qcd_ml -- some machine learning layers for QCD

written using torch.

DOCUMENTATION

Contents

• Getting Started
  • Installation
• Functionality
  • Base
  • Neural Network
  • QCD
  • Utilities
  • Compatibility

Getting Started

Installation

The easiest way to install qcd_ml is by cloning the repository and installing the package using pip:

git clone https://github.com/daknuett/qcd_ml
cd qcd_ml
pip install .

This will install all dependencies (mostly numpy and pytorch) automatically.

If you need a specific version of numpy or pytorch, install them manually before installing qcd_ml.

Functionality

Base

• Various group operations of gauge and spin group.
• Gauge-equivariant vector hop.
• Gauge-equivariant paths for vector-like and matrix-like objects.
• Gauge-equivariant path-buffers for vector-like and matrix-like objects.

Neural Network

For vector-like objects the following layers are provided:

• v_PTC, v_LPTC, and v_LPTC_NG that implement 2302.05419.
• v_ProjectLayer that implements 2304.10438 parallel transport pooling.
• v_PT and v_Dense

For matrix-like objects the following layers are provided:

• LGE_Convolution
• LGE_Bilinear
• LGE_ReTrAct
• LGE_Exp
• PolyakovLoopGenerator and PositiveOrientationPlaquetteGenerator

See 10.1103/PhysRevLett.128.032003.

QCD

• Euclidean gamma matrices.
• Wilson Dirac operator and Wilson-Clover Dirac operator.
• Stout link smearing.
• Plaquette and topological charge field.

Utilities

• GMRES iterative solver.
• ZPP_Multigrid: Zero-Point-Projected Multigrid.
• Coarsened 9-point operators for Multigrid.

Compatibility

• Compatibility layer for lehner/gpt