mathlib-fp 1.3.0 release notes

Release date: 2026-07-23

Highlights

• MathBase.Complex introduces the value-record TComplex, TComplexArray, stable arithmetic, principal elementary functions, and inverse complex trigonometric and hyperbolic functions.
• AlgebraLib.VectorKernels introduces real and complex contiguous-array kernels through TVectorKit. The existing matrix-as-vector API remains source-compatible.
• EngineeringLib.Signal now uses TComplexArray as its FFT core while retaining split real/imaginary overloads for existing callers.

Vector kernels

Real vectors now provide elementwise multiplication/division, compensated Sum and Dot, Mean, Min, Max, stable Norm2, and normalization. Complex vectors provide non-conjugating and Hermitian dot products. Allocating transformations have ...Into variants so callers can reuse destination buffers in repeated computation. Inputs must be finite and paired vectors must have equal lengths. Empty sums, dots, and norms are zero; Mean, Min, Max, and normalization of a zero vector raise EVectorError.

Numerical behavior

Complex division is scale-safe for finite extreme-scale operands whenever the result is representable. Complex magnitude, exponential, and square root functions handle tested infinity and NaN limits without avoidable invalid operations. Principal CLog, CSqrt, and inverse functions distinguish branch sides through signed zero. Inverse functions preserve tiny first-order values and use scaled component or asymptotic large-input forms that avoid target-sensitive z*z cancellation. Tests cover reference values, extreme scales, branches, NaN/infinity behavior, array kernels, and complex FFT round trips.

For a full complex inverse transform, call FFT(Data, True) on a TComplexArray. The CalculateIFFT convenience overloads continue to return real-valued output arrays by design.

Compatibility

No existing public identifier was removed or renamed. IVector = IMatrix and the split real/imaginary FFT procedures remain supported. The new complex and array-vector APIs are additive.

See MathBase, AlgebraLib, and EngineeringLib for API details, or run examples/14_complex_vectors.pas for a compact end-to-end example.

The example shows inverse complex functions, elementwise real-vector arithmetic, AxpyInto destination reuse, a Hermitian dot product, and an in-place complex FFT/IFFT round trip.

Validation

The release contains 819 passing tests in Win64 normal, optimized, runtime-checked, and heap-traced local builds, with zero unfreed blocks; the optimized Win32 suite also passes. The Lazarus package builds for Win64 and Win32, all 15 examples compile and run, and representative benchmarks cover complex arithmetic, vector kernels, and the native complex FFT. Linux and Windows CI both pass. The reusable release process is documented in RELEASING.md.

See the 1.3.0 changelog for the complete release entry.