2007 08:00:04 -0500 (CDT) Subject: [FLASH-USERS] flash and absoft compiler In-Reply-To: References: Message-ID: <62247.75.3.114.195.1177678804.squirrel@flash.uchicago.edu> Dean, There is a site "skeeter.asci.uchicago.edu" whose Makefile.h is known to work with FLASH and absoft, though at first glance it doesn't look much different from the prototype. Your problem is most likely to be the MPI build, I remember that I had to tweak it a lot to get it to work on the mac with the xlf compiler. Usually your mpi and hdf libraries should be built with compilers from the same vendor that your fortran comes from. Shawn or Paul should be able to give you the exact options they used to build MPI for the absoft compiler on skeeter. Anshu > Hi everyone, > > I'm trying to help my friend Evan get flash working. He's got an > Absoft fortran compiler on a Mac, and starting from the Absoft > Makefile in the Prototypes directory we got as far as linking. But it > seems that this fortran compiler creates symbols in all caps but the C > functions are in lowercase (like every other compiler wants them). > > Does anyone know off-hand the way to make this work? > > I tried -YEXT_NAMES=LCS which changes all the fortan symbols to > lowercase, but then it has problems with the mpi functions. I guess > we need to rebuild mpich with a similar flag? And it makes > _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is > special. I'm just unsure if we are headed down the right path here. > > Compiler is Absoft Pro fortran version 8.0. > > Thanks for any help, > -Dean > From townsley at uchicago.edu Fri Apr 27 09:58:10 2007 From: townsley at uchicago.edu (Dean Townsley) Date: Fri, 27 Apr 2007 09:58:10 -0500 Subject: [FLASH-USERS] flash and absoft compiler In-Reply-To: Message from "Anshu Dubey" of "Fri, 27 Apr 2007 08:00:04 CDT." <62247.75.3.114.195.1177678804.squirrel@flash.uchicago.edu> References: <62247.75.3.114.195.1177678804.squirrel@flash.uchicago.edu> Message-ID: Thanks Anshu, We can rebuild mpi with the same options and maybe that will do the trick. We're so close! Cheers, -Dean On Fri, 27 Apr 2007 08:00:04 -0500 (CDT) "Anshu Dubey" wrote: > Dean, > > There is a site "skeeter.asci.uchicago.edu" whose Makefile.h is known > to work with FLASH and absoft, though at first glance it doesn't look > much different from the prototype. Your problem is most likely to be > the MPI build, I remember that I had to tweak it a lot to get it to > work on the mac with the xlf compiler. Usually your mpi and hdf libraries > should be built with compilers from the same vendor that your fortran > comes from. > > Shawn or Paul should be able to give you the exact > options they used to build MPI for the absoft compiler on skeeter. > > Anshu > > > Hi everyone, > > > > I'm trying to help my friend Evan get flash working. He's got an > > Absoft fortran compiler on a Mac, and starting from the Absoft > > Makefile in the Prototypes directory we got as far as linking. But it > > seems that this fortran compiler creates symbols in all caps but the C > > functions are in lowercase (like every other compiler wants them). > > > > Does anyone know off-hand the way to make this work? > > > > I tried -YEXT_NAMES=LCS which changes all the fortan symbols to > > lowercase, but then it has problems with the mpi functions. I guess > > we need to rebuild mpich with a similar flag? And it makes > > _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is > > special. I'm just unsure if we are headed down the right path here. > > > > Compiler is Absoft Pro fortran version 8.0. > > > > Thanks for any help, > > -Dean > > > From cyoung at grad.physics.sunysb.edu Fri Apr 27 10:09:21 2007 From: cyoung at grad.physics.sunysb.edu (Clint Young) Date: Fri, 27 Apr 2007 11:09:21 -0400 (EDT) Subject: [FLASH-USERS] flash and absoft compiler In-Reply-To: Message-ID: On Thu, 26 Apr 2007, Dean Townsley wrote: > Hi everyone, > > I'm trying to help my friend Evan get flash working. He's got an > Absoft fortran compiler on a Mac, and starting from the Absoft > Makefile in the Prototypes directory we got as far as linking. But it > seems that this fortran compiler creates symbols in all caps but the C > functions are in lowercase (like every other compiler wants them). > > Does anyone know off-hand the way to make this work? > > I tried -YEXT_NAMES=LCS which changes all the fortan symbols to > lowercase, but then it has problems with the mpi functions. I guess > we need to rebuild mpich with a similar flag? And it makes > _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is > special. I'm just unsure if we are headed down the right path here. > > Compiler is Absoft Pro fortran version 8.0. > > Thanks for any help, > -Dean > From townsley at uchicago.edu Fri Apr 27 22:23:31 2007 From: townsley at uchicago.edu (Dean Townsley) Date: Fri, 27 Apr 2007 22:23:31 -0500 Subject: [FLASH-USERS] flash and absoft compiler In-Reply-To: Message from Dean Townsley of "Thu, 26 Apr 2007 23:42:43 CDT." References: Message-ID: Hello again, For everyone wondering how things turned out, we did finally get it working. Some info from Dan Sheeler on how he and Shawn Needham got Absoft working on machines at the Center was very helpful. Thanks to everyone for their thoughts! The big thing seems to be that we had a slightly older compiler, 8.0 instead of 9.0a. The general issue is that by default Absoft generates symbols which are all caps with no trailing underscore, but there are flags to change this behavior to something more normal. The highlights were: - There were flags in the Prototype makefile (-Bstatic, -rpath) that weren't recognized. But they seem to not be needed. - Both MPICH and flash have to be built with "-YEXT_SFX=_ -YEXT_NAME=LCS" to link with C properly. It seems that if you give these to mpich the right way they don't have to actually be in the Makefile.h, and so they aren't in the Prototype. - Use -lU77 on the link line to pick up the POSIX libraries. This is also not in the Prototype because MPICH was built with it here at the Center. - And the weird one... I had to add an "allocatable" array to the RuntimeParameters_data module to convince the compiler to generate a RUNTIMEPARAMETERS_DATA symbol. Even without it, the compiler tried to *use* the symbol, which was undefined. It does appear from other modules that some modules get these symbols and some don't. My current hypothesis is that this is a compiler bug that is fixed in 9.0a. And it runs! and there was much rejoicing :) Cheers, -Dean On Thu, 26 Apr 2007 23:42:43 -0500 Dean Townsley wrote: > Hi everyone, > > I'm trying to help my friend Evan get flash working. He's got an > Absoft fortran compiler on a Mac, and starting from the Absoft > Makefile in the Prototypes directory we got as far as linking. But it > seems that this fortran compiler creates symbols in all caps but the C > functions are in lowercase (like every other compiler wants them). > > Does anyone know off-hand the way to make this work? > > I tried -YEXT_NAMES=LCS which changes all the fortan symbols to > lowercase, but then it has problems with the mpi functions. I guess > we need to rebuild mpich with a similar flag? And it makes > _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is > special. I'm just unsure if we are headed down the right path here. > > Compiler is Absoft Pro fortran version 8.0. > > Thanks for any help, > -Dean From ipatov at dtm.ciw.edu Thu Apr 12 16:42:19 2007 From: ipatov at dtm.ciw.edu (ipatov at dtm.ciw.edu) Date: Thu, 12 Apr 2007 17:42:19 -0400 (EDT) Subject: [FLASH-USERS] Problems with multigrid in cylindrical coordinates Message-ID: <40097.10.101.1.203.1176414139.squirrel@www.dtm.ciw.edu> To: flash-users Problems with multigrid in cylindrical coordinates (we consider collapse+sod: collapse of a cloud together with a moving front): (1) Our runs with multipole were OK in FLASH2.5. When we tried to make the same runs with multigrid (i.e, used only -auto), then in FLASH2.5 we got errors after gmake [e.g., external subroutine MG-INIT CHARACTER*1 MPI_ARGVS_NULL(1,1) external subroutine MG_INIT_SRC CHARACTER*1 MPI_ARGVS_NULL(1,1) mg_unk(:,:,nguard+1,img_src,lb) & ] (2) For FLASH2.3, gmake was passed for multigrid (we use 2D cylindrical), but after initial plot I got "mg_init: unsupported grid geometry": [04-11-2007 14:42.00] *** plotfile: dustcoll_hdf5_plt_cnt_0000 [04-11-2007 14:42.00] *** wrote to dustcoll_hdf5_plt_cnt_0000 [04-11-2007 14:42.00] step 1 t= 0.000000E+00 dt= 1.000000E+00 [04-11-2007 14:42.00] abort_flash(): called by PE 0; message follows: [04-11-2007 14:42.00] mg_init: unsupported grid geometry Best regards Sergei From sgu at oact.inaf.it Tue Apr 17 11:15:31 2007 From: sgu at oact.inaf.it (Salvo Guglielmino) Date: Tue, 17 Apr 2007 18:15:31 +0200 Subject: [FLASH-USERS] external force References: <40097.10.101.1.203.1176414139.squirrel@www.dtm.ciw.edu> Message-ID: <00a901c7810b$9ef52dc0$99116197@Lothlorien> Dear flash-users, I'm using MHD and hydrodinamic modules, and I'm trying to include an external force, which should mantain a velocity profile against the Lorentz force. Is there any possibility to include in FLASH2.5 such a force, different from gravity, of course? Thanks in advance Salvo ********************************************************************************** Salvo Guglielmino Ph.D. Student in Physics Dipartimento di Fisica e Astronomia - Sez. Astrofisica Universit? di Catania - Via Santa Sofia 78 I-95123 Catania (Italy) Tel: +39-095-7332-330 E-mail: sgu at oact.inaf.it From rfisher at flash.uchicago.edu Tue Apr 17 11:37:03 2007 From: rfisher at flash.uchicago.edu (Robert Fisher) Date: Tue, 17 Apr 2007 11:37:03 -0500 (CDT) Subject: [FLASH-USERS] external force In-Reply-To: <00a901c7810b$9ef52dc0$99116197@Lothlorien> References: <40097.10.101.1.203.1176414139.squirrel@www.dtm.ciw.edu> <00a901c7810b$9ef52dc0$99116197@Lothlorien> Message-ID: Ciao Salvo : Both FLASH2 anf FLASH3 have source term modules to force a turbulent fluid. These is contained in /FLASH2/source/source_terms/stirring The key routine to focus on is stir, stored in stir.F90. The remainder of the files are part of the machinery to stochastically vary the driving field. I believe should be able to use this module as a starting point from which to build a driving term of the kind you describe. Best wishes, Bob From ipatov at dtm.ciw.edu Fri Apr 20 20:03:06 2007 From: ipatov at dtm.ciw.edu (ipatov at dtm.ciw.edu) Date: Fri, 20 Apr 2007 21:03:06 -0400 (EDT) Subject: [FLASH-USERS] plots for 3D In-Reply-To: <40066.10.101.1.203.1176406019.squirrel@www.dtm.ciw.edu> References: <40066.10.101.1.203.1176406019.squirrel@www.dtm.ciw.edu> Message-ID: <33605.10.101.1.203.1177117386.squirrel@www.dtm.ciw.edu> I have made a 3D cartesian run for 2 levels in FLASH2.5 (sphere + moving front of material). Does somebody know that the below problems with making plots are real or is it possible to avoid them? It was not possible to make any plot for any files for ms_4 and other ms. It was not possible to show blocks and make contours. There were normal plots for x-y and y-z; but for the first plots for x-z, 'density' and 'pressure' were empty; they began to be present on later x-z plots, but only outside the initial sphere, which remained empty (but it was not empty for x-y and y-z). For full obtained files 'chk', I was able to plot only histograms and a plot for 'pres', but not any other plots. Plots of 'dens' were made only for 'plt' files, but I was not able to plot 'pres' for 'plt' files. For 2D, I had no such problems. Best regards Sergei PS. I tried to run ./setup sedov -auto -3d (to check plots for this problem), but it crashed before it made an initial plot. -------------- next part -------------- A non-text attachment was scrubbed... Name: dens-xy-0000.ps Type: application/postscript Size: 39807 bytes Desc: not available Url : http://flash.uchicago.edu/pipermail/flash-users/attachments/20070420/09ff1745/attachment-0003.ps -------------- next part -------------- A non-text attachment was scrubbed... Name: dens-xz-0000.ps Type: application/postscript Size: 38034 bytes Desc: not available Url : http://flash.uchicago.edu/pipermail/flash-users/attachments/20070420/09ff1745/attachment-0004.ps -------------- next part -------------- A non-text attachment was scrubbed... Name: dens-yz-0000.ps Type: application/postscript Size: 39890 bytes Desc: not available Url : http://flash.uchicago.edu/pipermail/flash-users/attachments/20070420/09ff1745/attachment-0005.ps From duffindf at univmail.cis.mcmaster.ca Mon Apr 23 14:57:28 2007 From: duffindf at univmail.cis.mcmaster.ca (Dennis Duffin) Date: Mon, 23 Apr 2007 15:57:28 -0400 Subject: [FLASH-USERS] Resistive MHD question Message-ID: Hello all, I have a Resistive MHD question. I'm playing with a custom built module modeled after the resistive MHD stuff in FLASH2.5. The non-ideal physics I'm playing with is mainly ambipolar diffusion (which kinda looks like ohmic diffusion). Anyhow, midway through the simulation the run will crash (getting negative densities; though even if I make these positive it will crash on the same step). I have no idea why. One thing I forgot to add were the divB terms and I noticed that the ohmic diffusion module includes them (as the MHD method is not assuming divB = 0). Is there any chance this could cause my program to crash (note: I will soon find out, but there are no div B anomalies in my crashed run.)??? However, if anything else pops to mind I appreciate the input! I figure similar things may have happened in ohmic diffusion runs. thanks for any help! Dennis From rfisher at flash.uchicago.edu Mon Apr 23 15:54:40 2007 From: rfisher at flash.uchicago.edu (Robert Fisher) Date: Mon, 23 Apr 2007 15:54:40 -0500 (CDT) Subject: [FLASH-USERS] Resistive MHD question In-Reply-To: References: Message-ID: Dennis : Obtaining negative mass densitites is a classic signature of a numerical instability -- you are advecting more mass out of a cell in a timestep than it has in total. This can happen in a few different ways, but the most common problem (barring major bugs) is that you are using an timestep criterion which is not sufficient. How have you generalized CFL in this case? Also, note that CFL is itself not sufficient if the accelerations are large, or if there is sudden heating in the system. Best wishes, Bob From dongwook at flash.uchicago.edu Mon Apr 23 16:20:27 2007 From: dongwook at flash.uchicago.edu (Dongwook Lee) Date: Mon, 23 Apr 2007 16:20:27 -0500 (CDT) Subject: [FLASH-USERS] Resistive MHD question In-Reply-To: References: Message-ID: Hi Dennis, I'd like to point out several things: In FLASH2.5's 8wave solver (see mhd_interpolate.F90) tries to prevent those unphysical states by reducing interpolation gradients when constructing left and right states. But if you take a close look at line 307 in the interpolation routine, you'll see the algorithm is just exiting out the DO loop when "counter" is negative, without fixing the possible unphysical state(s). So the current mhd_interpolate.F90 does not fix the issue if any negative states were introduced during the interpolation step. If you want to fix it you might want to try modifying lines to ... if( counter < 0 ) then Ux = 0 Um(:,i) = Uc(:,i) Up(:,i) = Uc(:,i) exit else ... As for the divB terms in resistive flux implementation, the inclusion of divB meant to improve numerical behaviour of the 8wave solver, because the solver does not assume divB=0. Since the divB gets bigger near discontinuities, and if you're using 8wave solver for your resistive mhd implementation, I'd suggest you put those terms to make the scheme stable. Other choice would be to use the robust HLLE solver to provide positivity. Hope this helps. Dongwook ---------------------------------------------------- Dongwook Lee, Ph.D., Research Associate ASC Center for Astrophysical Thermonuclear Flashes University of Chicago 5640 S. Ellis, RI 484 (773) 834-6830 On Mon, 23 Apr 2007, Dennis Duffin wrote: > Hello all, > > I have a Resistive MHD question. > > I'm playing with a custom built module modeled after the resistive MHD > stuff in FLASH2.5. The non-ideal physics I'm playing with is mainly > ambipolar diffusion (which kinda looks like ohmic diffusion). > > Anyhow, midway through the simulation the run will crash (getting > negative densities; though even if I make these positive it will crash > on the same step). I have no idea why. > > One thing I forgot to add were the divB terms and I noticed that the > ohmic diffusion module includes them (as the MHD method is not assuming > divB = 0). Is there any chance this could cause my program to crash > (note: I will soon find out, but there are no div B anomalies in my > crashed run.)??? > > However, if anything else pops to mind I appreciate the input! I > figure similar things may have happened in ohmic diffusion runs. > > thanks for any help! > Dennis > From townsley at uchicago.edu Thu Apr 26 23:42:43 2007 From: townsley at uchicago.edu (Dean Townsley) Date: Thu, 26 Apr 2007 23:42:43 -0500 Subject: [FLASH-USERS] flash and absoft compiler Message-ID: Hi everyone, I'm trying to help my friend Evan get flash working. He's got an Absoft fortran compiler on a Mac, and starting from the Absoft Makefile in the Prototypes directory we got as far as linking. But it seems that this fortran compiler creates symbols in all caps but the C functions are in lowercase (like every other compiler wants them). Does anyone know off-hand the way to make this work? I tried -YEXT_NAMES=LCS which changes all the fortan symbols to lowercase, but then it has problems with the mpi functions. I guess we need to rebuild mpich with a similar flag? And it makes _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is special. I'm just unsure if we are headed down the right path here. Compiler is Absoft Pro fortran version 8.0. Thanks for any help, -Dean From dubey at flash.uchicago.edu Fri Apr 27 08:00:04 2007 From: dubey at flash.uchicago.edu (Anshu Dubey) Date: Fri, 27 Apr 2007 08:00:04 -0500 (CDT) Subject: [FLASH-USERS] flash and absoft compiler In-Reply-To: References: Message-ID: <62247.75.3.114.195.1177678804.squirrel@flash.uchicago.edu> Dean, There is a site "skeeter.asci.uchicago.edu" whose Makefile.h is known to work with FLASH and absoft, though at first glance it doesn't look much different from the prototype. Your problem is most likely to be the MPI build, I remember that I had to tweak it a lot to get it to work on the mac with the xlf compiler. Usually your mpi and hdf libraries should be built with compilers from the same vendor that your fortran comes from. Shawn or Paul should be able to give you the exact options they used to build MPI for the absoft compiler on skeeter. Anshu > Hi everyone, > > I'm trying to help my friend Evan get flash working. He's got an > Absoft fortran compiler on a Mac, and starting from the Absoft > Makefile in the Prototypes directory we got as far as linking. But it > seems that this fortran compiler creates symbols in all caps but the C > functions are in lowercase (like every other compiler wants them). > > Does anyone know off-hand the way to make this work? > > I tried -YEXT_NAMES=LCS which changes all the fortan symbols to > lowercase, but then it has problems with the mpi functions. I guess > we need to rebuild mpich with a similar flag? And it makes > _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is > special. I'm just unsure if we are headed down the right path here. > > Compiler is Absoft Pro fortran version 8.0. > > Thanks for any help, > -Dean > From townsley at uchicago.edu Fri Apr 27 09:58:10 2007 From: townsley at uchicago.edu (Dean Townsley) Date: Fri, 27 Apr 2007 09:58:10 -0500 Subject: [FLASH-USERS] flash and absoft compiler In-Reply-To: Message from "Anshu Dubey" of "Fri, 27 Apr 2007 08:00:04 CDT." <62247.75.3.114.195.1177678804.squirrel@flash.uchicago.edu> References: <62247.75.3.114.195.1177678804.squirrel@flash.uchicago.edu> Message-ID: Thanks Anshu, We can rebuild mpi with the same options and maybe that will do the trick. We're so close! Cheers, -Dean On Fri, 27 Apr 2007 08:00:04 -0500 (CDT) "Anshu Dubey" wrote: > Dean, > > There is a site "skeeter.asci.uchicago.edu" whose Makefile.h is known > to work with FLASH and absoft, though at first glance it doesn't look > much different from the prototype. Your problem is most likely to be > the MPI build, I remember that I had to tweak it a lot to get it to > work on the mac with the xlf compiler. Usually your mpi and hdf libraries > should be built with compilers from the same vendor that your fortran > comes from. > > Shawn or Paul should be able to give you the exact > options they used to build MPI for the absoft compiler on skeeter. > > Anshu > > > Hi everyone, > > > > I'm trying to help my friend Evan get flash working. He's got an > > Absoft fortran compiler on a Mac, and starting from the Absoft > > Makefile in the Prototypes directory we got as far as linking. But it > > seems that this fortran compiler creates symbols in all caps but the C > > functions are in lowercase (like every other compiler wants them). > > > > Does anyone know off-hand the way to make this work? > > > > I tried -YEXT_NAMES=LCS which changes all the fortan symbols to > > lowercase, but then it has problems with the mpi functions. I guess > > we need to rebuild mpich with a similar flag? And it makes > > _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is > > special. I'm just unsure if we are headed down the right path here. > > > > Compiler is Absoft Pro fortran version 8.0. > > > > Thanks for any help, > > -Dean > > > From cyoung at grad.physics.sunysb.edu Fri Apr 27 10:09:21 2007 From: cyoung at grad.physics.sunysb.edu (Clint Young) Date: Fri, 27 Apr 2007 11:09:21 -0400 (EDT) Subject: [FLASH-USERS] flash and absoft compiler In-Reply-To: Message-ID: On Thu, 26 Apr 2007, Dean Townsley wrote: > Hi everyone, > > I'm trying to help my friend Evan get flash working. He's got an > Absoft fortran compiler on a Mac, and starting from the Absoft > Makefile in the Prototypes directory we got as far as linking. But it > seems that this fortran compiler creates symbols in all caps but the C > functions are in lowercase (like every other compiler wants them). > > Does anyone know off-hand the way to make this work? > > I tried -YEXT_NAMES=LCS which changes all the fortan symbols to > lowercase, but then it has problems with the mpi functions. I guess > we need to rebuild mpich with a similar flag? And it makes > _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is > special. I'm just unsure if we are headed down the right path here. > > Compiler is Absoft Pro fortran version 8.0. > > Thanks for any help, > -Dean > From townsley at uchicago.edu Fri Apr 27 22:23:31 2007 From: townsley at uchicago.edu (Dean Townsley) Date: Fri, 27 Apr 2007 22:23:31 -0500 Subject: [FLASH-USERS] flash and absoft compiler In-Reply-To: Message from Dean Townsley of "Thu, 26 Apr 2007 23:42:43 CDT." References: Message-ID: Hello again, For everyone wondering how things turned out, we did finally get it working. Some info from Dan Sheeler on how he and Shawn Needham got Absoft working on machines at the Center was very helpful. Thanks to everyone for their thoughts! The big thing seems to be that we had a slightly older compiler, 8.0 instead of 9.0a. The general issue is that by default Absoft generates symbols which are all caps with no trailing underscore, but there are flags to change this behavior to something more normal. The highlights were: - There were flags in the Prototype makefile (-Bstatic, -rpath) that weren't recognized. But they seem to not be needed. - Both MPICH and flash have to be built with "-YEXT_SFX=_ -YEXT_NAME=LCS" to link with C properly. It seems that if you give these to mpich the right way they don't have to actually be in the Makefile.h, and so they aren't in the Prototype. - Use -lU77 on the link line to pick up the POSIX libraries. This is also not in the Prototype because MPICH was built with it here at the Center. - And the weird one... I had to add an "allocatable" array to the RuntimeParameters_data module to convince the compiler to generate a RUNTIMEPARAMETERS_DATA symbol. Even without it, the compiler tried to *use* the symbol, which was undefined. It does appear from other modules that some modules get these symbols and some don't. My current hypothesis is that this is a compiler bug that is fixed in 9.0a. And it runs! and there was much rejoicing :) Cheers, -Dean On Thu, 26 Apr 2007 23:42:43 -0500 Dean Townsley wrote: > Hi everyone, > > I'm trying to help my friend Evan get flash working. He's got an > Absoft fortran compiler on a Mac, and starting from the Absoft > Makefile in the Prototypes directory we got as far as linking. But it > seems that this fortran compiler creates symbols in all caps but the C > functions are in lowercase (like every other compiler wants them). > > Does anyone know off-hand the way to make this work? > > I tried -YEXT_NAMES=LCS which changes all the fortan symbols to > lowercase, but then it has problems with the mpi functions. I guess > we need to rebuild mpich with a similar flag? And it makes > _RUNTIMEPARAMETERS_DATA not lowercase, but I don't know why this is > special. I'm just unsure if we are headed down the right path here. > > Compiler is Absoft Pro fortran version 8.0. > > Thanks for any help, > -Dean