Maximum Theorem¶

Guarantee continuity of a parametric optimum's value and upper hemicontinuity of its maximizer correspondence when objective and compact feasible correspondence vary continuously.

Version: v2 · 2026-09-06 · History
Domain-specific #: 2249
Origin domain: mathematics
Subdomain: parametric optimization
Aliases: Berge maximum theorem, Berge's maximum theorem, Theorem of the maximum

Core Idea¶

The Maximum Theorem, commonly called Berge's Maximum Theorem, gives stability conditions for a parameterized optimization problem. Let \(X\) and \(\Theta\) be suitable topological spaces, let \(f:X\times\Theta\to\mathbb{R}\) be continuous, and let \(C:\Theta\rightrightarrows X\) be a nonempty compact-valued correspondence that is both upper and lower hemicontinuous. Define the value function \(v(\theta)=\max_{x\in C(\theta)} f(x,\theta)\) and the argmax correspondence \(M(\theta)=\{x\in C(\theta):f(x,\theta)=v(\theta)\}\). Then \(v\) is continuous, while \(M\) is nonempty, compact-valued, and upper hemicontinuous.^[1]

The theorem distinguishes value stability from optimizer selection stability. Small parameter changes cannot cause a jump in the optimized value under the stated hypotheses, and maximizers cannot suddenly appear far from the previous maximizer set in the upper-hemicontinuity sense. But the argmax correspondence need not be lower hemicontinuous, and no continuous single-valued optimizer follows without additional uniqueness and regularity. If strict quasiconcavity and convex feasible sets yield a unique maximizer, upper hemicontinuity of the singleton-valued correspondence can imply continuity under standard settings. Aliprantis and Border give a modern correspondence-based treatment and make the topology and compactness obligations explicit.^[2]

The autonomous residual is the coupled result continuous objective + continuous compact feasible correspondence → continuous value + upper-hemicontinuous compact argmax. The extreme value theorem alone establishes existence at a fixed parameter but says nothing about variation with the parameter. The envelope theorem differentiates values under stronger assumptions and is not a synonym. Minimization follows by applying the theorem to \(-f\). Noncompact feasible sets require replacement hypotheses such as inf-compactness or compactness localized to relevant levels; omitting compactness can allow optimizers to escape. The strict parent is Continuity because the theorem's distinctive work is preserving controlled no-jump behavior through optimization.

Structural Signature¶

A parameter space \(\Theta\). Its topology determines what nearby optimization problems mean.
A choice space \(X\). Candidate decisions live in a topology compatible with the correspondence results.
A continuous objective \(f\). Payoff varies continuously in both choice and parameter.
A feasible correspondence \(C\). Each parameter is mapped to its admissible choice set.
Nonempty compact values. A feasible optimum exists and escaping maximizing sequences are controlled.
Upper hemicontinuity of feasibility. Nearby feasible sets do not acquire remote limit behavior unexpectedly.
Lower hemicontinuity of feasibility. Feasible choices can be approximated from nearby parameter values.
A value function. The optimal objective level summarizes each parameterized problem.
An argmax correspondence. All maximizing choices are retained instead of forcing an arbitrary selector.
Asymmetric continuity dividend. Value is continuous; argmax is guaranteed upper but not generally lower hemicontinuity.

What It Is Not¶

Not the extreme value theorem. Compactness gives an optimum at one parameter but not parametric stability.
Not the envelope theorem. The maximum theorem proves continuity, not a derivative formula for the value.
Not a maximum principle. Control and convex-analysis principles use different conditions and conclusions.
Not a uniqueness theorem. Multiple maximizers are allowed and represented as a correspondence.
Not continuity of every optimizer selection. A selected branch may jump even when the argmax set is upper hemicontinuous.
Not Berge's theorem in matching theory. That names a graph-theoretic augmenting-path criterion.

Scope of Application¶

The Maximum Theorem supports comparative statics and existence arguments wherever feasible sets and objectives vary with parameters.

Consumer theory. Establishing continuity and upper hemicontinuity of demand under changing prices and wealth.
Game theory. Proving best-response correspondence regularity before a fixed-point argument.
Optimal control. Studying how value and optimal-control sets vary with state or model parameters.
General equilibrium. Supplying continuity properties needed for equilibrium existence machinery.
Statistical decision theory. Controlling parameterized minimizers after changing sign or loss convention.
Numerical optimization. Distinguishing stable objective values from unstable or set-valued optimizers.

Clarity¶

Specify the topology on parameter and choice spaces. Define upper and lower hemicontinuity using one consistent convention. State that the feasible correspondence has nonempty compact values, not merely a closed graph. Use maximum rather than supremum only after attainment is secured. Keep the value function and argmax correspondence typographically distinct. Do not call the argmax continuous unless lower hemicontinuity or single-valued conditions are separately established. When using a local or noncompact variant, state its replacement compactness hypothesis and do not present it as the classical theorem without qualification. For minimization, apply the result to the negative objective and preserve all feasibility conditions. Distinguish strict quasiconcavity, which can provide uniqueness, from continuity assumptions, which provide stability. If a computed optimizer jumps between two equal optima, check the full set-valued correspondence before declaring failure. Also distinguish the parameter topology from the topology inherited by feasible choices; continuity is not a purely metric assertion. When the objective or correspondence is only semicontinuous, state which half of the classical conclusion survives. An application should identify where compactness, joint continuity, and both correspondence directions enter rather than citing the theorem as an undifferentiated regularity slogan.

Manages Complexity¶

Parametric optimization composes two varying objects: a payoff surface and a feasible set. Taking a maximum is nonlinear, and choosing an optimizer can magnify tiny parameter changes. The theorem decomposes this complexity into separate topological obligations. Upper hemicontinuity prevents feasible mass from appearing remotely; lower hemicontinuity prevents existing choices from disappearing without nearby approximants; compactness traps maximizing sequences; objective continuity transports payoff comparisons. These components yield a continuous scalar value even when the solution remains set-valued. Retaining all maximizers avoids arbitrary tie-breaking and makes the weaker upper-hemicontinuity result exactly visible. Additional convexity and strict quasiconcavity can then be layered on to obtain convexity or uniqueness. The theorem thus creates a modular stability certificate rather than pretending optimization automatically preserves continuity.

Abstract Reasoning¶

Define parameter, choice, objective, and feasible correspondence with their topologies.
Verify nonemptiness and compactness for every feasible set in the relevant parameter domain.
Prove upper hemicontinuity and lower hemicontinuity of the feasible correspondence separately.
Verify joint continuity, or the appropriate semicontinuity halves, of the objective.
Use compactness to replace supremum by maximum and establish nonempty compact argmax values.
Prove upper and lower semicontinuity of the value and combine them into continuity.
Prove upper hemicontinuity of the argmax through closedness and compact containment.
Add uniqueness or convexity only from separately stated curvature assumptions and test boundary variants explicitly.

Knowledge Transfer¶

The strict parent is Continuity. The theorem identifies topologies on inputs and outputs, controls closeness of feasible sets through two hemicontinuities, and proves that optimization preserves no-jump behavior of the value while giving a precise weaker continuity property for maximizers. The transferable lesson is to track stability of both a score and its optimizing states after a nonlinear extremization. Compact-valued correspondences, argmax sets, and quasiconcavity are domain-specific mathematical accent.

Examples¶

Canonical¶

Let \(C(\theta)=[0,\theta]\) for \(\theta\in[0,1]\) and \(f(x,\theta)=-(x-1/2)^2\). The feasible correspondence is continuous with nonempty compact values. The maximizer is \(M(\theta)=\{\theta\}\) for \(\theta<1/2\) and \(M(\theta)=\{1/2\}\) for \(\theta\ge 1/2\). The value changes continuously, and here the unique optimizer is continuous as well.

Mapped back: continuous expanding interval + continuous objective → attained optimum → continuous value and stable argmax.

Applied / In Practice¶

A consumer's budget set changes with prices and wealth while utility is continuous. Under the appropriate compactness and correspondence-continuity conditions, indirect utility is continuous and the demand correspondence is nonempty, compact-valued, and upper hemicontinuous. Multiple equally preferred bundles can prevent a unique continuous demand function, which is why the set-valued result is the correct default.^[2]

Mapped back: parameters → continuous budget correspondence + utility → continuous indirect utility + upper-hemicontinuous demand set.

Structural Tensions¶

Value stability vs. optimizer stability. The scalar optimum can be smooth while choices switch. Diagnostic: Are value and argmax claims stated separately?
Compactness vs. realistic unbounded domains. Compact sets make the theorem clean while applications may be noncompact. Diagnostic: What prevents maximizing sequences from escaping?
Upper vs. lower hemicontinuity. Calling a correspondence ‘continuous’ can hide two different obligations. Diagnostic: Were both directions proved for feasibility?
Set-valued truth vs. selected convenience. One selector is easier to use but may introduce artificial jumps. Diagnostic: Is multiplicity retained until uniqueness is proven?
Autonomous theorem vs. Continuity plus Optimization. Those primes do not alone yield the asymmetric value/argmax result. Diagnostic: Does the compact correspondence theorem perform indispensable work?

Structural–Framed Character¶

Parametric objective, continuous compact feasible correspondence, value continuity, and argmax upper hemicontinuity are structural. Choice of application, topology, curvature assumptions, and noncompact generalization is framed. The abstraction is domain-specific because it is a theorem of set-valued analysis and optimization.

Structural Core vs. Domain Accent¶

The portable core is nearby inputs + controlled feasible variation → no-jump output. The domain accent is compact-valued correspondence, extremization, value function, argmax set, and asymmetric hemicontinuity conclusion. Removing it leaves Continuity; retaining it yields Maximum Theorem.

Continuity is the strict parent because Berge's theorem establishes precisely when small parameter changes produce small value changes and controlled set-valued changes of optimizers. Optimization is a close neighbor and application context, but continuity preservation is the theorem's distinctive dividend.

The prospective workspace queue contains one strict upward edge to prime:continuity. No live DAG mutation is authorized.

Relationships to Other Abstractions¶

Current abstraction Maximum Theorem Domain-specific

Parents (1) — more general patterns this builds on

Maximum Theorem is a kind of Continuity Prime

Continuity is the strict parent because Berge's theorem establishes precisely when small parameter changes produce small value changes and controlled set-valued changes of optimizers.

Hierarchy paths (2) — routes to 2 parentless roots

Maximum Theorem → Continuity → Neighborhood → Topology

Show alternative path (1)

Neighborhood in Abstraction Space¶

Maximum Theorem sits in a sparse region of the domain-specific corpus (86^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (1565 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Extreme value theorem. Ensures attainment for one continuous function on one compact set.
Envelope theorem. Gives derivative information for optimized values under stronger assumptions.
Bauer maximum principle. Locates maxima of convex functions on extreme points.
Maximum principle in control. Supplies necessary conditions for optimal controls.
Berge's matching theorem. Characterizes maximum matchings through augmenting paths.
Minimax theorem. Relates max–min and min–max values under convexity conditions.

References¶

[1] Claude Berge, Topological Spaces, Including a Treatment of Multi-Valued Functions, Vector Spaces and Convexity, trans. E. M. Patterson (Oliver & Boyd, 1963), Chapter VI, §3. registry ↩

[2] Charalambos D. Aliprantis and Kim C. Border, Infinite Dimensional Analysis: A Hitchhiker's Guide, 3^rd ed. (Springer, 2006), Theorem 17.31, https://doi.org/10.1007/3-540-29587-9. registry ↩a ↩b