Hi Joakim,

There is no function like runScheduleADI(). Adjoints are not yet supported, but 
we have committed to adding such support.

Perhaps we should have a short video meeting to discuss these things? If so, 
please send me an off-list email (anyone else who would like to join the 
discussion is welcome, please email me).

Atgeirr



28. jun. 2017 kl. 15.36 skrev Joakim R. Andersen 
<[email protected]<mailto:[email protected]>>:

Dear Atgeirr and Alf,

  Thanks for your quick response.

  If you are familiar with MRST I would like to have access to a function 
similar to runScheduleADI.  Further, if possible I would like to have access or 
to build adjoint variables for the simulation, however I'm aware that this last 
functionality is not implemented yet.

   At the moment I would like to implement an interface to the fully-implicit 
simulators that given the initial states of the reservoir (pressure, 
saturations, rs, rv) provides the intermediate states for all the actual 
simulated timesteps (not only report steps) together with their Jacobians at 
convergence.

    In addition to that, I will need to build the Jacobians of the implicit 
steps with respect to the initial conditions and well controls settings to 
implement gradient-based optimization methods.  I am not sure if these are 
readily available and not interfaced or if I have to figure out how to build 
them.


- Joakim


2017-06-28 4:56 GMT-03:00 Atgeirr Rasmussen 
<[email protected]<mailto:[email protected]>>:

28. jun. 2017 kl. 09.32 skrev Alf Birger Rustad 
<[email protected]<mailto:[email protected]>>:

Hi Joakim,

There is currently no library, we have basically viewed ourselves as 
downstream. Still, the code that makes up the application flow is modular. We 
have gone through great lengths to make the various parts reusable, so to what 
extent you can use it for your tasks really depends on your needs.

To customize the simulator code currently you may have to go somewhat deep into 
it. It is not necessarily hard, but it can be difficult to find the right place 
to modify. If you would like to share a little about what you want to 
accomplish, we could point you in the right direction! In the long term, I'd 
like to see a higher-level API that is well-suited to scripting, but we do not 
have that at the moment.

Atgeirr



_______________________________________________
Opm mailing list
[email protected]
http://opm-project.org/cgi-bin/mailman/listinfo/opm

Reply via email to