Awesome

The Helsinki Discrete Element Model (HiDEM)

Developers: Jan Åström & Joe Todd

This is HiDEM, the Helsinki Discrete Element Model, a particle model for simulating elastic behaviour, fracture and calving at marine terminating glaciers. Due to its computational demands, this code is designed to be run on parallel HPC facilities.

See the model in action here and here.

The model physics is described in detail in Åström et al. 2013.

Compilation

Configuration, compilation and installation is handled by CMake. Example installation scripts for Cray and Ubuntu systems are located in scripts/compilation. These scripts invoke cmake with toolchain files located in scripts/toolchains, which set default compilers etc.

By default, compilation produces a single binary 'HiDEM' in the top level of the 'build' directory.

If you generate your own toolchain/compilation scripts for different systems, please get in touch or make a pull request!

Input Files

On starting, HiDEM reads the name of an input file from HIDEM_STARTINFO (e.g. 'inp.dat').

inp.dat - defines various parameters for the simulation

The user must also supply the initial geometry in gridded format: X,Y,SURF,BASE,BED,FRICTION

This initial geometry file is specified in the input file as 'Geometry File':

Geometry File = "mass3.dat"

Head of a sample mass3.dat file:

15851
0.000000000000000000e+00	0.000000000000000000e+00	9.248271630519999462e+02	9.248271630519999462e+02	9.248271630519999462e+02	1.833796257881020755e+07  
5.000000000000000000e+01	0.000000000000000000e+00	9.248386672000000317e+02	9.248386672000000317e+02	9.248386672000000317e+02	2.784068307483682781e+07  
1.000000000000000000e+02	0.000000000000000000e+00	9.248503686919999609e+02	9.248503686919999609e+02	9.248503686919999609e+02	4.768270480512356758e+07

See more information in section 'Geometry File'

Note: To avoid spurious single particles appearing at the edge of the domain, interpolation of particle locations is not permitted where not all 4 supporting points have valid surface and bed values. To permit this, set Strict Domain Interpolation = False in inp.dat.

Important Points

Typical simulation domains:

• 100m x 100m * 100m - typical particle size 2-10m
• 100km x 100km * 1km - typical particle size 30-100m

Boundary conditions:

• Domain in x-y plane
• Main flow in y-direction
• Forces acting on edge should be defined

Particle size:

• Should have at least 10 particles in thickness direction (preferably 20-30).
• Maximum ~10 million particles
• Max timestep size scales with particle size:
  • Too long timestep => instability
  • Too short timestep => long execution times

Getting data ready

Replace no data with zero

Adjust the waterline (lowest point in domain must be non-negative)

Choose the number of cores

Do something to the restitution coefficient

Point to the input file (e.g. testinp.dat) using HIDEM_STARTINFO

Test Case

A test case is provided in the test directory. This is a simple rectangular domain with a sloping upper surface. This simulation will produce fracturing behaviour within a 6 hour simulation on 130 CPUs. The user should edit example.job to provide a valid PBS budget account, and the correct number of nodes and cores (currently 6, 130) depending on system architecture (cores per node).

The behaviour of the test case can be changed by modifying Max Load and Friction Scale in inp.dat.

Running the model

The model runs in parallel using MPI, so the simulation should be started using mpirun or aprun:

mpirun -n 70 HiDEM

An example PBS job script is provided in example.job

The user may specify a run name which is preprended on all output files:

Run Name = Example_Simulation

Restarting

It is possible to restart one run from another:

Restart = 1

If the run name has changed, specify the run from which to restart:

Run Name = Example_Simulation_Part2
Restart from Run Name = Example_Simulation

Melange

It is possible to use HiDEM to investigate glacier/melange interaction. Melange is generated by running an initial 'melange simulation' until the domain calves to produce icebergs & brash. Then a subsequent simulation can read in this broken ice from the 'melange simulation'. This is controlled by the keyword:

Melange Run Name = "Your_previous_simulation_run_name"

The 'melange simulation' setup could be identical to the standard calving simulation, or one could choose to increase the porosity or lower the maximum load to encourage breakup. It is not necessary for the melange simulation and the subsequent simulation to run on the same number of cores, but it is expected that the particle size (SCL) would be the same.

Note: unlike a regular simulation restart, the melange is assumed to be 'at rest' when read into the new sim.

Troubleshooting

If the simulation explodes (particles moving too far):

• Set a higher Translational Damping or Rotational Damping in the input file - multiply by two? Very high values for damping might be e.g. 16E4 and 8E4

• Set Fracture After Time = 50.0 or so, to allow the simulation to settle before permitting fracture

• Change friction scale in input - this scales the bed friction

• Change the timestep

• Smooth gradients in basal friction or geometry (in geometry file)

• Kill all motion every x timesteps

Files, Variables and Parameters

Structure of the repository

• src - all the code required to compile the HiDEM binary.
• Within the scripts directory:
  • analysis - various fortran and python scripts for post-processing
  • compilation - example compilation scripts
  • job_scripts - example PBS scripts
  • io - fortran programs for modifying input
  • paraview - macros for working with HiDEM data in Paraview

HiDEM Source Code

These are part of the compiled code:

Other Files

ave*.f - these compute averages of various outputs

rc2.f90, rc3.f90 - compute the calved size distrib

inp.dat parameters

Geometry file

The geometry file contains the input configuration, in format:

x, y, surface, base, bed, friction, geom_mask (optional)

x and y must be gridded and start at zero, must have a (0,0) corner.

Friction has units of Newton seconds per metre

output transformation matrix which takes from Elmer domain to HiDEM domain.

make sure the bed is buffered beyond the edge of the ice, and define these regions by setting surf and base equal to bed.

User may optionally specify a 'geom_mask' column in geometry input file, which tells the model which regions are ice (=1), fjord (=2), bedrock(=0). This is required for imposing lateral boundary conditions (Fixed Lateral Margins).

Internal variables

Output - JYR and STR files

By default the model produces particle information in .vtu format (readable in Paraview (v5.5) - use HiDEM_load.py macro) and bond strain information in binary format (readable by the python script rh.py). For CSV output, use the 'CSV Output' option in the inp.dat (beware this produces much larger files which make visualisation time consuming).

JYR files list the position of all particles in x,y,z, every 2 seconds.
Read this in paraview quite easily.

STR file list the midpoint position and strain of each bond between two particles, for each node connection in initial geometry (including broken bonds)

dtop* files - these show the total energy in the system for different parts

dtop00 - T,WENS,ENMS+ENMS0,KINS,MGHS-MGH0
dtop01 - T,DPES,DMPENS,PSUMS,GSUMS
dtopr - T,system energy, damping energy?

Time
WENS - elastic energy of the spheres - imagine the particles overlap and are deformed against each other. Then they might bounce back apart.
ENMS+ENMS0 - elastic deformation energy - the energy held in a deformed/bent system
MGHS-MGH0 - potential energy
Kins - translational kinetic energy
Kins2 - rotational kinetic energy

DPES,DMPENS - guess energy lost to drag and damping?
PSUMS - something like pressure
GSUMS - energy of bed interaction

Processing

rc2.f90 computes the size distribution of calved blocks

size count

1 452000 2 4560 3 985 ... largest_block 1

'maxi' is the same as jyr but for the largest intact block (e.g. the remaining glacier)

rh2.f90 - computes velocity
rh.f - computes strain

INFI1 - first JYR
INFI2 - second JYR
N - wc -l JYR0001.csv
max - 10?

Paraview

First simply load csv, with blank space delimiter, merge delimiters, has no header
Then apply table to points filter
Then ensure enable ospray and shadows, and pick point size that makes them touch

For the sea: add a box, set size and centrepoints

For the bed: get bed.csv, read it in same as other CSVs
then apply Delaunay2D filter.

TO DO

• BC strategy

• Translate & Rotate input

Notes

Domain needs to be orientated in XY because:

Boundary conditions are applied in WSY (, WSX?) components (need to compute normal? Ask