NOTE-038: Structural Realism (Stanford Encyclopedia of Philosophy)

Contribution

The entry is a survey, and it adds organisation. It supplies:

• a single taxonomy separating ESR (restricting belief to structure) from OSR (inflating the ontological priority of structure);
• three graded ESR theses: unknowable individuals; unknowable intrinsic properties; knowledge only of second-order structure (Russell/Carnap);
• seven OSR theses, listed below;
• case-study sections on group theory, QFT, spacetime and the special sciences;
• seven catalogued objections with replies.

The seven OSR theses are:

1. eliminativism about individuals;
2. relations that do not supervene on intrinsic properties;
3. no intrinsic natures, so no haecceities;
4. individuals without irreducibly intrinsic properties (Esfeld’s “moderate structural realism”);
5. identity and diversity grounded in relational structure (Saunders’ weak discernibility);
6. no self-subsistent objects;
7. objects as constructs.

The 2023 revision widens OSR’s reach to the special sciences through “effective ontology” and “real patterns” (§4.4). It also includes the author’s reply to the objection that OSR is not worked out: “The wealth of work referred to above makes these charges harder to sustain now” (§5(3)).

Key insight

The theory-change argument threatens realism about ontology, not about relations among phenomena. Examples: Fresnel’s elastic ether to Maxwell’s field, phlogiston to oxygen (the same grouping of combustion, respiration and calcination, and reciprocal smelting), Newton’s tidal-lock explanation surviving into GR. In each, what survives is the equations or the modal structure: which counterfactuals, explanations and predictions hold in a regime (§2, §4 opening). The two forms differ in what they do with that. The epistemic reading says we can know only that structure. When it is cashed out as Ramsey sentences it collapses into empiricism plus a cardinality claim (Demopoulos & Friedman: “only cardinality questions are open to discovery!”, §3.2). The ontic reading says structure, modal structure in particular, is what there is. That escapes Newman by refusing an extensional notion of relation, and it pays for it with the relata, causation and mathematical-vs-physical objections.

Assumptions

These are the entry’s framing commitments, since it has no formal results of its own.

• The dialectic is set by the no-miracles argument (Putnam 1975) against the pessimistic meta-induction, in Laudan’s 1981 three-step form (§2). Other anti-realist routes are set aside as less compelling: underdetermination, Stanford’s unconceived alternatives, base-rate objections to no-miracles.
• “Structure” is left plural. The entry says “there many different notions of structure” (§1): mathematical/equational (Worrall), second-order/Ramsey (Russell, Maxwell), group-theoretic invariants (§4.1), modal structure (§4), real patterns (§4.4). Objection (7) in §5 is that this plurality is a defect.
• OSR’s “modal structure” is “an abstraction of causal structure and nomological structure … intended to also include symmetry principles and perhaps more” (§4). This commitment to natural necessity, not Humean supervenience (Berenstain & Ladyman 2012), is what the entry uses to separate OSR from standard realism in §5(1).
• Scope. Most case material is physics. §4.4 extends it to the special sciences via Ladyman & Ross 2007, but it records dissent that SR applies only to the mathematical sciences (Gower 2000; Mark Newman 2005).

Key results

These are results the entry reports; the proofs are in the cited sources.

• The Newman / Demopoulos–Friedman result (§3.2). If a first-order theory Π(O₁…O_n; T₁…T_m) is consistent and all its observational consequences are true, its Ramsey sentence ∃t₁…t_m Π(O; t) is true as a theorem of second-order logic or set theory, given a domain of the right cardinality. So Ramsey-sentence ESR cannot distinguish theories with the same observational consequences. English 1973: incompatible Ramsey sentences cannot share all observational consequences. Theoretical equivalence therefore collapses into empirical equivalence. The claimed escape through un-Ramsified mixed predicates (Cruse, Melia & Saatsi) is answered by Demopoulos 2011: the content is still observational content plus cardinality.
• Discernibility (§4, thesis (5)).
  • Saunders: electrons in the singlet state are weakly discernible, since “has opposite spin to” is irreflexive.
  • Ladyman, Linnebo & Pettigrew 2012: “weak discernibility in a language without names is equivalent to absolute discernibility in a language with names”.
• Schurz 2009 “proves a structural correspondence theorem showing that successive theories that share empirical content also share theoretical content” (§2). The entry reports this and does not state the theorem.
• QFT (§4.2). Field quanta are not individuals, so the French–Redhead underdetermination does not arise. Malament 1996, Halvorson & Clifton 2002 and Fraser 2008 (Haag’s theorem) argue against localisable particles. Unitarily inequivalent representations pose a problem for algebraic structure (Howard 2011, Ruetsche 2011); French 2012 and Glick 2016 respond.
• Spacetime (§4.3). General covariance and the hole argument (Earman & Norton 1987) push toward structural or “sophisticated” substantivalism (Stein 1968, DiSalle 1994, Pooley). Pooley 2006 denies that the analogy with quantum individuality is deep.

Claims

id	claim	strength	support
C1	Structural realism is “considered by many realists and antirealists alike as the most defensible form of scientific realism”	weak	assertion in the preamble; no survey or citation; the entry’s most-cited sentence and its least supported
C2	Across major theory changes (Fresnel→Maxwell, phlogiston→oxygen, Newton→GR) structure or modal structure is retained while ontology is not	moderate	case studies cited (Worrall 1989; Schurz 2009; Ladyman 2011; Saunders 1993a; Batterman 2002, etc.); informal argument in the entry
C3	Ramsey-sentence ESR is trivialised by the Newman objection: its content reduces to observational content plus a cardinality claim	moderate	reported formal argument (Demopoulos & Friedman 1985; Ketland 2004), with contested replies (Melia & Saatsi 2006; Friedman 2011); the entry does not adjudicate
C4	OSR avoids the Newman objection because it does not construe relations extensionally	weak	assertion attributed to Ladyman 1998 and Ladyman & Ross 2007
C5	Quantum entanglement implies relations that do not supervene on intrinsic properties of the relata, contrary to Humean supervenience	moderate	informal argument from Cleland 1984, Maudlin 1998, Teller 1989, reported
C6	Quantum particles are at least weakly discernible, so PII in weak form survives	moderate	reported argument (Saunders 2003a,b, 2006), with the Ladyman–Linnebo–Pettigrew equivalence result
C7	OSR’s rivals are no more “worked out” than OSR	weak	assertion in the author’s own voice (§5(3))
C8	Causation is “very often the pragmatically indispensable proxy for modal structure in the sciences”	weak	assertion attributed to Ladyman & Ross 2007 (§5(4))
C9	Ontology is scale-relative; there may be no fundamental level	weak	position of Ladyman & Ross 2007 reported sympathetically (§4.2, §4.4); argued there, not here

Concepts

• Epistemic structural realism (ESR): belief is restricted to what theories say about the relations among unobservables. It is agnostic about their natures (Psillos’s “restrictive structural realism”).
• Ontic structural realism (OSR): any SR “incorporating an ontological or metaphysical thesis that inflates the ontological priority of structure and relations”. It comes in the seven theses above.
• Upward and downward paths (Psillos): the upward path goes to SR from empiricist epistemology (Russell’s percepts plus the Helmholtz–Weyl principle). The downward path goes by weakening standard realism (Worrall).
• Ramsey sentence: ∃t₁…t_m Π(O₁…O_n; t₁…t_m). Theoretical terms are replaced by bound predicate variables.
• Newman objection: any collection of the right cardinality realises any purely formal structure, so structural claims are trivial apart from cardinality.
• Modal structure: the structure of necessity, possibility, counterfactual dependence, symmetry and probability among phenomena. It is broader than causal or dispositional structure.
• Weak discernibility: a and b are weakly discernible iff some relation holding between them is irreflexive.
• Moderate structural realism (Esfeld): there are objects, but all their properties are relational. Neither objects nor relations are ontologically prior.
• Real pattern (Dennett, Ross, Ladyman & Ross): an entity of any category that features non-redundantly in projectable regularities. It is often glossed as compressibility.
• Effective ontology: ontology that holds within a regime of energy, length or time scale. On Ladyman & Ross this is the only kind outside “ultimate physics”.
• Rainforest realism: realism about the effective ontologies of the special sciences conjoined with OSR about physics (Ladyman & Ross 2007).

Connections

LIT-037 (Cuffaro & Hartmann), §6 item 4, says OSR “seems to rely on a distinction between extrinsic and intrinsic properties (Ladyman & Ross, 2007, p. 151)”, which the open systems view would dissolve. This entry shows how much work that distinction does in OSR:

• Thesis (2) is a claim about relations failing to supervene on intrinsic properties.
• Theses (3)–(4) deny irreducibly intrinsic properties to individuals.
• ESR in the Jackson/Langton form turns on intrinsic natures being unknowable (§3.1).

So Cuffaro & Hartmann’s worry lands on some OSR variants (those that state themselves as denials about intrinsic properties) more than others. Their speculation that OSR is compatible with the open view is not tested by either text. The two texts also meet on Everett. The entry cites Wallace 2003 and Saunders for a structuralist and real-patterns answer to the preferred-basis problem (§4, §4.4). That is the Everettian reading LIT-037 §3 relies on when it treats reduced density operators as “just as real” as the universal state.

The entry’s only contact with the GPT/operational-theories programme is its pointer to Adlam 2022 (operational theories as specifications of modal structure) and to Constructor Theory (§4 thesis (2)). LIT-073 is the framework those pointers presuppose. The Spekkens programme (LIT-007, LIT-019, LIT-054, LIT-003) and ψ-ontology (LIT-062, LIT-090) are not mentioned. Two apparent resonances are the reader’s analogies, not links the entry draws:

• between ESR’s “we know only structure, not natures” and an epistemic restriction on classical states;
• between OSR’s non-supervenient entanglement relations and the nonclassicality residue those papers isolate.

Bearing on the record

It carries nothing for ML practice. It holds no instruction, and it has no argument about learned representations (see corrections). A THEORY document that wanted to say “what transfers across model changes is relational structure” could cite §2 for the philosophical form of that claim. It would have to supply the ML argument itself.

For filing: the entry’s primary subject is philosophy of science (scientific realism, theory change, the Ramsey-sentence view of theories). philosophy-of-science (ADR-004) is its primary topic; metaphysics fits §4 and §6, and quantum-foundations fits §4 theses (2)–(5) and §4.2. identity would be justified by thesis (5) and the PII discussion. LIT-045 should not be cited as support for OSR, since it is a survey with an authorial stance. It should be cited for the taxonomy and the Newman result.

Note of 2026-10-16: THEORY-036 now cites LIT-045 as a source.

Limitations

• It is written by OSR’s originator (Ladyman 1998 is the source of the ESR/OSR distinction). The §5 replies and several §4 framings are advocacy: “The wealth of work … makes these charges harder to sustain now”; “it is not clear that OSR’s rivals are ‘worked out’ in any sense that OSR is not”.
• The preamble’s headline, that SR is “the most defensible form of scientific realism” by many lights, is unsupported.
• As a survey it reports arguments without reconstructing them. The Newman result, Schurz’s theorem and the discernibility equivalence are stated, not proved.
• The notion of “structure” is never fixed. The entry concedes this in objection §5(7) and in thesis (1)‘s remark that debates “are not productive unless the terminology is univocal”.
• Coverage is physics-heavy. The special-sciences section (§4.4) is programmatic: triangulating real patterns to get objects is “a subject for further work” (§5(5)), and so is the cognitive role of objects (§4(7)).
• It contains typographic slips (“there many different notions”, “in ever form”, “to the extant”, “There in no general agreement”) but no substantive errors that I found.

Open questions

The entry names these:

• whether contextual identity depends on the whole structure or on part of it, and which of identity or diversity is primitive (§4(5));
• a non-ad hoc account of why science refers to and generalises over objects, possibly from cognitive science (§4(7));
• how real patterns, triangulated across detection modes, yield objects rather than “minimal real patterns” (§5(5));
• how ontology, causation and law at different scales relate (§4.4);
• whether group structure or category theory is the right formal notion of structure (§4.1, §4.3; Roberts 2011, Bain 2013, Lal & Teh 2017).

A precise, agreed notion of modal structure would close §5(1) and §5(7) together. So would a demonstration that it distinguishes OSR from standard realism with a commitment to natural necessity.

Corrections to the seeded skim

• The dossier puts “French & Redhead on metaphysical underdetermination” in §4.2 (QFT). In the entry the underdetermination dilemma belongs to non-relativistic many-particle QM: either quantum particles are not individuals, or they are individuated by an empirically transcendent haecceity. It appears under OSR thesis (3), in the discussion of the Principle of the Identity of Indiscernibles. §4.2 says the reverse for QFT: the underdetermination “is obviated in the context of quantum field theory, because a particles-as-individuals view is not an option” (citing Bigaj 2018).

• The dossier lists “the need for relata (Chakravartty, Psillos, Cao …)” among the §5 objections. The relations-without-relata objection is discussed in §4, under OSR thesis (1), eliminativism. That is also where the entry gives its two replies: relations as universals (Platonic OSR, after Stein), and “relations all the way down”, with no fundamental relata (Saunders, Ladyman & Ross). §5 has seven different objections:

  1. collapse into standard realism (Psillos, Papineau, Stanford);
  2. not all structure is retained;
  3. too metaphysically revisionary;
  4. cannot account for causation;
  5. “coincidence or object?” (Chakravartty);
  6. collapse of the mathematical/physical distinction (van Fraassen);
  7. no clear notion of structure.

• The dossier suggests the entry’s “structure is what transfers” and real-patterns strands “connect to arguments about what representations learn”. The entry makes no such connection. The nearest material:

  • A one-line mention (§4.4) of Beni 2019’s “cognitive structural realism … in terms of connectionist networks and activation patterns”.
  • A remark (§4.4) that real patterns are “natural[ly]” characterised by compressibility or by a reduction in effective degrees of freedom. Millhouse 2021 is cited against compressibility for coarse-grained patterns.

  Any ML-representation reading is the reader’s own.

• The dossier quotes “Newman objection trivialises the Ramsey sentence formulation of ESR” as if it were the entry’s verdict. The entry attributes it to Ketland 2004 (“Jeffery Ketland (2004) argues in detail that…”). It then reports replies: Worrall & Zahar 2001, Cruse 2005, Melia & Saatsi 2006 on mixed predicates, Friedman 2011 on Carnap. It also reports Demopoulos’s 2011 counter-reply. The entry leaves the question open.

• The dossier’s §6 summary omits Dunlap 2022 (information-theoretic interpretations of QM as OSR) and Isaac 2014 (structural realism for secondary qualities). It also omits the §4 item (2) pointer to operational theories: Adlam 2022 reads the constraints on correlations in different operational theories as “different specifications of modal structure”. That pointer is the entry’s only contact with the GPT literature.