Changelog

All notable changes to FEMTISE.jl are documented in this file.

The format follows Keep a Changelog and the project adheres to Semantic Versioning. While the version is 0.x, an increase of the minor number (e.g. 0.1 → 0.2) may contain breaking changes; patch releases (e.g. 0.1.0 → 0.1.1) only contain backward-compatible bug fixes.

Each entry is classified as Added, Changed, Deprecated, Removed, Fixed or Security. Changes that modify numerical results are always marked explicitly.

[Unreleased]

Nothing yet.

[0.2.0] - 2026-10-18

New features for general masses, tabulated potentials, two-parameter studies, wave-packet spectra and three-dimensional problems. Existing scripts and input files keep working. The changes that can affect existing code are listed under Changed.

Added

  • solve_tise(model, potential; masses, order, quadrature_degree, nev, sigma): general solver for 1D, 2D and 3D meshes, returning a TISESolution with the eigenvalues sorted by energy, the eigenfunctions, measures and spaces. It computes a few extra eigenpairs (guard_states) so that degenerate levels at the edge of the window are complete.
  • Particles of any mass: masses keyword (one mass per coordinate, in units of the electron mass); kinetic_coefficients, hamiltonian_forms (anisotropic kinetic term in any dimension) and discretization.
  • Potentials from data files: tabulated_potential (1D linear or natural cubic spline, 2D bilinear; validation of the table and of the evaluation range).
  • Parameter studies: sweep_eigenvalues for one or two potential parameters, reusing mesh and spaces.
  • Plots (package extension, loaded with any Makie backend such as CairoMakie): plot_energy_map (eigenvalue or gap over two parameters) and plot_energy_curves, readable in black and white.
  • Wave-packet dynamics and spectra: expansion_coefficients, autocorrelation, autocorrelation_spectrum (Lorentzian or Gaussian lines) and fourier_transform of sampled autocorrelation functions, with consistent conventions.
  • Cavity QED: pauli_fierz_potential (length gauge, dipole approximation, optional dipole self-energy) for matter coupled to one photon mode, e.g. nucleus–electron– photon problems on 3D meshes.
  • Input files: optional settings after line 21, read by name (masses, element_order, quadrature_degree, potential_data_file, interpolation, analysis_param_2, Lz, nz), and dimension = 3D. Files that use them are solved with solve_tise; two-parameter sweeps are reloaded as AnalysisParams2D. Files without optional settings are solved exactly as before.
  • Documentation: manual pages on masses, data files and sweeps; autocorrelation and spectra; three coordinates and cavity QED. New tutorials: vibrational levels from a tabulated potential, autocorrelation and spectrum of a wave packet, two-parameter map of a double well, and a nucleus–electron–photon (Pauli–Fierz) model.

Changed

  • Errors in input files now raise an ArgumentError with a description of the problem. Previously the driver printed a message and called exit(0), which closed the Julia session.
  • InputData1D and InputData2D have a new field options::DriverOptions. The previous constructors are kept.
  • Makie exports a function named density, like FEMTISE. Code that loads both packages with using must write FEMTISE.density.

Fixed

  • The docstring of run_default_eigen_problem was detached from the function (blank line in between) and did not appear in the help or in the API reference.

[0.1.1] - 2026-10-18

Bug-fix release. The public API is unchanged; code written for 0.1.0 keeps working. Two fixes change numerical results (marked below).

Fixed

  • (changes results) time_indep_entropy(ψ, U, dΩ, α) now computes the Rényi entropy with $ρ^α$; 0.1.0 evaluated $ρ^{α/\ln 2}$. The Shannon entropy is unchanged.
  • (changes results) reduced_density / reduced_integration now normalize the approximate Dirac delta with the side lengths of the domain. In 0.1.0 the span of the sampling points ($L-\Delta x$) was used, which scaled the reduced densities by $1-\Delta x/L$.
  • collect_result_data can reload non-sweep results written in the bin eigen format: DefaultBinEigenProblem and DefaultBinEigenProblemReducedDensity now store the sampled eigenfunctions and reduced densities as matrices.
  • make_boundary_conditions("simple_rectangle_v2", ...) returns its boundary tags (an == was used instead of =).
  • solve(::EigenProblem) is now a method of Gridap's solve instead of a separate function with the same name, so using FEMTISE, Gridap no longer makes solve ambiguous. FEMTISE.solve keeps working.

Changed

  • [compat] bounds for every dependency (required for registration in the General registry).
  • Tutorials in the documentation rewritten (1D and 2D harmonic oscillators, finite Kronig–Penney chain, two-electron helium model): each one now runs from the FEMTISE_TUTORIAL repository with a single script, checks the results against exact or physical references, and shows figures generated from those simulations (black-and-white and colour-blind safe).

Removed

  • Unused dependencies: DataInterpolations, KeywordDispatch, Documenter, DocumenterTools and Pkg. Loading FEMTISE is faster and no longer installs the documentation tools.

[0.1.0] - 2026-10-18

First versioned release. It freezes the state of the package used in the publications listed in the README, so that their results can be reproduced with add https://github.com/mendzmartin/FEMTISE.jl#v0.1.0.

Added

  • Solution of the time-independent Schrödinger equation in Sturm–Liouville form with conforming Lagrange finite elements (Gridap.jl) and the shift-and-invert Arnoldi eigensolver of ARPACK (Arpack.jl).
  • Low-level interface: make_model, make_boundary_conditions, measures, fe_spaces, bilineal_forms, eigen_problem, solve, eigen_values_and_eigen_vectors, space_coord.
  • Anisotropic kinetic term for two particles of different masses in 2D (bilineal_forms(p₁, p₂, q, r, dΩ)).
  • Input-file driver (run_default_eigen_problem, set_type_potential, input_data) for 1D and 2D problems with user-defined potentials, and an interactive console with built-in potentials (1D/2D harmonic oscillator, finite well, symmetric finite Kronig–Penney chain).
  • Parameter sweeps (analysis_param) that reuse the mesh for every value of a potential parameter.
  • Output as binary arrays (write_bin/read_bin) or JLD2 files with the finite element objects; reloading with collect_result_data.
  • Post-processing: norm_l2, orthogonality_check, eigenstates_normalization, density, reduced_density, time_indep_entropy (Shannon and Rényi), reduced_time_indep_entropy, time_indep_diff_mutual_information, state_population, coord_first_moment, coord_second_moment, coord_standar_deviation.
  • Documentation reorganized into Getting started, Manual (theory, architecture, input-file reference, low-level API, outputs, post-processing, accuracy and best practices) and an API reference grouped by module; versioned documentation (dev, stable and one page per release).
  • This changelog.

Changed

  • Test suite reorganized into validation tests with analytic references: upper bounds and $h^{2\ell}$ convergence for $\ell = 1, 2, 3$, position-dependent mass (BenDaniel–Duke), two particles of different masses, post-processing observables, binary I/O, and the input-file driver end to end (1D, 2D, sweeps, JLD2 and binary outputs). Test files no longer activate environments themselves.
  • The orthogonality check of the 2D oscillator example skips pairs of degenerate eigenvectors (they are not guaranteed to be orthogonal, see Known issues); the previous check could fail randomly.
  • CI: updated GitHub Actions (checkout@v4, setup-julia@v2, cache@v2) and coverage upload (julia-processcoverage + codecov-action@v4).
  • Removed the stale test/Manifest.toml (Julia 1.9).

Known issues

These issues are present in 0.1.0 and are scheduled for the next releases. They are documented in the Accuracy and best practices page of the manual.

  • time_indep_entropy(ψ, U, dΩ, α) (Rényi entropy) evaluates exp∘(α*(log2∘ρ)), i.e. $ρ^{α/\ln 2}$ instead of $ρ^α$. The Shannon entropy is not affected. The fix will change numerical results.
  • make_boundary_conditions("simple_rectangle_v2", ...) uses == instead of = when setting the boundary tags.
  • FEMTISE.solve and Gridap.solve are both exported; code that loads both packages must call FEMTISE.solve explicitly.
  • The input-file driver integrates with a quadrature of degree 3 for quadratic elements, which is not exact for polynomial potentials. Convergence remains $O(h^4)$ but the error is about twice that of exact quadrature, and the eigenvalues are no longer guaranteed to be upper bounds.
  • In parameter sweeps, the first computed value is $λ_i + Δλ$ rather than $λ_i$.
  • collect_result_data cannot reload non-sweep results written in the bin eigen format (DefaultBinEigenProblem* declare the eigenfunctions as Vector{CellField}); read the files with read_bin.
  • reduced_density (through reduced_integration) normalizes with the span of the sampling points, $L-\Delta x$, instead of the domain length $L$: the reduced densities are too small by the factor $1-\Delta x/L$. Their shape is correct.
  • Eigenvectors belonging to a degenerate eigenvalue are linearly independent but not necessarily orthogonal (complex, non-Hermitian ARPACK driver).
  • When the input-file driver is called from inside a function, the potential defined by the potential file is "too new" for that function; use Base.invokelatest(run_default_eigen_problem, set_type_potential(path)).

[Unreleased]: https://github.com/mendzmartin/FEMTISE.jl/compare/v0.2.0...HEAD [0.2.0]: https://github.com/mendzmartin/FEMTISE.jl/compare/v0.1.1...v0.2.0 [0.1.1]: https://github.com/mendzmartin/FEMTISE.jl/compare/v0.1.0...v0.1.1 [0.1.0]: https://github.com/mendzmartin/FEMTISE.jl/releases/tag/v0.1.0