v0.12.0-beta — implicit stiff mode for SymmetricOperator (#16)

Implicit Stiff Mode for SymmetricOperator

The new path="implicit" option adds a dependency-free preconditioned-CG backward-Euler solver for externally-assembled symmetric operators, resolving the O(λ_max·t) timeout on stiff FEM operators.

Highlights

• Flagship benchmark: ~4 ms / sup_error 3.3e-12 on a Neumann N=400 ×10⁷ stiff operator
  • Compare: explicit Lanczos ~56 s, scipy expm_multiply >90 s timeout
• Cost caveat: governed by √κ of Jacobi-preconditioned system, not strictly λ_max-independent
• Zero new dependencies: PCG + Jacobi preconditioning are dependency-free
• API: SymmetricOperator.evolve_batched, mass_lumped_evolve, MassKOperator.evolve all support path="implicit"

CI Hardening

Alongside the #16 feature:

• Flagship Gates: Resolvent argument parsing + bounded L_RESOLVENT tick budget turned green
• Nightly miri: Scalar path (src/scalar/) validated
• Nightly bench-regression: Scoped core families + baseline guard to prevent false positives

Full Changelog

See CHANGELOG.md for complete details.

v0.11.0-beta — Python bindings for the 0.10.0 wave (#15)

[0.11.0-beta] — 2026-06-27

Python binding surface for the 0.10.0-beta feature wave (#11–#14 + A1).
Issues #11, #12, #13, #14, and the graph Krylov additions from the 0.10.x patch
series are now usable from Python via the semiflow-pde wheel. No breaking
changes to the existing Python surface or the Rust public API.

Added (Python bindings, #15)

• SymmetricOperator.from_csr / SymmetricOperator.evolve_batched —
  externally-assembled symmetric PSD sparse operators (FEM stiffness, Robin BC)
  accepted as CSR arrays; batched time-stepping across multiple right-hand sides.
• symmetric_op_expmv_frechet — combined matrix-exponential action +
  per-entry Fréchet gradient in one call. EntrySensitivity is not exposed as a
  separate Python class; the gradient is returned directly from this function.
• MassKOperator / mass_lumped_evolve — generalized-eigenproblem (M, K)
  propagation via the congruence chain  = R⁻ᵀKR⁻¹; lumped-mass 3-liner path.
• ConservativeDiffusionChernoff / assemble_conservative_csr_1d —
  conservative divergence-form diffusion with harmonic-mean face conductivities;
  to_symmetric_operator() bridge to the Krylov engine.
• phi_action / phi_action_batched / Etdrk4 — ETD φ-functions and the
  Cox–Matthews ETDRK4 semilinear integrator. Nonlinearity is menu-based
  (AllenCahn, Burgers, GrayScott, KuramotoSivashinsky); arbitrary
  per-step Python callbacks are deferred (ADR-0179 / ADR-0189 wall preserved).
• Laplacian.from_csr — direct CSR-based Laplacian construction path.
• PEP 561 stubs updated to cover all new symbols; 16/16 smoke tests green.

Changed

• Registry metadata: Changelog and Release Notes links now appear on the
  PyPI sidebar (semiflow-pde [project.urls]), the crates.io semiflow page
  (README ## Changelog section), and the npm @semiflow/wasm page
  (README ## Changelog / Release notes section).

v0.10.2-beta — Lanczos OOB fix at t≥4

Fixed

Lanczos OOB panic at t≥4 (graph_krylov)

• Off-by-one bug: wrote past MAX_LANCZOS_DIM=18 when Krylov dimension maxed out at large t·λ_max
• Bug present since A1/ADR-0185; only triggered at deep t
• Fix: Lanczos now matches Chebyshev/dense expm at t∈{4,8,16} (max|Δ|≤1e-9, no panic)
• Gate lanczos_large_t_no_oob validates the fix
• The --no-lanczos workaround is no longer needed

Clippy cast_precision_loss in graph_par_speedup example

• Cleaned up compiler warning in --all-targets build

---

See CHANGELOG.md for full context.

v0.10.1-beta — Channel-parallel graph path (multicore)

Performance

Channel-parallel graph evolve

• ADR-0184 D3: evolve_batched and batched adjoint now split channels across cores via std::thread::scope
• Gated behind parallel feature (C≥2); each worker has its own ScratchPool
• Graph Krylov ML path was previously single-threaded (parallel feature only covered Strang2D/3D grid kernels)
• Bit-identical to serial (0-ULP; forward + ascending-index gradient reduction per ADR-0184 D4/D5)
• Measured ~5–6× on i7-12700K (612–1942% CPU)
• Closes the "speed" half of the memory↔speed contradiction for graph diffusion / SSM layers

---

See CHANGELOG.md for full technical details and gate validation.

v0.10.0-beta — Krylov/Fréchet ML backend + 4 issues (#11/#12/#13/#14)

Summary

Five-feature wave (issues #11, #12, #13, #14 + A1 stiff fix): generic symmetric operators, conservative divergence-form diffusion, stiff multilayer conduction, ETD/φ semilinear integrator, and depth-independent Krylov + Fréchet for graph semigroups. All additive; no breaking changes to the 0.9.1-beta public surface.

Added

A1/A2 — Depth-independent graph-semigroup Krylov action + edge-weight Fréchet gradient

• GraphKrylovChernoff, graph_expmv_frechet; ADR-0185, math §54
• graph_expmv(L, v, t, ε) computes e^{−tL_G}·v without stepping through each time unit
• Chebyshev default (O(1) working vectors) or Lanczos adaptive (O(m·N) basis, m ≈ 20–40, flat in t)
• graph_expmv_frechet returns ∂J/∂w for all edge weights via one augmented Krylov solve (Al-Mohy & Higham 2009)
• Closes forward and backward speed ceilings exposed by v0.9.1-beta (#10)
• PyO3: GraphKrylov pyclass + graph_expmv_frechet pyfunction in one GIL-releasing call
• Honest limits: symmetric L_G only; time-varying L(t) out of scope

#13 — Generic externally-assembled symmetric-operator entry point

• SymmetricOperator, MassKOperator, EntrySensitivity; ADR-0186, math §55
• SymmetricOperator::from_csr accepts any externally-assembled symmetric PSD sparse operator
• FEM stiffness with Robin BC, anisotropic conductivity — not just graph Laplacians
• MassKOperator propagates e^{−τM⁻¹K}·v via congruence chain without forming M⁻¹K explicitly
• Lumped (M,K) path is a 3-liner
• Honest limits: symmetric PSD required; consistent-mass differentiability deferred

#11 — Conservative (divergence-form) variable-coefficient diffusion

• ConservativeDiffusionChernoff, assemble_conservative_csr_1d; ADR-0187, math §56
• Harmonic-mean face conductivities reproduce series-resistance network at machine precision
• Handles sharp material interfaces (k-contrast 100:1 to 3025:1) where non-conservative methods give ≥50% error
• Crank–Nicolson wrapper (order()=2, A-stable, no CFL, one O(n) Thomas solve)
• Honest limits: k > 0; full-tensor non-separable ∂_x(k∂_y) out of scope

#14 — Stiff multilayer conduction via mass-weighted Krylov

• MultilayerStack, multilayer_evolve, MassWeightedConservativeChernoff; ADR-0188, math §57
• Maps physical [(thickness, k, ρc)] → node arrays on single uniform grid
• Propagates in one depth-flat Krylov action for any integration span
• Beats explicit CFL by ~28000× on Shuttle TPS stack (k-contrast ≈ 3025×, ρc-contrast ≈ 27×)
• Optional A-stable CN convenience for O(1)-vector memory
• Honest limits: 1-D first; one global dx; node-centered lumped mass only

#12 — ETD φ-functions and ETDRK4 semilinear integrator

• phi_action, phi_action_batched, Etdrk4, Nonlinearity; ADR-0189, math §58
• phi_action_batched computes φ₀…φ_p(τA)·v simultaneously via ONE augmented block-triangular matvec-only Taylor action
• Etdrk4 (Cox–Matthews 2002 / Kassam–Trefethen 2005) integrates ∂ₜu = Lu + N(u) at order 4
• N(u) is declarative Nonlinearity trait or fixed enum menu (AllenCahn, Burgers, GrayScott, KuramotoSivashinsky)
• Never a per-step Python callback; ADR-0179 wall preserved
• NonlinearityDiff enables end-to-end adjoint ∂J/∂param through one step
• Honest limits: L 1-D divergence-form or symmetric graph only; deferred beyond this scope

Fixed

A1 stiff operator NaN bug (commit 9e5f557)

• GraphKrylovChernoff's Chebyshev path silently returned garbage for stiff operators (λ_max ≳ 1400)
• Fix: substep s = ⌈z/Z_SAFE⌉ (Z_SAFE = 200), mirroring Lanczos scaling
• Plus fail-loud finiteness guard returning SemiflowError::DomainViolation
• Conservative high-contrast diffusion (λ_max ∼ 10⁴) was affected
• New test cheb_stiff_regime (z ≈ 6400) was red before, green after (sup_error 1.0e-11)

---

See CHANGELOG.md for ADRs, validation gates, and complete honest limitations.

v0.9.1-beta

semiflow v0.9.1-beta

v0.9.0-beta.3

semiflow v0.9.0-beta.3

v0.9.0-beta.2

semiflow v0.9.0-beta.2

v0.9.0-beta.1

Release v0.9.0-beta.1 — manylinux x86_64 + macOS arm64 + Windows whee…