CHANGELOG.md

Changelog

All notable changes to this project are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Added

• Add SO2 methods norm and normalized @elias-taufer (#239)
• Rigid-body inertia and the free joint. SpatialInertia for a body's mass, balance point, and resistance to spinning, and FreeJointState for the pose and velocity of a body free to move in all six directions, with a SpatialError for the mass properties that do not describe a usable body. @kmolan
• MuJoCo model loading. A multicalc-mjcf crate reading one free-floating body out of an MJCF file, working the mass out from the shapes where the file does not state it, checked against MuJoCo's own compile of the same file. Ships with a model_ingestion demo. @kmolan
• Vector2D / Vector3D / Vector6D and Matrix2D / Matrix3D / Matrix4D / Matrix6D aliases for the vector and matrix sizes used most, so a call site writes Vector3D<T> rather than Vector<3, T>. Each names the same type as the long spelling, so the two mix freely and nothing existing has to change. @kmolan
• Quaternion::from_two_vectors, Quaternion::rotation_angle_to, and Quaternion::inverse_transform_point. (#214)
• Add Default trait implementation for spatial types Quaternion, SO2, SO3, SE2, SE3, Twist, Wrench. @elias-taufer (#244)

Fixed

• ResamplingScheme::resample_indices now leaves output indices untouched for empty weights instead of panicking. (#204)

0.9.0 - 2026-07-26

A feature release adding state estimation, feedback control, wheeled kinematics, and waypoint paths, plus a prelude and one-call entry points for the calculus core.

Added

• Const-evaluable solver constructors, AutoDiff new(), and Gaussian quadrature node tables / nodes. @rtmongold (#168)
• Add Vector/Matrix get / get_mut, Matrix::try_row / try_column, and Matrix::as_mut_slice_rows. @rtmongold (#185)
• Kinematics. A kinematics module with DifferentialDrive, a unicycle model, and dead-reckoning odometry, plus a kinematics demo. @kmolan (#170)
• State estimation. An estimation module with KalmanFilter and ExtendedKalmanFilter (the latter differentiating its nonlinear models for the Jacobians each step), a CovarianceUpdate choice for how the covariance is recomputed, cross-validation fixtures, and an estimation demo. @kmolan (#175)
• Particle filter. A bootstrap/SIR ParticleFilter with pluggable resampling and measurement likelihood, behind the alloc feature. @kmolan (#188)
• Control. A control module with Pid, a one-pole derivative filter (OnePoleLowPass), and the pure-pursuit path-following law. @kmolan (#179)
• Follow-the-gap. A reactive obstacle-avoidance controller in control, with an avoidance demo. @kmolan (#188)
• Waypoint paths. A motion module with PolylinePath and its arc-length, closest-point, and lookahead queries. @kmolan (#179)
• Random numbers. A random module with the RandomSource trait and the seedable Pcg32 generator, giving uniform and normal draws without a heap or std. @kmolan (#188)
• Prelude and one-call functions. multicalc::prelude for the traits and short functions worth importing together, plus derivative, second_derivative, partial, and integral, which need no configured backend. @kmolan (#198)
• General N×N determinant and inverse. Sizes past 4×4 now factor through LU instead of being unsupported; small sizes keep their closed forms. @kirloo (#184)
• 3D surface flux integral. flux_integral_3d and flux_integral_3d_custom. @mubasharameen485-cloud (#172)
• Property tests that LU and QR factorizations reconstruct their input matrix. @yvoolab (#189)
• A localization and obstacle-avoidance lap demo, with simulated lidar, an occupancy grid, a particle-filter localizer, latency benchmarks, and embedded smoke fixtures. @kmolan (#192)
• docs.rs now builds with all features so gated items appear in the published docs. @kmolan (#197)
• Add Matrix methods: trace, frobenius_norm, from_diagonal @birchmd (#191)
• Add Vector methods: normalize, normalized, map @birchmd (#191)
• Add impl Div where T: Numeric for Matrix and Vector @birchmd (#191)

Fixed

• Iterative integrators return IntegrateError::LimitsIllDefined instead of NaN if classify fails after validation. @rtmongold (#171)
• RK45 now handles zero-dimensional states without producing NaN norms. @terminalchai (#174)
• 2×2/3×3/4×4 Matrix::inverse reject near-singular inputs via an EPSILON-scaled det check instead of exact det == 0. @rtmongold (#181)
• Spatial small-angle Taylor cutoffs use 30 · EPSILON and (30 · EPSILON)² instead of fixed 1e-6 / 1e-12. @rtmongold (#190)
• The iterative and Gaussian integrators check their samples and return IntegrateError::NonFinite, instead of letting a blown-up integrand propagate into a garbage result. @SAY-5 (#180)
• The multi-variable integrators bounds-check the variable index and return the new IntegrateError::IndexOutOfRange. @Devil1716 (#178)

Changed

• Keep panicking Index / IndexMut as the ergonomic Matrix/Vector access path. Remove panicking row / column (use try_row / try_column). @rtmongold (#185)
• Vector-field functions renamed and re-exported. curl::get_2d, line_integral::get_2d, and the rest are now curl_2d, line_integral_2d, and so on, imported straight from vector_field. @kmolan (#199)
• Kalman filter takes a KalmanModel. The four model matrices are one named struct instead of four positional constructor arguments. @kmolan (#198)
• The crate root re-exports the differentiation, integration, and approximation types, so they can be named without their module paths. @kmolan (#198)
• Rewrote the test suite for readability: table-driven cases, named expectations, and shared helpers. @kmolan (#200)
• Golden fixture regeneration runs from a script in a local virtualenv with locked, hash-checked dependencies and a pinned timestamp, so a rerun produces the same bytes. @kmolan (#201)
• Dropped CI caching, which was causing flaky pipeline failures. @kmolan (#173)

0.8.0 - 2026-07-16

A feature release adding spatial math (quaternions, Lie groups, twists/wrenches), ODE solvers, matrix-based discretization, and per-module error enums.

Added

• vector! and matrix! macros for building [Vector] and [Matrix] literals, e.g. vector![1.0, 2.0, 3.0] and matrix![[1.0, 2.0], [3.0, 4.0]]. @rtmongold (#37)
• Autodiff scalars example. A runnable walkthrough (examples/autodiff_scalars.rs) that uses Dual and HyperDual directly (no derivator). @rtmongold (#106)
• Fixed examples/README.md table entry for optimization_solvers (was an orphan bullet). @rtmongold (#117)
• Primal scalar projection. New to_f64 / to_f32 trait for edge use (viz, logging, etc.), kept separate from Numeric. showcase/viz now uses it instead of its private Plottable trait. @ProtoFN (#131)
• Property tests for linear/quadratic approximation invariants (exactness on matching-degree polynomials, metrics consistency). @rtmongold (#130)
• Property tests that autodiff and central finite-difference derivatives agree on random polynomial compositions. @rtmongold (#129)
• Opt-in Kahan compensated summation for iterative integration and approximation metrics via .with_kahan_summation(); pairwise remains the default. @rtmongold (#134)
• RISC-V embedded smoke. embedded-smoke runs on riscv32imc-unknown-none-elf under QEMU virt, gated with its own flash/stack budgets. @rtmongold (#140)
• Quaternion type. A spatial::quaternion module with unit-quaternion rotations. @kmolan (#128)
• Lie groups. A spatial::lie module with SO(2), SE(2), SO(3), and SE(3) (exp/log maps, composition, and group action), plus an examples/lie_groups.rs walkthrough. @kmolan (#133)
• Twists and wrenches. spatial::twist and spatial::wrench screw-theory types. @kmolan (#137)
• Discretization and matrix exponential. A new discretization module and a linear_algebra::expm matrix exponential. @kmolan (#137)
• ODE solvers. A new ode module with a fixed-step RK4 and an adaptive RK45 integrator, plus an examples/ode.rs walkthrough and an ode benchmark. @kmolan (#135)
• Extended Numeric trait. Inverse-trig (asin/acos/atan/atan2), hyperbolic (sinh/cosh/tanh), and copysign/floor/signum/hypot/powf/mul_add/ recip, so autodiff scalar types support them. @kmolan (#120)
• Property tests for numerical integration invariants. @kirloo (#122)
• Showcase demos. A 3D arm IK demo and a 2D arm IK demo (renamed from ik_servo), with a CSV/Rerun sink abstraction in the demos crate. @kmolan (#145)

Changed

• Per-module error enums. Fallible operations now return family-specific enums (LinalgError, DiffError, IntegrateError, SolveError), all convertible into the CalcError umbrella, replacing the monolithic error codes. @kmolan (#136)
• Faster Jacobians for large inputs: the multi-variable Jacobian now uses a single column solve instead of per-column solves. @kmolan (#138)
• Reorganized the benchmark suites and harness. @kmolan (#143)
• Refreshed the crate README and updated its links and media. @kmolan (#144, #146, #147)
• Internal refactor of the QA test and oracle scaffolding. @kmolan (#141)
• Updated the ignored advisories in deny.toml. @kmolan (#142)

Fixed

• Fixed misspelled examples/ link labels in optimization/README.md. @rtmongold
• Corrected the CONTRIBUTING.md link from the README. @kirloo (#121)

0.7.2 - 2026-07-09

Added

• f32 pseudo-inverse identity tests. The SVD pseudo-inverse is now checked against the Moore–Penrose conditions in f32, complementing the existing f64 coverage. @KSHITIZ6341 (#110)
• Gauss-Newton example. A runnable walkthrough (examples/optimization_solvers.rs) and accompanying docs for the Gauss-Newton solver. @tapheret2 (#108)
• A pull-request template and a workflow that greets first-time issue and PR authors.
• Two showcase demo GIFs (1 kHz IK servo, Newton fractal) in the crate README.

Changed

• Refreshed crates.io metadata (description, keywords, categories) for discoverability.
• Renamed CONTRIBUTIONS.md to CONTRIBUTING.md and rewrote it with a host-only quickstart and contribution workflow.

0.7.1 - 2026-07-07

A tooling and infrastructure release. The library API is unchanged; the repository is now a Cargo workspace with cross-validation, embedded, and visualization crates alongside the core multicalc crate.

Added

• Workspace layout. The repository is now a Cargo workspace; the library lives in crates/multicalc, with supporting crates and tools split out. (#90)
• Embedded smoke tests. A no_std firmware crate (crates/embedded-smoke) that builds and runs core APIs on Cortex-M targets, guarding embedded compatibility in CI. (#92, #100)
• Oracle. Cross-validation fixtures for linear algebra, optimization, and quadrature, generated from reference implementations, with tests checking multicalc against them. (#93)
• multicalc-viz showcase. Rerun-based visualization examples under showcase/viz (Fourier Ferris, gradient marbles, IK servo, Newton fractal, live/recorded curve fit), plus demo reels in the README. (#97)
• Nightly CI and expanded build matrix. A nightly workflow and additional target builds. (#94)

Changed

• Benchmark suites are unified under a single benchmarks target: use criterion substring filters (cargo bench -- linear_algebra) instead of --bench <suite>. @rtmongold (#96)
• Improved the README project description and switched image sources to direct GitHub links so they render off-crates.io. (#91, #98)
• Miscellaneous CI cleanup. (#95)

0.7.0 - 2026-07-06

A feature release adding automatic differentiation, a linear-algebra core, nonlinear optimization, and root finding, while staying strict no_std.

Added

• Automatic differentiation. New scalar module with dual, hyper-dual, and Jet number types built on a shared numeric-traits abstraction. Autodiff is now the default backend for derivatives, gradients, Jacobians, and Hessians, with finite differences still available.
• Linear algebra. Fixed-size, const-generic Vector and Matrix types plus small-matrix solves: inverse, LU, Cholesky, QR, and SVD.
• Optimization. A nonlinear least-squares optimization module with Gauss-Newton, Levenberg-Marquardt, and trust-region solvers.
• Root finding. A root_finding module with bisection, Newton's method, and Newton for systems of equations.
• Pairwise summation utility for numerically stable sums, applied across iterative integration and approximation.
• An embedded curve-fitting example and an embedded feature marker that keeps the firmware example off host builds.
• A README for every source module, and benchmark suites split by area (calculus, linear algebra, optimization, root finding).

Changed

• Autodiff replaces finite differences as the default differentiation method.
• Public integration tests moved from src/**/test.rs into top-level tests/, and BENCHMARKS.md was split into per-area files under benches/.

0.6.0 - 2026-06-28

A breaking rewrite focused on real-time latency, ease of use, maintainability, and strict no_std.

Added

• Infinite and semi-infinite integration limits for the iterative integrators (use f64::INFINITY / f64::NEG_INFINITY), via a domain transform. Accurate for convergent, decaying integrands.
• A runnable example for every module under examples/, and BENCHMARKS.md with accuracy and latency figures.
• A criterion benchmark suite under benches/, plus deterministic work-count regression tests that fail if a latency optimization is reverted.
• CalcError implements core::error::Error and Display.

Changed

• f64-only API. Dropped the num_complex::ComplexFloat generic; transcendental functions now come from libm, so the crate is genuinely no_std.
• Errors are a typed CalcError enum instead of &'static str strings.
• Functions and integrands are taken as generic F: Fn(...) (monomorphized) rather than &dyn Fn.
• Removed the redundant runtime "number of integrations" / "derivative order" arguments; these are now inferred from the const-generic array lengths.
• Renamed the solver types to FiniteDifferenceSingle/Multi, IterativeSingle/Multi, and GaussianSingle/Multi.
• Unified the vector-field calling convention: line and flux integrals now take &[&dyn Fn(&[f64; N]) -> f64; N], matching curl and divergence.
• Gaussian quadrature takes the bare integrand; the tabulated weights carry the e^{-x^2} / e^{-x} weighting factor.
• Per-solver settings are now a plain-data config field (e.g. solver.config.method = ...) instead of getters and setters.
• Renamed the heap feature to alloc.
• Default iterative interval count is now 120 (a multiple of 12) so the Boole and Simpson rules stay exact.

Fixed

• Gauss-Hermite and Gauss-Laguerre no longer double-apply the weighting function.
• Gauss-Laguerre multi-fold recursion (it read the Legendre table and failed to decrement the fold).
• The 3D line-integral z-component, and the reversed / equal / NaN limit checks (negative ranges such as [-2.0, 1.0] are now accepted).
• The linear and quadratic approximation formulas and their goodness-of-fit metrics.

Removed

• The num-complex dependency, the ComplexFloat generic, and f32 / complex-number support.