Math

On the construction of the Lebesgue-Stieltjes measure

Step-by-step mathematics solution: On the construction of the Lebesgue-Stieltjes measure

As an Amazon Associate, I earn from qualifying purchases. For more practice problems like this, see Schaum’s Outline of Calculus, 7th Edition.


1. Restatement of the Problem

When constructing the Lebesgue-Stieltjes measure, we usually start with the family of bounded left-closed, right-open intervals $\mathcal{E} = {(a,b] \mid a, b \in \mathbb{R}}$ and a set function $\nu((a,b]) = F(b) - F(a)$. To apply Carathéodory’s Extension Theorem or use standard uniqueness arguments (like the Uniqueness of Extensions Lemma or Dynkin’s $\pi-\lambda$ theorem), we need our initial set function to be defined on an algebra of sets, not just a semiring.

The author of your notes expands $\mathcal{E}$ to a family $\mathcal{F}$ consisting of finite disjoint unions of intervals, where these intervals can be bounded, unbounded (like $(-\infty, b]$ or $(a, \infty)$), or even $\mathbb{R}$. The student’s confusion is twofold:

  1. Why does $\mathcal{F}$ contain unbounded intervals like $(-\infty, a]$ if $\mathcal{E}$ only contains bounded intervals?
  2. Why is $\mathcal{F}$ the smallest algebra containing $\mathcal{E}$? Specifically, how can we be sure that any algebra containing $\mathcal{E}$ must also contain the unbounded components needed to make $\mathcal{F}$ closed under complements?

2. Step-by-Step Solution

To clear up the confusion, let us rigorously define the objects involved and trace what the properties of an algebra require.

Step 1: Definition of an Algebra

A collection of subsets $\mathcal{A}$ of a set $X$ is called an algebra of sets if:

  1. The empty set $\emptyset \in \mathcal{A}$ (and hence $X \in \mathcal{A}$).
  2. If $A \in \mathcal{A}$, then its complement $A^c \in \mathcal{A}$ (closed under complementation).
  3. If $A, B \in \mathcal{A}$, then $A \cup B \in \mathcal{A}$ (closed under finite unions, and consequently finite intersections).

Step 2: What must any algebra containing $\mathcal{E}$ contain?

Let $\mathcal{A}$ be any arbitrary algebra on $\mathbb{R}$ that contains the family of bounded intervals $\mathcal{E} = {(a,b] \mid a, b \in \mathbb{R}}$.

By definition of an algebra, $\mathcal{A}$ must be closed under taking complements. Let us take any interval $(a, b] \in \mathcal{E}$. Its complement in $\mathbb{R}$ is: \((a,b]^c = (-\infty, a] \cup (b, \infty)\) Because $\mathcal{A}$ is an algebra and $(a,b] \in \mathcal{A}$, it must contain $(a,b]^c$.

Now, can we isolate the pieces $(-\infty, a]$ and $(b, \infty)$ using only the operations allowed in an algebra (finite unions, intersections, and complements)?

  • Notice that $(-\infty, a]$ is the union of all intervals of the form $(n, a]$ as $n \to -\infty$, but more importantly, in an algebra generated by $\mathcal{E}$, can we write $(-\infty, a]$ as a finite combination?
  • Actually, look at the intersection of complements of different intervals. For instance, take $a < c < b$. We have bounded intervals.
  • More directly, consider the unbounded ray $(-\infty, b]$. Can we get it? Yes! Notice that for any $a < b$, we can form the union: \((-\infty, b] = (-\infty, a] \cup (a, b]\) Wait, the student’s exact worry was: “Why must it contain the individual pieces $(-\infty, a]$ or $(b, \infty)$? As far as I can check, these cannot be obtained as finite unions of bounded intervals.”

Let us address this directly.

Step 3: Proving that unbounded rays must belong to any algebra containing $\mathcal{E}$

Let $\mathcal{A}$ be an algebra containing $\mathcal{E} = {(a,b] \mid a,b \in \mathbb{R}}$.

  1. For any $b \in \mathbb{R}$, consider the sequence of bounded intervals $(-n, b]$ for integers $n > 0$. Each of these is in $\mathcal{E}$, hence in $\mathcal{A}$.
  2. However, an algebra is only guaranteed to be closed under finite unions and intersections, not countable ones. So we cannot simply take $\bigcup_{n=1}^\infty (-n, b]$.
  3. Correction/Refinement: How do we obtain $(-\infty, b]$ using finite operations from $\mathcal{E}$? We cannot! An algebra containing only bounded left-closed, right-open intervals does not automatically contain $(-\infty, b]$ unless we explicitly expand our base space or include them.

This reveals that the author of your notes either:

  • Implicitly defined $\mathcal{E}$ from the very beginning to include the extended real line $-\infty$ and $\infty$ (i.e., allowing $a = -\infty$ or $b = \infty$), OR
  • Is defining the algebra $\mathcal{F}$ as the collection of finite disjoint unions of elements from an extended family $\mathcal{E}^* = {(a,b] \mid -\infty \leq a < b \leq \infty}$.

If the author defines $\mathcal{E}$ strictly as bounded intervals ($a, b \in \mathbb{R}$), then the student is entirely correct: the complement of $(a,b]$ is $(-\infty, a] \cup (b, \infty)$, which cannot be formed from bounded intervals alone. Therefore, to make it an algebra, unbounded intervals must be included in the starting family.

Step 4: The Corrected Foundation for Lebesgue-Stieltjes

To fix this gap (which is common in introductory measure theory texts when authors gloss over the endpoints), we define the foundational family of intervals to include points at infinity: \(\mathcal{E} = \{(a,b] \mid -\infty \le a < b \le \infty\}\) (Note: $(a, \infty]$ is interpreted as $(a, \infty)$, and $(-\infty, b]$ includes all $x \le b$.)

With this proper definition of $\mathcal{E}$ including the extended reals:

  1. Complements: The complement of $(a, b]$ (for finite $a, b$) is $(-\infty, a] \cup (b, \infty)$, which is a disjoint union of two rays from $\mathcal{E}$.
  2. Intersections: The intersection of two intervals $(a, b]$ and $(c, d]$ is another interval of the same type (possibly empty).
  3. Finite Disjoint Unions: Let $\mathcal{F}$ be the family of all sets that can be written as a finite disjoint union of intervals from $\mathcal{E}$.

We now show that $\mathcal{F}$ is an algebra:

  • Empty set: $\emptyset \in \mathcal{F}$ (as a 0-term union, or $(a,a]$).
  • Complements: If $A \in \mathcal{F}$, $A = \bigcup_{i=1}^n I_i$ (disjoint union of intervals $I_i \in \mathcal{E}$). Then $A^c = \left(\bigcup_{i=1}^n I_i\right)^c = \bigcap_{i=1}^n I_i^c$. Since the complement of any interval in $\mathcal{E}$ is a finite disjoint union of intervals in $\mathcal{E}$, and $\mathcal{F}$ is closed under finite intersections, $A^c \in \mathcal{F}$.
  • Unions: If $A, B \in \mathcal{F}$, they are finite disjoint unions of intervals. By cutting them against each other, their union can also be expressed as a finite disjoint union of intervals in $\mathcal{E}$, so $A \cup B \in \mathcal{F}$.

Step 5: Why is $\mathcal{F}$ the smallest algebra containing $\mathcal{E}$?

Any algebra $\mathcal{A}$ containing $\mathcal{E}$ must contain:

  1. All individual intervals in $\mathcal{E}$ (by assumption).
  2. All finite unions of these intervals (by closure under finite unions).
  3. All finite intersections and complements of these intervals.

Because any finite disjoint union of intervals in $\mathcal{E}$ is built using finite unions and intersections of elements in $\mathcal{E}$, every element of $\mathcal{F}$ must belong to any algebra $\mathcal{A}$ containing $\mathcal{E}$. Thus, $\mathcal{F} \subseteq \mathcal{A}$.

Since $\mathcal{F}$ is itself an algebra containing $\mathcal{E}$, it is the smallest such algebra.


3. Final Answer

The student’s intuition is entirely correct: if $\mathcal{E}$ is strictly restricted to bounded intervals with real endpoints ($a, b \in \mathbb{R}$), then the collection of finite disjoint unions cannot form an algebra because the complement of a bounded interval yields unbounded rays $(-\infty, a]$ and $(b, \infty)$ which cannot be formed from bounded intervals alone.

To make the construction work, the base family $\mathcal{E}$ must be defined to include extended intervals with endpoints at $-\infty$ and $\infty$: \(\mathcal{E} = \{(a,b] \mid -\infty \le a < b \le \infty\}\)

With $\mathcal{E}$ properly defined, the family $\mathcal{F}$ of finite disjoint unions of these intervals is indeed an algebra. It is the smallest algebra containing $\mathcal{E}$ because the axioms of an algebra (specifically closure under complements and finite unions) force any containing algebra to incorporate these unbounded rays and all their finite disjoint combinations.


4. Common Mistakes

  1. Confusing Bounded and Unbounded Intervals: Forgetting that the complement of a bounded left-closed, right-open interval $(a,b]$ in $\mathbb{R}$ is $(-\infty, a] \cup (b, \infty)$. You cannot stay strictly within bounded intervals if you want the family to be closed under complements.
  2. Confusing Algebras with $\sigma$-Algebras: Assuming you need countable unions to generate the algebra. An algebra only requires closure under finite unions and complements. Finite disjoint unions of intervals are sufficient to form an algebra, whereas countable unions generate the Borel $\sigma$-algebra.
  3. Omitting the Endpoints at Infinity: Leaving $-\infty$ and $\infty$ out of the initial interval family $\mathcal{E}$, which breaks the algebraic property of complementation as the student rightly noticed.

Original question: On the construction of the Lebesgue-Stieltjes measure on Mathematics Stack Exchange, licensed CC BY-SA.