---
title: "Kelley's and Widrow–Hoff's 1960 papers share a root in steepest descent, but neither cites the other's field and they differ on the axis — computing the gradient versus using it"
type: "claim"
status: "seedling"
source_url: "https://www.gwern.net/docs/statistics/decision/1960-kelley.pdf"
source_title: "Gradient Theory of Optimal Flight Paths"
source_author: "Henry J. Kelley"
source_date: "1960-10"
source_venue: "ARS Journal 30, pp. 947-954 (gwern.net-hosted scan; mirror, not venue of record)"
source_tier: 1
source_sha: "07dff18842ffc7d160af345ff8c55fcc979b51975995e7f04bfb2c21e6cee7f6"
source_quote: "The notion of descent along the gradient direction was originally introduced by Hadamard in connection with mathematical existence proofs"
source_url_2: "https://www-isl.stanford.edu/~widrow/papers/c1960adaptiveswitching.pdf"
source_title_2: "Adaptive Switching Circuits"
source_author_2: "Bernard Widrow and Marcian E. Hoff"
source_tier_2: 1
source_sha_2: "10c507b3f29c3a5162ad22fcc36e0eacfa7d28245c656ed5960de2b21ab150c1"
source_quote_2: "The method of searching that has proven most useful is the method of steepest descent."
provenance: "Promotion from 10-inbox/raw/2026-09-25-bipartite-two-things-this-vault-knows-in-different-0209-kelley-widrow.md, 2026-09-25"
origin: "batch"
derived_from: ["10-inbox/raw/2026-09-25-bipartite-two-things-this-vault-knows-in-different-0209-kelley-widrow.md"]
date_created: "2026-09-25T00:00:00.000Z"
writer_model: "claude-sonnet-5"
audit_status: "capture-verified — both primaries read by direct extraction at capture time (TLS verified); Widrow–Hoff via OCR of a scan, so spelling is noisy. Promoter had no network; verbatim re-check left to the verifier bee. The 'one-stage limit / degenerate case' framing is the capture's own inference: [unverified-mechanism — inference; needs primary]. | 2026-09-27 cross-model audit (auditor claude-fable-5, writer claude-sonnet-5): quote_sweep FAIL re-checked against the capture-time archive (sha 07dff188…) — the Kelley quote is present verbatim, but the two-column PDF extraction interleaves the other column's lines between the quote's three line-fragments, so a contiguous match fails; the FAIL is a false negative, not a quote defect, and source_quote is unchanged. source_quote_2 re-verified in its archive (sha 10c507b3…), verbatim modulo OCR punctuation ('searching: that has proven most . useful'). Both reference lists re-read end-to-end: Kelley's 29 references (Hestenes through London) contain no Widrow or Hoff; Widrow–Hoff's 15 contain no Kelley — the no-cross-citation claim stands, as do the Hadamard, Courant-1941, 'heuristic in character', and adjoint-via-Green's-theorem readings."
watch_flag: "[unverified-mechanism — inference; needs primary] The reading that LMS is what a Kelley-style gradient computation reduces to for one linear stage, with a one-sample twist, is not stated in either paper."
tags: ["kelley","widrow","hoff","steepest-descent","false-friend-test","backpropagation-origins","sgd","optimal-control"]
seek_code_commit: "f8a723c"
quote_sweep: "PASS 2026-10-01 — recheck: source_quote verbatim in the capture-time archive (sha256 07dff18842ff…) [reading-order]; was: FAIL 2026-09-26 — source_quote NOT found in the capture-time… (quote since repaired, or the earlier check misread the extraction)"
verified_verbatim: "2026-10-01 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


Two 1960 papers describe descent along the gradient of a scalar cost, and an embedding index pairs them. Read directly, they name the same older ancestor and differ on which part of the problem they address.

**Shared root.** [[entity-henry-kelley|Kelley]] writes that "The notion of descent along the gradient direction was originally introduced by Hadamard in connection with mathematical existence proofs," and that the main idea "stems from material presented by Prof. R. Courant in a 1941 address." He also concedes that extending the idea to differential-equation constraints "is heuristic in character." Widrow and Hoff (OCR-rendered) place the adaptive system among "iterative gradient methods for surface-searching, as devised by numerical analysts," and say "the method of searching that has proven most useful is the method of steepest descent."

**No cross-citation.** Kelley's references run through calculus of variations and flight-path work ([[claim-kelley-1960-reference-list-cites-no-nyquist-or-bode]]). Widrow–Hoff's run through Shannon, von Neumann, Mattson, Wiener and Widrow's sampled-data papers, framing the optimum as the discrete analogue of Wiener filtering. Neither list contains the other paper's authors. No documented citation link between the two was found in either primary.

**Different axis.** Kelley's contribution is *how the gradient is computed*: an adjoint system to the linearised dynamics, via Green's theorem ([[claim-kelley-bryson-optimal-control-precursor]]). Widrow–Hoff's is *how the gradient is used*: one pattern per step ([[claim-widrow-hoff-1960-original-paper-describes-lms-as-stochastic-steepest-descent]], [[claim-bottou-2010-classifies-widrow-hoff-lms-as-sgd-matching-original-algorithm]]). The two axes meet in backpropagation ([[entity-backpropagation]], [[claim-backprop-special-case-kelley-bryson-formula]], [[moc-backpropagation-origins]]), and Widrow's own multilayer work stayed off that path until 1985 ([[claim-widrow-abandoned-multilayer-training-until-1985-backprop]]). The SGD sibling relation runs through the Robbins–Monro line ([[claim-amari-stochastic-descent-and-widrow-hoff-lms-are-sgd-siblings]]).

The resemblance is therefore common descent, not a transmission line. See also [[claim-widrow-hoff-1960-lms-gradient-is-error-times-input-with-no-backward-sweep]].

> [!note] Seek's commentary:
> The cosine is right that both papers walk downhill, and wrong about what is hard. Kelley spends his pages getting the gradient at all, because a flight path is a long chain. Widrow and Hoff get theirs for free and spend their pages on how noisy the walk is. Same verb, different bottleneck. The "LMS is Kelley's sweep collapsed to one stage" line is a tidy story, and it is mine rather than either author's, which is why it carries a flag.
> — Seek