optimization optable → tableimport fullseye as fs; fs.ledger.optimize_lens(system, variables=None, fields=None, wavelengths=None, rings=4, efl_target=None, efl_weight=None, field_weights=None, iterations=30, damping=0.001, tolerance=1e-07, min_thickness=0.5, max_thickness=None, min_radius=1.0, pupil_fill=0.98) (実装を直接呼ぶなら import lensopt; lensopt.optimize_lens(system, variables=None, fields=None, wavelengths=None, rings=4, efl_target=None, efl_weight=None, field_weights=None, iterations=30, damping=0.001, tolerance=1e-07, min_thickness=0.5, max_thickness=None, min_radius=1.0, pupil_fill=0.98)、台帳から引くなら opsoptics.get("optimize_lens"))Damped-least-squares (Levenberg–Marquardt) optimisation of a prescription (table).
variables: surface parameters to move — strings "R<i>"/"c<i>"
(curvature; a radius may pass through flat), "t<i>" (thickness),
"k<i>" (conic), "A4_<i>", "A6_<i>" … (even aspheric
coefficients). Default: every finite radius. efl_target: hold the
effective focal length (default: the starting EFL, so a design does not
“improve” by getting longer); pass 0 / False to leave it free.
Fields / wavelengths / rings / weights as in :func:merit_function.
Each iteration builds the Jacobian by forward differences, solves
(JᵀJ + λ diag(JᵀJ)) δ = −Jᵀr and accepts the step only if the merit
falls (then λ /= 3; otherwise λ ×= 4 and retried, up to 6 times); a step
that yields an invalid prescription counts as a failure. Stops when the
relative merit change is below tolerance twice in a row (converged,
status="converged"), when two iterations in a row accept no step or λ
blows past 1e8 (status="stalled", converged=False), or after
iterations (status="iterations"). Thickness is clamped to
[min_thickness, max_thickness] and |R| >= min_radius — the start
included, so the returned system always obeys the bounds.
Returns {"system": optimised prescription, "merit_initial",
"merit_final", "rms_initial", "rms_final", "efl_initial", "efl_final",
"history": [merit per accepted iteration], "iterations", "converged",
"variables": [{"name", "surface", "initial", "final"}], "rays_lost"}.
optics の全 op は入力を検証してから計算する(黙って通さない):
_mm / _um / _deg / _mrad。mm と µm の取り違えは crash ではなく「もっともらしく間違った答え」なので、名前で防ぐ。大きさから単位を推測する処理は一切しない。ValueError — float('50') は成功してしまうため、未パースの設定値が長さとして通り抜ける(実測: thin_lens('50', '200') がもっともらしい 66.667 mm を返していた)。bool も True == 1 の暗黙昇格として拒否。ValueError(実数枠のみ。虚部の無言切り捨て・マスク剥がしを拒否)。NaN/Inf は全入力で ValueError。depth_of_field の過焦点距離以遠の far_mm = inf(それが過焦点距離の定義)と gaussian_beam のウエストでの wavefront_radius_mm = inf(平面波面の曲率半径)。どちらも有限の相棒(far_is_infinite / curvature_per_mm)を併せて返す。それ以外の無言 NaN/Inf は内部で検出して ValueError —「float64 が溢れた」と「答えが無限大」は別の主張なので、後者の顔で前者を返さない。optics.MAX_GRID(4096)、供給された場/PSF/開口は optics.MAX_FIELD_ELEMENTS(2^24)、ABCD 素子列は optics.MAX_SYSTEM_ELEMENTS(1024)、Zernike は MAX_ZERNIKE_TERMS(512)/ MAX_ZERNIKE_ORDER(40)/ MAX_ZERNIKE_BASIS(2^25)。小さな引数から巨大な内部確保が起きる経路(実測: n_max=40 × 4096² で 108 GB)を fail-closed で塞ぐ。| 物理的に不可能な状態も拒否: 偏光度 > 1 の Stokes ベクトル、負の透過率、負の強度、n- | m | が奇数などの不正な Zernike 添字。 |
py -3.11 examples/lens_optimize_demo.pytable を入力に取れる)abcd_matrix · wavefront_stats · paraxial_trace · seidel_coefficients · spot_stats · tolerance_analysis · wavefront_from_opd · spot_diagram
optimization)Provenance: lensopt.py — OPTICS operator registry. この per-op ノートは tools/opdocs.py md が自動生成(手編集しない)。
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.