A signed integral can cancel: positive and negative contributions may each be substantial even when the result is zero. A signed measure records this cancellation on every measurable set. Its total variation records the mass before cancellation and provides the correct way to control errors. We construct these objects for the finite class described below, then connect them to densities.
Throughout is a measurable space. We work with the following explicit class, which avoids any unproved claim about arbitrary extended-valued signed measures.
Definition 3.3.1
(Finite signed measure presented as a difference)
A finite signed measure on is a set function of the form , that is for , where are finite nonnegative measures on .
For disjoint , , so countable additivity involves an absolutely convergent real series. Continuity from above and below follows by subtracting the corresponding identities for the two finite measures.
Positive and negative sets
The sign of a single value says little about the subsets of , so positivity is defined through all subsets.
Definition 3.3.2
(Positive and negative sets)
A measurable set is positive for if every measurable satisfies . A measurable set is negative for if every measurable satisfies .
Common mistake
(A positive total does not make a set positive)
The sign of alone is not enough. On with masses at and at , the whole set has total mass but contains the negative subset , so it is not a positive set.
The next theorem splits the whole space into one positive and one negative set.
Theorem 3.3.3
(Hahn decomposition for finite signed measures)
If is a finite signed measure, there are disjoint measurable with , where is positive and is negative.
The proof looks for a set on which is as large as possible.
Proof
(A set attaining the supremum)
Put . It is finite because . Choose measurable with . For finitely many such sets, the identity
and the fact that every set has measure at most give , and so, inductively,
Let . Continuity from above gives . The sets increase, so satisfies by continuity from below. If had , then , impossible. If had , then , also impossible. Thus is positive and is negative.
Proof X-ray
Why a maximizing set detects the sign of every subset
Finiteness enters twice. It makes a real number, and it makes the subtractions and the inclusion–exclusion identity legitimate, with no . The construction has two limits in a deliberate order. Intersections are taken first because the bound loses only the small defects , whose tail sums to ; the resulting sets increase, so a union then gives a set attaining exactly.
The key move is to use maximality against every subset at once. Removing a negative subset from , or adding a positive subset of its complement, would exceed the supremum. That is why one extremal value yields the two universal sign conditions, and why no subset-by-subset search is needed.
The sets need not be unique. A set on which the signed measure vanishes on every measurable subset can be moved between them without affecting the sign property. The measures constructed next, however, are unique.
Jordan decomposition and total variation
Fix a Hahn decomposition .
Definition 3.3.4
(Jordan parts and total variation)
The positive and negative parts of are
The total variation measure is . Two nonnegative measures are mutually singular, , if some measurable has and .
These are finite nonnegative measures, mutually singular because they are concentrated on the disjoint sets and , and . This is the Jordan decomposition. Its parts do not depend on the chosen Hahn sets.
Theorem 3.3.5
(Uniqueness of the Jordan decomposition)
If with finite nonnegative mutually singular measures, then for every measurable
and .
Both sides are compared with the same supremum, which mentions neither the Hahn sets nor the set .
Proof
(Both parts equal the same supremum)
First, . Since is positive and , the latter is at most ; the choice attains equality. So the supremum equals .
Conversely, for all . Let witness mutual singularity, so and . Then , so the supremum equals . Thus , and finite subtraction gives .
Counterexample mode
Without mutual singularity a difference is not unique
The claim “a finite signed measure has only one expression with ” is false. On let . Also : on both sides give , on both give . Here and both charge , so they are not mutually singular, and . Adding any finite to both parts produces further examples. The repair is the hypothesis . Every decomposition still satisfies and , because and . So the Jordan parts are the minimal choice.
The total variation can also be recovered without choosing a Hahn decomposition.
Proposition 3.3.6
(Total variation as a supremum over partitions)
For every measurable ,
In particular .
Proof
(An upper bound attained by the Hahn split)
For any finite partition, , and summing gives at most . The partition gives , so the supremum is attained. The trivial partition gives .
Common mistake
(The variation of a set is not the absolute value of its mass)
It is tempting to write . Only the inequality holds, and cancellation within can make it strict: in the first worked example below while . The map is in general not even additive.
Concept lensAnalytic
Total variation as the largest integral of a bounded function
For bounded measurable put . Then
Every such satisfies , and attains the bound. A finite partition is the special test function ; allowing every bounded test function gives nothing larger.
Signed densities and variation
Let be a finite nonnegative measure. We write if implies .
Corollary 3.3.7
(Signed Radon–Nikodym theorem)
If is a finite signed measure and , there is , unique -almost everywhere, with
Moreover with , , and .
The proof applies the nonnegative theorem to each Jordan part.
Proof
(Apply the positive theorem to each part)
For any -null , every measurable subset is also -null, so . The supremum formula therefore gives ; the same formula applied to , whose positive part is , gives . Both Jordan parts are absolutely continuous with respect to .
By the finite nonnegative Radon–Nikodym theorem, there are with and . Hence with .
Since , we get , so almost everywhere on ; likewise almost everywhere on . Consequently and almost everywhere, and .
For uniqueness, if two integrable functions have the same integral over every measurable set, apply the equality to and . Each has measure zero, and their countable union is .
This is a corollary of the already proved positive-measure theorem; it does not require differentiating a distribution function.
Worked examples
Worked example 3.3.8
(Cancellation without zero variation)
On define . The density is nonnegative exactly on , so one Hahn choice is , . Then
So , yet the signed measure is not zero: for instance .
Worked example 3.3.9
(An atom as well as a density)
Let on . The point carries mass although it is Lebesgue-null. The measurable sets and are positive and negative: subsets of have nonpositive integral and miss the atom. The total positive mass is , the total negative mass is , and . This signed measure is not absolutely continuous with respect to Lebesgue measure, since it assigns mass one to the null singleton . Its density describes only the integral part.
The Hahn sets can move when vanishes on a region, while the Jordan parts cannot.
Worked example 3.3.10
(A flat region: many Hahn sets, one Jordan decomposition)
Let on , on and on , and set on . Every measurable has on all its subsets. Hence , is a Hahn decomposition for every such , including the edge cases and . Whatever is chosen,
because . Thus while .
Practice with solutions
Quick check
(On let and . Find its Jordan parts and total variation.)
Find a Hahn decomposition first.
Solution · Answer
Take and . Then , , and , so while .
Quick check
(On let . Compute and .)
Use the density formula for the total variation.
Solution · Answer
, while .
Quick check
(Suppose and for every measurable . Why does , and how does supply this hypothesis?)
Use the supremum characterization of .
Solution · Answer
is the supremum of over measurable , and every such value is zero. Under , every measurable subset of the -null set is -null and therefore has -mass zero.
Quick check
(Can a Hahn positive set have negative signed mass on one of its measurable subsets?)
Compare with the definition of a positive set.
Solution · Answer
No: nonnegativity on every measurable subset is exactly the definition of a positive set. A positive total alone is weaker, as the two-point example shows.