In the mathematical discipline of descriptive set theory, a scale is a certain kind of object defined on a set of points in some Polish space (for example, a scale might be defined on a set of real numbers). Scales were originally isolated as a concept in the theory of uniformization, but have found wide applicability in descriptive set theory, with applications such as establishing bounds on the possible lengths of wellorderings of a given complexity, and showing (under certain assumptions) that there are largest countable sets of certain complexities.
Given a pointset A contained in some product space
where each X<sub>k</sub> is either the Baire space or a countably infinite discrete set, we say that a norm on A is a map from A into the ordinal numbers. Each norm has an associated prewellordering, where one element of A precedes another element if the norm of the first is less than the norm of the second.
A scale on A is a countably infinite collection of norms
with the following properties:
By itself, at least granted the axiom of choice, the existence of a scale on a pointset is trivial, as A can be wellordered and each φ<sub>n</sub> can simply enumerate A. To make the concept useful, a definability criterion must be imposed on the norms (individually and together). Here "definability" is understood in the usual sense of descriptive set theory; it need not be definability in an absolute sense, but rather indicates membership in some pointclass of sets of reals. The norms φ<sub>n</sub> themselves are not sets of reals, but the corresponding prewellorderings are (at least in essence).
The idea is that, for a given pointclass Γ, we want the prewellorderings below a given point in A to be uniformly represented both as a set in Γ and as one in the dual pointclass of Γ, relative to the "larger" point being an element of A. Formally, we say that the φ<sub>n</sub> form a Γ-scale on A if they form a scale on A and there are ternary relations S and T such that, if y is an element of A, then
where S is in Γ and T is in the dual pointclass of Γ (that is, the complement of T is in Γ). Note here that we think of φ<sub>n</sub>(x) as being ∞ whenever x∉A; thus the condition φ<sub>n</sub>(x)≤φ<sub>n</sub>(y), for y∈A, also implies x∈A.
The definition does not imply that the collection of norms is in the intersection of Γ with the dual pointclass of Γ. This is because the three-way equivalence is conditional on y being an element of A. For y not in A, it might be the case that one or both of S(n,x,y) or T(n,x,y) fail to hold, even if x is in A (and therefore automatically φ<sub>n</sub>(x)≤φ<sub>n</sub>(y)=∞).
The scale property is a strengthening of the prewellordering property. For pointclasses of a certain form, it implies that relations in the given pointclass have a uniformization that is also in the pointclass.