Marginal Realism: A Restricted Realist View

• Marginal realism is a form of restricted realism that affirms an independent reality only under explicitly limited conditions, diverging from full-blown classical realism.
• It spans diverse fields—quantum mechanics, dark matter studies, and temporal ontology—where it imposes operational or semantic constraints to traditional realist positions.
• The concept underpins practical frameworks in information theory and cosmology, offering measurable constraints and guiding debates on the scope of ontological commitment.

Searching arXiv for papers relevant to "marginal realism" and adjacent realism frameworks. Marginal realism denotes a restricted form of realist commitment in which reality is affirmed, but only under explicitly limited conditions. In the surveyed literature, the term does not function as a single canonical doctrine. In information theory it has a precise distributional meaning, requiring the reconstruction to match the source marginal law; in philosophy of cosmology it designates a weak or provisional realism justified under scarce empirical evidence; in quantum foundations and temporal ontology it is a looser label for positions that preserve realism while rejecting classical assumptions such as separability, non-contextuality, intrinsic properties, or a global present (Hamdi et al., 20 Jul 2025, Allzén, 19 Nov 2025, Karakostas, 2012, Rovelli, 2019). A unifying theme is that realism survives, but only after the scope of ontological or epistemic commitment has been narrowed.

1. Terminological scope and conceptual profile

The literature does not present marginal realism as a stable, universally recognized term. One paper explicitly states that it is not a standard named concept in the way that “local realism” or “macroscopic realism” are used in quantum foundations, and another states that it does not use “Marginal Realism” as a technical label at all (Fucci et al., 2024, Dawid et al., 18 Dec 2025). This suggests a family-resemblance concept rather than a single doctrine: realism is retained, but only at a constrained level of reference, context, ontology, or statistical structure.

Domain	Sense of realism retained	Principal restriction
Quantum foundations	Mind-independent reality	Properties are context-dependent
Dark matter realism	Weak or provisional commitment	Canonical referential success is not secured
Temporal ontology	Reality of becoming	No global, objective present
Rate-distortion theory	Output realism at the marginal level	Joint source-side-information structure need not be preserved

Across these uses, marginal realism differs from anti-realism because it does not deny a mind-independent world, and it differs from classical realism because it denies that reality is fully accessible through intrinsic, context-free, or uniquely reference-fixing structure. In that sense, the term marks a contraction of the classical realist package rather than its abandonment.

2. Quantum contextualization and operational realism

In quantum foundations, the closest analogue to marginal realism is the contextual realism defended in “Realism and Objectivism in Quantum Mechanics” (Karakostas, 2012). The paper argues that standard quantum mechanics can, and indeed should, be understood as a realist theory within its domain of application, but only if one accepts the “abandonment or radical revision of the classical conception of physical reality.” What is rejected is the classical package comprising separability, definite values, non-contextuality, intertemporal individuality, reductionism, strict subject-object separation, and the idea that the world is fully describable as a collection of self-contained individuals. The quantum world is instead treated as a non-separable whole in which parts acquire determinate descriptions only relative to an experimental context. On this view, quantum objects are carriers of dispositional or potential properties, and objectivity is preserved not as context-independence but as intersubjective validity once the conditions of observation and disentanglement are fixed.

The operationalization of restricted realism is developed further in “Quantifying continuous-variable realism” (1904.02490). There, position and momentum are treated as operationally discretized observables, so that the Bilobran–Angelo criterion can be extended from finite-dimensional settings to continuous variables. For an observable $A$, realism is expressed by invariance under an unrevealed projective measurement,

$\Phi_A(\rho)=\rho,$

and the corresponding irreality measure is

$\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$

For the discretized canonical pair, the framework yields the position-momentum irreality uncertainty relation

$$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$

which is interpreted as showing that quantum mechanics forbids both position and momentum from being simultaneously elements of reality. Here the restriction is quantitative rather than merely philosophical: classical realism fails for conjugate observables because simultaneous zero irreality is excluded.

A more abstract generalization appears in “Theory-Independent Realism” (Fucci et al., 2024). Using GPTs, the paper defines realism for a property $\mathcal Y$ by the requirement that an unrevealed measurement of $\mathcal Y$ leave the outcome probabilities of every subsequent property $\mathcal X$ unchanged: $$p_\epsilon(x_i)=p_{\Phi_\mathcal{Y}(\epsilon)}(x_i)\qquad \left(\forall\,\mathcal{X}=\{x_i\}_{i=1}^d\right).$$ The paper explicitly states that any connection to “marginal” ideas is only indirect, but structurally the criterion is an invariance condition on outcome distributions. In this sense, the quantum and GPT literatures support a restricted realism grounded not in classical intrinsic properties but in contextual stability, operational invariance, and measurement-relative objectivity.

3. Semantics, underdetermination, and cosmological posits

In the philosophy of dark matter, marginal realism is used more explicitly to designate a realism that is weaker than canonical scientific realism. “Dark Matter Realism: How Referential Semantics Restricts Realism in Contemporary Fundamental Physics” argues that a Psillos-style causal-descriptive semantics is a poor fit for dark matter because it presupposes well-confirmed natural-kind structure and fine-grained empirical specificity that are not presently available (Allzén, 19 Nov 2025). The target proposal is the semantic package

$$\text{SDMR: } \text{dark matter refers to the entity possessing } (i)\text{ NB},\ (ii)\text{ EN},\ (iii)\text{ GR},\ (iv)\text{ CF},$$

where the proposed kind-constitutive properties are non-baryonic, electromagnetically neutral, gravitationally interactive, and acting like a collisionless fluid.

The paper’s criticism is twofold. First, the semantics relies on hidden metaphysics: intrinsicness, natural-kind essentialism, objective kind boundaries, and a sufficiency/necessity structure linking core causal description to reference. Second, the proposed descriptors are ill-suited to that role. “Non-baryonic” and “electromagnetically neutral” are exclusionary; “gravitationally interactive” is too general; and “collisionless fluid” is relational and emergent rather than intrinsic. The resulting extension is both too broad and too narrow. It is too broad because standard model neutrinos satisfy the proposed package while failing to play the relevant causal-explanatory role in $\phi_{\mathrm{dark}}$; it is too narrow because it excludes live candidates such as self-interacting dark matter, dissipative or double-disk dark matter, mirror dark matter, superfluid dark matter, and SIMPs.

The paper formalizes this semantic failure by distinguishing two mistake types: $\Phi_A(\rho)=\rho,$0 and

$\Phi_A(\rho)=\rho,$1

These formulas express, respectively, that the proposed conditions are not necessary and not sufficient for reference. The broader conclusion is that in contexts of sparse evidence one may reasonably adopt only a pragmatic or provisional realism: dark matter may well be real, but the canonical scientific-realist machinery does not yet secure full referential success. Here marginal realism is a realism of epistemic restraint, driven by underdetermination and the determinable/determinate gap.

4. Temporal ontology and local becoming

In the metaphysics of time, the closest counterpart to marginal realism is the position developed in “Neither Presentism nor Eternalism” (Rovelli, 2019). The paper argues that Presentism and Eternalism form a false dilemma. Presentism fails because relativity rules out a universal, observer-independent present; Eternalism is rejected because it treats becoming as unreal and interprets four-dimensional spacetime through an overly static metaphysics. The proposed alternative is that reality has a richer temporal structure in which becoming is real, but only locally and without a globally privileged present.

The paper states that there is no global notion of present, but there is a local becoming at every point of spacetime. The set $\Phi_A(\rho)=\rho,$2 of events at spacelike distance from an event $\Phi_A(\rho)=\rho,$3 may be called the “extended present” of $\Phi_A(\rho)=\rho,$4, but it cannot serve as a genuine global present because it contains events in one another’s future. By contrast, local notions of present remain well defined in context and approximation, including Einstein simultaneity relative to a worldline, a “bubble present” of radius

$\Phi_A(\rho)=\rho,$5

and a “diamond present” for extended events. The paper also emphasizes that fundamental becoming is “local and unoriented”: directionality is not built into the basic temporal structure in the form of a universal flow.

If this position is described as marginal realism, that characterization is loose rather than technical. The realist commitment is to temporal becoming, but only under relativistically local conditions. The view is therefore realist about temporal structure while anti-realist about a universal present. Its restriction is not epistemic in the dark-matter sense and not statistical in the information-theoretic sense; it is geometric and causal.

5. Strong realism constraints in rate-distortion theory

The most explicit technical use of the term appears in “Rate-Distortion-Perception Trade-off with Strong Realism Constraints: Role of Side Information and Common Randomness” (Hamdi et al., 20 Jul 2025). There, marginal realism is a distributional constraint in lossy compression with side information. Perfect marginal realism requires, for all large enough blocklengths $\Phi_A(\rho)=\rho,$6,

$\Phi_A(\rho)=\rho,$7

so the reconstruction sequence has exactly the same marginal distribution as the source sequence. Near-perfect marginal realism weakens this to

$\Phi_A(\rho)=\rho,$8

The paper distinguishes this from joint realism, which requires

$\Phi_A(\rho)=\rho,$9

Only the output alone must “look like” the source under marginal realism; under joint realism the reconstruction must also preserve the statistical relation to the side information.

Realism notion	Formal constraint	Strength
Perfect marginal realism	$\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$0	Exact marginal matching
Near-perfect marginal realism	$\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$1	Asymptotic marginal matching
Joint realism	$\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$2	Preserves joint law

The system model considers memoryless $\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$3 with two side-information settings: $\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$4 only at the decoder, and $\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$5 at both encoder and decoder. In the E-D setting, the achievable region for marginal realism is characterized by joint laws $\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$6 satisfying

$\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$7

together with

$\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$8

$\mathfrak{I}(A|\rho)=S(\Phi_A(\rho))-S(\rho).$9

$$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$0

The term $$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$1 captures the dual role of side information: it helps compression and can also contribute randomness for synthesizing realistic outputs. In the D setting, the paper gives an inner bound with

$$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$2

$$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$3

$$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$4

$$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$5

This technical notion of marginal realism is not merely terminological. It changes the Wyner–Ziv problem into an overview-constrained rate-distortion problem. The paper shows that near-perfect realism and perfect realism are equivalent under a mild uniform integrability condition, and that with sufficiently large common randomness rate one obtains exact single-letter characterizations. For jointly Gaussian $$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$6 with correlation $$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$7 and squared-error distortion, the paper proves that

$$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$8

provided common randomness is available at a sufficiently large rate. The main restriction here is therefore precise and operational: realism is imposed only at the level of the output marginal, not necessarily at the joint level.

6. Comparative interpretation and related realist positions

The surveyed uses of marginal realism are non-equivalent, but they converge on a shared pattern. Realism is preserved, while at least one classical demand is denied: intrinsic context-free properties in quantum mechanics, unique reference-fixing essence in dark matter semantics, a universal present in temporal ontology, or unrestricted freedom to ignore distributional realism in compression theory. A plausible implication is that marginal realism functions as a general strategy for retaining ontological commitment under conditions where full classical realism is blocked.

This restrictedness should not be confused with instrumentalism. The contextual realist interpretation of quantum mechanics explicitly insists on a mind-independent world, even while denying access to “reality in itself” within the quantum domain (Karakostas, 2012). The dark-matter case likewise does not say that dark matter is unreal; it says only that full-blown realism justified by canonical scientific-realist standards is not yet licensed (Allzén, 19 Nov 2025). Rovelli’s temporal view does not demote becoming to appearance; it treats becoming as real, but local (Rovelli, 2019). In the information-theoretic setting, marginal realism is not a metaphysical weakening but a mathematically exact constraint on $$\mathfrak{I}(Q|\rho)+\mathfrak{I}(P|\rho)\ge \ln(2\pi e),$$9 (Hamdi et al., 20 Jul 2025).

The term also intersects, without coinciding, with other realism programs. “Theory-Independent Realism” treats realism as invariance of outcome probabilities under unrevealed measurement and explicitly notes that any connection to “marginal” notions is only indirect (Fucci et al., 2024). “Realism and Ontology in Quantum Mechanics and String Theory” separates realist commitment from ontological commitment, arguing that realism should attach to a theory’s full formal structure while ontology should be tied to specific empirical contexts via an “observer-based ontology” (Dawid et al., 18 Dec 2025). That position does not employ marginal realism as a technical label, but it is consonant with a broader tendency in the literature: robust commitment at one level of theory, restricted commitment at the level of ontology, reference, or empirical access.

In this comparative sense, marginal realism is best understood not as a single thesis but as a recurrent form of qualified realism. It marks positions that affirm reality while denying that realism requires exhaustive, context-free, globally privileged, or uniquely essence-fixing access to what is real.