MATLAB legacy library

SIES started as a MATLAB library written by Han Wang (ENS Paris). The original code is preserved in this repository (directories +shape/, +ops/, +asymp/, +PDE/, +dico/, +acq/, +tools/, +tracking/, +wavelet/ and examples/).

Correspondence between MATLAB and Python

MATLAB	Python	Status
+shape/@C2boundary and shape classes	sies.shapes	✅ ported
+ops (layer potentials)	sies.operators	✅ ported (P0 elements)
+asymp/+CGPT	sies.asymptotics	✅ ported
+acq (acquisition)	sies.acquisition	✅ ported (single-Dirac sources)
+PDE/@Conductivity_R2	sies.pde.ConductivityR2	✅ ported
+dico/+CGPT	sies.dictionary	✅ ported
+tracking	sies.tracking	✅ ported
+tools	sies.greens, sies.utils	✅ ported (the needed subset)
+PDE/@Helmholtz_R2, +asymp/+SCT (echolocation)	—	not yet ported
+PDE/@ElectricFish, +PDE/@PulseImaging_R2	—	not yet ported
+wavelet (wavelet polarization tensors)	—	not yet ported
examples/ (demo scripts)	notebooks/ (Marimo)	✅ ported

Notable differences

• Indices — Python indices are 0-based; source s in MATLAB corresponds to source s - 1 in Python.
• Spline resampling — shapes with corners are smoothed with periodic cubic splines (scipy.interpolate.CubicSpline), where MATLAB used not-a-knot splines (csapi); the periodic variant is the natural choice for closed curves.
• Flower perturbation exponent — the second derivative of the flower boundary for perturbation exponents k > 1 follows the correct analytic formula (the MATLAB code had a typo in a branch never exercised, since it always used k = 1).
• Operator overloading — MATLAB overloaded < for rotations; Python uses the explicit shape.rotate(phi). Translation (+, -) and scaling (*) are kept.
• Randomness — all stochastic functions accept a numpy.random.Generator for reproducibility.