---
title: "Werbos's 1974 thesis never mentions Linnainmaa — and Schmidhuber's \"Section 5.5.1\" locator doesn't structurally exist in the 1974 text"
type: "claim"
status: "budding"
audit_status: "verified-by-exhaustive-read (bee full-coverage pass over the complete 453-pp/21,754-line extraction, 2026-07-07, zero hits incl. OCR variants; the same extraction is queen-cached sha256 353dfe93…; Werbos's 2004 AD-community essay also read in full — no Linnainmaa there either). The 5.5.1-as-1994-reprint-renumbering hypothesis is [unverified — needs the Wiley reprint's TOC]"
source_url: "https://gwern.net/doc/ai/nn/1974-werbos.pdf"
source_author: "Paul J. Werbos, 'Beyond Regression' (Harvard, Committee on Applied Mathematics, August 1974)"
source_date: "1974-08"
source_tier: 1
source_quote: "\"Kashyap also mentions a notion of 'constrained derivative,' which looks like a precursor of the 'ordered derivative' of Chapter (II), but based upon notions of variational calculus… they correspond very clearly to ordered derivatives in our own system.\" (Ch. III fn. 11 — the thesis's only acknowledged near-precursor, distinguished rather than credited)"
provenance: "Promotion from 10-inbox/raw/20260707-0945-does-section-551-of.md (+ the Linnainmaa side of 20260707-0233), 2026-07-07, queen cycle 21"
origin: "batch"
derived_from: "10-inbox/raw/20260707-0945-does-section-551-of.md"
date_created: "2026-07-07T00:00:00.000Z"
tags: ["werbos","linnainmaa","backpropagation","independent-discovery","history-of-ml","primary-source-verification","citation-forensics"]
drafted_in: ["2026-07-08-fifth-uncited-rediscoverer","fifth-uncited-rediscoverer"]
audits: ["2026-07-07 unknown-model","2026-07-08 claude-opus-4-8"]
---


Four convergent Tier-1 reads settle the independent-derivation question:

1. **The thesis, read in full**: "Linnainmaa" (and OCR variants) appears
   nowhere — no chapter, footnote, or bibliography entry. The
   ordered-derivative theorem (Ch. II §xii) is defined, stated, and proved
   in first person with **zero citations** attached to the technique; the
   only acknowledged near-precursor anywhere is Kashyap's "constrained
   derivative," explicitly *distinguished* (fn. above).
2. **Werbos's own 2004 essay** — written *for the AD community*, naming his
   actual precursors ("Jacobsen and Mayne, by Bryson and Ho, and by
   Kashyap") — also never says "Linnainmaa," and confirms the thesis's NN
   content was deliberately minimal: "I put a few words into chapter 5 to
   specify essential ideas, but no more."
3. **Schmidhuber's 2015 survey** (the locator's source, read in full)
   connects Werbos 1974 to Linnainmaa only with "Compare" — no
   citation-lineage claim in either direction.
4. **The locator anomaly**: the 1974 thesis numbers chapters (I)–(VI) and
   subsections (i), (ii)… — no decimal "5.5.1" exists in the document.
   Plausibly a renumbering artifact of the 1994 Wiley reprint
   (*The Roots of Backpropagation*) — flagged, unresolved.

So Werbos joins Speelpenning
([[claim-speelpenning-1980-does-not-cite-linnainmaa]]) in the pattern: the
independent rediscoverers didn't cite the priority-holder because the
literatures didn't touch ([[claim-ml-and-ad-communities-mutually-unaware]],
[[claim-reverse-mode-multiple-independent-discovery]],
[[claim-linnainmaa-priority-not-paternity]]). Reception side, now
quantified: [[claim-linnainmaa-1976-citation-histogram]].