1 Definition and basic notation

An iterated integral is an integral in several variables obtained by carrying out integration one variable at a time. One specifies an order, integrates first with respect to an “inner” variable, and treats the remaining variables as parameters. The outcome is then integrated with respect to the next variable, continuing until all variables have been integrated.

1.1 Single-variable integral inside a multivariable integral

A standard two-variable example uses an order such as “first \(x\), then \(y\).” If \(f(x,y)\) is defined on a set where both variables range, and for each fixed \(y\) the inner bounds for \(x\) are specified, one writes \[ \int_{y=a}^{b}\left(\int_{x=g_1(y)}^{g_2(y)} f(x,y)\,dx\right)dy. \] Here the inner integral produces a function of \(y\), \[ F(y)=\int_{g_1(y)}^{g_2(y)} f(x,y)\,dx, \] and the outer integral computes the total accumulated value of \(F(y)\) across the allowed values of \(y\).

1.2 Different orders of integration

Given the same integrand and region, choosing the order of integration changes the structure of the iterated integral. For instance, integrating “first \(y\), then \(x\)” gives \[ \int_{x=c}^{d}\left(\int_{y=h_1(x)}^{h_2(x)} f(x,y)\,dy\right)dx. \] When the hypotheses of interchange theorems hold, both iterated integrals agree; otherwise, they may differ or one may fail to exist.

1.3 Region vs. bounds descriptions (rectangular and non-rectangular domains)

Iterated integrals can be presented either by describing a domain directly or by giving bounds. For rectangular domains in \(\mathbb{R}^n\), bounds are constants, such as \[ \int_{a_1}^{b_1}\int_{a_2}^{b_2}\cdots\int_{a_n}^{b_n} f(x_1,\dots,x_n)\,dx_n\cdots dx_2\,dx_1. \] For non-rectangular regions, bounds typically depend on previously integrated variables, producing limits like \(g_1(y)\), \(g_2(y)\) in the earlier example. A region can also be described as a set of points satisfying inequalities; an iterated-integral form corresponds to slicing the region according to the chosen order.

1.4 Integrability assumptions and conventions

In practice, one distinguishes between cases where ordinary (“proper”) integrals are finite and cases involving infinite limits (“improper” integrals). In the more general measure-theoretic setting, integrals are defined for measurable functions, with distinctions such as integrability of \(f\) versus mere integrability of \(f\). Conventions vary across texts: some define iterated integrals using limits of truncated integrals; others phrase results in terms of Lebesgue integration and appeal to theorems ensuring equality and finiteness.

2 Geometric and analytical interpretation

Iterated integration can be understood as a procedure for accumulating contributions from “slices” of a multidimensional region. The choice of slice direction affects how the computation is organized.

2.1 Integral as accumulated area/volume (intuition)

When \(f\ge 0\) and the integrand models density or height, the double integral over a region can be viewed as total accumulated “mass.” Performing the inner integral first corresponds to accumulating mass along lines (or hyperplane slices) of constant values of the outer variable. The outer integral then aggregates those slice totals across the remaining direction.

2.2 Slices and conditional integration

For each fixed value of the outer variable, the inner bounds describe a cross-section of the region. The inner integral sums the contributions within that cross-section. The resulting slice total can be interpreted as a function that depends on the outer variable, often analogous to a conditional quantity in probability or a marginal density in statistics.

2.3 Dependence on the choice of variable order

If the integrand has sufficient regularity and the integral is well-behaved (for instance, absolute integrability in Lebesgue terms), different orders typically yield the same result. Without such conditions, the calculated values may depend on the integration order because truncation and convergence behavior differ among the iterated expressions.

3 Fubini’s theorem and equality of iterated integrals

Fubini’s theorem provides the central criterion under which the order of integration can be interchanged without changing the value of the multiple integral.

3.1 Statement of Fubini’s theorem (integrable functions)

In its common measurable-function form, Fubini’s theorem states that if \(f\) is integrable over a product domain (meaning \(\intf<\infty\) in the Lebesgue setting), then the iterated integrals exist and are equal to the multiple integral. Concretely, for a two-variable setting on a product region,

\[ \int\!\!\int f(x,y)\,dx\,dy = \int \left(\int f(x,y)\,dx\right)dy = \int \left(\int f(x,y)\,dy\right)dx, \] with the equality interpreted under the theorem’s hypotheses.

3.2 Measurability requirements

The theorem assumes measurability compatible with the underlying integral definition. In the Lebesgue framework, one typically requires \(f\) to be measurable with respect to the product \(\sigma\)-algebra so that integrals over slices are well-defined for almost every parameter value.

3.3 Interchanging order: when it is valid

Under absolute integrability (or other equivalent sufficient hypotheses), interchanging the order is justified. Analytically, this prevents cancellation effects from causing divergence in one iterated direction while convergence holds in another. As a result, the computed totals coincide regardless of which variable is integrated first.

3.4 Counterexamples outside hypotheses

When \(f\) fails the integrability conditions required by Fubini, iterated integrals can behave inconsistently. A classical phenomenon is that one iterated order may converge while the other diverges, or both may converge to different values due to the integrand’s oscillations and conditional convergence.

4 Tonelli’s theorem for nonnegative functions

Tonelli’s theorem is the nonnegative analogue of Fubini’s theorem. It ensures that slice-by-slice integration is always consistent when the integrand never takes negative values.

4.1 Statement of Tonelli’s theorem (nonnegative case)

If \(f\ge 0\) is measurable on a product domain, then the multiple integral equals the iterated integrals in the sense that \[ \int\!\!\int f = \int\left(\int f\,dx\right)dy = \int\left(\int f\,dy\right)dx, \] where each side is allowed to take the value \(+\infty\).

4.2 Allowing infinite values

Because nonnegative integrals cannot suffer cancellation, divergence manifests as \(+\infty\) rather than undefined subtraction. Tonelli’s theorem therefore applies even when the total integral is infinite, guaranteeing that either order produces the same extended value.

4.3 Monotone convergence perspective

Tonelli’s theorem aligns naturally with the monotone convergence theorem: nonnegative measurable functions can be approximated from below by increasing sequences of simple functions. The equality between iterated integrals and the multiple integral follows by passing to limits along such approximations.

4.4 Practical consequences for computation

For integrands that are nonnegative, one may compute iterated integrals without first proving absolute integrability. This often simplifies calculations in applications such as probability and areas involving densities, where nonnegativity is typical.

5 Product measure viewpoint

Modern formulations of iterated integrals are elegantly described using product measures. This framework clarifies the relationship between iterating over variables and integrating with respect to combined measures.

5.1 Iterated integrals as integrals with respect to product measures

Let \((X,\mathcal{A},\mu)\) and \((Y,\mathcal{B},\nu)\) be measure spaces. On the product space \(X\times Y\), the product measure \(\mu\times \nu\) assigns to suitable sets a combined measure. The multiple integral of \(f(x,y)\) over \(X\times Y\) becomes the integral of \(f\) with respect to \(\mu\times\nu\). Iterated integrals correspond to first integrating over one factor and then over the other.

5.2 Marginal functions and projection

Fixing one variable and integrating over the other yields a marginal function. For example, integrating \(f\) over \(x\) produces \[ m(y)=\int_X f(x,y)\,d\mu(x), \] which can be interpreted as the projection of the “mass” onto the \(y\)-coordinate. Iterating once more integrates this marginal over \(y\).

5.3 Relation to expectation in probability

In probability, expectations over a joint distribution are integrals over a product measure. For a random pair \((X,Y)\) with joint density or distribution, expected values are computed by integrating the function of \((X,Y)\) against the joint law. Conditioning and marginalization correspond closely to computing appropriate iterated integrals.

5.4 σ-finite conditions and their role

Product measure constructions and several versions of Fubini/Tonelli require σ-finiteness of the underlying measures. σ-finiteness ensures the space can be decomposed into countably many finite-measure pieces, allowing integral identities to be established without pathological measure-theoretic obstructions.

6 Improper iterated integrals

Iterated integrals commonly arise with infinite limits or unbounded domains, requiring careful attention to convergence.

6.1 Infinite limits and improper domains

An iterated integral may involve bounds like \(a\) to \(\infty\) for some variable, or an unbounded region described by inequalities. Typically, the integral is defined by truncating one or more limits (e.g., replacing \(\infty\) by \(R\)) and then taking a limit as \(R\to\infty\), with the iterated structure fixing which limit is taken first.

6.2 Absolute vs. conditional convergence

A key distinction is whether the integrand is absolutely integrable. Conditional convergence can mask cancellations between positive and negative contributions. As a result, improper iterated integrals may depend on the order because the truncation regions differ and because cancellation may occur in one order but not another.

6.3 Criteria ensuring convergence

Sufficient conditions for equality and convergence include absolute integrability or domination by an integrable function. In nonnegative contexts, Tonelli’s theorem extends naturally and ensures consistent computation even if the integral is infinite.

6.4 Independence issues with changing order

For improper integrals, changing the order of integration is not automatically valid. Equality may fail even when both iterated integrals exist individually, especially if only conditional convergence holds or if integrability assumptions are violated.

7 Methods of evaluation

Evaluation techniques for iterated integrals depend on how the region and bounds are expressed and whether changes of variables can simplify the integrand and limits.

7.1 Computing rectangular-domain iterated integrals

On rectangles (product sets), one often separates variables or directly integrates step-by-step. With integrands like \(f(x,y)=g(x)h(y)\), the inner integral produces a factor depending only on the outer variable, and the outer integral multiplies the remaining factor, leading to a product of one-dimensional integrals.

7.2 Changing variables (general framework)

When a change of variables transforms the region and the integrand, the iterated structure may become simpler. A typical strategy is to select a substitution tailored to the geometry of the domain (for example, rescaling, shifting, or converting a slanted region into an axis-aligned one) and then apply the corresponding transformation to the bounds and measure factor.

7.3 Jacobians in iterated integration

For differentiable transformations, the Jacobian determinant adjusts for how volumes scale under the substitution. In iterated form, this adjustment appears in the integrand so that integrating with respect to the new variables corresponds to integrating over the original region with the correct scaling.

7.4 Symmetry and exploiting separability

Symmetry can reduce computation by restricting attention to equal contributions from symmetric subregions or by showing an integrand’s oddness leads to cancellation. Separability remains a common simplification: when \(f\) factors into parts depending on different variables, iterated integration often collapses to repeated one-dimensional integrals.

7.5 Using Fubini/Tonelli to simplify domains

Sometimes it is easier to integrate over a larger product domain and use indicator functions to represent the original region. Under the conditions of Fubini/Tonelli, one can legitimately move between “integrate over a region” and “integrate over an ambient box with a cutoff,” turning a complicated domain into a more manageable product structure.

8 Applications in analysis

Iterated integrals serve as a bridge between geometry, probability, harmonic analysis prerequisites, and classical integral forms.

Many geometric quantities, such as volumes of solids with known cross-sections, are computed by integrating slice areas. In simpler settings, an area as a function of one coordinate becomes the integrand for an outer integral. More generally, iterated integrals underpin the derivation of surface-area formulas and related quantities by accumulating contributions across parameter directions.

8.2 Deriving marginal distributions (probability interpretation)

In probability models with joint distributions, marginal distributions are obtained by integrating out one variable. This corresponds exactly to computing an iterated integral where the inner integration produces a marginal density or distribution function and the outer integration handles the remaining variable.

8.3 Applications to Fourier analysis prerequisites (integrability basics)

Fourier analytic techniques rely on understanding when functions are integrable and how norms behave under multidimensional integration. Iterated integrals provide explicit ways to verify integrability conditions such as those needed for applying convergence theorems and for defining transforms on spaces of functions with appropriate decay or regularity.

While Green’s theorem itself is about line integrals and double integrals over planar regions, its use motivates iterated integration as a method for converting geometric boundary information into an area integral. Choosing an iterated-integral description of the region can make the double integral in Green’s theorem computationally accessible.

Beyond Fubini and Tonelli, convergence theorems for integrals justify exchanging limits with integration and support approximation methods.

9.1 Dominated convergence and iterated integrals

Dominated convergence provides a route to interchange limiting processes with integration when functions converge pointwise and are bounded by an integrable dominating function. In iterated-integral contexts, this helps justify taking limits in one variable or simultaneously in multiple variables while preserving integral values.

9.2 Convergence of sequences of integrals (limit interchange)

Many proofs in analysis involve sequences of functions \(f_n\) where one wants \[ \lim_{n\to\infty}\int f_n = \int \lim_{n\to\infty} f_n. \] When the hypotheses of convergence theorems hold, these equalities are valid, ensuring that computations using iterated integrals remain stable under approximation and discretization.

9.3 Approximation by simple functions

Measurable functions can be approximated by simple functions, which take finitely many values on measurable sets. Since simple functions admit straightforward computation of integrals—often compatible with iterated forms—approximation becomes a practical and theoretical foundation for establishing identities such as those in Fubini/Tonelli and for extending results to broader classes of functions.

9.4 Relationship to Lebesgue’s integral theory

Iterated integrals align closely with Lebesgue’s approach because both rely on measurability, integrability, and convergence. Lebesgue’s framework supplies theorems that clarify when iterated integration matches integration over product spaces and when order issues arise due to conditional behavior.

10 Examples and worked computations

Concrete examples illustrate how iterated integrals are computed and how order dependence can appear when conditions are not satisfied.

10.1 Power-function integrands on simple regions

On a rectangle \([0,a]\times[0,b]\), integrands of the form \(f(x,y)=x^m y^n\) separate cleanly: \[ \int_0^b\int_0^a x^m y^n\,dx\,dy = \left(\int_0^a x^m\,dx\right)\left(\int_0^b y^n\,dy\right). \] The result follows from straightforward one-dimensional antiderivatives, with finiteness controlled by whether the exponents permit convergence at the endpoints.

10.2 Integrands with separable variables

For products \(f(x,y)=g(x)h(y)\) on product domains, the inner integral yields \(h(y)\int g(x)\,dx\), and the outer integral multiplies \(\int h(y)\,dy\). This structure generalizes to higher dimensions and often reduces multidimensional problems to a sequence of one-dimensional tasks.

10.3 Integrals over triangular and other non-rectangular regions

For a triangular region in the plane, bounds typically depend on the outer variable. For instance, the region \(0\le x\le 1\) and \(0\le y\le x\) leads to \[ \int_{x=0}^{1}\left(\int_{y=0}^{x} f(x,y)\,dy\right)dx. \] Alternatively, rewriting the same region as \(0\le y\le 1\) and \(y\le x\le 1\) gives a different iterated form. Under appropriate hypotheses, both yield the same value, illustrating how the slice direction can change the appearance of bounds without changing the integral’s value.

10.4 Examples illustrating order dependence and when it disappears

When the integrand is absolutely integrable (e.g., nonnegative or dominated by an integrable function), order dependence disappears: iterated integrals agree with the multiple integral. By contrast, for carefully constructed oscillatory functions that are not absolutely integrable, one can obtain different outcomes from different integration orders or observe that one order produces divergence while another converges. Such examples underscore the role of the assumptions in Fubini’s theorem and the need for caution in improper settings.