4D-Var's adjoint method reached meteorology by documented transmission — a seminar and borrowed optimization algorithms — not independent rediscovery, per Talagrand's own account

automatic-differentiationhistory-of-mlmultiple-discoverymeteorologydata-assimilationoptimal-controladjoint-methodpontryagin

The capture this note is drawn from initially framed 4D-Var (Le Dimet & Talagrand 1986, Tellus 38A) as a sixth, undocumented independent rediscovery of reverse-mode differentiation, alongside the five fields in Griewank's list. On Talagrand's own fuller account, this does not hold. Around 1979–80, while Talagrand was in ECMWF's Assimilation Section, G. R. Kontarev — one of two visitors from Academician G. I. Marchuk's group (the other was V. N. Lykossov) — gave an ECMWF seminar on "the Technique of Adjoint Equation," introducing the method. Back in France in 1981, Talagrand met François-Xavier Le Dimet, who "had studied optimal control with J.-L. Lions, and was interested in applying optimization methods to meteorology," and who supplied the connecting algorithm: he "gave [Talagrand] a number of documents, describing various algorithms for optimization. Among these, found one, based on adjoint equations". The same slides also record Penenko & Obraztsov (1976, Soviet variational data assimilation) as already using the adjoint approach before the 1986 paper — introduced by Talagrand as something learned of afterwards ("In the mean time, I became aware of a number of facts"): prior art bearing on non-priority, not a transmission channel into the 1986 work.

4D-Var is therefore meteorology's documented application of an already-circulating adjoint-equation technique, not a case of the method being independently re-derived from scratch inside a field with no contact to the others — the same equivalence-vs-transmission distinction the vault's optimal-control/backprop cluster already enforces elsewhere (see claim-recht-adjoints-equivalence-not-transmission for the mirror case: mathematical equivalence asserted without a transmission claim). It does not extend Griewank's independent-discovery list. What does survive from the same source, as a genuinely independent fact, is the unawareness pattern recorded separately in claim-talagrand-ad-and-optimization-had-little-contact.