Hi, I'd just like to add that I think it makes a lot of sense to experiment in a less complex setting first, before adding these capabilities to Flow. Also, since we are going to refactor the Flow code to enable easier extension it may be a moving target for a while. In the long term, I definitely think we will see adjoint support in Flow.
I did not know about TAO before it was mentioned here, looking forward to see how that plays out! Atgeirr 20. apr. 2015 kl. 09:05 skrev Alf Birger Rustad <[email protected]>: > Hi Tobias, > > Access to adjoints is interesting for more of us. We have not been able to > prioritise it, but getting adjoint output is quickly moving up the list. > However, I would be most interested in seeing the functionality in our most > capable simulator, which today is flow (found within the opm-autodiff > repository). > > Cheers, > Alf > > From: Opm [[email protected]] on behalf of Tobias Ritschel > [[email protected]] > Sent: Sunday, April 19, 2015 1:08 PM > To: [email protected] > Subject: Re: [OPM] Opm Digest, Vol 31, Issue 5 > > Alright, I'll continue with the sim_example.cpp, I just wanted to make sure. > I think I'll have a look at the MRST code, I guess it should be a good > starting point. Some of my fellow students are using MRST for optimization > and it seems to work quite well, but my master thesis (which I am currently > working on) is concerned with high performance computing so I'll stick to OPM > and use PETSc/TAO for the optimization. For now I am only concerned with > two-phase immiscible flow but I'll let you know of my progress. > > Kind Regards > > On 19 April 2015 at 12:00, <[email protected]> wrote: > Send Opm mailing list submissions to > [email protected] > > To subscribe or unsubscribe via the World Wide Web, visit > http://www.opm-project.org/mailman/listinfo/opm > or, via email, send a message with subject or body 'help' to > [email protected] > > You can reach the person managing the list at > [email protected] > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of Opm digest..." > > > Today's Topics: > > 1. Derivatives from solvers in opm-autodiff (Tobias Ritschel) > 2. Re: Derivatives from solvers in opm-autodiff (Atgeirr Rasmussen) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Sat, 18 Apr 2015 15:11:26 +0200 > From: Tobias Ritschel <[email protected]> > To: [email protected] > Subject: [OPM] Derivatives from solvers in opm-autodiff > Message-ID: > <CAJcuCB34EoqUw6c8-10_W=zttsm2sugm0+yzduwniqs4qrk...@mail.gmail.com> > Content-Type: text/plain; charset="utf-8" > > Hi > > I would like to use OPM for optimal control of injection and production. As > such I need the derivatives/sensitivities of the a given objective function > with respect to controls. This can be calculated using the discrete adjoint > method which in turn requires derivatives of the residuals with respect to > the dynamic variables and controls. > > I can see that there are some solvers such as SimulatorIncompTwophaseAd and > ImpesTPFAAD, which I guess uses Automatic Differentiation. Is it possible > to extract relevant derivatives from these functions, e.g. by using the > function LinearizedBlackoilResidual and FullyImplicitBlackoilSolver? > > I am currently modifying the sim_simple.cpp example but if the > functionality is implemented in one of the solvers in any way, I would > prefer to use them. > > Kind Regards > Tobias > -------------- next part -------------- > An HTML attachment was scrubbed... > URL: > <http://www.opm-project.org/pipermail/opm/attachments/20150418/5a19d9f9/attachment-0001.html> > > ------------------------------ > > Message: 2 > Date: Sun, 19 Apr 2015 08:18:25 +0000 > From: Atgeirr Rasmussen <[email protected]> > To: OPM Mailing List <[email protected]> > Subject: Re: [OPM] Derivatives from solvers in opm-autodiff > Message-ID: <[email protected]> > Content-Type: text/plain; charset="us-ascii" > > Hi! > > We have not yet implemented the adjoint method in OPM. > We do indeed use Automatic Differentiation (AD) which should make it > relatively easy (compared to without AD) to implement adjoints. If you'd > implement this I would certainly be interested! > > It should also be noted that if you have access to Matlab, there are > adjoint implementations in MRST's solvers. > > Atgeirr > > > 18. apr. 2015 kl. 15:11 skrev Tobias Ritschel <[email protected]>: > > > Hi > > > > I would like to use OPM for optimal control of injection and production. As > > such I need the derivatives/sensitivities of the a given objective function > > with respect to controls. This can be calculated using the discrete adjoint > > method which in turn requires derivatives of the residuals with respect to > > the dynamic variables and controls. > > > > I can see that there are some solvers such as SimulatorIncompTwophaseAd and > > ImpesTPFAAD, which I guess uses Automatic Differentiation. Is it possible > > to extract relevant derivatives from these functions, e.g. by using the > > function LinearizedBlackoilResidual and FullyImplicitBlackoilSolver? > > > > I am currently modifying the sim_simple.cpp example but if the > > functionality is implemented in one of the solvers in any way, I would > > prefer to use them. > > > > Kind Regards > > Tobias > > _______________________________________________ > > Opm mailing list > > [email protected] > > http://www.opm-project.org/mailman/listinfo/opm > > > > > ------------------------------ > > Subject: Digest Footer > > _______________________________________________ > Opm mailing list > [email protected] > http://www.opm-project.org/mailman/listinfo/opm > > > ------------------------------ > > End of Opm Digest, Vol 31, Issue 5 > ********************************** > > > > ------------------------------------------------------------------- > The information contained in this message may be CONFIDENTIAL and is > intended for the addressee only. Any unauthorised use, dissemination of the > information or copying of this message is prohibited. If you are not the > addressee, please notify the sender immediately by return e-mail and delete > this message. > Thank you > _______________________________________________ > Opm mailing list > [email protected] > http://www.opm-project.org/mailman/listinfo/opm _______________________________________________ Opm mailing list [email protected] http://www.opm-project.org/mailman/listinfo/opm
