ReleaseNotes

Release Notes July 2025

Enforcing post ice thickness can, as before, be done using the penalty method, by setting

  CtrlVar.ThicknessPenalty=1;

This causes and additional implicit mass-balance term to be added

$a^{\star} = a_1 (h-h_{\min}) + a_2 (h-h_{\min})^2 + a_3 (h-h_{\min})^3$

where

$h < h_{\min}$

and where:

a_1 = CtrlVar.ThicknessPenaltyMassBalanceFeedbackCoeffLin
a_2 = CtrlVar.ThicknessPenaltyMassBalanceFeedbackCoeffQuad;
a_3 = CtrlVar.ThicknessPenaltyMassBalanceFeedbackCoeffCubic;

Note: $a_1$ and $a_3$ need to be negative and $a_2$ positive, however, this is check internally so actually the sign on input is immaterial.

By setting:

CtrlVar.ActiveSet.ExcludeNodesOfBoundaryElements=true;

nodes of boundary elements can now be excluded from the active set. By default this is false, which is the previous behavior.

The parameter:

CtrlVar.LevelSetMethodAutomaticallyDeactivateElementsRunStepInterval

is no longer used. Instead the interval is now identical to the general adapt-meshing interval. See further explanation in Ua2D_DefaultParameters.m

Release Notes May 2025

The call to

DefineDesiredEleSize.m

now only accepts one combination of the number of input and output variables.

UaUtilities contains a new function:

ElementErrorEstimator(CtrlVar,MUA,F)

which provides a "Recovery-based error estimator". An example of how to use this error indicator for automated mesh refinement and un-refinement is provided in "1dIceStream" example within the UaExamples repository.

Also within UaUtilities is the function

InfluxOutfluxNodes.m

which can be used to find nodes along the boundary where the velocity vectors are oriented into the domain (influx/inflow nodes) and out of the domain (outflux/outflow nodes). This can be used, for example, in when defining boundary condition to ensure that flux into the domain is some given value, e.g. zero.

Release Notes March 2025

The uv assembly was changed slightly to make the assembly with respect to ice thickness, h, better consistent with the uvh assembly. This can lead to uv solve to be different if the ice density is varying greatly spatially. The basic difference is that the flotation condition, which involves the product rho and h, is now evaluated at integration pSainanoints directly. Previously in the uv solve, it was evaluated by first forming the product, rho h, at nodes and then interpolating the product to the integration points.

Calls to gcp have been replaced with gcp('nocreate'). The implication is that a parallel pool should be defined and started ahead of a call to Ua. However, not that this might also depend on user settings for the parallel pool. For example if the user setting imply automated start of a parallel pool whenever parfor or smpd is encountered. If no parallel pool is found, all parallel options are turned off, and the run proceeds in non-parallel mode.

Unless the mass balance/thickness feedback is activated, MassBalance evaluation now lags behind by one time step. This was done to reduce calls to DefineMassBalance, and to make sure that the mass balance of the Úa fields, F0, was same as the F from previous time step. Note that when the mass balance/thickness feedback is activated, the mass balance function is called within the assembly loop and then the mass balance will not lag behind.

The start and end times can now be specified using (new option):

CtrlVar.StartTime
CtrlVar.EndTime

Previously this was done using (old option):

CtrlVar.time
CtlrVar.TotalTime

The previous option still works, but the new option is recommended.

When not inverting for both A and C, the variable InvValues was not updated in last call, leading to a possible mismatch between F.A and InvValues.A. Thanks to Camilla Schelpe for identifying this.

The implementation of the ice-thickness barrier function has been simplified, and is now similar to the barrier function used in the level-set solver.

The MassContinuity solver, (only used when solving for h alone, and not in uv or uvh solves), now uses the active-set method to enforce positive thickness.

The model was tested with MATLAB R2024b

Release Notes September 2024

For comparison purposes the semi-implicit solver has been updated.

This option is NOT recommended and the (fully) implicit approach continues to be the default. The semi-implicit approach is both slower and less accurate that the implicit one. This option has now been updated and made available for use as apparently most (all?) other ice-flow models appear to be using this semi-implicit approach.

This option is activated by setting:

  CtrlVar.ForwardTimeIntegration="-uv-h-" ;

The default value of

CtrlVar.LevelSetMethodThicknessConstraints;

is now set to true (previously set to false). Therefore if one uses the active-set method and the level-set method, min thickness constraints will be automatically applied to all nodes downstream of calving fronts, unless of course this parameter is set to false by the user.

Release Notes July 2024

The utility

[tbx,tby,tb,eta] = CalcBasalTraction(CtrlVar,UserVar,MUA,F,options)

has been updated to allow for calculation at integration points (before only calculated nodal values).

An inconsistent input parameter test when using different sliding laws for basal drag and ocean drag calculations corrected. Thanks to Sainan Sun for pointing this out.

*A situation where a composite variable was used inside of spmd, resulting in a warning but no errors, has been corrected.

Release Notes March 2024

Positive thickness constraints --- added dynamically when using the active-set option --- are now not added to nodes already contained in a user-defined constraint. Hence, any user-defined thickness constraints are respected, even if this means that some thickness go below the minimum specified thickness. Previously, thickness constraints were added to all nodes with thickness less than CtrlVar.ThicMin.

Release Notes February 2024

Parallel spmd assembly option now improved and shows good scalability, although speedup always somewhat problem dependent. On local workstations with 12 workers, speedup ranging from 6 to 10 seems easily obtainable, and on machines with 24 workers a speedup of 20 has been observed.

To switch on the smpd parallel assembly do:

CtrlVar.Parallel.uvhAssembly.spmd.isOn=true;    % uvh assembly in parallel using spmd over sub-domain (domain decomposition)
CtrlVar.Parallel.uvAssembly.spmd.isOn=true;     % uv assembly in parallel using spmd over sub-domain (domain decomposition)

The linear system can now also be solved using distributed arrays, although not all cases yet implemented. Turn this on/off as:

CtrlVar.Parallel.Distribute=true/false;                       % linear system is solved using distributed arrays.

Again, the speedup is problem dependent and never particularly large for low density sparse matrices as typically generated by Úa and other FE programs.

CtrlVar.Parallel.isTest=false/true;             % Runs both with and without parallel approach, and prints out some information on relative performance.
                                                % Good for testing if switching on the parallel options speeds things up, and by how much.

Release Notes October 2023

A (rare) case where the Cauchy direction is not a direction of descent was incorrectly updated. This has now been addressed.

Release Notes September 2023

Thanks to Sebastian Rosier for providing an example where the backtracking algorithm failed. This has now been corrected.

Release Notes July 2023

uv and uvh solver now uses dog-leg search if Newton back-tracking results in small steps.

Thanks to Sainan Sun for spotting that in an adaptive mesh step, call to calving was ahead of call to geometry and densities. This has now been corrected.

Release Notes July 2022

Call to DefineOutputs is now only done at the beginning and end of runs if the variables

     CtrlVar.CreateOutputsBeginningOfRun=true;   % If true, then call DefineOutputs at the beginning of a run, that is ahead of the runstep/transient loop.
     CtrlVar.CreateOutputsEndOfRun=true;         % If true, then call DefineOutputs at the end of a run, that is after the runstep/transient loop

are set to true.

Calving options have been greatly improved and a flexible framework for implementing calving laws implemented. Although still considered not fully tested, this option appears to work quite well. Further details can be found in Ua2D_DefaultParameters.m and in the DefineCalving.m files.

In the active-set method a minimum number of new active nodes can be specified for the active set to be updated ahead of a new uvh solve, e.g.:

CtrlVar.MinNumberOfNewlyIntroducedActiveThicknessConstraints=5;

UaPlots is a new plotting utility function.

Some minor bugs have been corrected, but nothing that should affect calculated solutions.

This version requires MATLAB2021a or later.

Release Notes Sept 2021

Calving

Calving can now be prescribed by defining a ice/ocean mask at each time step. The implementation is more general and faster than other similar manual calving options such as deactivating elements or prescribing an additional mass balance term.

The ice-free areas are automatically melted away using a mass-balance feedback option implemented at the integration points. Second-order NR convergence is obtained even if the resulting mass balance distribution varies significantly spatially within an element

To use this calving option set:

    CtrlVar.LevelSetMethod=1;

and then define the ice/ocean mask at each time step using the m-file

    DefineCalving.m

NOTE: Currently the only supported calving option is based on directly prescribing the LSF at each time step

If you wanted, for example to get rid of all floating ice for x>500e3, do:

       [UserVar,LSF,CalvingRate]=DefineCalving(UserVar,CtrlVar,MUA,F,BCs)
         F.GF=IceSheetIceShelves(CtrlVar,MUA,F.GF);
         OceanNodes=MUA.coordinates(:,1)>500e3 & F.GF.NodesDownstreamOfGroundingLines;
         LSF=zeros(MUA.Nnodes,1)+ 1 ;
         LSF(OceanNodes)=-1;
       end

Think of the LSF as a ice/ocean nodal mask, positive for ice and negative for ocean.

The the ice thickness in ice-free areas is

CtrlVar.LevelSetMinIceThickness

While this could in principle be set to zero, doing so might create isolated islands of ice, resulting in an undetermined system. So put this to a some small positive value (small compared to typical ice thicknesses of interest). By default:

  CtrlVar.LevelSetMinIceThickness=CtrlVar.ThickMin+1;

where

   CtrlVar.ThickMin=1;

Automated mesh refinement based on distance to calving from is possible in similar way as around grounding lines. For example to locally refine all elements with 10e3 distance units from the calving front to 1e3, use:

CtrlVar.MeshAdapt.CFrange=[10e3 1e3] ;

Remember also to set

CtrlVar.AdaptMesh=1;

as always needed to activate the remeshing options.

New argument list option with F as an input

New:

    DefineMassBalance(UserVar,CtrlVar,MUA,time,F)

Old:

    DefineMassBalance(UserVar,CtrlVar,MUA,time,s,b,h,S,B,rho,rhow,GF)

The old approach still works and is generally recommended.

Geometry and densities can now be defined in the same m-file, using

    DefineGeometryAndDensities.m

The previous approach of using separate m-files to define geometry (DefineGeometry.m) and densities (DefineDensities.m) still works. But if you have a "DefineGeometryAndDensities.m" in the run folder, only "DefineGeometryAndDensities.m" is used and "DefineGeometry.m" and "DefineDensities.m" ignored.

x and y nodal values now a field of F, ie F.x and F.y

The (x,y) coordinates of the nodes are now also accessible through F as

(F.x,F.y)

These are always identical to MUA.coordinates, that is

F.x=MUA.coordinates(:,1)
F.y=MUA.coordinates(:,2)

This was simply done for convenience, and now one can, for example, create a plot showing the upper surface, s, as a function of x as

plot(F.x.F,s,'.')

Default inverse algorithm has changed. Now the default options is a Hessian-based inversion. The older-approach is still available by selecting a Gradient-based inversion. This is specified using the CtrlVar field:

     CtrlVar.Inverse.MinimisationMethod

When using this Hessian-based inversion, there is no need to specify the value of

     CtrlVar.Inverse.AdjointGradientPreMultiplier

and this field is not used.

Ocean and wind-induced drag over floating ice can be included. This is defined in

DefineSeaIceParameters.m

and you need to switch this option on by setting

CtrlVar.IncludeMelangeModelPhysics=true;

in DefineInitialInputs.m

The quadrature degree can now be directly specified using

CtrlVar.QuadratureRuleDegree=N

where N is a number that can be as high as 25 (Previously the highest possible degree was 13). Generally, the default option should be fine, but there might be instances were increasing the degree could help with obtaining second-order convergence of the Newton-Raphson system.

The shallow-ice sheet (SIA/SSHEET) option now includes basal sliding.

The SSHEET option is implemented for the Weertman sliding law only. The transient SSHEET solution is done implicitly with respect to the thickness using the NR method. When using SSHEET you will, in general, need to specify boundary conditions for both the deformational and the basal sliding velocities. Note the SSHEET option is based on the shallow-ice approximation.

Release Notes June 2020

Naming of some user input m-files has changed:

     Ua2D_InitialUserInput.m

is now named:

     DefineInitialInputs.m

And

     UaOutputs.m

is now:

     DefineOutputs.m

These changes were done so that names of all user-input files start with 'Define'

Note: To systematically change the names of all your old Ua2D_InitialUserInput.m and UaOutputs.m files you can, for example, consider using the m-file utility file:

     RenameFileRecursively.m

Naming of a few CtrlVar fields has changed. For example

     CtrlVar.ATStimeStepTarget

is for example no longer used. Use:

     CtrlVar.ATSdtMax

instead.

You can now specify a minimum selected time step in the automated-time-stepping (ATM) algorithm settting the field

     CtrlVar.ATSdtMin

Also

CtrlVar.UaOutputsDt

has also been replaced by

CtrlVar.DefineOutputsDt

If you define these old fields in your (new) DefineInitialInputs.m, �a will spot this and complain bitterly.

Those using Unix might want to systematically change names of some of the CtrlVar fields in there old input files. You might be able to use something like:

find . -name "DefineInitialInputs.m" -exec sed -i 's/CtrlVar.ATStimeStepTarget/CtrlVar.ATSdtMax/g' {} +
find . -name "Define*.m" -exec sed -i 's/CtrlVar.UaOutputsDt/CtrlVar.DefineOutputsDt/g' {} +

Exit criteria for the non-linear uv and uvh loops are now more flexible and allow for more options. See comments in

     Ua2D_DefaultParameters.m

for more details.

The

RunInfo

variable now contains more information about the run, such as time step and number of non-linear iterations per runstep.

Most of the changes since last Feb 2020 have been 'under the hood'. For example solving a KKT type system can now be done with a much more flexible pre-elimination method than before.

All line searches are now done using one single backtracking algorithm (before at least 4 different backtracking routines were used.)

The KKT system is now always solved using the primal-dual method. (Previously the initial iterative was made feasible and the primal method then used.)

Release Notes February 2020

Úa can now be called with CtrlVar as second argument, e.g

Ua([],CtrlVar)

in which case the fields of the CtrlVar on input will be those used in the run even if these same fields are also defined in Ua2D_InitialUserInput.m

The combination of local mesh refinement using the newest-vertex bisection method with mesh deactivation is now very flexible and allows for both reduction and extension of the computational domain during the run.

Several new sliding laws are now implemented. These include Weertman, Coulomb, Budd, Tsai, Cornford, and Umbi. Refer to the Úa compendium for definition of these different sliding laws. Inversion is also possible for all these sliding laws except for the Coulomb sliding.

The semi-implicit time-stepping method now uses SUPG by default and when used the automated time-stepping algorithm ensures that time step fulfils the CFL condition.

When using automated mesh refinement/coarsening in a time-dependent run, the upper surfaces are mapped between meshes and the thickness calculated afterwards. (Previously, the thickness was mapped and the surface calculated afterwards.) This may cause a potential violation of mass conservation. However, this also ensures smooth upper surfaces even when refining mesh over areas with very uneven bedrock topography.

Inversion can be done using all implemented sliding laws except Coulomb.

u, v and h residuals now calculated in the L2 norm instead of the l2 norm as in the past.

New user input file options:

DefineRunStopCriterion.m

DefineFinalReturnedValueOfUserVar.m

DefineOutsideValues

Several new computational utilities

LakeOrOcean3.m

is both fast and robust (Thanks to Sebastian Rosier).

EleFlooding.m

finds all nodes/elements connected to a given node fulfilling some additional criterion.