Empirical Structure Realism Explained

• Empirical Structure Realism is a philosophical stance that prioritizes observable structures—such as relational patterns, invariants, and contextual observables—over unobservable intrinsic properties.
• It is operationalized in theory comparison, simulation, and cosmology to ensure that successive theories preserve and refine the empirical discriminations of earlier frameworks.
• The approach impacts the interpretation of quantum mechanics and cosmology by emphasizing invariant structures that remain robust across dual descriptions and evolving theoretical models.

Empirical Structure Realism is a family of realist positions according to which the most defensible scientific commitments concern empirically anchored structure—relations, invariants, contextual patterns, and discriminations preserved in successful inquiry—rather than the intrinsic natures of putative underlying entities. In recent work, the term ranges from an explicit thesis that earlier physical theories supervene on later ones through refinement of empirical structure, to relational ontologies of matter points, contextual accounts of quantum objects, information-theoretic invariants in cosmology, and operational criteria for judging whether synthetic models preserve the relevant structure of real systems (Gyenis, 26 Jul 2025, Esfeld et al., 2015, Karakostas, 2012, Profumo, 19 Jun 2026).

1. Conceptual profile and range of usage

The contemporary literature does not treat empirical structure realism as a single canonical doctrine. Rather, it groups together several closely related claims about what realism should target when ontology is unstable, underdetermined, or highly theory-dependent. In philosophy of physics, the emphasis typically falls on relations, contextual observables, or invariant structures. In intertheoretic analysis, the emphasis falls on preservation, refinement, or common-core structure across theory change or duality. In simulation and modeling, the emphasis becomes operational: a surrogate is realistic when it preserves the relevant empirical or behavioral structure of the target system (Haro, 2021, Karra et al., 2018, Ahmed et al., 18 Dec 2025, Münker et al., 27 Jun 2025).

Setting	Structure treated as central	Realist or validation criterion
Foundations of physics	Relational, contextual, or modal structure	Commit to what successful theory requires for empirical phenomena
Theory comparison	Preserved, refined, or invariant structure	Later or dual formulations retain empirical discriminations
Simulation and synthesis	Behavioral, graph, or communication structure	Synthetic artifacts match real systems in measured structure

Across these uses, empirical structure realism is typically anti-essentialist. It resists the inference from empirical success to intrinsic nature, while also resisting the conclusion that science yields only prediction. This suggests a characteristic middle position: stronger than instrumentalism, weaker than naive ontology-first realism, and often more selective than global ontic structural realism.

2. Relational ontology and contextual objects

A paradigmatic metaphysical articulation appears in the defense of a “moderate ontic structural realism” about matter points. On this view, there are fundamental physical objects, but their identity is exhausted by the spatial relations in which they stand: “all there is to the matter points are the spatial relations in which they stand.” Absolute space and intrinsic properties are rejected, and change is represented by a dynamical structure governing the evolution of relational configurations, for example

$\frac{d}{dt}\Delta_t = v_t(\Delta_t).$

Masses, charges, initial velocities, and in Bohmian mechanics the initial wave function are treated as dynamical parameters rather than intrinsic essences; the wave function “traces” the evolution of the spatial relations among particles (Esfeld et al., 2015).

This relationalism is structurally realist without being object-free. The view explicitly rejects the stronger thesis that only structure exists, insisting instead that if there are relations, there are objects that stand in them. Empirical discipline enters through parsimony and adequacy: the ontology is meant to match classical mechanics and quantum mechanics while positing no more than matter points, their spatial relations, and dynamical structure (Esfeld et al., 2015).

A related but distinct line appears in contextual realism about quantum mechanics. Here the target is not matter points but quantum objects and entangled wholes. Classical separability, the definite-values principle, and non-contextual value attribution are rejected. A quantum object is real only within a context that selects a set of co-measurable observables; the Heisenberg cut is therefore methodological rather than a metaphysical partition already present in nature. What science discloses is a network of entangled interrelations and contextual objects of empirical reality, while “reality in itself” remains inaccessible in principle (Karakostas, 2012).

Taken together, these views define a central strand of empirical structure realism: realism attaches to relational or contextual organization sufficient for empirical science, not to a stock of intrinsically individuated substances.

3. Formalizing empirical structure and theory change

The most explicit formalization defines a theory’s empirical content region by region. If $H^T$ is the set of history-descriptions of a theory $T$, and $EA^T(h,R)$ means that $h$ is empirically adequate to a spacetime region $R$, then the empirical content of $R$ according to $T$ is

$e^T_R \doteq \{ h \in H^T \mid EA^T(h,R) \}.$

The empirical structure of $T$ is the partition of spacetime regions induced by equality of empirical content:

$H^T$0

One theory supervenes on another when the latter refines the former’s empirical structure:

$H^T$1

In this framework, empirical structure realism is the thesis that earlier theories of physics supervene on later theories of physics, and an “optimistic meta-induction” supports the expectation that future physics will refine the empirical structure of current physics (Gyenis, 26 Jul 2025).

This approach shifts the focus from ontology or full formal content to empirical discrimination. A later theory need not preserve the old ontology; it must preserve, and possibly sharpen, the theory’s ability to distinguish empirically different regions. Examples proposed in this spirit include the claim that Newtonian gravitation supervenes on General Relativity and that Classical Thermodynamics strongly supervenes on Statistical Mechanics (Gyenis, 26 Jul 2025).

A historically different but closely aligned defense appears in the treatment of Sommerfeld’s 1916 fine-structure derivation. That case seems hostile to realism because the exact successful formula was obtained using elliptical electron orbits and other now-rejected elements. The proposed realist response is not to defend the obsolete ontology but to identify “sufficient continuity of relevant structure” across theory change. The distinction between working structure and idle structure is central: the predictive success is explained by structural features preserved into later quantum theory, not by the truth of the orbit picture as such (Vickers, 2019).

Empirical structure realism in this sense is selective. It takes theory change to preserve what is structurally responsible for success, while allowing discontinuity in entities, imagery, and even parts of formalism.

4. Duality, equivalence, and cross-representational invariants

Dualities intensify the central problem of empirical structure realism because they present empirically equivalent descriptions with differing interpretations. One influential response distinguishes internal from external interpretation. Dual bare theories may be related by a duality map $H^T$2, and once interpretation maps are added, empirical underdetermination arises when the theories are empirically equivalent but theoretically inequivalent. The recommended cautious realism is then realism about the shared common core accessed by internal interpretation rather than about incompatible external ontologies (Haro, 2021).

That common-core strategy has, however, been explicitly challenged. A recent alternative argues that realism should attach not to the invariant common core alone but to the theory’s full formal structure, while ontological commitment should be tied to empirically accessible contexts. In quantum mechanics this is illustrated by the existence of many inequivalent tensor product structures, each fixing different subsystem ontologies; in string theory it is illustrated by dual perturbative or near-classical regimes. The common core may be mathematically extractable, but it can be too thin to carry empirical interpretation or explanatory force (Dawid et al., 18 Dec 2025).

A closely related dispute appears in the claim that the Schrödinger and Heisenberg pictures are only instrumentally equivalent. Since they yield the same Born-rule predictions, instrumentalism treats them as interchangeable. Under realism, however, the relevant question is whether the underlying structures are isomorphic. The proposal here is that Deutsch–Hayden descriptors constitute a richer, local, separable structure that surjects onto the Schrödinger state without reducing to it, so the two pictures are not equivalent on a realist reading (Bédard, 2 Oct 2025).

In cosmology, the same issue is recast in information-theoretic form. The objective content of cosmology is proposed to lie in cross-representational informational invariants such as correlation structure, distinguishability structure, and limits on accessible information. A model family $H^T$3 induces a Fisher information matrix

$H^T$4

and hence the Fisher–Rao metric $H^T$5, while empirical equivalence classes can be described using $H^T$6. The realist commitment is then to what remains invariant across empirically adequate representations, not to the ontology of any single favored formalism (Profumo, 19 Jun 2026).

5. Empirical access, internal views, and emergent observables

A recurrent objection to structural realism is that it may fail to recover the empirical world of observers, measurements, and localized entities. The problem becomes sharp in quantum gravity, where fundamental spacetime may be absent. A theory is empirically incoherent if its truth undermines empirical justification for believing it true, and the worry is that only spacetime-localized “local beables” can be observed. The proposed response is that theories without fundamental spacetime are not automatically empirically incoherent if they can derive structures that play the empirical role of localized observables, whether by approximation, limiting procedures, duality, or algebraic representation (Huggett et al., 2012).

Relational observables in general relativity and quantum gravity provide a positive structuralist program. In the partial/complete observables formalism, partial observables are gauge-variant quantities associated with measurement procedures, whereas complete observables are gauge-invariant relational quantities. In the quantum reference frame formalism, one begins from a perspective-neutral state and recovers states only relative to a chosen subsystem. The key conceptual move is the “internal view”: apparently non-modal or intrinsic-looking features arise only relative to a subsystem functioning as a frame, while the underlying ontology is modal and structural (Adlam, 27 Jul 2025).

Not all authors accept that observerhood can itself be reduced to structure. An explicit anti-reductionist theorem argues that if all observer-like structures across all parametrizations counted as genuine observers, their memory would be no better than random guess with respect to the external world. The conclusion is that there must be more to observers than structure alone, and that the unique correspondence between observables and physical properties becomes manifest through actual observers rather than through formal structure by itself (Stoica, 2023).

This disagreement marks an important fault line. One strand of empirical structure realism tries to derive empirical access from relational observables and internal viewpoints; another insists that bare structure underdetermines the conditions of observation.

6. Operational uses in simulation, networks, and machine learning

Outside foundational metaphysics, empirical structure realism has been operationalized as a validation principle for synthetic or surrogate systems. In synthetic social contact networks, realism is evaluated by comparing global structural signatures with empirical graphs from the literature. The networks are represented by 34 features derived from 11 graph-complexity measures, including clustering, centrality distributions, degree statistics, entropy-like measures, and connectivity or distance features. K-means clustering with six clusters and PCA visualization are then used to compare synthetic, stylized, and empirical graphs; the main result is that agent-synthetic graphs cluster with real-world graphs more often than stylized baselines do (Karra et al., 2018).

In deep learning mutation analysis, realism is treated as preservation of fault-detection structure. The benchmark uses 86 reproducible real faults from CleanML, DeepFD, DeepLocalize, and defect4ML, and defines two realism measures: Coupling Strength (CS) and Intersection over Union (IoU). On this basis, pre-training mutants are found to be more realistic than post-training mutants: pre-training mutants had the highest median CS in 56 bugs (65%) and the highest median IoU in 64 bugs (74%). The philosophical significance is limited and methodological: a surrogate is realistic when the pattern of killing inputs overlaps with that of real faults in the relevant empirical sense (Ahmed et al., 18 Dec 2025).

A parallel development appears in generative-agent simulations of social network communication. The TWON framework distinguishes agent behavior $H^T$7, state update $H^T$8, and platform-level mediation $H^T$9, and realism is benchmarked through losses on user behavior and network mechanics, $T$0 and $T$1. The principal methodological claim is that simulations should be validated by their empirical realism in the same setting in which components were fitted. The empirical results are strongly setting-dependent: reply-likelihood prediction achieved weighted $T$2 in English and approximately $T$3 in German (Münker et al., 27 Jun 2025).

These applications do not advance a metaphysical doctrine about what ultimately exists. They instead operationalize a structurally realist intuition: plausible appearance is insufficient, and validity depends on preserving the empirically relevant relational pattern.

7. Major disputes and philosophical significance

The first major dispute concerns whether structure can stand alone. Moderate structural realists retain relata, whether as matter points or contextual quantum objects, and deny that relations eliminate objects altogether (Esfeld et al., 2015, Karakostas, 2012). Anti-reductionist arguments about observers strengthen this resistance to object-free formulations (Stoica, 2023). By contrast, modal and relational approaches in quantum foundations sometimes use structural considerations to revise direct realism rather than locality, treating branch relations, light-cone constraints, or internal relational descriptions as primary (Vongehr, 2011).

A second dispute concerns the proper target of realism under empirical equivalence. One option is common-core realism: commit to what empirically equivalent or dual theories share (Haro, 2021). Another is realism about full formal structure with ontology indexed to empirical context (Dawid et al., 18 Dec 2025). A third holds that empirical equivalence is too weak for realist equivalence because non-isomorphic structures can support the same predictions (Bédard, 2 Oct 2025). In cosmology, this pressure motivates realism about informational invariants across representations rather than about any single model ontology (Profumo, 19 Jun 2026).

A third dispute concerns the scope of the “empirical.” In some formulations it means the partition of spacetime regions by empirical adequacy and discrimination (Gyenis, 26 Jul 2025). In others it means contextual accessibility of quantum properties (Karakostas, 2012), derivability of local beables from non-spatiotemporal theory (Huggett et al., 2012), or measurable structural similarity in benchmarks for synthetic systems (Karra et al., 2018, Ahmed et al., 18 Dec 2025, Münker et al., 27 Jun 2025). This suggests that empirical structure realism is unified less by one formal criterion than by a common constraint: realist commitment must be answerable to structure that successful inquiry can in some principled sense recover, preserve, or validate.

In that broad sense, empirical structure realism remains a durable strategy for reconciling realism with theory change, duality, contextuality, emergent spacetime, and the increasing use of synthetic models. It does so by relocating epistemic and ontological weight from intrinsic essence to empirically disciplined structure, while leaving open a substantive and unresolved question: exactly which structures are preserved, which are merely representational, and which are sufficient for realism.