NOTE-297: Structural Realism: The Best of Both Worlds?

Contribution

The paper sets up the realism debate as a tension between two old arguments, both of which Worrall finds in Poincaré and Duhem (p. 101).

• The no-miracles argument. Predictive success would be a miracle if the theory were not “essentially” correct.
• The argument from scientific revolutions. The pessimistic induction: predictively successful theories have been overturned at the theoretical level while their empirical content accumulated.

It then argues that neither of the two established resolutions keeps both arguments, and that a third one does.

• Convergent realism (Boyd, Putnam) denies the historical premise. It fails, because Fresnel’s elastic-solid ether was “entirely overthrown” (p. 116).
• Anti-realism comes in two forms. Pragmatism or instrumentalism, and Popperian conjectural realism, which Worrall’s own 1982 view was, both give up the no-miracles argument (pp. 110–111).
• Structural realism, taken from Poincaré, keeps both. Mature theories’ mathematical structure is retained across revolutions, so their success is no miracle, while their ontology is not retained.

The paper adds two pieces of apparatus along the way. One is a non-ad-hoc criterion of scientific “maturity”: the science has produced a theory that predicts new types of phenomena not written into it (p. 114). The other is a reading of the correspondence principle as evidence for structural realism, not for full realism (pp. 120–121).

Key insight

The continuity that realists such as Boyd point to is real, but it sits at the wrong level for full realism. It is “one of form or structure, not of content” (p. 117).

• Fresnel→Maxwell. Fresnel’s four equations for the reflected and refracted amplitude ratios are “directly and fully entailed by Maxwell’s theory” once the amplitudes are read as those of the electric vector (p. 119). Fresnel was wrong about what vibrates but right that the optical phenomena “depend on something or other that undergoes periodic change at right angles to the light” (p. 120).
• Other cases. Where equations are not carried over intact, the correspondence principle recovers them as limits (Newton from GR). That is “approximate continuity” of structure with no continuity of mechanism. Worrall sees “no clear sense” in which action at a distance “approximates” space-time curvature (p. 121).

Assumptions

• Scope. Realism is “modified realism”: grounds for approximate truth of present theories in mature science (p. 104).
• Predictive success. The relevant success is prediction of phenomena not used in constructing the theory. Temporal novelty is not required: Worrall 1985 and 1989 are cited (p. 102 and fn. 5).
• Empirical content. Empirical content accumulates across revolutions when “properly construed”. Genuine Kuhn loss is “remarkably thin on the ground” (fn. 15). This is asserted, with criticisms of Feyerabend and Kuhn deferred to other papers.
• Interpretation. How a theory ought to be interpreted is independent of what its creator believed (fn. 34). This licenses reading Poincaré’s gloss on Fresnel against Fresnel’s own realist intent (fn. 28).
• Continuity of equations. Continuity of equations (identity or limiting case) is taken as the mark of retained structure. “Structure” is not otherwise defined.

Key results

The paper argues. It proves nothing.

1. Against inference to the best explanation (p. 102). The no-miracles argument cannot be run as IBE against an anti-realist, who rejects IBE (Laudan, Fine). Realism’s “explanation” of success also cannot be independently tested. The argument therefore gives only “a prima facie plausibility argument”.

2. Approximate truth (pp. 105–106). It is formally problematic: Popper’s verisimilitude fails (Tichý, Miller). Realism needs it to be transitive, and the tadpole-to-frog photograph series shows that intuitive approximation is not transitive. These problems are set aside, not solved.

3. Maturity (pp. 113–115). Defining maturity by strong predictive success disposes of some of Laudan’s counterexamples, such as the ether theories of Hartley and LeSage. Fresnel’s ether survives as the sharpest case.

4. Against Hardin & Rosenberg (pp. 116–117). Their claim that Fresnel was “really” talking about the electromagnetic field is “taking ‘rational reconstruction’ too far”.

5. Fresnel’s equations (p. 119). In Worrall’s notation (magnitudes, signs suppressed):

   • R/I = tan(i−r)/tan(i+r) and X/I = 2 sin r cos i/(sin(i+r) cos(i−r)), for the component “polarised in the plane of incidence”;
   • R′/I′ = sin(i−r)/sin(i+r) and X′/I′ = 2 sin r cos i/sin(i+r), for the perpendicular component.

   These are carried over unchanged into Maxwell’s theory. Worrall calls this case “unrepresentative”. The common pattern is limiting-case recovery (p. 120).

6. The correspondence principle (pp. 120–121). It operates “purely at the mathematical level”. It is also a heuristic (Zahar 1983b; Worrall 1985). Its general applicability is “not evidence for full-blown realism – but instead only for structural realism” (p. 121).

7. Quantum mechanics (pp. 121–123). Fine’s “realism is dead” (1984) saddles realism with Einstein’s classical-metaphysics demands. The structural realist takes the quantum state as primitive, as Newtonian gravity came to be taken, and asserts only that “the structure of the universe is (probably) something like quantum mechanical” (p. 123). Fn. 40 suggests Fine’s NOA may not differ substantially from structural realism, a “banner” under which realists and anti-realists could unite.

My check of result 5. The four ratios match the standard Fresnel amplitude coefficients, with the tan-form pair for the electric vector in the plane of incidence (p-polarisation) and the sin-form pair for the electric vector perpendicular to it (s-polarisation), when “polarised in the plane of incidence” is read as Worrall reads it, as “vibration in the plane of incidence” (p. 119). That is the modern convention. Under the nineteenth-century convention, “plane of polarisation” is perpendicular to the vibration, so a reader taking the old usage would find the labels swapped. Worrall’s text is internally consistent. Also, Maxwell’s theory yields exactly these forms only for media of equal magnetic permeability, the optical case. The paper does not state that condition.

Claims

id	claim	strength	support
C1	The two most persuasive arguments in the realism debate are both old, found in Poincaré and Duhem, and pull in opposite directions (p. 101)	assertion	quotations from Duhem 1906 and Poincaré 1905 (pp. 102–103)
C2	The no-miracles argument gives only prima facie plausibility; it cannot be an IBE against the anti-realist and realism’s “explanation” is not independently testable (p. 102)	informal argument	appeal to Laudan 1981, Fine 1984
C3	Only novel-type predictive success, not accommodation, sustains no-miracles intuitions; a science is “mature” once it has such theories (p. 114)	informal argument	examples: Neptune, white spot, conical refraction vs. Gosse creationism, Biot’s corpuscular theory
C4	Fresnel’s elastic-solid ether was not approximately retained in later physics; Hardin & Rosenberg’s reinterpretation is untenable (pp. 115–117)	informal argument	historical case; Whittaker 1951 acknowledged on the displacement discontinuity (fn. 28)
C5	Fresnel’s reflection/refraction equations are entailed by Maxwell’s theory when I, R, X are read as electric-vector amplitudes (p. 119)	assertion (standard physics)	none given in the paper; correct for equal permeabilities (my check)
C6	The usual pattern of retention is that old equations reappear as limiting cases of new ones (correspondence principle) (p. 120)	assertion	one example (Newton/GR); Zahar 1983b, Worrall 1985 cited on its heuristic role
C7	The correspondence principle is evidence only for structural, not full, realism (p. 121)	informal argument	follows from C6 plus the claim that mechanisms are not limits of one another
C8	Structural realism is “the only available account” that holds out promise of the best of both worlds (p. 121)	assertion	rests on the elimination of pragmatism, conjectural realism and convergent realism; stronger than the one worked case supports
C9	Fine’s quantum argument against realism does not touch structural realism (pp. 121–123)	informal argument	analogy with the acceptance of action at a distance as a primitive
C10	The empirical level of mature science is essentially cumulative; genuine Kuhn loss is rare (p. 108, fn. 15)	assertion	one example (rectilinear propagation, fn. 16); detailed arguments deferred to Worrall 1978 and 1989a

Concepts

• No-miracles argument: the claim that predictive success would be a cosmic coincidence unless the theory were essentially correct. Here it is weakened to a prima facie plausibility consideration.
• Pessimistic induction / argument from scientific revolutions: past predictively successful theories were overturned at the theoretical level, so present ones probably will be. Worrall attributes it to Poincaré.
• Maturity: the possession of theories that predict new types of phenomena not written into them.
• Conjectural realism: Popper’s realism without verisimilitude. Worrall held it earlier (1982) and gives it up here in favour of a stronger position (fn. 19).
• Structural (syntactic) realism: we have grounds to believe the mathematical structure of mature theories, carried over identically or in a limit, and not their interpretation of the terms.
• Correspondence principle: a new theory’s equations must reduce to the old ones as some quantity tends to a limit. Here it is read as retention of structure.

Connections

• Poincaré 1905 (Science and Hypothesis, pp. 160–2) is the position’s source. Worrall’s reading of Poincaré as a realist follows Zahar 1983a.
• Grover Maxwell 1970a/b (Russellian “structural realism”) is acknowledged as related. Clarifying the relation is left “for further research” (fn. 32).
• Interlocutors: Laudan 1981 (“A Confutation of Convergent Realism”), Boyd 1973/1984, Putnam 1978, Hardin & Rosenberg 1982, Fine 1984 (NOA), van Fraassen 1980.
• Successors: Ladyman 1998 (LIT-342, not yet read) turns this position ontic. Worrall & Zahar 2001 moves it to Ramsey sentences, a move the SEP entry LIT-045 reports and this paper does not contain.

Bearing on the record

What it supplies for the map’s row 9. It is the right citation for the modern revival of structural realism, and for structure as what survives theory change. It is the wrong citation for “to be = to be an invariant of structure” or for “ontology is the epistemology of invariants”.

• It is epistemic, not ontic. Worrall’s structural realist is agnostic about “the basic furniture of the universe” (p. 122) and endorses Poincaré’s “real objects which Nature will hide for ever”. The move from what we can know to what there is belongs to Ladyman 1998 (LIT-342) and 2001 (LIT-217), and to the Ladyman–Ross programme (LIT-301, LIT-219, LIT-184).
• Its “invariance” is diachronic. It means equations preserved, or recovered as limits, across theory change. It does not mean quantities invariant under a group of transformations. The map’s slogan is closer to the group-theoretic sense of structure, which the SEP entry (LIT-045 / NOTE-038) lists separately (§4.1) and which this paper never uses. So the map’s “KNOWN (position)” verdict holds for the OSR literature, but should cite Worrall only for the retention-across-theory-change sense.
• Where the map could lean on it. Worrall’s correspondence-principle argument has the same shape as a claim that what is real in a learned representation is what is preserved across retrainings or reparametrisations, not the particular coordinates. The analogy is the owner’s to draw, not Worrall’s.
• THEORY-017. The analogy is closest to THEORY-017: only unitarily invariant structure is intrinsic, and a basis is extra data. A basis choice plays the role Worrall gives to the “interpretation of the terms”. But Worrall’s structure is equations up to limits, not invariants of a symmetry group. The link is an analogy and should be cited as one.

Other entries.

• LIT-045 / NOTE-038. The paper confirms the SEP entry’s framing (C2 there: Fresnel→Maxwell retention of structure but not ontology). It confirms that the equational notion of structure is Worrall’s. It shows that the Ramsey-sentence formulation and the Newman objection are not in the 1989 paper.
• LIT-215 (Karpenko; NOTE-096) places Worrall in the Poincaré → Maxwell → Worrall ESR line, and the text bears that out.
• LIT-220 (Dennett, “Real Patterns”) is not mentioned. The bridge to real patterns is Ladyman–Ross’s.
• LIT-267 (Xiong, FCA lattice): no genuine connection.

For ML practice. It carries nothing for ML practice in the Anthology.

Limitations

• One detailed case. The Fresnel→Maxwell case carries the argument, and Worrall himself calls it unrepresentative, since the equations survive intact (p. 120). The general claim rests on the correspondence principle, supported by one further example (Newton/GR) and no survey.
• Undefined terms. “Structure” and “structural approximation” are undefined. Worrall concedes that “much clarificatory work needs to be done … concerning the notion of one theory’s structure approximating that of another” (p. 121). His own objection to approximate truth, that it is intransitive (the tadpole and the frog), is not shown to spare structural approximation.
• The structure/content distinction. How is structure separated from content without trivialising the view (the later Newman worry)? The paper does not address it.
• Overstatement. “The only available account” (p. 121) and “the only hopeful way” (p. 117) claim more than the elimination argument shows.
• Deferred support. The empirical-cumulativity premise is asserted and defended elsewhere (Worrall 1978, 1989a), not here.

Open questions

• Precision. Can “one theory’s structure approximates another’s” be made precise, for instance as a limit relation between models or as a category-theoretic reduction? And is that notion transitive where approximate truth is not?
• Two senses of structure. Is retention of equations across theory change the same thing as invariance under a symmetry group, or only sometimes coincident with it? The map’s slogan needs the second. Worrall supplies the first.
• The QM claim. Does the structural reading of quantum mechanics (“the structure of the universe is (probably) something like quantum mechanical”) survive the interpretation-dependence of which structure is retained?

Corrections to the seeded skim

• The seed says the paper is seeded from metadata, and its summary holds: the paper reconciles the no-miracles argument with the pessimistic induction, and what survives Fresnel→Maxwell is mathematical structure, not ontology. Two refinements from the text. (i) Worrall does not claim to originate the position. He attributes it to Poincaré (Science and Hypothesis, 1905, pp. 160–2, quoted pp. 118), credits the term “structural realism” to Grover Maxwell (1970a, 1970b; fn. 32), and calls it “an old and hitherto mostly neglected position” (p. 101). “Founding paper” is right for the modern debate, not for the idea. (ii) He also attributes the pessimistic induction to Poincaré, “usually regarded as a recent methodological discovery but in fact already stated clearly by Poincaré” (p. 109).
• The paper’s own name for the view is “structural or syntactic realism” (p. 112; “syntactic or structural realism”, p. 117).
• The paper contains nothing on Ramsey sentences, the Newman objection or invariance under transformation groups. Those belong to later literature, including Worrall & Zahar 2001 as the SEP entry (LIT-045) reports. “Structure” here means the mathematical equations of a theory, identical or related by a limit. Descriptions that fold Ramsey-sentence ESR into “Worrall 1989” are describing later work.
• The map’s row 9 files this under “reality = invariance; ‘ontology is the epistemology of invariants’”. The paper is explicitly epistemic, not ontological. It says it is “a mistake to think that we can ever ‘understand’ the nature of the basic furniture of the universe” (p. 122), and it quotes Poincaré’s “real objects which Nature will hide for ever from our eyes” (p. 118) with approval. The only invariance in the paper is preservation of equations across theory change. Invariance under symmetry transformations does not appear.
• Metadata. Crossref and the published page agree on Dialectica 43(1–2), pp. 99–124, 1989. A table of contents found by search dates the double issue to June 1989 (unverified against the issue itself), so the seed’s year precision is right. The paper was delivered at a Neuchâtel conference on scientific realism: it thanks Howard Stein “for his comments on the version delivered at Neuchâtel” and answers papers by Agazzi, Stein, Boyd and d’Espagnat given there (fns. 6, 11, 20, 40). The same double issue carries Boyd, Newton-Smith, Stein, d’Espagnat and others. Author affiliation: LSE, Department of Philosophy, Logic and Scientific Method.
• The internal cross-references (“see below, pp. 15–17”, “pp. 18–19”, “quoted below p. 20”; fns. 14, 27, 18) point to manuscript pagination, not to the journal pages. They correspond to journal pp. 113–115, 115–117 and 118.
• The tag metaphysics is defensible, because the paper is about what theories commit us to. But the position defended is epistemic, and philosophy-of-science and epistemology carry it. I suggest keeping metaphysics third, as seeded.