---
title: "Applied Optimal Control (1969) was compiled from six years of Harvard, MIT, and AIAA courses — the adjoint method's curriculum vector, by Bryson's own account"
type: "claim"
status: "budding"
audit_status: "verified-verbatim (Bryson's 1980 Citation Classic retrospective, direct fetch at capture level 2026-07-07; Dreyfus-1990 corroboration from the directly-read PDF; the '520 citations by 1980' figure is [unverified-quant — SCI tally not independently checked])"
source_url: "https://garfield.library.upenn.edu/classics1980/A1980JE96000001.pdf"
source_author: "Arthur E. Bryson, Jr. ('This Week's Citation Classic,' Current Contents #8, 1980-02-25)"
source_date: "1980-02-25T00:00:00.000Z"
source_tier: 1
source_quote: "As a result of the fine reception given to these courses, we were encouraged to extend our class notes and put them into book form."
provenance: "Promotion from 10-inbox/raw/20260707-0208-did-bryson-hos-applied.md, 2026-07-07, queen cycle 21"
origin: "batch"
derived_from: "10-inbox/raw/20260707-0208-did-bryson-hos-applied.md"
date_created: "2026-07-07T00:00:00.000Z"
tags: ["bryson-ho","optimal-control","adjoint-method","curriculum","history-of-ml","backpropagation"]
drafted_in: ["the-wall-they-agreed-on"]
---


Bryson's first-person 1980 retrospective documents the teaching-to-textbook
pipeline directly: a 1963 two-week Harvard summer course ("attended by 100
engineers, most of whom were from industry"), a Harvard graduate course from
1963 also given at MIT in 1966, and a repeated 1967 AIAA continuing-education
course afterward distributed "in the form of audio-cassettes and a
slide-book" — then the book, extended from those class notes. Cited "over
520 times since 1969" as of 1980 [unverified-quant]. That the book's content
carries the goods is confirmed from the Dreyfus 1990 primary (read
directly): "Bryson and Ho explicitly gave in their 1969 book gradient
formulas for exactly the multistage, free-terminal-state problem."

**An unbundling worth keeping** (the capture's real find): the harvested
topic assumed the *argmin* blog makes this curriculum argument. It doesn't.
Recht's "Mates of Costate" (2016) cites the book only as the accessible
written record of the Lagrangian/adjoint derivation, standing in for
Bryson's unobtainable 1961 symposium paper — a derivational-lineage claim,
no pedagogy framing (and he dates the book 1968; standard record and
Dreyfus say 1969 — discrepancy preserved, not corrected). The
curriculum-vector story is real but it is Bryson's own, not argmin's. A
topic question's pre-bundled attribution is itself a claim to check.

That this gradient paradigm traveled richly *within* control while remaining
off AI's map is documented from the AI side: even as the field diagnosed its
own failure as combinatorial explosion
([[claim-lighthill-1973-blamed-combinatorial-explosion]]), McCarthy's 1973
rebuttal to Lighthill dismissed control theory as having "little relevance to
AI" ([[claim-mccarthy-1973-control-theory-little-relevance-to-ai]]).

So the optimal-control thread of [[moc-backpropagation-origins]] gains its
transmission mechanism: Kelley/Bryson's gradient procedure
([[claim-kelley-bryson-optimal-control-precursor]]) reached a generation of
engineers through courses and this textbook — while the neural-network
community rediscovered the mathematics separately
([[claim-rhw-1986-reference-list-four-works]]). Hecht-Nielsen's 1990
*Neurocomputing* is the attested first to flag the book as a [[entity-backpropagation|backprop]]
presage in the NN literature (Tier 4 pointer, not load-bearing).

> [!note] Seek's commentary:
> This is the positive-control case for the disciplinary-wall thesis, and the more damning one. Linnainmaa didn't transmit because he was obscure — Finnish, wrong journal, wrong framing. Bryson was the opposite of obscure: the adjoint gradient method reached a whole generation of engineers through a 1963 Harvard summer course of 100 people, an MIT graduate course, a repeated AIAA continuing-ed course pressed onto audio-cassettes and a slide-book, and a textbook cited 500+ times. Transmission *succeeded*, richly, through every pedagogical channel the 1960s had — and it still didn't cross into the neural-network community, which re-derived the same mathematics from scratch two decades later. So the wall between fields isn't a filter that only stops weak signals; it stops a firehose. An idea can be taught to hundreds, printed in a classic, and remain perfectly invisible one discipline over. That's how load-bearing the boundary is — and why [[claim-linnainmaa-field-numerical-analysis|disciplinary contact, not citation]] is the real unit of transmission.
> — Seek