Solver diverging after some iteration

[elinlarsson744-ux]

Hi again, I am using this case file

{
  "version": 1.0,
  "case": {
    "mesh_file": "new_mesh.nmsh",
    "output_at_end": true,
    "output_boundary": true,
    "output_checkpoints": false,
    "nsamples": 10,
    "numerics": {
      "polynomial_order": 5,
      "time_order": 3,
      "dealias": true
    },
    "fluid": {
      "scheme": "pnpn",
      "Re": 5600.0,
      "boundary_conditions": [
        {
          "type": "no_slip",
          "zone_indices": [1]
        },
        {
          "type": "velocity_value",
          "value": [3.0, 0.0, 0.0],
          "zone_indices": [2]
        },
        {
          "type": "outflow+dong",
          "zone_indices": [3]
        }
      ],
      "initial_condition": {
        "type": "uniform",
        "value": [3.0, 0.0, 0.0]
      },
      "velocity_solver": {
        "type": "cg",
        "preconditioner": {
          "type": "jacobi"
        },
        "projection_space_size": 0,
        "absolute_tolerance": 1e-7,
        "max_iterations": 800
      },
      "pressure_solver": {
        "type": "gmres",
        "preconditioner": {
          "type": "hsmg"
        },
        "projection_space_size": 5,
        "absolute_tolerance": 1e-3,
        "max_iterations": 800
      },
      "output_control": "simulationtime",
      "output_value": 0.5
    },
    "time": {
      "end_time": 45.0,
      "variable_timestep": true,
      "target_cfl": 0.4,
      "max_timestep": 1e-4
    },
    "simulation_components": [
      {
        "type": "lambda2"
      },
      {
        "type": "curl",
        "fields": ["u", "v", "w"]
      }
    ]
  }
}

but unfortunately the solution diverged after some, can someone help me with the case file what is incorrect? I am sharing the log and error log below. Thanks

neko_hill_4209266.log

neko_hill_error_4209266.log

[vbaconnet]

This could be due to many different things, but the first thing I'd do in this situation do is to have a look at the flow field: do you see any abnormal values? If so, where are they located (on the boundaries?)? Check the walls, is the no-slip imposed properly? Does the flow flow in the right direction? In other words: does the flow look like what you expect? You may attach some snapshots here for us to look at if you are unsure.

On another note, for this case I would recommend you set the projection space for the pressure solver to 0. Notice that projections are not being used at the beginning of your simulation to "get the residuals down", and the variable time-stepping is resetting your projection space every time dt is changed. Once the simulation has more or less reached a state of "converged" residuals, notice how the projections are not (or barely) helping to get the pressure residuals down. Look at the Original res. of the projection and compare it to the Start residual of the Pressure solver.

------------------------------------------------------------------------------
 Step =        8355 t =   0.4915807E+00                             [   1.09% ]
 ------------------------------------------------------------------------------

   --------Preprocessing---------
   CFL:  0.3762  dt: 1.8532825E-05

   ------------Fluid-------------
   8355         | Projection: Pressure, Vectors:   5, Original res.: 0.1952E-02

   Step:        | Field              Iters:   Start residual:   Final residual:
   8355         | Pressure                3   0.230727092E-02   0.880441194E-03
   8355         | X-Velocity              3   0.221742803E-02   0.459066499E-08
   8355         | Y-Velocity              2   0.730544847E-02   0.765104740E-07
   8355         | Z-Velocity              2   0.373816903E-03   0.269589464E-07
   Fluid step time (s):     0.1354437E+00

   ---------Step summary---------
   Total time for step     8355 (s):   0.1513643E+00
   Total elapsed time (s):             0.1568080E+04

It saves you at most 1 or 2 pressure iterations. In my experience projections are only helpful to get residuals down at the beginning of a simulation if it is a difficult case to get to "converge". In your case you don't need them.

I also see from your case file that you only have three boundary zones: 1,2 and 3. I would expect more boundaries for a channel case: two walls, inflow and outflow. I suspect that the two zones corresponding to the walls got merged into labeled zone 1, this is a known behavior of the rea2nbin converter.

I don't think that is the source of your problem, but if you'd want to try to separate the zones into two separate boundaries, you can set a "dummy" label ( e.g. "s") in the python script that I gave you to artificially separate the two wall zones.

For example:

from mshconvert import *
convert('new_mesh.msh',
        bcs={
            10:'W',
            13:'W', # <-- replace one of the 'w'/'W' with 's', a dummy label that rea2nbin will still recognize
            12:'v',
            15:'o'})

That way, the rea2nbin converter will "see" two different boundaries and put them into different labeled zones.

Hopefully this can help you move in the right direction, but unfortunately there is no recipe for how to solve these things.

[elinlarsson744-ux]

Okay thank you very much for the suggestions, though I have changed the inlet velocity now to bit lower and the simulation works now, I started to use neko very recently so these are just cases to get to know the program. But I have one question, @vbaconnet when I run the simulation can I export velocity field at some plane as a contour and monitor the contours?

[vbaconnet]

when I run the simulation can I export velocity field at some plane as a contour and monitor the contours?

Yes, please have a look at the probes documentation page.

You can generate a set of points yourself based on your domain and sample the fields u,v,w at a given frequency.

[elinlarsson744-ux]

Ahh, Okay I see Thanks a lot for your help