RFSurfHMC

Joint inversion of Receiver Function and Surface Wave Disperion by Hamiltonian Monte Carlo Method

Requirements

RFSurfHMC has been built upon a few modern packages for its performance and sustainability, as listed below:

name	version
Pybind11	>=0.4
FFTW3	>=3.3
GCC	>=7.5
CMAKE	>=3.10.0
mpi4py	>=3.0

Installation

• This package requires some py-package as follow:
  • mpi4py
  • pybind11
  • numpy
  • h5py
  • pyyaml we recommand install those packages by anaconda

conda create -n hmc_inv python=3.9
conda install mpi4py pybind11-global numpy h5py pyyaml

• Then use the following to compile the code

mkdir -p build
cd build
cmake .. 
make -j4 
make install

Usage

• set your parameters in param.yaml
• For naive HMC, try: mpiexec -n 4 python main_base.py for parallel running
• For HMC with dual averaging (hoffman 2014,algorithm 5), try mpiexec -n 4 python main_DA.py
• python plot.py for drawing figures.

New Features

• Python extensions for surface wave dispersion, receiver functions and their Frechet kernels.

• A more general framework for HMC.

Update

• 2023-08-23 we update the forward calculation of receiver function. RF could be calculated in time domain or frequency domain either.
• 2025-07-26 update pybind11 interface, and add dual averaging, which significantly reduces the need for manual hyperparameter tunings on dt and L.

References

Junliu Suwen, Qi‐Fu Chen, Nanqiao Du; Joint Inversion of Receiver Function and Surface Wave Dispersion by Hamiltonian Monte Carlo Sampling. Seismological Research Letters 2022;; 94 (1): 369–384. doi: https://doi.org/10.1785/0220220044