Skip to content

README.md

Latest commit

 

History

History
407 lines (356 loc) · 20.7 KB

README.md

File metadata and controls

407 lines (356 loc) · 20.7 KB
Raw

iontrap-dynamics

Open-system quantum dynamics of trapped-ion spin-motion systems.

New here? See docs/overview.md for an accessible introduction to what we do, how we work, and what we don't claim — written with three reading levels for students, collaborators, and curious visitors.

iontrap-dynamics is a domain-specific Python library for modelling trapped-ion spin-motion physics with explicit, typed configuration objects for species, drives, modes, and measurement conventions. The project is built around a hard separation between physics, apparatus, and observation layers. QuTiP is the reference backend; a JAX / Dynamiqs backend is available on the same public solver surface via backend="jax".

Status

v0.6.0 released 2026-06-04 — adds the reduced light–matter models surface (Jaynes–Cummings / anti-Jaynes–Cummings / quantum-Rabi builders plus a model_deviation comparison harness) and completes the two-mode / motional surface, on top of the v0.5.0 estimation / quantum-Darwinism and two-mode / motional service surfaces and the device-neutral JAX spectrum backend. Phase 0 foundations, Phase 1 (dynamics core + measurement layer + systematics layer + registered entanglement observables), Phase 2 (performance track + JAX / Dynamiqs backend end-to-end), the Clos 2016 integration (full-Lamb–Dicke carrier dynamics, exact-diagonalization spectrum tools, N = 1 / 2 / 3 reproduction inside declared tolerances), and the v0.5.0 / v0.6.0 service surfaces are all shipped on main. CONVENTIONS.md carries §1–25: §19–24 (Fisher information / quantum Darwinism, two-mode squeezing / SU(1,1), motional CPTP channels) sealed at the v0.3 Convention Freeze (CONVENTION_VERSION 0.2 → 0.3), and §25 (reduced light–matter models) sealed by WP-03 (CONVENTION_VERSION 0.3 → 0.4). Additive only: existing v0.2.0 callers see no behaviour change — every new backend= kwarg defaults to "qutip", every channels= argument defaults to none, every JAX entry is behind an opt-in backend="jax".

The v0.6.0 cut bundles two things. The two-mode / motional service surface is complete: interferometric fringe observables (MCD), Lamb–Dicke regime helpers — Debye–Waller factor, a deep/intermediate/beyond classifier, and the all-orders sideband-Rabi forms (MCE), a probe-QFI benchmark that consumes the estimation primitive (MCF), and a motional ModeFrequencyDrift systematic (MCG). And the reduced light–matter models surface (WP-03) is new: the reduced_models JC / AJC / QRM builders, a model_deviation comparison helper, an analytic oracle suite (the four hierarchy cases), Tutorial 18 with a deterministic comparison benchmark, and the vendored model-hierarchy companion note (docs/models-hierarchy.md), anchored by CONVENTIONS §25 — all additive.

End-to-end stacks work dynamics-through-statistics, on either backend:

  • DriveConfig → carrier_hamiltonian → solve(...) → expected π-pulse flip (Phase 1 core; QuTiP default, or backend="jax" for the Dynamiqs path — cross-backend agreement under 1e-3 across all builder families).
  • TrajectoryResult → SpinReadout.run → MeasurementResult → Wilson CI band on finite-shot estimator (Phase 1 measurement).
  • Base DriveConfig → perturb_carrier_rabi → ensemble of solves → inhomogeneous-dephasing signature (Phase 1 systematics).
  • detuned_carrier_hamiltonian(..., backend="jax") → solve(..., backend="jax") → TrajectoryResult(backend_name="jax-dynamiqs") with StorageMode.LAZY per-index loader (Phase 2 JAX backend end-to-end).

Phase 0 + Phase 1 + Phase 2 artefacts delivered:

  • Public conventions frozen in CONVENTIONS.md v0.3 (§1–24 complete; §19–24 added at the v0.3 Convention Freeze, post-v0.4.0).
  • Three-layer regression harness populated: migration (5 / 5 scenarios with legacy qc.py-generated references, bit-identical across three runs; 3 / 5 active comparisons, 2 / 5 skipped with probe-informed blockers), analytic (6 closed-form formulas), invariant (9 checks).
  • Cache-integrity contract + tests.
  • CI with ruff, ruff-format, mypy strict, pytest, pa11y WCAG 2 Level A (hard gate, AA advisory per v0.3.2 amendment).

Today the importable code surface covers:

Foundation (Phase 0)

  • iontrap_dynamics.exceptions — canonical exception hierarchy (IonTrapError, ConventionError, BackendError, IntegrityError, ConvergenceError)
  • iontrap_dynamics.results — frozen TrajectoryResult schema with storage-mode consistency enforcement
  • iontrap_dynamics.cache — hash-verified .npz + JSON persistence
  • iontrap_dynamics.conventions — CONVENTION_VERSION marker
  • iontrap_dynamics.invariants — density-matrix / state-vector validators
  • iontrap_dynamics.analytic — closed-form reference formulas (carrier Rabi, sideband rates, Lamb–Dicke parameter, coherent-state occupation), plus the Lamb–Dicke regime helpers (debye_waller_factor, lamb_dicke_regime → LambDickeRegime, all-orders {red,blue}_sideband_rabi_frequency_full_ld)

Configuration layer (Phase 1)

  • iontrap_dynamics.operators — single-ion Pauli set in the atomic-physics convention (sigma_z_ion, sigma_plus_ion, ...; see CONVENTIONS.md §3)
  • iontrap_dynamics.species — IonSpecies, Transition, TransitionType and factories for ²⁵Mg⁺, ⁴⁰Ca⁺, ⁴³Ca⁺
  • iontrap_dynamics.drives — DriveConfig (wavevector, Rabi, detuning, ...)
  • iontrap_dynamics.modes — ModeConfig with CONVENTIONS.md §11 normalisation enforced at construction
  • iontrap_dynamics.system — IonSystem composition with cross-validation
  • iontrap_dynamics.hilbert — HilbertSpace implementing the §2 tensor ordering, operator embedding helpers, motional primitives (a, a†, n̂)
  • iontrap_dynamics.states — ground_state ket, compose_density, coherent_mode / squeezed_vacuum_mode / squeezed_coherent_mode factories (CONVENTIONS.md §6, §7), the two-mode two_mode_squeezed_vacuum factory (§23), and the ghz_state / cat_mode entangled-probe factories
  • iontrap_dynamics.observables — named Observable records and expectations_over_time(...); spin x/y/z, multi-ion parity (σ_z product), mode number, plus the interferometric fringe analysis fringe_visibility / fit_fringe → FringeFit
  • iontrap_dynamics.entanglement — nonlinear trajectory evaluators for concurrence, entanglement_of_formation, and log_negativity (with partition="spins" | "modes" for bipartite splits). Consume TrajectoryResult.states under storage_mode=EAGER
  • iontrap_dynamics.systematics — dynamics-side noise models (§18 — frozen at v0.2). Jitter primitives RabiJitter (multiplicative on Ω), DetuningJitter (additive on δ), PhaseJitter (additive on φ) with perturb_* ensemble helpers. Parallel drift primitives RabiDrift, DetuningDrift, PhaseDrift (deterministic single-value offsets) with apply_*_drift composition helpers — plus the motional ModeFrequencyDrift (ω_m → ω_m·(1+δ) on a ModeConfig) and the correlated CommonModePhase channel. SPAM primitives SpinPreparationError, ThermalPreparationError produce per-subsystem density matrices via imperfect_spin_ground / imperfect_motional_ground that compose into a full initial state via states.compose_density.
  • iontrap_dynamics.information — the estimation surface (§19–20): SLD quantum Fisher information, classical Fisher information, Cramér–Rao bound and a linear-Gaussian oracle (quantum_fisher_information_trajectory, classical_fisher_information, cramer_rao_bound), plus quantum-Darwinism redundancy and Schumacher–Nielsen recoverability.

Dynamics (Phase 1, full builder family)

The public Hamiltonian surface is symmetric across four families:

exact (time-indep. Qobj) detuned (list format)
carrier carrier_hamiltonian detuned_carrier_hamiltonian
red sideband red_sideband_hamiltonian detuned_red_sideband_hamiltonian
blue sideband blue_sideband_hamiltonian detuned_blue_sideband_hamiltonian
MS gate ms_gate_hamiltonian detuned_ms_gate_hamiltonian

Plus modulated_carrier_hamiltonian (time-dependent envelope primitive), two_ion_{red,blue}_sideband_hamiltonian (single-tone shared-mode), the two-mode two_mode_squeezing_hamiltonian / beamsplitter_hamiltonian (SU(1,1) / SU(2), §23), and a full_lamb_dicke: bool flag on the sideband builders (Wineland–Itano all-orders operator via matrix exponentiation). Solver entry point: iontrap_dynamics.sequences.solve(...) — accepts both Qobj and QuTiP list-format Hamiltonians, enforces the §13 Fock-saturation ladder on every call, and accepts typed motional CPTP channels via solve(channels=…) (iontrap_dynamics.channels: AmplitudeDamping, Heating, Dephasing, with optional time windows; §24).

Measurement (Phase 1, v0.2 — frozen)

  • iontrap_dynamics.results.MeasurementResult — Result sibling carrying the ideal / sampled dual-view mandated by WORKPLAN §5; enforces shots ≥ 1 and storage_mode = OMITTED at construction.
  • iontrap_dynamics.measurement.BernoulliChannel — per-shot Bernoulli sampler; returns (shots, n_inputs) bits with leading shot axis (CONVENTIONS.md §17.1).
  • iontrap_dynamics.measurement.BinomialChannel — aggregated sampler; returns (n_inputs,) int64 counts (CONVENTIONS.md §17.7, shape classes). Distributionally equivalent to summing Bernoulli bits, not bit-identical under matched seed.
  • iontrap_dynamics.measurement.PoissonChannel — per-shot photon- counting channel; consumes non-negative rates (not probabilities) and returns (shots, n_inputs) int64 counts via rng.poisson(λ).
  • iontrap_dynamics.measurement.DetectorConfig — detector-response parameters (efficiency η, dark-count rate γ_d, threshold N̂) with apply(rate), discriminate(counts), and classification_fidelity(lambda_bright, lambda_dark) methods. Composes explicitly with PoissonChannel via Poisson thinning plus additive background (exact; §17.8).
  • iontrap_dynamics.measurement.sample_outcome(channel, inputs=..., shots, seed, upstream) — orchestrator that seeds the RNG, dispatches uniformly on channel type, stores inputs under ideal_outcome[channel.ideal_label] ("probability" or "rate"), and inherits upstream-trajectory metadata when supplied.
  • iontrap_dynamics.measurement.SpinReadout — first protocol-layer composer. .run(trajectory, shots, seed) executes the projective- shot readout model (§17.9) and returns a MeasurementResult with per-shot counts / bits / bright-fraction plus the ideal p_↑ and TP · p_↑ + (1 − TN) · (1 − p_↑) envelope.
  • iontrap_dynamics.measurement.ParityScan — two-ion joint readout protocol (§17.10). Reconstructs P(s_0, s_1) from ⟨σ_z^i⟩, ⟨σ_z^j⟩, and ⟨σ_z^i σ_z^j⟩, draws joint categorical samples so entangled-state correlations survive, and returns parity estimate
    • envelope shrunk by (TP + TN − 1)². Requires the new iontrap_dynamics.observables.parity factory.
  • iontrap_dynamics.measurement.SidebandInference — motional-state thermometry protocol (§17.11). Takes paired RSB / BSB trajectories and reports fidelity-corrected n̄ = r / (1 − r) via the short-time Leibfried–Wineland ratio; independent RNG streams per sideband; NaN propagates on singular ratios.
  • iontrap_dynamics.measurement.wilson_interval, clopper_pearson_interval, and binomial_summary / the BinomialSummary dataclass (§17.12) — vectorised confidence intervals on binomial shot counts. Wilson is the recommended default; Clopper–Pearson is exact and conservative.

The measurement track is complete: CONVENTIONS.md §17.1–17.12 close the read-through, with §17 frozen at the v0.2.0 Convention Freeze gate.

Phase 2 — performance and JAX backend (v0.3)

  • sequences.solve(...) gains a backend: str = "qutip" keyword- only parameter; backend="jax" dispatches to the Dynamiqs integrator via iontrap_dynamics.backends.jax. Solver / backend compatibility is validated (explicit solver="sesolve" or "mesolve" with backend="jax" raises ConventionError; only solver="auto" is accepted on the JAX path).
  • iontrap_dynamics.backends.jax.solve_via_jax(...) — opt-in JAX-backend entry. Dispatches to dynamiqs.sesolve for ket inputs, dynamiqs.mesolve for density matrices. Honours all three StorageMode values (OMITTED / EAGER / LAZY — the LAZY loader closes over the Dynamiqs-returned JAX array, materialises one Qobj per index on demand, bounds-checks against JAX's silent-clamp behaviour). Forces JAX x64 at solve entry for complex128 arithmetic (CONVENTIONS.md §1 unit commitment).
  • Results on the JAX backend are tagged with ResultMetadata.backend_name="jax-dynamiqs" and backend_version=f"dynamiqs-{ver}+jax-{ver}". The backend_name string is a schema-commitment tag: a future integrator swap requires a new string (not a suffix), so users' cache manifests stay consistent. convention_version is read from hilbert.system.convention_version, honouring archival pins exactly like the QuTiP path.
  • Every time-dependent Hamiltonian builder gains a backend= kwarg on the same pattern (carrier, RSB, BSB, MS gate, modulated carrier). The four structured detuning builders share backends.jax._coefficients.timeqarray_cos_sin for dq.modulated(cos, H_static) + dq.modulated(sin, H_quadrature) assembly; modulated_carrier_hamiltonian takes a user-supplied envelope_jax keyword (JAX-traceable mirror of envelope) for arbitrary pulse shapes. Missing envelope_jax on backend="jax" raises ConventionError — no silent translation.
  • Cross-backend numeric equivalence validated at library-default integrator tolerances across all five time-dependent builders: worst-case 1.35e-5 absolute expectation delta, well under the 1e-3 design-target tolerance. Honest performance null result at dim ≥ 100 / 5000 steps (QuTiP 5 is ~2.8× faster than Dynamiqs + JAX on CPU at current library scales; see docs/benchmarks.md). The JAX backend's value is positioning, cross-backend consistency checking, and forward-looking capability (autograd scaffolding ready via envelope_jax coefficients; GPU / TPU dispatch if the user installs a CUDA / Metal JAX build).

Demo tools (tools/run_*.py with canonical manifest.json + arrays.npz + demo_report.json artefacts under benchmarks/data/): run_benchmark_sideband, run_demo_carrier, run_demo_gaussian_pulse, run_demo_ms_gate, run_demo_bernoulli_readout, run_demo_binomial_readout, run_demo_poisson_readout, run_demo_detected_readout, run_demo_spin_readout, run_demo_parity_scan, run_demo_sideband_inference, run_demo_wilson_ci, run_demo_bell_entanglement, run_demo_rabi_jitter, run_demo_detuning_jitter, run_demo_rabi_drift_scan, run_demo_spam_prep.

Phase 2 benchmark tools (tools/run_benchmark_*.py with report.json + plot.png artefacts under benchmarks/data/): run_benchmark_sesolve_speedup (Dispatch X — sesolve / mesolve parity on QuTiP 5), run_benchmark_ensemble_parallel (Dispatch Y — serial / loky / threading crossover), run_benchmark_sparse_vs_dense (Dispatch OO — CSR / dense operator-dtype baseline; closes the Phase 2 sparse-matrix-tuning open item), run_benchmark_jax_timedep (Dispatch YY / β.4.5 — cross-backend QuTiP-vs-JAX at dim ≥ 100 / 5000 steps across all five time-dependent builders; needs the [jax] extras).

Test suite at v0.6.0 (base CI, no extras):

  • 1319 passed, 3 skipped. Two skipped tests are migration-tier builder-comparison slots (scenarios 3 and 4) with probe-informed blockers (see CHANGELOG.md); one skipped module is the Dynamiqs-gated integration test file (tests/unit/test_backends_jax_dynamiqs.py, gated on pytest.importorskip("dynamiqs")).
  • With the [jax] extras the suite additionally runs the Dynamiqs-gated integration tests (cross-backend numeric equivalence, result metadata, storage modes, Fock-saturation check, time-dependent builders, user-envelope dual-callable contract, and solve_ensemble on JAX); run under the extra to confirm the current total.

Docs site scaffold:

  • mkdocs.yml configures the public-facing documentation build
  • docs/index.md — welcome page
  • docs/getting-started.md — install + first run
  • docs/framework.md — high-level design rules
  • docs/conventions.md — rendered live from root CONVENTIONS.md (single source of truth via pymdownx.snippets)
  • docs/phase-1-architecture.md — concrete public-API reference (module map, per-module surface, extension points, non-goals; now also hosts the result-family vs backend-variety decision record from D5 / Dispatch NN)
  • docs/phase-2-jax-backend-design.md — deliberation note for the Phase 2 JAX / Dynamiqs backend (design axes A-D, chosen Option β, ten open questions + their recorded answers)
  • docs/phase-2-jax-time-dep-design.md — β.4 staging note for the time-dependent Hamiltonian track (Option X parallel JAX-native builders; 5-sub-dispatch plan; scope inventory)
  • docs/benchmarks.md — honest performance baselines for every Phase 2 dispatch including the β.4.5 cross-backend benchmark at dim ≥ 100 / 5000 steps
  • docs/boundary-decision-tree.md — contributor scope rules (closes D8)
  • docs/tutorials/ — task-oriented walkthroughs. Twelve tutorials shipped (Tutorials 1–12 cover the full public-surface pipeline end-to-end, from carrier Rabi + Wilson CIs through two-ion Bell-state entanglement); see docs/tutorials/index.md
  • docs/stylesheets/tokens.css — vendored from threehouse-plus-ec/cd-rules

The authoritative project documents are:

  • WORKPLAN_v0.3.md for scope, architecture, milestones, and governance
  • CONVENTIONS.md for physical, numerical, and notational rules
  • LICENCE for the repository split-licence declaration

Scope

Planned capabilities include:

  • Unitary and dissipative dynamics for coupled spin-motion systems
  • Standard ion-trap Hamiltonians: carrier, sideband, Mølmer-Sørensen, parametric modulation, and stroboscopic drives
  • Standard state preparations and observables for spins and motional modes
  • Backend-agnostic architecture: QuTiP reference backend + JAX / Dynamiqs backend (opt-in via backend="jax" — see docs/benchmarks.md for when each is the right choice)

Explicitly out of scope:

  • Trap geometry simulation
  • Molecular dynamics of ion crystals
  • Pulse-sequence compilers and hardware-control stacks
  • Electromagnetic field modelling

Development

Python 3.11+ is required.

Editable install:

python -m pip install -e ".[dev]"

Optional groups:

  • .[docs] for documentation tooling
  • .[plot] for plotting helpers used by examples and tutorials
  • .[jax] for the JAX / Dynamiqs backend (Phase 2 β.1–β.4 on main: opt in via backend="jax" on solve and on the time-dependent Hamiltonian builders)

Example:

python -m pip install -e ".[dev,docs]"

Repository Layout

  • src/iontrap_dynamics/ — Python package (core + measurement/ subpackage)
  • tests/ — unit/, conventions/, regression/{analytic,invariants,migration}/, benchmarks/
  • tools/ — maintenance scripts (asset fetch / checksum, SPDX check, pa11y config, migration-reference generator) and demo runners
  • benchmarks/data/ — canonical manifest + arrays + report artefacts per demo / benchmark
  • docs/ — mkdocs-material source for the documentation site
  • assets/ — design assets consumed from threehouse-plus-ec/cd-rules
  • legacy/ — pinned legacy qc.py used by migration-tier regression
  • WORKPLAN_v0.3.md — project workplan (v0.3.7 amendments applied: §4.0 repo-hosting, §5.0 release-mapping, §5.1 v0.2 release, §5.2 post-v0.2 on-main, §5.3 β.4 as v0.3.x follow-up, §5.4 estimation/Darwinism and §5.5 two-mode/motional open-system surfaces)
  • CONVENTIONS.md — binding conventions document (v0.3 frozen: §17 measurement and §18 systematics closed at the v0.2.0 release; §19–24 estimation/Darwinism + two-mode/motional closed at the v0.3 Convention Freeze, post-v0.4.0)
  • CHANGELOG.md — Keep-a-Changelog log of dispatches on main

Licence

The distributable Python package is MIT-licensed. The repository as a whole uses a split-licence architecture declared in LICENCE; design documents and tutorial material do not all share the same terms.

Endorsement Marker

Local candidate framework under active stewardship. No external endorsement is implied.