Proper Convex Function¶

An extended-real convex function whose effective domain is nonempty and which nowhere takes negative infinity, excluding the two degenerate functions that break convex-analytic operations.

Version: v1 · 2026-08-30 · History
Domain-specific #: 2555
Origin domain: convex analysis
Subdomain: extended real valued functions

Core Idea¶

A proper convex function is an extended-real-valued convex function that is finite somewhere and never equals negative infinity. With the minimization convention, write the effective domain as dom f = {x : f(x) < +infinity}. Properness says that this set is nonempty and that f(x) > -infinity everywhere. Equivalently, it excludes the identically positive-infinite function and every function attaining negative infinity.^[1]^[2]

The qualifier is not a synonym for convex. It is a domain-admissibility condition that makes extended-value notation useful without allowing its two destructive degeneracies. Positive infinity can encode exclusion from the feasible domain, while finite values retain ordinary objective information. Properness is therefore the gate through which many conjugacy, subdifferential, and optimization statements enter convex analysis. Extra hypotheses such as closedness, lower semicontinuity, coercivity, differentiability, or strict convexity remain separate.^[1]^[3]

Structural Signature¶

Recognition roles:

ambient convex domain: a real vector space, commonly finite-dimensional Euclidean or a Banach/Hilbert space;
extended-real codomain: values in the real numbers together with positive infinity, and sometimes negative infinity before properness is imposed;
convexity invariant: f(theta*x + (1-theta)*y) <= theta*f(x) + (1-theta)*f(y) for admissible weights;
effective domain: the points where the value is below positive infinity;
nonvacuity gate: at least one point has a finite value;
lower degeneracy exclusion: the function is nowhere negative infinity; and
operational consequence: indicator functions, constraints, conjugates, epigraphs, and subdifferentials can be manipulated without the excluded cases trivializing them.

The identity survives changes of formula, dimension, or application. A real-valued convex function is automatically proper when its declared domain is nonempty. An extended-real function that is convex but identically positive infinity is not.

What It Is Not¶

Properness is not closedness. A proper convex function may fail to be lower semicontinuous at a boundary, while a closed proper convex function has an appropriately closed epigraph. It is not strict or strong convexity, both of which strengthen the chord inequality. It is not coercivity, which concerns growth toward infinity, and not essential smoothness or essential strict convexity.

Nor does “proper” mean statistically correct, physically meaningful, or well posed for a particular application. The word has a precise convex-analytic role. A nonempty effective domain does not guarantee that a minimum exists, is unique, or is numerically easy to locate. Boyd and Vandenberghe's extended-value convention shows how positive infinity records points outside an ordinary function's domain; the convention alone does not supply properness or lower semicontinuity.^[4]

Scope of Application¶

The abstraction belongs to convex analysis, variational analysis, optimization, monotone operator theory, and convex duality. It appears wherever constraints are folded into objectives through extended-real indicator functions. For a nonempty convex set C, the indicator delta_C equals zero on C and positive infinity outside; it is proper exactly when C is nonempty. This representation lets one state constrained minimization as unconstrained minimization of an extended-valued function.^[4]^[3]

Proper convex functions also organize Fenchel conjugacy, proximal mappings, subdifferentials, and dual problems. The qualification should not be transported casually to arbitrary uses of “proper function” in topology, algebraic geometry, category theory, or programming. Those are different established senses with different invariants.

Clarity¶

The fastest diagnostic is two questions. First, is f(x) finite for at least one x? Second, does f avoid negative infinity everywhere? If both answers are yes and convexity is already established, f is proper. The test explains why positive infinity is permitted: it marks exclusion from the effective domain. What is disallowed is having no finite point at all or using negative infinity as an attained objective value.

For example, f(x)=x^2 on the real line is proper. So is the extended function equal to x^2 for nonnegative x and positive infinity otherwise. The identically positive-infinite function fails nonvacuity. A function assigning negative infinity at even one point fails the lower exclusion, even if a formal inequality can be made to look convex under a chosen extended-arithmetic convention.

Manages Complexity¶

Extended values unify objectives and constraints. Instead of carrying “minimize g(x) subject to x in C” through every theorem, one studies g + delta_C. The effective domain records feasibility, addition intersects admissible domains, epigraphs encode value and feasibility geometrically, and conjugation converts the combined object into a dual representation.^[4]

Properness keeps this compression informative. If every value is positive infinity, the model has no feasible point. If negative infinity is attained, infimal operations and conjugacy can collapse. The gate does not solve the model, but it prevents elementary degeneracies from masquerading as legitimate convex objects.

Abstract Reasoning¶

Properness licenses controlled inferences. A proper convex function has a nonempty convex effective domain because convexity preserves finite-valued mixtures. It has at least one finite epigraph point. Its indicator-function examples correspond to nonempty convex sets. These claims follow directly from the recognition roles.

Other conclusions need additional hypotheses. Fenchel–Moreau recovery generally invokes properness together with lower semicontinuity and convexity. Existence of minimizers may require compact sublevel sets or coercivity. A nonempty subdifferential at every domain point is not automatic; relative-interior conditions matter. The abstraction improves reasoning precisely by making these hypothesis boundaries explicit rather than smuggling them into the adjective “proper.”^[1]^[3]

Knowledge Transfer¶

Within convex analysis the role package transfers literally among finite-dimensional optimization, function spaces, optimal control, signal recovery, statistics, and economics. A loss function plus a regularizer plus indicator constraints can remain proper when their effective domains share a suitable finite point. The terminology, however, stays mathematical.

The broader transferable structure is a nondegeneracy gate: retain sentinel values that encode exclusion while banning states that erase the object. That portable intuition is already captured by catalog-level Constraint, Boundary, and Convexity. It does not justify treating every discipline's word “proper” as an instance of this node.

Examples¶

Constraint indicator. For a nonempty convex set C, define delta_C(x)=0 on C and +infinity outside it. Its effective domain is exactly C; it never takes negative infinity, so it is proper and convex. If C is empty, the same definition is identically positive infinity and fails properness.^[4]

Restricted logarithmic barrier. Define f(x)=-log(x) for positive x and positive infinity otherwise. It is proper and convex, with effective domain the positive ray. It is not finite everywhere, and its boundary behavior should not be confused with the properness test.

Nonclosed example. On the real line, let f(x)=0 for positive x, f(0)=1, and positive infinity for negative x. It is proper and convex under the chord test but is not lower semicontinuous at zero. The example separates properness from closedness.

Degenerate counterexamples. The constant positive-infinite function has empty effective domain. Any extended-real function that attains negative infinity violates the second clause. Neither belongs to the proper class.

Structural Tensions¶

Unified notation versus hidden feasibility. Positive infinity elegantly encodes constraints but can hide an empty feasible domain. Diagnostic: explicitly exhibit one finite point before invoking proper-function results.
Minimal gate versus stronger regularity. Properness is often printed beside “closed,” encouraging readers to merge them. Diagnostic: test epigraph closedness separately from nonempty effective domain.
Permitted positive infinity versus forbidden negative infinity. The asymmetry can seem arbitrary. Diagnostic: identify the minimization convention and reverse the convention carefully for concave maximization.
Formal convexity versus informative analysis. Degenerate extended functions may satisfy weakened formal inequalities. Diagnostic: apply both properness clauses before treating a convex inequality as analytically useful.
Autonomy versus Convexity coverage. Convexity supplies the mixture inequality, but not the extended-value nondegeneracy gate. Diagnostic: remove finite-somewhere and no-negative-infinity; if the admitted class now includes vacuous functions, the proper-function residual is real.

Structural–Framed Character¶

The identity is highly structural. It depends on codomain order, effective domain, two infinity exclusions, and convexity, not on an application narrative. Framing enters through the minimization convention and the choice of extended-real arithmetic. Sources that work with functions valued only in real numbers may treat properness as automatic rather than foreground it.

Structural Core vs. Domain Accent¶

The structural core is nonvacuous admissibility under sentinel-valued representation. The domain accent is decisive: extended real numbers, epigraphs, effective domains, convex conjugates, subdifferentials, and minimization conventions. Without those objects, “proper” becomes an analogy or a different technical homonym.

The node is therefore domain-specific. The broader prime Convexity covers the chord and mixture invariant across many domains, while this candidate records the specialist gate applied to convex functions.

Proper Convex Function is a strict specialization of prime:convexity: every member carries the convex-function chord inequality, with additional extended-value admissibility conditions. It also relates to Boundary through the effective domain, Constraint through positive-infinity indicators, and Nondegeneracy through exclusion of vacuous cases. Only Convexity is proposed as the minimal parent.

Relationships to Other Abstractions¶

Current abstraction Proper Convex Function Domain-specific

Parents (1) — more general patterns this builds on

Proper Convex Function is a kind of Convexity Prime

Proper Convex Function is a strict specialization of prime:convexity: every member carries the convex-function chord inequality, with additional extended-value admissibility conditions.

Hierarchy path (1) — routes to 1 parentless root

Proper Convex Function → Convexity → Optimization

Neighborhood in Abstraction Space¶

Proper Convex Function sits in a sparse region of the domain-specific corpus (72^nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Applied Linear & Special Functions (18 abstractions)

Nearest neighbors

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

Not to Be Confused With¶

Convex function: may be used broadly enough to include improper extended-valued cases; properness adds two gates.
Closed convex function: adds lower semicontinuity or closed-epigraph requirements, depending on convention.
Strictly/strongly convex function: strengthens curvature, not domain admissibility.
Coercive function: grows along escaping sequences and concerns existence/compactness behavior.
Proper map: an unrelated topological notion based on inverse images of compact sets.
Proper morphism: an algebro-geometric notion, not an extended-value function class.
Concave proper function: conventionally handled by negating the function and reversing the relevant infinities.

References¶

[1] R. Tyrrell Rockafellar, Convex Analysis, Princeton University Press, 1970, especially the extended-real and proper-function definitions in the opening chapters, https://press.princeton.edu/books/paperback/9780691015866/convex-analysis. registry ↩a ↩b ↩c

[2] R. Tyrrell Rockafellar and Roger J-B Wets, Variational Analysis, Springer, 1998/2009 printing, Chapter 1 definitions of effective domain and proper extended-real functions, https://doi.org/10.1007/978-3-642-02431-3. registry ↩

[3] Heinz H. Bauschke and Patrick L. Combettes, Convex Analysis and Monotone Operator Theory in Hilbert Spaces, second edition, Springer, 2017, foundational notation and proper lower-semicontinuous convex function class, https://doi.org/10.1007/978-3-319-48311-5. registry ↩a ↩b ↩c

[4] Stephen Boyd and Lieven Vandenberghe, Convex Optimization, Cambridge University Press, 2004, section 3.1.2 on extended-value extensions and indicator functions, https://web.stanford.edu/~boyd/cvxbook/bv_cvxbook.pdf. registry ↩a ↩b ↩c ↩d