Adding new Saint-Venant-Exner coupled finite volume solver for sediment transport in 1D: Arrakis

See merge request !234

[mascaret][courlis] Bug fix in mascaret.dico (following merge !238)

See merge request !239

This fixes a bug on Windows that caused the contour of some meshes to be
extracted incorrectly.

[VnV] Changes for the IMDC system

See merge request !237

Alexander BREUGEM authored 2 months ago and Chi-Tuân Pham committed 2 months ago

+ add missing med tag for the T3D lock-exchange sensitivity study

(1 argument forgotten for GAIA + arguments for T2D and T3D called with optional
syntax)

[fix] Time Series extraction for longitude/latitude meshes across IDL
(International Date Line)

Closes #1034

See merge request !202

Thomas SAILLOUR authored 2 months ago and Chi-Tuân Pham committed 2 months ago

(International Date Line)

MASCARET new keyword VARIABLES_PRECISION

Closes #1078

See merge request !238

- NERD for the advection of velocities (as for tracers) rather than
characteristics to decrease differences between sequential and parallel runs
- use of the non-hydrostatic version (but more CPU time consuming)
and conjugate gradient + preconditiong = 34 = 2*17 to solve PPE (more efficient
choice than GMRES here)
- use of preconditioning 34 = 2*17 for the diffusion of tracers + velocities
so that accuracy for the diffusion of tracers can be 1.E-15 without significant
increase of CPU time
- increase of MAXIMUM NUMBER OF ITERATIONS FOR ADVECTION SCHEMES to 100

Update of epsilons (1 order of magnitude less than before for tracers)
Smaller GRAPHIC PRINTOUT PERIOD to draw time series for figure

(Intel Turbo Boost on EDF clusters)

- TREATMENT OF NEGATIVE DEPTHS = 2 (new default value since release 9.0)
to improve mass conservation (water depth + temperature, water depth up to
machine accuracy)
- solver and accuracy for propagation set to default (GMRES + 1.E-8)
- accuracy for diffusion of tracers can be set to 1.E-15 without significant
increase of CPU time
- preconditioning for diffusion = 34 = 2*17 for tracers + velocities for
more efficiency (less iterations)
- solver and preconditioning for PPE = conjugate gradient (1) and 34
- non-hydrostatic version (new default since 8.0)

- default options to solve propagation (Krylov space dimensions, accuracy)
without significant increase of CPU time
- no diffusion along x and y with keyword
COEFFICIENT FOR HORIZONTAL DIFFUSION OF VELOCITIES = 0. rather than
SCHEME FOR DIFFUSION OF VELOCITIES = 0

This allows CPUs to operate at a higher clock rate, to speed up CPU-bound calculations

- SOLVER FOR PROPAGATION = 7 (GMRES = default) as quite more efficient than
old choice 2 (conjugate residual) or 1 (conjugate gradient)
This choice enables to set ACCURACY FOR PROPAGATION to its default value 1.E-8
rather than 1.E-6 with quite the same CPU times

- TIDAL FLATS = YES (default value) to improve mass conservation (water depth
+ tracer, water depth up to machine accuracy)
- SOLVER FOR PROPAGATION = 7 (GMRES = default) as the most efficient
- with theses choices of solvers, accuracies for propagation and PPE can be set
to default = 1.E-8
- ==> less differences between sequential and parallel runs

- TREATMENT OF NEGATIVE DEPTHS = 2 (new default value since release 9.0)
to improve mass conservation (water depth + temperature, water depth up to
machine accuracy)
- SCHEME FOR ADVECTION OF VELOCITIES = 5 (PSI) rather than 1 (characteristics)
for better mass conservation + reduced differences between seq/par runs
- use of gradient conjugate + preconditioning 34 = 2*17 to solve PPE, more
efficient than GMRES for this example
- accuracies for diffusion of velocities/tracers/k-epsilon + PPE set to
1.E-15 as easily and efficiently reached with this set-up

- TIDAL FLATS = YES (default value) to improve mass conservation (water depth
+ tracer, water depth up to machine accuracy)
- use preconditioning 34 for the diffusion of velocities (a little bit more
efficient)

- use of gradient conjugate + preconditioning 34 = 2*17 to solve PPE, more
efficient than GMRES or CGSTAB (previous choice) for this example

- use of gradient conjugate + preconditioning 34 = 2*17 to solve PPE, more
efficient than GMRES for this example
- switch from characteristics to defaul advection scheme for velocities (LIPS)
to decrease differences between sequential and parallel runs
- use preconditioning 34 for the diffusion of velocities (a little bit more
efficient)
- default accuracies for propagation and PPE with this setting of solvers
(1.E-8)

- TREATMENT OF NEGATIVE DEPTHS = 2 (new default value since release 9.0) and
TIDAL FLATS = YES (default value) to improve mass conservation of water
- use of gradient conjugate + preconditioning 34 = 2*17 to solve PPE, more
efficient than GMRES for this example
- accuracy for propagation decreased to 1.E-15 (improved differences between
sequential and parallel runs)
- switch to non-hydrostatic version

Vertical velocity W along z, rather than horizontal velocity V along x

- SOLVER FOR DIFFUSION OF VELOCITIES = 1 (conjugate gradient)
+ PRECONDITIONING FOR DIFFUSION OF VELOCITIES = 34 (= 2*17) rather than GMRES
- accuracy for propagation and diffusion of velocities = 1.E-15 easily reached
with previous choices
- ACCURACY FOR PPE set to 1.E-9

- SOLVER FOR PROPAGATION = 7 (GMRES) + SOLVER FOR PPE = 1 (conjugate gradient)
+ PRECONDITIONING FOR PPE = 34 (=2*17) rather than conjugate residual (= 2) for
propagation and GMRES for PPE
- accuracies for propagation and PPE set to default values (1.E-8)

 - TREATMENT OF NEGATIVE DEPTHS = 2 (new default value since release 9.0) and
TIDAL FLATS = YES (default value) to improve mass conservation
- use of gradient conjugate + preconditioning 34 = 2*17 to solve PPE, more
efficient than GMRES for this example
- use of preconditioning 34 for diffusion of every variable (more efficient)
- accuracy for PPE and propagation decreased from 10^-10 to 10^-12

(including non-hydrostatic version: no differences except for W, light for W)
IMPLICITATION FOR VELOCITIES has no impact

- TREATMENT OF NEGATIVE DEPTHS = 2 (new default value since release 9.0) and
TIDAL FLATS = YES (default value) to improve mass conservation of water
in particular for prismatic elements (also chosen for tetrahedra)

- TREATMENT OF NEGATIVE DEPTHS = 2 (new default value since release 9.0) and
TIDAL FLATS = YES (default value) to improve mass conservation of water
- use of gradient conjugate + preconditioning 34 = 2*17 to solve PPE, more
efficient than GMRES for this example (very few iterations are needed)
- ACCURACY FOR PROPAGATION = 1.E-15 as easily solved (only 2 iterations needed)
- ACCURACY FOR PPE = 1.E-9 (less than 10 iterations needed)
- PRECONDITIONING FOR DIFFUSION OF K-EPSILON = 34 to improve efficiency
(turbulent variable)
- PRECONDITIONING FOR DIFFUSION OF VELOCITIES set to default
(not understood why 0 previously, no clear differences)

- TREATMENT OF NEGATIVE DEPTHS = 2 (new default value since release 9.0) and
TIDAL FLATS = YES (default value) to improve mass conservation of water
- use of gradient conjugate + preconditioning 34 = 2*17 to solve PPE, more
efficient than GMRES for this example (very few iterations are needed)
- ACCURACY FOR DIFFUSION OF VELOCITIES = 1.E-15 as easily solved
+ PRECONDITIONING FOR DIFFUSION OF VELOCITIES = 34 to improve efficiency

- TREATMENT OF NEGATIVE DEPTHS = 2 (new default value since release 9.0) and
TIDAL FLATS = YES (default value) to improve mass conservation of water

- TREATMENT OF NEGATIVE DEPTHS = 2 (new default value since release 9.0)
to improve mass conservation of water + tracer
- use of gradient conjugate + preconditioning 34 = 2*17 to solve PPE, quite
more efficient than GMRES for this example
- switch to non-hydrostatic version (default) but no significant differences