Existence of the Lebesgue measure, proof thereof.

1k Views Asked by At

On page 10 of Stefan Grosskinsky's lecture notes, Theorem 1.7 states the following.

There exists an unique Borel measure $\mu$ on $(\mathbb{R}, \mathcal{B})$ such that $\mu((a, b]) = b - a$ for all $a < b$ in $\mathbb{R}$.

The Borel measure claimed to exist is called the Lebesgue measure. In the proof, a set function is defined on the ring of half-open intervals, $$\mathcal{R} := \{\bigcup_{n=1}^{N} (a_n, b_n]: N \in \mathbb{N}, \forall n, a_n < b_n\}.$$ For a set $\bigcup_{n=1}^{N} (a_n, b_n]$ in $\mathcal{R}$ with disjoint intervals, define $$\mu(\bigcup_{n=1}^{N} (a_n, b_n]) := \sum_{n=1}^{N} (b_n - a_n).$$ Clearly this $\mu$ satisfies $\mu((a, b]) = b - a$. The proof then shows that $\mu$ is countably additive. We also know that $\sigma(\mathcal{R}) = \mathcal{B}$, and so by Caratheodory extension theorem, $\mu$ extends to a measure in $\mathcal{B}$.

My question is about the way we prove that $\mu$ is countably additive. In Grosskinsky's lecture notes, this is done by showing that $\mu$ is continuous from above at $\emptyset$. Since we also clearly have $\mu(A) < \infty$ for all $A \in \mathcal{R}$, this implies that $\mu$ is countably additive. However, I've managed to convince myself for the moment that, by defining $\mu$ differently in the first place, we can prove the countable additivity of $\mu$ directly, without resorting to the equivalence between continuity (under certain conditions) and countable additivity. Here it goes.

Define set function $\lambda$ for all unions (finite or countably infinite) of disjoint intervals, $$\lambda(\bigcup_{n \in \mathbb{N}} (a_n, b_n]) := \sum_{n \in \mathbb{N}} (b_n - a_n).$$ Note that $\lambda(\cdot)$ may be $\infty$. This definition is only slightly different from that of $\mu$ above, in that $\lambda$ is defined on possibly (countably) infinite unions of disjoint intervals, whereas $\mu$ is only defined on finite unions of disjoint intervals. We see that $\lambda$ also clearly satisfies $\lambda((a, b]) = b - a$. Moreover, $\lambda$ restricted to $\mathcal{R}$, which we call $\lambda|_{\mathcal{R}}$ for convenience, is the same as $\mu$. The point is that $\lambda$ is countably additive by definition. Consider disjoint sets $(A_n)_{n \in \mathbb{N}}$ in $\mathcal{R}$. The union $\bigcup_{n \in \mathbb{N}} A_n$ is precisely a possibly countably infinite union of disjoint intervals, since each $A_n$ is a finite union of disjoint intervals and the $A_n$ are mutually disjoint. That is to say, $$\bigcup_{n \in \mathbb{N}} A_n = \bigcup_{i \in \mathbb{N}} (c_i, d_i],$$ for some suitable choice of $c_i$ and $d_i$, where the $(c_i, d_i]$ are mutually disjoint. But by definition of $\lambda$, we have $$\lambda(\bigcup_{n \in \mathbb{N}} A_n) = \lambda(\bigcup_{i \in \mathbb{N}}(c_i, d_i]) = \sum_{i \in \mathbb{N}}(d_i - c_i) = \sum_{n \in \mathbb{N}} \lambda(A_n).$$ This is the statement that $\lambda$ is countably additive, as required.

My hunch is that this proof must be wrong, but I can't see what's wrong at the moment. Can you?