Option for restart mode for GAIA like in TELEMAC-2D and TELEMAC-3D

As already done for TELEMAC-3D and more recently for TELEMAC-2D, a restart mode may be useful to do continued computations without loss of accuracy.

Moreover, it secures the saving of useful/mandatory variables to correctly restart a computation coupled with GAIA.

added module:gaia type:feature labels

Up to v8p4, some variables are forced to be written in the RESULTS FILE of GAIA, even if not asked by the user with the keyword VARIABLES FOR GRAPHIC PRINTOUTS. Generating a dedicated RESTART FILE can decrease the size of RESULTS FILE. This RESTART FILE is the assumed to store every variable useful/mandatory to restart the computation without loss of accuracy.

In the continued computation step, what is necessary to be read to initialise the next computation is to be correctly done (e.g. overloading previous initial conditions).

The 1st implementation is based on Rajae and Sara's feedback to save a few variables. The GAIA user manual indicates:

• the bottom (B),
• the layer thickness (*ES),
• non erodable bottom (R),
• the masses of sand (S) if modelling sand,
• the masses of mud (M) if modelling mud.

Based on what is available in v8p4 in lecdon_gaia subroutine for mud only:

• mud concentration (*CONC),
• critical erosion shear stress of the mud (*TOCEMU),
• Partheniades erosion coefficient (*PARTHE),
• mass transfer (*MTRANS) with BED MODEL = 2 (multilayers with consolidation).

I do not remember why, but additional variables are also stored in the RESTART FILE currently:

• solid discharges (M, N, P),
• evolution (E),
• friction coefficient (KS) not read by TELEMAC-2D?
• MU,
• D50,
• ZRL if coupling with NESTOR.

I wonder if every variable from the latest block is necessary for a perfect restart (ZRL is OK).

A shift of index of 1 seems to be a bug from index 11 to 23 when used with ALIRE array (in GAIA_INIT subroutine). Are variables with index 14 to 20 in NOMVAR_GAIA (M, N, P, E, KS, MU and D50) are really mandatory to be stored and read or could they be computed correctly with other variables?

The bottom variable has to be saved in the RESTART FILE of TELEMAC-2D when coupling with GAIA, otherwise the continued computation crashes.

Validation of a 1st implementation has started.

A perfect restart is OK for the following GAIA test cases (whole simulation or part depending of the CPU time):

• bosse-t2d (FE+FV),
• bump-t2d,
• hippodrome-t2d (OK for 1COs, 1NCOs, 1NCOb, 1NCOb_vf, 4NCOb and 4NCOb_vf BUT NOK for 1COs_vf and 4NCOb_strat_vf),
• mud_conservation-t2d,
• yen-t2d (OK for FE and FV, BUT NOK for multi layers).

Differences appear for restart in 2D for:

• cvsm-t2d : some additional are surely expected to be stored in the RESTART FILE,
• guenter-t2d,
• hippodrome-t2d : NOK for 1COs_vf and 4NCOb_strat_vf (for this last one, other variables to be stored?),
• sandpit-t2d (with the use of wave file),
• wilcock_crowe-t2d : depending on variables, some are OK, other are not good,
• yen-t2d : NOK multi layer.

Is there anybody who has ideas of what additional variables to save for some of the not working examples?

When coupling with TELEMAC-3D, 2 examples are nearly OK for restart without loss of accuracy:

• rouse-t3d,
• tidal_flats-t3d. For which the differences only appears for hydrodynamic variables (velocities V, W).

2 examples are currently NOK (hippodrome-t3d tested for 1COs and 1NCOs only + bosse-t3d) but there are already issues for restart computation using TELEMAC-3D alone.

3D to be investigated after solving some issues of rest for TELEMAC-3D alone.

Not tested yet:

• lyn-t3d,
• turbidity-3d.

Other 2D GAIA examples not tested yet:

• flume_bc-t2d: no reproducibility of results file, depending on the run.
• littoral-t2d-tom will be tested after improvement when coupling without TOMAWAC.

Good news for littoral-t2d-tom, it is OK, the only thing to do with current implementation is to store GLOBAL RESULT FILE FORMAT in double precision (when RESTART MODE = YES in the TELEMAC + GAIA steering files) and then read this file in double precision from TOMAWAC with 3 keywords (PREVIOUS COMPUTATION FILE, PREVIOUS COMPUTATION FILE FORMAT = SERAFIND or MED and NEXT COMPUTATION = YES) in the TOMAWAC steering file (+ other expected keywords for TELEMAC and GAIA as other test cases).

For yen multi layer examples, the loss of accuracy appears when the keyword ACTIVE LAYER THICKNESS is not set to default value. Any idea what additional variable(s) to save in that case?

Update: same behaviour for guenter-t2d and hippodrome-t2d with 4NCOb_strat_vf config. Each time: ACTIVE LAYER THICKNESS is set not to default value + several classes.

Associated branch called restart_gaia

Thanks to @rebekka.kopmann this feature is now OK for yen multi (every formula), guenter-t2d, hippodrome-t2d with 4NCOb_strat_vf config (no calls to BED1_UPDATE_ACTIVELAYER_HIRANO nor 2nd call to BED1_UPDATE in GAIA_INIT subroutine in case of continued computation). No regression for previously OK examples. It also works for new examples (sliding + yen with slope effect) created after Spring hackathon. The branch has been updated as well (and rebased with main).

marked this issue as related to #756 (closed)

mentioned in issue #756 (closed)

Issue firstly initiated in #756 (closed)

created branch 899-option-for-restart-mode-for-gaia-like-in-telemac-2d-and-telemac-3d to address this issue

mentioned in merge request !121 (merged)

closed with merge request !121 (merged)

mentioned in commit 5cde0487

Issue reopened as only partially solved (WIP for a few options).

reopened

mentioned in commit 9d4e4100