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

Re: [petsc-dev] Fwd: [petsc-maint #70591] Are the petsc4py directions current?
by Matthew Knepley 18 Apr '11

18 Apr '11

On Mon, Apr 18, 2011 at 6:24 PM, Lisandro Dalcin <dalcinl(a)gmail.com> wrote: > On 17 April 2011 14:19, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > > > > > Begin forwarded message: > > > > From: Jed Brown <jed(a)59A2.org> > > Date: April 17, 2011 12:14:03 PM CDT > > To: petsc-maint <petsc-maint(a)mcs.anl.gov> > > Subject: [petsc-maint #70591] Are the petsc4py directions current? > > Reply-To: petsc-maint(a)mcs.anl.gov, Jed Brown <jed(a)59A2.org> > > > > > http://www.mcs.anl.gov/petsc/petsc-as/documentation/faq.html#install-petsc4… > > > > I just had a (Mac) user tell me that this isn't working. He was silently > > getting some much older install of petsc4py that he had forgotten about, > so > > we ended up wasting some time before confirming that he really had not > > loaded petsc4py dev. > > > > When someone follows this procedure, where does petsc4py actually go? > > > > > http://www.mcs.anl.gov/petsc/petsc-as/documentation/faq.html#install-petsc4… > > I just had a (Mac) user tell me that this isn't working. He was silently > > getting some much older install of petsc4py that he had forgotten about, > so > > we ended up wasting some time before confirming that he really had not > > loaded petsc4py dev. > > When someone follows this procedure, where does petsc4py actually go? > > > > Well, these instructions are a bit outdated. The 'make cython' step is > not required, Cython is run automatically. And I never use the > --download-petsc4py stuff. This should be as easy as: > > $ pip install Cython # or manual "python setup.py install" from tarball > $ hg clone https://petsc4py.googlecode.com/hg/ petsc4py-dev > $ export PETSC_DIR=/path/to/petsc-dev PETSC_ARCH=arch-xxx ARCHFLAGS="" > $ python setup.py install --user > > # Using --user with Python >=2.6, gets installed at > ~/.local/lib/python2.6/site-packages, this is automatically added to > Python's sys.path. > > However, this model is perhaps more appropriate for PETSc releases, > for petsc-dev the --download-petsc4py could be more convenient for > some users that also use petsc-3.1 on a regular basis. However, the > instructions in the FAQ forget to ask users to set PYTHONPATH. > Nope, I use the above model for petsc-dev and its in my Chile slides. Matt > -- > Lisandro Dalcin > --------------- > CIMEC (INTEC/CONICET-UNL) > Predio CONICET-Santa Fe > Colectora RN 168 Km 472, Paraje El Pozo > 3000 Santa Fe, Argentina > Tel: +54-342-4511594 (ext 1011) > Tel/Fax: +54-342-4511169 > -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener

1 0

Re: [petsc-dev] Fwd: [petsc-maint #70591] Are the petsc4py directions current?
by Lisandro Dalcin 18 Apr '11

18 Apr '11

On 17 April 2011 14:19, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > > Begin forwarded message: > > From: Jed Brown <jed(a)59A2.org> > Date: April 17, 2011 12:14:03 PM CDT > To: petsc-maint <petsc-maint(a)mcs.anl.gov> > Subject: [petsc-maint #70591] Are the petsc4py directions current? > Reply-To: petsc-maint(a)mcs.anl.gov, Jed Brown <jed(a)59A2.org> > > http://www.mcs.anl.gov/petsc/petsc-as/documentation/faq.html#install-petsc4… > > I just had a (Mac) user tell me that this isn't working. He was silently > getting some much older install of petsc4py that he had forgotten about, so > we ended up wasting some time before confirming that he really had not > loaded petsc4py dev. > > When someone follows this procedure, where does petsc4py actually go? > > http://www.mcs.anl.gov/petsc/petsc-as/documentation/faq.html#install-petsc4… > I just had a (Mac) user tell me that this isn't working. He was silently > getting some much older install of petsc4py that he had forgotten about, so > we ended up wasting some time before confirming that he really had not > loaded petsc4py dev. > When someone follows this procedure, where does petsc4py actually go? > Well, these instructions are a bit outdated. The 'make cython' step is not required, Cython is run automatically. And I never use the --download-petsc4py stuff. This should be as easy as: $ pip install Cython # or manual "python setup.py install" from tarball $ hg clone https://petsc4py.googlecode.com/hg/ petsc4py-dev $ export PETSC_DIR=/path/to/petsc-dev PETSC_ARCH=arch-xxx ARCHFLAGS="" $ python setup.py install --user # Using --user with Python >=2.6, gets installed at ~/.local/lib/python2.6/site-packages, this is automatically added to Python's sys.path. However, this model is perhaps more appropriate for PETSc releases, for petsc-dev the --download-petsc4py could be more convenient for some users that also use petsc-3.1 on a regular basis. However, the instructions in the FAQ forget to ask users to set PYTHONPATH. -- Lisandro Dalcin --------------- CIMEC (INTEC/CONICET-UNL) Predio CONICET-Santa Fe Colectora RN 168 Km 472, Paraje El Pozo 3000 Santa Fe, Argentina Tel: +54-342-4511594 (ext 1011) Tel/Fax: +54-342-4511169

1 0

Re: [petsc-dev] Fwd: [petsc-maint #70675] update of the timestep in theta method
by Lisandro Dalcin 18 Apr '11

18 Apr '11

On 18 April 2011 19:35, Jed Brown <jed(a)59a2.org> wrote: > On Tue, Apr 19, 2011 at 00:33, Jeremy Riousset <riousset(a)gatech.edu> wrote: >> >> n order, I call >> ierr = >> PetscOptionsGetReal(PETSC_NULL,"-ts_dt",&user.dt,PETSC_NULL);CHKERRQ(ierr); > > TS gets this option itself, inside TSSetFromOptions(). You can use TSALPHA. You can set a callback with TSAlphaSetAdapt(), then you can control the time step and even accept/reject steps. With default parameters, ALPHA is equivalent to THETA. However, it costs a bit more because of some extra Vec operations. We should really extend this API to many TS implementations in order to support simple time step adaptativity. -- Lisandro Dalcin --------------- CIMEC (INTEC/CONICET-UNL) Predio CONICET-Santa Fe Colectora RN 168 Km 472, Paraje El Pozo 3000 Santa Fe, Argentina Tel: +54-342-4511594 (ext 1011) Tel/Fax: +54-342-4511169

1 0

Re: [petsc-dev] Fwd: [petsc-maint #70675] update of the timestep in theta method
by Jed Brown 18 Apr '11

18 Apr '11

On Tue, Apr 19, 2011 at 00:33, Jeremy Riousset <riousset(a)gatech.edu> wrote: > n order, I call > ierr = > PetscOptionsGetReal(PETSC_NULL,"-ts_dt",&user.dt,PETSC_NULL);CHKERRQ(ierr); > TS gets this option itself, inside TSSetFromOptions().

1 0

Re: [petsc-dev] Fwd: [petsc-maint #70675] update of the timestep in theta method
by Jeremy Riousset 18 Apr '11

18 Apr '11

Hi, It seems that when I want to modify the time step before the call of TSSetInitialTimeStep, I run into an error. I compare a value read from an input file to a hard coded value and the code systematically use the value from the file. Any idea why? in order, I call ierr = PetscOptionsGetReal(PETSC_NULL,"-ts_dt",&user.dt,PETSC_NULL);CHKERRQ(ierr); then define user.dt = something smaller than what is put in the file then print user.dt (which returns the modified value) then call ierr = TSSetInitialTimeStep(ts,user.ti,user.dt);CHKERRQ(ierr); but the value read by PetscOptionsGetReal is used by TSSetInitialTimeStep Thanks for your help, --Jeremy On 04/18/2011 03:19 PM, Jed Brown wrote: > On Mon, Apr 18, 2011 at 21:17, Jeremy Riousset <riousset(a)gatech.edu > <mailto:[email protected]>> wrote: > > I was thinking to call it in the Monitor function. > > > You can also do that, but then the output of other monitors is > dependent on whether it comes before or after your "monitor" changes > the step size. Using TSSetPostStep is better in that regard because it > is consistent. >

1 0

Re: [petsc-dev] Fwd: [petsc-maint #70675] update of the timestep in theta method
by Vijay S. Mahadevan 18 Apr '11

18 Apr '11

If you provide TSMonitorStep or something along those lines, I do believe this could be solved. And the interface should also hold for future modifications where somebody comes up with a custom adaptor to modify time steps based on physics restrictions/a-priori error estimates/adaptive posteriori errors etc.. I have used the TSSetTimeStep with a post-step monitor function before but I feel knowing the current and previous step solutions would be relevant arguments for the user to decide an optimal step. My 2 cents. Vijay On Mon, Apr 18, 2011 at 2:15 PM, Jed Brown <jed(a)59a2.org> wrote: > On Mon, Apr 18, 2011 at 21:06, Jeremy Riousset <riousset(a)gatech.edu> wrote: >> >> What difference between this and the use of TSSetTimeStep? > > Where would you like to call TSSetTimeStep() from? It doesn't make much > sense to call it from inside residual evaluation, for example. So you can > register a post-step function that looks at whatever it wants and calls > TSSetTimeStep().

1 0

Re: [petsc-dev] Fwd: [petsc-maint #70675] update of the timestep in theta method
by Jed Brown 18 Apr '11

18 Apr '11

On Mon, Apr 18, 2011 at 21:17, Jeremy Riousset <riousset(a)gatech.edu> wrote: > I was thinking to call it in the Monitor function. You can also do that, but then the output of other monitors is dependent on whether it comes before or after your "monitor" changes the step size. Using TSSetPostStep is better in that regard because it is consistent.

1 0

Re: [petsc-dev] Fwd: [petsc-maint #70675] update of the timestep in theta method
by Jed Brown 18 Apr '11

18 Apr '11

On Mon, Apr 18, 2011 at 21:06, Jeremy Riousset <riousset(a)gatech.edu> wrote: > What difference between this and the use of TSSetTimeStep? Where would you like to call TSSetTimeStep() from? It doesn't make much sense to call it from inside residual evaluation, for example. So you can register a post-step function that looks at whatever it wants and calls TSSetTimeStep().

1 0

Re: [petsc-dev] Fwd: [petsc-maint #70675] update of the timestep in theta method
by Jed Brown 18 Apr '11

18 Apr '11

On Mon, Apr 18, 2011 at 20:19, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > I was wondering if I could adjust the time step as the simulations run > > using DAEs & theta method. > We have longer term plans to add a proper interface for using adaptive controllers with all the methods in TS, but TSTheta is not currently set up for it so you have to do something slightly dirty. You can use TSSetPostStep() to register a callback which you can use to modify the step size. When this function was added in 2001, it didn't have a context: changeset: 20:3a389a7b1f12 user: knepley(a)spinetta.mcs.anl.gov date: Thu Oct 11 06:07:09 2001 -0600 summary: bk-changeset-1.21.1.2 It should definitely get a context. Using TSPostStep() to change the step size is actually not ideal because (in this case) the monitor will see the wrong time step. Alternatively, you can use TSSetPreStep() to choose the step size. Unfortunately, it has confusing semantics near the end of the interval. for (i=0; i<ts->max_steps; i++) { if (ts->ptime + ts->time_step > ts->max_time) break; ierr = TSPreStep(ts);CHKERRQ(ierr);

1 0

Fwd: [petsc-maint #70675] update of the timestep in theta method
by Barry Smith 18 Apr '11

18 Apr '11

Begin forwarded message: > From: Jeremy Riousset <riousset(a)gatech.edu> > Date: April 18, 2011 1:17:23 PM CDT > To: petsc-maint <petsc-maint(a)mcs.anl.gov> > Subject: [petsc-maint #70675] update of the timestep in theta method > Reply-To: petsc-maint(a)mcs.anl.gov, Jeremy Riousset <riousset(a)gatech.edu> > > Hi, > > I was wondering if I could adjust the time step as the simulations run > using DAEs & theta method. > > best, > > --Jeremy >

1 0