open TELEMAC-MASCARET The mathematically superior suite of solvers

Dear all,

I would like to get your expertise on the following case. I am trying to run a sedimentological simulation but I am facing a problem of solution convergence (i.e. exceeding maximum iterations permissible).

For the moment, I've made a very simple case with no fortran file included and the setup is as follows: Only cohesive sediments are modelled. There is only one class of sediment, one bed layer and no mud consolidation model. Only suspended load has to be computed. The sisyphe run is internally coupled with telemac2d. Regarding the boundary conditions, a sinusoidal tidal elevation is prescribed by means of a liquid boundary file supplementing the boundary condition file (.cli) for the hydro BCs. Then, prescribed suspended sediment concentration (SSC) are imposed as boundary conditions in the same boundary section as for the hydrodynamic boundary conditions. Both initial condition of SSC and water elevation are uniform. Different friction coefficient are defined by zones in the computational domain which have been created in Blue Kenue and saved in the geometry file.

Since I am modelling a domain representing a mud flat with an average elevation situated at mean sea level, the wetting and drying process plays a big part in the computation since the domain is alternatively dried and wet during a tidal cycle, and I fear like this problem of exceeding maximum iterations is due to this wetting/drying process. As a matter of fact, if I choose another boundary condition making sure that the domain is always wet during the computation, then the simulation is able to go until the end. Regarding the wetting/drying option (so called tidal flats in telemac), at first, I was using the option 3 (with the inclusion of a porosity parameter) for telemac2d .cas file and the option 1 (correction on free surface gradient) for sisyphe .cas file. But it didn't work. Therefore I decided to use the same tidal flat option (i.e., option 1) for both steering files (t2d_hydro_01.cas and sis_susp_01.cas files) for a matter of consistency. But still it doesn't work. Another problem is that, by going through the log file, I also realised that after two time step the SSC drops to 0, which is strange knowing that I imposed initial and boundary concentration values.

Maybe although this case is rather simple, I may have skipped or forgot a step during the setup of the steering file, therefore I would be very grateful if you could take a quick look at my two cas files (i.e., telemac2d and sisyphe cas files), perhaps you could provide me with some hints on how to solve this problem.

You can find my file in attachement

Thank you for your help and sorry for this long post,

Jean-Philippe

Dear all,

Ok, I think I know why the concentration values drop to zero after two timesteps. It is because I activated the keyword equilibrium inflow concentration which recalculates the concentration values considering equilibrium conditions. This is why the values I imposed in the keywords initial suspension concentrations and concentration per class at boundaries were not selected. My apology for that.

Therefore I turned off the keyword equilibrium inflow concentration. I also imposed more realistic initial and boundary suspended concentrations as I realised that I made a mistake in the units used in the steering file (one needs to impose volume concentration which is dimensionless). By the way, can you confirm me that the relation between mass concentration and volume concentration is as follows: volume concentration [-] = Mass concentration [kg/m3] / Sediment density [kg/m3] ?

In any case, this still does not solve the problem of exceeding maximum iteration. And I saw that this porblem of convergence of the solution appears after 6h40min. This corresponds to the moment where the tide is half way lowering (during the ebb phase), exposing most of the domain to dry condition. Therefore, I am pretty sure this has to do with the wetting and drying process.

I would really appreciate if you could provide me with some hints on that latter point.

Kind regards,

Jean-Philippe

Dear all,

I have made further progress with my morphodynamic model as I am getting more familiar with the different options offered by the TELEMAC system. I am now able to run my case over long duration. However, as I mentioned in a previous post (Calculating dispersion coefficient in sisyphe) I have problems with convection and/or diffusion of the suspended sediment concentrations that I prescribed at the open boundary condition. Basically, all the sediment entering into the domain directly deposits near the boundary section.

I tried to play on the dispersion term but it has no effect. I tried as well to reduce the settling velocity but it has no effect either. What is strange is that my velocities are in the order of 0.1-0.2 m/s over the domain with locally 0.6-0.7 m/s. Knowing that my domain is 500 m long over 400 m large, these velocities should still be able to transport the suspended sediment further in my domain. I also applied a uniform and steady wind of 20 km/h in the direction where I want my suspended sediment concentration to be transported. This wind effect should enhance the transport of the suspended sediment concentration. But I do not notice any change. To make sure whether this was a physical or numerical problem, lately, I made a further test with imposing a settling velocity equal to 0. By this, one expects no sediment deposition. This what I actually see in my model when checking the listing printout. However, after 9 days, I get again this problem of convegence of the solution:

GRACJG (BIEF) : EXCEEDING MAXIMUM ITERATIONS: 100 RELATIVE PRECISION: NaN

I tried to reduce the sediment size from 100 microns to 50 microns (typical of cohesive sediments), in an attempt of reducing this deposition at the boundary section but my model shows the same error message after 9 days.
What do you think is the cause of this problem of convergence of solution ? I already reduce my time step to 1sec. My domain does not depict any high bathymetry gradient (it is basically a mud flat with a channel at the open boundary section which smooth bank slope). I lately checked some numerical parameters playing on solving the linear system (e.g. preconditioning). I use option 2 for the preconditioning. It worked pretty good for the hydrodynamic simulation. But when it comes to coupling with sisyphe, I get often this convergence problem. Do you think I should use another option? Or should I combine different preconditionings for fastening the convergence of the solution ? Is there any other numerical related keyboard that should make me get ride of this problem ?

I attached my t2d and sis cas file if you want to have a further look.







Thank you for your kind reply,

Jean-Philippe

Dear Jean-Philippe,
Did you take a look to the values of the shear velocity u*?
There are some very rough but useful expressions to estimate approximative time scales and travelling downstream distance at mean flow velocity that could help you to see the range of applicability of your problem.

For example, the transversal time scale can be estimated according to:
tt = W^2/hu*, with h the water depth and W the channel width and the length scale for longitudinal dispersion Xd = 500hu*/V, with V the mean flow velocity.

I hope it helps,

Pablo

Dear Pablo,

I was actually trying to play on reducing the settling velocity and sediment diameters, but looking at the result, it looks like the shear velocity is too big leading to a high critical shear stress for deposition. I will reduce it therefore and see if I can get a further transport of my suspended sediments as a result. I'll let you know.

Thanks

Jean-Philippe

Ok, it seems to work better now.

I also read that you cannot apply a morphological factor with suspended load as this type of transport requires the use of the same time step as for the hydrodynamics. However, is there any way you could recommend in order to accelerate the morphological computation?

Indeed, I need model long term period simulation and in my environment, typically, morphodynamics acts on a longer timescale compared with hydrodynamics, justifying therefore the use of a morphological factor.

Dear Jean-Philippe,
I'm glad to heard it is working better now.

Regarding the use of the morphological factor, it is true that it is not recommended for morphodynamic simulations involved suspended load.
I don't have many experience in the use of this acceleration techniques. Anyway, I would suggest the proceedings article by Knaape M. and Joustra R. (2012) Morphological acceleration factor: usability, accuracy and run time reductions, from the proceedings book of the last Telemac User's conference.

all the best,

Pablo

I will take a look. Thanks

Jean-Philippe,

There is not much difference between suspended load and bedload regarding the use of the morphological factor. It will just include the assumtpion that the suspended load is constant (usually the errors involved with this assumption are not too big).

However, as the paper tha pablo refers to (www.opentelemac.org/index.php/2012) explains, the use of the morphological factor as is at the moment assumes that there is no variation in the hydrodyanmic forcing during the period between two consequetive morphological updates. I think in literature this is referred to as tide stretching.

Michiel