GitHub

UTDKernels.jl

A branch-safe and differentiable implementation of the Uniform Theory of Diffraction for PEC wedges.

Overview

UTDKernels.jl provides a numerically robust Julia implementation of the Kouyoumjian–Pathak (KP) uniform theory of diffraction (UTD) for perfectly electrically conducting (PEC) wedges. The package evaluates the UTD transition function via the scaled complementary error function (erfcx) to maintain full floating-point accuracy across all parameter regimes, and includes a regularised cotangent–transition-function product that eliminates the $\infty \cdot 0$ singularity at shadow boundaries.

Key features:

  • Overflow-free transition function via the erfcx identity
  • Regularised cot–$F$ product for finite values at shadow boundaries
  • Documented principal-branch policy for all square roots
  • End-to-end automatic differentiation via a ForwardDiff.jl package extension

Installation

using Pkg
Pkg.add("UTDKernels")

For development:

Pkg.develop(path="path/to/UTDKernels.jl")

Quick start

using UTDKernels

# Define a half-plane wedge (exterior angle α = 2π)
w = Wedge(2π)

# Set ray angles: observation φ = 90°, incident φ' = 45°
ang = RayAngles(π/2, π/4)

# Wavenumber and effective distance
k = 10.0
L = 1.0

# Compute soft and hard diffraction coefficients
Ds, Dh = pec_wedge_DsDh(w, ang, k, L)
println("Ds = $Ds")
println("Dh = $Dh")
Ds = 0.07634584116031005 - 0.04771875264397024im
Dh = -0.17545379565978506 + 0.14123983292699288im

Tutorial outline

This documentation is structured as a self-contained tutorial that follows and expands on the companion paper. Every equation is derived in full detail with no intermediate steps omitted.

  1. Maxwell's Equations and the Helmholtz Equation – Time-harmonic convention, frequency-domain Maxwell's equations, and the scalar Helmholtz equation for 2D diffraction problems.
  2. Wedge Geometry and Geometrical Optics – Canonical PEC wedge, shadow/reflection boundaries, ray geometry, spreading factor, and diffraction dyadic.
  3. The UTD Transition Function – Full step-by-step derivation of $F(x)$ from the Fresnel integral to the erfc form to the numerically stable erfcx form.
  4. Kouyoumjian–Pathak Diffraction Coefficients – The four-term KP structure: cotangent arguments, boundary-tracking integers, distance parameters, sign factors, and the full $D_{s/h}$ formula.
  5. Numerical Methods – The three numerical challenges (overflow, singularity, branch cuts) and their solutions.
  6. Automatic Differentiation – Derivation of the erfcx derivative rule, the complex chain rule for ForwardDiff, and end-to-end gradient examples.
  7. Validation – Comparison with the exact Sommerfeld half-plane solution, GTD convergence, reciprocity, and shadow-boundary continuity.