|
Mushy Layer
1.0
|
This document explains how to use the mushy layer code for various physical applications.
It assumes you already have a compiled version of the code which you can run.
There are over a hundred options used in the code, which are stored on the AMRLevelMushyLayer class in one of two places:
AMRLevelMushyLayer.m_parameters which is a MushyLayerParams object. This is where most physical parameters (e.g. Rayleigh number, Lewis number) are stored.AMRLevelMushyLayer.m_opt, which is a struct of type MushyLayerOptions and contains options on how to solve the equations (e.g. solver thresholds, how to do projection, and lots of debugging options).Inputs are loaded in the code via the ParmParse class, e.g.
which would load some inputs like
We do most of this in either /mushy-layer/srcSubcycle/MushyLayerSubcycleUtils.cpp or /mushy-layer/src/MushyLayerParams.cpp.
Most of these options/parameters can just be left at their default values. The inputs file at /mushy-layer/examples/meltponds/inputsGrowSeaIce illustrates the main options you may wish to change, which are described in more detail below under some roughly suitable headings.
main.verbosity=2 larger verbosity means there will be more text output produced
main.output_folder=. folder to save plot/checkpoint files to a period . means the directory from which the code was executed
main.plot_prefix=plt prefix for plot files, which will then be like plt001234.2d.hdf5
main.chk_prefix=chk prefix for checkpoint files
main.plot_interval=n produce plot files every n steps (-1 = don't use this option)
main.checkpoint_interval=n produce checkpoints files every n steps
main.plot_period=0.005 time interval at which to write out plot files
main.debug=false set to true to write more fields to the plot files
main.cfl=0.1 max allowed CFL number
main.initial_cfl=2e-05 start simulations with a smaller CFL number. The timestep increases thereafter as determined by main.max_dt_growth
main.max_dt=0.1 max allowed timestep
main.max_dt_growth=1.05 max fractional increase in $ t$ allowed from one timestep to the next
main.max_step=10000 stop simulations after this many timesteps
main.max_time=10.0 stop simulations once this time has been reached
main.min_time=0.1 ensure simulations run until at least this time (even if a steady state condition is reached)
main.dt_tolerance_factor=1.01 Set the factor by which the current dt must exceed the new (max) dt for time subcycling to occur (i.e., reduction of the current dt by powers of 2).
main.num_cells=Nx Ny Nz e.g. main.num_cells=64 128 size of the grid on the coarsest mesh. Make sure each dimensions is a power of 2 unless you want the multigrid to be slow. Only need to specify 2 dimensions if the code is compiled for 2 dimensions. The final dimension is always the 'vertical' with respect to gravity.
main.domain_height=1.0 domain height (width is then calculated from the number of grid cells specified)
main.periodic_bc=1 0 whether to enforce periodic boundaries (1) or not (0) in each dimension
main.max_grid_size=256 split domain into boxes (for sending to different processors) of max size in any dimension given by this parameter.
See the separate document.
main.nondimensionalisation=0 choose nondimensionalisation, different options are: 0. Diffusive timescale, advective velocity scale
parameters.problem_type=0 can specify different problem types, which can tell the code to solve some of the equations differently. Lots of the options aren't really used anymore (and maybe don't work). Ones which definitely should work are in bold. 0. mushyLayer
parameters.compositionRatio=1.18 composition ratio, should be >= 1.0.
parameters.rayleighComp=500 rayleigh number for compositional differences. When just solving Darcy's equation, this is the mushy layer number. Otherwise, it's the fluid rayleigh number.
parameters.rayleighTemp=25.0 rayleigh number for temperature.
parameters.stefan=5.0 stefan number
parameters.K=1 ratio of solid/liquid heat diffusivity
parameters.heatConductivityRatio=1 ratio of solid/liquid heat conductivity
parameters.specificHeatRatio=1 ratio of solid.liquid specific heat
parameters.lewis=100 lewis number (liquid heat diffusivity/liquid salt diffusivity)
parameters.darcy=0.0 darcy number. Set to 0 to solve Darcy's equation rather than Darcy-Brinkman
parameters.prandtl=0 prandtl number
parameters.heleShaw=true constrain permeabilities by a Hele-Shaw cell
parameters.nonDimReluctance=0.1 inverse of the dimensionless Hele-Shaw permeabilty. Smaller means wider hele-shaw gap width. Alternatively, you can now specify the dimensionless Hele-Shaw cell permeability e.g. parameters.heleShawPermeability=10.0
parameters.nonDimVel=0.0 dimensionless frame advection velocity parameters.permeabilityFunction=2 mathematical form of the permeability function. Options include: 0. pureFluid (permeability=1)
parameters.waterDistributionCoeff=1e-5 water distribution coefficient
parameters.eutecticComposition=230
parameters.eutecticTemp=-23
parameters.initialComposition=30
parameters.liquidusSlope=-0.1
main.max_level=0 max allowed level for AMR simulations.
main.block_factor=8 all boxes must be divisible by at least this length in all dimensions.
main.use_subcycling=1 whether or not to use subcycling in time
main.ref_ratio=2 2 2 refinement ratios between successive levels
main.refine_thresh=0.1 refinement threshold
main.regrid_interval=8 8 8 regrid interval on each level
main.tag_buffer_size=4 number of cells by which to grow the set of cells tagged for refinement, before creating the new grids
main.fill_ratio=0.75 should be between 0 and 1. A value of 1 means the grids generated on refined levels will be as tightly grouped as possible to the cells tagged for refinement. A smaller value means that Chombo will cover a larger area on refined levels with grid cells.
main.grid_buffer_size=0 0 0 this is the 'padding' between grids on different levels of refinement
projection.eta=0.0 Freestream correction coefficient. Should be less than 1 for stability, but close to 1 for accuracy.
The projection solver has some tolerance specified by projection.solverTol. If the unprojected velocity has some divergence $d$, then the projected velocity will have some divergence of order $d $projection.solverTol. In reality, we care more about the absolute value of the final divergence than how much it is has been reduced. Therefore we introduce an option to adapt the solver tolerance to achieve a particular final divergence:
projection.adaptSolverParamsDivU = 1 turns on this option (set 0, or don't set at all, to not use this option) projection.divergence_tolerance = 1e-9 absolute divergence we wish to aim for
If you are struggling to achieve this final divergence, you can try increasing either the number of multigrid smooths or number of multigrid iterations:
projection.numSmoothUp=64 projection.maxIter=40
Additionally, you can try using the pressure from the previous timestep to remove a significant ammount of the divergence before projection:
projection.useIncrementalPressure=true
i.e. after computing the unprojected velocity ${U}^*$, then subtract off the old pressure gradient to find ${U}^* - p$ before projecting this to find a (hopefully) small extra pressure correction. This can significantly speed up the solve.
1.8.13