从ℝ到ℝ的非可测函数病态程度及相关可测性疑问

Let’s tackle your questions one by one—non-measurable functions are a fascinatingly weird corner of measure theory, so it’s totally reasonable to wonder about their "pathological" nature and real-world relevance.

How Pathological Are Non-Measurable Functions? #

First, let’s ground this: a function ( f: \mathbb{R} \to \mathbb{R} ) is non-measurable if there exists some Borel set ( B \subseteq \mathbb{R} ) such that ( f^{-1}(B) ) is not a Lebesgue-measurable set.

So what makes them so pathological? They break almost every nice property we rely on in analysis:

• Core theorems like the Lebesgue Dominated Convergence Theorem, Fubini’s Theorem, or basic monotone convergence results don’t apply to them. These theorems all hinge on measurability to guarantee consistent, predictable behavior with integrals.
• Their algebraic behavior is erratic: the sum or product of two non-measurable functions might be measurable, or it might not—there’s no general rule to follow. Contrast that with measurable functions, where sums, products, limits, and compositions (with measurable functions) all stay measurable.
• Most importantly, non-measurable functions can’t exist without the Axiom of Choice (AC). That’s a big clue: they’re not "natural" objects that pop up in standard math or applications—you have to actively construct them using AC, which already puts them in the realm of highly abstract, non-constructive math.

Are All Real-World Functions Measurable? #

Short answer: Yes, for all practical purposes.

Every function you’ll encounter in engineering, physics, statistics, or even most pure math applications is measurable:

• Continuous functions, polynomials, trigonometric functions, exponentials, logarithms—all measurable.
• Piecewise-defined functions, limits of measurable functions, integrals of measurable functions—still measurable.
• Even "weird" functions like the Dirichlet function (1 on rationals, 0 on irrationals) are measurable (it’s the indicator function of a measurable set, since rationals have Lebesgue measure 0).

Non-measurable functions are purely theoretical constructs. You’d never run into one in a real-world problem because they require a level of abstraction (and reliance on AC) that doesn’t map to any physical or computational scenario.

Do You Need Cantor-Style Constructions to Get Non-Measurable Functions? #

Nope—though Cantor sets are useful for building other pathological objects (like the Cantor function, which is continuous but not absolutely continuous), the classic non-measurable function uses a different construction: the Vitali set.

Here’s the gist of the Vitali construction:

1. Partition the interval ([0,1]) into equivalence classes where two numbers are equivalent if their difference is rational.
2. Using the Axiom of Choice, pick exactly one number from each equivalence class to form a set ( V ) (the Vitali set).
3. The indicator function ( \chi_V(x) ) (1 if ( x \in V ), 0 otherwise) is non-measurable, because ( V ) itself is a non-measurable set.

That said, there are other non-measurable functions built using different non-measurable sets (like Bernstein sets), but none of these rely on Cantor sets specifically. The common thread is that all non-measurable functions are tied to non-measurable sets, which require the Axiom of Choice to construct.

内容的提问来源于stack exchange，提问作者user56834