1.3 Double and Triple Integrals: Area, Volume and Surface Area
Double and triple integrals from Riemann sums to Fubini's theorem, changing the order of integration, polar, cylindrical and spherical coordinates through the Jacobian, and the applications GATE asks for — area, volume, mass, centroid and surface area — ending with improper integrals over the whole plane and the Gaussian integral.
A one-variable integral adds up a quantity along an interval: the area under a curve, the work done by a varying force, the mass of a thin rod. Most quantities in science live on regions of the plane or of space instead — the rainfall over a catchment, the mass of a metal plate whose thickness varies, the volume of a reservoir, the charge in a cloud, the probability that a random point lands in a set. The double and triple integrals of this section add up such quantities. They are defined by Riemann sums exactly as in one variable, but they are computed by two ideas that have no one-variable counterpart. Fubini's theorem turns a double integral into two single integrals taken one after the other, in either order — and choosing the order well can turn an impossible integral into an easy one. The change-of-variables formula lets us replace $x,y$ by coordinates adapted to the region (polar, cylindrical, spherical, or a linear substitution) at the price of one factor, the absolute value of the Jacobian, which measures how the new coordinates stretch area.
GATE MA asks about this section every year, almost always as computation. MCQs and NATs ask for the value of an integral after reversing the order or switching to polar coordinates, for the volume of a solid bounded by a paraboloid, a cone, a sphere or a cylinder, for a centroid or a mass, or for the area of a piece of a surface, correct to two decimal places. MSQs test the hypotheses: when two iterated integrals agree, why the Jacobian carries an absolute value, which angle ranges are allowed in spherical coordinates, when an improper double integral converges. The derivative tools of §1.1 — Jacobians, the chain rule, the inverse function theorem — are used throughout, and the surface-area formulas here are the starting point of the surface integrals in §2.2.
Double Integrals and Fubini's Theorem
We start on a rectangle, where partitions are easy to describe, and then move to general regions by extending functions by zero.
Let $R=[a,b]\times[c,d]$ and let $f:R\to\mathbb{R}$ be bounded. A partition $P$ of $R$ is given by points $a=x_0\lt x_1\lt\cdots\lt x_m=b$ and $c=y_0\lt y_1\lt\cdots\lt y_n=d$; it cuts $R$ into the cells $R_{ij}=[x_{i-1},x_i]\times[y_{j-1},y_j]$ of area $\Delta A_{ij}=\Delta x_i\,\Delta y_j$. With $m_{ij}=\inf_{R_{ij}}f$ and $M_{ij}=\sup_{R_{ij}}f$, the lower and upper sums are
$$L(f,P)=\sum_{i,j}m_{ij}\,\Delta A_{ij},\qquad U(f,P)=\sum_{i,j}M_{ij}\,\Delta A_{ij}.$$$f$ is (Riemann) integrable on $R$ if $\sup_PL(f,P)=\inf_PU(f,P)$; the common value is the double integral $\iint_Rf\,dA$. Equivalently (Riemann's criterion), for every $\varepsilon\gt0$ there is a partition with $U(f,P)-L(f,P)\lt\varepsilon$. As in one variable, the Riemann sums $\sum f(\xi_{ij})\Delta A_{ij}$, with any sample points $\xi_{ij}\in R_{ij}$, then converge to $\iint_Rf\,dA$ as the largest cell diameter tends to $0$. If $f\ge0$, the integral is the volume of the solid between the graph of $f$ and $R$.
- A set $Z\subseteq\mathbb{R}^2$ has zero area if for every $\varepsilon\gt0$ it can be covered by finitely many rectangles of total area less than $\varepsilon$. (Finite sets, segments and, as we shall prove, graphs of continuous functions have zero area.)
- Let $D$ be bounded, $R\supseteq D$ a rectangle, and $f:D\to\mathbb{R}$ bounded. Put $f_D=f$ on $D$ and $f_D=0$ on $R\setminus D$. Then $\iint_Df\,dA=\iint_Rf_D\,dA$ whenever the right side exists; it does not depend on the choice of $R$. A bounded set $D$ whose boundary has zero area is a region, and its area is $\operatorname{Area}(D)=\iint_D1\,dA$.
- A type I region is $D=\{(x,y):a\le x\le b,\ g_1(x)\le y\le g_2(x)\}$ with $g_1\le g_2$ continuous on $[a,b]$ (vertical strips); a type II region is $D=\{(x,y):c\le y\le d,\ h_1(y)\le x\le h_2(y)\}$ with $h_1\le h_2$ continuous (horizontal strips).
- Every continuous function on a rectangle $R$ is integrable on $R$.
- The graph $\{(x,g(x)):a\le x\le b\}$ of a continuous function $g$ has zero area. Hence the boundary of a type I or type II region has zero area.
- If $f$ is bounded on $R$ and continuous at every point outside a set of zero area, then $f$ is integrable on $R$. In particular every continuous function on a type I or type II region $D$ — or on a finite union of such regions — is integrable on $D$.
Proof. (1) $R$ is closed and bounded, so $f$ is uniformly continuous on it: given $\varepsilon\gt0$ there is $\delta\gt0$ with $|f(p)-f(q)|\lt\varepsilon/\operatorname{Area}(R)$ whenever $\|p-q\|\lt\delta$. Take a partition whose cells have diameter less than $\delta$. On each (closed) cell $f$ attains its maximum and minimum at two points less than $\delta$ apart, so $M_{ij}-m_{ij}\lt\varepsilon/\operatorname{Area}(R)$, and $U-L\lt\varepsilon$.
(2) Given $\varepsilon\gt0$, uniform continuity of $g$ gives $n$ with $|g(s)-g(t)|\lt\varepsilon$ whenever $|s-t|\le(b-a)/n$. Cut $[a,b]$ into $n$ equal pieces $[t_{k-1},t_k]$. Over each piece the graph lies in the rectangle $[t_{k-1},t_k]\times[g(t_k)-\varepsilon,g(t_k)+\varepsilon]$, and these $n$ rectangles have total area $n\cdot\frac{b-a}n\cdot2\varepsilon=2\varepsilon(b-a)$, as small as we like. The boundary of a type I region consists of two such graphs and two vertical segments.
(3) Let $|f|\le M$ and let $Z$ be the set where $f$ is discontinuous. Given $\varepsilon\gt0$, cover $Z$ by finitely many rectangles of total area less than $\varepsilon$, and enlarge each slightly so that $Z$ lies in the union $W$ of their interiors and the total area is still less than $2\varepsilon$. The set $K=R\setminus W$ is closed and bounded, and $f$ is continuous on $K$, hence uniformly continuous there. Take a partition $P$ whose grid lines include all edges of the covering rectangles and whose cells are so small that $M_{ij}-m_{ij}\lt\varepsilon$ for every cell lying in $K$. Every cell either lies in a covering rectangle or lies in $K$. The first kind contribute at most $2M\cdot2\varepsilon$ to $U-L$, the second kind at most $\varepsilon\operatorname{Area}(R)$. So $U-L\le\varepsilon\left(4M+\operatorname{Area}(R)\right)$, which is arbitrarily small. For a continuous $f$ on a type I region $D$, the extension $f_D$ is continuous except possibly on $\partial D$, which has zero area by (2). ∎
Let $f,g$ be integrable on a region $D$ and $c\in\mathbb{R}$.
- Linearity: $\iint_D(f+cg)\,dA=\iint_Df\,dA+c\iint_Dg\,dA$.
- Monotonicity: $f\le g$ on $D$ implies $\iint_Df\le\iint_Dg$; also $|f|$ is integrable and $\left|\iint_Df\right|\le\iint_D|f|$.
- Sets of zero area do not count: changing $f$ on a set of zero area does not change $\iint_Df$; if $D=D_1\cup D_2$ where $D_1\cap D_2$ has zero area, then $\iint_D=\iint_{D_1}+\iint_{D_2}$.
- Mean value theorem: if $D$ is a closed, bounded, connected region and $f$ is continuous, there is $p\in D$ with $\iint_Df\,dA=f(p)\operatorname{Area}(D)$.
Proof. (1) and the first part of (2) hold for every Riemann sum and pass to the limit. On each cell the oscillation $\sup|f|-\inf|f|$ is at most $\sup f-\inf f$, so $|f|$ satisfies Riemann's criterion whenever $f$ does; then $-|f|\le f\le|f|$ gives the inequality. (3) If $h$ vanishes outside a set $Z$ of zero area and $|h|\le M$, cover $Z$ as in the proof of Theorem 1.16(3): only cells inside the covering rectangles meet $Z$, so $|U(h,P)|,|L(h,P)|\le2M\varepsilon$, and $\iint h=0$. Apply this to the difference of the two functions, and to $f_D-f_{D_1}-f_{D_2}$, which vanishes outside $D_1\cap D_2$. (4) By the extreme value theorem $f$ has a minimum $m=f(p_1)$ and a maximum $M=f(p_2)$ on $D$, and by (2) $m\operatorname{Area}(D)\le\iint_Df\le M\operatorname{Area}(D)$. If $\operatorname{Area}(D)=0$ any $p$ works. Otherwise the number $\iint_Df/\operatorname{Area}(D)$ lies in $[m,M]$, and the continuous image $f(D)$ of the connected set $D$ is an interval containing $m$ and $M$, hence containing it. ∎
The number $\frac{1}{\operatorname{Area}(D)}\iint_Df\,dA$ is the average value of $f$ on $D$. Definitions are not how we compute; this is.
(a) Let $f$ be integrable on $R=[a,b]\times[c,d]$, and suppose that for every $x\in[a,b]$ the function $y\mapsto f(x,y)$ is integrable on $[c,d]$. Then $F(x)=\int_c^df(x,y)\,dy$ is integrable on $[a,b]$ and
$$\iint_Rf\,dA=\int_a^b\left(\int_c^df(x,y)\,dy\right)dx.$$The same holds with the roles of $x$ and $y$ exchanged. In particular, for $f$ continuous on $R$ both iterated integrals exist and equal $\iint_Rf\,dA$.
(b) If $f$ is continuous on a type I region $D$, then $\displaystyle\iint_Df\,dA=\int_a^b\int_{g_1(x)}^{g_2(x)}f(x,y)\,dy\,dx$; on a type II region, $\displaystyle\iint_Df\,dA=\int_c^d\int_{h_1(y)}^{h_2(y)}f(x,y)\,dx\,dy$. For a region of both types the two iterated integrals are equal — this is changing the order of integration.
Proof. (a) Take partitions $P_1$ of $[a,b]$, $P_2$ of $[c,d]$ and $P=P_1\times P_2$. Fix a cell column $[x_{i-1},x_i]$ and a point $x$ in it. For each $j$, $m_{ij}\le f(x,y)\le M_{ij}$ for $y\in[y_{j-1},y_j]$; integrating in $y$ and adding over $j$,
$$\sum_jm_{ij}\,\Delta y_j\ \le\ F(x)\ \le\ \sum_jM_{ij}\,\Delta y_j .$$This holds for every $x$ in the column, so the infimum and the supremum of $F$ over $[x_{i-1},x_i]$ also lie between these two numbers. Multiply by $\Delta x_i$ and add over $i$:
$$L(f,P)\le L(F,P_1)\le U(F,P_1)\le U(f,P).$$Because $f$ is integrable, the outer two numbers can be brought within $\varepsilon$ of each other and of $\iint_Rf$ by choosing $P$ fine. Then $U(F,P_1)-L(F,P_1)\lt\varepsilon$, so $F$ is integrable, and its integral, trapped between $L(f,P)$ and $U(f,P)$ for every $P$, equals $\iint_Rf\,dA$. For continuous $f$ both hypotheses hold (Theorem 1.16(1), and continuity of $y\mapsto f(x,y)$).
(b) Enclose $D$ in a rectangle $R=[a,b]\times[c,d]$. The extension $f_D$ is integrable by Theorem 1.16(3). For fixed $x$, the function $y\mapsto f_D(x,y)$ is continuous on $[c,d]$ except possibly at the two points $g_1(x)$, $g_2(x)$, so it is integrable, with $\int_c^df_D(x,y)\,dy=\int_{g_1(x)}^{g_2(x)}f(x,y)\,dy$. Now apply (a). ∎
- Sketch the region. Find the intersection points of the boundary curves; mark which curve is on top, which is on the right.
- Vertical strips ($dy\,dx$): the outer limits are the extreme values of $x$ (constants); for a fixed $x$ the strip runs from the lower curve $y=g_1(x)$ to the upper curve $y=g_2(x)$. Horizontal strips ($dx\,dy$): outer limits are the extreme values of $y$; the strip runs from the left curve $x=h_1(y)$ to the right curve $x=h_2(y)$.
- To reverse an order, never manipulate the limits algebraically: read the region off the given limits, draw it, and describe it again by strips the other way. If the "left" or "right" curve changes part-way, split the outer interval.
- When to reverse: the inner integrand has no elementary antiderivative ($e^{y^2}$, $\sin(y^3)$, $\frac{\sin y}{y}$, $\sqrt{1+x^3}$), or one order needs several pieces while the other needs one.
- Check: integrate $1$ over the region both ways — you should get the same area.
The hypothesis "$f$ integrable on $R$" in Fubini's theorem is not decoration. Without it the two iterated integrals can both exist and be different.
Change of Variables: Polar Coordinates and the Jacobian
A disc is a rectangle in polar coordinates; a parallelogram is a square in the right linear coordinates. To use this we must know how a change of variables distorts area. For a linear map the answer is exact: if $A$ is a $2\times2$ matrix and $Q$ a rectangle of sides $h,k$, then $A(Q)$ is the parallelogram spanned by $hAe_1$ and $kAe_2$, whose area is $|\det(hAe_1,\,kAe_2)|=|\det A|\,hk$. (The area of the parallelogram spanned by $(p,q)$ and $(r,s)$ is $|ps-qr|$; in $\mathbb{R}^3$ the volume of a parallelepiped is the absolute value of the determinant of its three edge vectors.) A differentiable map is linear to first order (§1.1), so on a tiny square it multiplies area by about $|\det DT|$ — the absolute value of its Jacobian.
Let $T:U\to\mathbb{R}^n$ ($n=2$ or $3$) be $C^1$ on an open set $U$, and let $D^*\subseteq U$ be a closed bounded region (boundary of zero area or zero volume). Suppose that $T$ is injective on the interior of $D^*$ and that its Jacobian $J_T=\det DT$ is nonzero there. Then $D=T(D^*)$ is a region, and for every $f$ continuous on $D$
$$\int_Df(x)\,dx=\int_{D^*}f\big(T(u)\big)\,\big|J_T(u)\big|\,du .$$For $n=2$, with $(x,y)=T(u,v)$: $\ \iint_Df\,dx\,dy=\iint_{D^*}f\big(x(u,v),y(u,v)\big)\left|\dfrac{\partial(x,y)}{\partial(u,v)}\right|du\,dv$.
Polar coordinates. $x=r\cos\theta$, $y=r\sin\theta$ has $\dfrac{\partial(x,y)}{\partial(r,\theta)}=r$. If $D=\{(r\cos\theta,r\sin\theta):\alpha\le\theta\le\beta,\ h_1(\theta)\le r\le h_2(\theta)\}$ with $0\le h_1\le h_2$ continuous and $\beta-\alpha\le2\pi$, then $$\iint_Df\,dA=\int_\alpha^\beta\int_{h_1(\theta)}^{h_2(\theta)}f(r\cos\theta,r\sin\theta)\,r\,dr\,d\theta .$$
Proof (careful sketch). Step 1 (linear maps). For a linear $T=A$ the image of a square $Q$ has area (volume) exactly $|\det A|$ times that of $Q$, as computed above.
Step 2 (small squares). On a compact subset of the interior of $D^*$, $T$ is uniformly differentiable: for every $\eta\gt0$ there is $\delta_0$ such that $\|T(q)-T(p)-DT(p)(q-p)\|\le\eta\|q-p\|$ whenever $\|q-p\|\le\delta_0\sqrt n$. So if $Q$ is a square of side $\delta\le\delta_0$ centred at $p$, the curved region $T(Q)$ lies within distance $\eta\sqrt n\,\delta$ of the parallelogram $T(p)+DT(p)(Q-p)$, whose area is $|J_T(p)|\delta^n$ by Step 1. A band of width $\eta\sqrt n\,\delta$ around the boundary of that parallelogram has area at most $C\eta\delta^n$ ($C$ depending only on a bound for $\|DT\|$), hence
$$\big|\operatorname{Area}T(Q)-|J_T(p)|\,\delta^n\big|\le C\eta\,\delta^n .$$Step 3 (adding up). Cover $D^*$ by a grid of squares of side $\delta$. The squares meeting $\partial D^*$ have total area tending to $0$ with $\delta$, and since $T$ is Lipschitz on compact sets their images have total area tending to $0$ as well. The images of the interior squares $Q_k$ overlap only along their edges (injectivity), so $\int_Df=\sum_k\int_{T(Q_k)}f+o(1)$. By uniform continuity of $f$, $\int_{T(Q_k)}f=f(T(p_k))\operatorname{Area}T(Q_k)+o(\delta^n)$, and by Step 2 this is $f(T(p_k))\,|J_T(p_k)|\,\delta^n+o(\delta^n)$. There are $O(\delta^{-n})$ squares, so $\int_Df=\sum_kf(T(p_k))|J_T(p_k)|\delta^n+o(1)$ — a Riemann sum of $f(T)\,|J_T|$ over $D^*$, which tends to the right-hand side as $\delta\to0$. Where injectivity or $J_T\ne0$ fails only on the boundary of $D^*$ (a set of zero area), the argument is unaffected. ∎
For the polar map, $T(r,\theta)=(r\cos\theta,r\sin\theta)$ on $[0,R]\times[0,2\pi]$ is not injective on the edges $r=0$ and $\theta\in\{0,2\pi\}$ and has $J=r=0$ at $r=0$; these edges have zero area, so the theorem applies. Its Jacobian is $\det\begin{pmatrix}\cos\theta&-r\sin\theta\\\sin\theta&r\cos\theta\end{pmatrix}=r$. The factor $r$ has a picture: the polar "rectangle" $[r,r+dr]\times[\theta,\theta+d\theta]$ is a piece of annulus of width $dr$ and arc length $r\,d\theta$.
- Polar when the region is a disc, an annulus, a sector or bounded by curves $r=h(\theta)$ (circles through the origin $r=2a\cos\theta$, cardioids, rose petals), or when the integrand contains $x^2+y^2$. Never forget the factor $r$.
- Linear substitution $u=ax+by$, $v=cx+dy$ when the region is a parallelogram bounded by lines $ax+by=$ const, $cx+dy=$ const; then $dA=\dfrac{du\,dv}{|ad-bc|}$.
- Adapted curvilinear coordinates ($u=xy$, $v=y/x$; $u=x^2-y^2$, $v=xy$; …) when the boundary consists of level curves of two functions. Compute whichever Jacobian is easier and take the reciprocal if necessary.
- Cylindrical for solids with an axis of symmetry (paraboloids, cylinders, cones with vertical axis); spherical for balls, spherical shells and cones with vertex at the origin.
- Always use $|J|$, and check that the new region is covered exactly once.
Triple Integrals: Cylindrical and Spherical Coordinates
(a) If $E$ is $z$-simple and $f$ is continuous on $E$, then
$$\iiint_Ef\,dV=\iint_D\left(\int_{\varphi_1(x,y)}^{\varphi_2(x,y)}f(x,y,z)\,dz\right)dA,$$and the double integral over $D$ is then computed by Theorem 1.18. (b) Cavalieri. If $E$ lies between the planes $z=e$ and $z=g$, and each cross-section $E_z=\{(x,y):(x,y,z)\in E\}$ is a region of area $A(z)$, then $\operatorname{Vol}(E)=\int_e^gA(z)\,dz$ (here $E$ is a solid region, as are all solids bounded by finitely many graphs of continuous functions).
Proof. (a) Write a partition of a box $B=R\times[e,g]\supseteq E$ as the product of a partition of the rectangle $R$ into cells $R_k$ and a partition of $[e,g]$ into intervals $I_l$. Exactly as in the proof of Theorem 1.18(a), for $(x,y)\in R_k$ the inner integral $F(x,y)=\int_e^gf_E(x,y,z)\,dz$ lies between $\sum_lm_{kl}|I_l|$ and $\sum_lM_{kl}|I_l|$, which gives $L(f_E,P)\le L(F,P')\le U(F,P')\le U(f_E,P)$; and $f_E$ is integrable because it is continuous off the graphs of $\varphi_1,\varphi_2$ (and the vertical walls over $\partial D$), which have zero volume by the argument of Theorem 1.16(2). For fixed $(x,y)\in D$, $F(x,y)=\int_{\varphi_1}^{\varphi_2}f\,dz$. (b) is the same argument with the variables grouped as $(x,y)$ inside and $z$ outside, applied to $f=1$: $\iiint_E1\,dV=\int_e^g\left(\iint_{E_z}1\,dA\right)dz$. ∎
- Cylindrical coordinates $(r,\theta,z)$: $x=r\cos\theta$, $y=r\sin\theta$, $z=z$, with $r\ge0$, $0\le\theta\le2\pi$. The surfaces $r=a$ are cylinders about the $z$-axis.
- Spherical coordinates $(\rho,\varphi,\theta)$: $x=\rho\sin\varphi\cos\theta$, $y=\rho\sin\varphi\sin\theta$, $z=\rho\cos\varphi$, with $\rho\ge0$, $0\le\varphi\le\pi$, $0\le\theta\le2\pi$. Here $\rho=\sqrt{x^2+y^2+z^2}$ is the distance to the origin, $\varphi$ is the angle from the positive $z$-axis and $\theta$ is the polar angle of $(x,y)$; the distance to the $z$-axis is $r=\rho\sin\varphi$. The surface $\rho=a$ is a sphere, $\varphi=\alpha$ is a cone with vertex at the origin ($\varphi=\tfrac\pi2$ is the $xy$-plane), and $\theta=\beta$ is a half-plane.
Proof. For cylindrical coordinates the Jacobian matrix is the polar one with a $1$ added on the diagonal, so its determinant is $r$. For spherical coordinates,
$$\frac{\partial(x,y,z)}{\partial(\rho,\varphi,\theta)}=\det\begin{pmatrix}\sin\varphi\cos\theta&\rho\cos\varphi\cos\theta&-\rho\sin\varphi\sin\theta\\\sin\varphi\sin\theta&\rho\cos\varphi\sin\theta&\rho\sin\varphi\cos\theta\\\cos\varphi&-\rho\sin\varphi&0\end{pmatrix}.$$Expand along the third row. The cofactor of $\cos\varphi$ is $\rho^2\cos\varphi\sin\varphi\left(\cos^2\theta+\sin^2\theta\right)=\rho^2\sin\varphi\cos\varphi$; the cofactor of $-\rho\sin\varphi$ is $-\left(\rho\sin^2\varphi\cos^2\theta+\rho\sin^2\varphi\sin^2\theta\right)=-\rho\sin^2\varphi$. So the determinant is $\rho^2\sin\varphi\cos^2\varphi+\rho^2\sin^3\varphi=\rho^2\sin\varphi$, which is $\ge0$ because $0\le\varphi\le\pi$. The map is injective with nonzero Jacobian on $\rho\gt0$, $0\lt\varphi\lt\pi$, $0\lt\theta\lt2\pi$; the excluded faces have zero volume. Finally
$$\int_0^{2\pi}\!\!\int_0^\pi\!\!\int_0^a\rho^2\sin\varphi\,d\rho\,d\varphi\,d\theta=2\pi\cdot2\cdot\frac{a^3}3=\frac43\pi a^3.\ ∎$$Geometrically, the spherical "box" of sides $d\rho$ (radially), $\rho\,d\varphi$ (along a meridian) and $\rho\sin\varphi\,d\theta$ (along a circle of latitude of radius $r=\rho\sin\varphi$) has volume $\rho^2\sin\varphi\,d\rho\,d\varphi\,d\theta$.
Applications: Mass, Centroid, Volume and Surface Area
A thin plate (lamina) occupying a region $D$ with continuous density $\delta(x,y)\ge0$ (mass per unit area) has
- mass $M=\iint_D\delta\,dA$ and first moments $M_y=\iint_Dx\,\delta\,dA$, $M_x=\iint_Dy\,\delta\,dA$;
- centre of mass $(\bar x,\bar y)=\left(\dfrac{M_y}M,\dfrac{M_x}M\right)$, called the centroid of $D$ when $\delta$ is constant (then $\bar x=\frac1{\operatorname{Area}(D)}\iint_Dx\,dA$);
- moment of inertia about an axis $L$: $I_L=\iint_Dd_L^2\,\delta\,dA$, with $d_L$ the distance to $L$; e.g. $I_0=\iint_D(x^2+y^2)\,\delta\,dA$ about the $z$-axis.
For a solid $E$ with density $\delta(x,y,z)$ the same definitions hold with triple integrals: $M=\iiint_E\delta\,dV$, $\bar z=\frac1M\iiint_Ez\,\delta\,dV$, and so on. Symmetry helps: if $D$ is symmetric under $x\mapsto-x$ and $\delta(-x,y)=\delta(x,y)$, then $\bar x=0$.
Proof. In cylindrical coordinates $E=\{(r,\theta,z):(r,z)\in D,\ 0\le\theta\le2\pi\}$, so by Theorems 1.19 and 1.21
$$\operatorname{Vol}(E)=\int_0^{2\pi}\!\!\iint_Dr\,dr\,dz\,d\theta=2\pi\iint_Dx\,dA=2\pi\,\bar x\operatorname{Area}(D).\ ∎$$Two uses. The torus made by rotating a disc of radius $a$ whose centre is at distance $R\ge a$ from the axis has volume $2\pi R\cdot\pi a^2=2\pi^2Ra^2$. Conversely, rotating the half-disc $x^2+z^2\le a^2$, $x\ge0$, about the $z$-axis gives the ball, so $\frac43\pi a^3=2\pi\bar x\cdot\frac{\pi a^2}2$ and the centroid of a half-disc lies at distance $\bar x=\frac{4a}{3\pi}$ from its diameter.
For surface area, cut a parameter domain into small rectangles of sides $\Delta u,\Delta v$. A $C^1$ map $\mathbf{r}(u,v)$ sends each one to a small curved patch which, to first order, is the parallelogram spanned by $\mathbf{r}_u\Delta u$ and $\mathbf{r}_v\Delta v$ — of area $|\mathbf{r}_u\times\mathbf{r}_v|\,\Delta u\,\Delta v$. Adding up suggests:
- Graph. If $g$ is $C^1$ on a region $D$, the graph $z=g(x,y)$ over $D$ has area $\displaystyle A=\iint_D\sqrt{1+g_x^2+g_y^2}\,dA=\iint_D\sec\gamma\,dA$, where $\gamma$ is the angle between the upward normal and the $z$-axis.
- Level surface. If the surface $F(x,y,z)=0$ is, over $D$, the graph of a $C^1$ function and $F_z\ne0$ there, its area is $\displaystyle\iint_D\frac{|\nabla F|}{|F_z|}\,dA$.
- Surface of revolution. Rotating the graph of a $C^1$ function $y=f(x)\ge0$, $a\le x\le b$, about the $x$-axis gives a surface of area $\displaystyle2\pi\int_a^bf(x)\sqrt{1+f'(x)^2}\,dx$.
Proof. (1) Parametrise by $\mathbf{r}(x,y)=(x,y,g(x,y))$: $\mathbf{r}_x=(1,0,g_x)$, $\mathbf{r}_y=(0,1,g_y)$, and $\mathbf{r}_x\times\mathbf{r}_y=(-g_x,-g_y,1)$, of length $\sqrt{1+g_x^2+g_y^2}$. This normal has $z$-component $1$, so $\cos\gamma=1/\sqrt{1+g_x^2+g_y^2}$. (2) By the implicit function theorem (§1.1), $g_x=-F_x/F_z$ and $g_y=-F_y/F_z$, so $1+g_x^2+g_y^2=\left(F_x^2+F_y^2+F_z^2\right)/F_z^2$. (3) Parametrise by $\mathbf{r}(x,\theta)=\left(x,\ f(x)\cos\theta,\ f(x)\sin\theta\right)$, $a\le x\le b$, $0\le\theta\le2\pi$. Then $\mathbf{r}_x=(1,f'\cos\theta,f'\sin\theta)$, $\mathbf{r}_\theta=(0,-f\sin\theta,f\cos\theta)$, and $\mathbf{r}_x\times\mathbf{r}_\theta=\left(ff',-f\cos\theta,-f\sin\theta\right)$, of length $f\sqrt{1+f'^2}$. Integrate over $\theta$. ∎
Improper Multiple Integrals and the Gaussian Integral
The integral $\int_{-\infty}^\infty e^{-x^2}dx$, the normalising constant of the normal distribution, has no elementary antiderivative behind it; it is computed by squaring it into a double integral over the whole plane and switching to polar coordinates. To do that honestly we need improper double integrals.
Proof. Since $f\ge0$ and the $D_k$ increase, $a_k=\iint_{D_k}f$ is an increasing sequence (Theorem 1.17(2)), so it has a limit $L\in[0,\infty]$. Let $(E_m)$ be another exhaustion with limit $L'$. Each $E_m$ is a closed bounded subset of $D$, so $E_m\subseteq D_k$ for some $k$, and $\iint_{E_m}f\le\iint_{D_k}f\le L$. Letting $m\to\infty$, $L'\le L$; by symmetry $L\le L'$. For the extensions, apply the bounded versions on each $D_k$ (to $f^+$ and $f^-$ separately) and let $k\to\infty$. ∎
(a) $\displaystyle\int_{-\infty}^\infty e^{-x^2}dx=\sqrt\pi$. Consequently $\int_0^\infty e^{-x^2}dx=\frac{\sqrt\pi}2$, $\int_{-\infty}^\infty e^{-ax^2}dx=\sqrt{\pi/a}$ for $a\gt0$, $\int_{-\infty}^\infty e^{-(x-\mu)^2/(2\sigma^2)}dx=\sigma\sqrt{2\pi}$, and $\Gamma\left(\tfrac12\right)=\sqrt\pi$.
(b) If $A$ is a real symmetric positive definite $n\times n$ matrix, then $\displaystyle\int_{\mathbb{R}^n}e^{-x^{\mathsf T}Ax}\,dx=\frac{\pi^{n/2}}{\sqrt{\det A}}$.
Proof. (a) Let $I(R)=\int_{-R}^Re^{-x^2}dx$. On the square $S_R=[-R,R]^2$, Fubini gives
$$\iint_{S_R}e^{-x^2-y^2}dA=\int_{-R}^Re^{-x^2}dx\int_{-R}^Re^{-y^2}dy=I(R)^2 .$$On the disc $D_\rho$ of radius $\rho$, polar coordinates give
$$\iint_{D_\rho}e^{-x^2-y^2}dA=\int_0^{2\pi}\!\!\int_0^\rho e^{-r^2}r\,dr\,d\theta=\pi\left(1-e^{-\rho^2}\right).$$Since $D_R\subseteq S_R\subseteq D_{R\sqrt2}$ and the integrand is positive,
$$\pi\left(1-e^{-R^2}\right)\le I(R)^2\le\pi\left(1-e^{-2R^2}\right).$$Letting $R\to\infty$, $I(R)^2\to\pi$, and $I(R)\gt0$, so $\int_{-\infty}^\infty e^{-x^2}dx=\sqrt\pi$. The consequences follow by the substitutions $x\mapsto\sqrt a\,x$, $x\mapsto(x-\mu)/(\sigma\sqrt2)$, and $t=x^2$ in $\Gamma\left(\tfrac12\right)=\int_0^\infty t^{-1/2}e^{-t}dt=2\int_0^\infty e^{-x^2}dx$.
(b) By the spectral theorem (§3.3) $A=Q\Lambda Q^{\mathsf T}$ with $Q$ orthogonal and $\Lambda=\operatorname{diag}(\lambda_1,\dots,\lambda_n)$, all $\lambda_i\gt0$. The substitution $x=Qy$ has $|\det Q|=1$ and maps balls centred at $0$ onto themselves, so it respects the exhaustion by balls (Theorem 1.24), and $x^{\mathsf T}Ax=\sum\lambda_iy_i^2$. Exhausting now by cubes and using Fubini, the integral is $\prod_i\int_{-\infty}^\infty e^{-\lambda_it^2}dt=\prod_i\sqrt{\pi/\lambda_i}=\pi^{n/2}/\sqrt{\det A}$. ∎
- Iterated integrals can exist and differ. $\frac{x-y}{(x+y)^3}$ on $[0,1]^2$ gives $\frac12$ and $-\frac12$ (Example 2); $\frac{x^2-y^2}{(x^2+y^2)^2}$ gives $\frac\pi4$ and $-\frac\pi4$ on $[0,1]^2$ and $-\frac\pi4$, $\frac\pi4$ on $[1,\infty)^2$. Both functions fail to be absolutely integrable.
- Equal iterated integrals do not make $f$ integrable. Let $S=\{(\frac mp,\frac np):p\text{ prime},\ 0\lt m,n\lt p\}$ and $f=1$ on $S$, $0$ elsewhere in $[0,1]^2$. Every vertical and horizontal line meets $S$ in finitely many points, so both iterated integrals are $0$; but $S$ is dense in the square, so every upper sum is $1$ and every lower sum is $0$.
- A double integral can exist when an inner integral does not. $f=1$ on $\{\tfrac12\}\times(\mathbb{Q}\cap[0,1])$, $0$ elsewhere: $f$ vanishes off a segment (zero area), so $\iint f=0$, but $\int_0^1f(\tfrac12,y)\,dy$ does not exist. This is why Theorem 1.18(a) assumes the inner integrals exist.
- Forgetting the Jacobian. $\int_0^{2\pi}\int_0^1dr\,d\theta=2\pi$ is not the area of the unit disc; $\int_0^{2\pi}\int_0^1r\,dr\,d\theta=\pi$ is. In spherical coordinates the factor is $\rho^2\sin\varphi$, not $\rho^2$ or $\rho\sin\varphi$.
- $J$ versus $|J|$, and which Jacobian. For $u=x+y$, $v=x-2y$, $\frac{\partial(u,v)}{\partial(x,y)}=-3$; the area element is $dx\,dy=\frac13\,du\,dv$ — not $-\frac13$, and not $3$.
- Covering a region twice. $\int_0^{4\pi}\int_0^1r\,dr\,d\theta=2\pi$: the map is not injective, so Theorem 1.19 does not apply. Likewise, in spherical coordinates $\varphi$ runs over $[0,\pi]$; letting $\varphi$ run over $[0,2\pi]$ makes $\sin\varphi$ negative on half the range, and the "volume" of a ball comes out $0$.
- Outer limits must be constants. After reversing the order, the inner limits may depend only on the outer variable; a variable appearing in the outer limits signals a mistake.
- $\sqrt{\sin^2\theta}=|\sin\theta|$. In Viviani-type problems, $(4-4\cos^2\theta)^{3/2}=8|\sin\theta|^3$; dropping the absolute value changes the answer (Example 3(b)).
- Sign-changing improper integrals depend on how you exhaust. For $\sin(x^2+y^2)$ on $\mathbb{R}^2$, discs give $\pi\left(1-\cos R^2\right)$, which oscillates, while squares give $\to\pi$ (by the Fresnel integrals $\int_{\mathbb{R}}\sin x^2=\int_{\mathbb{R}}\cos x^2=\sqrt{\pi/2}$). The function is not absolutely integrable, and Theorem 1.24 needs $f\ge0$.
- Convergence depends on dimension. $\iint_{x^2+y^2\le1}(x^2+y^2)^{-\alpha}dA$ converges iff $\alpha\lt1$, while $\int_0^1x^{-2\alpha}dx$ converges iff $\alpha\lt\frac12$; in $\mathbb{R}^3$, $\iiint_{\rho\le1}\rho^{-s}dV$ converges iff $s\lt3$.
- Surface area is not the area of the shadow. The factor $\sec\gamma=\sqrt{1+g_x^2+g_y^2}\ge1$ is essential; and it blows up where the surface becomes vertical (the equator of a hemisphere), making the integral improper — usually still convergent.
- Symmetry needs both region and integrand. $\iint_Dx\,dA=0$ when $D$ is symmetric under $x\mapsto-x$; for the disc $x^2+y^2\le2x$ it equals $\pi$ (Example 3(a)).
Two Labs
Exercises
- 1 Compute $\iint_Rxe^{xy}\,dA$ over $R=[0,1]\times[0,2]$. Which order is easier?
- 2 Compute $\iint_D(x+y)\,dA$, where $D$ is the triangle with vertices $(0,0)$, $(2,0)$, $(2,1)$.
- 3 Use a double integral to find the area between $y=x^2$ and $y=x+2$.
- 4 Reverse the order: $\displaystyle\int_0^4\int_{\sqrt x}^2f(x,y)\,dy\,dx$.
- 5 Reverse the order: $\displaystyle\int_{-1}^2\int_{x^2}^{x+2}f(x,y)\,dy\,dx$.
- 6 Evaluate $\displaystyle\int_0^1\int_x^1x^2e^{y^4}\,dy\,dx$.
- 7 Evaluate $\displaystyle\int_0^\pi\int_x^\pi\frac{\sin y}{y}\,dy\,dx$.
- 8 Evaluate $\displaystyle\int_0^8\int_{\sqrt[3]y}^2\frac{dx}{1+x^4}\,dy$.
- 9 Find the average value of $f=x^2+y^2$ on the square $[0,2]\times[0,2]$, and a point where $f$ takes it (Theorem 1.17(4)).
- 1 Compute $\iint_{x^2+y^2\le4}e^{-(x^2+y^2)}\,dA$.
- 2 Compute $\iint_D(x^2+y^2)^{3/2}\,dA$ over the annulus $1\le x^2+y^2\le4$.
- 3 Find the area of one petal of the rose $r=\cos3\theta$.
- 4 Evaluate $\displaystyle\int_0^1\int_x^{\sqrt{2-x^2}}(x+y)\,dy\,dx$ by converting to polar coordinates.
- 5 Compute $\iint_De^{x+y}\,dA$ over the square $|x|+|y|\le1$ using $u=x+y$, $v=x-y$.
- 6 Find the area of the region in the first quadrant bounded by $xy=2$, $xy=4$, $y=x$ and $y=3x$.
- 7 Using $x=ar\cos\theta$, $y=br\sin\theta$, show that the ellipse $\frac{x^2}{a^2}+\frac{y^2}{b^2}\le1$ has area $\pi ab$.
- 8 Modelling. The population density of a circular city of radius $6$ km is $\delta(r)=3000\,e^{-r/2}$ people per $\text{km}^2$, at distance $r$ km from the centre. Estimate the population.
- 1 Compute $\iiint_Ez\,dV$ over the tetrahedron $x,y,z\ge0$, $x+y+z\le1$.
- 2 Find the volume of the solid inside the cylinder $x^2+y^2=1$, above $z=0$ and below the plane $z=2+x$.
- 3 Compute $\iiint_B(x^2+y^2+z^2)\,dV$ over the ball $B$ of radius $a$ centred at the origin.
- 4 Find the mass of the ball of radius $2$ centred at the origin with density $\delta=\dfrac1{\sqrt{x^2+y^2+z^2}}$.
- 5 Find the centroid of the solid hemisphere $x^2+y^2+z^2\le a^2$, $z\ge0$.
- 6 Find the centroid of the region under $y=\sin x$, $0\le x\le\pi$.
- 7 Find the area of the part of the plane $x+2y+2z=6$ in the first octant.
- 8 Find the area of the part of the cone $z=\sqrt{x^2+y^2}$ between the planes $z=1$ and $z=3$.
- 9 The torus is $\mathbf{r}(\theta,\varphi)=\big((R+a\cos\varphi)\cos\theta,\ (R+a\cos\varphi)\sin\theta,\ a\sin\varphi\big)$, $0\lt a\lt R$. Show that its surface area is $4\pi^2Ra$, and find its volume by Pappus's theorem.
- 1 Evaluate $\displaystyle\int_0^\infty\int_0^\infty(x^2+y^2)\,e^{-(x^2+y^2)}\,dx\,dy$.
- 2 For which $\alpha\gt0$ does $\iint_{0\lt x^2+y^2\le1}(x^2+y^2)^{-\alpha}\,dA$ converge? Find its value.
- 3 Show that $\int_0^\infty x^2e^{-x^2}dx=\frac{\sqrt\pi}4$, and deduce $\Gamma\left(\frac32\right)$.
- 4 Using $\Gamma(p)=2\int_0^\infty x^{2p-1}e^{-x^2}dx$ and polar coordinates in the first quadrant, show that $\Gamma(p)\Gamma(q)=\Gamma(p+q)\,B(p,q)$ for $p,q\gt0$, where $B(p,q)=2\int_0^{\pi/2}\cos^{2p-1}\theta\,\sin^{2q-1}\theta\,d\theta$. Deduce $B\left(\frac12,\frac12\right)=\pi$.
- 5 Prove or disprove: if both iterated integrals of a bounded $f$ on $[0,1]^2$ exist and are equal, then $f$ is integrable on $[0,1]^2$.
- 6 Prove or disprove: a bounded function on $[0,1]^2$ that is continuous except at finitely many points is integrable.
- 7 Compute both iterated integrals of $\dfrac{x^2-y^2}{(x^2+y^2)^2}$ over $[1,\infty)^2$. What do you conclude?
- 8 Prove: if $f\ge0$ is continuous on a rectangle $R$ and $\iint_Rf\,dA=0$, then $f=0$ on $R$.
- 9 A student computes the area of the unit disc as $\int_0^{4\pi}\int_0^1r\,dr\,d\theta=2\pi$. Which hypothesis of Theorem 1.19 is violated?
E1. GATE MA · MCQ The value of $\displaystyle\int_0^1\int_{\sqrt y}^1\sqrt{1+x^3}\,dx\,dy$ is (A) $\frac29\left(2\sqrt2-1\right)$; (B) $\frac23\left(2\sqrt2-1\right)$; (C) $\frac19\left(2\sqrt2-1\right)$; (D) $\frac29\left(\sqrt2-1\right)$.
E2. GATE MA · MCQ The volume of the solid bounded by the paraboloid $z=x^2+y^2$ and the plane $z=2y$ is (A) $\frac\pi4$; (B) $\frac\pi2$; (C) $\pi$; (D) $2\pi$.
E3. GATE MA · MSQ Which of the following is/are true? (One or more options may be correct.) (A) $\iint_{x^2+y^2\le1}xy^3e^{x^2+y^2}\,dA=0$. (B) $\iint_{[0,2]^2}\lfloor x+y\rfloor\,dA=4$, where $\lfloor\cdot\rfloor$ is the greatest integer function. (C) The volume of the unit ball equals $\int_0^{2\pi}\int_0^\pi\int_0^1\rho^2\sin\varphi\,d\rho\,d\varphi\,d\theta$. (D) $\iint_{[0,1]^2}\dfrac{dA}{x^2+y^2}$ is finite.
E4. GATE MA · MSQ Let $S$ be the part of the sphere $x^2+y^2+z^2=4$ with $1\le z\le2$. Which of the following is/are true? (One or more options may be correct.) (A) The area of $S$ is $4\pi$. (B) The area of $S$ equals the area of the part of the sphere with $-1\le z\le0$. (C) The part of the sphere with $0\le z\le1$ has smaller area than $S$. (D) The area of $S$ equals $\displaystyle\iint_{x^2+y^2\le3}\frac{2}{\sqrt{4-x^2-y^2}}\,dA$.
E5. GATE MA · NAT The area of the part of the surface $z=xy$ that lies inside the cylinder $x^2+y^2=1$ is ______ (correct to two decimal places).
E6. GATE MA · NAT Let $E$ be the solid bounded below by the cone $z=\sqrt{x^2+y^2}$ and above by the plane $z=2$. The value of $\iiint_E(x^2+y^2)\,dV$ is ______ (correct to two decimal places).
Suggested plan — 3 lectures × 60 min.
- Definition and Fubini. Definitions 1.10–1.11 (draw the cells and the extension by zero). Theorem 1.16 with the proofs of (1) and (2); state (3) and explain the idea. Theorem 1.18 with the sandwich proof. The strategy box, Example 1 on the board, then project Interactive 1 on all three problems — stop the strip in problem 2 at $y=1$ and ask what changes. Example 2 as a "spot the hypothesis" discussion.
- Changing variables. The parallelogram-area computation, Theorem 1.19 with Steps 1–3, and Interactive 2 (polar, then hyperbolic: shrink $h$ and read the ratio). Examples 3 and 4; let the class find the $|\sin\theta|$ error in Example 3(b) themselves. The coordinates strategy box.
- Three dimensions and applications. Theorem 1.20, Definition 1.13, Theorem 1.21 (derive $\rho^2\sin\varphi$ from the spherical box first, then the determinant). Example 5, Pappus with the half-disc centroid, Definition 1.15, Theorem 1.23 and Example 6. Theorem 1.25 with the square–disc sandwich, Example 7(b). GATE corner E2 and E5 live.
Discussion questions. (1) Why does Fubini's theorem need $f$ to be integrable on the rectangle, when each iterated integral only uses one-variable integrals? Use Example 2 and pitfall 2. (2) The polar map is not injective and has $J=0$ at $r=0$ — why is Theorem 1.19 still usable? What would go wrong if the bad set had positive area? (3) Archimedes' hat-box theorem says all zones of equal height have equal area. Explain this from $dS=a^2\sin\varphi\,d\varphi\,d\theta$ without integrating.
Misconceptions to watch for. Dropping the Jacobian (especially $r$ and $\rho^2\sin\varphi$); using $J$ instead of $|J|$ or the reciprocal Jacobian in the wrong direction; reversing limits algebraically without a sketch; outer limits that depend on a variable; $\varphi\in[0,2\pi]$ in spherical coordinates; $\sqrt{\sin^2\theta}=\sin\theta$; using the area of the shadow as surface area; treating a sign-changing improper integral as convergent because one exhaustion has a limit.
Suggested homework. I(1, 4, 5, 6, 8), II(3, 4, 6, 8), III(2, 5, 7, 9), IV(2, 4, 5, 7); GATE corner E1, E3, E6.
Research question. A radiologist estimates the volume of an organ or a tumour from a stack of scan slices. How accurate is the slice method (Cavalieri's principle with the trapezoidal rule), how does it compare with random-point (Monte Carlo) estimation, and how does each error shrink as more work is done?
Why it matters. Tumour volume measured from CT or MRI slices decides treatment response in oncology trials worldwide, and the same slice-and-add method is used by geologists for reservoir volumes and by hydrologists for lake storage. Theorem 1.20(b) says $V=\int A(z)\,dz$; in practice only finitely many slices are available, so $V$ is estimated by a sum, and the question is how fast that sum converges.
What to do.
- Model the organ as the ellipsoid $\frac{x^2}{36}+\frac{y^2}{16}+\frac{z^2}9\le1$ (lengths in cm). Using the substitution $x=6u$, $y=4v$, $z=3w$ and Theorem 1.19, show that its volume is $\frac43\pi\cdot6\cdot4\cdot3=96\pi\approx301.59\ \text{cm}^3$.
- Slice perpendicular to the long axis: show that the slice at $x$ has area $A(x)=12\pi\left(1-\frac{x^2}{36}\right)$. Apply the trapezoidal rule to $\int_{-6}^6A(x)\,dx$ with slice spacing $h=2,1,0.5$ cm and record the errors.
- Prove that for this quadratic $A$ the error is exactly $-\frac{\pi bc\,h^2}{3a}$ (with $a=6$, $b=4$, $c=3$), so it falls by a factor of $4$ each time $h$ is halved — and that Simpson's rule is exact.
- Estimate the volume by Monte Carlo: draw $N$ uniform random points in the box $[-6,6]\times[-4,4]\times[-3,3]$ (volume $576\ \text{cm}^3$) and multiply $576$ by the fraction inside. Use $N=10^3,10^4,10^5$; compare the errors with the predicted standard error $576\sqrt{p(1-p)/N}$, $p=\pi/6$.
Tools: Python with numpy (a few lines each), or a spreadsheet for the slice sums.
Hand in: a table of errors for both methods; a log–log plot of error against work (number of slices, number of points) with fitted slopes; a paragraph on which method you would trust for a real, irregular organ, and why.
Going further. Repeat in higher dimensions: the volume of the unit ball in $\mathbb{R}^d$ is $\pi^{d/2}/\Gamma\left(\frac d2+1\right)$ (derive it from Theorem 1.25). Grid methods need $n^d$ points, while the Monte Carlo error is $O(N^{-1/2})$ in every dimension — the reason Monte Carlo dominates in finance and physics. Also find the volume common to three perpendicular cylinders of radius $1$ (exact value $8\left(2-\sqrt2\right)\approx4.686$).
🧪 Self-check quiz
Q1. With the order reversed, $\displaystyle\int_0^1\int_0^xf(x,y)\,dy\,dx$ equals
Q2. In spherical coordinates ($\varphi$ measured from the positive $z$-axis) the volume element is
Q3. $\displaystyle\int_0^\infty e^{-x^2}dx=$
Q4. Under $u=x+y$, $v=x-y$ a region $D$ of the $xy$-plane corresponds to a region of area $6$ in the $uv$-plane. The area of $D$ is
Q5. A disc of radius $1$ whose centre is at distance $3$ from an axis in its plane is rotated about that axis. The volume of the torus is