A function f on a convex domain is quasiconvex if all its sublevel sets {x : f(x) ≤ alpha} are convex for every alpha; equivalently, f(tx + (1-t)y) ≤ max{f(x), f(y)} for t in [0,1].

Principle

Generalizes convexity by preserving convexity of sublevel geometry rather than linear interpolation inequalities; it captures unimodal or plateau-like behaviors while retaining global minimum structure.

Demonstration

The piecewise function f(x)=1 for x≤0 and f(x)=x+1 for x>0 on R has convex sublevel sets (intervals) and is quasiconvex but not convex because it lacks the Jensen linear bound across the kink at 0.

Misapplication

Assuming quasiconvexity implies all convex-optimization tools apply (e.g., subgradient descent guarantees); quasiconvex problems may lack the same duality and smoothness properties as convex problems.

Consequence

Quasiconvexity ensures every local minimizer is a global minimizer on a convex domain and that level-set methods and certain bisection strategies remain valid for global optimization.

Reversal

Convex functions are a strict subclass: convexity imposes linear combination bounds (Jensen) and stronger second-order conditions; replacing '≤ max' by '≤ weighted average' yields convexity.

Boundary

Defined on convex domains and concerned with sublevel-set convexity; does not imply differentiability, strict convexity, or the full suite of convex duality results; not equivalent to quasiconcavity.

Semantic Tension

Tension arises between quasiconvexity and convexity in algorithm design: some descent guarantees hold, but rates, dual formulations, and separation theorems are weaker or absent for quasiconvex functions.

Synthesis

A quasiconvex function is one whose sublevel sets are convex: it generalizes convexity by preserving global-minimum geometry while allowing nonconvex pointwise behavior that still admits certain global optimization strategies.