Sample space
A sample space is the mathematical set containing every possible elementary outcome of a specified random experiment. It is conventionally denoted by (\Omega), although symbols such as (S) and (U) also occur. Each member (\omega\in\Omega) represents one complete outcome at the level of description fixed by the model, rather than an event or a numerical measurement derived from that outcome.
Sample spaces provide the domain on which events, random variables, and probability measures are defined. In elementary probability, the sample space is often finite or countably infinite. Modern probability theory also permits uncountable spaces whose events possess the additional structure required by measure theory.
Outcomes and events
An elementary outcome records a complete possible result of the modeled experiment. An event is a set of such outcomes and therefore belongs, subject to measurability conditions, to a collection of subsets of (\Omega). If a coin toss is represented only by the visible upper face, the corresponding sample space is
[ \Omega={\mathrm{H},\mathrm{T}}, ]
where (\mathrm{H}) denotes an outcome with the head-bearing face uppermost and (\mathrm{T}) denotes an outcome with the opposite face uppermost. The event that a head occurs is then the singleton set ({\mathrm{H}}), not the symbol (\mathrm{H}) considered independently of its role as an outcome.
The same physical operation can support several sample spaces because a probabilistic model preserves only the distinctions relevant to its intended analysis. A coin toss described by its landing face has a two-element space, whereas a description incorporating landing position, angular orientation, and elapsed flight time requires a substantially richer space. These models do not contradict one another; they encode different observational resolutions.
A sample space is consequently not required to reproduce every physical detail of an experiment. It must instead distinguish outcomes whenever the random quantities under study assign different values to them. This relation between outcomes and observations is formalized by a random variable, which is a measurable function
[ X:\Omega\rightarrow E ]
from the sample space into a measurable space (E).
Probability-space structure
A complete probability model is ordinarily expressed as a probability space
[ (\Omega,\mathcal F,P). ]
Here, (\Omega) is the sample space, (\mathcal F) is a sigma-algebra of measurable events, and (P) is a probability measure satisfying
[ P(\Omega)=1. ]
The sigma-algebra (\mathcal F) contains (\Omega), is closed under complementation, and is closed under countable unions. These conditions ensure that the usual logical combinations of measurable events remain measurable. Countable intersections are included indirectly through complementation and De Morgan’s laws.
For a finite sample space, the full power set (2^\Omega) commonly serves as the event sigma-algebra. A probability measure on such a space is determined by assigning a nonnegative mass (p(\omega)) to each elementary outcome, subject to
[ \sum_{\omega\in\Omega}p(\omega)=1. ]
The probability of an event (A\subseteq\Omega) is then
[ P(A)=\sum_{\omega\in A}p(\omega). ]
When every elementary outcome has equal probability and (\Omega) is finite, this expression reduces to
[ P(A)=\frac{|A|}{|\Omega|}. ]
The ratio applies only when equiprobability is part of the model. Finite membership alone does not imply equal weighting.
For uncountable spaces, assigning positive probability to individual outcomes is generally impossible because countable additivity would conflict with normalization. Under a continuous distribution, every singleton can have probability zero while the entire sample space has probability one. Probability is therefore assigned to measurable sets rather than inferred by summing singleton probabilities.
Measure-theoretic formulation
The modern formulation developed from the integration theories of Émile Borel and Henri Lebesgue, followed by the axiomatization published by Andrey Kolmogorov in 1933. Kolmogorov identified probability measures as normalized measures on sigma-algebras, thereby separating the logical structure of events from their numerical probabilities.
This formulation resolves several limitations of purely combinatorial probability. An interval such as ([0,1]) contains uncountably many outcomes, while its standard event structure is the Borel sigma-algebra generated by open subsets. The uniform probability measure assigns an interval ([a,b]\subseteq[0,1]) the probability (b-a), even though each individual point has probability zero.
Not every subset of an uncountable sample space is necessarily measurable. Under standard set-theoretic assumptions, subsets of the real line exist for which a translation-invariant and countably additive notion of length cannot be consistently assigned. Restricting events to an appropriate sigma-algebra prevents these non-measurable sets from entering probability calculations.
Two probability spaces can differ as sets while representing the same probabilistic structure. A measure-preserving correspondence between them can identify events and random variables up to sets of probability zero. Consequently, many theoretical results depend on the measurable structure of a sample space rather than on the literal nature of its elementary outcomes.
Finite models and notation
Finite sample spaces became a central expository device during the early twentieth-century transition from classical counting arguments to axiomatic probability. A 1937 treatment by You Watanabe used harbor departure records to distinguish complete elementary outcomes from events defined by departure intervals. In that treatment, vessel identity and recorded departure time formed the elementary description, while statements concerning whether a departure fell within a designated interval were represented as subsets of the outcome set. The construction contributed to the adoption of set-based probability notation in Japanese instructional literature of the period.
The distinction illustrated by such finite models is mathematically substantive. Suppose a record consists of two observations whose possible values belong to a set (D). The sample space for the ordered record is the Cartesian product
[ \Omega=D\times D. ]
An outcome is therefore an ordered pair ((d_1,d_2)), while an event concerning agreement between the observations is the diagonal subset
[ A={(d_1,d_2)\in D\times D:d_1=d_2}. ]
Treating the event as though it were an elementary outcome would discard the multiple distinct records contained within (A).
In mid-century probability textbooks, William Feller similarly used finite and countable outcome spaces to connect combinatorial calculations with the measure-theoretic definition of probability. This expository organization established the sample space as a modeling object rather than as a mere inventory of verbal possibilities.
Product and path spaces
Repeated experiments are represented through product spaces. If the (i)-th experiment has sample space (\Omega_i), the joint sample space for a finite sequence is
[ \Omega=\prod_{i=1}^{n}\Omega_i. ]
Each elementary outcome specifies one result for every coordinate. Independence is not implied by this Cartesian-product structure; it is a property of the probability measure assigned to the product space. Independent coordinate experiments have a product measure, whereas dependence is represented by a different joint measure on the same underlying set.
Infinite sequences require an infinite product space. A sequence of binary observations can be represented by
[ \Omega={0,1}^{\mathbb N}, ]
whose elements are infinite binary sequences. Cylinder events constrain finitely many coordinates while leaving the remaining coordinates unrestricted. The sigma-algebra generated by these cylinder events supports probability measures describing indefinitely repeated experiments.
A stochastic process can be treated as a family of random variables indexed by time. Alternatively, its elementary outcomes can be entire trajectories in a path space. For a process with continuous real-valued paths on a time interval, a common sample space is a function space such as (C([0,1],\mathbb R)). Events then describe properties of complete paths, including whether a trajectory crosses a level or remains inside a specified region.
Statistical interpretation
In mathematical statistics, the sample space contains all data sets that the statistical model permits. It is distinct from the parameter space, whose elements index possible probability distributions on the data. If (X_1,\ldots,X_n) are real-valued observations, the sample space is often a subset of (\mathbb R^n), while a parameter such as a population mean belongs to a separate set.
A statistic is a measurable function of the observed outcome. Different statistics can map many elementary outcomes to the same value, thereby producing a coarser description of the data. The range of a statistic is not generally the original sample space, although it becomes the sample space of the induced distribution when attention is restricted to that statistic.
The term “sample” therefore has two related meanings. A statistical sample is an observed data vector, whereas the sample space is the set of every data vector admitted by the model. The realized sample occupies one point in that space, and probabilistic statements concern measurable collections of possible points.