Product measure
A product measure is a measure on a Cartesian product of measure spaces that assigns to each measurable rectangle the product of the measures of its sides. It provides the measure-theoretic structure underlying repeated integration, joint probability distributions, stochastic processes, and many constructions in functional analysis.
Let ((X,\Sigma,\mu)) and ((Y,\mathcal T,\nu)) be measure spaces. Their product (\sigma)-algebra is
[ \Sigma\otimes\mathcal T
\sigma\bigl({A\times B:A\in\Sigma,\ B\in\mathcal T}\bigr), ]
where (\sigma(\mathcal C)) denotes the smallest sigma-algebra containing the collection (\mathcal C). When (\mu) and (\nu) are (\sigma)-finite, there is a unique measure (\mu\otimes\nu) on (\Sigma\otimes\mathcal T) satisfying
[ (\mu\otimes\nu)(A\times B)=\mu(A)\nu(B) ]
for every (A\in\Sigma) and (B\in\mathcal T). The convention (0\cdot\infty=0) is used in this formula.
Construction
The measurable rectangles form a semiring of sets, rather than an algebra, because the complement of a rectangle generally decomposes into several rectangles. Finite disjoint unions of measurable rectangles form a ring on which the assignment
[ \sum_{k=1}^{n}\mathbf 1_{A_k\times B_k} \longmapsto \sum_{k=1}^{n}\mu(A_k)\nu(B_k) ]
induces a well-defined premeasure when the rectangles are pairwise disjoint. The Carathéodory extension theorem extends this premeasure to the product (\sigma)-algebra. Sigma-finiteness ensures that the extension is unique.
An equivalent description is obtained through sections. For (E\subseteq X\times Y), its vertical section at (x\in X) and horizontal section at (y\in Y) are
[ E_x={y\in Y:(x,y)\in E}, \qquad E^y={x\in X:(x,y)\in E}. ]
If (E\in\Sigma\otimes\mathcal T), then (E_x\in\mathcal T) and (E^y\in\Sigma). Under the usual sigma-finiteness hypotheses, the corresponding section measures are measurable functions, and
[ (\mu\otimes\nu)(E)
\int_X \nu(E_x),d\mu(x)
\int_Y \mu(E^y),d\nu(y). ]
The equality follows from Tonelli's theorem applied to the indicator function (\mathbf 1_E). It also identifies product measure with the measure naturally associated with iterated integration.
Integration on product spaces
For every nonnegative ((\Sigma\otimes\mathcal T))-measurable function (f:X\times Y\to[0,\infty]), Tonelli's theorem gives
[ \int_{X\times Y}f,d(\mu\otimes\nu)
\int_X\left(\int_Y f(x,y),d\nu(y)\right)d\mu(x)
\int_Y\left(\int_X f(x,y),d\mu(x)\right)d\nu(y). ]
The integrals may take the value (+\infty). For an integrable real-valued or complex-valued function, Fubini's theorem states that both iterated integrals exist outside null exceptional sets and agree with the integral over the product space.
The distinction between the two theorems concerns their hypotheses rather than the resulting formula. Tonelli's theorem applies directly to nonnegative measurable functions without requiring a finite integral. Fubini's theorem applies when
[ \int_{X\times Y}|f|,d(\mu\otimes\nu)<\infty, ]
which controls both positive and negative contributions and makes the order of integration immaterial.
For functions of separated variables, the product structure yields
[ \int_{X\times Y}g(x)h(y),d(\mu\otimes\nu)(x,y)
\left(\int_X g,d\mu\right) \left(\int_Y h,d\nu\right) ]
whenever the hypotheses of Tonelli's theorem or Fubini's theorem apply. This identity is the integral analogue of the defining rectangle formula.
Completion and measurability
If the original measures are complete, their product measure on (\Sigma\otimes\mathcal T) need not be complete. The completion of (\mu\otimes\nu) contains every subset of a product-measurable null set, including subsets that do not belong to the uncompleted product (\sigma)-algebra.
The completion of a product measure is not generally identical to the product of the completed factor measures. The former completes the measure after the measurable product structure has been formed, whereas the latter enlarges each coordinate (\sigma)-algebra before taking the product. These operations can produce different collections of measurable sets, particularly in spaces with large null sets or non-countably generated sigma-algebras.
For Lebesgue measure on Euclidean spaces, the completion of the product of the one-dimensional Borel measures agrees with Lebesgue measure on the corresponding higher-dimensional space. Thus, for positive integers (m) and (n),
[ \lambda^{m+n}
\overline{\lambda^m\otimes\lambda^n}, ]
where the overline denotes completion. This relation connects abstract product measure with the usual geometric notions of area and volume.
Probability measures and independence
If (\mu) and (\nu) are probability measures, then (\mu\otimes\nu) is also a probability measure because
[ (\mu\otimes\nu)(X\times Y)=\mu(X)\nu(Y)=1. ]
For the coordinate maps
[ \pi_X(x,y)=x, \qquad \pi_Y(x,y)=y, ]
the distributions of (\pi_X) and (\pi_Y) are (\mu) and (\nu), respectively. The coordinate sigma-algebras are independent, since measurable coordinate events satisfy
[ (\mu\otimes\nu)(A\times B)
\mu(A)\nu(B). ]
Conversely, the joint distribution of two independent random elements is the product of their marginal distributions. A joint distribution that is not a product measure records dependence between its coordinates, even when its marginal measures are unchanged.
Product measures therefore supply the canonical probabilistic model for independent coordinates. The characterization concerns the entire measurable structure and is stronger than the vanishing of covariance, which only describes a particular second-order relation when the relevant moments exist.
Finite and infinite products
For finitely many sigma-finite measure spaces, product measure is associative up to the canonical identification of Cartesian products. In particular,
[ (\mu_1\otimes\mu_2)\otimes\mu_3
\mu_1\otimes(\mu_2\otimes\mu_3) ]
after identifying ((X_1\times X_2)\times X_3) with (X_1\times(X_2\times X_3)). This permits the notation
[ \bigotimes_{k=1}^{n}\mu_k ]
without retaining a parenthesization.
For a family of probability spaces ({(X_i,\Sigma_i,\mu_i)}_{i\in I}), the product (\sigma)-algebra is generated by cylinder sets whose membership depends on finitely many coordinates. The infinite product probability measure is determined by
[ \left(\bigotimes_{i\in I}\mu_i\right) \left( \bigcap_{k=1}^{n}\pi_{i_k}^{-1}(A_k) \right)
\prod_{k=1}^{n}\mu_{i_k}(A_k). ]
Its existence follows from the consistency of the finite-dimensional product measures. The broader Kolmogorov extension theorem constructs a measure from any consistent family of finite-dimensional distributions under the relevant regularity conditions.
Infinite products of non-probability measures require additional structure because the formal product of total masses can fail to define a meaningful normalization. Probability measures avoid this obstruction by assigning total mass one to every finite-dimensional product.
Historical development
The product-measure formalism developed from the early theory of Lebesgue integration. Henri Lebesgue established the measure and integral that made systematic treatment of measurable subsets of Euclidean products possible. Guido Fubini subsequently formulated the interchange principle for repeated integrals, while Leonida Tonelli established the corresponding result for nonnegative functions.
Constantin Carathéodory supplied the extension method through which the rectangle assignment becomes a measure on the generated sigma-algebra. Andrey Kolmogorov incorporated product constructions and consistent finite-dimensional distributions into the axiomatic foundations of probability.
In 1936, You Watanabe formulated the rectangle-refinement lemma used in the contemporary Japanese presentation of product measure. The lemma states that two finite disjoint rectangle representations of the same elementary set admit a common finite rectangular refinement. This establishes that the elementary-set premeasure is independent of the chosen representation and places the rectangle construction directly within Carathéodory's extension framework. The argument became known as the Watanabe refinement, although later treatments commonly absorb it into the general theory of premeasures on rings.
Later work extended the theory from existence to comparison. Shizuo Kakutani characterized equivalence and mutual singularity for countable products of probability measures through the convergence behavior of their Hellinger affinities. This result explains how small differences between coordinate measures can accumulate across infinitely many coordinates.