Welcome, Guest
Username: Password: Remember me
Forgot your password? Forgot your username? Create an account
Forum
System Modules
TELEMAC-2D
NaN problem after 31 of simulation
Flat Threaded

TOPIC: NaN problem after 31 of simulation

NaN problem after 31 of simulation 12 years 6 months ago #4437

  • mtayel
  • mtayel's Avatar
Hello for all,
the following error appear to me after 31 days of simulation with wind
I try different numerical solvers but the error also appear as NaN in all of the flux boundaries

ITERATION 45180 TIME: 31 D 9 H 0 MN 0.0000 S ( 2710800.0000 S)
LIQUID BOUNDARY: Q(1) = 171.290000000000
LIQUID BOUNDARY: Q(2) = 53.7600000000000
LIQUID BOUNDARY: Q(3) = 191.375000000000
LIQUID BOUNDARY: Q(4) = 486.625000000000

ADVECTION STEP

DIFFUSION-PROPAGATION STEP
GMRES (BIEF) : 3 ITERATIONS, RELATIVE PRECISION: 0.2755447E-05

ADVECTION STEP

DIFFUSION-PROPAGATION STEP
GMRES (BIEF) : 1 ITERATIONS, RELATIVE PRECISION: 0.2825900E-05

BALANCE OF WATER VOLUME
VOLUME IN THE DOMAIN : 0.5326128E+14 M3
FLUX BOUNDARY 1: 171.1030 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 2: 53.72187 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 3: 191.2279 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 4: 486.1909 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 5: -1450927. M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 6: -423740.3 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 7: 966271.1 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 8: 8537379. M3/S ( >0 : ENTERING <0 : EXITING
RELATIVE ERROR IN VOLUME AT T = 0.2711E+07 S : 0.1471624E-06

METEO DATA RECORD NB= 755
(SUB. METEO) METEO DATA READ AT TIME = 2714400

=============================================================================

ITERATION 45210 TIME: 31 D 9 H 30 MN 0.0000 S ( 2712600.0000 S)
LIQUID BOUNDARY: Q(1) = 171.280000000000
LIQUID BOUNDARY: Q(2) = 53.6250000000000
LIQUID BOUNDARY: Q(3) = 191.396000000000
LIQUID BOUNDARY: Q(4) = 485.854000000000

ADVECTION STEP

DIFFUSION-PROPAGATION STEP
GMRES (BIEF) : ALGORITHM FAILED R= 0.000000
IF THE MATRIX IS DIAGONAL, CHOOSE
THE CONJUGATE GRADIENT SOLVER.
GMRES (BIEF) : 1 ITERATIONS, RELATIVE PRECISION: NaN

ADVECTION STEP

DIFFUSION-PROPAGATION STEP
GMRES (BIEF) : ALGORITHM FAILED R= 0.000000
IF THE MATRIX IS DIAGONAL, CHOOSE
THE CONJUGATE GRADIENT SOLVER.
GMRES (BIEF) : 1 ITERATIONS, RELATIVE PRECISION: NaN

BALANCE OF WATER VOLUME
VOLUME IN THE DOMAIN : 0.8040356+103 M3
FLUX BOUNDARY 1: 171.0763 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 2: 53.58509 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 3: 191.2504 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 4: 485.3973 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 5: -905212.7 M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 6: 1697381. M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 7: 1497763. M3/S ( >0 : ENTERING <0 : EXITING
FLUX BOUNDARY 8: 5363698. M3/S ( >0 : ENTERING <0 : EXITING
RELATIVE ERROR IN VOLUME AT T = 0.2713E+07 S : -0.3333333E-01
Attachments:
The administrator has disabled public write access.

Re: NaN problem after 31 of simulation 12 years 6 months ago #4438

  • mtayel
  • mtayel's Avatar
hello,
after changing the numerical scheme the same error appear
please see the attachment
Attachments:
The administrator has disabled public write access.

Re: NaN problem after 31 of simulation 12 years 6 months ago #4440

  • jmhervouet
  • jmhervouet's Avatar
Hello again,

I looked a the listing more closely, it is rather:

SOLVER : 1

that you must try, because it is the diffusion-propagation step which is involved. However there could be something else occuring in your computation a that time.

JMH
The administrator has disabled public write access.

Re: NaN problem after 31 of simulation 12 years 6 months ago #4439

  • jmhervouet
  • jmhervouet's Avatar
Hello,

You should not ask for the GMRES solver for a diffusion equation, as says the listing (when the matrix of the diffusion equation tends to a diagonal, e.g. if the diffusion coefficient tends to 0, it causes the GMRES method to crash, as paradoxically, GMRES is suited for very difficult linear systems, but by construction will fail if given a diagonal to invert). So choose:

SOLVER FOR DIFFUSION OF TRACERS : 1

I cannot prohibit GMRES for diffusion because it is efficient when the SUPG advection scheme is asked.

I hope it solves the matter.

Jean-Michel Hervouet
The administrator has disabled public write access.

Re: NaN problem after 31 of simulation 12 years 6 months ago #4443

  • mtayel
  • mtayel's Avatar
hello,
thanks for your care
i have two wind results from two wind models over the north sea, one result is for remo and the other from clm model
- i ran telemac succssefuly with remo wind for 40 days (even for the whole year of 2006) with cashipocas1.txt and results cashipo1.r .
- then i just change the wind to clm modle (metwind2) and ran the same cas parameters, but the error appear after 31 days of simulations (cashipocas2.r)
- then i changed the solver as you advised me to 1 but the simulation hang for 4 hours and stoped without any advancement cashipocas3.r

I have checked the wind data a lot but no error in it een checked if there is missing values.

please if there are any ideas or hints to solve this probleme since i think it comes from numerical parameters
thanks for your help
Attachments:
The administrator has disabled public write access.

Re: NaN problem after 31 of simulation 12 years 6 months ago #4444

  • mtayel
  • mtayel's Avatar
here is the results
Attachments:
The administrator has disabled public write access.

Re: NaN problem after 31 of simulation 12 years 6 months ago #4445

  • jmhervouet
  • jmhervouet's Avatar
Hello,

I would suggest that you try the following changes :

CONTINUITY CORRECTION : YES (this is just for better evaluating the fluxes when depth is imposed, it is mandatory to have a good mass conservation and tracer conservation).

A conservative advection scheme for tracers :

TYPE OF ADVECTION : 1;5;3 (or 1;5;13 if you have tidal flats) with diffusion coefficient of tracers set to 0. this should give mass conservation.

A number of changes to have a better stability :

SUPG OPTION ; 0;0 (effect removed and replaced by the following)
COMPATIBILITY OF FREE SURFACE GRADIENT : 0.9 (you can go down to 0. to stabilize)

MASS-LUMPING ON H : 1.
MASS-LUMPING ON VELOCITIES : 1. (but done automatically by the program)

Then if you still have a crash after a long computation time I would investigate the possibility of errors due to your specific programming in parallel mode, or try to do a restart just before the crash (if the restart works it gives a hint of some parallel problem).

With best regards,

Jean-Michel Hervouet
The administrator has disabled public write access.

Re: NaN problem after 31 of simulation 12 years 6 months ago #4448

  • mtayel
  • mtayel's Avatar
hello,
i have tried the suggestions you give to me but the simulation also stopped, please see cas4 and cas4r.
i have changed the advection type an try again, please see cas1,cas2,cas3.
attached also the fortran file please feel free if it has any errors.

note: the simulation worked with remo wind data without any problems

thanks for your help
Attachments:
The administrator has disabled public write access.

Re: NaN problem after 31 of simulation 12 years 6 months ago #4454

  • c.coulet
  • c.coulet's Avatar
  • OFFLINE
  • Moderator
  • Posts: 3722
  • Thank you received: 1031
Hi

Just an idea after having a look to the fortran file.
In the meteo subroutine, you search the variable "WIND_X M/S". Keep in mind that in telemac the name is in the first 16 characters and the unit after (between 17 and 32).
So maybe you don't read correctly the wind...

Regards
Christophe
The administrator has disabled public write access.

Re: NaN problem after 31 of simulation 12 years 6 months ago #4458

  • mtayel
  • mtayel's Avatar
Hi
thank you all for real help
yes it is now working without any problems
i just change the names as c.coulet said to be of length 16. also may be the space inside the names of the variables cause the problem.

at last if you suspect any other errors inside the fortran file please say to me.

the thing that I cannot believe until now it was working perfectly on the remo data before.
thanks
The administrator has disabled public write access.
Flat Threaded
Moderators: pham
Forum
System Modules
TELEMAC-2D
NaN problem after 31 of simulation

The open TELEMAC-MASCARET template for Joomla!2.5, the HTML 4 version.