AutoMovie API
    Preparing search index...

    Interface IAutoMovieQuadraticPolishingResult

    One polishing attempt on a solved quadratic program and its diagnostics, as polishAutoMovieQuadraticProgram returns it inside a refinement result.

    Polishing takes one of two methods, chosen by whether the native primal already satisfies feasibility and complementarity on every original row: if so, it keeps that primal and sweeps the signed multipliers toward stationarity; otherwise it runs active-set steps on exact interval groups. Every step counts against the caller's one maximumRefinements budget. primal and dual are only a candidate: refineAutoMovieQuadraticProgram remeasures every original row and alone sets accepted, and a rejected candidate never replaces the native solution. Arrays are new and owned by the caller.

    interface IAutoMovieQuadraticPolishingResult {
        accepted: boolean;
        activeRows: number[][];
        dual: number[];
        dualPasses: number;
        dualUpdates: number[];
        iterations: number;
        jointMerits: number[];
        linearResiduals: number[];
        merits: number[];
        method: "fixed-primal-dual" | "interval-joint";
        primal: number[];
        reason: string;
        stepSizes: number[];
        trialDual?: number[];
        trialPrimal?: number[];
        workingChanges: string[];
        workingRanks: number[];
    }
    Index
    accepted: boolean

    Whether the refinement's full-row admission adopted the candidate.

    activeRows: number[][]

    Original row numbers of the working equations at each working solve.

    dual: number[]

    Candidate multipliers in original row order, signed so a positive value presses on the upper endpoint and a negative one on the lower endpoint.

    dualPasses: number

    Dual sweeps started; zero for "interval-joint".

    dualUpdates: number[]

    Multipliers changed in each dual sweep.

    iterations: number

    Steps started, dual sweeps or working solves; at most the budget.

    jointMerits: number[]

    Full-row KKT merit of each working candidate measured, in order.

    linearResiduals: number[]

    Largest of the stationarity residual and the working-equation residual after each working step that produced multipliers.

    merits: number[]

    Full-row KKT merit of the native point, then after each dual sweep or of the working candidate that reached the tolerance.

    method: "fixed-primal-dual" | "interval-joint"

    "fixed-primal-dual" when the native primal was already feasible and complementary, "interval-joint" otherwise.

    primal: number[]

    Candidate primal in the program's variables: the native primal, or a working-set point whose full-row merit reached the tolerance.

    reason: string

    Why polishing stopped, as a sentence or a kebab-case refusal code.

    stepSizes: number[]

    Fraction of each taken step, 1 for a full step and less when blocked.

    trialDual?: number[]

    Multipliers paired with trialPrimal; absent likewise.

    trialPrimal?: number[]

    Last working-set trial primal; absent when no working step started.

    workingChanges: string[]

    Working-set changes in order, leave-wrong-sign:<row> or enter-blocking:<row> with the original row number.

    workingRanks: number[]

    Rank of the working basis at each working solve, including refused ones.