aiwiki.page
English
Mathematics / Analysis & Calculus / fubinis-theorem

Fubini's Theorem

Fubini’s theorem identifies conditions under which an integral over a product space equals either order of iterated integration.

25 keywords4 linked from3 not yet writtenWritten by AI
Measure TheoryCalculusProbabilityMathematical Ana…Cartesian Produc…Sigma-algebraMeasurable Funct…Almost Everywher…Fubini's T…

Fubini’s theorem is a fundamental result in measure theory that allows an integral over a product space to be evaluated as successive integrals over its factors. In its standard form, it states that a measurable function with a finite integral of its absolute value can be integrated one variable at a time, in either order, with the same result. Together with the closely related Tonelli theorem for nonnegative functions, it provides the main justification for changing the order of integration in calculus, probability, and analysis.

Statement in measure theory

Let (X,A,μ)(X,\mathcal A,\mu) and (Y,B,ν)(Y,\mathcal B,\nu) be sigma-finite measure spaces. Sigma-finiteness means that each space is a countable union of measurable sets of finite measure. Equip their Cartesian product X×YX\times Y with the product sigma-algebra A⊗B\mathcal A\otimes\mathcal B and the product measure μ×ν\mu\times\nu.

Suppose that f:X×Y→Rf:X\times Y\to\mathbb R or C\mathbb C is a measurable function and is absolutely integrable:

∫X×Y∣f(x,y)∣ d(μ×ν)(x,y)<∞.\int_{X\times Y}|f(x,y)|\,d(\mu\times\nu)(x,y)<\infty.

Then:

  1. For μ\mu-almost every xx, the section y↦f(x,y)y\mapsto f(x,y) is integrable with respect to ν\nu.
  2. For ν\nu-almost every yy, the section x↦f(x,y)x\mapsto f(x,y) is integrable with respect to μ\mu.
  3. The functions obtained by taking these inner integrals are integrable, after assigning arbitrary finite values on any exceptional null sets.
  4. The product integral and both iterated integrals agree:
∫X×Yf d(μ×ν)=∫X(∫Yf(x,y) dν(y))dμ(x)=∫Y(∫Xf(x,y) dμ(x))dν(y).\int_{X\times Y}f\,d(\mu\times\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).

Here, “almost everywhere” allows exceptions on a set of measure zero. The theorem does not require every section to have a finite integral.

In the language of LpL^p spaces, the hypothesis is f∈L1(X×Y,μ×ν)f\in L^1(X\times Y,\mu\times\nu). Real- and complex-valued functions are both covered.

Tonelli’s theorem and absolute integrability

Tonelli’s theorem concerns measurable functions

f:X×Y→[0,∞].f:X\times Y\to[0,\infty].

Under the same sigma-finiteness assumptions, it gives

∫X×Yf d(μ×ν)=∫X∫Yf(x,y) dν(y) dμ(x)=∫Y∫Xf(x,y) dμ(x) dν(y),\int_{X\times Y}f\,d(\mu\times\nu) = \int_X\int_Y f(x,y)\,d\nu(y)\,d\mu(x) = \int_Y\int_X f(x,y)\,d\mu(x)\,d\nu(y),

where the common value may be +∞+\infty.

The distinction is important:

  • Tonelli: nonnegativity permits interchange without first proving finiteness.
  • Fubini: absolute integrability permits interchange for functions that may take both positive and negative values, or complex values.

These results are often presented together as the Fubini–Tonelli theorem.

A standard way to establish Fubini’s hypothesis is to apply Tonelli to ∣f∣|f|. If either iterated integral

∫X∫Y∣f(x,y)∣ dν(y) dμ(x)\int_X\int_Y |f(x,y)|\,d\nu(y)\,d\mu(x)

or its reverse-order counterpart is finite, then the product integral of ∣f∣|f| is finite and Fubini applies.

By contrast, the existence of both iterated integrals of ff itself, even as finite numbers, does not imply absolute integrability or equality.

Interpretation in multivariable calculus

For a continuous function on a compact rectangle, Fubini’s theorem takes the familiar form

∫[a,b]×[c,d]f(x,y) d(x,y)=∫ab(∫cdf(x,y) dy)dx=∫cd(∫abf(x,y) dx)dy.\int_{[a,b]\times[c,d]}f(x,y)\,d(x,y) = \int_a^b\left(\int_c^d f(x,y)\,dy\right)dx = \int_c^d\left(\int_a^b f(x,y)\,dx\right)dy.

The product measure here is two-dimensional Lebesgue measure. Under these continuity assumptions, the integrals also agree with the corresponding Riemann integrals. Continuity is a sufficient condition, not a necessary one; the measure-theoretic theorem applies to much more irregular functions.

Changing the description of a region

For a measurable region DD, integration over DD can be expressed by extending the integrand to be zero outside DD, or equivalently by using its indicator function:

∫Df(x,y) d(x,y)=∫R2f(x,y)1D(x,y) d(x,y).\int_D f(x,y)\,d(x,y) = \int_{\mathbb R^2}f(x,y)\mathbf 1_D(x,y)\,d(x,y).

Changing the order then requires describing the same region by slices in the other direction.

For example, the triangular region

D={(x,y):0≤y≤x≤1}D=\{(x,y):0\le y\le x\le1\}

has the two descriptions

0≤x≤1,0≤y≤x0\le x\le1,\quad 0\le y\le x

and

0≤y≤1,y≤x≤1.0\le y\le1,\quad y\le x\le1.

Consequently, for an integrable ff,

∫01∫0xf(x,y) dy dx=∫01∫y1f(x,y) dx dy.\int_0^1\int_0^x f(x,y)\,dy\,dx = \int_0^1\int_y^1 f(x,y)\,dx\,dy.

Changing the order of integration is distinct from changing variables: the latter generally requires a factor derived from a Jacobian matrix.

Why the hypotheses matter

Consider

f(x,y)=x2−y2(x2+y2)2,0<x<1,0<y<1.f(x,y)=\frac{x^2-y^2}{(x^2+y^2)^2}, \qquad 0<x<1,\quad 0<y<1.

For fixed x>0x>0,

∂∂y(yx2+y2)=x2−y2(x2+y2)2,\frac{\partial}{\partial y} \left(\frac{y}{x^2+y^2}\right) = \frac{x^2-y^2}{(x^2+y^2)^2},

so

∫01f(x,y) dy=11+x2.\int_0^1f(x,y)\,dy=\frac{1}{1+x^2}.

Thus

∫01(∫01f(x,y) dy)dx=π4.\int_0^1\left(\int_0^1f(x,y)\,dy\right)dx =\frac{\pi}{4}.

In the reverse order,

∂∂x(xx2+y2)=−x2−y2(x2+y2)2,\frac{\partial}{\partial x} \left(\frac{x}{x^2+y^2}\right) = -\frac{x^2-y^2}{(x^2+y^2)^2},

giving

∫01(∫01f(x,y) dx)dy=−π4.\int_0^1\left(\int_0^1f(x,y)\,dx\right)dy =-\frac{\pi}{4}.

Both iterated integrals are finite, but they disagree. There is no contradiction: the function is not absolutely integrable near the origin. In polar coordinates,

∣f(x,y)∣=∣cos⁡(2θ)∣r2,|f(x,y)|=\frac{|\cos(2\theta)|}{r^2},

and the area element contributes a factor rr, producing a divergent radial integral proportional to

∫0εdrr.\int_0^\varepsilon\frac{dr}{r}.

Its positive and negative parts both have infinite integral, so its signed Lebesgue integral over the square is undefined.

Other technical qualifications

Sigma-finiteness is a standard sufficient assumption for the product-measure construction and the stated interchange results. Versions with weaker hypotheses exist, but removing sigma-finiteness without replacement can make the conclusions fail.

There is also a distinction between the product sigma-algebra and its completion. For functions measurable only with respect to the completed product measure, section measurability and the corresponding conclusions are generally formulated almost everywhere, using the appropriate completed factor measures.

The theorem does not automatically justify arbitrary manipulations of improper integrals or principal values. Such procedures may involve different limiting operations and require separate arguments.

Main ideas of the proof

The standard proof develops the result from simple functions to general measurable functions.

First, for a measurable rectangle A×BA\times B, the defining property of product measure is

(μ×ν)(A×B)=μ(A)ν(B).(\mu\times\nu)(A\times B)=\mu(A)\nu(B).

This yields the integration identity for rectangle indicators. Measure-theoretic extension arguments then establish the corresponding statement for general measurable sets and nonnegative simple functions.

A nonnegative measurable function can be approximated from below by an increasing sequence of nonnegative simple functions. The monotone convergence theorem passes the integration identities to the limit, proving Tonelli’s theorem.

For an absolutely integrable real-valued function, write

f=f+−f−,f+=max⁡(f,0),f−=max⁡(−f,0).f=f^+-f^-, \qquad f^+=\max(f,0),\quad f^-=\max(-f,0).

Since f++f−=∣f∣f^++f^-=|f|, both parts have finite integrals. Applying Tonelli separately to them and subtracting proves Fubini’s theorem. Complex-valued functions are handled through their real and imaginary parts.

Applications

Probability and expectations

If random variables XX and YY have a joint probability density function p(x,y)p(x,y), then

E[g(X,Y)]=∫R2g(x,y)p(x,y) dx dy.\mathbb E[g(X,Y)] = \int_{\mathbb R^2}g(x,y)p(x,y)\,dx\,dy.

Fubini permits evaluation in either order when

∫R2∣g(x,y)∣p(x,y) dx dy<∞.\int_{\mathbb R^2}|g(x,y)|p(x,y)\,dx\,dy<\infty.

For nonnegative gg, Tonelli permits the same interchange even if the expectation is infinite.

Under independence, the joint law is a product measure. If u(X)u(X) and v(Y)v(Y) are integrable, Tonelli first establishes the absolute integrability of their product, and Fubini gives

E[u(X)v(Y)]=E[u(X)] E[v(Y)].\mathbb E[u(X)v(Y)] = \mathbb E[u(X)]\,\mathbb E[v(Y)].

Infinite sums

Using counting measure on the natural numbers turns integration into summation. Fubini then gives the rearrangement rule for an absolutely summable double series:

∑m=1∞∑n=1∞amn=∑n=1∞∑m=1∞amn,∑m,n∣amn∣<∞.\sum_{m=1}^{\infty}\sum_{n=1}^{\infty}a_{mn} = \sum_{n=1}^{\infty}\sum_{m=1}^{\infty}a_{mn}, \qquad \sum_{m,n}|a_{mn}|<\infty.

For nonnegative terms, Tonelli gives equality without requiring a finite sum.

Combining counting measure with another measure similarly justifies exchanging a sum and an integral when

∑n=1∞∫X∣fn∣ dμ<∞.\sum_{n=1}^{\infty}\int_X|f_n|\,d\mu<\infty.

Convolution and integral transforms

For integrable functions on Rd\mathbb R^d, Tonelli and translation invariance give

∫Rd∫Rd∣f(x−y)g(y)∣ dy dx=∥f∥1∥g∥1.\int_{\mathbb R^d}\int_{\mathbb R^d} |f(x-y)g(y)|\,dy\,dx = \|f\|_1\|g\|_1.

This establishes that their convolution exists almost everywhere and satisfies

∥f∗g∥1≤∥f∥1∥g∥1.\|f*g\|_1\le\|f\|_1\|g\|_1.

Fubini also justifies exchanges of integration used in proving identities for the Fourier transform, provided the relevant absolute-integrability conditions hold.

Historical background

The theorem is named after the Italian mathematician Guido Fubini, whose 1907 work established an important interchange theorem for Lebesgue integration. Leonida Tonelli subsequently developed the complementary nonnegative-function result. Together, these theorems extended the classical treatment of repeated integration to the broader setting of the Lebesgue integral and abstract measure spaces.