feynlag 0.3.0

Added

• diagonalize_svd(method="auto"|"symbolic"|"numeric", dps=50): a numeric
  matrix that is complex, floating-point or larger than 2×2 is decomposed with
  mpmath's complex SVD, returning unitary rotations with
  R_L M R_R† = diag(m ≥ 0) (rows by increasing mass, a round-off singular
  value as an exact zero). Generic complex 3×3 Yukawas now give their masses
  and V_CKM = R_uL·R_dL† inside feynlag (FG-6, #31). Exact 2×2 calls are unchanged.
• Rotation.bar(old_bar, new_bar): the bar-leg partner of a fermion rotation,
  with the conjugate matrix R* (identical to R for a real rotation) (#31).
• GlobalU1: a global (ungauged) U(1) whose assign takes one charge per
  flavour for a fermion (e.g. Froggatt–Nielsen Q_L: (3, 2, 0)), checked by
  the new check_global_invariance and declared on a model with
  Model(global_groups=[...]) (check_invariance/validate report
  global:<name>). Not part of the anomaly check (FG-7, #32). Legs sharing a
  symbolic flavour index are charged flavour by flavour, so a flavour-diagonal
  current ψ̄_i Γ ψ_i (a kinetic term, fermion_gauge_current) is neutral even
  with flavour-dependent charges; an out-of-range integer flavour index, or a
  symbolic index shared by fields with different flavour counts, raises a
  ValueError.

Changed

• The SU(N) groups tutorial (#15) and the Particle Decays tutorial (#30) are
  rebuilt in the derive-then-check style: a prediction before each section,
  and an assert behind every claim instead of printed booleans. They add
  basis-independent invariants, real/complex/pseudo-real representations and
  singlet counting (SU(N)), hand-built Dirac-matrix cross-checks of the
  covariant traces (decays), and web-verified reference lists.

Fixed

• diagonalize_svd no longer returns a silently wrong answer on a complex
  matrix: the symbolic route raises on an explicit I. It also no longer
  crashes with TypeError on a real floating-point input (#31).

feynlag 0.2.0

Added

• diagonalize_takagi(method="auto"|"symbolic"|"numeric", dps=50): for a
  numeric matrix larger than 2×2, auto factorises with mpmath's eigsy at
  high precision (columns by increasing mass, round-off eigenvalues as exact
  zero modes), so a generic 3+2 seesaw 5×5 finishes instead of stalling in the
  exact route. Exact 2×2 and symbolic calls are unchanged.
• Per-symbol LaTeX: component_tex=[...] on Scalar, WeylFermion/
  MajoranaFermion, GaugeBoson and GaugeGroup.bosons(), tex= on
  conjugate_pair, and a parameter's existing tex= now reach plain
  sympy.latex through the symbol itself (TexSymbol), so downstream code
  needs no name→LaTeX map. A one-component field's tex= also names its
  component, and fermion bar legs print as \overline{…}.
• Opt-in LaTeX names in the SM builders and Scalar.expand_vev: each takes
  tex=, a map from the name of a symbol it creates to a LaTeX name
  (electroweak_gauge, higgs_doublet, electroweak_scaffold,
  weinberg_rotation, charged_current_rotation, to_physical_basis,
  standard_ckm), so one global table covers the couplings, the Higgs
  parameters, the doublet and its fluctuations (H0_r→h), and the physical
  Z, A, W±, G⁻. Names missing from the map stay plain. An explicit
  symbol_names printer setting still overrides a symbol's own tex.
  GaugeGroup.bosons() raises if a later call asks for a different tex than
  the cached bosons carry.
• A TexSymbol sorts exactly like a plain Symbol of the same name, so
  adding a tex never reorders vertex legs or printed terms.

Fixed

• The UFO writer now escapes texname/antitexname. It wrote them as bare
  '...' literals, so a tex containing a quote (g') made parameters.py
  a SyntaxError, and a backslash was read as a Python escape (\tau →
  TAB + au).

Changed

• A field component or parameter declared with a tex is a TexSymbol, which
  is not equal to a plain Symbol of the same name. Code that rebuilds such a
  symbol by name (sp.Symbol("lam")) must use the object instead. Without a
  tex nothing changes.

feynlag 0.1.0

First public release (beta).

Model building and checks

• Parameters split into external and internal, with dependencies resolved in
  order; scalar, Weyl-fermion, Majorana-fermion and gauge-boson fields.
• Gauge groups U1 and SUN (any SU(N) irrep, generators built in the
  Gelfand–Tsetlin basis); discrete groups ZN and S3 with Clebsch–Gordan
  products.
• Lagrangian building blocks: Dmu, dag, Bilinear, MajoranaBilinear,
  fermion_gauge_current.
• Invariance checks (gauge, discrete, hermiticity, mass dimension with an
  EFT max_dim flag) and a Model.validate() umbrella adding anomaly
  cancellation, electric-charge conservation and a UFO round-trip.
• Operator suggestion (suggest_potential, suggest_yukawa,
  suggest_kinetic, build_lagrangian), including the dim-5 Weinberg
  operator.
• Reusable Standard Model builders (feynlag.models) and CKM mixing
  (standard_ckm).

Symmetry breaking, masses and vertices

• VEV expansion, tadpole extraction and solving, and scalar, charged,
  gauge-boson, Dirac and Majorana mass matrices, including the type-I seesaw.
• Diagonalization: orthogonal and analytic 2×2, SVD (biunitary), Takagi,
  and MajoranaRotation for Majorana mass eigenstates.
• Momentum-space Feynman rules for SSS, SSSS, VSS, VVS, VVSS, VVV, VVVV, FFS,
  FFV, four-fermion (FFFF) and Majorana vertices.

Export

• LaTeX vertex tables.
• UFO model directories for MadGraph, with SU(3) colour tensors and
  four-fermion Lorentz structures; the exported SM reproduces MadGraph's
  stock sm cross sections (docs/benchmark.md).

Phenomenology (feynlag.pheno)

• Tree-level 1→2 partial widths, total widths and branching ratios, with a
  DiracParticle abstraction for fermions.
• Off-shell h→WW*/ZZ* widths (1→3 with a Breit–Wigner propagator).
• Loop-induced h→gg, h→γγ and h→Zγ from standard effective vertices.
• Tree-level single-diagram 2→2 cross sections, including the γ₅ (ε·ε) term
  and the forward–backward asymmetry.

Known limitations

• No R_ξ gauge fixing or ghosts.
• Majorana vertices are not exported to UFO.
• 2→2 scattering does not yet sum interfering diagrams; VVV decays are not
  implemented.