Why does the definition of a sigma-algebra require closure under countable unions rather than under all possible (uncountably infinite) unions?
ACountable unions are computationally tractable while uncountable unions are not, making sigma-algebras more practical
BUncountably infinite unions can generate non-measurable sets — like Vitali sets — that cannot be assigned a consistent probability, leading to contradictions
CThe axiom of choice is only invoked for uncountable unions, which is philosophically unacceptable in standard probability theory
DProbability measures are only defined on finite or countably infinite sample spaces, making uncountable closure unnecessary
The Vitali set construction shows that if you allow probability to be defined on all subsets of [0,1], you cannot simultaneously maintain non-negativity, countable additivity, and translation invariance — the three properties we require. The resolution is to restrict attention to a carefully chosen collection of 'measurable' sets — the sigma-algebra — that excludes the pathological sets. Requiring only countable closure keeps the algebra rich enough to contain all practically needed sets (open intervals, singletons, all Borel sets) while excluding the paradoxical ones.
Question 2 Multiple Choice
Which of the following collections is a valid sigma-algebra on Ω = {a, b, c}?
A{∅, {a}, {b}, {c}} — contains all singletons and the empty set
B{∅, {a}, {b, c}, Ω} — the empty set, one singleton, its complement, and the full set
C{∅, {a, b}, {b, c}, Ω} — contains four sets including two overlapping pairs
D{{a}, {b}, {c}, Ω} — all singletons and the full set but not the empty set
A sigma-algebra must contain Ω, be closed under complementation, and be closed under countable unions. Option B: {∅, {a}, {b,c}, Ω} — check: Ω ✓; complement of {a} is {b,c} ✓; complement of {b,c} is {a} ✓; union of {a} and {b,c} is Ω ✓. This works. Option A fails because complement of {a} = {b,c} is not in the collection. Option C fails because complement of {a,b} = {c} is not included. Option D fails because ∅ is missing (complement of Ω must be in the collection). The valid sigma-algebra here is exactly the algebra generated by the partition {{a}, {b,c}}.
Question 3 True / False
The trivial sigma-algebra {∅, Ω} satisfies all three formal requirements of a sigma-algebra, even though it only permits two events to be assigned probabilities.
TTrue
FFalse
Answer: True
The trivial sigma-algebra is a genuine sigma-algebra: it contains Ω ✓; the complement of ∅ is Ω ✓ and vice versa ✓; any countable union of ∅ and Ω produces only ∅ or Ω ✓. It is mathematically valid but informationally useless — the only events that have probabilities are 'nothing happens' (probability 0) and 'something happens' (probability 1). It represents a state of complete ignorance about the sample space. The concept of a sigma-algebra includes this degenerate case because generality requires it, not because it is useful on its own.
Question 4 True / False
For probability on the real line, the standard practice is to use the power set of most subsets of ℝ as the sigma-algebra, since it is the largest possible collection.
TTrue
FFalse
Answer: False
The power set of ℝ contains non-measurable sets — Vitali sets and others — that cannot be assigned a consistent Lebesgue measure or probability. Using the power set leads to contradictions. The correct choice is the Borel sigma-algebra on ℝ, generated by all open intervals. The Borel sigma-algebra includes all open, closed, and half-open intervals, all singletons, all countable sets, and vastly more — but it excludes the pathological sets. The power set is fine for finite or countably infinite sample spaces; for uncountable spaces like ℝ, a proper sub-collection is required.
Question 5 Short Answer
What problem do non-measurable sets like Vitali sets create for probability theory, and how does restricting attention to a sigma-algebra resolve the problem?
Think about your answer, then reveal below.
Model answer: The Vitali construction shows that if you try to assign a probability (or length) to every subset of [0,1] while requiring three natural properties — non-negativity, countable additivity, and translation invariance — you arrive at a contradiction: the probabilities can neither sum to 1 nor maintain consistency across partitions. You cannot consistently measure everything. The sigma-algebra resolves this by designating in advance which sets are 'measurable' — a collection closed under complementation and countable unions that excludes the pathological sets. Probability is only defined for sets in this collection. Since the problematic sets (Vitali sets, etc.) are not constructible without the axiom of choice and never appear in practice, the restriction is mathematically necessary but practically invisible.
The insight is that the problem is not with probability theory — it is with the assumption that all subsets deserve a probability. By working within a sigma-algebra, measure-theoretic probability achieves both mathematical consistency and practical completeness: the Borel sigma-algebra on ℝ contains every set that arises naturally in analysis and probability, with the pathological exceptions safely excluded.