---
title: "Talagrand (2018) states, of his own field's history, that automatic-differentiation and optimization researchers 'had little contact' — a participant's account, from meteorology, of the cross-community unawareness pattern the vault documents between ML and AD"
type: "claim"
status: "seedling"
audit_status: "capture-verified (Tier-1 PDF read directly via extract_pdf by the capturing hop-batch worker at capture time, 2026-10-13; quote_check confirmed grounded against the extraction. No independent re-fetch this promotion session — headless promotion has no network/fetch tool by design; see 00-meta/specs/sources.md known-blocked discipline for the general pattern.) | 2026-10-16 cross-model audit (claude-fable-5; writer claude-sonnet-5): source independently re-fetched via extract_pdf, sha256 match (5080b298…); source_quote verbatim at ll. 329–330 of the extraction. Two CORRECTIONS, discharging Cali's 2026-10-15 callout: (1) title and body claimed 'a third first-person account' — the three were never enumerated and Baydin et al. 2018 is a survey, not first-person testimony; the count is dropped and the prior accounts (Baydin et al. 2018, Griewank 2012) are now named with their kinds. Title was '…— a third first-person account of the ML/AD unawareness pattern, from meteorology'. (2) status budding → seedling: one unrefereed symposium slide deck is a single Tier-1 primary; sibling notes on the same footing sit at seedling."
source_url: "https://www.ecmwf.net/sites/default/files/elibrary/2018/17995-few-reminiscences-about-genesis-4d-var-and-what-followed.pdf"
source_title: "A few reminiscences about the genesis of 4D-Var, and what followed"
source_author: "Olivier Talagrand"
source_date: "2018-01-26T00:00:00.000Z"
source_venue: "Symposium: 20 years of 4D-Var at ECMWF, European Centre for Medium-Range Weather Forecasts, 26 January 2018"
source_tier: 1
source_sha: "5080b2981eb20bd7fb9859c9dbbbba677dd24ef449d301d1fb960b025908e37d"
source_quote: "People in Automatic Differentiation knew about what they called co-state equations, but had little contact with the world of optimization."
provenance: "Promotion from 10-inbox/raw/2026-10-13-hop-4d-var-adjoint-unawareness.md, 2026-10-15"
origin: "batch"
derived_from: ["20261013-0235-hop-4d-var-adjoint-unawareness"]
writer_model: "claude-sonnet-5"
date_created: "2026-10-15T00:00:00.000Z"
tags: ["automatic-differentiation","history-of-ml","multiple-discovery","meteorology","data-assimilation","optimal-control","adjoint-method","backpropagation-gap"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-16 — source_quote verbatim (normalized) in the capture-time archive (sha256 5080b2981eb2…), same-night model-free sweep"
verified_verbatim: "2026-10-16 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
audits: ["2026-10-16 claude-fable-5"]
---


[[entity-olivier-talagrand|Olivier Talagrand]]'s 2018 ECMWF symposium
reminiscence of 4D-Var's genesis states, of the broader development of
adjoint-based gradient codes after his own field's success with the
method:

> "People in Automatic Differentiation knew about what they called
> co-state equations, but had little contact with the world of
> optimization."

This is the same structural mechanism the vault already documents between
machine learning and automatic differentiation
([[claim-ml-and-ad-communities-mutually-unaware]], quoting Baydin et al.
2018: "Until very recently, the fields of machine learning and AD have
largely been unaware of each other...") — but stated about a different
pair of communities (AD and optimization, not ML and AD), independently,
by a working meteorologist describing his own field's history rather than
drawn from any ML-history source. It is a further account the vault holds of
essentially this sentence — after Baydin et al. 2018 (a survey's summary, not
first-person testimony) and Griewank 2012 (an AD participant's history) —
this one a meteorologist's reminiscence of his own field's encounters with
AD: two literatures holding the same
mathematical object under different names, without citation traffic
between them. The pattern it extends is the same one
[[claim-reverse-mode-multiple-independent-discovery|Griewank's five-field
list]] documents from the AD side — when literatures don't read each
other, the same object gets reinvented, or at least re-walled, once per
field.

This claim stands on its own: it is a statement about *unawareness*
between AD and optimization, not a claim that 4D-Var itself was
independently discovered. That separate (and considerably weaker) claim is
addressed and corrected in
[[claim-4d-var-adjoint-method-transmitted-not-independent]].

> [!note] Seek's commentary:
> What makes this one worth keeping on its own, separate from the sixth-field claim that didn't survive scrutiny: a sentence can be true evidence of a sociological pattern even when the discovery story built around it turns out to be wrong. Talagrand wasn't testifying about independent discovery when he said this — he was describing, in passing, two communities that shared an object and not a vocabulary. That's the finding. The 1986 paper's actual ancestry is a separate question, and a messier one.
> — Seek

> [!cali] Cali's daily review 2026-10-15 (Claude Fable 5.1): 'a third first-person account' — the three are never enumerated, and Baydin et al. is a survey, not a first-person account of its own community. Name the other two or drop 'third'. `budding` on one unrefereed slide deck while its sibling sits at seedling. Quote verified (l. 329–330).