Supervenience (Philosophy)

• Supervenience is a property-based dependence relation where a family A supervenes on family B if no difference in A occurs without a difference in B, highlighting key characteristics like asymmetry and various formulations such as explicit versus implicit definability, strong versus weak supervenience.
• Supervenience plays a crucial role in logical formulations, the quantum mechanical phenomenon of entanglement, physical theories like Boltzmannian statistical mechanics, and debates over emergence and reduction; applications span counterfactual dependence, nomological necessity, categorical versus structural understanding, and scientific representations.
• Examples of supervenience include the global nature of determinate response predicates in empirical systems, use in logical systems, and different interpretations based on variations, dependence, and determinism in various frameworks.

Supervenience is a dependence relation between families of properties, facts, states, or structures. A family $A$ supervenes on a family $B$ when there can be no difference in $A$ without some difference in $B$; equivalently, once the $B$-facts are fixed, the $A$-facts are fixed as well. In Lewis’s formulation, “there can be no difference in respect of so-and-so without difference in respect of such-and-such.” Supervenience is directional, and its strength depends on the relevant base, target, possibility space, language, and model class. It is distinct from identity, causation, explanation, reduction, and logical implication, although particular formulations connect it with each of these notions.

1. Definition, structure, and varieties

The standard property-based formulation considers a domain of objects $O$ and two families of properties, $B$ and $T$. $T$ supervenes on $B$0 if any two objects that match in every $B$1-property also match in every $B$2-property. In contrapositive form:

$B$3

The corresponding determination formulation is:

$B$4

Determination makes explicit that the $B$5-description is sufficient to fix the $B$6-description. Supervenience itself does not entail that the $B$7-properties are identical with the $B$8-properties, that they are finitely definable, or that the base explains them.

Supervenience may be local, concerning properties of individual objects, or global, concerning entire worlds, structures, or model classes. Global supervenience compares possible worlds or interpretations: if two worlds agree in their lower-level descriptions, including relevant relations, they must agree in their higher-level descriptions. A model-theoretic formulation divides a language into a lower-level vocabulary $B$9 and a higher-level vocabulary $A$0. Explicit definability requires each $A$1-predicate to be definable by a formula using only $A$2-predicates. Implicit definability requires that any two interpretations agreeing on $A$3 also agree on $A$4.

Supervenience is asymmetric. If $A$5 determines $A$6, $A$7 need not determine $A$8. A conditional such as $A$9 does not normally express that $B$0 determines $B$1, because both $B$2 and $B$3 remain possible. A biconditional $B$4 imposes mutual determination, but is stronger and symmetric. Supervenience is instead a cross-object or cross-world condition concerning patterns of variation.

Several distinctions concern the strength of the dependence:

• Weak supervenience: sameness of the base within one world guarantees sameness of the higher-level properties.
• Strong supervenience: the dependence is stable across possible worlds.
• Global supervenience: the total distribution or pattern of higher-level facts is fixed by the total lower-level structure.
• Strong non-supervenience: a relation is neither dependent upon nor determined by non-relational properties of its relata.
• Weak non-supervenience: a relation depends on non-relational properties but is not determined by them.

The distinction between dependence and determination is important. A relation may require suitable properties of its relata without being fixed by those properties. This is the sense in which quantum entanglement can be weakly non-supervenient under a density-operator description, while it is argued to be strongly non-supervenient under pure-state assignment (Karakostas, 2009).

Supervenience also depends on the specified possibility space. A higher-level property may supervene over a narrow class of dynamically permitted models while failing to supervene over a broader class of kinematically possible models. Caulton argues that supervenience relevant to reduction and theoretical equivalence should be assessed over sufficiently broad kinematical possibilities generated by a theory’s ontology, rather than merely over solutions selected by its dynamical laws (Caulton, 2024).

2. Supervenience, reduction, and emergence

Supervenience is often described as weaker than Nagelian reduction. Reduction requires deduction using explicit definitions or bridge principles:

$B$5

where $B$6 is the basic theory, $B$7 the higher-level theory, and $B$8 a set of definitions. Supervenience is traditionally regarded as allowing infinitely long definitions: the higher-level property may be determined by an exhaustive disjunction of lower-level descriptions without possessing a short or manageable definition.

Butterfield challenges the assumption that this distinction automatically yields “emergence as supervenience without reduction.” In a first-order, finitary, extensional language, when the relevant class of interpretations is a theory-class, Beth’s theorem establishes the equivalence of implicit and explicit definability:

$B$9

Thus, under those assumptions, global supervenience collapses into explicit definability. The resulting definition may be extremely long, heterogeneous, or practically unusable, but Beth’s theorem says that a finite first-order definition exists. The collapse depends on first-order finitary logic, extensionality, and the model class being the class of models of a theory. It may fail for higher-order or infinitary languages, or when the scientifically relevant class of structures is not a theory-class (Butterfield, 2011).

Butterfield defines emergence as behavior that is novel and robust relative to a comparison class, while treating reduction essentially as deduction. The two notions are logically independent. Emergence may occur with reduction when deduction is performed after taking a limit, such as $B$0; it may also occur without reduction. Conversely, supervenience may occur without emergence when it records a dependence that is neither novel nor robust. His examples include arbitrary functions, fractals, quantum superselection, and phase transitions (Butterfield, 2011).

In these examples, supervenience is often true but explanatorily weak. It does not specify the structure of the dependence, the relevant limiting process, the organization of possible realizations, or the source of robustness. In the method of arbitrary functions, for example, equiprobability emerges because increasingly intricate partitions wash out the peaks and troughs of an unknown density. The statement that the resulting probability supervenes on the microscopic details does not explain this limiting mechanism. Similar limitations arise for fractal dimensions, superselection sectors, and phase transitions.

Multiple realizability does not by itself refute reduction. A higher-level property may have heterogeneous lower-level realizations, represented schematically as:

$B$1

Butterfield argues that a disjunctive definition can still support deduction. Multiple realization therefore does not establish either emergence or irreducibility without further assumptions. The classification of a phenomenon depends on the lower-level vocabulary, the comparison class, and the model space.

3. Quantum nonseparability and Humean supervenience

Humean supervenience is the thesis that the world is a “vast mosaic of local matters of particular fact.” Its basis consists of intrinsic qualitative properties instantiated at spacetime points or regions, together with spatiotemporal relations among them. Laws, chances, causal relations, global properties, and possibly mental or modal facts are supposed to supervene on this basis.

The doctrine incorporates the separability principle of classical physics:

Each physical system possesses its own distinct, separate state, and the whole state of a compound system is completely determined by the separate states of its subsystems and their spatiotemporal relations.

In classical particle mechanics, a system of $B$2 particles is represented by positions and momenta, or by an $B$3-tuple of subsystem states. Global quantities such as total momentum, angular momentum, kinetic energy, center-of-mass motion, and gravitational potential energy are functions or sums of quantities associated with the parts. Classical field theory likewise assigns field values at spacetime points, so the field in a region is determined by its values throughout that region.

Quantum mechanics does not generally preserve this structure. For a composite system,

$B$4

and a pure state may have Schmidt form

$B$5

A product state has only one nonzero Schmidt coefficient:

$B$6

When at least two coefficients are nonzero, the state is entangled and cannot be represented as a tensor product of independently specified pure states. For an entangled system, the composite has a definite pure state, while the components do not possess corresponding individual pure states within that state.

The singlet state illustrates the point:

$B$7

The pair has total spin zero and exhibits perfect anticorrelation, but neither particle has a separately definite spin value in the singlet state. The global property “having total spin zero” is therefore not constructed from independently possessed local spin properties.

Density operators alter, but do not remove, the problem. The reduced states of the singlet are:

$B$8

with both subsystems maximally mixed. Yet the pair of reduced density operators does not determine the global state:

$B$9

A singlet and a suitable triplet can have identical reduced states and spatial separation while differing in their joint properties. The local states satisfy dependence, because the subsystems possess reduced density operators, but fail determination, because those operators do not fix the composite state. This motivates the classification of entanglement as weakly non-supervenient under density-operator assignment and strongly non-supervenient under pure-state assignment (Karakostas, 2009).

The empirical significance of Bell correlations reinforces the structural point. Entangled states violate Bell inequalities, and such violations have been experimentally observed, including over large distances. The claim is not that entanglement enables faster-than-light signaling, but that the correlations are not products of independently existing local states connected only by ordinary spatiotemporal relations.

Bohmian mechanics offers a qualified defense of Humean supervenience. Its fundamental ontology consists of particles with definite positions and trajectories, while the universal wave function may be treated as nomological within a Humean Best System. A subsystem’s effective wave function depends on its configuration relative to the environment. This can preserve ordinary Separability if subsystem quantum states are denied intrinsic status. However, such a strategy may violate Strong Supervenience, according to which the complete physical state of a system supervenes on the intrinsic states and spatiotemporal relations of its parts. The effective state of a subsystem can depend on external configuration, producing a necessary connection between the subsystem and its environment (Bigaj et al., 2020).

4. Supervenience in physical theory and scientific representation

Supervenience plays a central role in debates over intertheoretic reduction. If thermodynamic quantities are to reduce to Boltzmannian statistical mechanics, thermodynamic differences must be reflected in differences in the relevant Boltzmannian description. A reduction requires, at minimum, that the reduced properties supervene on the properties of the reducing theory.

The standard Boltzmannian framework partitions a finite available phase space into sharp, fixed macroregions:

$A$0

Every microstate belongs to exactly one macroregion. The corresponding macrovalues are therefore piecewise constant along a continuous microstate trajectory. Thermodynamic temperature, by contrast, can change continuously. For an ideal monatomic gas,

$A$1

so an arbitrarily small change in a particle velocity can change the thermodynamic temperature while leaving the microstate inside the same BSM macroregion. Consequently,

$A$2

A single BSM macrovalue can correspond to multiple thermodynamic macrovalues. This is not the usual form of multiple realizability, in which many lower-level states realize one higher-level property; here, one BSM value is compatible with many thermodynamic values. The issue is coarse-grained underdetermination, not merely a boundary artifact. Refining the finite partition does not solve it so long as each region contains multiple microstates (Ehmann, 2022).

Caulton develops a broader criterion for reduction and theoretical equivalence. A physical theory should be represented by both a category of kinematically possible models and a subcategory of dynamically possible models:

$A$3

Two theories may agree on dynamical solutions while differing in their kinematical possibility spaces. Generalized holonomy and principal connections, or Faraday-tensor and potential formulations of electromagnetism, may be dynamically related while failing Caulton’s stronger kinematical supervenience criterion. In contrast, standard and geometrized Newtonian gravitation may preserve the same possibilities once their structures are appropriately reconstructed (Caulton, 2024).

If spacetime is emergent, the traditional Humean mosaic becomes problematic. Humeanism requires a basis of non-modal facts, while laws are selected by a Best System analysis. If the basis is traditionally a spatiotemporal arrangement of local qualities, emergent spacetime threatens the basis itself, not merely the subsequent analysis of lawhood. Possible replacements include non-spatiotemporal glueing relations, configuration or Hilbert spaces, causal structures, and entanglement. Each raises questions about whether the proposed basis remains categorical and free of necessary connections (Lam et al., 2021).

Caulton’s Humean dictum connects ontology with independent variation. An ontology determines a space of kinematical possibilities by allowing its distinct degrees of freedom to vary independently, subject to internal constitutive restrictions. Supervenience is then assessed over this possibility space rather than merely over actual or dynamically allowed models. The approach is intended to support intensional semantics, intertheoretic reduction, and theoretical equivalence without forcing physical theories into first-order languages (Caulton, 2024).

5. Supervenience in logic, sorites, and probabilistic dependence

Supervenience has been formalized as a logical operator. A dyadic operator $A$4 can be read as “$A$5 supervenes on $A$6.” In a global semantics,

$A$7

iff every pair of worlds agreeing on $A$8 also agrees on $A$9:

$O$0

The operator is reflexive and transitive but not symmetric. The agreement operator is recovered as:

$O$1

The dyadic supervenience logic is strictly more expressive than the corresponding agreement logic. It can express conditional agreement—agreement on $O$2 entails agreement on $O$3—rather than merely global agreement on $O$4. A sound Hilbert system is provided, while strong completeness for the dyadic system remains open (Fan, 2016).

In inquisitive first-order modal logic, global supervenience is expressed through questions about predicate extensions:

$O$5

This formula says that throughout the relevant successor worlds, information sufficient to settle which individuals are $O$6 also suffices to settle which individuals are $O$7. The logic evaluates formulas relative to information states or teams rather than individual worlds. A complete labelled sequent calculus provides soundness, completeness, strong completeness, invertibility, and cut admissibility for the resulting system (Ciardelli et al., 30 Jun 2026).

Supervenience also structures formal analyses of sorites. Let a system be:

$O$8

where $O$9 assigns each stimulus a determinate response. Deterministic supervenience ensures that response predicates are classical. Tolerance requires $B$0 to be locally constant, while connectedness requires finite chains of overlapping vicinities connecting stimuli with different responses. Deterministic supervenience, tolerance, and connectedness cannot all hold. The system must fail tolerance at some boundary or fail connectedness.

Empirical systems may instead exhibit probabilistic supervenience. A stimulus $B$1 is mapped to a probability measure over responses:

$B$2

In the binary case, the response probability is $B$3, with probabilities $B$4 and $B$5. Higher-order probability distributions reduce to ordinary probabilities by integration:

$B$6

Thus, a distribution over probability measures yields an ordinary measure over responses. Probabilistic supervenience avoids the soritical contradiction because the response distribution need not be locally constant (Dzhafarov et al., 2015).

The reverse-mathematical analysis distinguishes two formalizations of the sorites. The implication from tolerance to non-connectedness in the vicinity space induced by $B$7 is provable in $B$8. The implication from connectedness to failure of tolerance for arbitrary vicinity spaces is equivalent over $B$9 to $T$0. The difference arises from the representation and quantifier structure of the vicinity systems, not from an inconsistency in supervenience itself (Dzhafarov, 2016).

A recent stochastic formulation generalizes deterministic supervenience by replacing a function $T$1 with a Markov kernel:

$T$2

Each base state determines a probability measure over higher-level configurations:

$T$3

Deterministic supervenience is recovered as the Dirac case:

$T$4

The framework treats distributional fixation as the basic relation and uses normalized mutual information, divergence spectra, tail sensitivity, and intervention-sensitive measures as diagnostics of structured stochasticity and distributional multiple realization (Zhang, 20 Dec 2025).

6. Supervenience, causation, mind, and agency

Supervenience does not by itself entail causation. This point is central to debates over mental causation. A mental state may depend on a physical state without that dependence being causal. The Dual-Laws Model distinguishes physical determinism from physical causal closure: physical determinism concerns lawful evolution in the Physical Stance, whereas causal closure is an interventionist claim belonging to the Causal Stance.

The model retains the supervenience principle that mental differences require physical differences:

$T$5

It nevertheless proposes that higher-level causal efficacy can arise through a distinction between supervenient functions and the index sequences that select or compose them. The supervenient functions remain physically realized, but an index sequence $T$6 determines which expression is active. A selected expression produces a feedback error:

$T$7

where $T$8 is a lower-level physical state. Lower-level dynamics reduce the error, while higher-level dynamics alter $T$9:

$T$0

and lower-level dynamics respond to the resulting error. The proposed causal chain is:

$T$1

This model attempts to preserve physical implementation while assigning independent dynamics to the higher level. Its unresolved issue is whether the higher-level law $T$2 is genuinely independent or merely a compact redescription of lower-level mechanisms (Ohmura et al., 6 Jan 2026, Ohmura et al., 12 Feb 2026).

Quantum approaches to consciousness use the failure of classical local supervenience to challenge combination arguments against panpsychism and panprotopsychism. Classical supervenience assumes that the properties of a whole are functions of the properties and relations of independently characterized parts:

$T$3

Entangled quantum systems violate this compositional picture. Their Hilbert space is a tensor product,

$T$4

but generic states and projectors are not products of subsystem states and projectors. For $T$5 two-level systems, the state space has dimension $T$6, providing an exponentially expanding range of collective states and properties. On this basis, quantum panprotopsychism proposes that phenomenal aspects may be associated with quantum states and events whose properties are holistic rather than reducible to local classical properties (Gambini et al., 2024, Gambini et al., 13 May 2025).

These proposals do not establish a psychophysical mapping from particular quantum states to particular phenomenal qualities. They instead argue that quantum nonseparability removes a classical constraint: the whole need not be assembled from independently determinate microproperties or microsubjects. Consciousness, self-consciousness, and personal identity are proposed to involve entangled quantum systems coupled to neural organization, but the neural implementation, decoherence problem, subject combination problem, and status of phenomenal causation remain unresolved.

Supervenience is therefore compatible with several positions:

• Reductive physicalism, if the higher-level facts are identified with or finitely derived from physical facts.
• Non-reductive physicalism, if higher-level facts depend on physical facts but retain distinct predicates or explanatory roles.
• Humeanism, if laws and modal facts supervene on a suitably specified non-modal base.
• Emergentism, if higher-level behavior is novel and robust, whether or not it reduces.
• Panprotopsychism, if phenomenal aspects are associated with fundamental states or events but are not reducible to local classical properties.
• Dual-laws models, if higher-level organizational dynamics alter lower-level evolution without violating physical implementation.

Across these applications, supervenience remains a dependence relation rather than a complete metaphysical theory. It constrains possible variation, but does not alone determine whether the dependence is explanatory, causal, reductive, nomological, probabilistic, or merely covariational. Its interpretation depends on what counts as the base, what possibility space is relevant, whether the determinandum is a value or a probability distribution, and whether the higher-level structure is represented by properties, relations, questions, functions, or dynamically selected organizational patterns.