Sparse Quadrature Wave

This repository contains two closely related workflows for stochastic acoustic wave propagation:

• acoustic_MCS/ implements Monte Carlo simulations (serial and MPI) together with scripts for generating Karhunen–Loève/PCE data and meshes.
• NISP_wave/ implements non-intrusive spectral projection (NISP) with sparse quadrature, sharing the same preprocessing utilities.
• matlab_scripts/ provides MATLAB post-processing utilities for inspecting Monte Carlo outputs (MCS_pdf, MCS_variance) and for comparing solver results against reference data (MCS_process).

---

Repository Layout

• acoustic_MCS/
  • preprocess.sh – builds KLE/PCE data (klePceData/) and meshes (meshData/) via gfortran and gmsh.
  • acoustic_parallel_MCS.py, acoustic_parallel_3RV.py, acoustic_pdf.py – Monte Carlo drivers (MPI-enabled).
  • klePceData/, meshData/, ssfem_solver/ – generated input data, mesh converters, and reference CSV outputs from the SSFEM solver.
• NISP_wave/
  • preprocess.sh, quad.sh – same preprocessing pipeline plus UQTk quadrature generation.
  • nisp_serial.py, nisp_parallel*.py, nisp_SparseQuad.py – NISP solvers matching different execution modes.
  • serial/, quadData/, klePceData/, meshData/ – solver outputs and reusable data.
• matlab_scripts/
  • MCS_pdf.m – aggregates PDF samples u_pdf_*.mat from outputs/<case>/... to a single u_pdf.mat.
  • MCS_variance.m – computes confidence bounds across experiments stored under outputs/3rv_sigma1/3rv_pdf_<idx>/.
  • MCS_process.m – loads Monte Carlo results plus SSFEM reference CSV files to compare mean/variance histories.
• .gitignore – ignores heavy generated data (outputs/, solver .mat files, etc.).

---

Requirements

• Python 3 with dolfin/FEniCS, numpy, scipy, mpi4py.
• GNU Fortran (gfortran) for building KLE/PCE data generators.
• gmsh for mesh generation.
• UQTk binaries for quadrature generation (referenced in NISP_wave/quad.sh).
• MATLAB (or Octave) for the scripts inside matlab_scripts/.
• MPI runtime (e.g., OpenMPI) for parallel simulations.

---

Typical Workflow

1. Preprocess shared data

   cd acoustic_MCS
   ./preprocess.sh        # builds klePceData/ and meshData/
   cd ../NISP_wave
   ./preprocess.sh        # builds equivalent data for NISP
   ./quad.sh              # generates quadrature points via UQTk

   Adjust mesh density (lc) and UQTk parameters inside the scripts as needed.

2. Run simulations

   • Monte Carlo (MPI):

     mpirun -np <ranks> python3 acoustic_parallel_MCS.py

     This produces u_MCS_mean.mat, u_MCS_sd.mat, nS.mat, etc., which remain untracked because of .gitignore.
   • NISP:

     python3 nisp_serial.py
     # or mpirun -np <ranks> python3 nisp_parallel1.py

     Outputs such as serial/d3l*/u_nisp_{mean,sd}.mat are likewise ignored.

3. Post-process in MATLAB

   • matlab_scripts/MCS_pdf.m expects Monte Carlo pdf chunks under outputs/3rv_sigma3/3rv_sigma3_pdf_20/.
   • matlab_scripts/MCS_variance.m reads aggregated pdf files under outputs/3rv_sigma1/3rv_pdf_<idx>/.
   • matlab_scripts/MCS_process.m compares Monte Carlo results against reference CSVs under acoustic_MCS/ssfem_solver/13k_3rv_3s/. Each script now infers the repository root automatically:

   script_dir = fileparts(mfilename('fullpath'));
   repo_root = fileparts(script_dir);

   Update the case-specific subdirectories at the top of each script to match your experiment layout.

---

Regenerating Removed Files

All deleted .mat artifacts are solver outputs and can be regenerated:

• NISP results (NISP_wave/serial/d3l*/u_nisp_mean.mat, u_nisp_sd.mat): rerun nisp_serial.py or the desired parallel variant.
• Monte Carlo summaries (acoustic_MCS/u_MCS_mean.mat, u_MCS_sd.mat, nS.mat): rerun acoustic_parallel_MCS.py (or related scripts).
• PDF chunk consolidation: run matlab_scripts/MCS_pdf.m after Monte Carlo chunk generation.

If additional heavy outputs should remain local only, add their patterns to .gitignore.

---

Questions ?

Contact : Sudhi Sharma P V : sudhisharmapadillath@gmail.com