NOTE-178: De Haro & Butterfield — Philosophy and Physics of Duality

Contribution

The first book-length philosophical account of dualities, meant for both philosophers and physicists. Part I states and motivates a formal “Schema” for duality and an account of interpreting physical theories, with elementary examples: position–momentum, wave–particle, classical and Dirac electric–magnetic, and Kramers–Wannier. Part II works through the physics up to string theory and gauge–gravity duality:

• bosonization and sine-Gordon/Thirring;
• particle–vortex;
• the ‘t Hooft–Mandelstam dual-superconductor picture of confinement;
• Montonen–Olive and Seiberg–Witten;
• T-, S- and U-dualities and M-theory;
• AdS/CFT;
• the hole argument with a cosmological constant;
• the Strominger–Vafa entropy count.

Part III uses the Schema in general philosophy of science:

• a defence of isomorphism as the formal criterion of theoretical equivalence (ch. 11);
• a definition of physical equivalence and of the common core (ch. 12);
• “cautious scientific realism”, underdetermination and unextendability (ch. 13);
• the geometric view of theories, emergence and fundamentality (ch. 14);
• understanding and explanation (ch. 15).

Key insight

Separate what is formal from what is interpretative, and give each its own map. A duality is a formal relation, an isomorphism of model roots. Physical (theoretical) equivalence is that plus an interpretative condition, that the interpretation maps commute with the duality (i1 = i2∘d, the “triangle”). Most disputes about dualities, such as whether they are gauge symmetries, favour structural realism, create underdetermination or show spacetime emerges, run these two together. Once they are separated, the logico-semantic relations settle most of what a realist should commit to. The duals entail the common core (M1 ⊨ T, M2 ⊨ T), while inequivalent duals contradict each other on their specific structure.

Assumptions

• Theories as triples ⟨S, Q, D⟩ (eq. 2.1), and models as ⟨m; M̄⟩: a model root plus specific structure (eq. 2.4). The model-root/specific-structure cut is relative to the choice of bare theory and “not God-given”. “Theory” and “model” are used “one level up”: the common core is the theory, the duals are models.
• Referential, intensional semantics. Interpretations are partial structure-preserving maps to a domain of application, with intension and extension components (Carnap/Lewis possible worlds). The world is taken to consist of objects with identity criteria “in some stable sense” (p. 44).
• Scientific realism in the “cautious” form: mind-independence plus warranted, defeasible belief (§13.1). The Schema is said not to depend on it, but ch. 13’s conclusions (fn. 51, p. 457) and unextendability do.
• Univocality (“No deus ex machina”, §11.3.4, p. 391): interpretation may forget distinctions the formalism makes but may not add any.
• Nagelian reduction as the correct notion of reduction: formal, respecting extensions but not intensions (§14.3.2). Motivationalism: to claim theoretical equivalence one must first explicate the common core’s ontology (p. 91).
• Physics status. Many Part II dualities are conjectures. Montonen–Olive has “very good evidence” in N = 4 SYM but no proof (p. 243). The AdS/CFT Hilbert space is unknown beyond limits (p. 319). “The main string dualities remain unproven” (p. 353). The Schema is illustrated “if gauge-gravity duality is true” (p. 320).

Key results

• Schema (§3.1). d_S: S1→S2 and d_Q: Q1→Q2 are isomorphisms with values matching, ⟨Q1, s1⟩ = ⟨d_Q(Q1), d_S(s1)⟩ (eq. 3.3), and equivariant with the dynamics. The common core is required to be a triple. A duality is not a gauge symmetry, contra Rickles. It is analogous to a symmetry, and the analogy yields “same physics” only with internal interpretations.
• Physical equivalence (§12.1.2, p. 402) iff (A) duals and (B) same domain of application, more strongly i1 = i2∘d. Kramers–Wannier is dual but physically inequivalent, because the heat bath is an external standard of temperature. Common core by abstraction (cross out specific structure) or augmentation (bosonization vertex operators; Lagrange multipliers in EM). The desirable logical strength is T ≅ m1 ≅ m2 (§12.2.3).
• Against weaker criteria (ch. 11). Logical equivalence is too strict. Definitional equivalence is necessary, not sufficient. Categorical equivalence is “too weak”. Barrett’s signature argument is diagnosed as conflating formal and interpretative choices (§11.3).
• Underdetermination (ch. 13). There are no genuine cases in the examples. In a hypothetical case the cautious realist commits to the common core. Underdetermination “as usually stated is too strong”, and the antirealist must show that empirically equivalent rivals lack a compatible common core (pp. 459–460, 528).
• Structural realism (§13.2.4). The evidence “does not vindicate a structural realist view” (p. 454). Common cores are rich:
  • bosonization’s contains both bosonic and fermionic states;
  • the electric–magnetic one contains a Faraday-type tensor, with “purely electric” a relational fact;
  • T-duality’s has momentum and winding as non-independent properties.
• Unextendability (§13.3). Sufficient for warrant: the domain of application is already the whole relevant possible world. Two weakenings, backed by stipulated symmetries: position–momentum quantum mechanics (linearity and unitarity), and N = 4 SYM fixed by maximal supersymmetry.
• Hole argument with Λ (§10.1). Hole diffeomorphisms are proper symmetries of the gravity model (specific structure), invisible to the CFT. Boundary conformal diffeomorphisms are stipulated symmetries of the common core. Leibniz equivalence as usually stated is “false” because of large diffeomorphisms (p. 341).
• Emergence (§14.3). Ontological emergence is a non-commuting diagram, i_b ≠ i_t∘link, where link is a non-injective partial homomorphism. Reduction is a surjective link. Reduction (formal) and emergence (interpretative) are compatible. Duality is incompatible with both reductive and irreducible ontological emergence (equinumerosity, surjectivity). There are three ways to combine them: emergence beside duality (AdS/CFT RG), effective duality, and epistemic emergence (irreducibility, or weak “in-practice” emergence).
• Fundamentality (§14.4). Duality is symmetric, so in general it gives no priority. Emergence alone does not imply priority. Reductive emergence gives definitional-dependence priority (thermodynamics under statistical mechanics; M-theory “in so far as” it reduces string and supergravity theories).
• Geometric view (§14.2). Quasi-duals are charts (M_i, φ_i) with moduli as coordinates, and quasi-duality maps are transition functions. The Ising moduli space is two disconnected segments; Seiberg–Witten’s is a thrice-punctured plane. Exact dualities are the “minimal case”, with T = Tc, T ⊂ Tc or Tc ⊂ T.
• Understanding (ch. 15). Adopts de Regt’s pragmatic account. Dualities feed three tools for making spacetime-less quantum gravity intelligible: approximation, similarity and internal comparison. Strominger–Vafa is explanatory in the DN sense without requiring “sameness” of the systems (§10.2.4).

Claims

id	claim	strength	support
C1	Isomorphism of model roots (with a triple as common core) is the right formal criterion of theoretical equivalence; weaker criteria (categorical, definitional) are too weak	moderate	argued against Barrett, Weatherall and Halvorson (ch. 11) and illustrated on every Part I–II example; the authors “never saw a convincing case against” it (p. 393); rests on univocality, itself a stipulated principle
C2	Duals are physically equivalent iff dual and i1 = i2∘d; internal interpretations suffice	strong (as a definition)	a definition plus examples; Kramers–Wannier as a worked counterexample to “dual implies equivalent”
C3	No genuine cases of in-principle underdetermination arise from dualities	moderate	case-by-case over the book’s examples; theoretically inequivalent duals turn out empirically inequivalent
C4	Where duals contradict each other, a cautious realist is warranted in committing only to the common core	moderate	logico-semantic entailment M_i ⊨ T (§12.3); holds only where an interpreted common core exists, which the authors say is “not true in general”
C5	Dualities do not favour structural realism; common-core ontologies are rich	moderate	three arguments (§13.2.4); strongest for exact cases (bosonization, EM); for AdS/CFT the common core is “partial … ongoing work in physics!”
C6	Unextendability, or its weakenings plus stipulated symmetries, warrants a physical-equivalence verdict	weak–moderate	conceptual proposal with QM and N = 4 SYM illustrations; the M-theory case is “currently out of reach”
C7	Exact duality is incompatible with ontological emergence between the duals	moderate	formal argument from equinumerosity and surjectivity; strictly proven for finite sets, with a physical-grounds argument for the infinite case (fn. 57)
C8	Reduction (Nagelian, formal) and ontological emergence are compatible	moderate	follows from the formal/interpretative split; examples: Ising spontaneous magnetisation, massless vs massive particle, Seiberg–Witten
C9	In the hole argument with Λ, Leibniz equivalence as usually stated is false; it holds for hole diffeomorphisms and, on an internal interpretation, for boundary symmetries	moderate	classification of diffeomorphism classes (De Haro 2017b, Thm 3); depends on a large-diffeomorphism argument and on the conformal anomaly vanishing
C10	The Strominger–Vafa count explains black-hole entropy without the weak- and strong-coupling systems being “the same”	weak–moderate	DN and unification accounts of explanation plus BPS protection; relies on the conjectural string dualities
C11	Quasi-duals are like charts on a manifold whose transition functions are the quasi-duality maps	weak (heuristic)	two worked examples (Ising, Seiberg–Witten); explicitly “undogmatic” and “is like”, not “is”

Method

Conceptual analysis and explication in Carnap’s sense, framed set-theoretically (the semantic view). Chapters 11–12 add syntactic reformulations: bosonization is axiomatized as T = ⟨Q, A1–A5⟩ with model-specific axioms B6/F6 acting as bridge laws to classical limits. The book works through extended physics case studies taken from the textbook and research literature, with commuting and non-commuting interpretation diagrams as the recurring tool. Arguments are informal but tightly definitional. There are no new physics results. Some material reuses earlier papers by the authors: De Haro & Butterfield 2021, De Haro 2019a/2020a/2020c, and De Haro & De Regt 2018/2020.

Concepts

• Bare theory / model root / specific structure: the triple ⟨S, Q, D⟩ common to duals, and each model’s extra structure.
• Internal vs external interpretation: interprets only the model root, or the specific structure too. Only internal interpretations guarantee a commuting triangle.
• Common core: the bare theory the duals represent. Obtained by abstraction or augmentation, ideally with T ≅ m1 ≅ m2.
• Stipulated / accidental / proper symmetry: symmetries of the bare theory, of the model root only, or of the specific structure.
• Quasi-duality / effective duality: an approximate, partial or limiting duality with no exact isomorphism.
• Motivationalism vs interpretationalism (Read & Møller-Nielsen): whether equivalence waits on an explicated common-core ontology. The authors endorse motivationalism.
• Cautious scientific realism: realism that fixes commitments by the logico-semantic relations between a theory and its models.
• Unextendability: the domain of application is already the whole relevant possible world.
• Linkage map: link: T_b → T_t, non-injective and a partial homomorphism. It is reductive iff surjective.
• Ontological vs epistemic emergence: novelty in the domain of application, vs novelty only in the bare theories (irreducibility or computational intractability).
• Geometric view / comprehensive theory Tc / successor theory TS.
• Univocality: semantics should not cut finer than the bare theory.

Connections

• LIT-113 (Le Bihan & Read 2018). Cited and its Fig. 1 reused as Fig. 13.3 (p. 446). The book’s entailment argument eliminates two of the six options there. Le Bihan & Read define duality only by empirical equivalence, which is weaker than the Schema, and they deliberately set structural realism aside. The two works agree that the common core alone does not settle actuality without further principles. The book supplies the entailment argument that Le Bihan & Read lack.
• LIT-115 (Corfield, duality as a category-theoretic concept). The book treats category theory as reformulation and judges categorical equivalence “too weak” (§§11.2.3, 12.1.2). It is the natural counterpoint.
• LIT-151 (Wallace, Stating Structural Realism), cited as Wallace 2022c. LIT-045 (Ladyman, Structural Realism), whose later SEP revision is cited as Ladyman 2014. LIT-184 (Ladyman, Effective OSR), LIT-196 (ontological priority, OSR), LIT-152 and LIT-215 (informational structural realism), and LIT-173 (Morganti, coherentism). The book argues dualities give no support to structural realism (§13.2.4). fn. 42 doubts that syntactic informational structural realism (Beni 2018) keeps SR’s aim. So the book bears against any OSR/ISR entry that cites duality as motivation.
• LIT-122 (Belot, time and change). The book does not cite this chapter. It does cite Belot 1998 (networks of theories, §13.2.1) and Belot 2018 (“Fifty million Elvis fans”, on Leibniz equivalence with Λ, which ch. 10 disputes).
• LIT-150 (Rizi, what is emergence) and LIT-027 (emergence as conversion of information). These are other emergence accounts in the record. The book’s emergence is Butterfield-style novelty with the robustness condition dropped, not an information measure. fn. 80 points to the literature on quantitative measures.

Bearing on the record

It is the anchor for the duality and theoretical-equivalence cluster. Any entry that says “X and Y are dual, therefore the same physics” should cite §12.1.2 and check the interpretative condition. Any OSR/ISR entry that claims dualities as evidence should be read against §13.2.4. The record could use the book’s formal/interpretative split (bare theory + interpretation map) as a shared vocabulary for claims about equivalence, reduction and emergence. The emergence account (§14.3) is a precise, citable alternative to information-theoretic emergence accounts, and the two are not in conflict because they answer different questions. It has no direct bearing on ML practice.

Limitations

• Much of the physics is conjectural: AdS/CFT, Montonen–Olive, M-theory and the main string dualities are “unproven”. The Schema is illustrated conditionally, and the common core of gauge–gravity duality is “only a class of asymptotic operators and a conformal class of … metrics” (p. 70).
• The realism is a starting position. Ch. 13 defends it against the underdetermination threat from dualities, not in general, and it defers the pessimistic meta-induction to De Haro 2020c.
• The “rich common core” claim is argued from exact, well-controlled cases (bosonization, EM). It does not reach the cases (AdS/CFT, M-theory) where structural-realist readings were proposed. The reply to Grimmer et al. (fn. 43) addresses one example, not their general worry.
• Univocality and the isomorphism criterion depend on theories being formulated set-theoretically. The authors admit they know of no counterexample “in an example of a scientific theory formulated in set theory” (p. 393), which is the scope of the claim, not evidence beyond it.
• The geometric view and the Lusternik–Schnirelmann suggestion (fn. 44) are heuristic sketches.
• Editorial slips that a careful reader will notice:
  • T-dual radius given as “R̃ := R/α′” on p. 278, against α′/R elsewhere.
  • “non-zero expectation value of the magnetisation above the critical temperature” on p. 483, which should be below.
  • The AdS/CMT superconductor dual described as a four-dimensional black hole for a three-dimensional system on p. 499, but as a five-dimensional black hole for a four-dimensional strange metal on p. 522.
• The preface discloses that “some text suggestions have been taken from www.bing.com” (p. 2).

Open questions

• What is the explicit common core of the string and gauge–gravity dualities, beyond the conformal fixed point and perturbative regimes? The authors list this first.
• Does the “rich common core” verdict survive for AdS/CFT once a non-perturbative formulation exists, or does Dawid’s “virtually empty” worry return there?
• Can unextendability be applied to any empirically relevant (non-maximally-supersymmetric) theory, or is it only a limiting ideal?
• Can the geometric view be made precise beyond manifolds (algebraic varieties, Hilbert manifolds), and does moduli-space topology really track ontological emergence, as fn. 44 suggests?
• Do syntactic reconstructions like the bosonization axiomatization (§12.4) extend to non-exact cases, where the entailment argument for realism would be needed most?
• Would a cautious realist’s commitment to T hold up against a case of “transient” historical underdetermination? The authors suggest the lesson applies (p. 450) but do not show it.

Corrections to the seeded skim

• The skim covered only §1.1, §3.6 and ch. 16 and asked for chapters 3, 12 and 14. All three are now read, and the whole book with them. The dossier’s summary is accurate in substance. The citation line says “De Haro et al. (2025)”, but the book has exactly two authors, so “De Haro & Butterfield (2025)” is the right form.
• “Dualities by themselves do not indicate emergence or fundamentality, AdS/CFT excepted” misplaces the exception. §16.2 (p. 529) says dualities are “by themselves not indicative of emergence or fundamentality, nor are they—except in the significant case of AdS-CFT, where an extra dimension appears—compatible with anything other than an epistemic conception of emergence”. AdS/CFT is the exception to incompatibility with non-epistemic emergence (“emergence beside duality”, §14.3.4(1), Fig. 14.8, where the emergence map runs along the RG/radial direction and not along the duality map). It is not an exception to “not indicative”. §14.3.3 (p. 489) states that on a theoretically-equivalent reading, Horowitz & Polchinski’s “emergent gravity” in AdS/CFT “cannot be understood as ontological emergence”.
• “Duality is an isomorphism of models with respect to the structure of a common core” is right, but the dossier leaves out the condition that does the work. The common core must itself be a bare theory, a triple ⟨S, Q, D⟩ (§3.1, pp. 68–73). This is what blocks trivial dualities, and it is what fn. 43 (p. 452) relies on to dismiss Grimmer, Cinti & Jaksland (2024). The book concedes (fn. 15, ch. 2) that not every common core is rich enough to be a theory, and restricts the Schema to cases where it is.
• The underdetermination claim is stronger than the dossier’s paraphrase. §13.4 (p. 459) says “Strictly speaking, we do not find any genuine cases of under-determination”, because in the examples, theoretically inequivalent duals turn out empirically inequivalent (e.g. Kramers–Wannier temperatures). The “cautious realist commits to the common core” argument (§13.2.3, pp. 444–450, via M1 ⊨ T, M2 ⊨ T, M1 ⊨ ¬M2) is the fallback for a hypothetical sui generis case. It makes the underdetermination “benign”, not absent.
• On structural realism, the dossier’s “no evidence that dualities favour structural realism” (from §16.1) understates §13.2.4 (pp. 450–454). That section gives three positive arguments against Matsubara’s (2013) and Dawid’s (2013) structural-realist readings: the hole-argument analogy (motivationalism resembles sophisticated substantivalism); that a referential, intensional semantics for bare theories “goes beyond what a structural realist would say”; and that common-core ontologies are “rich, rather than empty”. It also endorses Read & Møller-Nielsen’s motivationalism (p. 91, §12.1.1, fn. 39 of ch. 13), which neither the dossier nor the LIT entry mentions.
• The book also proposes unextendability as a sufficient condition for epistemic warrant of a physical-equivalence verdict (§13.3, pp. 454–459). The dossier omits it, and it is the book’s main answer to “when are we entitled to say duals are equivalent?”.
• The preface (p. 2) states: “Some text suggestions have been taken from www.bing.com.” No record entry notes this disclosure.
• metaphysics tag: justified. The book argues about ontology and realism: what the common core’s ontology is and whether it is “virtually empty”, sophisticated substantivalism and Leibniz equivalence in the hole argument with Λ (§10.1, where Leibniz equivalence “as usually stated” is argued false because of large diffeomorphisms, p. 341), ontological emergence as novel reference, and fundamentality including grounding (§14.4). Someone browsing metaphysics would be right to expect it. It is secondary, though: the book treats emergence as a “semantic or interpretative, and not … epistemic or metaphysical” condition (p. 479), and it sets grounding aside (§14.4.1(iv)).
• primary topic: philosophy-of-science (keep as listed). The load-bearing questions are theoretical equivalence, theory individuation, scientific realism and underdetermination, reduction and emergence, and explanation and understanding. quantum-foundations fits second, for the QFT and string physics of Part II. metaphysics fits third.