/** Options accepted by {@link svds}. */
export interface SvdsOptions {
    /** Convergence tolerance forwarded to the Lanczos eigensolver (default 1e-10). */
    tol?: number;
    /** Maximum Lanczos steps (default `min(max(2k + 20, 20), dim)`). */
    maxIter?: number;
}
/**
 * Result of {@link svds}. Singular values are returned **descending**
 * (`s[0]` largest — matching this library's full {@link svd}); note this is the
 * opposite of `scipy.sparse.linalg.svds`, which returns them ascending.
 * Singular vectors are stored as **columns**: `U[i][j]` is the `i`-th component
 * of the `j`-th left singular vector (for `s[j]`), and likewise `V` for the
 * right singular vectors.
 */
export interface SvdsResult {
    /** Left singular vectors, `m × k`, as columns. */
    U: number[][];
    /** The `k` largest singular values, descending. */
    s: number[];
    /** Right singular vectors, `n × k`, as columns. */
    V: number[][];
}
/**
 * The `k` largest singular triplets of `A` (dense `m × n`) via Lanczos on the
 * smaller normal operator.
 *
 * @example
 * svds([[1,2,0],[0,3,1],[4,0,2]], 2) // => { U, s: [σ₁, σ₂] (descending), V }
 */
export declare function svds(A: number[][], k?: number, opts?: SvdsOptions): SvdsResult;
//# sourceMappingURL=svds.d.ts.map