Read and write lattice QCD data in Python with NumPy.
This project was originally the I/O part of PyQUDA, and I simplified the code to create this repository. This might be helpful to analysis lattice QCD data in Python.
This repository contains Python implementations to read and write different lattice QCD data such as gauge configurations and propagators generated by Chroma software.
Just loading data into NumPy format is useless, and you should use these I/O samples along with your analysis script.
The propagator will be rotated to DeGrand-Rossi basis.
| Type | shape | dtype | Comment |
|---|---|---|---|
| Gauge | (Nd, Lt, Lz, Ly, Lx, Nc, Nc) |
complex128 |
row-colum order |
| Propagator | (Lt, Lz, Ly, Lx, Ns, Ns, Nc, Nc) |
complex128 |
sink-source order |
| Staggered propagator | (Lt, Lz, Ly, Lx, Nc, Nc) |
complex128 |
sink-source order |
| Fermion | (Lt, Lz, Ly, Lx, Ns, Nc) |
complex128 |
|
| Staggered fermion | (Lt, Lz, Ly, Lx, Nc) |
complex128 |
| Software | Gauge | Propagator | Other |
|---|---|---|---|
| Chroma | Read | Read | |
| MILC | Read | Read | |
| KYU | Read/Write | Read/Write | |
| openQCD | Read/Write* | ||
| NERSC | Read | ||
| IO General | Read |
*Requires the plaquette value.
The repository requires mpi4py to read lattice QCD data in parallel. You could setup the grid by calling setGrid(grid_size). If you do not call this function, each process will read the entire field.
test.mpi.py:
import pylatticeio as io
io.setGrid([1, 1, 2, 2])
gauge = io.readChromaQIOGauge("./data/weak_field.lime")
print(gauge.shape)And then run the script by
mpiexec -n 4 python test.mpi.pyYou should get the shape of gauge to be (Nd, Lt/2, Lz/2, Ly, Lx, Nc, Nc), since the gauge is divided in z and t directions.
The rank of a process varies fastest in the t direction. For example, with grid_size=[1, 1, 2, 2], the grid coordinate and the corresponding rank of a process is
| item | ||||
|---|---|---|---|---|
| rank | 0 |
1 |
2 |
3 |
| grid coordinate | 0,0,0,0 |
0,0,0,1 |
0,0,1,0 |
0,0,1,1 |
You can also check this by calling io.getCoordFromRank and io.getRankFromCoord.