Lebesgue Measure Introduction

Measuring length seems straightforward until you confront pathological sets like the rational numbers on the real line. The classical Riemann integral, foundational to calculus, stumbles precisely because it relies on partitioning the domain—the x-axis. To integrate a wider universe of functions and solidify the foundations of probability and analysis, a more sophisticated notion of "size" is required. Lebesgue measure provides this revolutionary extension, generalizing length, area, and volume to a vast family of subsets of real numbers, enabling the powerful Lebesgue integral.

From Riemann's Limitations to Lebesgue's Insight

The Riemann integral works by chopping the domain into intervals, approximating the function's value on each, and summing areas of rectangles. Its critical flaw is domain-sensitivity. Consider the Dirichlet function: $f(x)=1$ if $x$ is rational, and $0$ if $x$ is irrational, on the interval $[0,1]$. Any interval, no matter how small, contains both rational and irrational numbers. Therefore, upper Riemann sums always equal 1 and lower sums equal 0, preventing Riemann integrability. The function seems conceptually simple, yet Riemann's method fails because it cannot coherently handle the "size" of the set of rationals within each partition interval.

Henri Lebesgue's seminal idea was to partition the range of the function, not its domain. For the Dirichlet function, the range is simply $0,1$. The integration problem then reduces to measuring the set of points where $f(x)=1$ (the rationals) and the set where $f(x)=0$ (the irrationals). This approach demands a consistent way to assign a "measure" or "length" to such bizarre sets. Lebesgue measure, built upon the concept of Lebesgue outer measure, fulfills this demand and extends integration to a class of functions far beyond Riemann's reach.

Lebesgue Outer Measure: The First Step

We begin by defining the size of any subset of the real line. For a set $E\subset \mathbb{R}$, its Lebesgue outer measure $m^{\ast }(E)$ is defined as the infimum (greatest lower bound) of the total lengths of countable coverings of $E$ by open intervals. Formally:

$$m^{\ast }(E)=\inf \left\{\sum _{k=1}^{\infty }\ell (I_k):E\subset \bigcup _{k=1}^{\infty }{I}_k\right\}$$

Here, $I_k$ is any countable collection of open intervals covering $E$, and $\ell (I_k)$ is the standard length of an interval. The outer measure has key properties: it is non-negative ($m^{\ast }(E)\ge 0$), monotonic (if $A\subset B$, then $m^{\ast }(A)\le m^{\ast }(B)$), and countably subadditive (for any countable collection of sets $E_k$, $m^{\ast }({\bigcup }_k{E}_k)\le {\sum }_k{m}^{\ast }(E_k)$).

For an interval $[a,b]$, the outer measure equals its length $b-a$. For the set of rational numbers $\mathbb{Q}\cap [0,1]$, we can exploit countability. List the rationals as $q_1,q_2,q_3,\dots$. Cover $q_1$ with an interval of length $ϵ/2$, $q_2$ with an interval of length $ϵ/4$, and so on. The total length of this covering is at most $ϵ$, which can be made arbitrarily small. Thus, $m^{\ast }(\mathbb{Q}\cap [0,1])=0$.

Measurable Sets and Carathéodory's Criterion

A critical problem arises: outer measure is not countably additive. Ideally, if we have disjoint sets $A_1,A_2,\dots$, we want the measure of their union to equal the sum of their measures. Outer measure's subadditivity only guarantees $\le$, not equality. To restore additivity, we restrict to a "well-behaved" family of sets called the Lebesgue measurable sets.

A set $E\subset \mathbb{R}$ is Lebesgue measurable if, for every test set $A\subset \mathbb{R}$, it satisfies Carathéodory's criterion:

$$m^{\ast }(A)=m^{\ast }(A\cap E)+m^{\ast }(A\cap E^c)$$

where $E^c$ is the complement of $E$. This condition says that $E$ splits any other set $A$ additively with respect to outer measure. The collection $\mathcal{M}$ of all Lebesgue measurable sets forms a sigma-algebra, meaning it is closed under complements and countable unions (and hence countable intersections). For measurable sets, we drop the "*" and simply write $m(E)$ for the Lebesgue measure.

All open sets, closed sets, intervals, and sets constructed from them via countable operations are measurable. The measure $m$ is now countably additive: if $E_1,E_2,\dots$ are disjoint measurable sets, then $m({\bigcup }_{k=1}^{\infty }{E}_k)={\sum }_{k=1}^{\infty }m(E_k)$. This property is the cornerstone of modern integration and probability theory.

Null Sets and Completeness

A set of Lebesgue measure zero is called a null set. Any countable set, like the rationals $\mathbb{Q}$, is a null set. However, uncountable null sets also exist, such as the famous Cantor set. Null sets are the "negligible" sets in measure theory. If a property holds everywhere except on a null set, we say it holds almost everywhere (a.e.). This concept is crucial; for instance, the Dirichlet function is equal to the zero function almost everywhere (since the rationals are a null set), which aligns with its Lebesgue integral being zero.

The Lebesgue measure is complete: every subset of a null set is itself measurable (and has measure zero). This is a desirable property not automatically present in all measure spaces. In probability, a null event is one with probability zero, and completeness ensures any sub-event of an impossible event is also formally assigned probability zero.

Sigma-Algebras: The Formal Foundation

The measurable sets $\mathcal{M}$ constitute a sigma-algebra on $\mathbb{R}$. A sigma-algebra on a set $X$ is a collection of subsets of $X$ that includes $X$ itself, is closed under complements, and is closed under countable unions. This structure is the rigorous domain for a measure, which is a countably additive function from the sigma-algebra to $[0,\infty ]$.

This framework is the bedrock of modern probability theory, where $X$ is the sample space $\Omega$, the sigma-algebra represents the collection of possible events ($\mathcal{F}$), and the measure is the probability measure $P$ satisfying $P(\Omega )=1$. Lebesgue measure on $\mathbb{R}$ is the prototype for continuous probability distributions, like the uniform distribution on $[0,1]$, where the probability of an interval is simply its length.

Common Pitfalls

1. Confusing Outer Measure with Measure: A common error is to assume $m^{\ast }(E)$ is the measure of $E$. Outer measure is defined for all subsets, but it is not additive. The true Lebesgue measure $m(E)$ is only defined for sets in the sigma-algebra $\mathcal{M}$ (the measurable sets). Always check measurability before applying additive properties.

1. Assuming All Sets Are Measurable: It is tempting to believe every subset of $\mathbb{R}$ has a well-defined length. This is false. Using the Axiom of Choice, one can construct non-measurable sets, such as the Vitali set, which do not satisfy Carathéodory's criterion and cannot be assigned a consistent Lebesgue measure. This is not a flaw but a deep result showing the necessity of restricting to a sigma-algebra.

1. Misinterpreting "Almost Everywhere": The statement "$f(x)=g(x)$ almost everywhere" does not mean they are equal at all but finitely many points. It means the set where they differ has measure zero, which could be uncountable (like the Cantor set). Convergence almost everywhere is a fundamentally different, and generally weaker, concept than pointwise convergence.

1. Overlooking Completeness: When working with other measures or in abstract settings, the property of completeness (all subsets of null sets being measurable) is not guaranteed. The Borel sigma-algebra, generated by open sets, is not complete. The Lebesgue sigma-algebra $\mathcal{M}$ is actually the completion of the Borel sigma-algebra. For many proofs, this distinction matters.

Summary

• Lebesgue measure generalizes length to a vast family of measurable sets, defined via Lebesgue outer measure and Carathéodory's criterion for measurability.
• It was developed to overcome the limitations of the Riemann integral, enabling the integration of functions by partitioning the range, which requires measuring the "size" of level sets.
• The collection of measurable sets forms a sigma-algebra, a structure closed under countable set operations, on which the measure is countably additive.
• Null sets (sets of measure zero) are negligible; properties holding almost everywhere are sufficient for integration theory.
• This measure-theoretic framework, with the sigma-algebra as the domain of events, is the rigorous foundation for modern probability theory and advanced analysis.