Iterated integrals and Fubini's theorem
A steel plate is thicker along one edge than along the others. Its mass is a double integral, a limit of sums over the whole plate. How do you actually compute that limit, and does it matter which way you slice the plate?
Where you are 6 in Multiple integrals
Why this exists
Mei and Hiro build steel frames in a fabrication shop. A triangular gusset plate braces each corner, and it is tapered: thick along the edge welded to the beam, thinner toward the tip. To place the lifting bolt they need the plate’s mass and its balance point. Both are Double integrals as volume, and the definition there is a limit of Riemann sums; the exact values in that lesson came from summation formulas or from integrands that separate into a function of x times a function of y.
Volumes with known cross sections already found volumes by slicing: if the slice at x has area A(x), the volume is \int_a^b A(x)\,dx. For the solid under a surface z = f(x, y), the slice at x is the region under a curve, so A(x) is itself an integral in y. A double integral becomes two single integrals, one inside the other, each done with the antiderivatives of Calculus 1. Fubini’s theorem says when this is legitimate, and that slicing the other way gives the same number.
Definition
Let R = [a, b] \times [c, d] and let f be bounded on R. Fix x in [a, b]. If y \mapsto f(x, y) is integrable on [c, d], write
the integral in y with x held constant, the reverse of taking a partial derivative. If A(x) is defined for every x in [a, b] and A is integrable on [a, b], the iterated integral is
The inner differential belongs to the inner integral sign: dy\,dx means y first. The other order is \int_c^d \int_a^b f(x, y)\,dx\,dy = \int_c^d B(y)\,dy with B(y) = \int_a^b f(x, y)\,dx.
Regions. A region is type I if it has the form
with g_1 \le g_2 continuous on [a, b]: it lies between two graphs over an interval of x. It is type II if D = \{(x, y) : c \le y \le d,\ h_1(y) \le x \le h_2(y)\} with h_1 \le h_2 continuous on [c, d]. Over a type I region the iterated integral is \int_a^b \int_{g_1(x)}^{g_2(x)} f(x, y)\,dy\,dx: the inner limits may depend on the outer variable, the outer limits are constants. Rectangles, triangles and disks are of both types.
Mass. A lamina (a thin plate) occupying a region D, with continuous density \delta(x, y) in mass per unit area, has mass and moments
and center of mass \displaystyle (\bar{x}, \bar{y}) = \left(\frac{M_y}{M}, \frac{M_x}{M}\right). The moment M_x about the x-axis weights each piece by its distance y from that axis. When \delta is constant the center of mass is the centroid of D; for a triangle it is the point of Centroid.
Intuition
Figure 1 shows a solid sliced by the plane x = 1. The slice is the region under the curve z = f(1, y), and its area is A(1). As x runs from a to b the slices sweep out the solid, and adding slabs of volume A(x)\,\Delta x gives \int_a^b A(x)\,dx.
In Riemann sum terms, a double sum on a grid can be added one column at a time:
The inner sum is a Riemann sum for A(x_i^*). For a finite sum the order of addition never matters; Fubini’s theorem says the order survives the limit. Over a type I region the slice at x runs from the lower boundary to the upper one, which is why its limits are g_1(x) and g_2(x).
Results
Theorem 1 (Fubini’s theorem, general form). Let f be integrable over R = [a, b] \times [c, d]. If for every x in [a, b] the function y \mapsto f(x, y) is integrable on [c, d], and A(x) = \int_c^d f(x, y)\,dy is integrable on [a, b], then
The same holds with the roles of x and y exchanged.
Lemma 2 (integrals with a parameter). If h is continuous on [a, b] \times [c, d], then H(x) = \int_c^d h(x, y)\,dy is continuous on [a, b].
Theorem 3 (Fubini’s theorem for continuous functions). If f is continuous on R = [a, b] \times [c, d], then
Theorem 4 (type I and type II regions). Let f be continuous on D. If D is type I, then \iint_D f\,dA = \int_a^b \int_{g_1(x)}^{g_2(x)} f(x, y)\,dy\,dx. If D is type II, then \iint_D f\,dA = \int_c^d \int_{h_1(y)}^{h_2(y)} f(x, y)\,dx\,dy. If D is of both types, the two iterated integrals are equal. With f = 1, A(D) = \int_a^b \left(g_2(x) - g_1(x)\right) dx, the area between two curves from Calculus 1.
Why it is true
Two facts from Real Analysis are used without proof: Result 1 of Double integrals as volume (a bounded function continuous except on finitely many graphs of continuous functions is integrable), and uniform continuity: a function h continuous on a closed bounded set K has, for every \varepsilon > 0, a \delta > 0 with \lvert h(\mathbf{p}) - h(\mathbf{q}) \rvert < \varepsilon whenever \mathbf{p}, \mathbf{q} lie in K and \lVert \mathbf{p} - \mathbf{q} \rVert < \delta.
Theorem 1. Let I = \iint_R f\,dA. For a partition P let m_{ij} and M_{ij} be the infimum and supremum of f on R_{ij} (finite, since f is bounded), and let L(P) = \sum m_{ij}\,\Delta A_{ij} and U(P) = \sum M_{ij}\,\Delta A_{ij}.
First, L(P) and U(P) are close to I. Given \varepsilon > 0, the definition gives \delta > 0 with I - \varepsilon < S(P) < I + \varepsilon for every Riemann sum with \lVert P \rVert < \delta. For any \eta > 0, choose each sample point with f(x_{ij}^*, y_{ij}^*) < m_{ij} + \eta; then S(P) < L(P) + \eta\,A(R), so L(P) > I - \varepsilon - \eta\,A(R). As \eta is arbitrary, L(P) \ge I - \varepsilon. Likewise U(P) \le I + \varepsilon.
Second, \int_a^b A(x)\,dx lies between them. Fix x in [x_{i-1}, x_i]. On [y_{j-1}, y_j], m_{ij} \le f(x, y) \le M_{ij}, so the bounds for single integrals and additivity over the intervals [y_{j-1}, y_j] give
Integrating over [x_{i-1}, x_i] multiplies the outer sides by \Delta x_i, and adding over i gives L(P) \le \int_a^b A(x)\,dx \le U(P). Hence I - \varepsilon \le \int_a^b A(x)\,dx \le I + \varepsilon for every \varepsilon > 0, and the two are equal. Exchanging x and y proves the other order. \blacksquare
Lemma 2. The rectangle is closed and bounded, so h is uniformly continuous. Given \varepsilon > 0, take \delta as above. If \lvert x - x' \rvert < \delta, the points (x, y) and (x', y) are within \delta for every y, so
Theorem 3. A continuous f is integrable over R (Result 1 of Double integrals as volume); each y \mapsto f(x, y) is continuous, hence integrable; and A is continuous by Lemma 2, hence integrable. Theorem 1 gives the first equality, and the same argument with x and y exchanged gives the second. \blacksquare
Theorem 4. Let D be type I. The continuous g_1, g_2 have a least and a greatest value on [a, b] (The extreme value theorem), so we may choose c \le g_1 and d \ge g_2 and put R = [a, b] \times [c, d]. The function f^D (f on D, 0 elsewhere in R) is bounded and continuous except possibly on the graphs of g_1 and g_2 and the lines x = a, x = b, so it is integrable and \iint_D f\,dA = \iint_R f^D\,dA. For fixed x, f^D(x, y) equals f(x, y) for g_1(x) \le y \le g_2(x) and 0 elsewhere in [c, d], so its integral over [c, d] is
To see that A is continuous, substitute y = g_1(x) + t\left(g_2(x) - g_1(x)\right), 0 \le t \le 1:
where k(x, t) = f\left(x,\ g_1(x) + t\left(g_2(x) - g_1(x)\right)\right) (both sides are 0 when g_1(x) = g_2(x)). The point inside f lies in D, and k is continuous on [a, b] \times [0, 1] as a composition of continuous functions. By Lemma 2 the integral in t is continuous in x, so A is continuous, hence integrable, and Theorem 1 applied to f^D gives the formula. Type II is the same with x and y exchanged. If D is of both types, both iterated integrals equal \iint_D f\,dA. \blacksquare
Worked examples
Both orders on a rectangle
Evaluate \iint_R (4 - x^2 + xy)\,dA over R = [0, 2] \times [0, 2] in both orders.
SolutionThe integrand is a polynomial, continuous on R, so Theorem 3 allows either order. With y first, x is constant in the inner integral, so 4 - x^2 integrates to (4 - x^2) \cdot 2 and xy to \displaystyle x \cdot \frac{y^2}{2}:
Then \displaystyle \int_0^2 (8 - 2x^2 + 2x)\,dx = 16 - \frac{16}{3} + 4 = \frac{44}{3}. With x first, y is constant:
and \displaystyle \int_0^2 \left(\frac{16}{3} + 2y\right) dy = \frac{32}{3} + 4 = \frac{44}{3}. The integrand is \ge 0 on R (it is at least 4 - x^2 \ge 0), so \displaystyle \frac{44}{3} is the volume of the solid in Figure 1. Its slice at x = 1 lies under z = 3 + y, with area A(1) = 8.
Choosing the order
Evaluate \iint_R y\cos(xy)\,dA over R = [0, 1] \times [0, \pi].
SolutionIntegrate in x first. For fixed y, \displaystyle \frac{\partial}{\partial x} \sin(xy) = y\cos(xy), so
and \int_0^{\pi} \sin y\,dy = 2. By Theorem 3 the integral is 2.
The other order also works but costs more. For x > 0, integration by parts with u = y and dv = \cos(xy)\,dy gives
which tends to \displaystyle \frac{\pi^2}{2} = \int_0^{\pi} y\,dy as x \to 0. This happens to be the derivative of \displaystyle \frac{1 - \cos(\pi x)}{x}, so its integral over [0, 1] is \displaystyle (1 - \cos \pi) - \lim_{x \to 0^+} \frac{1 - \cos(\pi x)}{x} = 2 - 0 = 2, as Theorem 3 promised. The order is a choice, and the integrand decides which one is short.
One region described both ways
Let D be the region between y = x^2 and y = x + 2. Evaluate \iint_D xy\,dA as a type I and as a type II integral.
SolutionThe curves meet where x^2 = x + 2, at x = -1 and x = 2, that is at (-1, 1) and (2, 4) (Figure 2). Between them x + 2 \ge x^2, so D is type I with -1 \le x \le 2, x^2 \le y \le x + 2. Holding x fixed,
and
As type II, a horizontal line at height y meets D from its left boundary to \displaystyle x = \sqrt{y}. For 0 \le y \le 1 the left boundary is the parabola, \displaystyle x = -\sqrt{y}; for 1 \le y \le 4 it is the line, x = y - 2. The formula for h_1 changes at y = 1, so split D there (additivity) and apply Theorem 4 to each piece:
The first inner integral is 0, since x is odd on an interval symmetric about 0. The second inner integral is \displaystyle \frac{y\left(y - (y - 2)^2\right)}{2} = \frac{-y^3 + 5y^2 - 4y}{2}, and
Reversing the order
Evaluate \displaystyle \int_0^4 \int_{\sqrt{x}}^{2} \sin(y^3)\,dy\,dx.
SolutionThe inner integral cannot be done as written: \sin(y^3) has no elementary antiderivative. The limits describe the type I region 0 \le x \le 4, \displaystyle \sqrt{x} \le y \le 2 (Figure 3). For x, y \ge 0, \displaystyle \sqrt{x} \le y is the same as x \le y^2, so the region is also the type II region 0 \le y \le 2, 0 \le x \le y^2. The integrand is continuous, so by Theorem 4 both iterated integrals equal \iint_D \sin(y^3)\,dA:
about 0.3818. The inner integral in x produced the factor y^2 that substitution with u = y^3 needs.
The gusset plate
Hiro’s gusset plate is the triangle with vertices (0, 0), (3, 0) and (0, 6), in decimeters, welded along its bottom edge. It is tapered, with density \displaystyle \delta(x, y) = \frac{8 - y}{4} kilograms per square decimeter. Find its mass and center of mass.
SolutionThe slanted side runs from (3, 0) to (0, 6), on the line y = 6 - 2x, that is \displaystyle x = 3 - \frac{y}{2}. A horizontal strip (Figure 4) runs from x = 0 to \displaystyle x = 3 - \frac{y}{2}, and the density depends only on y, so use the type II description 0 \le y \le 6, \displaystyle 0 \le x \le 3 - \frac{y}{2}. The inner integral of a function of y alone is that function times the strip’s length, and \displaystyle (8 - y)\left(3 - \frac{y}{2}\right) = 24 - 7y + \frac{y^2}{2}:
For M_y the inner integral is \displaystyle \int_0^{3 - y/2} x\,dx = \frac12\left(3 - \frac{y}{2}\right)^2, so
The mass is 13.5 kilograms, \displaystyle \bar{x} = \frac{117/8}{27/2} = \frac{13}{12} and \displaystyle \bar{y} = \frac{45/2}{27/2} = \frac53, so the center of mass is \displaystyle \left(\frac{13}{12}, \frac53\right).
A plate of constant density would balance at the centroid (1, 2), the average of the vertices. The extra metal near the bottom edge pulls the balance point down, and since the triangle is widest there, a little to the right.
A double integral that proves a single-variable formula
Prove that for f continuous on [a, b],
Then evaluate \int_0^1 \arctan x\,dx.
SolutionWork in the plane with coordinates (x, t) and let T = \{a \le x \le b,\ a \le t \le x\}, a triangle. It is type I in the form given, and type II as a \le t \le b, t \le x \le b. The function (x, t) \mapsto f(t) is continuous on T, so by Theorem 4
because in the inner integral f(t) is constant and the interval [t, b] has length b - t. \blacksquare
Take \displaystyle f(t) = \frac{1}{1 + t^2} on [0, 1]. The inner integral on the left is \displaystyle \int_0^x \frac{dt}{1 + t^2} = \arctan x, so
since \displaystyle \int_0^1 \frac{t}{1 + t^2}\,dt = \frac{\ln 2}{2} by the substitution u = 1 + t^2. No integration by parts was needed.
Boundaries
Integrability is a hypothesis. Let \displaystyle f(x, y) = \frac{x^2 - y^2}{(x^2 + y^2)^2} on [0, 1] \times [0, 1], away from the origin. For x > 0, \displaystyle \frac{y}{x^2 + y^2} is an antiderivative in y, so \displaystyle \int_0^1 f\,dy = \frac{1}{1 + x^2} and the dy\,dx integral is \displaystyle \frac{\pi}{4}. For y > 0, \displaystyle -\frac{x}{x^2 + y^2} is an antiderivative in x, so \displaystyle \int_0^1 f\,dx = -\frac{1}{1 + y^2} and the dx\,dy integral is \displaystyle -\frac{\pi}{4}. There is no contradiction: \displaystyle f(x, 0) = \frac{1}{x^2}, so f is unbounded, not integrable, and Theorem 1 does not apply.
Equal iterated integrals do not make f integrable. On [0, 1] \times [0, 1] let f = 1 at the points \displaystyle \left(\frac{p}{q}, \frac{r}{q}\right) with both fractions in lowest terms over the same positive integer q, and f = 0 elsewhere. A vertical line \displaystyle x = \frac{p}{q} meets at most q + 1 of these points and other vertical lines meet none, so every inner integral is 0; the same holds for horizontal lines, and both iterated integrals are 0. But every subrectangle contains such a point (take q a large prime) and a point with x irrational, so Riemann sums 0 and 1 occur for every partition and f is not integrable.
Improper integrals. The theorems here are for bounded functions on bounded regions. For unbounded ones, swapping the order is justified when the integral of \lvert f \rvert is finite (the Fubini and Tonelli theorems of measure theory, beyond this course); the first example shows what goes wrong without it.
Regions of neither type. The ring 1 \le x^2 + y^2 \le 4 meets the line x = 0 in two separate segments, so it is not type I, and likewise not type II. It splits into four pieces of one type or the other, or is better handled by Double integrals in polar coordinates.
Where this goes next
Double integrals in polar coordinates writes integrals over disks and sectors as iterated integrals in r and \theta. Triple integrals nests three single integrals, with Fubini’s theorem again justifying the order, and Change of variables and the Jacobian chooses coordinates in which the region becomes a rectangle. The proof of Green's theorem works on a region of both types, using Theorem 4 once in each direction. In Joint distributions, a marginal density is exactly an inner integral \int f(x, y)\,dy.
Exercises
Rectangles
- Evaluate \int_0^2 \int_0^3 (x^2 y + 2y)\,dy\,dx.
- Evaluate \displaystyle \iint_R \frac{1}{(x + y)^2}\,dA over R = [1, 2] \times [0, 1].
- Evaluate \iint_R x e^{xy}\,dA over R = [0, 1] \times [1, 2], choosing the easier order.
- Find the average value of \sin(x + y) over \displaystyle R = \left[0, \frac{\pi}{2}\right] \times \left[0, \frac{\pi}{2}\right].
General regions
- Let D be bounded by \displaystyle y = \sqrt{x}, y = 0 and x = 4. Evaluate \iint_D x\,dA in both orders.
- Let D be the triangle with vertices (0, 0), (2, 2) and (4, 0). Evaluate \iint_D y\,dA, and explain why one order needs a single iterated integral and the other needs two.
- Evaluate \iint_D e^{x/y}\,dA, where D = \{1 \le y \le 2,\ y \le x \le y^3\}.
- Find the volume of the solid under z = 1 + xy over the triangle with vertices (0, 0), (1, 0) and (1, 1).
Reversing the order
- Evaluate \int_0^2 \int_{y/2}^{1} e^{x^2}\,dx\,dy.
- Evaluate \displaystyle \int_0^1 \int_{\sqrt{y}}^{1} \sqrt{x^3 + 1}\,dx\,dy.
- Evaluate \displaystyle \int_0^{\pi} \int_x^{\pi} \frac{\sin y}{y}\,dy\,dx.
- Write each with the order reversed: (a) \displaystyle \int_0^1 \int_{x^2}^{\sqrt{x}} f(x, y)\,dy\,dx; (b) \displaystyle \int_{-2}^{2} \int_0^{\sqrt{4 - x^2}} f(x, y)\,dy\,dx; (c) \int_1^e \int_0^{\ln x} f(x, y)\,dy\,dx.
Mass and center of mass
- A lamina occupies the region between y = 4 - x^2 and the x-axis, with density \delta(x, y) = y. Find its mass and center of mass.
- Find the centroid of the quarter disk x^2 + y^2 \le 1, x \ge 0, y \ge 0.
When the theorem does not apply
- Compute both iterated integrals of \displaystyle f(x, y) = \frac{x - y}{(x + y)^3} over [0, 1] \times [0, 1], and explain why this does not contradict Theorem 1.
Proofs
- Derive Result 4 of Double integrals as volume, \iint_R g(x)h(y)\,dA = \left(\int_a^b g\right)\left(\int_c^d h\right) for continuous g and h, from Theorem 3.
- Let f be continuous on R = [a, b] \times [c, d] and F(x, y) = \int_a^x \int_c^y f(s, t)\,dt\,ds for (x, y) in R. Prove that F_{xy} = F_{yx} = f at every interior point of R.
- Let f and f_x be continuous on R = [a, b] \times [c, d] and H(x) = \int_c^d f(x, y)\,dy. Prove that H'(x) = \int_c^d f_x(x, y)\,dy for a < x < b.
- Prove that for f continuous on [a, b], \left(\int_a^b f(x)\,dx\right)^2 \le (b - a) \int_a^b f(x)^2\,dx, by integrating \left(f(x) - f(y)\right)^2 over [a, b] \times [a, b].
Answers
- 30. The inner integral is \displaystyle (x^2 + 2)\frac{9}{2}, and \displaystyle \frac92\int_0^2 (x^2 + 2)\,dx = \frac92 \cdot \frac{20}{3} = 30.
- \displaystyle \ln \frac43. The inner integral in y is \displaystyle \frac1x - \frac{1}{x + 1}, and \displaystyle \int_1^2 \left(\frac1x - \frac{1}{x + 1}\right) dx = \ln 2 - (\ln 3 - \ln 2), which is 2\ln 2 - \ln 3.
- \displaystyle \frac{(e - 1)^2}{2}. Integrate in y first: \int_1^2 x e^{xy}\,dy = e^{2x} - e^{x}, and \displaystyle \int_0^1 (e^{2x} - e^{x})\,dx = \frac{e^2 - 1}{2} - (e - 1), which is \displaystyle \frac{e^2 - 2e + 1}{2}. The x-first order needs integration by parts.
- \displaystyle \frac{8}{\pi^2}. The inner integral is \Big[-\cos(x + y)\Big]_{y=0}^{y=\pi/2} = \cos x + \sin x, whose integral over \displaystyle \left[0, \frac{\pi}{2}\right] is 2; the area is \displaystyle \frac{\pi^2}{4}.
- \displaystyle \frac{64}{5}. Type I: \displaystyle \int_0^4 \int_0^{\sqrt{x}} x\,dy\,dx = \int_0^4 x^{3/2}\,dx = \frac{64}{5}. Type II: 0 \le y \le 2, y^2 \le x \le 4, and \displaystyle \int_0^2 \frac{16 - y^4}{2}\,dy = \frac12\left(32 - \frac{32}{5}\right) = \frac{64}{5}.
- \displaystyle \frac83. As type II, 0 \le y \le 2, y \le x \le 4 - y, so \displaystyle \int_0^2 y(4 - 2y)\,dy = 8 - \frac{16}{3} = \frac83, one integral. As type I the top boundary is y = x for 0 \le x \le 2 and y = 4 - x for 2 \le x \le 4, a change of formula, so two integrals: \displaystyle \int_0^2 \frac{x^2}{2}\,dx + \int_2^4 \frac{(4 - x)^2}{2}\,dx = \frac43 + \frac43.
- \displaystyle \frac{e^4 - 4e}{2}. For fixed y, \int_y^{y^3} e^{x/y}\,dx = y\left(e^{y^2} - e\right), and \displaystyle \int_1^2 \left(y e^{y^2} - e y\right) dy = \frac{e^4 - e}{2} - \frac{3e}{2}.
- \displaystyle \frac58. The triangle is 0 \le x \le 1, 0 \le y \le x, and \displaystyle \int_0^1 \left(x + \frac{x^3}{2}\right) dx = \frac12 + \frac18.
- e - 1. The region 0 \le y \le 2, \displaystyle \frac{y}{2} \le x \le 1 is 0 \le x \le 1, 0 \le y \le 2x, so the integral is \int_0^1 2x e^{x^2}\,dx = e - 1.
- \displaystyle \frac29\left(2\sqrt{2} - 1\right). The region is 0 \le x \le 1, 0 \le y \le x^2, giving \displaystyle \int_0^1 x^2\sqrt{x^3 + 1}\,dx; with u = x^3 + 1 this is \displaystyle \frac13 \int_1^2 \sqrt{u}\,du = \frac29\left(2^{3/2} - 1\right).
- 2. The region 0 \le x \le \pi, x \le y \le \pi is 0 \le y \le \pi, 0 \le x \le y; with the integrand taken as 1 where y = 0 it is continuous, and \displaystyle \int_0^{\pi} \frac{\sin y}{y} \cdot y\,dy = 2.
- (a) \displaystyle \int_0^1 \int_{y^2}^{\sqrt{y}} f(x, y)\,dx\,dy, since between the curves y = x^2 and \displaystyle y = \sqrt{x} a horizontal line at height y runs from x = y^2 to \displaystyle x = \sqrt{y}. (b) \displaystyle \int_0^2 \int_{-\sqrt{4 - y^2}}^{\sqrt{4 - y^2}} f(x, y)\,dx\,dy, the upper half disk of radius 2. (c) \int_0^1 \int_{e^y}^{e} f(x, y)\,dx\,dy, since y \le \ln x means x \ge e^y.
- \displaystyle M = \frac{256}{15}, \displaystyle (\bar{x}, \bar{y}) = \left(0, \frac{16}{7}\right). \displaystyle M = \int_{-2}^{2} \frac{(4 - x^2)^2}{2}\,dx = \frac{256}{15}; M_y = 0, since xy is odd in x over a region symmetric about the y-axis; \displaystyle M_x = \int_{-2}^{2} \frac{(4 - x^2)^3}{3}\,dx = \frac{4096}{105}, and \displaystyle \frac{4096/105}{256/15} = \frac{16}{7}.
- \displaystyle \left(\frac{4}{3\pi}, \frac{4}{3\pi}\right). The area is \displaystyle \frac{\pi}{4} and \displaystyle \iint_D x\,dA = \int_0^1 x\sqrt{1 - x^2}\,dx = \frac13, so \displaystyle \bar{x} = \frac{1/3}{\pi/4}; \bar{y} = \bar{x} by symmetry in the line y = x.
- \displaystyle \frac12 in the order dy\,dx and \displaystyle -\frac12 in the order dx\,dy. For x > 0, \displaystyle \frac{y}{(x + y)^2} is an antiderivative in y, so the inner integral is \displaystyle \frac{1}{(1 + x)^2}, whose integral over [0, 1] is \displaystyle \frac12. For y > 0, \displaystyle -\frac{x}{(x + y)^2} is an antiderivative in x, giving \displaystyle -\frac{1}{(1 + y)^2} and \displaystyle -\frac12. Along y = 0, \displaystyle f(x, 0) = \frac{1}{x^2}, so f is unbounded on the square, hence not integrable, and Theorem 1 does not apply.
- g(x)h(y) is continuous on R, so Theorem 3 gives \iint_R g(x)h(y)\,dA = \int_a^b \left(\int_c^d g(x)h(y)\,dy\right) dx. In the inner integral g(x) is a constant, so it equals g(x)\int_c^d h(y)\,dy; the number \int_c^d h then comes out of the outer integral by linearity, leaving \left(\int_c^d h\right)\left(\int_a^b g\right).
- Fix y. By Lemma 2 applied on [a, b] \times [c, y] (trivial when y = c), G(s) = \int_c^y f(s, t)\,dt is continuous, so the fundamental theorem of calculus gives F_x(x, y) = G(x) = \int_c^y f(x, t)\,dt. For fixed x, t \mapsto f(x, t) is continuous, so the same theorem gives (F_x)_y = f(x, y), that is F_{xy} = f. By Theorem 3 on [a, x] \times [c, y], F(x, y) = \int_c^y \int_a^x f(s, t)\,ds\,dt, and the same argument with the variables exchanged gives F_{yx} = f.
- For a \le x \le b, apply Theorem 3 to f_x on [a, x] \times [c, d]: \int_a^x \left(\int_c^d f_x(s, y)\,dy\right) ds = \int_c^d \left(\int_a^x f_x(s, y)\,ds\right) dy. For fixed y, s \mapsto f(s, y) has continuous derivative f_x(s, y), so the inner integral on the right is f(x, y) - f(a, y), and the right side is H(x) - H(a). On the left, K(s) = \int_c^d f_x(s, y)\,dy is continuous by Lemma 2, so the fundamental theorem of calculus says the left side has derivative K(x). Hence H'(x) = K(x) = \int_c^d f_x(x, y)\,dy.
- Let S = [a, b] \times [a, b]. The function \left(f(x) - f(y)\right)^2 is continuous and \ge 0, so its integral over S is \ge 0 (monotonicity). Expanding, f(x)^2 - 2f(x)f(y) + f(y)^2, and by Theorem 3 (or Exercise 16), \iint_S f(x)^2\,dA = (b - a)\int_a^b f^2, the same for f(y)^2, and \iint_S f(x)f(y)\,dA = \left(\int_a^b f\right)^2. So 0 \le 2(b - a)\int_a^b f^2 - 2\left(\int_a^b f\right)^2, which is the inequality.