1 Definition

A strictly convex function is a function whose graph bends upward in a way that every nontrivial chord lies above the graph between its endpoints. In ordinary use, the term applies to functions defined on a convex set in a real vector space, often with values in the real numbers. The key feature is strict inequality: whenever two distinct points are chosen in the domain, the function value at any interior point on the segment joining them is strictly smaller than the corresponding weighted average of the endpoint values.

Strict convexity is stronger than convexity. It rules out flat segments on the graph and typically enforces uniqueness in optimization problems. In geometry, analysis, and economics, the notion is used to express curvature and to control how averages compare with pointwise values.

1.1 Jensen-style inequality form

The most common formulation states that for any two distinct points x and y in a convex domain and any t with 0 < t < 1, one has

f(tx + (1 - t)y) < t f(x) + (1 - t) f(y).

This is the strict analogue of the usual convexity inequality. It says that the function value at a proper convex combination of two distinct points is strictly below the linear interpolation of the endpoint values.

1.2 Chord-based geometric form

Geometrically, a strictly convex function has the property that the line segment connecting any two points on its graph lies entirely above the graph, except at the endpoints. Thus the graph never contains a nontrivial line segment. This visual description is especially useful for understanding the shape of one-variable functions and the epigraph of a function.

1.3 Comparison with convex functions

Every strictly convex function is convex, but not every convex function is strictly convex. Ordinary convexity allows equality in the defining inequality, which can occur along flat regions or linear pieces. Strict convexity eliminates these possibilities. As a result, a convex function may have many minimizers, while a strictly convex one can have at most one minimizer on a convex domain.

1.4 Strict vs. non-strict inequalities

The distinction between strict and non-strict inequalities is central. In convexity, equality may hold for many interior points and still preserve the definition. In strict convexity, equality is permitted only in trivial cases, such as when the two points coincide or when one endpoint weight is 0 or 1. This sharper inequality is what gives strict convexity its stronger consequences.

2 Basic properties

Strictly convex functions have several structural consequences that make them especially useful in analysis and optimization. Their graphs cannot contain linear stretches, and their minimization behavior is highly rigid. Many operations preserve strict convexity, though some do so only under additional conditions.

2.1 Uniqueness of minimizers

If a strictly convex function attains a minimum on a convex domain, that minimum is unique. The reason is simple: if two distinct minimizers existed, the strict inequality along the segment between them would force every interior point to have a strictly smaller value, contradicting minimality. This property is one of the main reasons strictly convex functions are central in optimization.

2.2 Local and global minima

For convex functions, any local minimum is automatically global. For strictly convex functions, the same principle holds, and the minimizer is additionally unique if it exists. A strictly convex function may fail to attain a minimum if the domain is not closed or if the infimum lies at the boundary, but any attained minimum is rigidly determined.

2.3 Behavior under affine transformations

Adding an affine function to a strictly convex function preserves strict convexity. Multiplying by a positive scalar also preserves it. By contrast, composing with an affine change of variable preserves strict convexity only when the map does not collapse distinct points in the domain. These transformations are often used to normalize problems without changing the essential curvature.

2.4 Closure properties

Strict convexity is stable under several common operations, although the details depend on the setting. Sums, positive scalar multiples, and suitable compositions often remain strictly convex. However, one must check that no operation introduces linear directions or identifications that destroy strictness.

2.4.1 Sum of strictly convex functions

The sum of two strictly convex functions defined on the same convex domain is strictly convex, provided both are finite-valued in the relevant region. More generally, the sum of a strictly convex function and a convex function is strictly convex. This is frequently used in regularization, where a curvature-inducing term is added to an objective function.

2.4.2 Composition with linear maps

If f is strictly convex and A is an injective linear map, then the composition f ∘ A is strictly convex on the preimage domain. If A is not injective, distinct points may collapse to the same image, and strict convexity can fail. Thus linear composition preserves strictness only when it does not destroy separation of points.

3 Characterizations

Strict convexity can be recognized in several equivalent or closely related ways. In one dimension, derivative-based tests are particularly convenient. In higher dimensions, gradients and Hessians provide practical criteria when differentiability assumptions are available.

3.1 One-dimensional characterizations

For functions of a single real variable, strict convexity has especially transparent descriptions in terms of slopes and derivatives. These criteria are often easier to verify than the definition itself.

3.1.1 Increasing derivative criterion

If a function is differentiable on an interval and its derivative is strictly increasing, then the function is strictly convex. Conversely, for differentiable convex functions, strict increase of the derivative is closely tied to strict convexity, though endpoint behavior may require attention depending on the domain.

3.1.2 Strictly increasing slope criterion

A one-variable function is strictly convex if the slope of the secant line between any two points increases as the interval shifts to the right. Equivalently, the average rate of change over intervals reflects a strict ordering. This captures the idea that the graph bends upward without flattening.

3.2 Differentiable characterizations

When a function is differentiable, strict convexity can often be described by the behavior of tangents and gradients. These formulations are common in optimization and variational analysis.

3.2.1 Gradient monotonicity

For a differentiable convex function, the gradient is monotone in the sense that the inner product of gradient differences with point differences is nonnegative. Strict convexity strengthens this behavior: equality is excluded unless the points coincide or the function has linearity in a forbidden direction. This monotonicity viewpoint is central in multivariable analysis.

3.2.2 Tangent line criterion

A differentiable function is convex if it lies above each of its tangent hyperplanes. For strict convexity, the graph lies strictly above the tangent plane at every other point of the domain, except where tangency occurs. This criterion gives a local geometric picture of the function’s curvature.

3.3 Twice differentiable characterizations

Second-derivative tests provide convenient sufficient conditions for strict convexity, and in many smooth settings they are also close to necessary conditions.

3.3.1 Positive second derivative

In one variable, a twice differentiable function with positive second derivative throughout an interval is strictly convex there. A nonnegative second derivative yields convexity, while strict positivity excludes flat curvature and leads to strictness.

3.3.2 Positive definite Hessian

In several variables, if a twice differentiable function has a positive definite Hessian at every point of a convex open set, then the function is strictly convex on that set. If the Hessian is merely positive semidefinite, the function may still be convex but not strictly convex. Positive definiteness is therefore a powerful sufficient condition.

4 Examples

Strictly convex functions appear in elementary algebra, classical analysis, and applied mathematics. Many familiar functions are strictly convex on natural domains, while others become so only after restriction.

4.1 Polynomial examples

The function f(x) = x^2 is a basic example of strict convexity on the real line. More generally, x^n is strictly convex on intervals where the second derivative is positive, such as x^4 on all of R and x^3 on intervals where it is convex. Quadratic forms with positive definite coefficients provide multivariable examples.

4.2 Exponential and power functions

The exponential function e^x is strictly convex on R. Power functions x^p are strictly convex on (0, ∞) for many values of p greater than 1. Logarithmic functions do the opposite: they are concave rather than convex on their natural domains.

The function x ↦ x^2 on R can be viewed as a squared norm. In higher dimensions, x ↦x^2 is strictly convex with respect to any norm induced by an inner product, and in many cases the squared Euclidean norm is a standard example. Distance-squared functions often play a similar role in least-squares problems and geometric fitting.

4.4 Piecewise-defined examples

Some piecewise-defined functions are strictly convex even though they are not smooth. For example, a function may have a kink at one point while remaining strictly convex overall, as long as it never contains a linear segment. Such examples show that smoothness is not required for strict convexity.

5 Non-examples and boundary cases

Not all convex functions are strictly convex. Boundary cases help clarify where strictness begins and ends, especially when a function has flat regions or affine behavior.

5.1 Convex but not strictly convex functions

The function f(x) =xis convex but not strictly convex on R, because it is linear on each side of the origin and fails the strict inequality on many pairs of points. More generally, any function with a nontrivial affine segment is not strictly convex. These examples are useful because they sit just below the threshold of strictness.

5.2 Linear and affine functions

Any linear or affine function is convex, but none is strictly convex on a domain containing more than one point. The reason is that the defining inequality becomes equality along every segment. Such functions represent the limiting case where curvature disappears entirely.

5.3 Functions with flat regions

A convex function that is constant or linear on a subinterval, or more generally on a line segment in its domain, cannot be strictly convex. Flat regions destroy the strict inequality. This phenomenon often appears in optimization when penalties are not sufficiently curved to force a unique solution.

Strict convexity is also used for sets and for the geometry of epigraphs. These ideas connect function theory with convex geometry and provide a useful bridge between algebraic and geometric viewpoints.

6.1 Strict convexity of sets

A strictly convex set is a convex set whose boundary contains no line segments. Equivalently, the open segment joining any two distinct boundary points lies entirely in the interior. This notion mirrors strict convexity of functions and is often studied in normed spaces and geometric analysis.

6.2 Relationship between function and set convexity

A function is convex if and only if its epigraph is a convex set. Strict convexity of the function corresponds to a stronger geometric property: the epigraph has no flat boundary pieces coming from the graph of the function. This relationship allows function problems to be translated into set-theoretic geometry.

6.3 Epigraph interpretation

The epigraph of a strictly convex function has a boundary that curves upward without containing line segments on the graph portion. In practical terms, this means that the region above the graph is convex in a way that reflects genuine bending. Epigraph methods are especially important in optimization and variational analysis.

7 Applications

Strict convexity is widely used because it guarantees stability and uniqueness in problems where many candidate solutions may otherwise exist. Its role is especially prominent in optimization, numerical methods, economics, and statistics.

7.1 Optimization theory

In optimization, strict convexity is a standard sufficient condition for uniqueness of minimizers. It also improves the behavior of descent methods by preventing entire flat valleys of solutions. Many theoretical results become simpler when the objective is strictly convex.

7.1.1 Existence and uniqueness of solutions

If a strictly convex objective function is minimized over a suitable compact convex set, existence and uniqueness may be obtained together. Even when existence requires additional assumptions, strict convexity ensures that any solution found is the only one. This is valuable in both finite-dimensional and infinite-dimensional settings.

7.1.2 Regularization methods

Regularization often adds a strictly convex penalty term to an otherwise ill-posed or non-unique problem. Examples include quadratic penalties and squared-norm terms. The added curvature can force uniqueness and improve numerical stability.

7.2 Numerical analysis

Strict convexity helps ensure that iterative algorithms converge toward a single target rather than wandering among multiple minimizers. It also improves conditioning in many least-squares and estimation procedures. In practice, this can make gradient-based methods more reliable and easier to analyze.

7.3 Economics and utility theory

In economics, strict convexity of preferences or utility-related functions is associated with a preference for diversification. When a utility function is strictly convex or strictly concave, it often yields unique optimal choices under constraints. The notion appears in consumer theory, production, and risk analysis.

7.4 Statistical estimation

Many statistical estimators arise as minimizers of loss functions. When the loss is strictly convex, the estimator is unique if it exists. This is particularly important in regression, maximum likelihood methods, and penalized estimation, where uniqueness simplifies interpretation and computation.

8 Generalizations

Several stronger or related notions extend strict convexity in different directions. These variants are useful when additional quantitative control is needed, or when the domain has a more specialized geometric structure.

8.1 Strong convexity

Strong convexity strengthens strict convexity by requiring the function to lie above its chords by a quadratic margin. This quantitative gap leads to faster convergence estimates in optimization and gives stronger control over curvature. Every strongly convex function is strictly convex, but not conversely.

8.2 Uniform convexity

Uniform convexity is a global geometric strengthening often used for Banach spaces and norms. It requires a uniform inward bending condition that applies across the entire domain. In many contexts, it yields robust convergence and stability properties beyond those implied by mere strictness.

8.3 Strict quasiconvexity

Strict quasiconvexity is weaker than strict convexity. It requires sublevel sets to be convex and excludes equality in a specific comparison sense, but it does not demand the full Jensen inequality with strictness. This concept is useful when level-set geometry matters more than linear interpolation of function values.

8.4 Convexity in higher-dimensional spaces

In higher-dimensional vector spaces, strict convexity is defined on convex subsets of R^n or more general linear spaces. The same idea extends to functions on manifolds or abstract spaces when there is a suitable notion of line segment or geodesic. In these settings, the precise formulation depends on the ambient geometry.

Strict convexity is part of a broader family of convexity notions. These related ideas often differ by reversing inequalities, changing the type of strictness, or focusing on supporting geometry rather than pointwise interpolation.

9.1 Concave functions

A concave function satisfies the opposite inequality to a convex function. Instead of lying below its chords, it lies above them. Concavity is the natural dual notion to convexity and is often used for utilities, entropy, and maximization problems.

9.2 Strictly concave functions

Strictly concave functions are the mirror image of strictly convex ones. They satisfy a strict Jensen inequality in the opposite direction and typically have unique maximizers on convex domains. Many arguments for strict convexity have direct dual versions for strict concavity.

9.3 Midpoint convexity

Midpoint convexity requires the convexity inequality only at the midpoint of two points. It is weaker than full convexity and much weaker than strict convexity. Under regularity assumptions such as continuity, midpoint convexity can lead to ordinary convexity, but not automatically to strictness.

9.4 Supporting hyperplanes

Supporting hyperplanes are affine hyperplanes that touch a convex set without cutting through it. For convex functions, supporting hyperplanes correspond to tangent approximations and subgradients. Strict convexity often means that a supporting hyperplane touches the graph in a more limited way, reflecting the absence of flat regions.