Measurability Through Threshold Events
A real-valued map is measurable when the inverse image of every Borel subset of \(\mathbb{R}\) is an event. In practice, it is often enough to test a much smaller family of sets: the strict upper rays \((a,\infty)\). This turns a condition involving all Borel sets into a direct test of the events \(\{X>a\}\). Moreover, the test can be restricted to rational \(a\), a countable family.
Throughout, \((\Omega,\mathcal{F},\mathbb{P})\) is a probability space, \(X:\Omega\to\mathbb{R}\) is real-valued at every \(\omega\in\Omega\), and an event means a set in \(\mathcal{F}\). Measurability here means Borel measurability: \(X\) is measurable if \(X^{-1}(B)=\{\omega:X(\omega)\in B\}\in\mathcal{F}\) for every \(B\in\mathcal{B}_{\mathbb{R}}\). No completeness assumption on the probability space is needed, and the criterion concerns pointwise measurability, not merely a version defined almost everywhere.
- \(X\) is measurable as a map into \((\mathbb{R},\mathcal{B}_{\mathbb{R}})\).
- For every \(a\in\mathbb{R}\), the set \(\{\omega\in\Omega:X(\omega)>a\}\) is an event.
Proof. Suppose first that \(X\) is measurable. For each \(a\in\mathbb{R}\), the set \((a,\infty)\) is open, hence belongs to \(\mathcal{B}_{\mathbb{R}}\). Therefore \(\{\omega:X(\omega)>a\}=X^{-1}((a,\infty))\in\mathcal{F}\).
Conversely, suppose \(X^{-1}((a,\infty))\in\mathcal{F}\) for every \(a\in\mathbb{R}\). We verify that these rays generate \(\mathcal{B}_{\mathbb{R}}\). First, for any \(b\in\mathbb{R}\), \([b,\infty)=\bigcap_{n=1}^{\infty}(b-1/n,\infty)\): a real number \(x\) belongs to every ray on the right exactly when \(x\ge b\). Consequently \((-\infty,b)\), the complement of \([b,\infty)\), belongs to the sigma-algebra generated by the rays \((a,\infty)\). Thus every open interval \((r,s)=(r,\infty)\cap(-\infty,s)\) belongs to that sigma-algebra.
Every open subset \(U\subseteq\mathbb{R}\) is the union of the rational-endpoint open intervals \((r,s)\) that it contains. There are only countably many such intervals, and they cover \(U\): for each \(x\in U\), openness gives an interval about \(x\) contained in \(U\), and density of \(\mathbb{Q}\) gives rational \(r<x<s\) inside that interval. Hence every open set, and therefore every Borel set, belongs to the sigma-algebra generated by the rays. Now consider \(\mathcal{C}=\{B\in\mathcal{B}_{\mathbb{R}}:X^{-1}(B)\in\mathcal{F}\}\). Inverse images preserve complements and countable unions, so \(\mathcal{C}\) is a sigma-algebra. It contains every ray \((a,\infty)\) by assumption, and those rays generate \(\mathcal{B}_{\mathbb{R}}\). Thus \(\mathcal{C}=\mathcal{B}_{\mathbb{R}}\), which is precisely measurability of \(X\). \(\square\)
Why Rational Thresholds Are Enough
Proof. If \(X\) is measurable, the theorem gives the event property for every real threshold and hence for every rational one. For the converse, assume all rational-threshold preimages are events. Fix any real \(a\). Then \[ \{\omega:X(\omega)>a\}=\bigcup_{\substack{q\in\mathbb{Q}\\q>a}}\{\omega:X(\omega)>q\}. \] For the inclusion from right to left, \(X(\omega)>q>a\) implies \(X(\omega)>a\). For the reverse inclusion, if \(X(\omega)>a\), density of \(\mathbb{Q}\) supplies a rational \(q\) with \(a<q<X(\omega)\), so \(\omega\) lies in the corresponding set on the right. The union is countable, so it is an event. This holds for every real \(a\); the threshold-event theorem now implies that \(X\) is measurable. \(\square\)
The rational test is useful because it reduces an uncountable family of conditions to a countable one. The reconstruction uses strict upper rays and density: between any two distinct real numbers there is a rational number. It does not say that an arbitrary sparse collection of cutoffs, such as the integers, is enough.
Worked Example: Thresholds on a Finite Probability Space
Let \(\Omega=\{1,2,3,4\}\), let \(\mathcal{F}\) be the power set of \(\Omega\), and assign probabilities \(1/10,2/10,3/10,4/10\) to the four outcomes, respectively. Define \(X(1)=-1\), \(X(2)=1/2\), and \(X(3)=X(4)=2\). Determine every strict upper-threshold event.
For a real threshold \(a\), compare each of the four values with \(a\). If \(a<-1\), all four values are strictly greater than \(a\), so \(\{X>a\}=\Omega\). If \(-1\le a<1/2\), the value \(-1\) does not exceed \(a\), while \(1/2,2,2\) do, so \(\{X>a\}=\{2,3,4\}\). If \(1/2\le a<2\), only the two values equal to \(2\) exceed \(a\), so \(\{X>a\}=\{3,4\}\). If \(a\ge2\), no value strictly exceeds \(a\), so the set is empty. Each set is in \(\mathcal{F}\), and the threshold-event criterion confirms that \(X\) is measurable.
Check: At \(a=1\), the event is \(\{3,4\}\), whose probability is \(3/10+4/10=7/10\). At \(a=2\), the event is empty and has probability \(0\), because equality does not satisfy the strict inequality. These values agree directly with the listed values of \(X\).
Worked Example: The Identity Map on an Interval
Take \(\Omega=[0,1]\), \(\mathcal{F}=\mathcal{B}_{[0,1]}\), and any probability measure \(\mathbb{P}\) on this measurable space. Define \(X(x)=x\). Compute \(\{X>a\}\) for every real \(a\) and use the criterion.
If \(a<0\), every \(x\in[0,1]\) satisfies \(x>a\), so \(\{X>a\}=[0,1]\). If \(0\le a<1\), the condition is \(x>a\) with \(x\in[0,1]\), so \(\{X>a\}=(a,1]\). If \(a\ge1\), no \(x\in[0,1]\) is strictly greater than \(a\), so \(\{X>a\}=\varnothing\). These sets are Borel subsets of \([0,1]\) in each case; therefore \(X\) is measurable.
For a specific rational-threshold reconstruction, take \(a=2/5\). The union of \(\{X>q\}\) over rationals \(q>2/5\) is \((2/5,1]\). Indeed, every \(x>q>2/5\) lies in that interval, and for every \(x>2/5\) there is a rational \(q\) strictly between \(2/5\) and \(x\). The endpoint \(x=2/5\) is excluded from every set in the union, as required.
Check: At \(a=0\), the threshold set is \((0,1]\), not \([0,1]\), because \(X(0)=0\) is not strictly greater than \(0\). At \(a=1\), the set is empty, also because equality does not count.
Worked Example: A Nonmeasurable Map Detected by Rational Cutoffs
Let \(\Omega=\{u,v\}\), \(\mathcal{F}=\{\varnothing,\Omega\}\), and let \(\mathbb{P}(\Omega)=1\). Define \(X(u)=1/5\) and \(X(v)=4/5\). Test the integer thresholds and then test the rational threshold \(q=1/2\).
For an integer \(k\le0\), both values are greater than \(k\), so \(\{X>k\}=\Omega\). For an integer \(k\ge1\), neither value is greater than \(k\), so \(\{X>k\}=\varnothing\). Thus all integer-threshold preimages are events. But at the rational threshold \(1/2\), \(1/5\) is not greater than \(1/2\) and \(4/5\) is greater than \(1/2\); hence \(\{X>1/2\}=\{v\}\notin\mathcal{F}\). The rational-threshold criterion shows that \(X\) is not measurable.
Check: The exact inequalities are \(1/5<1/2<4/5\), so the threshold set contains \(v\) and excludes \(u\). The only events available are \(\varnothing\) and \(\Omega\); therefore \(\{v\}\) is not an event. This verifies the failure without relying on the probability assigned to either singleton.
Strictness, Complements, and a Common Pitfall
Once the strict upper-threshold events are known, some related preimages follow by set operations. In any probability space \((\Omega,\mathcal{F},\mathbb{P})\), if \(X\) is real-valued and measurable, then \(\{X\le a\}=\Omega\setminus\{X>a\}\) and \(\{X<a\}=\bigcup_{n=1}^{\infty}\{X\le a-1/n\}\). For the second identity, \(X<a\) implies \(X\le a-1/n\) for some sufficiently large \(n\); conversely, \(X\le a-1/n<a\) implies \(X<a\). These identities are useful, but they do not permit replacing \(>\) by \(\ge\) at the same threshold.
Common mistake. In the finite example, a mistaken calculation might say \(\{X>2\}=\{3,4\}\) because \(X(3)=X(4)=2\). That calculation has used \(X\ge2\), not \(X>2\). The corrected calculation compares each value to \(2\): \(-1>2\) is false, \(1/2>2\) is false, and \(2>2\) is false for both remaining outcomes. Thus \(\{X>2\}=\varnothing\). In contrast, \(\{X\ge2\}=\{3,4\}\). Keeping the inequality and its boundary case explicit prevents a false threshold event.
Another tempting shortcut is to check only integer cutoffs. The two-point example shows why it fails: both values can lie between the same consecutive integers, so integer thresholds do not separate them, while a rational threshold between them does. The rational criterion works because it tests a dense set of cutoffs and countable unions recover every real cutoff.
Check Your Understanding
Use the threshold-event criterion and attend carefully to endpoints and strict inequalities.
- For a real-valued map \(X\) on \((\Omega,\mathcal{F},\mathbb{P})\), prove that \(\{X\le a\}\in\mathcal{F}\) for every real \(a\) if and only if \(X\) is measurable.
- On \(\Omega=[0,1]\) with its Borel sigma-algebra, compute \(\{X>a\}\) for \(X(x)=1-x\) and all real \(a\), then apply the criterion.
- Explain why the rational-threshold proof uses a countable union, and identify the role of density of \(\mathbb{Q}\).
- In the finite example, compute \(\{X>1/2\}\) and \(\{X\ge1/2\}\), including their probabilities.
- Give two distinct real values lying between the same consecutive integers and explain how a rational threshold can separate them when integer thresholds cannot.