Theory Supervenience

• Theory supervenience is a framework that defines a hierarchical empirical relationship where one theory’s capacity to differentiate spacetime regions is preserved by another.
• The concept employs precise conditions—such as the preservation of empirical partitions—to ensure that every distinction in the supervening theory is matched by the base theory.
• Canonical examples, like Newtonian gravitation versus General Relativity, illustrate how refining empirical content underlies inter-theoretic reduction and supports minimalist realism.

Theory supervenience is a reductive relationship between theories in which one theory supervenes on another when the empirical structure of the latter theory refines the empirical structure of the former (Gyenis, 26 Jul 2025). In this usage, the dependence relation is not stated in terms of ontology, bridge laws, or formal derivability, but in terms of which spacetime regions a theory can empirically distinguish. The notion is deployed to articulate a form of physicalism and realism that is explicitly empirical-structural, and it stands in close contact with broader supervenience frameworks that analyze determination via kinematical possibilities, definability, and reduction (Caulton, 2024).

1. Empirical structure as the basis of the relation

Gyenis formulates theory supervenience by first characterizing a scientific theory $T$ through a universe $H^T$ of history-descriptions and an empirical-adequacy relation $EA^T(h,R)$ between history-descriptions and actual spacetime regions $R \subseteq M$ (Gyenis, 26 Jul 2025). A history-description $h \in H^T$ is any candidate representation of how the relevant relata evolve or are arranged over a region, while $EA^T(h,R)$ holds when, according to $T$, $h$ can be idealized, localized, and approximated so as to save the observable data in $R$.

From $EA^T$, the empirical content of $H^T$0 at a region $H^T$1 is defined as

$H^T$2

This yields a full empirical content function $H^T$3. The empirical structure $H^T$4 of the theory is then obtained by grouping together exactly those regions with the same empirical content:

$H^T$5

Hence $H^T$6 is a partition of regions into equivalence classes that the theory cannot tell apart empirically.

This construction shifts the unit of comparison from propositions or models to empirically discriminable regions. A plausible implication is that supervenience between theories, so understood, is weaker than full theoretical reduction but stronger than mere coexistence, because it requires preservation of every empirical distinction made by the supervening theory.

2. Formal definition and equivalent characterizations

With empirical structures in place, theory supervenience is defined by the relation

$H^T$7

read as $H^T$8 supervenes on $H^T$9, when any of three equivalent conditions holds (Gyenis, 26 Jul 2025).

The first condition is:

$EA^T(h,R)$0

If $EA^T(h,R)$1 can empirically distinguish two regions, then $EA^T(h,R)$2 can distinguish them as well.

The second condition is the negated counterpart:

$EA^T(h,R)$3

There is no pair of regions that $EA^T(h,R)$4 tells apart but $EA^T(h,R)$5 treats alike.

The third condition states that the partition $EA^T(h,R)$6 refines the partition $EA^T(h,R)$7:

$EA^T(h,R)$8

Thus every equivalence class of regions according to $EA^T(h,R)$9 lies inside a single equivalence class according to $R \subseteq M$0.

Gyenis also defines two related relations. If both $R \subseteq M$1 and $R \subseteq M$2, then $R \subseteq M$3, meaning that the two theories share exactly the same empirical structure. If $R \subseteq M$4 but $R \subseteq M$5, then $R \subseteq M$6, called strong supervenience. The supervenience relation $R \subseteq M$7, so defined, is reflexive and transitive (Gyenis, 26 Jul 2025).

These equivalent formulations make clear that theory supervenience is a refinement relation on empirical partitions. It therefore captures empirical dispensability: once a second theory preserves all of the first theory’s empirical distinctions, the first contributes no irreducible empirical structure of its own.

3. Canonical examples and strong supervenience

Gyenis illustrates the definition through both diachronic and synchronic examples (Gyenis, 26 Jul 2025). In the diachronic case, Newtonian gravitation supervenes on General Relativity:

$R \subseteq M$8

Any two regions that Newton distinguishes are likewise distinguished by General Relativity. The converse fails, since General Relativity distinguishes, for example, perihelion-precession data that Newton treats as identical.

In the synchronic case, Classical Thermodynamics strongly supervenes on Statistical Mechanics:

$R \subseteq M$9

No two gas-dynamics histories can differ in pressure or temperature unless the micro-states of Statistical Mechanics differ, and Statistical Mechanics refines the empirical partition of Classical Thermodynamics strictly.

Gyenis also formulates a special-science application. If computer science is taken as the empirical theory of computational systems, then empirical-structure physicalism asserts that current special sciences supervene on current physics:

$h \in H^T$0

Relation	Status	Basis stated in the source
$h \in H^T$1	Supervenience	GR preserves Newtonian empirical distinctions
$h \in H^T$2	Strong supervenience	SM strictly refines CT’s empirical partition
Special sciences $h \in H^T$3 current physics	Physicalist thesis	Physics preserves the empirical distinctions of the special sciences

The examples show that the framework is intentionally neutral with respect to ontology. Newtonian gravitation and General Relativity can differ in ontology, and Classical Thermodynamics and Statistical Mechanics can differ in descriptive granularity, while the supervenience verdict turns only on empirical structure. This suggests that the framework is designed to track preservation of empirical discriminatory power rather than derivability or entity-by-entity identification.

4. Physicalism, future physics, and Hempel’s dilemma

The principal philosophical application of theory supervenience in Gyenis’s account is empirical structure physicalism, defined as the thesis that the current special sciences supervene both on current and on future physics (Gyenis, 26 Jul 2025). The argument proceeds through three claims. First, current special sciences supervene on current physics: $h \in H^T$4 Second, current physics supervenes on future physics: $h \in H^T$5 By transitivity, one obtains

$h \in H^T$6

This is presented as a way to avoid both horns of Hempel’s dilemma. The framework does not identify “the physical” with a finished ontology of present-day physics, nor does it rely on an indeterminate ideal future theory in an unrestricted metaphysical sense. Instead, it trades metaphysical reduction of entities or languages for empirical reduction of distinctions. On this basis, Gyenis argues that empirical structure physicalism is appropriately labeled as a type of physicalism and that it is compatible with multiple realization (Gyenis, 26 Jul 2025).

A further consequence stated in the abstract is that mental theories remain empirically dispensable in the future (Gyenis, 26 Jul 2025). In the empirical-structural idiom, this means that if future physics refines all empirical distinctions made by present mental theories, then those theories contribute no autonomous empirical partition not already preserved by physics. The paper also briefly addresses the knowledge argument to illustrate the plausibility of empirical structure physicalism.

5. Realism and the optimistic meta-induction

Theory supervenience also underwrites empirical structure realism, the thesis that earlier theories of physics supervene on later theories of physics (Gyenis, 26 Jul 2025). The motivating pattern is historical: no successor physics has ever lost an empirical distinction of its predecessor. Gyenis therefore proposes an optimistic meta-induction from past theory change to the claim that current physics will supervene on future physics.

The crucial point is that this optimistic meta-induction is presented as compatible with the pessimistic meta-induction. The reason is that supervenience is cast as a very weak structural relation. One can hold that theoretical ontologies change radically across scientific revolutions while also maintaining that the empirical distinctions of older theories are retained within later ones. The source states this contrast explicitly: there is no contradiction in “losing” ontology while “saving” empirical structure (Gyenis, 26 Jul 2025).

This use of theory supervenience yields a structurally minimalist realism. The realist commitment is not to the survival of specific posits such as caloric fluid or luminiferous aether, but to the preservation of empirical structure across theory succession. A plausible implication is that the framework is meant to stabilize realism at the level of preserved empirical discrimination while leaving open stronger metaphysical disputes about reference and ontology.

6. Relation to wider supervenience theory

The expression “theory supervenience” also appears in a related but distinct framework developed by Caulton, where supervenience is defined over a space of kinematical possibilities rather than over empirical partitions of regions (Caulton, 2024). If a theory’s ontology is given as a set $h \in H^T$7 of degrees of freedom with allowed values $h \in H^T$8, then its kinematical possibility space is

$h \in H^T$9

For collections of properties $EA^T(h,R)$0 and $EA^T(h,R)$1 on $EA^T(h,R)$2, supervenience is defined by

$EA^T(h,R)$3

Caulton presents this as a sufficiently demanding notion of supervenience to sustain plausible claims of inter-theoretic reduction and theoretical equivalence (Caulton, 2024).

The broader literature represented in the sources places these theory-level notions within long-standing debates about reduction and definability. Butterfield characterizes supervenience as a weakening of reduction, allowing infinitely long definitions, and argues that emergence is logically independent both of reduction and of supervenience (Butterfield, 2011). In a closely related formulation, local supervenience is given by

$EA^T(h,R)$4

while global supervenience is formulated model-theoretically through preservation under isomorphism (Butterfield, 2011). Beth’s theorem is then invoked to show that, in first-order finitary extensional settings, implicit definability and explicit definability coincide (Butterfield, 2011).

Taken together, these frameworks show that “theory supervenience” is not a single uniform doctrine. In Gyenis’s formulation, it is a new reductive relationship between theories defined by refinement of empirical structure (Gyenis, 26 Jul 2025). In Caulton’s formulation, it functions as a supervenience criterion over kinematical possibilities that supports inter-theoretic reduction and theoretical equivalence (Caulton, 2024). The shared core is the same difference-making clause: no difference at the higher or coarser level without a difference at the lower or finer level. The principal divergence lies in what counts as the relevant base—empirical content over spacetime regions, or ontology-generated kinematical possibility spaces.