Double integrals as volume
Water standing in a flooded field is deeper in some places than in others. How do you add up a quantity that changes from point to point over a region of the plane, and get the exact total rather than an estimate?
Where you are 6 in Multiple integrals
Why this exists
Amara and Samir flood a rice paddy before planting, a level field 60 meters long and 40 meters wide. The water is not equally deep everywhere, and they want its volume. With a constant depth the answer is base times height. Volumes with known cross sections handles solids whose slices have a known area, and The definite integral as a limit of Riemann sums found an area under a curve as a limit of Riemann sums. Here the height varies in two directions at once.
The double integral is the same idea one dimension up: cut the region into small rectangles, treat the depth as constant on each, add the volumes of the thin boxes, and take the limit. The same sums give a mass from a density and an average over a region.
Definition
Let R = [a, b] \times [c, d] = \{(x, y) : a \le x \le b,\ c \le y \le d\} and let f : R \to \mathbb{R} be a bounded function of two variables. A partition P of R is a choice of points
The lines x = x_i and y = y_j cut R into mn subrectangles R_{ij} = [x_{i-1}, x_i] \times [y_{j-1}, y_j] of area \Delta A_{ij} = \Delta x_i\,\Delta y_j, where \Delta x_i = x_i - x_{i-1} and \Delta y_j = y_j - y_{j-1}. The mesh \lVert P \rVert is the largest of all the \Delta x_i and \Delta y_j. With a sample point (x_{ij}^*, y_{ij}^*) chosen in each R_{ij}, the number
is a double Riemann sum for f over R. The function f is integrable over R, with double integral I, if for every \varepsilon > 0 there is a \delta > 0 such that |S(P) - I| < \varepsilon for every partition with \lVert P \rVert < \delta and every choice of sample points. We write
With equal widths \displaystyle \Delta x = \frac{b - a}{m} and \displaystyle \Delta y = \frac{d - c}{n}, every subrectangle has area \Delta A = \Delta x\,\Delta y, and the mesh tends to 0 exactly when m and n both grow without bound. Taking the center of each R_{ij} as its sample point gives the midpoint sum.
Other regions. Let D be a bounded set in the plane, f a function on D, and R a rectangle containing D. Extend f by zero,
and define \iint_D f\,dA = \iint_R f^D\,dA when the right side exists; terms outside D are 0, so the rectangle chosen does not matter. The area of D is A(D) = \iint_D 1\,dA, and when A(D) > 0 the average value of f over D is
Intuition
A term f(x_{ij}^*, y_{ij}^*)\,\Delta A_{ij} is the volume of a thin box: base R_{ij}, height the value of f at the sample point. In Figure 1 the box touches the surface at its front right corner and pokes above it elsewhere, because this f rises toward the front and the right. As the mesh shrinks, the mismatch on each tiny base shrinks faster than the number of boxes grows.
Where f is negative the box hangs below the xy-plane and its term is negative, so the double integral is a signed volume. If f is mass per unit area, the integral is the total mass. And if the paddy’s water settled to a flat surface over the same base, its depth would be the average depth.
Results
Call a bounded region D admissible when its boundary is made of finitely many graphs y = \phi(x) or x = \psi(y) of continuous functions on closed intervals. Rectangles, triangles and disks are admissible.
Result 1 (integrability, stated without proof). A function continuous on a closed rectangle R is integrable over R. More generally, if D is admissible and f is bounded on D and continuous except on finitely many graphs of continuous functions, then f is integrable over D, and A(D) exists. The proofs rest on uniform continuity and belong to Real Analysis.
Result 2 (volume). If f \ge 0 is continuous on an admissible region D, the solid \{(x, y, z) : (x, y) \in D,\ 0 \le z \le f(x, y)\} has volume \iint_D f\,dA. For f of either sign, \iint_D f\,dA is the volume above the xy-plane minus the volume below it.
Result 3 (properties). Let f and g be integrable over an admissible region D and \alpha, \beta constants.
- Linearity: \iint_D (\alpha f + \beta g)\,dA = \alpha \iint_D f\,dA + \beta \iint_D g\,dA.
- Monotonicity: if f \le g on D, then \iint_D f\,dA \le \iint_D g\,dA.
- Bounds: if m \le f \le M on D, then m\,A(D) \le \iint_D f\,dA \le M\,A(D).
- Additivity: if D = D_1 \cup D_2 with D_1, D_2 admissible and sharing only boundary points, then \iint_D f\,dA = \iint_{D_1} f\,dA + \iint_{D_2} f\,dA.
- For continuous f: \left\lvert \iint_D f\,dA \right\rvert \le \iint_D \lvert f \rvert\,dA.
Result 4 (separated variables). If g is continuous on [a, b] and h is continuous on [c, d], then
In particular \iint_R g(x)\,dA = (d - c)\int_a^b g(x)\,dx.
Result 5 (mean value theorem for double integrals). Let D be a closed admissible region with A(D) > 0 in which any two points are joined by a continuous path lying in D, and let f be continuous on D. Then f(x_0, y_0) = f_{\text{avg}} at some point (x_0, y_0) of D.
Why it is true
Result 3 on a rectangle. For one partition and one choice of sample points, a Riemann sum of \alpha f + \beta g is \alpha times that of f plus \beta times that of g, since finite sums may be regrouped and a common factor pulled out. If f \le g, each term for f is at most the matching term for g. Limits keep both facts. For a constant k, every Riemann sum is k \sum \Delta A_{ij} = k\,A(R), so \iint_R k\,dA = k\,A(R), and the bounds follow by comparing f with m and with M. For additivity, cut R by the line x = e into R_1 and R_2; f is integrable on all three (Result 1 for continuous f, Real Analysis in general). Partitions P_1 of R_1 and P_2 of R_2 with the same y-points together form a partition P of R whose x-points include e, with S(P) = S(P_1) + S(P_2) and \lVert P \rVert the larger of the two meshes. Letting both meshes go to 0 gives the sum rule. On an admissible D the same arguments apply to f^D; that the shared boundary of D_1 and D_2 adds nothing is part of Result 1. The absolute value inequality is Exercise 16.
Result 2. Take D = R, a rectangle. The extreme value theorem for a continuous function on a closed bounded set (the two-variable form of The extreme value theorem, proved in Real Analysis) gives a least value m_{ij} and a greatest value M_{ij} of f on each R_{ij}. The solid over R_{ij} contains the box of height m_{ij} and lies inside the box of height M_{ij}. Volume adds over pieces and a box has volume base times height, so
Both sides are Riemann sums, because m_{ij} and M_{ij} are values of f at points of R_{ij}, so both tend to \iint_R f\,dA as the mesh shrinks and squeeze V to that value. On an admissible D the squeeze runs on f^D with its infimum and supremum on each R_{ij}; that these sums still converge is part of Result 1. For f of either sign, write f = f^+ - f^- with f^+ = \max(f, 0) and f^- = \max(-f, 0), both continuous and \ge 0, and use linearity.
Result 4. The product g(x)h(y) is continuous on R, so it is integrable. Choose sample points on a grid, (x_i^*, y_j^*) with x_i^* in [x_{i-1}, x_i] and y_j^* in [y_{j-1}, y_j]. Then the double sum factors:
As the mesh tends to 0 the left side tends to the double integral, and the factors tend to the two single integrals, so the limit of the product is the product of the integrals. Taking h = 1 gives the special case.
Result 5. By the extreme value theorem f has a least value f(\mathbf{p}) = m and a greatest value f(\mathbf{q}) = M on D. Dividing the bounds of Result 3 by A(D) gives m \le f_{\text{avg}} \le M. Let \mathbf{r} : [0, 1] \to D be a continuous path with \mathbf{r}(0) = \mathbf{p} and \mathbf{r}(1) = \mathbf{q}. Then \varphi(t) = f(\mathbf{r}(t)) is continuous on [0, 1] with \varphi(0) = m and \varphi(1) = M, and the intermediate value theorem gives t_0 with \varphi(t_0) = f_{\text{avg}}. Take (x_0, y_0) = \mathbf{r}(t_0).
Worked examples
Estimating the water in the paddy
Amara models the depth of the water at the point x meters east and y meters north of the paddy’s southwest corner as
meters, on R = [0, 60] \times [0, 40]. Estimate the volume of water with the midpoint sum for m = 3 and n = 2.
SolutionThe widths are \displaystyle \Delta x = \frac{60}{3} = 20 and \displaystyle \Delta y = \frac{40}{2} = 20, so every subrectangle has area \Delta A = 400 square meters. The centers have x in \{10, 30, 50\} and y in \{10, 30\} (Figure 2).
Rather than evaluate d six times, add the depths part by part; a finite sum may be reordered and regrouped. The constant part contributes 6(0.04) = 0.24. The part \displaystyle \frac{x^2}{60{,}000} depends only on x, and each center x occurs in two rows: \displaystyle \frac{2(100 + 900 + 2500)}{60{,}000} = \frac{7}{60}. The part \displaystyle \frac{y}{800} depends only on y, and each center y occurs three times: \displaystyle \frac{3(10 + 30)}{800} = 0.15. The six depths add to \displaystyle 0.24 + \frac{7}{60} + 0.15 = \frac{38}{75} meters, so
A double integral from the definition
Compute \displaystyle \iint_R \left(x + \frac{y^2}{4}\right) dA over R = [0, 2] \times [0, 2] from the definition, with the upper right corner of each subrectangle as its sample point.
SolutionThe integrand is continuous, so it is integrable (Result 1) and the limit may be taken along any choice of sample points. With \displaystyle \Delta x = \frac{2}{m} and \displaystyle \Delta y = \frac{2}{n} the corners are \displaystyle x_i = \frac{2i}{m}, \displaystyle y_j = \frac{2j}{n}, where \displaystyle f(x_i, y_j) = \frac{2i}{m} + \frac{j^2}{n^2}, and \displaystyle \Delta A = \frac{4}{mn}. The first part does not involve j, so the sum over j multiplies it by n; the second does not involve i, so the sum over i multiplies it by m. Using \displaystyle \sum_{i=1}^{m} i = \frac{m(m+1)}{2} and \displaystyle \sum_{j=1}^{n} j^2 = \frac{n(n+1)(2n+1)}{6} from The definite integral as a limit of Riemann sums,
Each piece depends on one index only, so as m and n grow independently the limit is the sum of the two one-variable limits, \displaystyle 4 + \frac{4}{3}:
This is the volume under the surface of Figure 1. Check with Result 4: \iint_R x\,dA = 2\int_0^2 x\,dx = 4 and \displaystyle \iint_R \frac{y^2}{4}\,dA = 2 \cdot \frac{1}{4} \cdot \frac{8}{3} = \frac{4}{3}.
The paddy’s exact volume and average depth
Find the exact volume of water in Example 1, the average depth, and the points of the paddy where the depth equals the average.
SolutionSplit d by linearity (Result 3) and integrate each part with Result 4, using A(R) = 2400:
The volume is 96 + 48 + 60 = 204 cubic meters. The midpoint sum of Example 1 was \displaystyle \frac{4}{3} low; all of that error comes from the x^2 part, since midpoint sums are exact for linear functions (Exercise 15). The average depth is
The rectangle is closed and any two of its points are joined by a segment inside it, so Result 5 promises a point where d = 0.085. Multiplying \displaystyle 0.04 + \frac{x^2}{60{,}000} + \frac{y}{800} = 0.085 by 800 gives \displaystyle 32 + \frac{x^2}{75} + y = 68, so
This stays in R while y \ge 0, that is for \displaystyle 0 \le x \le 30\sqrt{3} \approx 52.0, the arc drawn in Figure 2.
Volume over a triangle
Let D be the triangle with vertices (0, 0), (4, 0) and (0, 3). Find \iint_D (12 - 3x - 4y)\,dA and the average value of 12 - 3x - 4y over D.
SolutionThe slanted side lies on 3x + 4y = 12, and D is the set of points with x \ge 0, y \ge 0, 3x + 4y \le 12. So f = 12 - 3x - 4y \ge 0 on D, with f = 0 on the slanted side. D is admissible and f is continuous, so Result 2 makes the integral the volume under the plane z = 12 - 3x - 4y over D. That solid is the tetrahedron with vertices (0, 0, 0), (4, 0, 0), (0, 3, 0) and (0, 0, 12), a pyramid with base area \displaystyle A(D) = \frac12 \cdot 4 \cdot 3 = 6 and height 12:
The average value is \displaystyle \frac{24}{6} = 4.
Odd symmetry cancels
Prove that if f is continuous on R = [-a, a] \times [c, d] and f(-x, y) = -f(x, y) on R, then \iint_R f\,dA = 0. Then evaluate \iint_R (x^3 e^{y} + y)\,dA over R = [-1, 1] \times [0, 2].
SolutionTake equal widths with m even. The x-points are then symmetric, x_{m-i} = -x_i, so the reflection (x, y) \mapsto (-x, y) carries R_{ij} onto R_{m+1-i,\,j}, a subrectangle of the same area. For \displaystyle i \le \frac{m}{2} choose any sample point (s, t) in R_{ij} and use (-s, t) in its mirror image. The two terms are f(s, t)\,\Delta A + f(-s, t)\,\Delta A = 0, so every such Riemann sum is 0. Since f is continuous it is integrable (Result 1), so these sums tend to \iint_R f\,dA as m (through even values) and n grow. A limit of zeros is 0. \blacksquare
For the integral, x^3 e^{y} is continuous and changes sign with x, so its integral is 0. By Result 4, \iint_R y\,dA = \left(\int_{-1}^{1} dx\right)\left(\int_0^2 y\,dy\right) = 2 \cdot 2 = 4. By linearity,
Trapping an integral between bounds
Bound I = \iint_R e^{-xy}\,dA over R = [0, 1] \times [0, 1], first in one step and then by splitting R into four squares.
SolutionOn R, 0 \le xy \le 1, so e^{-1} \le e^{-xy} \le 1. With A(R) = 1 the bounds of Result 3 give 0.368 \le I \le 1.
For a sharper result, cut R into four squares of side \displaystyle \frac12 and area \displaystyle \frac14 and add (additivity). For fixed y \ge 0, e^{-xy} does not increase as x grows, and likewise in y. So on a square \displaystyle [p, p + \frac12] \times [q, q + \frac12], moving from (x, y) to (p, y) and then to (p, q) never lowers f, and moving to the opposite corner never raises it: \displaystyle f(p + \frac12, q + \frac12) \le f(x, y) \le f(p, q). The lower left corners give 1, 1, 1, e^{-1/4} and the upper right corners give e^{-1/4}, e^{-1/2}, e^{-1/2}, e^{-1}. Applying the bounds on each square and adding,
In fact I \approx 0.7966.
Boundaries
Boundedness. The definition needs f bounded. For \displaystyle f(x, y) = \frac{1}{x^2 + y^2} on [0, 1] \times [0, 1], a sample point close to the origin in the corner subrectangle makes a Riemann sum as large as we like, so f is not integrable in this sense. Such integrals are improper, defined as limits of ordinary ones.
Too many discontinuities. A bounded function can still fail: the function equal to 1 where both coordinates are rational and 0 elsewhere has Riemann sums 0 and 1 for every partition (Exercise 18). Changing a continuous function at one point, by contrast, changes nothing (Exercise 17).
Sets without area. The points of the unit square with both coordinates rational form a bounded set with no area in this sense, since \iint_D 1\,dA is the integral of Exercise 18. The general theory is Real Analysis.
Signed volume. \iint_D f\,dA = 0 does not mean f = 0: positive and negative parts can cancel (Example 5, Exercise 7).
The mean value theorem needs a connected region. On two separate squares a continuous f can miss its own average (Exercise 19).
Where this goes next
Iterated integrals and Fubini's theorem proves Fubini’s theorem: a double integral over a rectangle, or a region between two graphs, is two single integrals one inside the other; Result 4 is the case where the integrand separates. Double integrals in polar coordinates handles disks, Change of variables and the Jacobian other coordinates, and Triple integrals repeats the construction in space. In Joint distributions a probability is a double integral of a joint density, and Green's theorem trades a double integral for a line integral around the boundary.
Exercises
Riemann sums
- Find the midpoint sum for \iint_R x y^2\,dA over R = [0, 4] \times [0, 2] with m = n = 2, and compare it with the exact value from Result 4.
- Find the Riemann sum for f(x, y) = 10 - x^2 - y^2 over [0, 2] \times [0, 2] with m = n = 2 and the lower left corner of each square as its sample point. Is it above or below the integral? Find the integral.
- A second paddy, 30 by 20 meters, is divided into six 10 by 10 meter plots. The depths in meters at the plot centers are 0.07, 0.08, 0.10 in the south row and 0.06, 0.09, 0.12 in the north row. Estimate the volume of water and the average depth.
Exact values
- Compute \iint_R xy\,dA over R = [0, 2] \times [0, 4] from the definition, with upper right corners as sample points.
- Use geometry to find \displaystyle \iint_R \sqrt{16 - x^2}\,dA over R = [-4, 4] \times [0, 3].
- Use geometry to find \displaystyle \iint_D \sqrt{9 - x^2 - y^2}\,dA, where D is the disk x^2 + y^2 \le 9.
- Find \iint_R (3 - x)\,dA over R = [0, 6] \times [0, 2] and explain the value with volumes.
- On a rectangle R of area 6, \iint_R f\,dA = 5 and \iint_R g\,dA = -2. Find \iint_R (3f - 4g + 1)\,dA.
- Evaluate \iint_R e^{y} \sin x\,dA over R = [0, \pi] \times [0, 1].
- Find the average value of f(x, y) = xy over R = [0, 4] \times [1, 3] and describe the points of R where f takes it.
Bounds and symmetry
- Give lower and upper bounds for \displaystyle \iint_R \sqrt{1 + x^3 + y^3}\,dA over R = [0, 1] \times [0, 1] from Result 3 alone.
- Sharpen the bounds in Exercise 11 by splitting R into four squares of side \displaystyle \frac12.
- Evaluate \iint_R (x y^4 + y^3 + 2)\,dA over R = [-1, 1] \times [-1, 1] without computing an antiderivative.
Proofs
- Let f be continuous with f \ge 0 on a rectangle R, and f(\mathbf{p}) > 0 at some point \mathbf{p} of R. Prove that \iint_R f\,dA > 0.
- Prove that for f(x, y) = \alpha + \beta x + \gamma y the midpoint sum over any rectangle, with any m and n, equals \iint_R f\,dA.
- Prove that \left\lvert \iint_R f\,dA \right\rvert \le \iint_R \lvert f \rvert\,dA for f continuous on a rectangle R.
- Let f be continuous on R and let g equal f except at one point \mathbf{p}, where g(\mathbf{p}) = k. Prove that g is integrable over R with the same integral as f.
- Prove that f(x, y) = 1 when x and y are both rational, f(x, y) = 0 otherwise, is not integrable over [0, 1] \times [0, 1].
- Let D = [0, 1] \times [0, 1] \cup [2, 3] \times [0, 1], with f = 0 on the first square and f = 1 on the second. Show that f is continuous on D but never equals its average value, and say which hypothesis of Result 5 fails.
Answers
- 20; exact \displaystyle \frac{64}{3}. The centers are x \in \{1, 3\}, \displaystyle y \in \{\frac12, \frac32\} and \Delta A = 2, so the sum is \displaystyle 2(1 + 3)(\frac14 + \frac94) = 20. Result 4 gives \displaystyle \left(\int_0^4 x\,dx\right)\left(\int_0^2 y^2\,dy\right) = 8 \cdot \frac83 = \frac{64}{3} \approx 21.3.
- 36, above the integral; \displaystyle \iint_R f\,dA = \frac{88}{3}. The corners (0, 0), (1, 0), (0, 1), (1, 1) give 10 + 9 + 9 + 8 = 36 times \Delta A = 1. On R, f decreases in each variable, so each lower left corner gives the largest value on its square. Exactly, \displaystyle 40 - 2 \cdot \frac83 - 2 \cdot \frac83 = \frac{88}{3} \approx 29.3 by Result 4.
- About 52 cubic meters, average about 0.087 meters. The depths add to 0.52 and each plot has area 100, so V \approx 52; dividing by the area 600 gives \displaystyle \frac{13}{150} \approx 0.087.
- 16. With \displaystyle x_i = \frac{2i}{m}, \displaystyle y_j = \frac{4j}{n} and \displaystyle \Delta A = \frac{8}{mn}, the sum factors as \displaystyle \frac{64}{m^2 n^2} times \displaystyle \frac{m(m+1)}{2} \cdot \frac{n(n+1)}{2}, that is \displaystyle \frac{16(m + 1)(n + 1)}{mn}, which tends to 16.
- 24\pi. The solid is half of a cylinder of radius 4 lying along the y-axis, length 3: \displaystyle \frac12 \pi \cdot 16 \cdot 3.
- 18\pi. The graph is the upper hemisphere of radius 3, with volume \displaystyle \frac12 \cdot \frac43 \pi \cdot 27.
- 0. Over [0, 3] \times [0, 2] the solid above the plane is a wedge with triangular cross section of area \displaystyle \frac12 \cdot 3 \cdot 3 and length 2, volume 9; over [3, 6] \times [0, 2] an identical wedge hangs below. Signed volume 9 - 9 = 0.
- 29. Linearity gives 3 \cdot 5 - 4(-2) + 1 \cdot 6 = 15 + 8 + 6.
- 2e - 2. By Result 4, \left(\int_0^{\pi} \sin x\,dx\right)\left(\int_0^1 e^{y}\,dy\right) = 2(e - 1).
- 4, on the arc \displaystyle y = \frac{4}{x} for \displaystyle \frac43 \le x \le 4. The integral is \left(\int_0^4 x\,dx\right)\left(\int_1^3 y\,dy\right) = 8 \cdot 4 = 32 and A(R) = 8. The curve xy = 4 meets R where \displaystyle 1 \le \frac{4}{x} \le 3.
- \displaystyle 1 \le I \le \sqrt{3}. On R, 1 \le 1 + x^3 + y^3 \le 3, and A(R) = 1.
- 1.060 \le I \le 1.441. The integrand increases in each variable, so on each square its least value is at the lower left corner and its greatest at the upper right. Lower: \displaystyle \frac14\left(1 + 2\sqrt{\frac98} + \sqrt{\frac54}\right) \approx 1.060. Upper: \displaystyle \frac14\left(\sqrt{\frac54} + 2\sqrt{\frac{17}{8}} + \sqrt{3}\right) \approx 1.441.
- 8. By Example 5, x y^4 (odd in x) integrates to 0, and so does y^3 by the same argument in y; \iint_R 2\,dA = 2 \cdot 4.
- By continuity at \mathbf{p} with \displaystyle \varepsilon = \frac12 f(\mathbf{p}) there is r > 0 with \displaystyle f > \frac12 f(\mathbf{p}) at every point of R within distance r of \mathbf{p}. Choose a closed rectangle Q \subseteq R of positive area, containing \mathbf{p} and within distance r of it. Extending the sides of Q cuts R into at most nine rectangles. By additivity \iint_R f\,dA is the sum of the integrals over them; by the bounds of Result 3 the one over Q is at least \displaystyle \frac12 f(\mathbf{p})\,A(Q) > 0 and the others are at least 0. So \iint_R f\,dA > 0.
- Let (u, v) be the center of R_{ij} = [u - h, u + h] \times [v - k, v + k]. Then f(x, y) = f(u, v) + \beta(x - u) + \gamma(y - v). By Result 4, \iint_{R_{ij}} (x - u)\,dA = 2k\int_{u-h}^{u+h} (x - u)\,dx = 0, and likewise for y - v. So by linearity \iint_{R_{ij}} f\,dA = f(u, v)\,\Delta A_{ij}, exactly the midpoint term. Adding over all i, j (additivity) gives \iint_R f\,dA equal to the midpoint sum.
- \lvert f \rvert is continuous, as a composition of continuous functions, so it is integrable. From -\lvert f \rvert \le f \le \lvert f \rvert, monotonicity and linearity give -\iint_R \lvert f \rvert\,dA \le \iint_R f\,dA \le \iint_R \lvert f \rvert\,dA, which is the claim.
- Let I = \iint_R f\,dA and c = \lvert k - f(\mathbf{p}) \rvert. For one partition and one choice of sample points, the Riemann sums of g and f differ only in terms whose sample point is \mathbf{p}. The point \mathbf{p} lies in at most four subrectangles, each of area at most \lVert P \rVert^2, so \lvert S_g(P) - S_f(P) \rvert \le 4c\lVert P \rVert^2. Given \varepsilon > 0, take \delta so that \displaystyle \lvert S_f(P) - I \rvert < \frac{\varepsilon}{2} when \lVert P \rVert < \delta, and also \displaystyle 4c\delta^2 < \frac{\varepsilon}{2}. Then \lvert S_g(P) - I \rvert < \varepsilon whenever \lVert P \rVert < \delta, so g is integrable with integral I.
- Every subrectangle has positive width and height, so it contains a point with both coordinates rational and a point with x irrational. For every partition, sample points of the first kind give S(P) = \sum \Delta A_{ij} = 1 and points of the second kind give S(P) = 0. No number I is within \displaystyle \frac12 of both 0 and 1, so no \delta works for \displaystyle \varepsilon = \frac12, and f is not integrable.
- Each point of D has a disk around it whose points in D all lie in the same square, where f is constant, so f is continuous on D. By additivity \iint_D f\,dA = 0 + 1 = 1 and A(D) = 2, so \displaystyle f_{\text{avg}} = \frac12, a value f never takes. The failing hypothesis is the path condition: a continuous path from the first square to the second would have an x-coordinate taking the value \displaystyle \frac32 by the intermediate value theorem, at a point outside D.