---
title: "M.V. Simkin"
type: "entity"
entity_kind: "person"
status: "hub"
canonical_name: "M.V. Simkin"
aliases: ["Mikhail V. Simkin","Simkin","Simkin & Roychowdhury"]
first_seen: "2026-09-26T00:00:00.000Z"
writer_model: "claude-opus-4-8"
connects_to: ["V.P. Roychowdhury","citation copying","bibliometrics","preferential attachment","Matthew effect"]
seek_code_commit: "f8a723c"
---


M.V. Simkin (UCLA, Dept. of Electrical Engineering, with co-author V.P. Roychowdhury) is a physicist of citation behaviour whose distinctive contribution the vault relies on twice: he turned copying — the fact that most citations are lifted from other papers' reference lists rather than earned by reading — from a suspicion into a measurable and then a generative object. That single idea is why he recurs here, across a paper-level method and a graph-level model.

His two load-bearing results in this vault:
- [[claim-simkin-roychowdhury-2002-identical-citation-misprints-are-copy-fingerprints-implying-about-20-percent-of-citers-read-the-original]] — repeated identical misprints as copy-fingerprints, implying only ~20% of citers read the original ("Read before you cite!", arXiv cond-mat/0212043).
- [[claim-simkin-roychowdhury-2005-random-copying-model-reproduces-preferential-attachment-and-sleeping-beauties]] — a random-copying model that reproduces preferential attachment and sleeping beauties ("A mathematical theory of citing", arXiv physics/0504094).

His copying premise also underwrites the vault's synthesis reading that a field's citation silence can be inherited rather than chosen: [[observation-copying-model-predicts-cross-field-citation-silence]].

Both papers come from the same group: the 2002 one is an unrefereed author preprint and counts under the vault's single-source concentration cap, while the 2005 one was refereed and published (*J. Am. Soc. Inf. Sci. Tech.* 58(11):1661–1673, 2007), so the cap does not reach it. His figures (the ~20% readership estimate; the "majority copied" premise) are model-dependent estimates on physics-citation data, not independently corroborated in the vault.

## References
- [[claim-simkin-roychowdhury-2002-identical-citation-misprints-are-copy-fingerprints-implying-about-20-percent-of-citers-read-the-original]] · [[claim-simkin-roychowdhury-2005-random-copying-model-reproduces-preferential-attachment-and-sleeping-beauties]]
- [[observation-copying-model-predicts-cross-field-citation-silence]]
- Related: [[entity-matthew-effect]] · [[claim-xu-2026-ghostcite-finds-citation-error-propagation-between-published-papers]] · [[claim-ansari-2026-contamination-inheritance-citation-error-propagates-across-models]]
- Captures: 2026-09-24-bipartite-two-things-this-vault-knows-in-different-kwakernaak-ansari, 2026-09-24-hop-copied-citations-make-zero-lineage-a-diffusion-prediction

## Updates
- 2026-10-18: A different Simkin & Roychowdhury paper than the one that first cited Kimura and Crow — "Re-inventing Willis" (arXiv physics/0601192, 2006, unrefereed) — independently restates and re-solves the Kimura–Crow 1964 allele-distribution problem from scratch, and checks the result against Sewall Wright's 1931 formula. This is the first vault note resting on physics/0601192 and the strongest evidence yet that his 2005 citation of the 1964 result reflects genuine engagement with its mathematics, not a loose name-drop. ([[claim-simkin-roychowdhury-2006-re-inventing-willis-independently-rederives-kimura-crow-1964-checks-against-wright-1931]])
- 2026-10-19 (propagation-repair): The 2026-10-18 entry above was corrected, following the 2026-10-19 scheduled cross-model audit of that claim-note. Was: "Re-inventing Willis" dated to 2006 and graded an unrefereed preprint, with the single-source concentration cap therefore in view. Now: the text carrying the re-derivation is the arXiv v3 of 17 Feb 2011, published as *Physics Reports* 502 (2011) 1–35 (doi:10.1016/j.physrep.2010.12.004) — a refereed review, which the cap does not reach, exactly as corrected on 2026-10-11 for the 2005 preprint / 2007 JASIST paper. The 25 Jan 2006 v1 is a 12-page text with no "Distribution of genes" section and no Kimura-and-Crow reference at all, so the re-derivation cannot be dated to 2006. Also: his paper's prose says Kimura and Crow built on "the 1930 work of Wright" while its reference [72] is Wright's 1931 *Genetics* paper. The re-derivation itself, and what it shows about his engagement with the 1964 mathematics, are unaffected.
- 2026-10-04: A bridge comparison against [[entity-sina-alemohammad|Alemohammad et al.]]'s 2026 recursive citation-dilution benchmark names his 2005 model's compounding, history-dependent feedback as the structural opposite of a mechanism that reaches the same concentration curve without one — the first time the vault has set his model's "copying needs history" premise explicitly against a mechanism that needs none. ([[observation-alemohammad-2026-simkin-2005-cosine-084-pairing-is-analytical-contrast]])
- 2026-10-10: His 2005 paper names its own ancestor — it credits [[entity-motoo-kimura|Kimura]] and Crow's 1964 neutral-allele model, jointly with Ewens (1964), as the prior derivation (by a different method) of its citation-count formula, locating the copying model in [[entity-neutral-drift-model|neutral-drift theory]] — and in the same four-sentence passage traces that theory forward into culture itself, naming Cavalli-Sforza & Feldman (1981) and then baby names (Hahn & Bentley, 2003) and dog breeds (Herzog, Bentley & Hahn, 2004). ([[claim-simkin-roychowdhury-credit-kimura-crow-1964-neutral-allele-model-as-prior-derivation]], [[observation-kimura-crow-1964-is-shared-named-ancestor-of-citation-copying-and-cultural-drift-models]])
- 2026-10-11 (propagation-repair): The 2026-10-10 entry above and the body's source-tier paragraph were corrected in place, following the 2026-10-11 cross-model audit of [[claim-simkin-roychowdhury-credit-kimura-crow-1964-neutral-allele-model-as-prior-derivation]]. Was: the 2005 paper credits Kimura and Crow alone as the prior derivation, and both of his papers here are unrefereed author preprints counting as one source under the single-source concentration cap. Now: the credit is joint — "by Kimura and Crow (1964) and by Ewens (1964)" — and the same passage names its own cultural-evolution descendants, so the genetics→baby-names leg is printed in his paper rather than assembled around it; and the 2005 paper is the refereed JASIST text (58(11):1661–1673, 2007; arxiv.org/pdf/physics/0504094 serves the post-refereeing v3), which the cap does not reach. The copying premise and the ~20% readership figure are unaffected.
- 2026-10-16: A targeted search for whether his 2005 copying mechanism specifically — not citation-count inflation generally — explains the training-corpus skew behind GPT-4's citation-popularity bias found no source that tests or even proposes the splice. His 2005 paper predates LLM citation behavior and makes no claim about machine training; Algaba et al.'s GPT-4 paper proposes training-data frequency as a bare fact about the corpus without engaging his or any copying-based account of why that frequency distribution looks the way it does. Routed as an open question rather than forced closed. ([[question-does-citation-copying-explain-training-corpus-skew-behind-llm-citation-bias]])
- 2026-10-18: A second, independent search session one day later reconfirms the 2026-10-16 finding rather than extending it: his 2005 paper is explicitly a model of human citer behavior — it predates LLM-based citation generation by roughly fifteen years and makes no claim about machine training corpora — and two more papers newly surveyed in this sub-literature ([[claim-niimi-2025-citation-count-predicts-llm-bibliographic-memorization-threshold|Niimi 2025]], [[claim-liu-chen-liu-lu-murray-2026-llms-overcite-popular-older-papers-affirms-algaba|Liu et al. 2026]]) still neither cite nor test his copying mechanism. The gap is corroborated, not new; still open.
- 2026-10-29: His 2002 copy-fingerprint mechanism was checked, for the first time, against a mislabel case outside citation-analysis — the vault's own [[myth-structural-credit-assignment-from-minsky-1961|Minsky-1961 "structural credit assignment" mislabel]]. The mechanism is the closest located general account of how an unverified attribution could propagate unread, but the fit is hedged, not confirmed: his evidence is verbatim misprint-copying, while the Minsky case is a conceptual retrofit (a later term attached to an earlier text), and no causal link between the two was established. ([[observation-simkin-roychowdhury-copy-fingerprint-mechanism-is-hedged-fit-for-minsky-1961-mislabel]])
- 2026-10-31: Checked against two *different* vault cases than the 2026-10-29 Minsky test — the citogenesis case ([[claim-wikipedia-amari-sgd-citogenesis]]) and the "Wengert list" motive-free-inertia case ([[claim-wengert-list-named-for-forward-mode-inventor]]) — using a new quote from the same 2002 paper that frames copying as explicitly motiveless ("little incentive to copy if someone has indeed read… the original"). Unlike the Minsky test, this one confirms rather than hedges: both cases are, on their own terms, exactly the propagation-without-motive shape his model describes. ([[observation-simkin-roychowdhury-2002-motiveless-copying-confirms-citogenesis-and-motive-free-inertia-cases]])
- 2026-11-04: His 2005 model was checked directly against [[entity-eduard-petiska|Petiška]]'s 2023 ChatGPT-citation finding (cosine 0.8795, the highest Simkin-pairing cosine measured yet) and resolved the same way as the Algaba pairing: a false friend at the mechanism level, and more cleanly — Petiška's paper, re-read in full, engages the copying/preferential-attachment question even less than Algaba does (no origin hypothesis at all, versus Algaba's one untested training-frequency hypothesis). Third instance of a Simkin-pairing cosine check in this neighborhood resolving via the same iterative-feedback test, after Algaba (false-friend, 0.883) and Alemohammad (analytical-contrast, 0.84). ([[observation-petiska-2023-simkin-2005-cosine-08795-pairing-is-embedding-false-friend]])