February 2011 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] documentation webpage is broken
by Satish Balay 08 Feb '11

08 Feb '11

On Tue, 8 Feb 2011, Barry Smith wrote: > > On Feb 8, 2011, at 4:12 PM, Aron Ahmadia wrote: > > > link to the source of DMCreateGlobalVector from http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/docs/manualpages/… > > > > goes to a broken url: http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/da/interfa… > > > > the functional url is: http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/interface/… > > > Satish, > > Something is seriously wrong with the development version of the online docs. The old location http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/da/interfa… was changed months ago. Is something not being properly rebuilt? the LOCDIR in the makefile was not updated [when the file was moved] - so the docs had the wrong path. I fixed the makefile - so we can check tomorrow to verify if the docs get the correct path with it. Satish

1 0

Re: [petsc-dev] documentation webpage is broken
by Barry Smith 08 Feb '11

08 Feb '11

On Feb 8, 2011, at 4:12 PM, Aron Ahmadia wrote: > link to the source of DMCreateGlobalVector from http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/docs/manualpages/… > > goes to a broken url: http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/da/interfa… > > the functional url is: http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/interface/… Satish, Something is seriously wrong with the development version of the online docs. The old location http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/da/interfa… was changed months ago. Is something not being properly rebuilt? > > Also, this really annoys me: > 234: PetscErrorCode DMCreateGlobalVector(DM dm,Vec *vec) > 235: > { > > > 239: > (*dm->ops->createglobalvector)(dm,vec); > > 240: return > (0); > > 241: } > I went through the documentation to find out how DMCreateGlobalVector is implemented. What's the mnemonic for finding the definition of the above function (or I guess, multiple definitions since this appears to be a 'virtual function' in C). > A You should use Emacs etags to do this kind of thing. See http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/docs/manual.pdf section 13.8 To find all the implementations use meta . functionname then meta 0 meta . to go to the next etc etc. No pain. Barry

1 0

Re: [petsc-dev] documentation webpage is broken
by Satish Balay 08 Feb '11

08 Feb '11

On Wed, 9 Feb 2011, Aron Ahmadia wrote: > link to the source of DMCreateGlobalVector from > http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/docs/manualpages/… > > goes to a broken url: > http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/da/interfa… > > the functional url is: > http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/interface/… > > Also, this really annoys me: > > 234: *PetscErrorCode > <http://www.mcs.anl.gov/petsc/petsc-as/snapshots/docs/manualpages/Sys/PetscE…> > DMCreateGlobalVector > <http://www.mcs.anl.gov/petsc/petsc-as/snapshots/docs/manualpages/DM/DMCreat…>(DM > <http://www.mcs.anl.gov/petsc/petsc-as/snapshots/docs/manualpages/DM/DM.html…> > dm,Vec <http://www.mcs.anl.gov/petsc/petsc-as/snapshots/docs/manualpages/Vec/Vec.ht…> > *vec)*235: { > 239: (*dm->ops->createglobalvector)(dm,vec);240: return(0);241: } pushed a fix - website will get updated tonight. [will see if the links are correct then] > I went through the documentation to find out how DMCreateGlobalVector > is implemented. What's the mnemonic for finding the definition of the > above function (or I guess, multiple definitions since this appears to > be a 'virtual function' in C). tags-search with emacs? satish

1 0

documentation webpage is broken
by Aron Ahmadia 08 Feb '11

08 Feb '11

link to the source of DMCreateGlobalVector from http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/docs/manualpages/… goes to a broken url: http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/da/interfa… the functional url is: http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/src/dm/interface/… Also, this really annoys me: 234: *PetscErrorCode <http://www.mcs.anl.gov/petsc/petsc-as/snapshots/docs/manualpages/Sys/PetscE…> DMCreateGlobalVector <http://www.mcs.anl.gov/petsc/petsc-as/snapshots/docs/manualpages/DM/DMCreat…>(DM <http://www.mcs.anl.gov/petsc/petsc-as/snapshots/docs/manualpages/DM/DM.html…> dm,Vec <http://www.mcs.anl.gov/petsc/petsc-as/snapshots/docs/manualpages/Vec/Vec.ht…> *vec)*235: { 239: (*dm->ops->createglobalvector)(dm,vec);240: return(0);241: } I went through the documentation to find out how DMCreateGlobalVector is implemented. What's the mnemonic for finding the definition of the above function (or I guess, multiple definitions since this appears to be a 'virtual function' in C). A

1 0

quad precision
by Jed Brown 08 Feb '11

08 Feb '11

> GCC now ships with the LGPL-licensed libquadmath library, which provides for targets with a __float128 datatype quad-precision mathematical functions. __float128 is available for targets on 32-bit x86, x86-64 and Itanium architectures. The libquadmath library is automatically built on such targets when building the Fortran compiler. http://gcc.gnu.org/gcc-4.6/changes.html#libquadmath http://gcc.gnu.org/onlinedocs/libquadmath/ Looks like this is something that actually works. GCC-4.6 is probably about two months away from a release, but this should be functional now. Perhaps it should be supported by petsc-3.2?

1 0

Re: [petsc-dev] manual page for TSSetIJacobian()
by Jed Brown 07 Feb '11

07 Feb '11

pushed On Mon, Feb 7, 2011 at 23:52, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > On Feb 7, 2011, at 4:46 PM, Jed Brown wrote: > > > On Mon, Feb 7, 2011 at 23:35, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > Well what I want is why is this "the derivative we need"? There is no > hint as to an explanation why the user providing this quantity has any > meaning a all. You could just as easily say g_x where g(x) = f(x, a*x^3 + > 94) and why would that be wrong, while what you write is correct? > > > > That is the reason why I like introducing g(x) = f(x,xdot) > > > > " where xdot is approximated by the integrator as z+a*x for some a and z > depending on the integration method, step size, and past states." > > This is exactly the kind of text I want. > > > But perhaps the user (as opposed to someone implementing a new TS) > doesn't care at all and would be happy to evaluate f_x + a*f_xdot without > any knowledge of why this beast is relevant. > > The problem is not that you are providing too much information and > should take some away. The problem is you are not providing enough. Just > writing G(U) = F(t,U,U0+a*U) is not enough, you need to explain as you have > above in your sentence that the schemes DO approximate xdot in that manner > and therefor the Jacobian needed by the schemes is given by dF/dU + > a*dF/dU_t > > Barry > >

1 0

Re: [petsc-dev] manual page for TSSetIJacobian()
by Barry Smith 07 Feb '11

07 Feb '11

On Feb 7, 2011, at 4:46 PM, Jed Brown wrote: > On Mon, Feb 7, 2011 at 23:35, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > Well what I want is why is this "the derivative we need"? There is no hint as to an explanation why the user providing this quantity has any meaning a all. You could just as easily say g_x where g(x) = f(x, a*x^3 + 94) and why would that be wrong, while what you write is correct? > > That is the reason why I like introducing g(x) = f(x,xdot) > " where xdot is approximated by the integrator as z+a*x for some a and z depending on the integration method, step size, and past states." This is exactly the kind of text I want. > But perhaps the user (as opposed to someone implementing a new TS) doesn't care at all and would be happy to evaluate f_x + a*f_xdot without any knowledge of why this beast is relevant. The problem is not that you are providing too much information and should take some away. The problem is you are not providing enough. Just writing G(U) = F(t,U,U0+a*U) is not enough, you need to explain as you have above in your sentence that the schemes DO approximate xdot in that manner and therefor the Jacobian needed by the schemes is given by dF/dU + a*dF/dU_t Barry

1 0

Re: [petsc-dev] manual page for TSSetIJacobian()
by Jed Brown 07 Feb '11

07 Feb '11

On Mon, Feb 7, 2011 at 23:35, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > Well what I want is why is this "the derivative we need"? There is no hint > as to an explanation why the user providing this quantity has any meaning a > all. You could just as easily say g_x where g(x) = f(x, a*x^3 + 94) and why > would that be wrong, while what you write is correct? That is the reason why I like introducing g(x) = f(x,xdot) where xdot is approximated by the integrator as z+a*x for some a and z depending on the integration method, step size, and past states. But perhaps the user (as opposed to someone implementing a new TS) doesn't care at all and would be happy to evaluate f_x + a*f_xdot without any knowledge of why this beast is relevant.

1 0

Re: [petsc-dev] manual page for TSSetIJacobian()
by Barry Smith 07 Feb '11

07 Feb '11

On Feb 7, 2011, at 3:47 PM, Jed Brown wrote: > On Mon, Feb 7, 2011 at 22:19, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > It uses U0 but never explains what it is. It also calls a the "shift" but never explains what it is. Perhaps some notes explaining this stuff would make the solver more transparent. And a reference to further material that uses this approach would be great if it exists. > > Suppose we have f(x,y). There are two ways to express the derivative we need > > 1. f_x(x,y) + a*f_y(x,y) > 2. g_x where g(x) = f(x,z+a*x) > > I used version 2 in the man page because I think 1 looks arbitrary and is less obvious how it arises from a time integrator. There is a note later in the page about the equivalence. I suppose the name "shift" is more natural for version 1. Do you think I should make version 1 primary (or remove any mention of 2)? > > Do you have a better name for "shift", or just want more text (I felt like elaborating would be describing a simple equation in words)? Well what I want is why is this "the derivative we need"? There is no hint as to an explanation why the user providing this quantity has any meaning a all. You could just as easily say g_x where g(x) = f(x, a*x^3 + 94) and why would that be wrong, while what you write is correct? Barry

1 0

Re: [petsc-dev] manual page for TSSetIJacobian()
by Jed Brown 07 Feb '11

07 Feb '11

On Mon, Feb 7, 2011 at 22:19, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > It uses U0 but never explains what it is. It also calls a the "shift" but > never explains what it is. Perhaps some notes explaining this stuff would > make the solver more transparent. And a reference to further material that > uses this approach would be great if it exists. Suppose we have f(x,y). There are two ways to express the derivative we need 1. f_x(x,y) + a*f_y(x,y) 2. g_x where g(x) = f(x,z+a*x) I used version 2 in the man page because I think 1 looks arbitrary and is less obvious how it arises from a time integrator. There is a note later in the page about the equivalence. I suppose the name "shift" is more natural for version 1. Do you think I should make version 1 primary (or remove any mention of 2)? Do you have a better name for "shift", or just want more text (I felt like elaborating would be describing a simple equation in words)?

1 0