Ckmeans.1d.dp: Optimal k-means Clustering in One Dimension by Dynamic Programming

Added by the independent DS19 audit (AUDITS/AUDIT-DS-PRACTICAL-CERTIFIED-SOLVER-001.md).

• Wang & Song (2011) implement the contiguous 1-D $k$-means DP in $O(kn^2)$ time and $O(kn)$ space — exactly the fixed-tilt evaluation DS19.2 describes.
• Grønlund et al. (2017, arXiv:1701.07204; pp. 1–2 read) survey the exact 1-D problem: the DP recurrence runs in $O(n\log n+kn)$, or $O(kn)$ on sorted input, by monotone-matrix search, and the weighted discrete scalar quantization form is covered. Transfers: the fixed-tilt bound is $O(KN)$ after sorting. Does not transfer: exact ties of tilted values are not treated; the audit's tie lemma (report §7.2) shows the DP value is tie-order independent.
• Megiddo (1983) parametric search and Toledo (1993; primary text read) fixed-dimension concave maximisation with polynomial comparisons. With $F=-v_K$ and the interval DP as a degree-2 evaluator, the exact tilt dual minimum is computable with polynomially many arithmetic operations for fixed $d_\lambda$ and variable $K$. Does not transfer: neither source bounds bit complexity; at $d_\lambda=1$ the audit's root-separation bisection supplies a polynomial-bit algorithm (report §7.5), and OPEN-DS-TILT-DUAL-EXACT-COMPLEXITY retains the bit question for fixed $d_\lambda\ge2$ and the whole question for variable $d_\lambda$.
• Carstensen (1983; secondary record) is the original $n^{\Omega(\log n)}$ parametric shortest-path breakpoint construction behind the Gajjar–Radhakrishnan warning: the envelope must not be materialised, which parametric search avoids. See LITERATURE/audits/AUDIT-DS-PRACTICAL-CERTIFIED-SOLVER-2-September-2026.md for the six-field table.