Capture: Speelpenning's 1980 thesis bibliography (leaf 74) does not cite Linnainmaa — full-text PDF retrieved from OSTI.gov

Bottom line up front: A free, full, text-extractable PDF of Speelpenning's thesis was located on OSTI.gov (a DOE open repository) and read in full via OCR. Leaf 74 — the thesis's REFERENCES page — contains eleven citations. None of them is Linnainmaa. This closes the specific sub-question the topic asked about: Speelpenning's 1980 thesis is not a case of "first to cite Linnainmaa," because it does not cite Linnainmaa at all, anywhere in the document (checked against both the References page and the Table of Contents' "Previous Work" chapter, which names only Warner 1975, Kedem 1977, and Joss 1976 as prior art). This directly supersedes the open gap left by an earlier capture (2026-07-01-speelpenning-1980-first-autodiff-cite-linnainmaa-high), which could not read leaf 74 because the only copy then accessible (Internet Archive) was an unreadable image scan.

Claim 1: A free full-text PDF of the Speelpenning 1980 thesis exists on OSTI.gov and is text-extractable via OCR

Claim type: bibliographic / historical (uncontested). Sourcing floor: Tier 3–4 acceptable; Tier 1 achieved (the document is itself the primary source).

The thesis is catalogued on OSTI.gov, the U.S. Department of Energy's open-access technical report repository, under OSTI ID 5254402. The bibliographic record page states:

"Title: Compiling fast partial derivatives of functions given by algorithms" / "Author: Speelpenning, B." / "Date: January 1, 1980" / Report Numbers: "COO-2383-0063; UILU-ENG-80-1702; UIUCDCS-R-80-1002" / "Publisher/Research Organization: Illinois University, Urbana, Department of Computer Science" / "DOI: https://doi.org/10.2172/5254402"

The record page links to a "View Technical Report" PDF at /servlets/purl/5254402. That PDF was fetched and OCR'd in this session (82 pages, sha256 409d6a7ca55517551db63ef552cdbbcb31fd77eb5d83a91665bdb1faa51757f4). The scanned document itself opens with the report number and DOE boilerplate:

"UIUCDCS-R-80-1002 ... COMPILING FAST PARTIAL DERIVATIVES OF FUNCTIONS GIVEN BY ALGORITHMS / by / Bert Speelpenning / January 1980 / DEPARTMENT OF COMPUTER SCIENCE / UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN ... Research ... supported by ... U.S. Department of Energy ... EY-76-S-02-2383 and submitted in partial fulfillment of the requirements of the Graduate College for the degree of Doctor of Philosophy."

This confirms the document retrieved is the actual thesis (not a summary or abstract), funded by DOE, held by the University of Illinois Department of Computer Science, and freely downloadable without paywall.

Claim 2: Leaf 74 of the thesis is the REFERENCES section, and it contains eleven citations — Linnainmaa is not among them

Claim type: technical-mechanism / bibliographic-historical (this is the load-bearing claim the topic asked about). Sourcing floor: Tier 1–2 required (specific claim about what a specific document contains). Source tier achieved: Tier 1 — direct read of the primary document itself.

The OCR'd text places page number "74" immediately before the heading "REFERENCES," which then lists the following works verbatim (OCR quality is generally clean; a few characters are garbled by the scan, noted in brackets where relevant):

"74 / REFERENCES / Bellman, Richard E., "Dynamic Programming", Princeton Univ. Press, 1957. / Comer, D., "MOUSE4: An improved implementation of the RATFOR preprocessor", Software-Practice and Experience, Vol. 8, 1978. / Dijkstra, Edsger W., "A Discipline of Programming", Prentice-Hall, 1976. / Johnson, S.C., "YACC - Yet Another Compiler-Compiler", C.S. Tech. Report 32, Bell Laboratories, July 1975. / Joss, Johan, "Algorithmisches Differenzieren", Ph.D. Thesis, ETH, Zurich, Switzerland, 1976. / Kam, J.B., Ullman, J., "Global Data Flow Analysis and Iterative Algorithms", JACM, Vol. 23, No. 1, Jan. 1976. / Kedem, Gershon, "Automatic Differentiation of Computer Programs", Proc. 1977 Army Numerical Analysis and Computer Conf., Madison, Wisc. 1977. / Kuck, David J., "The Structure of Computers and Computations", Vol. 1, John Wiley & Sons, 1978. / Standish, T., Harriman, D., Kibler, D. and Neighbors, J., "The Irvine Program Transformation Catalogue", C.S. Dept., U.C. Irvine, Irvine, Cal., Jan. 1976. / Strassen, Volker, "Gaussian Elimination is not optimal", Numer. Math. 13, 1969. / Warner, D.D., "A Partial Derivative Generator", C.S. Tech. Report 28, Bell Laboratories, April 1975."

This is the complete reference list — the page immediately following (numbered 75) is the "VITA" (biographical note), confirming REFERENCES is the thesis's final content section and leaf 74 is exhaustive, not a partial page. No entry by Linnainmaa (neither the 1970 University of Helsinki master's thesis nor the 1976 BIT paper) appears.

This is corroborated structurally by the thesis's own Table of Contents, which lists Chapter 2 as "SYMBOLIC DIFFERENTIATION OF ALGORITHMS: PREVIOUS WORK" with named subsections for "Warner, 1975" and "Kedem, 1977" (plus discussion of Joss's 1976 ETH thesis) as the entirety of the prior-work review — again, no section or subsection named for Linnainmaa:

"2. SYMBOLIC DIFFERENTIATION OF ALGORITHMS: PREVIOUS WORK ... 2.1. Warner, 1975 ... 2.2. [Joss/ETH material] ... Kedem, 1977 ... 2.3. Extensions to Gradients and Jacobians ... 2.5. Comparison of Joss with Numerical Differencing"

(Section numbering in this ToC line is partially OCR-garbled, but the named prior-work authors — Warner, Kedem, Joss — are unambiguous and match the References list exactly.)

This directly answers the topic's core question: leaf 74 closes the first-to-cite-Linnainmaa question by showing Speelpenning (1980) is not a citer of Linnainmaa at all — he cannot be "first to cite Linnainmaa" because he does not cite him. His stated intellectual lineage (per his own thesis) runs through Warner (1975, Bell Labs), Kedem (1977), and Joss (1976, ETH Zürich) — the U.S./compiler-optimization line of automatic-differentiation work — not through Linnainmaa's Finnish numerical-analysis/rounding-error line.

Claim 3: Griewank's own 2012 historical account corroborates that Linnainmaa's work had "little resonance" until Griewank contacted him directly in the late 1980s — consistent with Speelpenning not having encountered or cited it in 1980

Claim type: historical (contextual corroboration, not itself the load-bearing citation-count claim). Sourcing floor: Tier 1–2 required for a specific historical/technical-mechanism claim about reception. Source tier achieved: Tier 1 — Andreas Griewank, the field's own historian of automatic differentiation, writing in his own peer-reviewed account, published in Documenta Mathematica.

Griewank, Andreas. "Who Invented the Reverse Mode of Differentiation?" Documenta Mathematica, Extra Volume ISMP (2012), 389–400. Retrieved as a free full PDF from emis.icm.edu.pl (European Mathematical Information Service mirror).

On Linnainmaa's reception:

"Moreover, he did not market his approach as a method for cheaply evaluating gradients either, so there was little resonance until I called him up from Argonne in the late eighties. In fact, only in 1976 he published some of the results from his thesis in English."

On Speelpenning's own route to the reverse mode, described by Griewank without any mention of Linnainmaa as an influence:

"Bernt Speelpenning (Spe80) arrived at the reverse mode via compiler optimization when Bill Gear asked him to automatically generate efficient codes for Jacobians of stiff ODEs."

Griewank's own prologue frames the wider picture: multiple independent "incarnations" of the reverse-mode idea arose in different fields (Ostrowski in chemical engineering c.1965, Hachtel et al. in circuit design c.1971, Linnainmaa in numerical analysis 1970, Werbos in neural networks 1982, Speelpenning in compiler optimization 1980, Griewank himself independently in 1987) — consistent with Speelpenning's bibliography showing no awareness of Linnainmaa's specific prior work.

Related notes: claim-linnainmaa-priority-not-paternity, backpropagation-gap

Claim 4 (quantitative, recorded for completeness): Speelpenning's own headline performance result for "Jake"

Claim type: quantitative. Sourcing floor: Tier 1–2 required. Source tier achieved: Tier 1 — direct quote from the thesis's own abstract, as printed on the DOE "Bibliographic Data Sheet" page of the same PDF.

"The system 'Jake' described in this thesis produces gradients significantly faster than numerical differencing. Jake can handle algorithms Af with arbitrary flow of control. Measurements performed on one particular machine suggest that Jake is faster than numerical differencing for n > 8. Somewhat weaker results have been obtained for the problem of computing Jacobians of arbitrary shape."

This is tangential to the Linnainmaa question but preserved because it is the thesis's own stated headline result and was directly visible in the same source; no separate sourcing effort was spent on it beyond what retrieving the PDF already provided.

What was NOT established this run

• Whether any other pre-1980 or 1980-contemporary automatic-differentiation paper did cite Linnainmaa first was not investigated in this run. The topic's question was scoped specifically to Speelpenning's thesis and leaf 74; ruling Speelpenning out is a negative result, not a positive identification of who (if anyone) actually was first. Kedem (1977) — cited by Speelpenning himself — is a candidate worth checking in a future session, since it postdates Linnainmaa's 1976 BIT paper.
• The specific machine/instrumentation details behind the "n > 8" performance claim (what machine, what workload) were not chased further; the quote above is the full extent of what the abstract states.
• OCR fidelity was not independently cross-checked against a second scan. The single OSTI.gov copy was treated as authoritative. Given that (a) the page-number continuity (73→74→75), (b) the chapter structure in the Table of Contents, and (c) the VITA section immediately following all corroborate the same reading independently, the risk of an OCR-dropped Linnainmaa entry is assessed as low, but this is not a formal proof of completeness.

Central question: [RESOLVED]

Does Speelpenning's 1980 thesis (leaf 74, the bibliography) cite Linnainmaa, and does this settle the first-to-cite-Linnainmaa question? Yes, resolved, negatively: leaf 74 contains eleven references, none of them Linnainmaa's 1970 thesis or 1976 BIT paper. Speelpenning's thesis is therefore not a candidate for "first to cite Linnainmaa" — it does not cite him at all. This is corroborated by Griewank's own 2012 historical account, which independently describes Linnainmaa's work as having "little resonance" until Griewank made direct contact in the late 1980s, and which describes Speelpenning's route to the reverse mode as arising from compiler optimization for Gear's stiff-ODE work, with no mention of Linnainmaa as an influence.