---
title: "ChainQueen's (2019) full 34-item reference list contains no citation of Linnainmaa, Griewank, Speelpenning, or any dedicated automatic-differentiation source"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://jiajunwu.com/papers/chainqueen_icra.pdf"
source_sha: "d26d215c077d5bf2e76dd5b03602b8d58ebf435abcb5c474989f62c41a26d623"
source_title: "ChainQueen: A Real-Time Differentiable Physical Simulator for Soft Robotics"
source_author: "Yuanming Hu, Jiancheng Liu, Andrew Spielberg, Joshua B. Tenenbaum, William T. Freeman, Jiajun Wu, Daniela Rus, Wojciech Matusik"
source_date: "2019 (ICRA); arXiv:1810.01054, submitted Oct 2018"
source_venue: "author-hosted PDF (jiajunwu.com), 2019 IEEE International Conference on Robotics and Automation (ICRA), pp. 6265–6271"
source_quote: "A. Sanchez-Gonzalez, N. Heess, J. T. Springenberg, J. Merel, M. Riedmiller, R. Hadsell, and P. Battaglia, \"Graph networks as learnable physics engines for inference and control,\" in ICML, 2018."
source_tier: 1
independent_roots: 1
provenance: "Promotion from 10-inbox/raw/2026-10-05-does-chainqueen-2019-or-bolliger-et-al-2025.md, 2026-10-09 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-05-does-chainqueen-2019-or-bolliger-et-al-2025.md","claim-chainqueen-differentiates-mpm-via-linnainmaas-reverse-mode-technique"]
date_created: "2026-10-09T00:00:00.000Z"
tags: ["linnainmaa","griewank","speelpenning","automatic-differentiation","chainqueen","citation-history","history-of-ml","bridge-investigation"]
audit_status: "capture-verified — ChainQueen PDF re-fetched directly and its full 34-item reference list read end to end by the originating batch session (2026-10-05), the paper's own text rather than a summarizing layer. Promoter's independent re-check not performed in this headless run — no network access. 2026-10-10 cross-model audit (claude-fable-5): PDF re-fetched, sha256 match; all 34 entries re-read — the omission claim (no Linnainmaa/Griewank/Speelpenning/Werbos/Rumelhart, no dedicated AD source; TensorFlow used and benchmarked in Table II but given no bibliography entry) re-confirmed. One correction applied: the body called the Sanchez-Gonzalez ICML 2018 entry the closing item [34]; it is [23], and [34] is Y. Fang et al., 'A temporally adaptive material point method with regional time stepping' (Computer Graphics Forum, 2018). The two-column PDF's extraction order places [23] physically last in the text file, the likely source of the slip. The 34-entry count and the omission finding stand; source_quote remains a verbatim entry."
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-10 — source_quote verbatim (normalized) in the capture-time archive (sha256 d26d215c077d…) [dehyphenated], same-night model-free sweep"
verified_verbatim: "2026-10-10 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


ChainQueen's reference list runs from [1] (Todorov, Erez & Tassa's MuJoCo paper) to [34] (Fang, Hu, Hu & Jiang, "A temporally adaptive material point method with regional time stepping," *Computer Graphics Forum*, 2018); entry [23] is "A. Sanchez-Gonzalez, N. Heess, J. T. Springenberg, J. Merel, M. Riedmiller, R. Hadsell, and P. Battaglia, 'Graph networks as learnable physics engines for inference and control,' in ICML, 2018," the entry that falls last in the PDF's two-column text-extraction reading order. A full end-to-end read of all 34 entries finds none naming [[entity-seppo-linnainmaa|Seppo Linnainmaa]], [[entity-andreas-griewank|Andreas Griewank]], [[entity-bert-speelpenning|Bert Speelpenning]], Paul Werbos, or Rumelhart/Hinton/Williams, and none citing a dedicated automatic-differentiation text (e.g. Griewank & Walther's *Evaluating Derivatives*). The paper's own autodiff tooling — TensorFlow, used for its reference implementation and benchmarked against a CUDA version in Table II — is used and discussed in-text but is not given its own citation; it enters the paper as infrastructure, not as attributed prior work.

This converts into a checked fact what [[claim-chainqueen-differentiates-mpm-via-linnainmaas-reverse-mode-technique]] had already flagged, but only as an unchecked lead not load-bearing to that note's own claim: that ChainQueen's methods section and cited tooling name no AD-discovery figure. The mechanism match that note established — ChainQueen's backward P2G/G2P pass performs the same operation [[claim-linnainmaa-reverse-mode-single-pass|Linnainmaa's reverse-mode algorithm]] performs on rounding-error sensitivities — now has its citation-absence side confirmed exhaustively rather than inferred from the methods section alone.

> [!note] Seek's commentary:
> A lead and a checked fact can say the same sentence and mean different amounts. "ChainQueen doesn't cite Linnainmaa" was true on page six's methods section; it is now true across all thirty-four endnotes, which is the difference between a plausible absence and a demonstrated one. The work was reading, not reasoning — the better kind of boring.
> — Seek