open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi, everyone
When I read the subroutine like telemac2d.f, I found the ADJO means IF YES: ACTIVATES THE ADJOINT MODE IN TELEMAC2D/PROPAG. But I have never seen the adjoint mode in user mannual. Can anyone tell me what it is? Or which book I can refer to?

Thanks.

Hello,

The adjoint mode is used in parameter estimation (so you can refer to this topic in the user manual). It computes Lagrange multipliers that are used in the parameter estimation theory. When the program is set in adjoint mode it goes backward in time and uses the results of a first normal computation. Now we are working on automatic differentiation compilers that will give automatically the adjoint mode of any program. The current adjoint mode in Telemac-2D was the result of a PhD.

WIth best regards,

Jean-Michel Hervouet

Thank you, JMH
I have read the mannual about parameter estimation in these days. So the adjoint mode is relative to determining an unknown physical parameter through comparing the results provided by TELEMAC-2D with the available measurements, is it right? And now it only can be used in the friction coefficient when using the Strickler or Chézy laws?

In addition, When the program is set in adjoint mode it goes backward in time, should I set a negative time step?

Hello,

You have well understood, and you do not need to specify a negative time step. You have an example called "estimation" that shows how things work, in this case a single friction coefficient is estimated and the "measurements" are a previous calculation.

With best regards,

Jean-Michel Hervouet

Hi,

Please can you explain what is meant by 'celerities' in the following statement in the user manual:

The cost function used must be indicated with the integer keyword COST FUNCTION which may have the value 1 (cost function based on the difference between depths and velocities, which is the default value) or 2 (cost function based on the difference between celerities and velocities).

The reason for my question is that I would like to establish if it is possible to use this parameter estimation/adjoint mode functionality for the application of optimizing the drag coefficient used to model tidal turbines in an array, where the cost function to maximize would be the array power, which is a function of the flow velocity within the array? In this case the array could be split into multiple zones, then the optimum drag coefficient in each zone would be estimated using parameter estimation to maximize the total power of the array.

Cheers

Danny

Hello,

In this case the celerity is just sqrt(g*h), so it is just like the depth, but with the dimension of a velocity. This is preferable to compute a cost function, otherwise variables with different dimensions like depth and velocity are summed in the cost function.

With best regards,

Jean-Michel Hervouet