Set-theoretic limit

In set theory and measure theory, the set-theoretic limit of a sequence of sets \{A_n\}_{n=1}^\infty is defined as the set that arises when the limit inferior and limit superior of the sequence coincide, providing a way to extend the classical notion of sequence limits to non-numeric objects like sets.^[1]^[2] The limit inferior, denoted \liminf_{n \to \infty} A_n, consists of all elements that belong to all but finitely many of the sets A_n, formally expressed as \bigcup_{n=1}^\infty \bigcap_{k=n}^\infty A_k, while the limit superior, \limsup_{n \to \infty} A_n, includes elements that belong to infinitely many A_n, given by \bigcap_{n=1}^\infty \bigcup_{k=n}^\infty A_k.^[1]^[2] This framework always satisfies \liminf_{n \to \infty} A_n \subseteq \limsup_{n \to \infty} A_n, and the limit \lim_{n \to \infty} A_n exists precisely when equality holds.^[1]^[2] For monotonic sequences of sets, the set-theoretic limit simplifies significantly and always exists. In the non-decreasing case, where A_1 \subseteq A_2 \subseteq \cdots, the limit is the union \bigcup_{n=1}^\infty A_n, capturing all elements that eventually enter the sequence.^[1]^[2] Conversely, for non-increasing sequences A_1 \supseteq A_2 \supseteq \cdots, it is the intersection \bigcap_{n=1}^\infty A_n, representing elements that persist throughout.^[1]^[2] These properties leverage De Morgan's laws, relating complements such that (\liminf_{n \to \infty} A_n)^c = \limsup_{n \to \infty} A_n^c, which aids in computations within Boolean algebras of sets.^[1] The concept finds essential applications in probability and measure theory, where sequences of events or measurable sets are analyzed for convergence. For instance, the limit superior corresponds to the set of outcomes occurring infinitely often, expressible via indicator functions as \{\omega : \sum_{n=1}^\infty I_{A_n}(\omega) = \infty\}, while the limit inferior identifies outcomes occurring eventually always, \{\omega : \sum_{n=1}^\infty I_{A_n^c}(\omega) < \infty\}.^[1] This underpins theorems like the monotone convergence theorem for measures, enabling the passage to limits in integrals and probabilities, and is foundational in studying almost sure convergence of random variables.^[1] An example is the sequence A_k = [0, k/(k+1)) on the real line, whose set-theoretic limit is [0, 1), illustrating convergence to a half-open interval.^[1]

Formal Definitions

Union-Intersection Characterization

The union-intersection characterization provides a foundational approach to defining set-theoretic limits for a sequence of subsets \{A_n\}_{n=1}^\infty of a fixed ambient set X, relying solely on the basic set operations of union and intersection.^[3] The limit inferior of the sequence is defined as

\liminf_{n\to\infty} A_n = \bigcup_{n=1}^\infty \bigcap_{j\geq n} A_j.

This set consists precisely of those elements x\in X that belong to all but finitely many of the sets A_n, often denoted intuitively as the elements that occur "all but finitely often" (a.b.f.o.).^[3] Dually, the limit superior is given by

\limsup_{n\to\infty} A_n = \bigcap_{n=1}^\infty \bigcup_{j\geq n} A_j.

This captures the elements x\in X that belong to infinitely many of the sets A_n, or "infinitely often" (i.o.).^[3] The set-theoretic limit \lim_{n\to\infty} A_n exists if and only if \liminf_{n\to\infty} A_n = \limsup_{n\to\infty} A_n, in which case the limit equals this common set.^[3]

Indicator Function Approach

The indicator function of a set A \subseteq X, denoted \mathbb{1}_A: X \to \{0,1\}, is defined pointwise by \mathbb{1}_A(x) = 1 if x \in A and \mathbb{1}_A(x) = 0 otherwise.^[3] For a sequence of subsets \{A_n\}_{n=1}^\infty of a space X, the indicator functions \{\mathbb{1}_{A_n}\}_{n=1}^\infty form a sequence of functions taking values in \{0,1\}, allowing the set-theoretic limit inferior and limit superior to be characterized via pointwise limits of these indicators.^[4] Specifically, the limit inferior is given by

\liminf_{n\to\infty} A_n = \{ x \in X : \liminf_{n\to\infty} \mathbb{1}_{A_n}(x) = 1 \},

where \liminf_{n\to\infty} \mathbb{1}_{A_n}(x) is the standard limit inferior of the real sequence \{\mathbb{1}_{A_n}(x)\}_{n=1}^\infty.^[4] Similarly, the limit superior is

\limsup_{n\to\infty} A_n = \{ x \in X : \limsup_{n\to\infty} \mathbb{1}_{A_n}(x) = 1 \},

with \limsup_{n\to\infty} \mathbb{1}_{A_n}(x) denoting the limit superior of the same sequence. This formulation equates the indicator of the set-theoretic liminf (or limsup) to the pointwise liminf (or limsup) of the indicators, i.e., \mathbb{1}_{\liminf_{n\to\infty} A_n}(x) = \liminf_{n\to\infty} \mathbb{1}_{A_n}(x) and \mathbb{1}_{\limsup_{n\to\infty} A_n}(x) = \limsup_{n\to\infty} \mathbb{1}_{A_n}(x).^[4] The explicit form of the liminf indicator arises from the definition of the limit inferior for sequences:

\mathbb{1}_{\liminf_{n\to\infty} A_n}(x) = \sup_{n \geq 1} \inf_{j \geq n} \mathbb{1}_{A_j}(x).

This holds because x belongs to the liminf if and only if there exists some n such that \mathbb{1}_{A_j}(x) = 1 for all j \geq n, making the inner infimum 1 and the outer supremum 1.^[3] An analogous expression applies to the limsup indicator using infima of suprema. This indicator-based characterization is equivalent to the combinatorial union-intersection method, as the pointwise conditions on the indicators directly mirror the membership criteria for the nested unions and intersections defining the set limits.^[4] The equivalence relies on De Morgan's laws, which relate the limsup of sets to the liminf of their complements, and correspondingly link the limsup of indicators to the liminf of the complemented indicators (noting that \mathbb{1}_{A_n^c}(x) = 1 - \mathbb{1}_{A_n}(x)), ensuring consistency between the two approaches.^[3] The set-theoretic limit \lim_{n\to\infty} A_n exists if and only if \liminf_{n\to\infty} A_n = \limsup_{n\to\infty} A_n, which occurs precisely when \lim_{n\to\infty} \mathbb{1}_{A_n}(x) exists for every x \in X. In this case,

\lim_{n\to\infty} A_n = \{ x \in X : \lim_{n\to\infty} \mathbb{1}_{A_n}(x) = 1 \},

capturing the points where the indicators converge pointwise to 1.^[4]

Monotonicity Cases

In the case of a nonincreasing sequence of sets \{A_n\}_{n=1}^\infty, where A_{n+1} \subseteq A_n for all n, the set-theoretic limit exists and equals the intersection \bigcap_{n=1}^\infty A_n, as both the liminf and limsup coincide with this intersection.^[5] For a nondecreasing sequence \{A_n\}_{n=1}^\infty, satisfying A_n \subseteq A_{n+1} for all n, the limit similarly exists and is the union \bigcup_{n=1}^\infty A_n, with liminf and limsup both equal to this union.^[5] This simplification arises from the general definitions of liminf and limsup as tools for set limits. To verify for the nonincreasing case, note that the tail intersection \bigcap_{j \geq n} A_j = \bigcap_{j \geq 1} A_j holds for every n, so the liminf—defined as the union over n of these tails—reduces to the overall intersection; an analogous verification applies to the nondecreasing case, where tail unions stabilize to the full union.^[5] This continuity property for monotone sequences has motivated key applications in analysis, such as the Cantor set construction, where a nonincreasing sequence of nested closed intervals produces the set as their intersection.^[6] In general, monotonicity guarantees the existence of the set-theoretic limit for the sequence.^[5]

Key Properties

Inclusion Relations

In set theory, for a sequence of sets \{A_n\}_{n=1}^\infty in a universe \Omega, the set-theoretic limit inferior and limit superior satisfy the fundamental inclusion \liminf_{n\to\infty} A_n \subseteq \limsup_{n\to\infty} A_n. This relation holds universally, as every element belonging to all but finitely many A_n (characterizing the liminf) necessarily belongs to infinitely many A_n (characterizing the limsup). Equality in this inclusion occurs if and only if the set-theoretic limit \lim_{n\to\infty} A_n exists, defined as their common value.^[7]^[8] The liminf admits a useful characterization via tail sets B_n = \bigcap_{j \geq n} A_j for each n \in \mathbb{N}, which form a nondecreasing sequence (B_n \subseteq B_{n+1}). In this framework, \liminf_{n\to\infty} A_n = \bigcup_{n=1}^\infty B_n, and since the B_n are nested, this equals the limit \lim_{n\to\infty} B_n. Dually, the limsup can be expressed using the tail unions C_n = \bigcup_{j \geq n} A_j, yielding \limsup_{n\to\infty} A_n = \bigcap_{n=1}^\infty C_n, with the C_n nonincreasing. These tail representations underscore the inclusion, as the union of nested intersections is contained in the intersection of nested unions.^[8]^[7] A duality relation governs the interaction with complements: \liminf_{n\to\infty} A_n = \left( \limsup_{n\to\infty} A_n^c \right)^c, where A_n^c denotes the complement of A_n in \Omega. The dual identity is \limsup_{n\to\infty} A_n = \left( \liminf_{n\to\infty} A_n^c \right)^c. These follow from De Morgan's laws applied to the tail expressions, preserving the core inclusion under complementation.^[3]^[7] From a frequency perspective, an element \omega \in \Omega lies in \liminf_{n\to\infty} A_n if and only if \omega belongs to cofinitely many sets in the sequence, meaning \omega \in A_n for all sufficiently large n. In contrast, \omega \in \limsup_{n\to\infty} A_n if and only if \omega belongs to infinitely many A_n. This distinction highlights the stricter membership condition for the liminf within the broader limsup. In monotone sequences, such as nested increasing or decreasing sets, the inclusion often collapses to equality.^[3]^[7]

Limit Existence Criteria

The set-theoretic limit of a sequence of sets \{A_n\}_{n=1}^\infty in a universe X exists if and only if \liminf_{n\to\infty} A_n = \limsup_{n\to\infty} A_n, in which case the limit L is this common set, denoted \lim_{n\to\infty} A_n = L. This condition holds precisely when, for every x \in X, the indicator sequence \mathbb{1}_{A_n}(x) converges pointwise to either 0 or 1, as the limit set then consists of those points where the indicator converges to 1.^[3] If the limit exists and equals L, then necessarily \liminf_{n\to\infty} A_n = L = \limsup_{n\to\infty} A_n, reflecting the sequential convergence of the sets to L. The baseline inclusion \liminf_{n\to\infty} A_n \subseteq \limsup_{n\to\infty} A_n always holds, so existence requires equality in this relation. Conversely, the limit fails to exist if there exists some x \in X that belongs to infinitely many A_n but not to all but finitely many A_n, causing oscillation in the indicator sequence at that point. The limsup can also be characterized via the tails of the sequence: define C_n = \bigcup_{j \geq n} A_j for each n, yielding a nonincreasing sequence \{C_n\}_{n=1}^\infty. Then \limsup_{n\to\infty} A_n = \bigcap_{n=1}^\infty C_n = \lim_{n\to\infty} C_n, since the limit of a nonincreasing sequence of sets exists and equals the intersection. Non-convergence is common in oscillating sequences where liminf and limsup differ.

Measure-Theoretic Aspects

In measure spaces equipped with a σ-algebra \mathcal{F}, the set-theoretic limits preserve measurability when applied to sequences of measurable sets. Specifically, if \{A_n\}_{n=1}^\infty is a sequence of sets with each A_n \in \mathcal{F}, then both \liminf_{n \to \infty} A_n = \bigcup_{n=1}^\infty \bigcap_{k \geq n} A_k and \limsup_{n \to \infty} A_n = \bigcap_{n=1}^\infty \bigcup_{k \geq n} A_k belong to \mathcal{F}. This follows directly from the closure properties of σ-algebras under countable unions and intersections, ensuring that the limit sets are constructed from operations that maintain membership in \mathcal{F}.^[9]^[10] The measurability of these limits can also be understood through their indicator functions. The indicator function of the liminf satisfies \mathbb{1}_{\liminf_{n \to \infty} A_n}(x) = \sup_n \inf_{j \geq n} \mathbb{1}_{A_j}(x), and since each \mathbb{1}_{A_j} is a measurable function (taking values in {0,1}) when A_j \in \mathcal{F}, the pointwise supremum and infimum of such functions yield a measurable function. Thus, the set where this indicator equals 1—namely, \liminf_{n \to \infty} A_n—is measurable as the preimage of {1} under a measurable function. A similar argument applies to the limsup via its indicator \mathbb{1}_{\limsup_{n \to \infty} A_n}(x) = \inf_n \sup_{j \geq n} \mathbb{1}_{A_j}(x).^[9]^[10]^[11] For monotone sequences of sets, the preservation of measurability is particularly straightforward, assuming familiarity with σ-algebras as the foundational structure for measurability. If \{A_n\} is increasing (i.e., A_1 \subseteq A_2 \subseteq \cdots), then \lim_{n \to \infty} A_n = \bigcup_{n=1}^\infty A_n \in \mathcal{F} as a countable union of measurable sets. Similarly, for a decreasing sequence (A_1 \supseteq A_2 \supseteq \cdots), the limit is \bigcap_{n=1}^\infty A_n \in \mathcal{F} as a countable intersection. These cases highlight the continuity of set limits within the measurable structure.^[9]^[10] Overall, these properties ensure that set-theoretic limits behave well in measure-theoretic contexts, such as integration over measurable sets, by keeping the resulting sets within the σ-algebra without requiring additional measurability assumptions. This closure under limits supports rigorous analysis in spaces where measurability is essential.^[9]^[11]

Illustrative Examples

Simple Interval Sequences

A simple example illustrating the existence of a set-theoretic limit involves the sequence of half-open intervals A_n = (-1/n, 1 - 1/n] for n \geq 1, which are subsets of the real line \mathbb{R}. This sequence is neither strictly increasing nor decreasing, yet it converges in the set-theoretic sense. The limit inferior is defined as \liminf_{n \to \infty} A_n = \bigcup_{m=1}^\infty \bigcap_{n=m}^\infty A_n, the set of points that belong to all but finitely many A_n, while the limit superior is \limsup_{n \to \infty} A_n = \bigcap_{m=1}^\infty \bigcup_{n=m}^\infty A_n, the set of points that belong to infinitely many A_n. To compute \liminf_{n \to \infty} A_n, first determine the tail intersections \bigcap_{k \geq m} A_k for each fixed m. Each A_k = (-1/k, 1 - 1/k] has left endpoint -1/k increasing to 0 and right endpoint $1 - 1/k increasing to 1. The intersection is thus [0, 1 - 1/m], since points x \geq 0 satisfy x > -1/k for all finite k and x \leq 1 - 1/k for all k \geq m, while points x < 0 are excluded from sufficiently large k where -1/k > x. Union over m yields \liminf_{n \to \infty} A_n = [0, 1). For \limsup_{n \to \infty} A_n, consider the tail unions \bigcup_{k \geq m} A_k = (-1/m, 1), as the leftmost endpoint is -1/m and the right extends to but excludes 1. Intersecting over m gives \bigcap_{m=1}^\infty (-1/m, 1) = [0, 1), since points x < 0 are excluded from all large enough unions and x \geq 1 from all sets, while points in [0, 1) appear in infinitely many A_n. Since \liminf_{n \to \infty} A_n = \limsup_{n \to \infty} A_n = [0, 1), the set-theoretic limit exists and equals [0, 1). This example demonstrates convergence for a non-monotone sequence resembling nested intervals in its tail behavior.

Oscillating Intervals

The sequence of sets A_n = \left( \frac{(-1)^n}{n}, 1 - \frac{(-1)^n}{n} \right] for n \geq 1 provides an illustrative example of oscillation in set-theoretic limits, where the intervals alternate between slight expansions beyond the unit interval [0,1] for odd n (when (-1)^n = -1, yielding left endpoint -1/n and right endpoint $1 + 1/n) and contractions within (0,1) for even n (when (-1)^n = 1, yielding left endpoint $1/n and right endpoint $1 - 1/n). To determine whether the limit exists, compute the limit inferior and limit superior using their standard definitions: the liminf consists of points belonging to all but finitely many A_n, formally \liminf_{n \to \infty} A_n = \bigcup_{N=1}^\infty \bigcap_{n \geq N} A_n, while the limsup consists of points belonging to infinitely many A_n, formally \limsup_{n \to \infty} A_n = \bigcap_{N=1}^\infty \bigcup_{n \geq N} A_n. Consider first the liminf. For any interior point x \in (0,1), choose N > \max\left( \frac{1}{x}, \frac{1}{1-x} \right). Then, for all n \geq N, the distance from x to either boundary of [0,1] exceeds $1/n, so x lies within every A_n for n \geq N, regardless of parity. Thus, all such x belong to the liminf. However, the boundary point x = 0 fails to belong to any even-n interval, as the left endpoint $1/n > 0 excludes it, and there are infinitely many even indices; similarly, x = 1 is excluded from all even-n intervals, where the right endpoint $1 - 1/n < 1. For points x < 0 or x > 1, membership occurs only in finitely many odd-n intervals before the endpoints \pm 1/n shrink past x. Therefore, \liminf_{n \to \infty} A_n = (0,1). Now examine the limsup. Every point x \in (0,1) belongs to infinitely many A_n (in fact, all sufficiently large ones, as above). The boundary points x = 0 and x = 1 each belong to all odd-n intervals, hence infinitely often. Points outside [0,1] belong to only finitely many A_n, as the expanding endpoints for odd n approach 0 and 1 without encompassing fixed exterior points beyond some finite stage. Thus, \limsup_{n \to \infty} A_n = [0,1]. Since \liminf_{n \to \infty} A_n = (0,1) \subsetneq [0,1] = \limsup_{n \to \infty} A_n, the two differ, so the set-theoretic limit does not exist. This example underscores the sensitivity of set limits to boundary behavior in oscillating sequences, where inclusion relations become strict due to periodic exclusions at the endpoints.

Rational Number Approximations

Consider the sequence of sets A_n = \left\{ \frac{k}{n} \mid k = 0, 1, \dots, n \right\} \subseteq [0,1] for each positive integer n. These sets form increasingly fine grids of rational points spaced by $1/n, approximating the interval [0,1] with dyadic-like rationals whose denominators grow linearly.^[12] The set-theoretic liminf of this sequence is defined as \liminf_{n \to \infty} A_n = \bigcup_{N=1}^\infty \bigcap_{n \geq N} A_n, consisting of points that belong to all but finitely many of the sets A_n. For any fixed N, the tail intersection \bigcap_{n \geq N} A_n contains points x \in [0,1] such that n x is an integer for every n \geq N. Irrational points x satisfy n x \in \mathbb{Z} for no n \geq 1, so they are excluded from all A_n. For rational x = p/q in lowest terms with $0 < p < q, there exist infinitely many n \geq N (e.g., primes larger than q) where q does not divide n, making n x \notin \mathbb{Z}. Thus, only the endpoints x = 0 and x = 1 (where n x \in \mathbb{Z} for all n) lie in every tail intersection, so \bigcap_{n \geq N} A_n = \{0, 1\} for all N, and \liminf_{n \to \infty} A_n = \{0, 1\}.^[12] The limsup is \limsup_{n \to \infty} A_n = \bigcap_{N=1}^\infty \bigcup_{n \geq N} A_n, comprising points that belong to infinitely many A_n. A point x enters infinitely many A_n if there are infinitely many n with n x \in \mathbb{Z}. Irrationals again appear in none. For any rational x = p/q \in [0,1] in lowest terms, choose multiples n = q t for t = 1, 2, \dots; then x = (p t)/(q t), so x \in A_n with n x = p t \in \mathbb{Z}, yielding infinitely many such n. Hence, every rational in [0,1] belongs to infinitely many A_n, and \limsup_{n \to \infty} A_n = \mathbb{Q} \cap [0,1].^[12] Since \liminf_{n \to \infty} A_n = \{0, 1\} \neq \mathbb{Q} \cap [0,1] = \limsup_{n \to \infty} A_n, the sequence has no set-theoretic limit. The finite liminf captures only persistent endpoints, while the dense limsup reflects how rationals recur infinitely often due to the refining grid, demonstrating the role of density in determining "infinitely often" membership.^[12]

Probabilistic Applications

Continuity for Monotone Events

In probability theory, the continuity of measures for monotone sequences of events is a fundamental property that ensures the interchange of limits and probabilities under monotonicity. This requires a probability space (\Omega, \mathcal{F}, \mathbb{P}), where \Omega is the sample space, \mathcal{F} is a \sigma-algebra of events, and \mathbb{P} is a probability measure satisfying \mathbb{P}(\Omega) = 1 and countable additivity for disjoint events.^[13] The monotone continuity theorem states that for a nondecreasing sequence of events \{A_n\}_{n=1}^\infty with A_1 \subseteq A_2 \subseteq \cdots, the probability of the set-theoretic limit equals the limit of the probabilities: \mathbb{P}\left(\bigcup_{n=1}^\infty A_n\right) = \lim_{n \to \infty} \mathbb{P}(A_n). Similarly, for a nonincreasing sequence \{A_n\}_{n=1}^\infty with A_1 \supseteq A_2 \supseteq \cdots and \mathbb{P}(A_1) < \infty, \mathbb{P}\left(\bigcap_{n=1}^\infty A_n\right) = \lim_{n \to \infty} \mathbb{P}(A_n).^[13] A proof sketch for the nondecreasing case relies on countable additivity. Define disjoint events B_1 = A_1 and B_n = A_n \setminus A_{n-1} for n \geq 2. Then \bigcup_{n=1}^\infty A_n = \bigcup_{n=1}^\infty B_n, so \mathbb{P}\left(\bigcup_{n=1}^\infty A_n\right) = \sum_{n=1}^\infty \mathbb{P}(B_n). Moreover, A_m = \bigcup_{n=1}^m B_n, yielding \mathbb{P}(A_m) = \sum_{n=1}^m \mathbb{P}(B_n), and taking m \to \infty shows the limit equality. The nonincreasing case follows by applying the nondecreasing result to the complements \{A_n^c\} and using \mathbb{P}(A_n^c) = 1 - \mathbb{P}(A_n).^[13]^[14] This theorem holds in the general case because the set-theoretic limit exists for any monotone sequence of sets in a \sigma-algebra, providing the nested union or intersection structure needed for the probability limit.^[13] The property is essential for computing probabilities of limiting events, such as in the continuity of probability measures, where it underpins derivations like those for cumulative distribution functions approaching 1 as the argument tends to infinity.^[13]^[14]

Borel–Cantelli Lemmas

The Borel–Cantelli lemmas provide key insights into the asymptotic behavior of sequences of events in probability spaces, particularly regarding the set-theoretic limit superior, which corresponds to the event that the events occur infinitely often. These lemmas relate the summability of event probabilities to the measure of their limsup.^[15] The first Borel–Cantelli lemma, originally due to Émile Borel, asserts that for any sequence of events \{A_n\}_{n=1}^\infty in a probability space, if \sum_{n=1}^\infty \mathbb{P}(A_n) < \infty, then \mathbb{P}\left(\limsup_{n\to\infty} A_n\right) = 0.^[16]^[15] This implies that A_n occurs only finitely many times almost surely.^[15] The proof relies on the union bound:

\mathbb{P}\left(\limsup_{n\to\infty} A_n\right) \leq \mathbb{P}\left(\bigcup_{j \geq k} A_j\right) \leq \sum_{j \geq k} \mathbb{P}(A_j)

for any k \in \mathbb{N}, and the right-hand side tends to 0 as k \to \infty by the convergence of the series.^[15] Notably, this result holds without requiring independence among the events and mirrors convergence tests for infinite series.^[15] The second Borel–Cantelli lemma, due to Francesco Paolo Cantelli, addresses the divergent case under independence: if \{A_n\}_{n=1}^\infty consists of independent events and \sum_{n=1}^\infty \mathbb{P}(A_n) = \infty, then \mathbb{P}\left(\limsup_{n\to\infty} A_n\right) = 1.^[15]^[17] Thus, infinitely many A_n occur almost surely.^[15] The proof exploits independence to compute

\mathbb{P}\left(\bigcap_{j \geq k} A_j^c\right) = \prod_{j \geq k} \left(1 - \mathbb{P}(A_j)\right),

which converges to 0 as k \to \infty whenever the sum of \mathbb{P}(A_j) diverges, implying the complement of the limsup has probability 0.^[15] This lemma complements the first by handling the borderline case of non-summable probabilities.^[15]

Almost Sure Convergence

A sequence of random variables \{Y_n\} converges almost surely to a random variable Y if, for every \epsilon > 0,

\mathbb{P}\left( \limsup_{n \to \infty} |Y_n - Y| > \epsilon \right) = 0,

or equivalently,

\mathbb{P}\left( \{ \omega : \lim_{n \to \infty} Y_n(\omega) = Y(\omega) \} \right) = 1.[](https://sites.math.duke.edu/~rtd/PTE/PTE5_011119.pdf) This definition leverages the set-theoretic limit superior to capture [pointwise convergence](/page/Pointwise_convergence) on a set of probability one, emphasizing the exceptional set where convergence fails has measure zero.[](https://sites.math.duke.edu/~rtd/PTE/PTE5_011119.pdf) To connect this to events, define the deviation sets $ A_n = \{ |Y_n - Y| > \epsilon \} $ for fixed $ \epsilon > 0 $. Then, almost sure convergence holds if and only if \[ \mathbb{P}\left( \limsup_{n \to \infty} A_n \right) = 0,

where \limsup_{n \to \infty} A_n = \bigcap_{n=1}^\infty \bigcup_{m \geq n} A_m represents the set of outcomes belonging to infinitely many A_n.^[7] The full convergence set can be expressed using liminf concepts as \bigcap_{k=1}^\infty \bigcup_{n=1}^\infty \bigcap_{m \geq n} A_{m,k}^c, where A_{m,k} = \{ |Y_m - Y| > 1/k \}, but this is often simplified through the limsup of deviations to focus on the tail behavior of the sequence.^[7] In practice, the probability of the limsup event is frequently shown to be zero using the Borel–Cantelli lemmas, which bound the measure of outcomes in infinitely many deviation events based on the summability of their probabilities.^[7] This approach is particularly effective for independent sequences, where the first lemma applies directly if \sum \mathbb{P}(A_n) < \infty. Almost sure convergence is stronger than convergence in probability, as it requires pointwise convergence almost everywhere rather than merely controlling the probability of large deviations uniformly.^[7]