PINN Toolbox 1.1.0

Examples

• The example Live Scripts no longer require an open project. Paths to shipped
  files are built by a new helper, examples/helper/pinnToolboxRoot.m, which
  derives the toolbox root from its own location, so the examples run from a clone,
  from an installed .mltbx and in MATLAB Online alike. Replaces
  currentProject().RootFolder in priklad_kyvadlo / example_pendulum and
  priklad_teplo_2d / example_heat_2d.
• priklad_teplo_2d / example_heat_2d write the README GIF only when it is
  missing (refreshGif), and fall back to tempdir when the target folder is not
  writable (installed toolbox). getframe output is never byte-identical, so the
  previous unconditional write produced a spurious binary diff on every run.
• priklad_kyvadlo / example_pendulum no longer instruct users to overwrite the
  committed examples/data/kyvadlo.mat: record to any file outside the repository and
  point dataFile at it.
• Data generators are now functions with an optional output path —
  generate_kyvadlo_data(file) and generate_kyvadlo_data_simscape(file) write where
  told; called without an argument they ask via uiputfile. As scripts they polluted
  the caller's workspace and silently overwrote the committed data.
• generate_kyvadlo_data.m moved from examples/data/ to examples/helper/, so it is
  callable by name from anywhere and examples/data/ holds data only.

Fixed

• priklad_konvekce.m printed its curriculum phases without line breaks. The
  fprintf format string had lost the backslash of %%\n, so it printed a literal "n"
  and ran the phases together on one line; the defect was present in 1.0.0. The Live
  Editor strips one backslash directly after %% on every save (doubling it only delays
  the loss by one round), so both the Czech example and its English mirror now use
  %% \n with a space, which is stable.
• Math in the English mirrors normalised to the same backslash convention as the Czech
  originals (\\pi, \\nu, \\,), and two thin spaces lost to an earlier Live Editor
  round-trip restored in example_burgers.m and example_heat_2d.m.
• **example_neural_ode.m returned error "Error using extractBinaryBroadcastData>iGetData (line 108)
  Data must be of type double, single, or logical, or a gpuArray object of these types."
  The code now corresponds to the Czech original and the example runs correctly.

Project and tooling

• examples/helper added to the project path, which now mirrors ToolboxMatlabPath
  in tools/packageToolbox.m exactly.
• setupProject.m is authoritative for the full folder, file and path lists (it
  previously documented and restored only the repository root, so re-creating a lost
  .prj would have yielded a project that cannot run the examples); tools/ is
  registered with the project as well.
• The project file registry now matches the repository. Only 35 of 111 committed
  files were project members — addFolderIncludingChildFiles registers what exists at
  the time of the call, so everything added after the project was created stayed out
  (all of +PINN/+tools, most of +PINN/+utils, ten examples, most of examples/en,
  ten tests, all of tools/, every top-level file). setupProject.m now also registers
  the committed top-level files and prunes entries whose file no longer exists.
• setupProject.m is excluded from the packaged .mltbx — the repository root is on
  the installed path, so an installed user calling it would have created a MATLAB
  project in their current folder.
• tools/runAllExamples.m resolves the example folder via pinnToolboxRoot() and runs
  without an open project.
• New tests/test_ExamplePaths.m.

Documentation

• The Helmholtz example no longer calls its problem "stiff". A stationary
  Helmholtz BVP has no time scales, so it cannot be stiff in the usual sense; what is
  ill-conditioned is the PINN optimisation, and Wang et al. 2021 (paper 05) attribute
  the stiffness to the gradient flow of training rather than to the PDE. Relabelled in
  the Czech example, the English mirror, both tutorials and both tuning guides.
• Czech prose brought in line with doc/Slovnik_terminologie.md. Glossary
  violations replaced (multi-output, pipeline, rejection sampling, runaway λ,
  defaulty, papery, appka, dle), the Czech tuning guide moved off lab jargon
  onto the agreed terms, and the whole corpus switched to consistent vykání (formal
  address), including the Trainer App README and the Simulink script comments.
• Czech typography unified: en dash rather than a hyphen in prose and in ranges,
  non-breaking space before % and units, 1D/2D without a hyphen. Each convention
  is now a binding rule in §4 of the glossary — including the decision to keep the
  decimal point in prose, because the text quotes the very numbers MATLAB prints
  into the %[output:...] blocks a few lines below, and a comma would desync the two.
• README carries an Open in MATLAB Online badge, so the repository can be opened
  and run from a browser without a local installation.

PINN Toolbox 1.0.0

The first public release of the PINN Toolbox — a MATLAB namespace package (+PINN)
for solving forward and inverse PDE/ODE problems with Physics-Informed Neural
Networks, built on Deep Learning Toolbox automatic differentiation
(dlarray / dlnetwork / dlgradient / dlfeval).

What's in it

Core. Three value classes form the pipeline:

• PINN.core.NeuralNetwork — network wrapper with optional random Fourier features,
  a gated modified-MLP architecture (Wang et al. 2021) and a sin activation.
• PINN.core.LossFunction — combined data + physics loss from your own PDE residual
  handle, with optional adaptive grad-norm weighting and a divergence cap.
• PINN.core.Optimizer — Adam and L-BFGS training loops (forward and inverse) with
  dlaccelerate graph caching on by default, learning-rate schedules, early stopping,
  in-loop residual-adaptive resampling, and a non-finite-loss guard that rolls back
  and stops cleanly.

Utilities for collocation sampling (including complex geometries by rejection
sampling), hard boundary constraints, higher-order derivatives, a gradient-balance
diagnostic, and a name-value builder for the training inputs.

Examples. 19 narrated Live Scripts, one concept each, every one validated against
an analytic / pdepe / ode45 reference and reporting its relative L2 error. Czech
in examples/, with a full English mirror in examples/en/ and a guided tour in
both languages. They cover the basics, inverse problems from real sensor data, and
deliberate failure-mode studies with measured remedies — convection with curriculum
training (82.1 % → 9.93 %), stiff Helmholtz with adaptive weights and hard BCs
(34.7 % → 0.62 %), spectral bias with Fourier features (520.8 % → 5.23 %), and a
stiff reaction where the loss is near zero but the answer is 96 % wrong until
seq2seq time-marching fixes it (→ 0.12 %).

Interactive app. PINN.tools.launchTrainerApp — pick a preset or type your own
equation and data, configure the network, loss and optimizer, train, and export a
runnable script. Bilingual UI (Czech / English).

Optional companions. A Simscape Multibody digital twin that generates the pendulum
sensor data, and a trained PINN deployed as a Simulink block acting as a soft sensor.
Their generated artifacts are committed, so the core examples need only the Deep
Learning Toolbox.

Tests and tooling. 139 unit tests, including sentinels that verify numerical
equivalence with and without dlaccelerate caching, plus fast example-path smoke
tests. CI runs the suite on both supported releases.

Requirements

• MATLAB R2025a or newer — verified in CI on R2025a and R2026a, CPU only.
• Deep Learning Toolbox (required).
• Simulink, Simscape and Simscape Multibody only for the optional companion examples.
  The committed .slx models were saved in R2026a; on an older release regenerate them
  with the build_*.m scripts.

Install

From the repository (recommended — the examples resolve their data paths through
the MATLAB project):

% git clone https://github.com/JanStudnicka/pinn-toolbox.git
openProject("PINNToolbox.prj")
open examples/en/tutorial.m      % guided tour, English

As a toolbox: download PINNToolbox.mltbx below and double-click it, or run
matlab.addons.toolbox.installToolbox("PINNToolbox.mltbx").

Documentation

• doc/Tuning_Guide.md — symptom → diagnosis → fix, grounded in this repo's own
  measurements (Czech mirror: doc/Pruvodce_ladenim.md).
• doc/PINN_Architecture.md — architecture and a residual cookbook.
• Papers/README.md — the six publications the toolbox is built on.
• .claude/skills/pinn-toolbox/ — a usage skill for AI coding agents.

MIT licensed.