An outer measure is a set function that assigns a size in to every subset of a given set, assigns zero to the empty set, and satisfies monotonicity and countable subadditivity. Unlike a measure, it need not be additive on disjoint sets. In measure theory, outer measures provide a way to assign a preliminary size to all subsets and then identify a family of measurable sets on which this size becomes a complete measure. The construction of Lebesgue measure is the principal example.
Definition
Let be a set and let denote its power set. A function
is an outer measure if it satisfies the following conditions:
Zero on the empty set:
Monotonicity: if , then
Countable subadditivity: for every sequence of subsets of ,
The asterisk conventionally distinguishes an outer measure, defined on all subsets, from a measure defined on a specified σ-algebra. Infinite values are allowed.
Subadditivity says that the size of a union cannot exceed the total size of its pieces. It does not require equality, even when those pieces are disjoint. This weaker requirement makes it possible to assign sizes to sets that are not measurable in the usual sense.
Construction by countable coverings
A standard way to obtain an outer measure is to cover a set by simpler sets with assigned costs.
Let be a collection of subsets of , and let
assign a cost to each member. Define
An empty covering is allowed for the empty set and has cost zero; if no covering exists, the infimum is taken to be . With these conventions, the construction gives an outer measure.
The defining properties follow from the behavior of coverings:
- A covering of also covers every subset , giving monotonicity.
- Countable coverings of countably many sets can be combined into one countable covering of their union, giving countable subadditivity.
- The empty covering gives the empty set size zero.
When the relevant outer measures are finite, the subadditivity proof chooses coverings whose costs differ from their infima by at most . Combining them gives the required bound up to , and then is allowed to tend to zero.
The resulting outer measure need not equal the assigned cost on a covering set : a cheaper covering of may exist. Additional assumptions are needed to ensure agreement.
Carathéodory measurability
An outer measure determines which sets can split every other set without losing additivity.
A subset is Carathéodory measurable with respect to if, for every ,
Countable subadditivity already gives
Consequently, checking measurability amounts to establishing the reverse inequality for every test set .
The requirement is stronger than additivity for the single decomposition . A measurable set must split every subset of , not just the ambient space.
Carathéodory’s theorem
For any outer measure , the collection
is a σ-algebra. Moreover, the restriction
is a countably additive measure: if are pairwise disjoint, then
This measure is also complete: every subset of a measurable set of measure zero is measurable and has measure zero. Indeed, if , then every subset has outer measure zero by monotonicity, and the measurability criterion holds for . Thus all outer-measure-zero sets are null sets for the resulting measure.
This theorem supplies the bridge between a size function on all subsets and a genuine measure on an appropriate domain.
Lebesgue outer measure
For , Lebesgue outer measure is defined by
where the covering intervals have finite endpoints.
The definition measures from outside by countable interval coverings. In higher-dimensional Euclidean space, intervals are replaced by boxes and lengths by volumes.
Some basic examples are:
- A singleton has outer measure zero, because it can be covered by an arbitrarily short interval.
- Every countable set has outer measure zero, by countable subadditivity.
- An interval has outer measure equal to its length, irrespective of whether its endpoints are included.
- The usual middle-thirds Cantor set has outer measure zero, although it is uncountable.
The Carathéodory-measurable sets for are exactly the Lebesgue measurable sets. They include all sets in the Borel σ-algebra and all subsets of Borel null sets.
Outer measure is nevertheless defined on non-Lebesgue-measurable sets. For such a set, the failure is not the absence of an assigned size: it is the failure of the splitting identity required for measurability. In the usual set-theoretic framework with the axiom of choice, non-Lebesgue-measurable subsets of exist.
Extending a premeasure
A premeasure is a nonnegative set function on an algebra of sets that vanishes on the empty set and is countably additive whenever the union of the disjoint sequence remains in that algebra.
If is a premeasure on an algebra , define
Then:
- is an outer measure;
- for every ;
- every member of is Carathéodory measurable.
Restricting to the σ-algebra generated by therefore extends the premeasure to a measure. This is the central construction in the Carathéodory extension theorem.
If the premeasure is σ-finite—meaning that is a countable union of sets in with finite premeasure—the extension to the generated σ-algebra is unique. Without σ-finiteness, uniqueness can fail.
The full Carathéodory σ-algebra may be larger than the σ-algebra generated by . Under σ-finiteness, the measure on this larger domain is the completion of the extension on the generated σ-algebra.
Metric outer measures and Hausdorff measure
On a metric space , an outer measure is called a metric outer measure if
whenever and are separated by a positive distance:
This additional property ensures that every closed set, and hence every Borel set, is Carathéodory measurable.
An important example is Hausdorff measure. For , one defines, up to a conventional normalization factor,
and then sets
As decreases, fewer coverings are allowed, so the displayed limit exists in . The resulting function is a metric outer measure. Its restriction to Borel sets is particularly useful for studying curves, surfaces, and fractals, including sets whose natural dimension is not an integer.
Limitations and distinctions
Disjoint additivity can fail
Let have at least two elements and define
This is an outer measure. For disjoint nonempty and ,
Its only Carathéodory-measurable sets are and .
Continuity properties require care
Measures are continuous from below: if increases to , then their measures increase to the measure of . An arbitrary outer measure need not have this property.
For example, on , define
This is an outer measure. Yet for ,
The usual continuity results apply after restriction to measurable sets, where countable additivity is available. Continuity from above additionally requires a finite-measure assumption.
“Outer” does not always mean open approximation
Although Lebesgue outer measure can be computed using open coverings, an abstract outer measure need not involve a topology at all. The axioms alone do not imply approximation by open sets, compact sets, or measurable supersets of equal size. Such regularity properties require further hypotheses.
Historical development
Lebesgue’s work on measure and integration at the beginning of the twentieth century established the importance of assigning size through countable coverings. Carathéodory’s 1914 work formulated the general splitting criterion for measurability, separating the construction of an outer measure from the identification of its measurable sets. Hausdorff’s subsequent development of dimension-dependent measures extended covering methods beyond ordinary length, area, and volume.