18 minute read

Our next primary theme is the concept of differentiation: differentiating a measure $\nu$ with regard to another measure $\mu$ on the same $\sigma$-algebra. To do that, it is useful to extend the notion of measure so as to allow measures to have negative values. And signed measure forms a basic building block for steps required to deal with the differentiation, such as Radon-Nikodym theorem. It has the countable additivity property of measures, but are allowed to take negative as well as positive values.

There are two main results in this post.

  1. Hahn decomposition theorem
    If $\mu$ is a signed measure, then $X$ can be decomposed as the union of two disjoints sets $P$ and $N$ such that $\mu$ behaves like a positive measure (the measures we have been considering up until now) on $P$ and $-\mu$ behaves like a positive measure on $N$.
  2. Jordan decomposition theorem
    A signed measure can be written as the difference of two positive measures.


Signed measures

$\color{blue}{\mathbf{Definition.}}$ Signed Measure
Let $\mathcal{A}$ be a $\sigma$-algebra. A signed measure is a function $\mu: \mathcal{A} \to (-\infty, \infty]$ such that
  1. $\mu (\emptyset) = 0$
  2. countable additivity ($\sigma$-additivity): if $A_i \in \mathcal{A}$, $i = 1, 2, \dots$, are pairwise disjoint, i.e. $A_i \cap A_j = \emptyset$ if $i \neq j$, then $$ \begin{aligned} \mu \left( \bigcup_{i=1}^\infty A_i \right) = \sum_{i=1}^\infty \mu(A_i). \end{aligned} $$ where the series converges absolutely if $\mu \left( \bigcup_{i=1}^\infty A_i \right)$ is finite.
When we emphasize that a measure defined as so far and only takes non-negative values, we refer to it as a positive measure.


Here, we need the absolute convergence to ensure that the order of summation doesn’t matter.

Intuitively, the next definition defines regions of $X$ that behaves as if $\mu$ is a positive measure (on positive set), and $-\mu$ is a positive measure (on negative set).

$\color{blue}{\mathbf{Definition.}}$ Positive and Negative sets
Let $\mu$ be a signed measure on $\mathcal{A}$. A set $A \in \mathcal{A}$ is called a positive set for $\mu$ if $\mu (B) \geq 0$ whenever $B \subset A$ and $B \in \mathcal{A}$. On the other hand, a set $A \in \mathcal{A}$ is called a negative set for $\mu$ if $\mu (B) \leq 0$ whenever $B \subset A$ and $B \in \mathcal{A}$. A null set $A \in \mathcal{A}$ is one where $\mu (B) = 0$ whenever $B \subset A$ and $B \in \mathcal{A}$.


Note that a null set for a positive measure is of course a null set in the above sense. Also, in the same way with positive measures, we can show continuity from below and above of signed measure by exploiting $\sigma$-additivity.

$\color{#bf1100}{\mathbf{Proposition.}}$
  1. continuity from below: Suppose $A_i \in \mathcal{A}$ and $A_i \uparrow A$. Then $\mu (A) = \text{lim}_{n \to \infty} \mu (A_n)$.
  2. continuity from above: Suppose $A_i \in \mathcal{A}$ and $A_i \downarrow A$. If $\mu (A_1) < \infty$, then $\mu (A) = \text{lim}_{n \to \infty} \mu (A_n)$.

$\mathbf{Proof.}$

(1) Let $B_1 = A_1$, $B_2 = A_2 - A_1$, $B_3 = A_3 - (A_2 \cup A_1)$, $B_4 = A_4 - (A_3 \cup A_2 \cup A_1)$, and in general $B_i = A_i - \bigcup_{j=1}^{i-1} A_j$.

Obviously $B_i$ are pairwise disjoint, $\bigcup_{i=1}^n B_i = \bigcup_{i=1}^n A_i$, and $B_i \uparrow A$. Then $\bigcup_{i=1}^\infty B_i = A = \bigcup_{i=1}^\infty A_i$, therefore

\[\begin{aligned} \mu (A) & = \mu \left(\bigcup_{i=1}^\infty A_i\right) = \mu \left(\bigcup_{i=1}^\infty B_i\right) = \sum_{i=1}^\infty \mu (B_i) \\ & = \lim_{n \to \infty} \sum_{i=1}^n \mu (B_i) = \lim_{n \to \infty} \mu \left(\bigcup_{i=1}^n B_i\right) = \lim_{n \to \infty} \mu \left(\bigcup_{i=1}^n A_i \right) \\ \end{aligned}\]


(2) Define $B_1 = \emptyset$ and $B_i = A_{1} - A_{i}$ for $i \geq 2$. Then $B_i \uparrow A_1 - A$. Thus

\[\begin{aligned} \mu(A_1 - A) = \mu (A_1) - \mu(A) = \lim_{n \to \infty} \mu (A_1 - A_n) = \lim_{n \to \infty} \left[ \mu (A_1) - \mu(A_n) \right]. \end{aligned}\] \[\tag*{$\blacksquare$}\]


Here is an example of signed measure.

$\color{green}{\mathbf{Example.}}$

Suppose $m$ is Lebesgue measure and define

\[\begin{aligned} \mu (A) = \int_A f \; dm \end{aligned}\]

for some integrable $f$. If $f$ takes both positive and negative values, then $\mu$ will be a signed measure.

If we let

\[\begin{aligned} P = \{ x : f(x) \geq 0\} \text { and } N = \{ x: f(x) < 0 \} \end{aligned}\]

it is easy to show that $P$ and $N$ are positive and negative set, respectively.

The Hahn decomposition theorem provides a decomposition of $X$ into positive and negative sets $P$ and $N$. This decomposition is unique except that \(C = \{x : f (x) = 0 \}\) could be included in $N$ instead of $P$, or apportioned partially to $P$ and partially to $N$. Note that $C$ is a null set, however.

And the Jordan decomposition theorem provides a decomposition of $\mu$ into $\mu^+$ and $\mu^-$, where $\mu^+ (A) = \int_A f^+ \; dm$ and $\mu^- (A) = \int_A f^- \; dm$.

\[\tag*{$\blacksquare$}\]


Let’s conclude the section with the important results that we can always extract a negative subset with negative measure from any measurable set with negative measure.

$\color{#bf1100}{\mathbf{Proposition.}}$ Let $\mu$ be a signed measure. Let $E \in \mathcal{A}$ and $\mu (E) < 0$. Then, there exists a measurable subset $F \subset E$ that is a negative set with $\mu (F) < 0$.
$\mathbf{Proof.}$

Step 1. If $E$ is a negative set, we are done. If not, there exists a measureable subset of $E$ with positive measure. Then let $n_1$ be the smallest integer such that there exists $E_1 \subset E$ satisfying $E_1 \in \mathcal{A}$ and $\mu (E_1) \geq \frac{1}{n_1}$.

Step 2. And by induction, we can define pairwise disjoint measurable sets $E_2, E_3, \cdots$. Let $k \geq 2$. $E_1, \dots E_{k-1}$ are pairwise disjoint measurable contained in $E$, satisfying $\mu (E_i) > 0$ for $i = 1, \cdots, k-1$.

Let

\[\begin{aligned} F_k = E - \bigcup_{n=1}^{k-1} E_n. \end{aligned}\]

Then $\mu (E) = \mu (F_k) + \sum_{n=1}^{k-1} \mu (E_n)$, thus $\mu (F_k) \leq \mu (E) < 0$.

  • If $F_k$ is a negative set, then $F_k$ is the desired $F$. We stop the construction of $E_{k + 1}$ and $F_{k + 1}$.
  • If not, let $n_{k + 1} \leq n_{k}$ be the smallest positive integer such that there exists $E_{k + 1} \subset F_{k+1}$ satisfying $E_{k + 1} \in \mathcal{A}$ and $\mu (E_{k+1}) \geq \frac{1}{n_{k + 1}}$.

Step 3. But it is possible that we obtain infinitely many $F_k$. Then set

\[\begin{aligned} F = \bigcap_{k=1}^\infty F_k = E - \left( \bigcup_{k=1}^\infty E_k \right). \end{aligned}\]

Note that we have $E = F \cup (\bigcup_{k=1}^\infty E_k )$, therefore

\[\begin{aligned} \mu (E) = \mu (F) + \sum_{k=1}^\infty \mu (E_k) \end{aligned}\]

Since $\mu (E) < 0$ and $\mu (E_k) > 0$, $\mu (F) < \mu (E) < 0$. Note that from the definition of signed measure, $\mu (F) > -\infty$. Hence

\[\begin{aligned} \sum_{k=1}^\infty \mu (E_k) < \infty \end{aligned}\]

As $\mu (E_k) \geq \frac{1}{n_k}$, $\sum_{k=1}^\infty \frac{1}{n_k} < \infty$ thus $n_k \to \infty$ as $k \to \infty$. With this fact, we now claim that $F$ is a negative set.

$\mathbf{Claim.}$ $F$ is a negative set
$\mathbf{Proof.}$

Suppose $G \subset F$ is measurable. If $\mu (G) > 0$, then $\mu (G) > \frac{1}{N}$ for some $N \in \mathbb{N}$. Then, for some $k$, $n_k > N$ as $n_k \to \infty$ as $k \to \infty$.

Then we would have the set $G$ instead of $E_k$ at stage $k$ and $n_k \leq N$. But this contradicts the construction as $F$ is constructed by subtracting all $E_k$ from $E$. Hence $\mu (G) \leq 0$ whenever $G \subset F$ and $G \in \mathcal{A}$.

\[\tag*{$\blacksquare$}\]



Hahn decomposition theorem

Now we introduce Hahn decomposition theorem. This will be useful for proving the further theorems called Radon-Nikodym theorem that connects measure theory and differentiation.

$\color{red}{\mathbf{Theorem.}}$ Hahn Decomposition Theorem
Let $\mu$ be a signed measure taking values in $(- \infty, \infty]$.
  1. There exist disjoint measurable sets $E$ and $F$ such that $E \cup F = X$, $E$ is a negative set and $F$ is a positive set;
  2. If $E^\prime$ and $F^\prime$ are another such pair, then $E \Delta E^\prime = F \Delta F^\prime$ is a null set with regard to $\mu$;
  3. If $\mu$ is not a positive measure, then $\mu(E) < 0$. If $-\mu$ is not a positive measure, then $\mu(F) > 0$.
$\mathbf{Proof.}$

(1) Note that $\emptyset$ is a negative set. Let

\[\begin{aligned} L = \text{inf } \{ \mu (A) : A \text{ is a negative set } \}. \end{aligned}\]

therefore $L \leq 0$. Then by definition of infimum we can choose negative sets $A_n$ such that $\mu (A_n) \to L$ as $n \to \infty$.

Let

\[\begin{aligned} E & = \bigcup_{n=1}^\infty A_n \\ B_n & = A_n - \left( \bigcup_{k=1}^{n-1} A_k \right) \end{aligned}\]

Note that $B_n$ is a negative set and \(\{ B_n \}\) are pairwise disjoint, increasing to $E$. We claim that $E$ is a negative set.

$\mathbf{Claim.}$ $E$ is a negative set.
$\mathbf{Proof.}$

Suppose $C \subset E$ is measurable. Then $C = \bigcup_{n=1}^\infty (C \cap B_n)$. We have

\[\begin{aligned} \mu (C) & = \lim_{n \to \infty} \mu \left( C \cap \left( \bigcup_{k=1}^n B_k \right) \right) \\ & = \lim_{n \to \infty} \sum_{k=1}^n \mu \left( C \cap B_k \right) \\ & \leq 0. \end{aligned}\] \[\tag*{$\blacksquare$}\]


Hence $\mu (E) = \mu (A_n) + \mu (E - A_n) \leq \mu (A_n)$ for any $n \in \mathbb{N}$. Letting $n \to \infty, \mu (E) \leq L$. But since $E$ is negative set, we have $\mu (E) \geq L$ thus $\mu (E) = L > -\infty$.

Now, let $F = E^c$. We claim that $F$ is a positive set.

$\mathbf{Claim.}$ $F$ is a positive set.
$\mathbf{Proof.}$

Suppose not. Then there exists a subset $B \subset F$ such that $\mu (B) < 0$. By the proposition above, there exists a measurable set $C \subset B$ with $\mu (C) < 0$. Then $E \cup C$ is a negative set and $\mu (E \cup C) = \mu (E) + \mu (C) < \mu (E) = L$, which is contradiction.

\[\tag*{$\blacksquare$}\]



(2) Let $A \subset E - E^\prime$ and $A \in \mathcal{A}$. Since $A \subset E$, $\mu (A) \leq 0$. But $A \subset E - E^\prime = F^\prime - F \subset F^\prime$, thus $\mu (A) \geq 0$. Hence $\mu (A) = 0$.

Similarly, we can prove that $\mu (A) = 0$ if $A \subset E^\prime - E$ and $A \in \mathcal{A}$. Therefore if $A \subset E \Delta E^\prime$ and $A \in \mathcal{A}$, $\mu (A) = 0$.


(3) Suppose $\mu$ is not a positive measure but $\mu (E) = 0$. Since $\mu$ is not a positive measure, there exists $A \in \mathcal{A}$ such that $\mu (A) < 0$.

And such $A \in \mathcal{A}$, we have

\[\begin{aligned} \mu (A) & = \mu (A \cap E) + \mu (A \cap F) \\ & \geq \mu (E) + \mu (A \cap F) \\ & \geq 0 + \mu (A \cap F) \geq 0 \end{aligned}\]

since $F$ is a positive set, which is a contradiction.

\[\tag*{$\blacksquare$}\]



Jordan decomposition theorem

The decomposition of $X$ in Hahn decomposition theorem is usually called a Hahn decomposition for $\mu$, and not unique in general. But, it leads to the Jordan decomposition theorem which states that the existence and uniqueness of a canonical representation of $\mu$ as the difference of two positive measures.

To state the theorem, we introduce a new concept called mutually singular: Two measures that are mutually singular live on disjoint sets;

$\color{blue}{\mathbf{Definition.}}$ Mutually singular
Two positive measures $\mu$ and $\nu$ are called mutually singular and write $\mu \perp \nu$ if there exists $E, F \in \mathcal{A}$ with $$ \begin{aligned} E \cap F & = \emptyset \\ E \cup F & = X \end{aligned} \\ \begin{aligned} \mu (E) = \nu (F) = 0 \end{aligned} $$


For example, let $\mu$ be a Lebesgue measure. For $\nu$, consider Cantor-Lebesgue fuction $f$. Recall that it is increasing and continuous. Therefore, we can construct a Lebesgue-Stieltjes measure $\nu$ associated with $f$.

Then we claim that $\mu \perp \nu$ and $E = C$, $F = C^c$ where $C$ is Cantor set. Since $\mu (E) = 0$, we should show $\nu (F) = 0$. Note that Cantor set is closed, thus $F$ is open that implies $F$ can be written as disjoint countable union of open intervals.

Hence, it suffices to show that $\nu (I) = 0$ for every open intervals contained in $F$. And this will follow if we show $\nu (J) = 0$ for every interval of the form $J = (a, b]$ contained in $F$. But by construction of $f$, such $J$ must be satisfies $\nu (J) = f(b) - f(a) = 0$ as $b, a \in C^c$.

The following theorem, called Jordan decomposition theorem, is the second main result of this post. It can be derived with Hahn decomposition theorem.

$\color{red}{\mathbf{Theorem.}}$ Jordan Decomposition Theorem
If $\mu$ is a signed measure on a measure space $(X, \mathcal{A})$, there exist positive measures $\mu^+$ and $\mu^-$ such that $$ \begin{aligned} \mu = \mu^+ - \mu^- \end{aligned} $$ where $\mu^+$ and $\mu^-$ are mutually singular. And this decomposition is unique.
$\mathbf{Proof.}$
  1. Existence
    Let $X = E \cup F$ be a Hahn decomposition for $\mu$ where $E$ is a negative set and $F$ is a positive set. Define $$ \begin{aligned} \mu^+ (A) & := \mu (A \cap F) \\ \mu^- (A) & := - \mu (A \cap E) \end{aligned} $$ Then $\mu^+$ and $\mu^-$ satisfies the desired properties.
  2. Uniqueness
    Let $\mu = \nu^+ - \nu^-$ be another decomposition with $\nu^+ \perp \nu^-$. Since $\nu^+ \perp \nu^-$, let $E^\prime$ and $F^\prime$ be measurable sets such that $$ \begin{aligned} E^\prime \cup F^\prime & = X \\ E^\prime \cap F^\prime & = \emptyset \end{aligned} $$ and $\nu^+ (E^\prime) = 0$ and $\nu^- (F^\prime) = 0$. We claim that $F^\prime$ and $E^\prime$ are a positive set and negative set for $\mu$, respectively:
    $\color{black}{\mathbf{Claim.}}$ $F^\prime$ and $E^\prime$ are a positive set and negative set for $\mu$, respectively.
    $\mathbf{Proof.}$

    Let $A \subset F^\prime$ and $A \in \mathcal{A}$. Note that $\nu^-$ is a positive measure and $\nu^- (F^\prime) = 0$, so that $\nu^- (A^\prime) = 0$. Then

    \[\begin{aligned} \mu (A) = \nu^+ (A) - \nu^- (A) = \nu^+ (A) \geq 0 \end{aligned}\]

    which shows $F^\prime$ is a positive set for $\mu$. Similarly, we can show that $E^\prime$ is a negative set for $\mu$.

    \[\tag*{$\blacksquare$}\]

    Thus, by Hahn decomposition theorem, $F \Delta F^\prime = E \Delta E^\prime$ are null sets for $\mu$. With this fact, we can show the uniqueness of decomposition:
    $\color{black}{\mathbf{Claim.}}$ $\nu^+ (A) = \mu^+ (A)$ for any $A \in \mathcal{A}$.
    $\mathbf{Proof.}$

    Since $\nu^+ (E^\prime) = 0$ and $\nu^- (F^\prime) = 0$, we have

    \[\begin{aligned} \nu^+ (A) & = \nu^+ (A \cap F^\prime) + \nu^+ (A \cap E^\prime) \\ & = \nu^+ (A \cap F^\prime) + 0 \\ & = \nu^+ (A \cap F^\prime) - \nu^- (A \cap F^\prime) \\ & = \mu (A \cap F^\prime) \\ & = \mu (A \cap F) - \mu (A \cap (F - F^\prime)) \\ & = \mu (A \cap F) \\ & := \mu^+ (A) \end{aligned}\]

    Note that $\mu (A \cap (F - F^\prime)) = 0$ as $F \Delta F^\prime = (F - F^\prime) \cup (F^\prime - F)$ is null set for $\mu$.

    \[\tag*{$\blacksquare$}\]

    Similarly, we can show that $\nu^- = \mu^-$.
\[\tag*{$\blacksquare$}\]


In the proof, we constructed $\mu^+$ and $\mu^-$ as restrictions of $\mu$ to Hahn decomposition, but we can find the Jordan decomposition in another way:

$\color{#bf1100}{\mathbf{Proposition.}}$ Let $\mu$ be a signed measure on $(X, \mathcal{A})$. Then $$ \begin{gathered} \mu^{+}(A)=\sup \; \{\mu(B): B \in \mathcal{A}, B \subset A\} \\ \mu^{-}(A)=-\inf \; \{\mu(B): B \in \mathcal{A}, B \subset A\} \end{gathered} $$
$\mathbf{Proof.}$

Let any $B \in \mathcal{A}$ such that $B \subset A$ be given. Then

\[\begin{aligned} \mu (B) = \mu^+ (B) - \mu^- (B) \leq \mu^+ (B) \leq \mu^+ (A). \end{aligned}\]

Also, let $E \cup F = X$ be a Hahn decomposition that defines $\mu^+ (A) = \mu (A \cap F)$. Since $A \cap F \in \mathcal{A}$ and $A \cap F \subset A$, we have

\[\begin{aligned} \mu^+ (A) = \mu (A \cap F) \leq \sup \; \{\mu(B): B \in \mathcal{A}, B \subset A\} \end{aligned}\]

We can prove $\mu^-$ case in analogous way.

\[\tag*{$\blacksquare$}\]


Also, one can easily show that for a signed measure $\lambda$, $\lambda^+$ and $\lambda^-$ in Jordan decomposition are the smallest of those positive measures $\mu$ and $\nu$ that satisfy $\lambda = \mu - \nu$. Specifically,

$\color{#bf1100}{\mathbf{Proposition.}}$ Let $(X, \mathcal{A})$ be a measurable space. Suppose $\lambda = \mu - \nu$ where $\mu$ and $\nu$ are finite positive measures. Then $$ \begin{aligned} \mu (A) & \geq \lambda^+ (A) \\ \nu (A) & \geq \lambda^- (A) \\ \end{aligned} $$ Note that it is trivial when $\mu$ and $\nu$ are not finite. Hence we may assume they are finite.
$\mathbf{Proof.}$

First, we have to prove that $\lambda$ is a signed measure. $\lambda(\emptyset) = 0$ obviously. Let \(\{A_i \}\) be given pairwise disjoint in $\mathcal{A}$. Since $\mu$, $\nu$ are finite, $\lambda$ is also finite, and we have

\[\begin{aligned} \lambda\left(\cup_{i=1}^{\infty} A_i\right) & =\mu\left(\cup_{i=1}^{\infty} A_i\right)-\nu\left(\cup_{i=1}^{\infty} A_i\right) \\ & =\sum_{i=1}^{\infty} \mu\left(A_i\right)-\sum_{i=1}^{\infty} \nu\left(A_i\right) \\ & =\sum_{i=1}^{\infty} \mu\left(A_i\right)-\nu\left(A_i\right) \\ & =\sum_{i=1}^{\infty} \lambda\left(A_i\right) . \end{aligned}\]

Note that $\lambda = \mu − \nu$ is finite. And $\sum_{i=1}^\infty \lambda (A_i)$ is a difference of two absolutely convergence series, and thus it is absolutely converges. More precisely,

\[\begin{aligned} \sum_{i=1}^{\infty}\left|\lambda\left(A_i\right)\right| \leq \sum_{i=1}^{\infty}\left|\mu\left(A_i\right)\right|+\left|\nu\left(A_i\right)\right|<\infty \quad(\because \text { Def of finite signed measure }) \end{aligned}\]

Thus, $\lambda$ is a finite signed measure on $(X, \mathcal{A})$.

Let $\lambda = \lambda^+ - \lambda^-$ be Jordan decomposition with positive set $F$ and negative set $E$ such that $X = E \cup F$, $E \cap F = \emptyset$, and $\lambda^+ (E) = \lambda^+(F) = 0$. We have

\[\begin{aligned} & \mu (A) \geq \mu (A \cap F) \geq \mu (A \cap F) - \nu (A \cap F) = \lambda^+ (A) \\ & \nu (A) \geq \nu (A \cap E) \geq \nu (A \cap E) - \mu (A \cap E) = \lambda^- (A) \\ \end{aligned}\] \[\tag*{$\blacksquare$}\]


This proposition will be useful for further discussion. And we define total variation of measure that plays a similar role with modulus.

$\color{blue}{\mathbf{Definition.}}$ Total variation
In Jordan decomposition of $\mu$, $\mu^+$ and $\mu^-$ are called the positive and negative variations of $\mu$. Furthermore, we define the total variation of $\mu$ to be the measure $| \mu |$ defined by $$ \begin{aligned} | \mu | = \mu^+ + \mu^-. \end{aligned} $$


Equivalently, total variation of measure can be defined in different way:

$\color{red}{\mathbf{Theorem.}}$ Equivalent definitions
Let $\mu$ be a signed measure on $(X, \mathcal{A})$. Then $$ |\mu|(A)=\sup \left\{\sum_{j=1}^n\left|\mu(B_j)\right|: \operatorname{each} B_j \in \mathcal{A} \text{ the } B_j \text{ are disjoint, } \cup_{j=1}^n B_j=A \right\} $$
$\mathbf{Proof.}$

Let \(\{ B_j \}\) be any collection of pairwise disjoint measurable set, i.e. $B_j \in \mathcal{A}$, such that $\cup_{j=1}^n B_j = A$. Then

\[\begin{aligned} \sum_{j=1}^n | \mu (B_j) | \leq \sum_{j=1}^n | \mu | (B_j) = | \mu | (\cup_{j=1}^n B_j = A). \end{aligned}\]

Let $B_1 = A \cap E$ and $B_2 = A \cap F$ where $\mu^+ (A) = \mu (A \cap F)$ and $\mu^- (A) = \mu (A \cap E)$. Then

\[\begin{aligned} |\mu(B_1)| + |\mu(B_2)| = \mu^+ (A) + \mu^- (A) = | \mu | (A). \end{aligned}\] \[\tag*{$\blacksquare$}\]


Then, we have the following results:

$\color{#bf1100}{\mathbf{Proposition.}}$ Let $(X, \mathcal{A})$ be a measurable space. Suppose that $\mu$ and $\nu$ are finite signed measures on $(X, \mathcal{A})$. Then $$ \begin{aligned} | \mu + \nu | (A) & \leq | \mu | (A) + | \nu | (A) \\ | \mu - \nu | (A) & \leq | \mu | (A) + | \nu | (A) \end{aligned} $$ for every $A \in \mathcal{A}$.
$\mathbf{Proof.}$

For the first inequality, we should prove that $\mu + \nu$ is a finite signed measure and it is trivial. Let $\lambda = \mu + \nu$. Then $\lambda = (\mu^+ + \nu^+) - (\mu^- + \nu^-)$, where $\mu^+ + \nu^+$ and $\mu^- + \nu^-$ are finite positive measure. Thus we have

\[\begin{aligned} \lambda^+ (A) \leq \mu^+ (A) + \nu^+ (A) \\ \lambda^- (A) \leq \mu^- (A) + \nu^- (A) \end{aligned}\]

For the second inequality, we should prove that $\mu - \nu$ is a finite signed measure and it is trivial. We will show that $\mu^+(A) + \nu^−(A) \geq \lambda^+(A)$ and $\mu^−(A) + \nu^+(A) \geq \lambda^– (A)$ for every $A \in \mathcal{A}$.

\[\begin{aligned} \lambda^{+}(A) & =\lambda(A \cap F)=\mu(A \cap F)-\nu(A \cap F) \\ & =\mu^{+}(A \cap F)-\mu^{-}(A \cap F)-\nu^{+}(A \cap F)+\nu^{-}(A \cap F) \\ & \leq \mu^{+}(A \cap F)+\nu^{-}(A \cap F) \leq \mu^{+}(A)+\nu^{-}(A) \end{aligned}\]

Likewise,

\[\begin{aligned} \lambda^{-}(A) & =-\lambda(A \cap E)=-\mu(A \cap E)+\nu(A \cap E) \\ & =-\mu^{+}(A \cap E)+\mu^{-}(A \cap E)+\nu^{+}(A \cap E)-\nu^{-}(A \cap E) \\ & \leq \nu^{+}(A \cap E)+\mu^{-}(A \cap E) \leq \mu^{-}(A)+\nu^{+}(A) \end{aligned}\]

Thus we conclude that

\[\begin{aligned} | \lambda | (A) = | \mu - \nu | (A) \leq | \mu (A) | + | \nu (A) |. \end{aligned}\] \[\tag*{$\blacksquare$}\]


There are some useful properties related to $\mu$ and $| \mu |$. Here are examples.

$\color{#bf1100}{\mathbf{Proposition.}}$ For a signed measure $\mu$, $A$ is a null set for $\mu$ if and only if $| \mu | (A) = 0$.
$\mathbf{Proof.}$

Let $\mu^+ (B) = \mu (B \cap F)$ and $\mu^- (B) = \mu (B \cap E)$ as in the proof of Jordan decomposition theorem.

Suppose $A$ is a null set. Then $\mu (A \cap F) = \mu^+ (A) = 0$ and $\mu (A \cap E) = \mu^- (A) = 0$, thus $| \mu | (A) = 0$.

Suppose $| \mu | (A) = 0$. In other words, $\mu^+ (A) = 0$ and $\mu^- (A) = 0$. For any $B \subset A$ and $B \in \mathcal{A}$, note that $\mu^+(B) \leq \mu^+ (A) = 0$, and $\mu^-(B) \leq \mu^- (A) = 0$. Hence

\[\begin{aligned} \mu (B) = \mu^+ (B) - \mu^- (B) = 0. \end{aligned}\] \[\tag*{$\blacksquare$}\]





Reference

[1] Richard F. Bass, Real Analysis for Graduate Students, Version 4.3
[2] Folland, Gerald B. Real analysis: modern techniques and their applications. Vol. 40. John Wiley & Sons, 1999.
[3] W. Rudin. Real and Complex Analysis, 3rd ed. McGraw-Hill, New York, 1987.

Leave a comment