laddu

Amplitude analysis made short and sweet.

laddu brings the full analysis loop into one coherent interface: describe a reaction, compose a differentiable intensity, generate Monte Carlo, fit accepted data, and project the result. Models remain readable Python expressions while evaluation runs on parallel CPU, JIT-compiled CPU, WGPU, or MPI-backed execution.

import laddu as ld

execution = ld.Execution("auto", precision="f64")
likelihood = ld.Likelihood(
    [ld.NLL(model, data=data, accepted_mc=accepted_mc, name="signal")],
    execution=execution,
)
fit = likelihood.fit(
    terminators=[ld.ganesh.MaxSteps(500)],
)

Learn the workflow

Start with arrays or files, generate samples, build a likelihood, and perform a fit.

Tutorials

Build physical models

Combine topology, helicity amplitudes, line shapes, polarization, and coherent sums.

Building a polarized photoproduction model

Scale an analysis

Choose precision, automatic differentiation, CPU/JIT/GPU execution, and MPI partitioning.

Execution backends, memory, and MPI

Why laddu?

• One symbolic model. The same expression drives generation, likelihood evaluation, gradients, and projections.

• Physics-native building blocks. Four-vectors, reaction graphs, Wigner functions, relativistic line shapes, and coupled-channel amplitudes are first-class objects.

• Expression-based architecture. All mathematical models are built from expressions, so new users can write complex amplitudes in Python and let laddu handle mathematical optimizations, cached evaluations, and batched execution.

• Reproducible performance choices. Backends, precision, differentiation strategy, seeds, and partitioning are explicit.

• Platform-independent execution. Pipelines are automatically parallel and can be JIT compiled or even run directly on GPUs without writing a single kernel.

• Distributed programming. Code written in laddu is fully compatible with the MPI protocol for use on high-performance compute systems.