Subadditivity

Version

v2 · 2026-08-30 · History

Prime #

1496

Origin domain

Mathematics

Subdomain

analysis and inequalities → Mathematics

Also from

Statistics & Experimental Design, Information Theory, Economics & Finance, Computer Science & Software Engineering

Aliases

Subadditive Property, Subadditive Inequality

Related primes

Function (Mapping), Semigroup, Order, Measure, Metric, Economies Of Scope, Synergy and Antagonism

Core Idea

Subadditivity is the one-sided composition law that evaluating a combined whole never produces more than the sum of evaluating its parts separately. In its familiar additive form, a function \(f\) is subadditive when

\[ f(x+y) \leq f(x)+f(y) \]

for every admissible pair \(x,y\). The plus sign is not essential. The domain can combine objects by vector addition, set union, concatenation, joint formation, or another associative operation. What matters is that the same two routes can be compared: combine first and evaluate once, or evaluate the pieces and add their values. Subadditivity says the first route cannot exceed the second.

The law is weaker than additivity and stronger than an occasional saving. Equality is allowed: an additive map sits on subadditivity's boundary. Strict inequality records a combination slack

\[ \Delta_f(x,y)=f(x)+f(y)-f(x\oplus y) \geq 0, \]

which may represent overlap, shared structure, dependence, reuse, or simply the looseness of a decomposed bound. The law itself does not choose among those mechanisms. It only fixes the sign: combination may leave the total unchanged or reduce it, but never make it exceed the separate sum. A single favorable pair is therefore not enough. The defining commitment is universal over the stated scope.

That thin law does substantial work. Repeated application gives \(f(x_1\oplus\cdots\oplus x_n)\leq\sum_i f(x_i)\), so a hard whole can be upper-bounded by any convenient decomposition. For a subadditive sequence \(a_{m+n}\leq a_m+a_n\), Fekete's lemma identifies a stable asymptotic rate, \(\lim_n a_n/n=\inf_n a_n/n\), and Hammersley and Kingman extend the limiting logic to broader and stochastic subadditive settings.^[1][2][3] Later semigroup results show that this rate reasoning is not confined to ordinary numerical sequences.^[4] The recurring reasoning pattern is thus not merely “the whole can be cheaper.” It is: verify a universal combination inequality; decompose to obtain safe upper bounds; measure the slack; and, under repetition, extract a long-run per-unit rate.

Structural Signature

Sig role-phrases:

• the composable carrier — objects that can be joined by a defined binary operation, such as addition, union, concatenation, or joint formation
• the combination operation — the route \(x\oplus y\) that constructs one whole from two parts and remains closed on the stated domain
• the evaluation map — a single-valued function \(f\) assigning each carrier object a comparable cost, size, uncertainty, distance-like magnitude, or value
• the ordered additive codomain — a comparison scale whose addition is associative and order-compatible (monotone in each argument), so \(f(x\oplus y)\) and \(f(x)+f(y)\) are meaningful and inequalities may be substituted into finite sums
• the whole-below-sum invariant — \(f(x\oplus y)\leq f(x)+f(y)\) for every admissible pair, not merely for selected examples
• the decomposition closure — iteration of the binary law yields \(f(\bigoplus_i x_i)\leq\sum_i f(x_i)\), making every valid partition a certified upper bound
• the slack-and-boundary diagnostic — \(\Delta_f(x,y)\geq0\) measures departure from additivity; zero is the additive boundary, a negative value is a counterexample

The minimal formal setting is a carrier with an associative operation and a value map into an ordered additive codomain. Associativity in the carrier lets a many-part whole be regrouped without changing which whole is evaluated. In the codomain, addition must be associative and compatible with the order—monotone in each argument—so one may substitute a two-part inequality into a finite sum and iterate the bound. Neither commutativity nor positivity is required by this bare signature. Continuity, convexity, concavity, homogeneity, and monotonicity of the evaluation map \(f\) are likewise additional commitments. Particular families add them: a norm adds non-negativity, definiteness, and absolute homogeneity; a probability measure adds normalization and countable additivity on disjoint events; an economic cost function adds an output space and a production interpretation.

Three recognition tests prevent identity inflation. First, same object test: the combined object on the left must be the same whole represented by the parts on the right, not a different outcome chosen after seeing costs. Second, same evaluation test: one map and one unit must evaluate both routes; comparing dollars on the left with utility points on the right is not subadditivity. Third, scope test: the inequality must hold throughout an explicitly stated domain or class. If it holds only for one bundle, one range, or one empirical sample, the result is local subadditivity or an observed saving, not a global subadditive function.

What It Is Not

• Not the subadditivity_effect. That domain-specific psychology entry describes a signed bias in elicited probability judgments: separately naming an exhaustive event partition recruits more cognitive support, so the summed judgments exceed the packed judgment. Mathematical subadditivity is a property of the evaluated map itself and requires no judging agent, packed/unpacked description, or support-theory mechanism.
• Not additivity. Additivity requires equality, \(f(x\oplus y)=f(x)+f(y)\). Subadditivity permits equality but also permits positive slack. Treating it as “approximately additive” erases the one-sided guarantee that makes it useful.
• Not concavity or convexity. Convexity compares values under weighted averages; subadditivity compares a combined input with the sum of separate values. Extra hypotheses can connect them, but neither property entails the other in general. A ceiling function on non-negative inputs is subadditive without being concave, and a convex function need not be subadditive.
• Not submodularity. Submodularity imposes a diminishing-marginal or lattice inequality, \(f(A)+f(B)\geq f(A\cup B)+f(A\cap B)\). A non-negative normalized submodular set function is subadditive, but subadditivity alone says nothing about intersections or how marginal values change as a set grows. Subadditive valuations are a strictly broader complement-free class in algorithmic economics.^[5]
• Not economies of scale or scope. Those are economically interpreted mechanisms that can produce subadditive cost over a stated output region. Subadditivity is the general inequality property; it does not by itself identify fixed-cost spreading, shared inputs, production technology, or market structure.
• Not synergy_and_antagonism. That prime compares an observed joint effect with a chosen no-interaction baseline and signs the residual in either direction. Subadditivity is a universal one-sided law for one value map. It includes the additive boundary and does not require causal interaction or a measured baseline residual.
• Not “the whole is less than the sum” as a one-off observation. One favorable pair demonstrates slack at that pair. A subadditive classification requires the inequality for every pair in the declared scope; one negative slack value refutes it there.

Broad Use

Normed geometry and analysis. Every norm obeys \(\|x+y\|\leq\|x\|+\|y\|\): the triangle inequality is exactly subadditivity of the magnitude map on vector addition. Together with a norm's absolute homogeneity, the triangle inequality makes every norm ball convex; by itself it also turns local component bounds into a bound on a vector sum.^[6] The domain-specific catalog entry norm adds vector-space structure, homogeneity, and definiteness; subadditivity isolates the one constituent law that also survives far outside vector spaces.

Probability and measure. For events, the union bound is \(P(A\cup B)\leq P(A)+P(B)\), extending to countable families. It is subadditivity under set union. The slack is the overlap probability in the two-event case, because inclusion–exclusion gives \(P(A)+P(B)-P(A\cup B)=P(A\cap B)\). MIT's probability notes state both the two-event and generalized union bound explicitly.^[7] Here the decomposition tactic travels literally: replace a difficult failure event by a union of simpler events, sum their probabilities, and obtain a safe—even if loose—risk bound.

Information theory. Joint Shannon entropy satisfies \(H(X,Y)\leq H(X)+H(Y)\), with equality exactly when the variables are independent. The slack is mutual information: \(H(X)+H(Y)-H(X,Y)=I(X;Y)\). Shannon introduced the governing entropy quantities, and Stanford's information-theory notes present the inequality, equality condition, and mutual-information identity together.^[8][9] Quantum information retains the same family resemblance at a deeper level: von Neumann entropy is subadditive, and Lieb and Ruskai's strong-subadditivity theorem constrains tripartite quantum systems.^[10] The carrier and evaluator differ radically from probability of unions, but the operative diagnostic—joint whole, separate marginals, non-negative slack—does not.

Industrial organization and production cost. A cost function is subadditive over an output scope when one producer can supply a combined output vector at no greater cost than any partition of that output among separate producers. Baumol made strict global cost subadditivity the cost criterion for natural monopoly and stressed that declining average cost and economies of scale are not equivalent to it in the multiproduct case.^[11] The same inequality also underlies the narrower economies_of_scope: jointly producing distinct products costs less than producing them separately. The prime-level law says what must be tested; the economic nodes add why the slack may arise and what institutional consequence follows.

Algorithmic economics and combinatorial allocation. A set valuation is subadditive when \(v(S\cup T)\leq v(S)+v(T)\): the bidder has no positive complementarity that makes the union worth more than the separate bundles. This class is central in welfare maximization and combinatorial-auction mechanism design; Feige studies approximation when utility functions are subadditive rather than additive or submodular.^[5] Again, the vocabulary travels intact: bundles combine by union, a valuation evaluates them, the union is the whole, and the separate values supply the upper bound.

Stochastic growth and spatial processes. Passage time across a long random medium can be bounded by the passage time across consecutive segments; optimal path cost from the beginning to the end cannot exceed the cost of concatenating segmentwise paths. Kingman's theorem was developed for such subadditive stochastic processes, generalizing law-of-large-numbers reasoning to cases where totals are not sums but are controlled from above by sums.^[3] The consequence is a stable long-run growth or time constant even when increments interact and ordinary additivity fails.

Across these uses, “subadditive” is not ornamental mathematical language. It carries the same falsifiable inequality, the same partition-and-sum upper-bound method, the same zero-versus-positive slack distinction, and the same closure under repeated composition.

Clarity

Subadditivity clarifies a common ambiguity in claims that “combining helps.” That phrase can mean at least four different things: a one-off saving, a decreasing average, a causal synergy, or a universal upper-bound law. The abstraction asks for the exact object and exact comparison. What is the carrier? How are two objects combined? What map evaluates both the whole and the pieces? Over which scope is the inequality universal? Once those four fields are filled, the claim becomes testable rather than rhetorical.

The sharp operational diagnostic is the slack \(\Delta_f(x,y)\). Positive slack means the separate-sum route overestimates the combined evaluation; zero marks an additive pair; negative slack is a witness that the map is not subadditive on the claimed scope. This one scalar is interpreted differently across fields—event overlap, mutual information, shared production cost, or bundle redundancy—but its structural meaning remains “how much the separate sum exceeds the whole.” The distinction between the invariant and the interpretation prevents a frequent error: inferring a specific mechanism merely from the sign of the inequality.

The abstraction also makes qualifiers visible. “Globally subadditive” means every admissible decomposition passes. “Subadditive at \(q\)” in cost analysis means every relevant partition of one output vector passes, not that the cost function is globally subadditive. “Approximately” or “quasi-subadditive” means an explicit error term is permitted. “Strict” means positive slack on all nontrivial combinations. These are not stylistic refinements: each changes which deductions are licensed.

Manages Complexity

The central compression is to replace evaluation of a difficult composite object with a sum of tractable component evaluations. If a rare failure event is a union of many simpler events, the union bound replaces dependence-sensitive exact computation with a sum of marginal probabilities. If a long stochastic route is decomposed into segments, subadditivity replaces a global path optimization with segmentwise upper bounds. If a bundle's value or production cost is hard to compute jointly, separate valuations give a certified ceiling. One proved inequality therefore generates an entire family of safe bounds.

This compression is modular. After each component is bounded, the bounds can be recombined without reopening the internals of every component. A proof engineer can partition by cases; a reliability analyst can partition a system-failure event; an economist can partition an output vector; an information theorist can separate marginal entropies. The domain-specific details stay inside the component evaluations, while subadditivity supplies the common rule for recomposition.

Repeated composition yields another reduction: a potentially irregular sequence becomes a long-run rate. From \(a_{m+n}\leq a_m+a_n\), Fekete's lemma says the normalized sequence converges to its infimum.^[1] Kingman replaces deterministic terms with stationary random segment costs and still obtains an almost-sure asymptotic rate under its hypotheses.^[3] The complexity of all finite scales is compressed into one asymptotic constant. The price is that upper bounds can be loose: decomposition discards favorable interaction or overlap, and finer partitioning may improve tractability while worsening tightness. Subadditivity certifies safety, not sharpness.

Abstract Reasoning

Forward inference: decompose and bound. Once subadditivity is established, choose any legal decomposition and sum the easier values. The conclusion is guaranteed to sit above the whole. This is the union-bound move in probability, triangle-inequality move in analysis, and stand-alone-cost comparison in economics.

Refutation by witness. Because the claim is universal, one pair with \(f(x\oplus y)>f(x)+f(y)\) is decisive. This makes the abstraction unusually auditable: a global proof may be difficult, but a failure certificate is small. In applied work, the strongest search strategy is therefore adversarial—look for combinations with complementarity, congestion, setup duplication, or interaction that could drive the whole above the sum.

Slack inference. The sign is fixed but the magnitude is informative. Large probability slack identifies heavy event overlap; large entropy slack identifies strong dependence; large cost slack identifies a large joint-production advantage. The structural inference is only that separate evaluation double-counts or fails to credit combination. Naming the physical or causal source requires additional evidence.

Boundary inference. Equality isolates the additive regime. In entropy it diagnoses independence; for two events it diagnoses disjointness; for cost it means no cost advantage or disadvantage for that partition. Equality conditions therefore become domain-specific tests resting on one shared structural boundary.

Asymptotic inference. When the same kind of object repeats, subadditivity prevents per-unit cost from wandering indefinitely above its best finite-block rate. Blocking a large instance into copies of a good finite block yields a global upper envelope; Fekete and Kingman turn that envelope into a limiting rate. This is a stronger conclusion than “combination saves”: it predicts macroscopic regularity from a local composition law.

Intervention inference. If the goal is a safe certificate, partition into components with known values. If the bound is too loose, redesign the partition to reduce overlap or expose dependence. If subadditivity fails, identify the complementarity or interaction responsible rather than forcing an additive estimate. The abstraction thus routes practice among three moves: exploit, tighten, or reject the decomposed bound.

Knowledge Transfer

The transfer case is strongest when roles are mapped rather than examples merely listed.

• In a norm, vectors are the carrier, vector addition is combination, the norm is the evaluation, and the triangle inequality is the whole-below-sum invariant.
• In probability, events are the carrier, union is combination, probability is the evaluation, and event overlap is the non-negative slack.
• In information theory, random-variable systems are the carrier, joint formation is combination, entropy is the evaluation, and mutual information is the slack between marginal sum and joint whole.
• In production economics, output vectors are the carrier, vector addition is combination, minimum cost is the evaluation, and joint-production savings are the slack.
• In combinatorial allocation, item sets are the carrier, union is combination, a bidder's valuation is the evaluation, and absence of positive complementarity is the subadditive restriction.
• In a stochastic path process, adjacent intervals or path segments are the carrier, concatenation is combination, optimal passage time is the evaluation, and route re-optimization can only improve on the concatenated segment costs.

The carrier, operation, and interpretation all change; the diagnostic procedure does not. A practitioner transferring the abstraction asks the same questions in every field: Can the objects compose? Is the evaluator consistent across whole and parts? Does every combination satisfy the one-sided inequality? What does the slack measure here? Can a hard whole be partitioned into pieces with known values? Does repeated composition justify a long-run rate?

This is literal mechanism transfer, not metaphor. The union bound, entropy subadditivity, norm triangle inequality, cost-subadditivity test, and subadditive bundle valuation are mathematical instances of the same ordered homomorphism-like relaxation. A proof or algorithm that uses only that relaxation can often be transplanted after replacing the carrier and evaluation map. What does not transfer automatically is the explanation for strict slack: overlap, dependence, shared infrastructure, and redundancy are domain accents, not part of the prime.

Examples

Formal/abstract

Let \(A_1,\ldots,A_n\) be events. Probability is additive only on disjoint events, but it is subadditive on arbitrary unions:

\[ P\!\left(\bigcup_{i=1}^{n}A_i\right)\leq\sum_{i=1}^{n}P(A_i). \]

For two events, inclusion–exclusion makes the slack exact:

\[ P(A)+P(B)-P(A\cup B)=P(A\cap B)\geq0. \]

The carrier is the event algebra; combination is union; the evaluation is probability; the ordered codomain is the non-negative reals; and the slack is duplicated probability mass caused by overlap. Disjointness gives equality, while overlap gives strict subadditivity.^[7] The same roles recur for \(n\) events even though higher-order overlap makes the exact expression complicated. Subadditivity deliberately forgets that complicated overlap structure and retains only the safe direction.

Suppose a distributed system fails if any one of 100 component-level bad events occurs, and each event has probability at most \(10^{-6}\). Without knowing their dependence, the union bound immediately gives system-failure probability at most \(10^{-4}\). The estimate may be loose if the events overlap heavily, but it cannot understate risk under the stated marginal bounds. Mapped back: component failures are parts, their union is the system-level whole, probability is the evaluator, the marginal sum is the decomposed upper bound, and dependence appears only as slack. The example shows both the prime's power and its limit: it provides a certificate without solving the joint distribution, but it does not promise a tight certificate.

Applied/industry

Consider an industry required to supply an output vector \(q\). Let \(C(q)\) be the minimum cost for one firm to produce that vector. For every nontrivial partition \(q=q^1+\cdots+q^m\), compare one-firm cost with the cost of separate production:

\[ C(q)\leq C(q^1)+\cdots+C(q^m). \]

If the inequality holds for every relevant partition, cost is subadditive at \(q\); if it holds throughout the output domain, cost is globally subadditive. Baumol's natural-monopoly cost test uses precisely this comparison: one supplier is productively cheaper than every multifirm partition when the inequality is strict.^[11] A utility network can generate slack because trunk infrastructure, control systems, or rights-of-way are shared, but those mechanisms are evidence for why a particular cost function is subadditive, not part of subadditivity's definition.

The test is more discriminating than “average cost falls with volume.” A multiproduct technology can have economies of scale along individual output rays yet fail subadditivity across mixed output partitions; conversely, a cost function can be subadditive over a relevant scope without globally declining average cost. The regulator must therefore test the actual partition inequality rather than infer market structure from a familiar scale story. Mapped back: output bundles are the carrier, vector addition forms the demanded whole, minimum production cost is the evaluator, separate firms instantiate the decomposed route, and the cost saving is slack. One partition with cheaper separate production is a negative-slack witness that defeats the natural-monopoly cost claim at that output.

Structural Tensions

T1 — Universal law versus local evidence (scope). Subadditivity is universal over a declared scope, but applied evidence is collected at finitely many combinations. A cost study or benchmark may observe only favorable pairs and silently promote them to a global law. The useful abstraction is therefore hard to certify precisely where it is easiest to invoke. Diagnostic: Is the claim global, local to one object, or empirical over a sample, and what untested combinations could supply a negative-slack witness?

T2 — Safe bound versus useful bound (tightness). Decomposition guarantees an upper bound but can double-count overlap, dependence, or shared structure until the result is too loose to guide a decision. The most convenient partition is often the least informative. Diagnostic: How large is observed or estimable slack, and would a coarser grouping or dependence-aware partition preserve tractability while tightening the bound?

T3 — Additive boundary versus meaningful saving (identity). Every additive map is subadditive, so the class contains cases with no combination advantage at all. Calling every equality case an “economy” or “synergy” imports a mechanism that the law does not assert. Diagnostic: Is slack strictly positive in the relevant case, and if so what independent evidence identifies its source rather than merely its sign?

T4 — Subadditivity versus diminishing returns (neighbor boundary). Subadditivity constrains total bundle value; submodularity constrains marginal changes as context grows. A map can have no complementarity in the total sense while still violating diminishing marginal returns. Conflating them imports greedy-algorithm guarantees or comparative-statics claims that subadditivity alone cannot support. Diagnostic: Does the reasoning use only \(f(A\cup B)\leq f(A)+f(B)\), or does it quietly require the stronger intersection/marginal inequality?

T5 — Binary composition versus changing regime (closure). Repeated application assumes every intermediate whole remains inside the same domain governed by the same map. Real systems can cross capacity, phase, legal, or technological thresholds as they grow, invalidating the law mid-composition. Diagnostic: Is the carrier closed and the evaluator stable under every intermediate combination, or does some block cross into a new regime where the inequality must be re-established?

T6 — Exact law versus approximate implementation (robustness). Measurements, model error, and fixed interfaces often yield \(f(x\oplus y)\leq f(x)+f(y)+\varepsilon\), not exact subadditivity. Ignoring the error makes a false theorem; rejecting the structure entirely can discard a useful controlled approximation. Under repetition, however, per-combination errors may accumulate. Diagnostic: Is the permitted defect explicitly bounded, how does it scale with the number of compositions, and does the intended conclusion survive that accumulation?

Structural–Framed Character

Subadditivity sits at the structural pole of the structural–framed spectrum, with a pilot aggregate of 0.0. Its identity is exhausted by a composable carrier, an evaluation map, an ordered additive codomain, and one inequality. The same terms—whole, parts, combination, evaluation, upper bound, slack, equality—retain their roles for vectors, events, random-variable systems, output bundles, item sets, and stochastic paths. Domain vocabulary explains the carrier and the slack but does not alter the law.

Every criterion points structural. The vocabulary does not need to travel from one institutional practice; each field can state the relation in its native objects while preserving the exact role mapping. The inequality carries no approval or disapproval, so evaluative weight is zero. Its origin is formal rather than institutional. Physical and stochastic systems satisfy it without a human agent making it true. Invoking it recognizes a relation already present rather than importing a contested interpretation. The pilot grade is therefore unambiguous, while its provenance remains explicit so an independent grading pass can reproduce or challenge it.

Substrate Independence

Subadditivity is about as substrate-independent as a prime can be—composite 5 / 5. Domain breadth is maximal because literal, source-supported instances span mathematical analysis, probability, classical and quantum information, production economics, algorithmic allocation, and stochastic spatial processes. These are not several applications of one field's metaphor; their formal objects differ materially while satisfying the same inequality.

Structural abstraction is maximal: no vector space, event algebra, human judgment, market, or physical substrate is constitutive. The carrier needs only a composable operation, and the evaluation needs only an ordered additive codomain. Transfer evidence is also maximal: each domain uses the inequality operationally to prove bounds, characterize equality or slack, design algorithms, test market structure, or establish asymptotic rates. The cross-domain intervention—partition, evaluate parts, sum, and treat the result as an upper bound—travels with the vocabulary and diagnostics intact.

• Composite substrate independence — 5 / 5
• Domain breadth — 5 / 5
• Structural abstraction — 5 / 5
• Transfer evidence — 5 / 5

Relationships to Other Abstractions

Current abstraction Subadditivity Prime

Parents (3) — more general patterns this builds on

• Subadditivity is a kind of Function (Mapping) Prime

  The accepted reference-grade review places Subadditivity under Function (Mapping) because the child instantiates or depends on the parent's broader structure while retaining its own constitutive identity.

• Subadditivity presupposes Order Prime

  The accepted reference-grade review places Subadditivity under Order because the child instantiates or depends on the parent's broader structure while retaining its own constitutive identity.

• Subadditivity presupposes Semigroup Prime

  The accepted reference-grade review places Subadditivity under Semigroup because the child instantiates or depends on the parent's broader structure while retaining its own constitutive identity.

Hierarchy paths (8) — routes to 7 parentless roots

• Subadditivity → Function (Mapping)

Show alternative paths (7)

Neighborhood in Abstraction Space

Subadditivity sits among the more crowded primes in the catalog (25^th percentile for distinctiveness): several abstractions describe nearly the same structure, so a description that fits it will tend to fit its neighbors too — transporting it usually means disambiguating within this family rather than landing on it exactly.

Family — Foundational Mathematical Structures (23 primes)

Nearest neighbors

• Proof of impossibility — 0.73
• Coordinate-free — 0.73
• Utility — 0.73
• Multistability — 0.73
• Order — 0.72

Computed from structural-signature embeddings · 2026-09-10

Not to Be Confused With

subadditivity_effect. This is the mandatory identity boundary. The domain-specific entry is a judgment-and-decision-making phenomenon: one event receives a lower elicited probability when packed than the sum of its named unpacked alternatives because naming recruits cognitive support. Its roles include a judging agent, two descriptions, an elicitation procedure, an exhaustive partition, and support theory. subadditivity has none of those roles. It is satisfied by computed probabilities, entropies, norms, and costs with no psychological mechanism. The names are historically related because the bias produces an inequality of the same direction, but one is a domain-specific causal account of judged probabilities and the other is the substrate-independent inequality property. Neither should be an alias or recognized variant of the other.

synergy_and_antagonism. The neighbor asks whether a joint effect departs from a declared no-interaction baseline and then attributes a positive or negative interaction. Subadditivity asks whether one value map obeys a universal upper combination law. A subadditive cost can have strict slack without establishing causal synergy, and an additive map is still subadditive despite having no interaction residual. Conversely, synergy/antagonism allows both signs and multiple baseline models; subadditivity fixes one sign and one separate-sum comparator.

economies_of_scope and economies_of_scale. Economies of scope is a cost-specific child-shaped case: joint production of heterogeneous outputs costs less because a reusable substrate is shared. Economies of scale concerns average cost as volume grows. Both can support cost subadditivity, but neither exhausts it. Baumol's multiproduct analysis is especially important here: scale economies are not the same test as global cost subadditivity.^[11] The general node should therefore not absorb their mechanisms; they remain interpreted specializations and likely child/neighbor candidates.

measure. A measure is countably additive on disjoint sets and therefore subadditive on arbitrary unions. It adds a base set, sigma-algebra, non-negativity, empty-set zero, and exact disjoint additivity. Subadditivity is weaker and more general: its carrier need not be sets, its evaluator need not be non-negative or normalized, and exact additivity is unnecessary. measure is evidence of recurrence, not an alias.

metric and domain-specific norm. The triangle inequality is a subadditive-shaped constituent. A norm is literally a subadditive function on vector addition plus homogeneity and definiteness. A metric's inequality compares a direct pair with a two-leg path and additionally requires non-negativity, symmetry, and identity of indiscernibles. These nodes carry enough added structure and domain apparatus to remain separate. subadditivity names the one-sided combination law they reuse.

convexity. Convexity compares \(f(\theta x+(1-\theta)y)\) with a weighted average of \(f(x)\) and \(f(y)\); subadditivity compares \(f(x+y)\) with an unweighted sum. Positive homogeneity can connect the two, and every norm is both convex and subadditive, but the properties diverge without extra hypotheses. They should be neighbors, not aliases or a simple genus/species pair.

Recognized variants superadditivity and subadditive set function. The reviewed vocabulary overlay already routes these surfaces to the staged target as recognized variants, not aliases. A subadditive set function is the union-based specialization of the same law. Superadditivity reverses the inequality and is the sign/order dual; it shares the role package but is not coextensive with the name “subadditivity.” Promotion should preserve that variant status rather than adding either surface to aliases.

aggregation and compositionality. Aggregation collapses many items into a summary; compositionality says a whole's value is determined by parts and combination rules. Neither fixes the direction of a quantitative inequality. Subadditivity adds exactly that missing sign constraint. It can govern an aggregation or a compositional evaluation, but is not entailed by either.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.

References

[1] Fekete, Michael. “Über die Verteilung der Wurzeln bei gewissen algebraischen Gleichungen mit ganzzahligen Koeffizienten.” Mathematische Zeitschrift 17 (1923): 228–249. Original source associated with Fekete's subadditive lemma, which identifies the asymptotic per-unit rate of a subadditive sequence. registry ↩a ↩b

[2] Hammersley, J. M. “Generalization of the Fundamental Theorem on Subadditive Functions.” Proceedings of the Cambridge Philosophical Society 58, no. 2 (1962): 235–238. Develops the limiting-rate theorem for broader subadditive functions. registry ↩

[3] Kingman, J. F. C. “The Ergodic Theory of Subadditive Stochastic Processes.” Journal of the Royal Statistical Society: Series B 30, no. 3 (1968): 499–510. Develops subadditive ergodic theory for stochastic processes and its long-run rate conclusion. registry ↩a ↩b ↩c

[4] Ceccherini-Silberstein, Tullio, Michel Coornaert, and Fabrice Krieger. “An Analogue of Fekete's Lemma for Subadditive Functions on Cancellative Amenable Semigroups.” Journal d'Analyse Mathématique 124 (2014): 59–81. Extends subadditive-rate reasoning from sequences to semigroup/set-function settings. registry ↩

[5] Feige, Uriel. “On Maximizing Welfare When Utility Functions Are Subadditive.” SIAM Journal on Computing 39, no. 1 (2009): 122–142. Uses subadditive set valuations as the broad complement-free class in combinatorial welfare maximization. registry ↩a ↩b

[6] Boyd, Stephen, and Lieven Vandenberghe. Convex Optimization. Cambridge University Press, 2004, Appendix A.1.2. Author-hosted text stating the norm triangle inequality \(f(x+y)\leq f(x)+f(y)\) and its role in norm geometry. registry ↩

[7] Abel, Zachary, Benjamin Chapman, and Erik Demaine. “Lecture 19: Conditional Probability.” MIT OpenCourseWare, 6.1200J/18.062J, Spring 2024. States the two-event and generalized union bounds and contrasts them with exact additivity for disjoint events. registry ↩a ↩b

[8] Shannon, Claude E. “A Mathematical Theory of Communication.” Bell System Technical Journal 27 (1948): 379–423, 623–656. Foundational primary source for entropy and joint/conditional information quantities. registry ↩

[9] Weissman, Tsachy, with scribes Alon Devorah, David Hallac, and Kevin Shutzberg. “Lecture 3: Entropy, Relative Entropy, and Mutual Information.” Stanford EE376A, 2015. States \(H(X,Y)\leq H(X)+H(Y)\), equality under independence, and the mutual-information slack identity. registry ↩

[10] Lieb, Elliott H., and Mary Beth Ruskai. “Proof of the Strong Subadditivity of Quantum-Mechanical Entropy.” Recherche Coopérative sur Programme 25, vol. 19 (1973): 36–55; see also Journal of Mathematical Physics 14 (1973): 1938–1941, doi:10.1063/1.1666274. Primary proof of strong subadditivity for quantum entropy. registry ↩

[11] Baumol, William J. “On the Proper Cost Tests for Natural Monopoly in a Multiproduct Industry.” American Economic Review 67, no. 5 (1977): 809–822. Defines strict global cost subadditivity as the cost criterion for natural monopoly and distinguishes it from scale-economy tests. registry ↩a ↩b ↩c