Multiphase Plume Solver

GPU-accelerated, two-way coupled Euler-Lagrange solver for simulating subsea oil and gas blowout plumes in shallow water.

Combines Oceananigans.jl (non-hydrostatic Boussinesq LES) with custom KernelAbstractions.jl GPU kernels for Lagrangian particle tracking.

Key Features

• Two-way momentum coupling via cell-sorted force deposition (collision-free, no atomics)
• 5-fraction composition-dependent oil dissolution (aliphatic + aromatic tracers)
• Gas dissolution with Henry's law pressure dependence
• Bin-based DSD injection from TAMOC with mass conservation
• One-way coupling mode (--oneway) for comparison experiments
• Scalable ambient velocity (--uscale) for crossflow sensitivity studies
• Mass budget tracking (released, active, surfaced, dissolved)
• 4 grid presets: low (0.6M) to ultra (37.7M cells)

Structure

solver/          Julia solver code (Oceananigans + GPU kernels)
analysis/        Python post-processing and diagnostics
plotting/        Visualization and animation scripts
hpc/             SLURM batch scripts and deployment helpers
docs/            Theory and development documentation (markdown)

Quick Start

Local (CPU, low-res test)

cd solver
julia --project=. -e 'using Pkg; Pkg.instantiate()'
julia --project=. plume_multiphase.jl --cpu --low-res --duration=30 --outdir=test

Ixachi H200 (GPU)

bash hpc/sync_to_ixachi.sh
# On Ixachi:
srun --partition=oc_gpu --gres=gpu:H200:1 --time=00:30:00 --pty bash
cd /LUSTRE/adomingu/Multiphase_Plume
julia --project=solver plume_multiphase.jl --duration=600 --outdir=output_run

Analysis

python analysis/analyze_plume.py output_run
python plotting/animate_plume.py output_run --fps 10
python plotting/animate_particles_plume.py output_run

CLI Options

Flag	Default	Description
--cpu	GPU	Force CPU execution
--low-res	mid	Use 192x48x64 grid
--oneway	off	Disable particle-to-fluid feedback
--grid=PRESET	mid	low / mid / high / ultra
--duration=SEC	300	Simulation time (seconds)
--uscale=VAL	1.0	Scale ambient velocity U(z)
--outdir=DIR	output_multiphase	Output directory

See julia --project=solver plume_multiphase.jl --help for all options.

Output

File	Contents
plume_multiphase.nc	Eulerian fields (u, v, w, T, S, c, c_ali, c_arom, tke)
particles.zarr	Lagrangian trajectories (x, y, z, d, mass, composition, type)
mass_budget.csv	Mass budget time series
analysis_report.txt	Summary statistics

Dependencies

Julia (>= 1.10): Oceananigans, CUDA, KernelAbstractions, NCDatasets, NPZ, Zarr, StructArrays

Python (>= 3.9): numpy, xarray, zarr, matplotlib, cartopy (for maps)

Scenario

Nohoch Alfa platform, Cantarell complex, Bay of Campeche (19.70 N, 92.10 W). 50 m depth, Maya crude oil (API 21.3), methane gas, GLORYS12 winter ambient profiles, TAMOC-derived droplet size distributions.

Documentation

• Theory — governing equations, particle physics, coupling algorithm
• Development — code review, corrections, results, known issues

References

Dominguez, A., Sheinbaum, J., Socolofsky, S.A. (2026). GPU-accelerated two-way coupled Euler-Lagrange simulation of subsea oil and gas blowout plumes in shallow water. Ocean Modelling (in preparation).