Solutions To Exercises: J3) - J3) J3 - J3

Solutions to Exercises
The solutions here are intended to be instructive. They are therefore sprinkled
with comments, alternative approaches, and questions. Except for leaving such
questions to be pondered by the reader, these answers are complete.
Many of them were worked out by Sera Cremonini, summa cum laude gradu-
ate of City College, whose knowledge of mathematics is all the more wonderful
because it was not her major.
Chapter 1
Section 1.1
1. (a) The graph is a rectangular hyperbola--one whose asymptotes Y = ±x
are at right angles-crossing the x-axis at (±1, 0). See the figure for
this solution.
(b) No. (2, J3) and (2, -J3) both satisfy the equation, but J3 and
-J3 cannot both be g(2). (Note that an example is essential, and that
many exist.)
2. This is familiar from the definition of function. Keep Exercise 1 in mind.
(a) It must pass the "vertical-line test": No vertical line can have as many
as two points from the set. Observe that the graph in (1) fails.
(b) For any value of x, if (x, Yl) and (x, Y2) are both in the set, then
Yl = Y2·
156 Solutions to Exercises
y
z
---+-"111--1-+...... x
Exercise la Exercises 3a, c
3. (a) Algebraic experience tells us that the graph is a plane. Without worrying
too much about proof (one is suggested by Exercise 8 in Section 1.2),
we can give a good idea of the graph. First, the intercepts x = 4, y = 2,
1
and z = are evident. Second, the lines joining the intercepts are on
the graph. For example, points with x + 2y = 4 in the z = 0 plane
necessarily satisfy the equation. Hence the edges of the triangle in the
figure are in the solution set. The plane of that triangle is the graph of
the equation.
(b) Since (a) is a plane, this is the intersection of two planes. The planes are
evidently unequal, and setting Z = 0 gives xy-plane lines of unequal
slope. Therefore, the planes have nontrivial intersection; that is, they
intersect in a line.
(c) The system's augmented matrix reduces to
1 0 -1 -2
o 2 3
giving z = arbitrary t, Y = 3 - 2t, x = -2 + t. Those equations are

the parametric description of a line, shown dashed in the figure.
4. If a, b, and c are all 0, then R2 is the solution set. If a = b = 0 and c i= 0,

then the set is empty. Anything else-if either a or b is nonzero-allows
us to solve for either x or y and recognize the equation of a line. To put it
differently, if a or b is nonzero, then the set is a one-parameter family that
we can recognize as a Cartesian line.
5. (a) Same idea as in (4): a = b = c = d = 0 => all solutions; a =

b = c = 0 i= d => none; otherwise, solution has two free variables,
parametrizes a plane.
(b) If a through h are all zero, then R3 is the set. If either a, b, and care
zero with d i= 0, or if e, j, g are zero with h i= 0, then one of the
equations has no solutions; the system cannot have any. Otherwise,
three possibilities remain. Case 1: One equation is a multiple of the
Solutions to Exercises 157
other; either they both represent the same plane, and that plane is the
set, or the "one" says Ox + Oy + Oz = 0, and the plane of the "other" is
the solution. Case 2: One of (a, b, c) and (e, j, g) is a multiple of the
other, but d and h are not in the same ratio; in this case, the planes are
parallel, and the system has no solution. Case 3: Neither of (a, b, c)
and (e, j, g) is a multiple of the other, in which case ("one-parameter
solution") the planes intersect in a line, irrespective of d and h.
Section 1.2
1. (a) Slope 4 tells us that the line is in the direction of the vector c := (1, 4),
with b := (3,5).
(b) We calculate slope = ~. Then c has to be (2,1) or some multiple. We
can take b := (-2, -1),thencheckthat(4,2) = b+3c.Alternatively,
set b2 := (4,2); then (-2, -1) = b2 + (-3)c.
(c) The slope points us along c := (1, m), and the intercept locates b :=
(0, b).
2. (a) Point-slope form, where the slope has to be -1: y - 1 = -2(x + 2).
Alternatively, parametrically: x = -2 + 2t, y = 1 - 4t.
(b) y - 1 = (-~) (x + 2), which goes through the origin.
3. (a) Yes and yes. The line joining (1, 2) and (3, 4) consists of vectors of the
form (1- a)(l, 2) +a(3, 4). Is (5, 6) one such? Solve (1- a)(1, 2) +
a(3,4) = (5,6): a = 2, answering the first question. Then we need
not answer the second separately: Since (5, 6) = -1(1,2) + 2(3, 4),
we have (1, 2) = 2(3,4)+( -1)(5,6), putting (1,2) on the line joining
the other two.
(b) The argument is within (a): c is on the line joining a and b <:} c =
(1 - a)a + ab for some a <:} a = -a/(l - a)b + 1/(1 - a)b. (How
come a -:p I?) The last says that a is on the line joining band c, because
the coefficients add up to 1.
4. For comparison, calculate b - a and c - a for the vectors in Exercises 3a,

5, and 6.
:::} Assume that a, b, and c are on the line d + ((e)); that is, a = d + ae,
b = d + {3e, c = d + yeo Then b - a and c - a are mUltiples of e; they are
dependent.
~ Assume b - a = ae and c - a = {3e. Then a = a + Oe, b = a + ae, and
c = a + (3e are all on the line a + ((e)).
5. Yes and no. They have to be dependent; they are three vectors in R2. For
collinearity, Exercise 4 tells us to look at b - a = (2,2) and c - a = (5,3),
which are independent.
6. Yes and yes. Tum the vectors into rows in a matrix and do row reduction: first
row times -6 added to second, first row times -11 added to third, resulting
second row times - 2 added to third, third row is now all zeros. This means
thate-lla-2(b-6a) = la-2b+ Ie = 0, and the vectors are dependent.
Rewrite the last as e = 2b - la, and we see that e is on the line joining the
other two, because the coefficients sum to 1.
7. Algebraically: If three or more vectors are on the line d + ((e)), then they are
linear combinations of d and e; three combinations of two vectors are nec-
essarily dependent. Geometrically: If vectors lie along a line, then they are
on the plane (or line) determined by that line and the origin; that plane being
a two-dimensional subspace, the span of the vectors has at most dimension
2; they are dependent. For the converse, see Exercise 5.
8. Say a f.= O. The solutions are given by y = arbitrary = t,x = cia - (bla)t,
which can be written (x, y) = (cia, 0) + t( -bla, 1). (Check that the right
side is what the hint refers to.) The right side describes the line (cia, 0) +
(((-bla, 1»)).
9. Assume a f.= b and that they are both on the line e + ((d)). Thus, a = e + ad,
b = e + f3d. We solve: e = (f3a - ab)/(f3 - a), d = (b - a)/(f3 - a).
(Why is f3 f.= a?)
Now, suppose p := ka + lb is on the line joining a and b. We check that
p = (k + l)e + (ka + 1f3)d = e + md, which is on the line e + ((d)). (Why is
k + I = I?) That shows that "the line joining a and b" is a subset of e + ((d)).
Conversely, suppose q := e + rd is on the line e + ((d)). We check that
q = (f3 - r)/(f3 - a)a + (-a + r)/(f3 - a)b, a combination that is on the
line joining a and b. (Why?) Hence e + ((d)) is a subset of the line joining
a and b. This proves equality of the lines.
Section 1.3
1. To minimize the distance between (6, 8) and (x, xI2), minimize instead its
square, d 2 := (x - 6)2 + (x 12 - 8)2 = 5x 2 14 - 20x + 100. By calculus:
Its derivative 5x 12 - 20 is zero when x = 8, where its second derivative is
positive. By algebra: d 2 = 5([X_~2+16) is least when x = 8.
2. (a) By definition, the angle is cos- 1 (ae b/llall2l1bll2) = cos- 1 ([ J3J3 +

1( -1) + O]/[ A.J4]) = rr 13.
(b) The (your) sketch should show that the angle between a and -b is
the supplement of the angle between a and b. This is borne out by the
definition: because a • (-b) = - (a • b), we have (using the values
here) cos- 1 (-i) = rr - cos- 1 (i) = 2rr 13.
(c) From a + b = (2J3, 0, 0), the calculation is cos- 1(6/4J3) =
cos- 1 (J312) = rr 16. That is predictable. Vectors a and b, being
equally long, are sides of a rhombus. In the sketch, a + b is the diagonal

of the rhombus. Therefore, it bisects the angle between a and b.
3. This is vector-space material.
(a) x. 0 = x • (0 + 0) (vector identity) = x • 0 + x • 0 (Theorem

1.1(b». Subtract x • 0 from both sides and you get x. 0 = O.
(b) Yes. (1,2) • (-2,1) = 0 is an example in R2.
(c) No. If x is orthogonal to all, then in particular, x • x = 0, forcing
x = O. (Reason?)
(d) In R 1 , yes vacuously; there are no "vectors that are not .... " In dimen-
sion > 1, no: given x f= 0 and another vector y that is not a multiple
of x, it is impossible for x to be orthogonal to both y and to x + y,
because then x • x = x • (x + y) - x • y = o.
4. Angle is -IT { } x. y/llxl1211yll2 = -1 =} x. Y = -llxll2l1yll2 =} one

of x and y is a multiple of the other (Cauchy's inequality). Since they are
nonzero, we write x = ay. Then a(y • y) = (ay) • y = -liay 11211y 112 =
-lalllyll2l1yll2 implies a = -Ial, and a is negative. Conversely, x = ay
with a negative gives x • y IIIxll211yll2 = a/lal = -1, making the angle IT.
5. (a) By definition, IIx ± ylll = (x ± y) • (x ± y), which by numerous uses

of Theorem 1.1 match x • x ± 2x • y + y • y (all respectively). If x and
yare orthogonal, then we have Ilx + ylll = x. x + y. y = IIx - yll{
(b) In the figure, we see that the hypothesis says that x and yare the sides
of a rectangle. The equation says that in a rectangle, the diagonals are
equally long.
Ix-ay
I
I
\
\ ay - proj
y
Exercise 5b Exercise 6b Exercise 8c
6. (a) By multiple uses of Theorem 1.1, we show that (x + y) • (x - y) =

x • x - y • y = Ilxlll- lIyll{ If those last two norms match, then
x + y and x - yare orthogonal.
(b) In the figure, IIxll2 = IIyll2 says that x and yare the sides of a rhombus.
The equation says that in a rhombus, the diagonals are perpendicular.
(Compare Exercise 2c.)
7. (a) Symmetry: x. y := X1YI + ... + XnYn = Y1Xl + ... + YnXn = Y. x

by commutativity of real multiplication.
(b) Distributivity: (x + y) e Z := (Xl + Yl)ZI + ... + (Xn + Yn)Zn =

(XI ZI + X/lZ n ) + ... + (YIZI + YnZ/l) by several uses of the real-field
axioms; the last is x e Z + y e z. Similarly for the other half.
(c) Homogeneity: (ax)ey:= (axI)YI + .. ·+(axn)Yn = a(xIYI + ... +
xnYn) = a(x e y). The next-to-Iast quantity also matches XI (aYI) +
... + Xn (aYn) = x e (ay).
(d) Positive definiteness: x ex := XIXI + ... + X/lX n = x? + ... + xn2.
Squares being nonnegative, the last is a sum of nonnegative terms, so
x e x 2: O. Moreover, if any of the squares is positive, then the sum is
at least that positive. Hence if x e X = 0, then all the Xi have to be zero,
and x = o.
8. (a) [x-proj(x, y)]ey = xey-proj(x, y)ey := xey-[xey jyey]yey = 0

by parts (b) and (c) of Theorem 1.1.
(b) (See the answer to (5).) IIx - yll} := (x - y) e (x - y) = x e x -
2x e y + Y e Y (symmetry, homogeneity). If x is orthogonal to y, then
IIx - ylll = x e X + Y e Y = II x lll + lIyllt
(c) See the figure for this solution. Let ay be a multiple of y. Because
x - proj(x, y) is perpendicular to y (part (a)), it is perpendicular to all
mUltiples ofy, and therefore to ay -proj(x, y). (Why is x -proj(x, y)
perpendicular to multiples of y? Why is ay - proj(x, y) a multiple of
y?) By the Pythagorean theorem, IIx - aylll = IIx - proj(x, Y)lIl +
Ilay - proj(x, Y)lIt Clearly, the shortest length Ilx - aYl12 occurs
precisely for ay = proj(x, y).
9. We have seen that IIx - ylll = x e x - 2x e y + Y e y. On the right, x e x

and y e yare Ilxlll and lIylll by definition of norm, x ey is Ilxll2 lIy 112 cos()
by definition of angle.
Section 1.4
1. (a) lIaIl2:= )(12+22+42) =,J2f, IIbIl2:= )((-3)2+42 +5 2) =
5.J2, lIa + bll2 := ) (( _2)2 + 6 2 + 9 2) = 11.
(b) It suffices to check that the smaller two add up to more than the third:
,J2f + 5.J2 > 4.5 + 7> 11.
(c) By properties of norm, 113al12 = 311al12 = 3,J2f and 113a + 3bll2 =
311a + bl12 = 33.
2. (a) The calculations are direct: lIell2 = J84 = 2.J21, IIdll2 = 3,J2f,
lie + dll2 = ,J2f.
(b) Clearly, two of them sum to the third, so the vectors form a degenerate
triangle. More specifically: lIell2+ Ile+dll2 = IIdll2 = lI-dll2 = lIe-
(e + d) 112 forces (Theorem 1.6(a)) e + d = -,Be, and d = (-1 - ,B)e.
Indeed, we see d = -1.5e.
(c) In view of (b), we have II 30c+40dll2 = 1130c-60c1l2 = 11-30c1l2 =

30llcll2 = 30.J84.
3. (a) (x, y) has unit maxnorm iff Ixi ~ 1, lyl ~ 1, and one of them equals
1. Hence the "unit circle" is {(x, y): x = ±1 and -1 ~ y ~ 1, or
y = ± 1 and -1 ~ x ~ I}. In our usual language, that set is the square
with vertices at the four points (±1, ±1).
(b) If (x, y) is on the "circle," then 11(2,0) - (x, y)lIo := max{12 - xl,
Iy I} = 2 - x. This length is 1 iff x = 1; all the points on the right side
of the square fit the description.
4. Clearly, 11(1, 0) - (-1, 0) II 0 = 2, so this is the length we are seeking. We

noted that for (x, y) on the "circle," Ix I ~ 1 and Iy I ~ 1. Hence 11(1, 0) -
(x, y)lIo := max{11 - xl, Iyll can be 2 iff x = -1. Thus the chord from
(1, 0) to any point on the left side of the square is of length 2; no uniqueness.
5. (a) By Theorem l.l(c), x ex 2: 0, with equality iff x = O. The same holds

for (x. x)I/2.
(b) lIaxll2 := (ax. ax)I/2 = (aa)1/2(x. x)I/2 (Theorem l.l(c)) =
lalllxll2 (definitions).
6. Thelawofcosinessaysthatllx-ylll = Ilxlll+llylll-2I1xll2l1yll2cose.
Since cose 2: -1, we have Ilx - Yill ~ IIxlll + IIYlIl + 211xll211yl12 =
(lIxll2 + lIyIl2)2. The inequality follows.
7. Set x := a, y := -b.
(a) => (b): From lIa - bll2 ~ lIall2 + IIbll2, we get IIx + yll2 ~ IIxll2 +
11- yll2 = IlxIl2+llyI12, with equality iffy := -b = -(aa) = (-a)x
for an a ~ O.
(b) => (a): From Ilx + yll2 ~ IIxll2 + lIyll2, we get lIa - bib ~ lIall2 +
II - bll2 = lIall2 + Ilbll2, with equality iff a := x = f3y = (-f3)b with
f3 2: O.
8. (a) Positive definiteness: Ifx = (0, ... , 0), then IIxlio := max{O, ... , O} =
o. If instead x =1= 0, then some Xi =1= 0, so IXi I > 0; in this case,
IIxlio := max{lxII,···, IXnll2: Ixil > O.
(b) Radial homogeneity: lIaxlio := max{laxII, ... , laxnll = max{lallxll,
... , la Ilx nIl. Because the common factor la I is nonnegative, the biggest
IXi I gives the biggest la Ilxi I. Hence
Ilaxlio = lal max{lxII, ... , IXnll = lalllxllo.

(c) Triangle inequality: IIx + Yllo := max{lxI + YII,.··, IXn + Ynll. For
each i, we have IXi + Yi I ~ IXi 1+ IYi I ~ IIxlio + lIyllo. (Why?) Hence
max{lxi + Yi Il ~ IIxllo + lIyllo.
Section 1.5
1. Clearly, y - x = 52(1, 1, 1) and z - x = 16(1, 1, 1) are dependent, so x, y,
and z are on a line (Exercise 4 in Section 1.2). In fact, the sizes make clear that
z is between x and y. Therefore, it should work out that d(x, z) + d(z, y) =
d(x, y). Verify!
2. (a) It is the perpendicular bisector of their segment: the y-axis. The dis-
tances from (x, y) to (±1, 0) are lI(x, y) - (±1, 0)112 = (x ± 1)2 +
y2)1/2. Setting their squares equal, we obtain 2x = -2x, equivalent
tox = O.
(b) Not just a line: the y-axis is part of it, but it also contains the vee-shaped
region above y = 1 between the lines y = x + 1 and y = -x + 1,
plus its mirror image below the x-axis.
To see that, view only quadrant I and its edges. There, the two dis-
tances are do «x ,y), (±1, 0)) := max{lx =f lI, Iyj} = max{lx =f lI, y}.
Clearly, x = 0 makes the two distances match, max{l, y}. For x > 0,
Ix + 11 = x + 1 exceeds Ix - 11 = ±(x - 1); therefore, if y :=: x + 1,
then the two maxes are both y, and if y < x + 1, then one max is
x + 1 and the other is less. We conclude that in this quarter-plane, the
required points satisfy y :=: x + 1. By symmetry, the entire set satisfies
Iyl :=: Ixl + 1.
3. The three parts of Theorem 1.9 are named after three properties of norms,
and those norm properties are the explanations:
(a) d(x, y) := IIx - yll = d(y, x) because norms are symmetric;

(b) d(x, y) := IIx - yll > 0, except when x = y, because norms are
positive definite;
(c) d(x, z) := IIx - zll :s d(x, y) + d(y, z) because norms satisfy the
triangle inequality.
4. (a) d(ax, ay) := lIax-ayll = lalllx-yil (radial homogeneity of norms)

= lald(x, y).
(b) d(x, y) := IIx - yll = II (x + z) - (y + z) II (vector algebra) = d(x +
z,y+z).
Section 1.6
1. As x goes from 0 to 1, f decreases from 1 to -1, g increases from 1 to 2,
and f - g decreases from 0 to -3. Hence max If I = 1, max Igl = 2, and
max If - gl = 3; we have IIf - gllo = IIfllo + IIgllo· But IIf - gll2 =
IIfll2 + IIgll2 is impossible. It would violate Theorem 1.6, because f and g
are not multiples: f(,J'i/2) = 0 =F g(,J'i/2).
2. The cosines C2 and CI are 2 apart: At x = ~,cos 2rr x = -1 and cos 4rr x = 1;
therefore, 2 :s sup IC2 - cII :s sup IC21 + sup ICJ I = 2, and IIc2 - cIllo = 2.
Cosine CI and sine SI are ,J2 distant: cos 2rr x - sin 2rr x = ,J2 sin(rr /4 -
2rrx), so that llel - sllio := sup ICJ - sll = ,J2.
3. (a) From trigonometry, H = F - 2G.

(b) We cannot have f = ago Algebraic evidence: f has three zeros, but
g, being quadratic, cannot have that many; in fact, g ~ i~. Calculus
evidence: If f were ag, we would have f' = ag', f" = ag", ... ; but
gill is already 0, fill not.
4. Positive definite: IIfllo := sup If I ~ 0, and 111110 = 0 * If(x)1 = 0 for

every x * f = O.
Radially homogeneous: one way to proceed is to recall that sup If(x)1 is
some value If(b)l. Then laf(x)1 = lallf(x)1 :s lallf(b)1 for all x, with
equalityatx = b. Hence sup laf(x)1 = lallf(b)l, which says that II alii 0 =
laillfilo.
5. Both parts follow easily from properties of inner products. We have seen
that IIx + Yill ± IIx - Yill := (x + y) • (x + y) ± (x - y) • (x - y) =
Ilxlll + 2x. Y + lIylll ± Ilxlll =f 2x. y ± lIylll- The upper signs give (a),
the lower (b).
6. (a) Yes. The product of integrable functions is integrable.

(b) Yes, yes, no. It is symmetric, because fg = gf. Linear, by properties
of integrals. Not positive definite: F(x) := 0 for 0 :s x < 1, := 1 for
x = 1, defines an integrable function that is not the zero of the space
1[0, 1], but has F • • F = o.
7. (a) Induction: f(1) = If(I); and f(k) = kf(l) forces f(k + 1) =

f(k) + f(l) (additivity) = kf(l) + f(1) = (k + l)f(I).
(b) f(O) = f(O + 0) = f(O) + f(O) forces f(O) = Of (1). Likewise,
f(k) + f( -k) = f(k + -k) = 0 (just proved) forces f( -k) =
- f(k) = -(kf(I)) = -kf(I).
(c) We may assume m > O. For any s, mf(s) = f(s) + ... + f(s) =
f(s + ... + s) = f(ms). Hence mf(k/m) = f(k) = kf(I). Divide
by m to get f(k/m) = k/mf(l).
(d) (s-8 argument) Assume that f is continuous at x = b. Choose any
s > O. There is a corresponding 8 such that Ix - bl < 8 implies
If(x) - f(b)1 < s.
Now suppose Iy - 01 < 8. The point y + b is within 8 of b, so If(y +
b) - f(b)1 < s. On the other hand, fey + b) - feb) = fey). We
conclude that If (y) - f (0) I = If (y) I < s. We have proved that f is
continuous at y = o.
(e) (sequence argument) Assume that f is continuous at x = b. For any

sequence (Xi) converging to b, we have f(xi) -+ feb).
Now let Yi -+ c. We see that f(Yi) = f(Yi + [b - cD + fCc -
b) -+ feb) + fCc - b), because Yi + [b - c] -+ b. Since the sum
feb) + fCc - b) matches fCc), we have shown that f(Yi) -+ fCc),
proving that f is continuous at Y = c.
(t) Let z be irrational. By density of the rationals, there is a sequence
(qi) of rationals converging to z. By part (e), f is continuous at
x = z. Therefore, fez) = lim f(qi) = lim(qi f(l)) (by part (c))
= f(1) lim(qi) = f(1)z.
Chapter 2
Section 2.1
1. Write z := f3x + ay. We discussed in Section 1.2 that z is on the segment
from x toy. Thelengthsare IIx-zll = 11(1-f3)x-aYIl = allx-yll (because
a ~ 0) and liz - yll = lIf3x + (a - l)YII = lIf3x - (1- a)yll = f3l1x - YII·
The ratio is clear.
2. Write d := IIx - yll. The hypothesis YEN says that d < r.
(a) Pick a fixed s ~ r - d. Then r ~ d + s, and we are in the situation of

Theorem 2.2(c). Hence N(y, s) £; N.
(b) It suffices to show that N £; N(y, 2r). That zEN implies liz - yll ~
liz - xII + IIx - yll < r +d < 2r, which in turn implies Z E N(y,2r).
The inclusion follows.
3. (a) If r + t exceeded d, then the neighborhoods would intersect (Theorem

2.2(b)), and so would the balls. If r + t equaled d, then the point
t f(r + t)x + r /(r + t)y would be on both spheres (Verify!), and would
therefore be common to the balls. Hence r + t < d.
(b) and (c) Using the triangle inequality, we getd := IIx - yll ~ IIx - zll +
liz - wil + IIw - yll ~ r + liz - wll + t. Hence liz - wll ~ d - r - t,
and the inequality is strict if either IIx - zll < r or IIw - yll < t.
These parts prove that the least liz - wll might be is d - r - t, and that the
only way this minimum can be achieved is if z and ware on the surfaces of
the respective balls.
(d) For any point (1-a)x+ay on the line joining x andy, the distance tox
IIx - «l-a)x+ay) II = lalllx-Yil is r iff a = ±r/d. That is, theline
meets the sphere in just two points, of which u := (1 - r / d)x + r / d y
is the one on the segment (a > 0).
(e) Either reasoning as in (d), or observing that the point we want breaks
the segment from x to Y into the ratio (d - t) : t (compare Exercise
1), we get the same answer: v := t /d x + (1 - t /d)y.
(f) lJu-vlJ = 11(I-r/d)x+r/dy-t/dx-(I-t/d)ylJ = IJ(I-r/d-
t/d)(x - y)1J = (1 - rid - t/d)d.
Parts (b) and (c) established the smallest conceivable distance d - r - t.

Parts (d)-(f) show that this minimum is actually achieved along the segment,
and maybe other places.
(g) Assume that Z and ware on the two spheres. In (b), we saw that d :=
Ilx - yll :::: IJx - zlJ + IJz - wlJ + IJw - ylJ = r + IJz - wlJ + t. In an
inner product space, the inequality is strict, unless z is on the segment
from x to wand w is on the segment from z to y (Theorem 1.10). In
that case, z = (1 - a)x + aw and w = (1 - (3)z + (3y. Solving those
equations for z and w in terms of x and y, we deduce that z and w are
on the segment from x to y.
(h) By (g), this situation guarantees that z and ware on the segment from
x to y. By (d) and (e), the only such points on the spheres are z = u
and w = v.
4. By Exercise 2, some neighborhood N(u, p) is contained in N(x, r) and

some N(u, q) S; N(y, t). Then N(u, min{p, q}) fits within both.
Section 2.2
1. (a) Yes. Because i sin(I/ i) = sin(1/ i)/(I/ i) ~ 1 and I cos i / i I ::::
I/i ~ 0 as i ~ 00, we should expect (Xi) to converge to (1,0).
We verify: IJxi - (1, O)lJl = (i sin(1/i) - 1)2 + cos 2 i/i2 ~ O.
(b) No. The terms are (0,1), (-1,0), (0, -1), (1,0), (0, 1), .... Clearly,
YI, Y5, Y9,··· converges to (0,1), Y2, Y6, YIO,··· to (-1,0). With
unlike subsequences, (Yi) diverges (Theorem 2.4(a)).
(c) No. The terms begin (0, 1), (-2,0), (0, -3), .... Verify that IIzi 112 = i.
The sequence is unbounded, so it cannot have a finite limit (Theorem
2.7(c)).
(d) No limit. Terms are (0,0), (8, 8), (0, 0), (32, 32), ... ; the subsequence
of odd-numbered terms tends to 0, the evens to infinity.
2. Only (c). In the (c) answer, we said that IJzi II = i, which does approach 00.
In the others: (a) is bounded (it converges), so cannot approach 00; (b) is
bounded because each norm is 1; (d) has a subsequence not going to 00.
3. (a) Assume (Xi) ~ x and (Yi) ~ y. Then II(axi + (3Yi) - (ax + (3y)11 =
lIa(xi - x) + {3(Yi - y)11 :::: allxi - xii + {3I1Yi - yll ~ 0 + 0, which
says that aXi + {3Yi ~ ax + {3y.
(b) No. Take (Xi) to be (1, 1), (2,2), ... and Yi := (-l)i xi . They both
converge to 00, but (Xi + Yi) behaves like l(d).
4. (a) Since the difference of norms is at most the norm of the difference
(Theorem 1.8), we have I II Xi II - IIx III ::: IIxi - XII. If Xi -+ X, then the
right side tends to 0; so must the left.
(b) Xi := (_l)i (1,0) has II xiii -+ 11(1,0) II, but not Xi -+ (1,0). (Why?)
(c) Xi -+ 0 {} Ilxi - 011 -+ 0 {} Ilxi II -+ O.
5. By definition, Xi -+ X means that for any 10 > 0, there exists 1(10) such
that i ::: 1(10) implies II Xi - xII < e. Just substitute r for 10, then note that
IIxi - xII < r {} Xi E N(x, r).
6. (a) (Xi) is bounded {}thereis M with II xiii < Mforalli {} IIxi-OIi < M
for all i {} Xi E N(O, M) for all i {} {Xi} ~ N(O, M).
(b) Xi -+ 00 means IIxi II -+ 00. The latter is defined as follows: For any
M > 0, there exists I such that i ::: I implies IIxi II ::: M. Thus, the
numbers II Xi II have no upper bound. Hence (Xi) is unbounded.
7. (a) Exercise lb.

(b) Exercise 1d. See also the remarks in answer to Exercise 2.
Section 2.3
1. (a) Converges to (~, 1, ~); verify that the coordinate sequences converge
accordingly.
(b) Diverges, because the third coordinates diverge: tan 1i /4 = 1,
tan 31i/4 = -1, ....
(c) Converges to infinity, because the norm exceeds the second coordinate
ei .
2. (a) (1,0), (2,0), (3,0), .... The distance between any two terms is ::: 1.
Therefore, no subsequence (Xj(i» can have IIxj(i) - Xj(k) 112 -+ 0; no
subsequence can be Cauchy.
(b) A sequence has the property iff it converges to 00. If (Xi) -+ 00, then
IIxi - Xj 112 -+ 00 as j -+ 00, no matter how you choose i. Conversely,
if (Xi) does not tend to 00, then for some M, there is an infinity of terms
with norms below M (by definition). Those constitute a bounded subse-
quence. By BWT, the last must have in tum a subsequence converging
to a finite limit. The latter subsequence must be Cauchy.
3. (a) No. The x-coordinates are all 1,0, or -1. A convergent subsequence
cannot have an infinity of two of those, because then it would in tum
have unlike subsequences. Hence, eventually the x-coordinates be-
come constantly 1, forcing sublimit (1,0); or they become -1, forcing
sublimit = (-1, 0); or they become 0, and we apply similar reasoning

to deduce that y stabilizes at either 1 or -1.
(b) Similar to (a): a convergent subsequence cannot have an infinity of
odd-numbered terms and an infinity of evens, so eventually it takes the
road to 0 or to 00.
4. (a, c) They converge, so all subsequences converge to the parent limit (The-
orem 2.4(a)*, on p. 41). That is the only sublimit. (Compare Exercise 8a.)
(b) The coordinates look like x = sins/s -+ 1 (as s -+ 0), y = (1 -

cos s) / s2 -+ 1/2, and z = ± 1. Hence (reasoning as in Exercise 3a) a
convergent subsequence must approach (1,1/2,1) or (1,1/2, -1).
5. Yes, under our convention that R 1 is not the same as R. In R 1 , Xi := [_I]i i

has norm (= absolute value) i. This norm tends to infinity, so (Xi) -+ 00,
the only sublimit.
6. (a) Suppose IJr}(xi)1 -+ 00. Then IIxi III = Jrl (Xi)2 + ... + Jr n (Xi)2 ~
Jr j (Xi)2 -+ 00, and (Xi) -+ 00.
(b) (1,0), (0,2), (3,0), (0,4), ... converges to 00, because IIxi 112 = i;
but the two coordinate sequences oscillate.
7. (a) Assume (Xi) -+ x. Then for any £ > 0, there exists I such that i ~ I
implies II Xi - xii < £/2. By the triangle inequality, II Xi - X} II =
IIXi - X + X - x} II S II Xi - xII + IIx - x} II < £/2 + £/2 for every
i, j ~ I. Hence (Xi) is Cauchy.
(b) Use £ := 1 in the definition of Cauchy. There exists a corresponding I
suchthati,j ~ I ~ IIXi-X}1I < 1.SetM:= IlxllI+·"+lIxIII+1.
Then clearly, II x} II < M for j S I. For j > I, IIx}1I S IIx} - XIII +
IIXIIl < 1 + IIxl II SM. This says that (Xi) is bounded.
8. (a) A sublimit is the limit of some subsequence. If the parent sequence is

convergent, then (Theorem 2.4(a)* on p. 41) the limit of any subse-
quence is the parent limit. Hence the parent limit is the only sublimit.
(b) Assume that B is the unique sublimit of (Xi).
The argument is the same whether B is a vector or 00. Let N be any
neighborhood of B. There cannot be infinitely many Xi outside N:
Such a multitude would constitute a subsequence (x}(i); by the last
paragraph of this section, (X) (i) would in turn have a subsequence
(Xk(j(i))) converging to some (possibly infinite) C; C could not be B,
because then the sequence (Xk(j(i))) would have to enter N (Theorem
2.5*); C would be a second sublimit of (Xi). Hence only finitely many
terms are outside N. Thus, starting with some index I, we have Xl,
Xl+l, ... E N. This being true for arbitrary N, Theorem 2.5* tells us
that B is the limit of (Xi). The sequence converges. (Why does this
argument require Rn?)
9. => Assume that x is a sublimit of (Xi) and N a neighborhood of x. By

definition of sub limit, there exists a subsequence (x j (i)) converging to x. By
Theorem 2.5, there is I such that i 2: I => Xj(i) E N. Thus, N contains
Xj(l), Xj(I+I),····
{= Assume that every neighborhood of x has infinitely many terms. Then

N(x, 1) has some term Xj(l). Also, N (x,~) has terms Xi with i > j(1),
because only finitely many terms are numbered j (1) or less; let x j (2) be
such a term. Also N (x, 1) must have Xj(3) with j (3) > j (2), etc. Thus, we
recursively define an increasing sequence j (i) with Ilxj(i) - xii < 1/ i. That
is, (Xj(i)) -+ x, and x is a sublimit.
10. => Assume that x (or (0) is a sublimit of (Xi), so that some subsequence
(Xj(i)) goes to that value. Let 8 and I be chosen. By Theorem 2.5, there is a
starting pointl for which Xj(J), x j(1+I), ... , are all in N (x, 8) (respectively,
outside N(O, 2/8)). Then i := J + I satisfies i > I and IIxj(i) - xII < 8
(respectively, IIxj(i) II > 1/8).
{= Mimic the argument in Exercise 9. For finite x, set 8 := 1, I := 1, and
pick the corresponding index j (1); set 8 := ~, I := 2 + j (1), and pick
the corresponding index j (2); set 8 := 1,
I := 3 + j (2), and pick the
corresponding index j(3); etc. Then (Xj(i)) is a subsequence with Ilxj(i) -
x II < 1/ i , making x a sublimit.
Section 2.4
1. (a) For each graph, the upper horizontal segment is at y = 1, the lower at
y =0.
y y y
,,
,,
&------+x (9-----40 x
(b) Let 2k + 1 ~ i ~ 2k+1. We know that Hi 2: 0, and Hi is constantly 1

on an interval of length 1/2k. Hence fOI Hi 2 2: 1/2k. On the intervals
to either side, the rise/fall of Hi guarantees that f Hi 2 < f Hi =
.5 (1 /2k), and there are no more than two such intervals. Since Hi = 0
the rest of the time, we conclude that (1 + .5 + .5)(1/2k ) > fd H?
(c) Part (b) says that II Hi 112 -+ 0, so Hi -+ O. (Recall Exercise 4c in
Section 2.2.)
2. (a) Yes. 0 ::::: Hi ::::: 1, and 1 is a value of Hi, so II Hi 110 := sup IHi! = 1.
(b) No. It does not have Cauchy subsequences, because any two terms are
1 apart: If i < j, then -1 ::::: Hi - Hj ::::: 1 always; also, in part of
the subinterval in which Hi = 1, Hj is either constantly 0 or rising
from or falling to 0, so that somewhere Hi = 1 and Hj = 0; therefore,
IIHi - Hjllo = 1.
3. Nowhere. The upper horizontals for each group (graphs number 2k + 1 to

2k+1) cover the whole segment from (0, 1) to (1, 1), and the lower horizon-
tals cover (0,0) to (1,0) more than once. Therefore, at each x = a, there
exist an infinity of i for which Hi(a) = 1, along with even more j with
Hj(a) = O. The sequence (Hi (a» diverges.
4. (a) gl (x) == 0 and g2(X) == 1. If g(a) = 0 and g(b) = 1, then (intermedi-

!
ate value theorem) g = someplace. Hence if g is always 0 or 1, then
it cannot take on both values.
(b) If h is continuous at x = ~, then limi-HlO h(! - = h(!) = t)
limi~oo h(! + t). The values of h cannot then be 0 to the left of
x = ~ and 1 to the right.
5. (a) See the solution to Exercise 7b in Section 2.3.

(b) Assume (Xj(i) --* x, and let s > O. By convergence, there is I such
that i ~ I implies Ilxj(i) - xii < s/2; by Cauchyness, there is J such
that i, j ~ J implies IIxi - Xj II < s/2. Then i ~ 1+ J, which makes
j (i) also ~ 1+ J, forces IIxi - xII ::::: Ilxi - Xj(i) II + IIxj(i) - xII < s.
We conclude that (Xi) --* X.
(c) Assume that (Xi) is Cauchy. By (a), it is bounded. By BWT, it has a
subsequence converging to a vector. By (b), (Xi) itself must converge
to that vector.
Chapter 3
Section 3.1
1. => By definition, f := (fl, ... , fm) is of first degree if fi(X) = ailXI +
... + ainXn + bi for each i. Then
f(x) - fey) = [
fl(X)-fl(Y)]
... =
[all
fm(x) - fm(Y) ami
~ Iff(x) - fey) = A[x - y] for all vectors, then f(x) - Ax = fey) - Ay =

constant b, and we have f(x) = Ax + b. (Compare Example 2.) This says
that each row fi (x) is of the form ai I XI + ... + ainXn + bi.
2. (a) g\(x, y) := (1/[x 2 + y2], 1/[x 2 + y2], 1/[(x - 1)2 + (y - 1)2]) is

undefined at precisely (0,0) and (1,1).
(b) g2(X, y) := (x, y, 1/[y - ax - b]) is undefined along y = ax + b.
3. No. You would need f(x) = (P\ (x)/q\ (x), ... , Pm(x)/qm(x)) such that at
each place in the square, one of the denominators is zero. Then q(x) :=
q\ (x) ... qm (x) would be zero throughout the square. Organize q as the hint
says: q(x, y) = rk(x)l + ... + r\(x)y + ro(x). For a fixed t E [-1,1],
q(t, y) would be a polynomial in y with infinitely many zeros. Hence each
coefficient r j (t) would be zero. This being true for an infinity of t, each
polynomial rj(x) would have to be identically zero. Hence q(x, y) would
be identically zero, and f would have empty domain.
4. ae - bd -:j:. O. The hypothesis says that
h(X'Y)=(~ ~)(~)+(;).
Clearly, c and! are irrelevant; h is one-to-one iff the matrix is one-to-one,
which occurs iff it has nonzero determinant.
5. Yes. The question becomes whether the system represented by F(x, y, z) =

(13, 14, 15) has solutions. Reduce the augmented matrix
1 2
[ 5 6
9 10
The system is consistent.
6. Leti=(j,k, ... ,p),j=(l,m, ... ,q).
(a) (axi)(,Bxj) := (ax/xl ... x/)(,Bx/ x 2m ... x nq ) = (a,B)x/+ 1x 2k+ m

••• Xn
p+q = (a,B)xl'+'J.
(b) A polynomial is a sum of terms, and the term ax/ x 2k ... x/ is

ax(j,k, ... ,p).
(c) g(x, y, z) = x 2y 3Z4 + 5x 6y 7Z8 has degree 21. In general, ax i + ... +

,Bxj has degree max {j + k + ... + P, ... , I + m + ... + q }.
Section 3.2
1. (a) Yes. !(Xi) = IIXi III converges to 00 as Xi -+ 00.
(b) Yes, similarly. !(Xi) = IIXi III converges to 0 as Xi -+ O.
2. (a) Yes. If (Xi, Yi) -+ 00, then II (Xi , Yi)1I2 > 2 for i ~ some I. There-
after, max {x/, Yi 2 } must exceed 1, so that g(Xi, Yi) := x i 4 + Yi 4 ~
max {xi4, Yi4} > max {x/, Y/} ~ (x i 2 + y/)/2 -+ 00.
(b) Yes. If (Xi, Yi) ~ 0, then eventually lI(xi, Yi)1I2 < 1. Thereafter, xi 2
and Yi 2 are both ~ 1, so that g(Xi, Yi) := x i 4 + Yi 4 ~ x/ + Y/ ~ O.
(c) Yes, because g(Xi, Yi) ~ 0+ (part (b)) forces 1/g(Xi, Yi) ~ 00.
3. (a) No, because G(i, i) = i2 ~ 00, while G(i, 0) = 0 ~ O.

(b) Yes; 0 ~ IXiyi! ~ (X/+y/)/2 (property of rea Is) = II (Xi, Yi)lIl/2 ~
O.
(c) Yes, under our definition. By (b), (Xi, Yi) ~ 0 implies G(Xi, Yi) ~ O.
The last means that 11/ G(Xi , Yi) I ~ 00, which we accept as 1/ G ~
00, even though the actual values can go to either 00 or -00.
(d) Yes. ,JG is undefined in quadrants II and IV, but if Xi ~ 0 in the

domain of,JG, then ,JG(Xi) ~ 0 by part (b).
4. (a) At 0, limit is 00, because JXi Yi ~ 0+, as in Exercise 3d. At 00, no

limit; 11 (X, y) ~ 5 along the curve Y = l/x, ~ 00 along Y = l/x3.
(b) At 0, limit is 1: XiYi ~ 0, so h(xi, Yi) acts like sins/s ~ 1. At 00,
no limit: hex, y) = sin 1 along Y = l/x, but hex, y) ~ 0 along
Y =x.
(c) AtO,limitis(O,O): IIf3(xi,Yi)lll = (Xi2_Yi2)2+(2xiYi)2 = (x/+
Yi 2)2 ~ 0, so f3 (Xi, Yi) ~ O. The same equation shows that the limit
is 00 at 00.
(d) No limit at either place. Near 0, f4(X, y) ~ (0,0) along X = 0, but
Ilf4(x, y) III = 2x2/2x2 along Y = x. Near 00, f4(X, y) = (0, 1) along
Y = x, but tends to 00 along Y = O.
5. Let c be a closure point of S. By definition, some sequence (Xi) from S has

limit c. Given any neighborhood N of c, by Theorem 2.5, there exists I such
that XI, X/+ 1, ... are all in N. Thus, XI is a point from S in N.
6. (a) =} Assume that d is an accumulation point of S. The neighborhood

N(d, 1) has some Xl =1= d from S, the neighborhood N(d, 1/2) has
some X2 =1= d from S, .... We thereby define (Xi) from S with Xi =1= d
and lid-XiII < l/i, so that (Xi) ~ d.
¢= Assume (Xi) ~ d and d =1= Xi E S. Let N be a neighborhood of
d. By Theorem 2.5, XI E N for some I. Thus, N has elements of S
different from d. Hence d is an accumulation point of S.
(b) =} Assume that d is an accumulation point of S. The neighborhood
N (d, ~) has some Xl =1= d from S, the neighborhood N (d, II d - xIII /2)
has some X2 =1= d from S, .... For this sequence, each term is less than
half as far from d as the previous. Hence the (Xi) are different from
each other as well as from d, plus (Xi) ~ d.
Now let N be a neighborhood of d. As in Exercise 5, we find XI,
X/+l, ... , an infinity of members of Sin N.
{= Trivial; the hypothesis is stronger than the definition of accumula-

tion point.
7. (a) Along y = ax, H(x, y) = a 2 x 2 /x = a 2 x. Since (x, y) -+ (0,0)

forces x -+ 0 (limit of coordinates), we have H -+ 0 along the line.
(b) Along y = JX, H is constantly 1. Hence H -+ Ion the curve; H has
no limit at (0,0).
8. (a) No. If Xi := (_I)i Ii, then Xi E R# and Xi -+ 0, but K(Xi) = (-I)i

does not have a limit.
(b) Yes. For every member t of R+, It I = t, so K+(t) = 1. Thus, if
ti E R+ and ti -+ 0, then K+(ti) = 1 has a limit.
Understand the difference from (a) to (b): Restricting the domain eliminates
some testing sequences, exposing the restricted function to fewer tests, and
giving it increased chance to qualify. (See the similar comment in Example
1 of Section 3.4.)
Section 3.3
1. We need the result lims-+o s (ln s) = 0 (from L'Hospital's rule, or the more
elementary fact that t /21 -+ 0 as t -+ 00). It implies that for any e > 0,
there exists /1 > ofor which 0 < s < /1 => Is(lns)1 < e.ByTheorems3.7
and 3.3(b), we see that xy -+ 0 as (x, y) -+ (0,0). Hence (Theorem 3.4)
given /1, there is 8 such that 0 < II (x, y) 112 < 8 implies Ixy I < /1. We have
shown thateleads to8 with II (x, y)1I2 < 8 => Ixyl < /1 => Ixyln(xy)1 < e.
By Theorem 3.4, lim(x,y)-+(O,O) xy In(xy) = O.
2. Compare (1): (x, y) -+ (rr, 0) => x -+ rr, y -+ 0 (Theorem 3.7) => xy -+ 0

(limit of product), sinx/[x - rr] -+ -1 and [cosxy - 1]/x 2 y2 -+ -.5
(L'Hospital) => f(x, y) -+ (-1, -.5).
3. Let L := limx-+b f(x) < O. Set e := -L/2. By Theorem 3.4, there is a

neighborhood N(b, 8) in which b f=. xED implies If(x) - LI < e. For
such x, L - e < f(x) < L + e = L/2 = -e.
4. (a) By various theorems, lim f(x) = (lim 4 - [x - 1]2, lim 5 - [y - 2]2_

[z - 3]2) = (4,5); by Theorems 3.3, 3.7, and L'Hospital, lim g(y) =
6 + lim([u - 4] In[u - 4]) + lim([v - 5] In[v - 5]) = 6; but g(f(x» =
6 - (x - 1)2ln( -[x - 1]2) + ... is not defined anywhere, because the
two logarithms' arguments are ::s o.
(b) g+ matches g in the vicinity of (4,5), so irrespective of the value
of g+ at (4,5), limg+ = limg = 6. The composite g+(f) is unde-
fined, except at the place where f = (4,5), which is (1,2,3). Hence
(1,2,3) is an isolated point of the domain of g+(f), and lim g+(f) =
g+(f(1, 2,3» = 7.
(c) F is polynomial, so its limit is F(I, 2,3) = (4,5); lim G = 6, its value
in the vicinity of (4,5); G(F) is always defined, because G and Fare.
But G(F) has no limit: along the line x = 1, F has value (4,5), so
G (F) = 8; but along the line x = 1 + (, y = 2 + ( , Z = 3 + (, we have
F = (4 + (3, 5 + (3), G(F) = 6 for ( ;;6 O.
(d) Ft has a limit equal to its constant value; G(Ft(x)) is always defined
because both functions are; and lim G(Ft(x)) is the constant value 8.
5. Assume that b is an accumulation point of D; as usual, the isolated-point

case is trivial.
=} Suppose f has finite limit L at b. Then (Theorem 3.4) for any e > 0,
there is 8 > 0 such that IIx - bll2 < 8 =} IIf(x) - LI12 < e/2, lIy -
bl12 < 8 =} IIf(y) - Lib < e/2 (for x, y ;;6 b in D). For such x and y,
Ilf(x) - f(y) 112 .:::: IIf(x) - Lliz + ilL - f(y)112 < e.
<= Suppose (Xi) is a sequence from D converging to and not reaching b. Let
e > O. By hypothesis, there exists 8 such that IIf(x) - f(y) 112 < e for x and
y from D with 0 < IIx - bl12 < 8,0 < lIy - bl12 < 8. Because (Xi) must
be Cauchy, there exists J (e) such that i, j ~ J (e) implies II Xi - Xj 112 < 8.
Therefore,i,j ~ J(e) =} IIf(Xi)-f(xj)1I2 < e.Thisshowsthat(f(xi))isa
Cauchy sequence. Hence this value-sequence converges, proving that f has
a limit at b.
(Why was it essential to assume that f maps into Rm?)
6. Assume lim f(x) = u and lim g(x) = v. If (Xi) -+ b, then (af(xi) + .Bg(Xi))
converges to au +.Bv (Theorem 2.3(b)). Hence af +.Bg has a limit, and the
limit is au + .Bv.
7. See Example 2 in Section 3.4.
Section 3.4
1. If lIall < 1 (or lIall > 1) and (Xi) -+ a, then (Theorem 2.4(b)) lim IIXi II =
lIall < 1 (respectively, > 1). Hence (property of limits) beginning with
some J, II Xi II < (liall + 1)/2 (resp. ». Thus, i ~ J =} IIXili < 1 (resp. »
=} f(Xi) = 1 (resp. = 0) = !(a). We have shown that (Xi) -+ a implies
!(Xi) -+ !(a);! is continuous at a.
Ifinstead Ilbll = 1, thenYi := (1 + 1/ i)bconverges to b with IIYi II = 1+ 1/ i,
so !(Yi) = O. Hence !(Yi) does not approach 1 = !(b); ! is discontinuous
at b.
2. Fromllly-bll-lIx-blll.:::: lIy-xll(Theorem1.8),weseethatlly-bll-+
IIx - bll as y -+ x. Thus, the limit equals the value.
3. The domain of! is {x: Ilxll .:::: I} U {O}. On this domain, IIxll is continuous
(Exercise 2), IIxll - 1 is a continuous linear combination, and their product
is in the domain of the continuous real-variable function h(s) := ,.;s. By

Theorem 3.12, f is continuous. In general, if g(x) = G(lIxlD, where G(t)
is continuous for t ~ 0, then g is continuous on its domain.
4. The key is Cauchy's inequality: Iyeb-xebl = I(y-x)ebl ::: lIy-xll21lbll2.

so that y -+ x forces y e b -+ x e b. Since only properties of inner product
spaces are needed, the answer to (b) is yes.
5. (a) => (b) Because bED, b is trivially a closure point of D, so it is legal

to discuss limits at b. Assume that f is continuous at b. This means that for
any sequence (Xi) from D converging to b, f(Xi) -+ feb). By definition of
function limit (Section 3.2), f has a limit at b, and the limit is feb).
(b) => (c) Assume limx~b f(x) = feb).
Ifb isan isolated point of D, then there is some N(b, 8) in which bis the only
point of D. Hence xED with IIx - bll < 8 => x = b => IIf(x) - f(b) II =
o < e, no matter what e is.
Suppose instead that b is an accumulation point of D. Theorem 3.4 says that
given e > 0, there exists 8 such that Ilf(x) - f(b) II < e for all xED with
0< IIx - bll < 8. Clearly, we may drop the condition 0 < Ilx - bll.
(c) => (d) Part (d) is merely a translation into the language of neighborhoods
of what part (c) says in terms of inequalities.
(d) => (a) Assume part (d). Suppose (x;) -+ band N(f(b), e) is a neighbor-
hood of feb). By assumption, there is 8 such that any x E N (b, 8) n D has
f(x) E N(f(b), e). By convergence, there exists I such that i ~ I => Xi E
N(b,8) (Theorem 2.5). Hence i ~ I => f(Xi) E N(f(b), e). This shows
(Theorem 2.5 again) that f(x;) -+ feb); f is continuous at b.
6. (a) Assume that f is continuous at b. Then limx~b f(x) = feb) (Theorem

3.9). By Theorem 3.5(a), there are m and 8 with IIf(x) II < m for x =F b
in N(b, 8). Then M := m + IIf(b)II + 1 has IIf(x)1I < M for every x
in the neighborhood.
(b) The limit off(x) - y is feb) - y (linear combination) =F O. Theorem
3.5(b) guarantees some e1 and neighborhood N in which IIf(x) -yll >
e1, except maybe at b. Then IIf(x) -yll > e := min{e1,lIf(b) -YIl/2}
throughout N.
(c) Apply Theorem 3.9(c) to e := f(b)/2 (analogously if feb) is nega-
tive). Then in some neighborhood ofb,
If(x) - f(b)1 < e, so f(b)/2 = feb) - e < f(x) < feb) + e.
7. (a) BypropertiesofR,thereisanintegerjwithj::: lOs < j+l,andj =
lOs is ruled out by the irrationality of s. From j 110 < s < (j + 1)110,
we see that (j - 1)/10 and (j ± 2)/10, (j ± 3)/10, ... are more than
1110 away from s. Hence one or both of j 110, (j + 1) 1 lOis closest to
s, at distance 81 := min{s - j 110, (j + 1)/10 - s}.
(b) and (c) Same arguments as (a), beginning from j .::: 100s < j + 1,
etc., or j .::: J s < j + 1.
8. (a) If x = (x, y) is on the unit circle, then IIx - (3,0) 112 = ([x - 3f +
y2)1/2 > 3 - x unless y = 0, and 3 -x > 2 unless x = 1. Hence x :=
(1,0) gives the minimum IIx - (3,0)112 = 2, and this is necessarily
the infimum.
(b) As in (a), IIx - (3,0) 112 2: 2, but no point of the neighborhood gives
equality.However,xi := (1-I/i, O)isinN(O, I),and IIxi-(3, 0)112 =
2 + 1/ i. Hence inf Ilx - (3,0) 112 = 2.
(c) No and yes. First, let b := (1,0). From lib - (1 - I/i, 0)112 = I/i,
we judge that deb, N(O, 1» = O. Hence deb, N) = 0 does not imply
bEN. Second, let e E S. Then trivially dee, S) := inf{lIe - xII: x is
in S} = lie - ell = O.
(d) Assume S = {Xl, ... , xd. Then inf{lIb - xII: XES} = min{lIb-
xIII, ... , lib - xkll} = lib - xiII for some i. This is 0 iffb = Xi E S.
(e) In part (d), let b := .j2. The distance from b to T := {ijj E Q:
0< j < 101O0} is the same as the distance to the subset S := {i/j E
T: 0 < j < 10100 and j .::: i .::: 2j} (the ones between 1 and 2),
because the rest of T is more than 0.4 away from b. Since S is clearly
finite and b E S, part (d) tells us that ~ := deb, S) > O. Therefore,
i/j E T ::::} li/j - bl 2: deb, S) > .9~.
Section 3.5
1. Because IIG(Xi) - G(b)lIl = (GI(Xi) - GI(b»2 + ... +
(Gm(Xi) -
G m (b»2 .::: (Gj(Xi) - Gj(b»2 for each j, we see that G(Xi) -+ G(b)
as i -+ 00 iff G j (Xi) -+ G j (b) for each j. It follows that G is continuous
iff every G j is continuous. (Compare the argument in Theorem 3.13.)
2. We have seen that additive is subtractive: <l>(x - y) + <l>(y) = <l>(x) implies
<l>(x - y) = <l>(x) - <l>(y). If <l> is also of bounded magnification, then
II <l>(x) - <l>(y) II = II <l> (x - y)11 .::: Mllx - yll, and y -+ X forces <l>(y) -+
<l> (x); <l> is continuous.
3. K is a linear combination of functions that are separately continuous:

Kl (f) := -31 + 1 is of first degree; H(f) := 12 was proved continu-
ous in Co[O, 1] just below Theorem 3.14; and K2(f) := 1 3 is continuous
because II/i 3 - 1 3 110:= sup lii- Illf?+ l;f+ 1 2 1.::: llii- 1110(11 Ii 1102+
llii 110111110 + 11/1102) -+ 0 as Ii -+ I·
4. AhomogeneousmaphastomapOtoO:U(20) = 2U(0)forcesU(0) = O.
If U has unbounded magnification, then to each i there corresponds Xi with
IIU(xi)1I > ilixill·ThenYi :=xi/(iIiXill) (Why is Ilxili :j=.O?)hasYi -+ 0,
but IIU(Yi)11 = IIU(xi/[ilixillDll = IIU(xi)/[ilixiIIJII > 1, so that (U(Yi»
does not converge to U(O).
5. In any vector space, let b be a fixed nonzero vector. Then rex) := x + b is

of first degree, but not additive.
6. We assume that L is linear. (a) => (b) is trivial.

(b) => (c) Assume that L is continuous at b. Then Xi --+ 0 => Xi + b --+
b => L(xi) + L(b) = L(xi + b) --+ L(b) => L(xi) --+ 0 = L(O). That
makes L continuous at O.
(c) => (a) Assume that L is continuous at O. For any e, Xi
--+ e => Xi - e --+
o => L(xi) - L(e) = L(xi - e) --+ L(O). By Exercise 4, L(O) = O. Thus,
Xi --+ e => L(xi) - L(e) --+ 0 => L(xi) --+ L(e).
7. (a) Yes.Infaet,Tiseontraetive:Letmax/(x) = I(a),maxg(x) =g(b);

then T(f) - T(g) = I(a) - g(b) .:::: I(a) - g(a) .:::: III - gllo and
T(f)-T(g) ~ I(b)-g(b) ~ -1/(b)-g(b)1 ~ -1I/-gllo·Henee
IT(f) - T(g)1 .:::: III - gllo·
(b) No. For the functions hk discussed just above Theorem 3.15, hk --+ 0,
but T(hk) = hk (~) = 1.
8. No to (c), yes to the others. U is clearly linear, so we look at its magnification.

We have IU(f(x»1 = I/(x)llgo(x)1 for every x. Taking sups, we find that
IIU(f)lIo .:::: IIgoliollfllo, so that (a) U has bounded magnification on Co.
Doing integrals instead, we obtain either
IIU(f) 112:= ( fo I
[f(x)go(x)]2 dx
) 1/2
.:::: (
Ilgoili fo Il(x)2 dx)IP = II gO 110 II f112,
meaning that (b) U has bounded magnification on C2; or
I ) 1/2
IIU(f)112':::: ( 11/110210 g(x)2 dx = IIgo1l211/110,
which says that (d) U is bounded as a mapping from Co to C2. To get a

counterexample for (c), look at the functions hk defined above Theorem
3.15. They have IIhkll2 ~ ./3/2k/2 and IIhkllo = 1. Setting go := 1, we
obtain IIU(hk) 110 ~ 2k/2/./3l1hkIl2' and U is unbounded from C2 to Co.
Chapter 4
Section 4.1
1. (a) Yes. I is a composite of continuous functions, so it is continuous on
the box. By Theorem 4.3, it attains a maximum.
(b) No. g is not even bounded above: Along the line y = x, the value
g(1/ i, 1/ i) is 2 exp(i 2) / i 2, which tends to 00 as i --+ 00.
(c) Yes. h is undefined on the two axes, but its limit is 0 at each place
there: Near (a, 0), Ixy sin(l/xy)1 :::: Ixyl :::: (Ial + 1)Iyl -+ 0 as
(x, y) -+ (a, 0); and similarly near (0, b). Hence h can be extended
to a continuous h+, defined throughout the box. This h+ must have a
maximum h + (b) in the box. Point b is not on the axes, because h + = 0
on the axes, whereas h+ has such positive values as h+(1, 1) = sin 1.
Therefore, b is one of the places where h is defined. Then clearly, h (b)
is maximal.
2. (a) The biggest (x 2+ y2) 1/2-such a maximum must occur--comes when
Ixl and Iyl are maximal. The box is defined by -3 :::: x :::: 1,3 :::: y :::: 4.
Therefore, Ixl < 3, except where x = -3, and Iyl < 4, except where
y = 4. The farthest place is (-3,4).
(b) Generalizing (a) in R2: In [a, b], we have al :::: x :::: bl. If 0 :::: aI,
then Ix I < bl, except where x = bl; if bl :::: 0, then Ix I < -aI, except
wherex = -al; and if al < 0 < hI, then Ixl < max{ -aI, hI}, except
where x is the one of aI, bl farther from O. Similarly for y. Thus, the
outlying place has coordinates (al or bl, a2 or b2). Those name the
four "corners."
The argument extends to the 2n corners in Rn.
3. On [0, 00) (the first octant plus its edges):
(a) x 2 + y2 + Z2 is unbounded;
(b) tan- I (xy) sin Z is between -rr /2 and rr /2, but cannot reach either;
(c) x 2 is nonuniformly continuous. Set c := 1. No matter what 8 you
name,o := (1/8,0,0) and v := (1/8 + 8/2,0,0) have 110 - vl12 < 8,
but (1/8 + 8/2)2 - (1/8)2 = 1 + 82/4> c.
4. (a) For i ~ 2, Xi := a + (b - a)/ i is in (a, b), but Xi -+ a, which is not.

(b) I(x) := 1/lix - al12 is a continuous composite defined in (a, b) and
tending to 00 as X -+ a.
(c) g(x) := Ilx - allz can approach 0 but not reach it; and, because the
diagonal is the longest segment in the closed box, g can get near to,
but not as large as, lib - a1l2.
(d) The function in (b) is nonuniformly continuous, because for the se-
quence (Xi) in (a), the distances Ilxi - Xj II become small, but the value
differences I (Xi) - I (X j) get large.
5. (a) trivial.
(b) If each a j :::: b j and b j :::: a j, then a j = b j for every j, and a = b.
(c) If each a j :::: b j and b j :::: Cj, then a j :::: Cj for every j, and a :::: c.
6. "a is not to the left of b" denies that every a j :::: bj . Hence it means that some
ak exceeds the corresponding bk. Sinceak > bkonoxcanhaveak :::: Xk :::: bk.
This last is required for a :::: X :::: b.
Section 4.2
1. (a) Closed and unbounded. Closed because distinct members ofZ2 are.fi
or more apart. Hence any Cauchy sequence from Z2 must eventually
stabilize: (Xi) -+ X forces X = XI = X/+ I = ... E Z2. Unbounded
because (i, i) -+ 00; apply Theorem 4.6(b).
(b) Not closed, because (Iii, Iii) -+ 0 I/. I/Z2. Bounded, because
1I(1lx, Ily)1I2 ::: vItTI.
(c) Closed, because (Xi, 0) -+ (X, y) forces y = 0 (Theorem 2.8). Un-
bounded, because (i, 0) -+ 00.
(d) This part is deceptive, because the graph "approaches" an asymptote.
It is nevertheless closed: If (Xi, exp(Xi» -+ (X, y), then by Theorem
2.8 Xi -+ X and exp(Xi) -+ y; since !(s) := eS is continuous, the last
forces y = exp(x), which says that (x, y) is on the graph. The graph
is unbounded, because (i, ei ) -+ 00.
(e) Not closed, because (Iii, 1) -+ (0,1), which is not on the graph.
Unbounded, because (i, 1) -+ 00.
2. (a) This set T + T is bounded iff T is. If T is bounded, then Ix + yl :::

Ix 1+ Iy I ::: 2 sup Ix I, making T + T bounded. If T is unbounded, then
some (Xi) from T tends to 00, and so does (Xi + XI) from T + T.
(b) Same argument, with Ix - yl and sequence (Xi - x[).
(c) Same argument, with xy and sequence (XiXI), where XI is any nonzero
term in (Xi). Such terms must exist when (Xi) -+ 00. Note that we are
unworried about the sign of the product.
(d) New argument is essential: TIT is bounded iff T is bounded and does
not have 0 as accumulation point.
=} If either T is unbounded (some (Xi) -+ (0) or 0 is an accumulation
point (some (Yi) -+ 0 with Yi f=. 0), then xi/xI -+ 00 for appropriate
lor yt!Yi -+ 00.
<= If there are £ and M such that £ < IY I < M for those Y f=. 0 in T,
then Ix I Y I < M I £for xl YET IT.
3. Xi E (-00, b] means thatJl"[ (Xi) ::: bl , ... , Jrn (Xi) ::: bn . If (Xi) -+ C, then
Jrk(Xi) -+ Ck for each k (Theorem 2.8), forcing Ck ::: bk, or C E (-00, b].
Hence (-00, b] is closed. Similarly for [a, (0).
4. (a) Suppose (Xi) is a sequence from S := {ai, ... , ad. Necessarily some
member, say ai, is repeated infinitely often. Thus some subsequence
(Xj(i) has Xj(i) = ai, and converges to al.lfnow (Xi) -+ X, then
X = al E S (Theorem 2.4(a»; S is closed. The converse is false, since
R n is closed and infinite.
(b) If (Xi) comes from B(b, 8) and (Xi) -+ X, then IIx - bll = lim IIXi - bll
(Theorem 2.4(b» ::: 8, so that X E B(b, 8). The ball is closed.
(c) Similar argument with lim IIXi - bll = 8.

(d) Suppose that S and T are closed. If(Xi) comes from snT and (Xi) ~ X,
then Xi E S forces XES and Xi E T forces X E T, so XES n T; the
intersection is closed. If instead (Xi) comes from S U T, then it cannot
be that each of Sand T contributes finitely many terms; that is, a
subsequence (Xj(i) must come from just one of Sand T. Since (Xj(i)
must converge to X also, we have X in the same one, and XES U T.
The union is closed.
5. (a) We have seen (Exercise 2 in Section 3.4) that f(x) := IIx - bll2 is a
continuous function on S. By Theorem 4.8, it has a minimum value
f(c). Then deb, S) := inf{lIx - bib: XES} = IIc - bll2.
(b) Yes, there must be a maximum fed) = lid - bll2.
(c) G, ~).
Using vectors: Suppose x is on the circle. Let e be the angle between x
and (3, 4). By the law of cosines, IIx- (3,4) Iii = IIxlli+ II (3, 4)lIi-
211xII211 (3,4) 112 cos e = 26-lOcos e. Least value occurs when e = o.
Necessarily (Theorem 1.4(d» x = t(3, 4) for a positive t, which then
has to be ~.
Using calculus instead: d 2 := (3 - cos u)2 + (4 - sin u)2 is minimal
when tan u = 1,which describes the closest point (~, ~) and the
farthest (-~, -~).
(d) Your sketch will make the answer obvious; here is an analytical proof.
For any x in the ball, let e be the angle between x - a and b - a. By
the law of cosines (compare (c», IIx-blii = IIx-alli+llb-alli-
211x - all211b - a 112 cose = (lib - all2 -lix - a1l2)2 +211x - a11211b-
aIl2(1- cos e). Clearly, e = 0 is necessary for a minimum, so we need
x - a = t(b - a), or x := a + t(b - a).
If b is outside the ball, then lib - all2 > 8 :::: IIx - alb for every x,
so the minimum comes with IIx - all2 = 8, forcing t = 8/lib - a1l2.
That describes the place a + 8(b - a)/lib - all2 where the segment ab
crosses the sphere.
Ifb is in the ball, then obviously we should make IIx-all2 = Ilb-aI12.
That gives t = 1 and x = b, which makes sense: b is closest to b.
(e) Pick a fixed c E S. Let T := B(b, lib - C1l2) n S, which includes c.
Clearly, every point of S outside T is more than Ilb-cll2 from b. Hence
deb, S) := inf{llx - b112: XES} = inf{lly - b112: YET} = deb, T).
Since T is closed (Exercises 4b, d) and bounded (contained in a ball),
part (a) applies. Thus, there exists dE T with deb, S) = lib - dll2-
There need not be a farthest point: in R2, no point of the x-axis is
farthest from (0, 1).
6. {oJ is closed (Exercise 4a), but its set of accumulation points is 0.

7. (a)::::} (b) Suppose every XES satisfies IIxll ::: M. Then clearly, every term
of (Xi) from S satisfies the same inequality.
(b) ::::} (c) Assume that no neighborhood of the origin contains S. Then for
each i, there exists Xi E S outside of N (0, i). Since IIXi II > i, the sequence
(Xi) is unbounded.
(c) ::::} (d) trivial.
(d)::::} (a) Assume S ~ N(b, r). Since N(b, r) ~ N(O, IIbll +r) (Theorem
2.2(c», we have S ~ N(O, IIbll + r). Consequently, every XES satisfies
IIxll < IIbll + r.
(a) ::::} (e) Assume IIxlll := x l2 + ... + xn2 ::: M2 for all XES. Then
-M ::: Xk ::: M for each k, and X E [(-M, ... , -M), (M, ... , M)]. Hence
S is a subset of a box.
(e) ::::} (f) Let S ~[a, b). Then ak ::: Xk = 1Tk(X) ::: bk for each k and every
XES. Hence 1Tk(S) := (1Tk(X)} is bounded.
(f) ::::} (a) Assume that 1Tk(S) is bounded, say l 1Tk (x) I ::: Mk for all XES.
Then x E S::::} x/ + ... +xn2 ::: M/ + ... + Mn2, and S is bounded.
8. part (b) Let M := sup{f(x): XES}, which is finite by part (a). By definition
of sup, there is a sequence (f(Xi» ~ M. Because S is bounded, BWT
applies, and there is a subsequence (x j (i) converging to some b. S is closed,
so b E S. Since (f(Xj(i») must converge to the parent limit, we have
M = lim f(xj(i», which is f(b) by continuity. Hence the sup of f is a
value f(b), making f(b) the maximum. Similarly for min.
(c) Suppose there were an E > 0 for which no closeness 8 guaranteed

function differences less than E. Thus, for each i, there would be Xi
and Yi with IIxi - yilI2 < Iii, yet having IIf(Xi) - f(Yi)ll2 2: E.
From (Xi), find a subsequence (Xj(i» ~ b. Necessarily b E Sand
(Yj(i» ~ b. By continuity, f(xj(i» ~ f(b) and f(Yj(i» ~ f(b).
But then IIf(xj(i» - f(Yj(i»1I2 ~ 0, contradicting the function-
difference inequality.
9. (a)::::} (b) Assume (a), and let (Xi) be a bounded sequence. If {Xl, X2, ... } is
finite, then at least one Xk is repeated infinitely many times in the sequence.
That Xk qualifies as a sublimit. If instead {Xl, X2, ... } is an infinite set, then
it fits the hypothesis of (a), so it has an accumulation point b. N (b, 1) must
have infinitely many members of the set (Exercise 6b in Section 3.2). Call one
of them Xj(1), and write rl := lib - Xj(l) 11/2. N(b, rl) must have infinitely
many members; among them, there must be Xj(2) with j(2) > j(I). Set
r2 := lib - Xj(2) 11/2, and continue the selection process. Then (Xj(i» ~ b,
and b is a sub limit.
(b) =} (a) Assume (b), and suppose S is infinite and bounded. By the stated
definition, it is possible to find a sequence (Xi) from S of unequal vectors.
By Theorem 4.6(b), (Xi) is bounded. By hypothesis, it must have a sublimit
e = lim Xk(i). For each neighborhood N(e, 8), there is I such that the terms
Xk(l), Xk(l+l),··· are in N(e, 8). These terms being unequal, they constitute
an infinity of members of Sin N(e, 8). We conclude thate is an accumulation
point of S.
Section 4.3
1. Each function in Example 1 is a composite of xy (polynomial in R2), absolute
value, exponential, and inverse tangent (all continuous in R); that makes
them continuous.
The rangeofxy is (-00,00). Therefore, in (a), tan- 1 xy ranges over (-n /2,
n/2), without extremes, and I tan- 1 xyl ranges over [0, n/2). Similarly in
(b), eXY has range (0, 00), never hitting 0, and e1xyl has range [1, 00), value
= 1 at O.
2. Same words as in the solution to Exercise 8c in Section 4.2, with the conclu-
sions in the third and fourth sentences justified by sequential compactness.
3. (a) Continuity is automatic at isolated points.

(b) You have to have arbitrarily close points. Let T := {(1,0), (~,O),
G, 0), ... } ~ R 2 . (1/ i, 0) is isolated, because no other point of T is
within 1/ i -1/ (i + 1) of it. f(x, y) := I/x is continuous on T, but not
uniformly: Set E := 1, and see that no matter what 8 is, there are points
with II(I/i, 0) - (I/j, 0)112 < 8 but If(I/i, 0) - f(I/j, 0)1 >. E.
4. Since IIx - bll is a continuous function ofx E S, it must achieve greatest and
least values (Theorem 4.11). Those values occur at the farthest and closest
points.
5. Fix X and y. For Z E S, Ilx - zil s IIx - yll + Ily - zll. Hence d(x, S) :=
inf{lIx - z: z E S} s IIx - yll + inf{lly - zll: z E S} = Ilx - yll + dey, S).
Therefore, d(x, S) - dey, S) s IIx - YII, and the absolute value inequality
follows by symmetry.
6. By definition ofinf, there are sequences (Xi) from Sand (Yi) from T such
that IIXi -Yi II -+ deS, T). Let Sand T be sequentially compact. Then there
is a subsequence (Xj(i») -+ a E S and corresponding (Yk(j(i») -+ bET.
From IIXk{j(i» - Yk{j(i» II -+ lIa - bll, we judge that deS, T) = Iia - bll
(mUltiple uses of Theorem 2.4).
Alternative proof: By Exercise 5, d(x, T) is contractive on S, and therefore
continuous. Hence there is a minimum value dCa, T).
By Exercise 4, there is bET with dCa, T) = lIa - bll. Now take any XES,
YET. Then IIx - yll ~ d(x, T) ~ dCa, T) = lIa - bll. Of necessity,
deS, T) := inf IIx - yll = Iia - bll·
7. (a) If x, Y E B(b, r), then Ilx - yll ~ IIx - bll + lib - yll ~ 2r. Since
the points b ± ru, for any unit vector u, are in the ball 2r apart, we
conclude that sup Ilx - YII = 2r.
(b) The diameter d is finite, because S has to be bounded (Theorems 4.11
and 4.10). Let IIxi - Yi II -* d. As in Exercise 6, take subsequences
(Xj(i)) -* a, (Yk{j(i))) -* b. Then IIxk{j(i)) - Yk{j(i)) II tends to both d
and lIa - bll, so these last two must be equal.
8. One example fits all: S := quadrant I in R2, b := 0, T := {OJ.
9. (a) Since d(x, T) is contractive (Exercise 5) on S, it has a minimum value

dCa, T). That makes a as close to T as a member of S can get. Also,
XES, YET => IIx - yll ~ d(x, T) ~ dCa, T), giving deS, T) ~
dCa, T); while dCa, T) := inf lIa - YII ~ inf IIx - YII (inf of a smaller
set is larger) = deS, T).
(b) In (a), we found a with deS, T) = dCa, T). Were this distance 0, there
would be (Yi) from T such that lIa - Yi II -* 0; that would make a
a closure point of T, forcing a E T, violating S n T = 0. Hence
dCa, T) > O.
(c) S:= x-axis and T := {(x, y): y = eX} are both closed (Exercise Ic,d
in Section 4.2), but have points in arbitrarily close proximity: (-i, e- i )
on the graph is II ei from (-i, 0) on the x-axis.
(d) S:= {OJ and T := quadrant I in R2.
(e) By (a), there is a with deS, T) = dCa, T). By definition, there is a
sequence (Yi) in T with lIa - Yi II -* dCa, T). (Yi) has to be bounded,
because IIYi 112 ~ lIall2 + IIYi - a1l2, and the sequence on the right
converges. Since we are in Rn, some subsequence (y j(i)) -* bET,
and deS, T) = dCa, T) = lib - a1l2. This says that lib - all2
inf IIx - YII2 ~ inf Ilx - bib and so b is closest to S.
Section 4.4
1. (a) Open: If e := (a, b) is in quadrant I, so that a and b are positive, let r :=
min{a, b}; then N(e, r) s;: quadrant I, because (x, y) E N(e, r) =>
Ix - ai, Iy - bl both ~ (x - a)2 + (y - b)2 < r2 => x > 0, y > O.
Not closed: (11 i, II i) -* 0, which is not in quadrant I.
(b) Not open: (1, 1) is on the graph, but N((l, 1), r) has the point (1,1 +
rI2), which is not. Closed: Suppose Xi := (Xi, I/xi) is a sequence
from the graph converging to (a, b); then Xi -* a and Ilxi -* b; a
cannot be 0, because then 11 I Xi I would approach 00; by continuity
I/Xi -+ l/a; necessarily b = l/a, which says that (a, b) is on the

graph.
(c) Not open: (1, 0) (rectangular or polar) is on the graph, but (1 + r, 0)
is not for 0 < r < e2:n: - 1. Not closed: Xi := polar (e- i , -i) -+ 0,
which is not on the graph.
(d) Not open: (1/2n,0) is on, but (1/2n, r/2) is not. Not closed:
(1/ in, 0) -+ (0,0), and no point with x = 0 is on the graph.
2. (Compare the solution to Exercise la. Also, note that (a, 00) is not the
complement of (-00, a].)
Assume C E (a, 00), so that a < c. Let r be the least of C! - aI, ... , en - an.
Then X E N(c, r) ::::} each IXk - ql < r ::::} each Xk > q - r ~ ak ::::}
x E (a, 00). Hence N (c, r) ~ (a, 00), and the latter is open. Similarly with
c E (-00, b).
3. (Compare the solution to Exercise 6 in Section 4.1.) a < b means that each
ak < bk; to deny it is to say that some a j ~ b j. In that case, no x can satisfy
a j < x j < b j, so no x has a < x < b.
4. For every x E V, N (x, 1) ~ V trivially, so V is open. At the other extreme,

"x E 0 ::::} N (x, 1) S; 0" is true vacuously.
5. Only the empty set. If 0 is open and bE 0, then some N(b, 8) ~ 0, and
o has all the vectors b + (8/i)b/llbll, i ~ 2.
6. (a) Assume T S; W. In V, x r- I (T*) {} rex) E T* {} rex)
E fi T {}
x fi r- 1 (T) {} x E r- Hence r- I (T*) = r- I (T)*.
I (T)*.
(b) r- I (open) = open comes from Theorem 4.18 (whose proof, to be sure,
depends on Exercise 9 below). Suppose instead T is closed. Then T*
is open (Theorem 4.14, whose proof is Exercise 7). By Theorem 4.18,
r- I (T*) is open. By (a), r- I (T*) = r- I (T)*. Hence r- I (T) is closed.
7. ::::} Assume that 0 is open. Suppose (Xi) is a sequence from 0* converging
to b. If b were in 0, there would exist N (b, 8) ~ 0, so that (Theorem 2.5)
the Xi would eventually get into N (b, 8) and thereby into O. Hence b must
be in 0*. We have shown that 0* is closed.
{=: Assume that 0 * is closed. Let b EO. By Theorem 4.7, b is not a closure
point of 0*. By definition, there exists N (b, 8) containing no points from
0*. Necessarily N (b, 8) ~ 0, proving that 0 is open.
8. Let bED, c > O. The set 0 := N(r(b), c) is open, so by hypothesis there

exists an open P such that D n P is r- I (0). Since b E r- I (0) ~ P, there
is N(b, 8) ~ P. If now X E N(b,8) n D, then rex) E O. By Theorem
3.9(d), r is continuous at b.
9. (a) Ifx E OnP,bothopen,thenthereareN(x,r) ~ OandN(x,s) ~ P.

Let t := min{r, s}. Clearly, N(x, t) is contained in each of 0 and P,
so this neighborhood of x is a subset of 0 n P. Hence 0 n P is open.
(b) Same argument with t := min{Tj, ... , Tk}.
10. By one of DeMorgan's laws, (CI U··· U Ck)* = Cj n··· n C k. If each

Cj is closed, then each Cj is open (Theorem 4.14 = Exercise 7), so their
intersection is open (Exercise 9b). From (CI U ... U Ck)* being open, we
conclude that the union is closed.
Section 4.5
1. (a) Not connected, therefore not arc-connected. (0,0) is in the set, but
not (0, 1), even though y = sin x/x -+ 1 as x -+ O. Accordingly,
01 := N (0, IT /6) and 02 := {x: IIxll2 > IT /6} disconnect the set.
(b) Arc-connected, therefore connected. You would think it impossible to
get from (1, 0) (polar () = 0, r = 1) to the origin (polar () = -00, r =
0) along the spiral, since the distance appears to be infinite. Without
deciding whether that appearance is reality, we merely remark that you
can cover infinite distance in finite time by walking fast. In this case,
set() = tant,r = elant (rectangular x = rcos() = elantcos(tant),
y = r sin () = e lant sin (tan t)) for -IT /2 < t < IT /2. Clearly, the
mapping t r-+ (x, y) is continuous, stays on the spiral, covers the
whole spiral, and has x -+ 0, y -+ 0 as t -+ -IT /2. Therefore, we can
extend it by setting x := 0, y := 0 when t := -IT /2. The resulting
extension maps [ -IT /2, IT /2) continuously onto the set, so that we can
use it to define an arc from any point to another.
2. (a) Qn does not have the IVP, so it is not connected (Theorem 4.21).
For example, Ilxll} is continuous on Qn, has values IIOII} = 0 and
11(2,0, ... , O)II} = 4, but never reaches,J2.
(b) In R I, the irrationals are disconnected by 01 := {x > O} and 02 :=
{x < OJ. In R n , the not-all-rationals (complement of Qn) are arc-
connected. To illustrate, you travel from (,J2, y, z) to (x, ,J3, u) along
the segments from (,J2, y, z) to (,J2, ,J3, z) to (x, ,J3, z) to (x, ,J3, u),
staying in (Qn)*; and from (,J2, y, z) to (J5, v, w) along the seg-
ments from (,J2, y, z) to (,J2,,J3, z) to (J5,,J3, z) to (v'S, v, z) to
(v'S, v, w).
3. Projections are continuous (Theorem 3.8, among other places). By Theorem
4.23, the projected image of a connected set is connected. The converse is
false: In R 2, the union of the two lines y = x and y = x + 1 is a disconnected
set whose projections are connected.
4. 0 is trivially connected, because you cannot nontrivially partition it. V is

connected for many reasons; for example, it is convex: If a, b E V, then the
segment ab ~ V.
5. => Let 01 and 02 disconnect S. Write Cl := OJ, C2 := O~; these are

closed, by Theorem 4.14. Since 01 n Sand 02 n S partition S, we have
x E Cl n S {} XES and x ~ 01 {} X E 02 n S. Thus, Cl n S = 02 n S.
Similarly C2 n S = 01 n S. Hence {Cl n S, C2 n S} is a partition of S.
{= Same argument with C's interchanged with O's.
6. Suppose S is neither 0 nor V. Then Sand S* are both nonempty, so {S, S*}
is a partition of V. Since V is connected (Exercise 4), Sand S* cannot both
be open. Hence either S is not open, or S* is not open, in which case S is
not closed.
7. Let S be arc-connected. According to Theorem 4.19, we need to show that

S has the IVP. Suppose f is continuous on S, a and b E S, and f(a) <
Y < f(b). By the arc-connectedness, there exists a continuous function g
mapping some interval [r, s] to S with g(r) = a, g(s) = b. Then f(g(t» is
continuous on [r, s] and has f(g(r» < y < f(g(s». By the IVT for real
functions, there is a place u E [r, s] where f(g(u» = y. Set c := g(u), and
we have c E S with f(c) = y. Hence S has the IVP.
8. (a) f(t) := b, 0 S t S 1, defines an arc from b to b within S.

(b) If S is open and bE S, then there is N(b, 8) ~ S. If now x E N(b, 8),
then g(t) := (1 - t)x + tb, 0 S t S 1, defines an arc from x to b
within N(b, 8) (Theorem 2.1(b», and therefore within S.
(c) See the proof of Theorem 5.5(b).
9. (a) Assume that S is star-shaped and x, YES. By hypothesis, the segments

bx and by are subsets of S. As in Exercise 8b, these are arcs from b to
x and from b to y. As in 8c, xb is an arc from x to b. Hence we have
an arc from x to b and one from b to y. See the argument in Theorem
5.5(b) to establish an arc from x to y.
(b) No; the unit circle in R2 is itself an arc, but contains no line segments.
(c) No; draw a "star." Alternatively, look at the union of the two axes in
R2.
10. (a) Let x := alXl + ... + akXk and y := {31Yl + ... + {3jYj E CH(S).
The segment xy has vectors (1 - a)x + ay = (1 - a)alxl + ... +
(1 - a)akxk + a{31Yl + ... + a{3jYj· In this last combination, the
vectors are from S and the coefficients are nonnegative and sum to
(1-a)(al + .. ·+ak)+a({31 + .. '+{3j) = 1. Hence (1-a)x+ay E
CH(S), proving that CH(S) is convex.
(b) Ifx E S, then Ix E CH(S). Hence S ~ CH(S); the latter is one convex
(part (a» superset of S. Assume that T is any convex superset of S.
Then T has the convex combinations IXI of a single vector from S.
Suppose T includes all combinations of k - 1 or fewer vectors. Then
aIXI + ... + ak-IXk-I

x := aIXI + ... + akXk = (1 - ak) + akXk
(1- ak)
is on the segment from one vector in T to another, unless 1 - ak = 0,

in which case x = Xk E S. Either way x E T, since T is convex. By
induction, T possesses all the members of CH( S). Hence CH( S) ~ T,
and the hull is the smallest convex superset of S.
(c) {:::: If S = CH(S), then S is convex by part (a).
:::} If S is convex, then necessarily S is the smallest convex superset of
S; by (b), S = CH(S).
(d) The combination IXI consists of both endpoints of a degenerate seg-
ment. The combination aIXI + a2x2lies on a segment with one end-
point in S and the other also in S. The combination aIXI + a2X2 +
a3x3 = (1 - (3)(al [1 - a3]-I xi + a2[1 - a3]-I x2) + a3x3 lies on
a segment with one endpoint X3 in S and the other (al [1 - a3]-I xi +
a2[1- a3]-I x2) on a segment with one endpointx2 in S and the other
Xl also in S ....
(e) No, no, yes. The convex hull of a neighborhood or ball is the neighbor-
hood or ball, by part (c) and Example 3(a). A hull has to be connected;
being convex, it is arc-connected.
Section 4.6
1. (a) Positive-definiteness: Since IIxlli ~ each Ix} I, IIxlli = 0 forces x} = 0
for all j.
Radial homogeneity: lIax 111 := laxil + ... + laxnI = la I(ixil + ... +
Ixnl) = lalllxlII.
Subadditivity: IIx+ylli := IXI + YII + ... + IXn + Ynl ~ IXII + IYII +
... + IXnl + IYnl = IIxlli + lIyllI·
(b) Clearly, Ix}1 ~ (x? + ... + Xn2) 1/2 ~ (lIxlI?)1/2. Hence IIxllI ~
nllxII2 ~ nllxlIl. Set m := 1, M := n.
2. (a) Recall (Exercise 1 in Section 1.6) that in C2[0, 1], these sines and
cosines are orthogonal and have length I/.J2. Hence II f 112 = (a 2 /2 +
p2 /2)1/2. We can find the extremes of f
by calculus or trigonometry:
f(x) = (a 2 +p2)I/2 sin(2JTx+e),wheree:= ±cos- I (a/[a 2 +
p2]1/2) (same sign as (3), yielding sup If I = (a 2 + (32)1/2. Accord-
ingly, IIfllo = .J211f112.
(b) If (fi) converges uniformly to /, then Ii -+ / in Co[O, 1]. Since V,

being two-dimensional, must be a closed subset of Co[O, 1] (Theorem
4.30), we must have / E V.
3. The basic principle here is that the cubic polynomials form a 4-dimensional
subspace.
(a) By hypothesis, lllilil -+ 28. Necessarily (II Ii 112) is a bounded se-

quence of reaIs, so (Theorem 4.29(a» (II/illo) has to be bounded. If,
say, II Ii 110 < M, then IIi(x)1 < M for each i and every x.
(b) We just saw that (fi) is a bounded sequence in a 4-dimensional
subspace of Co[O, 1]. Hence some subsequence converges (Theorem
4.27(c» in Co[O, 1]; that is, some subsequence converges uniformly.
4. From the inequality, we conclude that (Xi) -+ X relative to II lIa <=> IIXi -
xlla -+ 0 (definition) <=> IIXi - Xllb -+ 0 <=> (Xi) -+ X relative to II lib.
5. We use the characterization in Theorem 4.25.

Reflexivity: ll1xlla S IIxlia S ll1xlla, so II lIa is equivalent to itself.
Symmetry: If mllxlla < IIxlib S Mllxll a, then M-Illxllb S IIxlia <
m-Illxllb'
Transitivity: If mllxlla < IIxlib S Mllxll a and kllxllb S IIxlic S Kllxllb,
then mkllxlla S IIxlic S M Kllxll a.
6. Write IIxll for the norm in V and IIxliB for the "Pythagorean norm," as defined
in the proof of Theorem 4.26. By Theorems 4.25 and 4.26, there are m and
Mwithmllxll S IIxliB S Mllxll for all x.
(a) :::::} If (Xi) converges to y := alVI + ... + anvn , then for each k,
lak-TIk(xi)f S (al-TII(xi»2+. +(an -TI n (Xi»2 = IIY-Xi IIi S
M 2 11y - Xi II~ -+ O. This says that each sequence (TIk(Xi» converges
to ak = TIk(Y).
{::: If, conversely, each coordinate sequence (TIk(Xi» converges to 13k.
then z := 131 VI + ... + f3nvn has liz -Xi II = II (131 - TIl (Xi»VI + ... +
(f3n- TI n(xi»)vn ll S If3l- TI I(Xi)llIvIII+·· ·+lf3n- TI n(Xi)llIvn ll-+
0, and (Xi) converges to z.
(c) FromITIk(Xi)1 S IIxillB S Mllxill,weconcludethatif(xi)isbounded,
then so is each sequence (TIdXi». As in the proof ofBWT (Theorem
2.10), we extract a subsequence (Xj(i» for which all the coordinate
sequences (TIk(Xj(i)) converge, and their limits are the coordinates
of a finite sublimit for (Xi).
7. Write II lIa and II lib for two norms in V. We now know there are m and M
with mllxlla S IIxlib S Mllxlla.
(a) If (Xi) -+ X relative to lilia, meaning that IIx - Xi lIa -+ 0, then

IIx - Xi lib S Mllx - Xi lIa -+ 0, and (Xi) -+ X relative to II lib·
(Conversely, by symmetry.)
(b) Case 1: b is isolated relative to II Ila. Thus, no member of D has

IIx-bli a < o.ThennomemberofDhasllx-blib < mo. Consequently,
b is isolated relative to II lib, and under both norms, the limit is feb).
Case 2: b is an accumulation point of D relative to II lIa. First, there is
a sequence (Xi) from D with 0 < II b - Xi II a -+ 0, and the limit of f is
c:= limf(Xi). By part (a), 0 < IIb-xillb -+ 0; that is, some sequence
from D converges to b, without reaching b, relative to II lib. Next, let
(Yi) be another such sequence. Then also lib - Yi lIa -+ O. Because f
has an a-limit, we have f(Yi) -+ c. We have established that for every
(Yi), the value sequence f(Yi) has limit c. It follows (Section 3.2) that
f has a b-limit matching its a-limit c.
(c) The conclusion says that Xi E D and lib - Xi lib -+ 0 imply IIf(b) -
f (Xi) II w -+ O. Its proof is immediate, because lib - Xi lib -+ 0 forces
lib - Xi lIa -+ O.
8. (a) Assume that S is a-closed, meaning that S possesses the a-limit of any
sequence from S. Thus, Xi E Sand IIx - Xi lIa -+ 0 =} XES. Since
IIx-xi lib -+ oforces IIx-xi Iia -+ o(Theorems 4.26 and 4.24), we see
that S possesses the b-limit of any sequence from S, and S is b-closed.
The conclusion about boundedness follows from IIxllb S Mllxlla.
(b) =} If S ~ V has the EVP, then it is closed and bounded by Theorem

4.10.
{= Assume that S is closed and bounded. Let (Xi) be a sequence from
S. It has to be bounded, so by Theorem 4.26(c), there is a convergent
subsequence. Because S is closed, the limit of the subsequence is in
S. Hence S is sequentially compact. By Theorem 4.11, S has the EVP.
(c) Suppose S is closed and bounded and f is continuous there. By (b), S

is sequentially compact. By Exercise 2 in Section 4.3, f is uniformly
continuous.
9. Let L: V -+ W, with norms II Ilv and II Ilw. By theorems from linear

algebra, the range L(V) is a finite-dimensional subspace of W. Let B :=
{VI, ... , vn } and C := {WI, ... , w m } be bases for V and L(V). There are
scalars CXjk with L(vk) = CXlkWI + ... + CXmkWm for each k. For any X =
th VI + ... + f3n vn , we then have

IIL(x)lI~ :::: M2I1L(x)IIE (by equivalence of norms in L(V))
= M 211f3I(allwl + ... +amIWm)
+ ... + f3n(al nwl + ... + amnWm) liE
:= M2 ([f3!a!! + ... + f3na!n]2 + ... + [f3!a ml + ... + f3namnf)
:::: M2([f3?+ ... + f3n2] [ad + ... + al;]

+ ... + [f3? + ... + f3n2] [ad + ... + al;])
by Cauchy's inequality. Writing K2 for the sum of all the a /f, we have
IIL(x)lI~ :::: M2[f312 + ... + f3n2]K2 = M2K211xll§ :::: M2K2m2I1xll~,
the last by equivalence of norms in V. This says that L is a bounded linear
map from V to W. By Theorem 3.16, L is continuous.
10. (a) =::} (b) If the norms are equivalent, then there are M and m with m IIxlia ::::
IIxlib :::: Mllxll a, so that IIxlla/llxllb :::: 11m and Ilxllblllxlla :::: M are both
bounded.
(b) =::} (c) If Ilxlla/llxllb :::: m and IIxllblllxlla :::: M, then II M :::: IIxlla/llxllb ::::
m says that the fraction is bounded and bounded away from zero.
(c) =::} (d) Assume 0 < 8 :::: IIxlla/llxllb :::: B. Then IlIxlial :::: Bllxllb says that
the a-norm is a function of bounded magnification relative to the b-norm;
and analogously IlIxllbl :::: (1/8)lIxll a ·
(d) =::} (e) Assume that the a-norm is a function of bounded magnification
relative to the b-norm, so that Ilxlia :::: Mllxllb.1f (Xi) -+ X in the b-norm,
then IlIxli a -llxi lIal :::: IIX-Xi lIa (difference of the norms) :::: Mllx-Xi lib -+
O. This says that IIxli a is a continuous function under the b-norm. Similarly
for the opposite direction.
(e) =::} (a) Assume that each norm is continuous relative to the other. Since
IIxlia is b-continuous at the origin, there exists 8 such that IIx - Ollb <
8 => Illxli a - 1I01la I < 1. Therefore, for every X :I 0, since the vector
y := 8x/(2I1xllb) satisfies lIyllb < 8, we know that lIylla < 1, or Ilxli a <
(218)lIxlib' Clearly, also 1I01la :::: (218)1I0Iib. Similarly, we find B such that
Ilxlib :::: (2IB)lIxll a , and the norms are equivalent (Theorem 4.25).
Chapter 5
Section 5.1
1. (a) This is just a consequence of continuity. Since the x-value 1TI (f(a))
is 1011T, there is (Theorem 3.1O(c)) a neighborhood N(a, 8) such that
1T! (f(t)) > 5 11T for tEN (a, 8) n [a, b) = [a, a + 8). All these points
f(t) are therefore on the curved part.
(b) Considering (a), let c := a +sup{8: rrj (f(t)) > 0 for all t E [a, a +8)}.
We show that this c is the first point of the arc on the straight part.
First, if aSs < c, then by definition of sup there exists 8 such that
s < a + 8 S c and f(t) is on the curved part for f E [a, a + 8);
in particular, f(s) is on the curved part. Second, rrj (f(c)) cannot be
positive. If it were, then we would have c < b, plus there would be a
neighborhood (c - /)., c + /).) in which rrj (f(f)) would stay positive,
contradicting the definition of c. Hence rrj (f(c)) = 0; f(c) is on the
y-axis.
Alternatively: Put c := inf{t: rrj (f(f)) = OJ. The indicated set is the
inverse image of {OJ under rrj (f), so it is closed. Therefore, its inf is its
least element. That makes rrj (f(c)) = 0 and rrl (f(t)) > 0 for f < c,
answering (a) along with (b).
(c) Let (s, sin lOis), 0 < s S IO/rr, be a point of the curved part up to
f(a). Since the x-value rrl(f(t)) is a continuous function of f, with
values 10/rr when f = a and 0 when f = c, it must achieve the
intermediate value s at some f = c. Since rrl (f(c)) = sand fCc) is on
the curve, necessarily fCc) = (s, sin lOis).
(d) Observe that the order has to be a < f5 < f9 < ... < c. That
is, rrl (f(t9)) = 20/9rr < 20/2rr = rrl (f(a)), so the intermediate
value 2015rr must have been achieved between f = a and f = f9;
that proves a < f5 < f9, and similarly for the other fj+4i. Thus,
(tl+4i) is an increasing sequence, converging to some d S c. Since
rrj(f(d)) = lim rrl (f(tj+4i)) = lim20/(rr +4rri) = 0, and c is the
first place where x = 0, we have d 2: c. We see that (tj +4i) -+ c.
(e) Same argument as (d).
(f) We now have rr2(f(c)) = limi-HlO rr2(f(fl+4i)) = 1 and rr2(f(c)) =
limi->oo rr2(f(t3+4i)) = -1.
2. The interval (00, -00) is empty, and by convention inf 0 = 00, sup 0 =
-00.
3. (a) See the proof of Theorem 5.5(b).

(b) See the solution to Exercise 8b in Section 4.5.
4. (a) Exercise 8a in Section 4.5 shows that JOINS is reflexive, and Exercise
8c shows that it is symmetric. Exercise 3a in this section shows it to
be transitive.
(b) Let la be an equivalence class. Any two x, y E la are related to a, so
there are arcs within 0 from a to x and from a to y. Clearly, any point
along these arcs is joined to a by part of that arc, so these points are in
la. Then we put the arcs together (Exercises 8c in Section 4.5 and 3
in Section 5.1) to give us one arc within la from x to y. We conclude
that la is arc-connected.
(c) Assume fa U {b} S; T with b ~ fa. By definition of equivalence class,

b is not related to a. That is, no arc within S joins a to b. Hence T is
not arc-connected, and fa is maximal.
Section 5.2
1. If t E (0, 1], then 0 < t ~ 1 < 1 + 1/ i for each i, so t is in the intersection.
Conversely, if t is in the intersection, then 0 < t < 1 + 1/ i for each i;
passing to the limit in the second inequality, we have t ~ 1, and t E (0, 1].
2. (a) Assume x < O. Then (1 - a)( -1,0) + a(x, y) = (a - 1 + ax, ay)

has (strictly) negative x-coordinate for every 0 ~ a ~ 1.
(b) Assume x > 0, and suppose g : [r, s] --+ R2 is a continuous function
with g(r) = (-1,0), g(s) = (x, y). Then 1T1 (g) is continuous on
[r, s], with 1T1 (g(r» = -1, 1T1 (g(s» = x > O. By the intermediate
value theorem, there exists C E (r, s) with 1T1 (g(c» = O. Then g(c) is
a point of the arc on the y-axis.
3. For x > 0, f = 0 at x = 1,~, t, ....

In each of the intervening inter-
vals, f must be of one sign. The intervals of positivity are (1, 00), (t, ~),
(~, !) , ....
Since f is even, we add the corresponding ones on the x < 0
side.
4. (a) The sum of the lengths is t + ~ + ~ + ... = (1) / (1 - ~) = 1.

(b) Let C := [0,1] - O. Since C = [0,1] n 0*, C is closed. C is
also infinite, since the components of 0 do not abut (do not share
endpoints), and those endpoints are in C. On the other hand, (a) says
that in some sense C occupies 0% of the space in [0, 1]. So C is weird,
and deserves more study.
5. (a) and (b) In the figure, the shaded portion shows Q2 U Q3 U Q4 near
the unit square. It makes clear that each Qi crosses the "main canal" Q2.
Accordingly, 0 is arc-connected, and has just one component.
3
(c) By DeMorgan's law, S - 0 = S n Q~ n Q n· ... For each power-of-
two group (2, then 3, 4, then 5, 6, 7,8), its intersection cuts the square(s)
from the group before into four times as many, each one-third as big.
In symbols, S n Q~ has four squares of size x 1 1, Sn Qi 3
n Q n Q:
has sixteen ~ x ~,and so on. Hence for each i, S - 0 is subset of a
union of squares with total area 4i /9 i ; 0 occupies 100% of the unit
area of S.
6. (a) In the figure, the shaded portion is PI U ... U PI +8+64. The picture
suggests that the gasket is a square with an infinity of square holes.
(b) Yes. Since the Pi are disjoint convex open sets, they must be the com-
ponents of O.
1
Exercise 5 Exercise 6
(c) PI has area (~) 2 , P2 through PI +8 all have area (~) 2 , and so on. Thus:
1 square of area (~) 2 , 8 squares of area (1/3 2 ) 2 , 64 squares of area
(1/3 3 )2, ... ; total area 1/32 + 8/34 + 64/3 6 + ... = (1/3 2 )/(1 -
8/3 2 ) = 1. The holes have all the area.
(d) No. All these constructions behave alike: If some neighborhood N (b, 8)
~ S were free of points of 0, then the area occupied by 0 would be
1 - Jr8 2 or less.
7. (a) This is trivial: The way we defined "component" (elements joined to

a) matches the definition of "equivalence class" (elements related to
a).
(b) A set harboring an equivalence relation is always the disjoint union of
the equivalence classes, and by (a) the classes are the components.
(c) See the solution to Exercise 4b in Section 5.1.
8. (a) T:= {OJ has only one equivalence class, and it is not open.
(b) Remember S from Example 1 in Section 5.1 and Exercise 1 in that
section. Clearly, you can join any point of the straight part to (0,0),
and you can join any two points in the curved part. We-decided that you
cannot join from the curved part to (0,0). Hence the MACS (equiva-
lence classes) are the curved part and the straight part. But neither is
alone a maximal connected subset, because S is connected.
(c) Let U := {(x, y): x is irrational}. Then each vertical line contained in
U is an equivalence class under JOINS, and the lines are uncountably
numerous.
Section 5.3
1. (a) Let ~ be a class of closed sets and S := nCEl: C their intersection.
Then each complement C* is open (Theorem 4.14), the union T :=
UCE'E C* of the complements must be open (Theorem 5.3(a», and

T = S* (DeMorgan). Thus, S has open complement, so S is closed.
(b) Same argument with 1: := {Cj, ... , Cd and Theorem 5.3(b).
2. (a) B(O, 1) U B (0,1) U B (0, ~) u··· = B(O, 1), which is closed (Ex-
ercise 4b in Section 4.2) but not open (any neighborhood N«I, 0), 8)
of (1, 0) has the point (1 + 812, 0) from outside B(O, 1».
(b) B (0, ~) U B (0, ~) U B (0, ~) U ... = N(O, 1) (Explain!), which
is open but not closed.
(c) The rings {(x, y): I/(i + 1) ~ x 2 + y2 ~ IIi}, i = 1,2, ... , add up
to {(x, y): 0 < x 2 + y2 ~ 1} (Explain!), which is neither open nor
closed (Explain, too!).
(d) B(O, 1) U B(O, 2) U B(O, 3) U· .. = R2. By Exercise 6 in Section 4.5,
the only open and closed candidates are R2 and 0, and it is impossible
to express 0 as the union of unequal sets.
3. (a) Tis not open, because every neighborhood N (0,8) has points (0,812)
from outside T. It is not closed, because Xi := ([1 + Ili]i, 0) -+
(e,O) rf. T.
(b) Every set is the union of its MACS (Exercise 7b in Section 5.2). lithe
MACS are closed and finitely numerous, then the set is the finite union
of closed sets, and is therefore closed.
(c) See the solution to Exercise 8c in Section 5.2. (Why is each line there,
like {(J2, Y): y E R}, closed?)
(d) No. MACS are disjoint, so a collection of open MACS is a collection
of disjoint open sets. The argument in Theorem 5.7 applies: Every such
collection has to be countable.
4. By Example 2(b), if b < c are in C, then there exists z rf. C with b < z < c.
Then {C n (-00, z), C n (z, oo)} is a disconnection of C.
5. See it geometrically and analytically. (Consult Example 2.)

For the geometric look, let c E C. Since no two component subintervals
from 0 abut, c cannot be the endpoint of subintervals on both sides. Thus,
on one side-say the right-every interval (c, c + £) has a point x E C.
The interval (c, x) has a point y from 0, because C contains no intervals
(Example 2(b». The component of 0 to which y belongs must be a subset
of (c, x), because otherwise (if it stretched left of c or right of x) either c or
x would be in it. That gives evidence for (c), which implies the others.
Analytically, let c E C have the ternamal #aja2 ... , with each ai = 0 or 2.
Then Xi := #aja2 ... ai 111 ... , which is not in C, is less than 3- i from c.
Hence Xi E N(c, £) for big enough i, proving (a). The same neighborhood
has Yi := #aj a2 ... ai 1000 ... and Zi := #aj a2 ... ai 1222 .... Those are
in C, because the l's are removable. The interval (Yi, Zi) is necessarily a
component of 0, because anything in it, having an irremovable 1 in temamal
place i + 1, is in O. That proves (b). It also proves (c), because no two Yi
match.
6. (a) Map the endpoints Xi, Yi of component Pi (defined in Example 3 in

Section 5.2) into 2i - 1, 2i, respectively. The mapping is well-defined,
because no component of 0 shares an endpoint with any of its prede-
cessors, and is clearly one-to-one.
(b) Expansions have two properties: Each real r E [0, 1] has a "binimal"
r = &hb2 ... := bj/2 + b2/4 + b3/8 + ... , where every bi is 0
or 1; and the binimal is unique if we require that the sequence (bi)
not end in 0, 0, .... With that principle in mind, we see that fer) :=
(2bj)/3 + (2b2)/9 + (2b3)/27 + ... = #(2bj)(2b2) ... maps r into
a temamal with only O's and 2's. Since two such temamals are equal
only if produced by identical sequences (2bi), f maps [0, 1] into C
one-to-one.
(c) If C were countable, there would be a one-to-one g: C ~ N. But then
g(f) (using f from part (b» would map [0, 1] one-to-one into N. That
would contradict the uncountability of [0, 1].
Section 5.4
1. (a) The y-axis has empty interior, because the neighborhood N«O, b), 8)
of point (0, b) has the point (8/2, b) noton the axis. Hence all its points
are boundary points. There are no other boundary points, because if
a =I 0, then N«a, b), lal/2) has no points ofthe axis. Hence bd(y-axis)
= y-axis.
(b) Quadrant I is an open set (Exercise 1 in Section 4.4), so it is its own
interior. Its boundary is the union ofthe nonnegative axes: Near (0, b),
with b 2: 0, we have (8/2, b) from quadrant I and (-8/2, b) from
outside it; similarly near (a, 0), with a 2: 0; and if either e < 0 or
d < 0, then the smaller of N«c, d), leI) and N«c, d), Idl) is devoid
of quadrant I points.
(c) We have seen (Example 2 in Section 3.4, among others) that every
neighborhood has points from Q2 and from its complement. That tells
us that bd(Q2) is all of R2. Then nothing is left for int(Q2), so this set
must be empty.
2. (a) Every ball is closed (Exercise 4b in Section 4.2); by Theorem 5.11(b),

a ball equals its closure. This means that its complement is open, so
its exterior := interior of complement = complement.
(b) cl(N (b, 8» = B(b, 8) and ext(N (b, 8» is the rest of R2 (outside the
ball). The ball B(b, 8) is a closed superset of N(b, 8), so B(b, 8) ~
cl(N(b, 8» (Theorem 5.11(d». By the argument in Example 2(a), the

points outside the ball are not closure points of the ball, let alone of
N (b, 8), socl(N(b, 8» £ R(b, 8). We conclude both thattheclosureis
the ball and that the complement of the ball, being open, is ext( N (b, 8».
(c) Closure = line, exterior = rest. Lines are closed sets: If (Xi, Yi) is a
sequence from the line given by ax + by = c and (Xi, Yi) -+ (s, t),
then as + bt = lim(axi + bYi) = c, so (s, t) is on the line. Hence
the closure is the line, and exterior = interior of (open) complement =
complement.
(d) Adapt the argument from the solution to Exercise l(b): The points on
the nonnegative axes {(x, 0): X ~ O} and {CO, y): Y ~ O} are bound-
ary points, and the points outside [0,00) are exterior. Consequently
cl(quadrant I) = [0,00), and the rest is the exterior.
(e) Since every point is a boundary point of Q2, the closure is R 2 and the
exterior is 0.
3. (a) int(int(S» = int(S). The interior is open (Theorem 5.9(c», so it is its

own interior (Theorem 5.9(b».
(b) int(ext(S» = ext(S). The exterior is by definition int(S*), which by (a)
is its own interior.
(c) ext(bd(S» = int(S) Uext(S). The boundary is closed (Theorem 5.10).
Hence its complement is open, is therefore its own interior, is there-
fore ext(bd(S». As our figures have shown, the complement bd(S)* is
int(S) U ext(S).
4. Exercises 4 and 5 are intended to show that the six combinations not covered
in Exercise 3 are as we would expect for a neighborhood, but can be strange
in general. Let N := N (b, 8) and S := S(b, 8). Then:
(a) int(bd(N» = 0. The boundary is the sphere: Points in N are interior

(N is open), points outside the ball are exterior (Exercise 2b), and
YES has nearby points Y ± .58 (y - b) /8 from inside and outside N.
Those same points are off the sphere, so they tell us that the sphere has
no interior points. Hence int(bd(N» = int(S) = 0.
(b) bd(int(N» = bd(N). The interior of N is N, so its boundary is that
of N (which is S by the previous paragraph).
(c) bd(bd(N» = bd(N). First, bd(N) = S. Second, we have seen that
S has no interior, so every point of S is a boundary point. Last, there
are no other boundary points of S, because points of N are surrounded
by N and points outside the ball are surrounded by outsiders. Hence
bd(S) = S.
(d) bd(ext(N» = bd(N). Points in N are surrounded by N, because N
is open. Similarly, points in ext(N) are surrounded by ext(N). That
leaves S, whose points have nearby points from ext(N) (Example 2(a)
or part (a) here). Hence bd(ext(N)) = S.
(e) ext(int(N)) = ext(N). int(N) is N, so its exterior is ext(N) (= com-
plement of ball, by Exercise 2a).
(f) ext(ext(N)) = int(N). We know ext(N)* = (complement ofball)* =
ball. Therefore, ext(ext(N)) := int(ext(N)*) = int(ball) = N = int(N).
5. (a) and (c) In Example 2(b), wesawthatbd(T) is the annulus consisting of the
unit circle, the radius-2 circle, and the points in between. From the arguments
in Exercise 4, it is clear that the in-betweens are interior to this set and the
two circles form its boundary. Hence (a) int(bd(T)) = {x: 1 < IIxII2 < 2},
which is not empty; and (c) bd(bd(T)) = {x: Ilxti2 = lor 2}, which is only
a subset of bd(T).
(b) and (e) Taking from T the part in the boundary annulus, we see that
int(T) = N(O, 1). Hence (b) bd(int(T)) = S(O, 1), a subset ofbd(T); and
(e) ext(int(T)) = complement of unit ball, a superset of ext(T).
(d) and (f) We have identified int(T) and bd(T), so what is left is exterior:
ext(T) = {x: IIxII2 > 2}. Hence (d) bd(ext(T» = radius-2 circle, a subset
ofbd(T); and (f) ext(ext(T» = N(O, 2), a superset of int(T).
(In these last three paragraphs, those extra remarks about "subset" and "su-
perset" are not coincidences; check that they are always true.)
6. For (c): We know that C is closed, so cl(C) = C. For the others, in view of
Exercise 5 in Section 5.3:
(a) int(C) is empty.

(b) bd(C) = C. Since there are no interior points, C ~ bd(C). Since the
definition of boundary implies that bd(C) ~ cl(C), we have bd(C) ~
C.
(d) Set is empty. Every C E C has nearby intervals from C*, an infinity
of these intervals ending in Cantor points 1= c. Hence no point of Cis
isolated.
(e) Set is C. By (d), every C E C is an accumulation point of C.
7. (c) Suppose (Xi) is a sequencefromcl(S) convergingtox. For any N(x, s),

there is I such that Xl , Xl +1, ... are in N (x, s). Since Xl is in the open
set N(x, s), there is a neighborhood N(XI, 8) ~ N(x, s). Because Xl
is a closure point of S, some YES must be in N (Xl, 8) (Exercise 5 of
Section 3.2). This Y is a member of S inside N(x, s). We have shown
that X E cl(S), leading to the conclusion that cl(S) is closed.
(d) By part (a), S ~ cl(S), and by (c), cl(S) is closed. Thus, cl(S) is a
closed set containing S.
Suppose T is another such set. If x E cl(S), then by definition there

exists (Xi) from S converging to x. Since S S; T, (Xi) comes from T.
The latter being closed, (Xi) ~ X forces X E T. We have proved that
cl(S) is the smallest closed superset.
8. The key is Theorem 3.1. Using that result, we have X E cl(S)* ¢> X rt cl(S)
¢> there exists N (x, E) devoid of points of S ¢> there exists N (x, E) S; S*
¢> x E int(S*).
9. (a) Apply Theorem 3.1 again: x E cl(S) ¢> every neighborhood ofx has
points from S ¢> (every neighborhood has points from both Sand S*)
or (every neighborhood has points from S and some neighborhoods
have points from only S). The first clause in parentheses says that
x E bd(S), the second says that x E int(S), and their disjunction
("or") amounts to x E bd(S) U int(S).
(b) By (a), cl(S) = bd(S) U int(S). We know that int(S) is contained in S.
Therefore, cl(S) S; S iff bd(S) S; S. By parts (a) and (b) of Theorem
5.11, bd(S) S; S iff S is closed.
(c) By Theorem 5.9(b), S is open iff every point of S is an interior point.
Since every set partitions into interior points and boundary points, we
conclude that S is open iff no point of S is in bd(S).
10. (a) Let (Xi) be a sequence of accumulation points of S converging to b. Any
neighborhood N (b, E) possesses some XI, X/+ 1, .... If XI = b, then
b is already an accumulation point of S. If not, let 8 := min{lIxI - bll,
E - IIxI - bJIl. Then (draw the picture) N(XI, 8) S; N(b, E) and does
not reach b. Within N (XI, 8), there must exist y =1= XI from S. This y
cannot be b, so we have found y =1= b from S in N(b, E). Hence b is
an accumulation point; the set of such points is closed.
(b) Partition the space into int(S), bd(S), and ext(S). The points of int(S)
are accumulation points of S but not S*; those in ext(S) are accumula-
tors of S* but not of S. In the boundary, there are points of S surrounded
by S* , meaning isolated points of S, which are accumulators of S* but
not S; points of S* surrounded by S, which are accumulators of S
but not S*; and points with neighbors from both Sand S*, which are
accumulation points for both, irrespective of which they belong to.
(c) No. S := {( i, 0) , (~, 0) , ... } has only isolated points, and is not
closed. (Why?)
11. (a) Yes to all. The closure of the interior of a ball is the ball (equal);
cl(int(line» is empty (smaller); cl(int(neighborhood» = ball (bigger);
S := N(O, 1) U {(2, O)} S; R2 has int(S) = N(O, 1), cl(int(S» =
B(O, 1) (not subset of S, not superset).
(b) Yes again: int( cl(neighborhood» = neighborhood; int( cl(line» = empty;
int(cl(Q2» = R2; S .- {x: 0 < IIxli ~ I} has int(cl(S» = {x:
Os IIxJJ < I}.
(c) First question: ext(cl(S)) = ext(S) by Theorem 5.12 and Exercise 3b.
Second question: cl( ext( S)) is int(ext( S)) U bd(ext( S)) by Exercise 9a.
Of the last two, int(ext( S)) is predictable, being ext( S) because ext( S)
is open; the other is not, as we see by Exercise 5d. It has to be a subset
of bd(S), because it cannot have any of int(S) or ext(S), and it would
include any isolated points of S.
12. (a) If S is bounded, say S ~ N (0, M), then B(O, M) is a closed superset
of S, so that cl(S) ~ B(O, M) (Theorem 5.11(d)). The closure is
therefore bounded.
(b) Assume that S is convex and x, Y E cl(S). By definition of closure,
there exist sequences from S with (Xi) ~ X and (Yi) ~ y. If (X and
{3 are nonnegative reals summing to 1, then «(xXi + {3Yi) is a sequence
from S (because S is convex) with lim «(xXi + {3Yi) = (Xx + {3y. This
shows that (Xx + {3y E cl(S), so that the closure is convex.
(c) The result is trivial if S = 0, so assume that S is connected and
nonempty. Let {R, T} be a partition of cl(S). Examine R n Sand
Tn S. If one of them is empty, say R n S, then S ~ T; that says
that R, which must have closure points of S, has closure points of
T. If instead both are nonempty, then they partition S. Because S is
connected, Theorem 5.13 tells us that one, say R n S, has closure points
of the other. This says that R has closure points of T. We have shown
that if {R, T} partitions cl(S), then one of R, T holds closure points
of the other. By Theorem 5.13, cl(S) is connected.
Section 5.5
1. (a) The hypotheses make S closed and bounded. By the Heine-Borel theo-
rem, a closed bounded subset ofR n is compact. Its continuous image is
compact (Theorem 5.19), and this compact image is closed (Theorem
5.15).
(b) No. Any constant function maps open sets into nonopen sets.
(c) First, r-I(T) = B(O, 1) - r-I(T*): X E r-I(T) #- rex) E T #-
rex) ~ T* #- X is in the part of B(O, 1) that does not map to T*.
Second, r- I (T*) is the intersection of B(O, 1) with some open set 0:
T is closed, so T* is open, so r- I (T*) is B(O, 1) n 0 (Theorem 4.17).
Therefore, r-I(T) = B(O, 1) - 0 = B(O, 1) n 0*. This last is an
intersection of closed sets, so it is closed. Being also a bounded set in
R n , it is compact.
You should see that the compactness of B(O, 1) is essential. Any con-
stant function will have r-I(range) = domain, with "range" compact
but not necessarily "domain."
2. (a) By Theorem 5.20, r can be extended continuously to cl(S). The closure
of a bounded set is bounded (Exercise 12 in Section 5.4), so cl(S) is
compact. Therefore, r( cl(S)) is compact (Theorem 5.19), and its subset

reS) must be bounded.
(b) No. R n is closed and f(x) := tan-I IIxll2 is uniformly continuous on
it (by the mean value theorem), but feRn) = (-T( /2, T( /2).
(c) Yes to both. If S is closed and bounded in R n , then it is compact, and
so must reS) be.
(d) No, yes. See Exercises 1b and 2a, respectively.
(e) Yes. f(x) := 1/(1- Ilxll2) maps N(O, 1) continuously onto [1, 00).
3. Assume that S is compact and T ~ S is closed. Let (Xi) be a sequence

from T. The Xi belong also to S, and S is sequentially compact (Theorem
5.15(b)). Hence there is XES with Xi ~ x. We now have a sequence from
T converging to x, forcing X E T. We have proved that T is sequentially
compact, and by the remarks following Theorem 5.16, T is compact.
4. We have here the hypothesis of Theorem 5.18, plus diameter ~ O. Con-

sequently, we know that TI n T2 n ... is not empty. The only question is
whether it can hold two different members. Suppose x, yare in the inter-
section. Then x, Y E Ii for each i, so that IIx - YII ::: diam(1i) ~ O. We
conclude that X = y.
5. (a) Assume that S is compact. Let n be a family of closed subsets of S, and

suppose that n has empty intersection: ncen C = O. By DeMorgan,
Ucen C* is the whole space. In particular, S is covered by {C*},
and these sets are open. Hence some finite subcollection Cr, ... , C;
suffices to cover S. That is, each XES is in q U ... u C;, so that
X E CI n ... n Ck is impossible. Thus, from the assumption that
ncen C = 0, we constructed a finite subfamily {CI, ... , Ck} having
empty intersection. By contraposition, if n has the finite intersection
property, then it has nonempty intersection.
(b) N(O, 1) ~ R n is not compact, but if n is a family of closed subsets
of N (0, 1) with the finite intersection property, then the members
of n are also subsets of B(O, 1) and must therefore (part (a)) have
nonempty intersection. (If this appears to contradict a theorem from
topology, bear in mind that "closed subsets of N (0, I)" means "closed
subsets ofRn that happen to be contained in N(O, 1).")
6. (a) LetybeanymemberofS.Recallthatifoy := IIx-yJJ/2,thentheneigh-

borhoods N(x, Oy) and N(y, Oy) are disjoint. The family {N(y, Oy):
YES} is clearly an open cover for S. There is therefore a finite
subcover; label it N(YI, 01), ... , N(Yk, Ok). If we set 0 := least of
01, ... ,Ok. then 01 := N(x, 0) does not intersect any of the N(Yj, OJ).
Consequently, 02 := N (YI, 01) u ... U N (Yk. Ok) is an open set with
S ~ 02, 01 n 02 = 0, and X E 01.
(b) Let Z be any member of S. By (a), there exist disjoint open sets 01 (z)
and 02(Z) with Z E 01 (z), T ~ 02(Z). The collection {Ol (z)} covers
S, so there exists a subcover 01 (ZI), ... , 01 (zm). Each corresponding
02(Zj) contains T, so T ~ 02 := 02(Zj) n··· n 02(Zm). 02, being
a subset of every 02(Zj), is disjoint from every 01 (Zj)' Hence 02 is
disjoint from their union: 01 := 01 (ZI) U· .. U 01 (zm) does not meet
02, and S ~ 01.
(c) In R2, x := 0, T := {OJ, S := quadrant I offer counterexamples for
(a) and (b).
7. We can look back to Exercise 9b in Section 4.3. The hypotheses here give
us S sequentially compact, T closed and disjoint from S. Hence 8 :=
deS, T)/2 > O. For any XES and YET, we must have IIx - yll ~ 28.
Therefore, N(x, 8) does not reach N(y, 8), and 01 := UXES N(x, 8) does
not intersect 02 := UYET N(y, 8). (Why are 01, 02 open?)
8. The distance function d(x, S) is a good start, since it is constantly 0 on

S and is at least c := deS, T) > 0 on T (Exercise 9b in Section 4.3).
It is also continuous, in fact, contractive (Exercise 5 in Section 4.3). Set
f(x) := d(x, S)/c if d(x, S) < c, := 1 if d(x, S) ~ c. Clearly, f = 0
on S, f = 1 on T. The reason f is continuous is that f is everywhere
the smaller of d(x, S)/c and 1. Whenever you have continuous functions
g and h, the function min{g(x), hex)} is necessarily continuous: min{g(x),
hex)} = (g(x) + h(x))/2 -Ig(x) - h(x)I/2, and the expression on the right
is a continuous composite.
Section 5.6
1. {:= Suppose V has finite dimension. We know, from considerable experience,
that R(O, 1) is closed and bounded. By Theorem 4.29(b), R(O, 1) has the
EVP. By Theorem 5.16, it is compact.
=} Suppose V has infinite dimension. By Theorem 5.23, V has a sequence
(Ui) of unit vectors separated by distance ~ 1. That separation means that
(Ui) has no Cauchy subsequences (same argument used in Theorem 5.24(a)).
Therefore (Ui) has no convergent subsequences. Thus, R(O, 1) has a se-
quence with no sublimits; it is not sequentially compact.
2. Compare the solution to Exercise 3 in Section 4.6.
(a) Let P2 be the set of all quadratic polynomials over the domain [0, 1].
P2 is a vector space of dimension 3. Hence all norms on P2 are
equivalent. In particular, IIp(x)11! := Ibl + lei + Idl and IIp(x) 110 :=
sUP-19:o1 Ip(x)1 are equivalent, so there exists M with IIplll :::: Mllp 110
independent of p. The subset (not subspace) P is precisely the II 110-
unit ball: P = {p: IIpllo :::: I}. Hence for any PEP, we have
Ibl :::: IIplll :::: Mllpllo:::: M.
(b) Set Pi (x) := (1 - x)i. Clearly, 0 ~ Pi (x) ~ 1 for every i and x, but
the second-degree coefficient in Pi (x) is i(i - 1)/2.
3. This is a standard vector-space result. The expression p in brackets is the
projection of x onto the subspace spanned by U1, ... , Uk. (See the remarks
between Theorems 5.21 and 5.22.)
For any j, (x - p) • Uj = x. Uj - p. Uj = x. Uj - ([x. u11u1 • Uj +
... + [x • Uk]Uk • Uj). In the parentheses, each of U1 • Uj, ... , Uk • Uj is
0, save for Uj • Uj, which is 1. Hence (x - p). Uj = x. Uj - x. Uj = O.
Thus, x - P is orthogonal to every Uj. By linearity of the dot product, x - p
is orthogonal to any combination a1 U1 + ... + akuk.
4. Refer to Section 1.6, especially Example 1.
(a) As we saw, the functions S j := sin 2j JT X are mutually orthogonal
and of length -J2 /2. Therefore {-J2s j } is an orthonormal set. Clearly,
a j = f • -J2s j and !k := a1-J2s1 + ... + ak-J2sk is the projection of
f onto the span of -J2s1, ... , -J2sk. By the principles cited between
Theorems 5.21 and 5.22, fk is the combination of -J2s1, ... , -J2sk
closest to f, making it likewise the closest combination of Sl, ... , Sk.
(b) and (c) This is just the Pythagorean theorem. We have f = (f - fk) + fk.
By Exercise 3, f - fk is orthogonal to fk. Hence IIflll = IIf - fklll +
t
II !k III ~ II fkll The inequality is strict, unless fk actually is f·
5. Write C := B(U2, 1/2) U B(U3, 1/3) U .... Since each Ui has unit norm,
the biggest possible norm in C is 1.5, so C is bounded. Also, since the
centers are 1 or more apart, the closest two of the component balls can get
is !- t = ~. Let (Xi) be a sequence from C converging to x. Then (Xi) is
Cauchy, so there is! such that II Xi - Xj II < ~ for any i, j ~ I. Of necessity,
all of Xl, X/+1, ... come from a single ball B(uk, 1/ k). Since this ball is
closed and (Xi) -+ x, we conclude that x E B(uk, 1/ k) S; C; the latter is
closed.
Now define f(y) := 2klly-ukll fory E B(uk, 1/ k),k = 2, 3, .... Then f is
continuous in each ball separately, because distance from Uk is continuous.
That makes f continuous on C, because the test sequences (those with
Yi -+ y) must, as described above, come eventually from a single ball.
However, f is not uniformly continuous, because the points [1 + .25/ k]Uk
and [1- .75/ k]Uk are "close together" (distance = 1/ k) in C, butthe function
difference f([1 - .75/k]uk) - f([1 + .25/k]uk) = 1 is not "small."
6. Let S := {OJ and T := {(4/3)U2, (5/4)U3, (6/5)U4, ... }. S obviously fits
the bill. T is closed, because its members are far apart: Each member comes
from a different one of the balls in Exercise 5, and those balls are at least ~
from one another; therefore, you cannot make Cauchy sequences from T,
except by repetition. But the members of T are at distances 1, i, ...from
S; there is no minimum.
References
We have made references to just four books. Here we list them, together with
comments on why they are our favorites.
Morris Kline, Mathematical ThoughtfromAncient to Modern Times, Oxford Uni-
versity Press, New York, 1972.
Kline's book is a monumental achievement among histories of the
sciences and mathematics. Although our focus on continuous func-
tions restricts our interest in history to Europeans during (roughly)
1810-1870, Kline is a wonderful source of information about the de-
velopment of mathematical ideas worldwide and over thousands of
years.
David C. Lay, Linear Algebra and Its Applications, 2nd edition, Addison Wesley
Longman, Reading MA, 1997.
Lay is a good source for the material we have assumed from elemen-
tary linear algebra.
Kenneth A. Ross, Elementary Analysis: The Theory of Calculus , Springer-Verlag,

New York, 1980.
This is the best introduction to advanced calculus we know. The sub-

title suggests that Ross's mission is an axiomatization of elementary
calculus. The book carries out that mission with an admirable com-
bination of mathematical rigor and attention to pedagogy, the latter
absolutely essential in a student's entry to this level of theory.
204 References
H. L. Royden, Real Analysis, 3rd edition, Macmillan, New York, 1988.
This graduate-level text is best described as "elegant." It may there-

fore seem like cheating to refer to it in a treatment that claims to be
elementary. Royden's first two chapters, however, are introductory,
and happen to cover the properties of continuous functions of one real
variable.
Index
accumulation point, 62 closure of a set, 136

additive function, 30 closure point, 60
angle between vectors, 11 compact set, 140
arc (from point to point), 108 complete normed space, 50
arc-connected set, 109 component
of a vector function, 56
ball, 33
of a vector, in a direction, 10
Bolzano-Weierstrass theorem (BWT),
of an open set, 124
45,47,48,95
composite function (continuity of),
boundary of a set, 135
76
bounded
connected set, 107, 138
function, 87
continuous function, 73
linear mapping, 82
continuity of a composite, 76
sequence, 41
continuity of a contractive func-
set, 90
tion,79
box, 86
continuity of a linear combina-
infinite box, 89,94
tion, product, quotient, 76
open box, 90
continuity on a set, 73
Cantor set, 131 contractive mapping, 44
Cantor, Georg, 133 convergence, 37
Cauchy's criterion, 44, 48, 49, 72 to infinity, 39
Cauchy's inequality, 11 uniform convergence, 52
closed convex
interval, 86, 101, 122, 141 combination, 110
set, 91 hull,110
206 Index
convex (continued) interval, 122

set, 108 closed interval, 101, 122
coordinate projections; See also pro- open interval, 101, 122
jection mappings, 42 inverse image, preimage, 103
countable set, 133 isolated point, 62, 138
Cremonini, Sera, 155
law of cosines, 12
diameter of a set, 100 left (a is to the left of b), 86
Dirichlet's function, 74 length,9
disconnected set, 107 Pythagorean length, 14
disconnection, 107 limit
distance of a composite, 70
between sets, 100 of a function, 64
from point to set, 77 of a linear combination or a prod-
distance, See also metric, 19 uct, 65
dot product, 8 of a quotient, 69
of a sequence, 38
equivalence relation, 118, 123 limit point; See closure point, 60
Euclidean space, 15 line, 4
exterior of a set, 136, 139 determined by points, 5
extreme value property (EVP), 96 line segment from a to b, 6
linear mapping, 82, 84
extreme value theorem, 87
bounded linear mapping, 82
Fernandez, Ted,45 linear space, 4
finite intersection property, 148 linear transformation; See linear map-
ping, 82
finite subcover(ing), 141
finite-dimensional subspace, 116 maximal arc-connected subsets, 123,
first-degree real or vector function, 129,130
56 maxnorm, 16,26
Fourier coefficients, 153 metric (as adjective), 119
metric; See also distance, 19
Gram-Schmidt process, 49,149 midpoint formula, 6
Heine-Borel property, 140 n-tuple, 1,2

Heine-Borel theorem, 143 neighborhood, 33
of infinity, 41
image of a set, 99, 103 nested intervals theorem, 148
inverse image, 103 nested sets theorem, 145
indexed family, 141 norm, 13
infinite set, 95, 133 equivalent norms, 111, 112
inner product, 8 mean-norm, 25
inner product space, 9, 10 standard norm, 15
interior of a set, 134 uniform norm, 52
intermediate val ue property (IVP), 106 normed linear space, normed vector
intermediate value theorem, 88, 107 space, 13
Index 207
open scalar function, scalar-valued func-

ball, 101 tion; See real function, 55
box or interval, 101, 122 scalar product, 8
cover(ing), 141 sequence, 38
set, 100 bounded sequence, 41
orthogonal vectors, 9 Cauchy sequence, 44
convergent sequence, 38
pair, ordered pair, 1 sequentially compact set, 97
parallelogram identity, 28 Siwanowicz, Jan, 98
partial order, 90 sphere, 33
partition, nontrivial partition, 107 star-shaped set, 110
perpendicular vectors, 9 subcover(ing), 141
persistence of sign, 68, 75 sublimit, 45
points, 4 subsequence, 38
polynomial real or vector function,
57 ternary (representation of reals), 131
projection, 11 topology, 119
mappings (same as projections), totall y disconnected set, 131
42 translate, 4
of a vector onto (line of) a vec- triangle inequality, 14
tor, 9 triple, ordered triple, 1
of a vector onto a subspace, 149
Pythagorean theorem, 10 uncountable set, 133
uniform continuity, 88
Rn,l uniform convergence, 52
rational real or vector function, 57
real function, real-valued function, vector function, vector-valued func-
55 tion, 55, 58
rectangle, 86 vector space, 4
relation, 1
rescaling, 111
restriction (and continuity), 73
Forthcoming by Alberto Guzman
Derivatives and Integrals of Multivariable Functions

ISBN: 0-8176-4274-9
This work examines derivatives and integrals of functions of several real vari-
ables. Topics from advanced calculus are covered, including: differentiability and
its relation to partial derivatives, directional derivatives and the gradient, sur-
faces, inverse and implicit functions, integrability and properties of integrals, and
the theorems of Fubini, Stokes, and Gauss. The order of topics reflects the order
of development in calculus: limits, continuity, derivatives, integrals---a sequenc-
ing that allows generalizations from and analogies to elementary results, such as
the second-derivative test and the Chain Rule.
Derivatives and Integrals of Multivariable Functions has a definition-theorem-

proof format, together with a conversational style, including historical comments,
an abundance of questions, and discussions about strategy, difficulties, and alter-
native paths. It is aimed at advanced undergraduate pure mathematics majors
whose next course will be real analysis with measure theory. Required background
includes theoretical work in linear algebra, one-variable calculus, properties of
continuous functions, and related topological material. The last two are used in
the context of Euclidean space, but a strong grounding in the corresponding real-
line topics will suffice.
Table of Contents:
1. Differentiability of Multivariable Functions
2. Derivatives of Scalar Functions
3. Derivatives of Vector Functions
4. Integrability of Multivariable Functions
5. Integrals of Scalar Functions
6. Vector Integrals and the Field Theorems
Bibliography
Index

Solutions To Exercises: J3) - J3) J3 - J3

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Solutions To Exercises: J3) - J3) J3 - J3

Uploaded by

Copyright:

Available Formats

Solutions to Exercises

2. This is familiar from the definition of function. Keep Exercise 1 in mind.

Exercise la Exercises 3a, c

giving z = arbitrary t, Y = 3 - 2t, x = -2 + t. Those equations are

4. If a, b, and c are all 0, then R2 is the solution set. If a = b = 0 and c i= 0,

5. (a) Same idea as in (4): a = b = c = d = 0 => all solutions; a =

4. For comparison, calculate b - a and c - a for the vectors in Exercises 3a,

2. (a) By definition, the angle is cos- 1 (ae b/llall2l1bll2) = cos- 1 ([ J3J3 +

equally long, are sides of a rhombus. In the sketch, a + b is the diagonal

3. This is vector-space material.

(a) x. 0 = x • (0 + 0) (vector identity) = x • 0 + x • 0 (Theorem

4. Angle is -IT { } x. y/llxl1211yll2 = -1 =} x. Y = -llxll2l1yll2 =} one

5. (a) By definition, IIx ± ylll = (x ± y) • (x ± y), which by numerous uses

Exercise 5b Exercise 6b Exercise 8c

6. (a) By multiple uses of Theorem 1.1, we show that (x + y) • (x - y) =

7. (a) Symmetry: x. y := X1YI + ... + XnYn = Y1Xl + ... + YnXn = Y. x

(b) Distributivity: (x + y) e Z := (Xl + Yl)ZI + ... + (Xn + Yn)Zn =

8. (a) [x-proj(x, y)]ey = xey-proj(x, y)ey := xey-[xey jyey]yey = 0

9. We have seen that IIx - ylll = x e x - 2x e y + Y e y. On the right, x e x

(c) In view of (b), we have II 30c+40dll2 = 1130c-60c1l2 = 11-30c1l2 =

4. Clearly, 11(1, 0) - (-1, 0) II 0 = 2, so this is the length we are seeking. We

5. (a) By Theorem l.l(c), x ex 2: 0, with equality iff x = O. The same holds

Ilaxlio = lal max{lxII, ... , IXnll = lalllxllo.

(a) d(x, y) := IIx - yll = d(y, x) because norms are symmetric;

4. (a) d(ax, ay) := lIax-ayll = lalllx-yil (radial homogeneity of norms)

3. (a) From trigonometry, H = F - 2G.

4. Positive definite: IIfllo := sup If I ~ 0, and 111110 = 0 * If(x)1 = 0 for

6. (a) Yes. The product of integrable functions is integrable.

7. (a) Induction: f(1) = If(I); and f(k) = kf(l) forces f(k + 1) =

(e) (sequence argument) Assume that f is continuous at x = b. For any

2. Write d := IIx - yll. The hypothesis YEN says that d < r.

(a) Pick a fixed s ~ r - d. Then r ~ d + s, and we are in the situation of

3. (a) If r + t exceeded d, then the neighborhoods would intersect (Theorem

Parts (b) and (c) established the smallest conceivable distance d - r - t.

4. By Exercise 2, some neighborhood N(u, p) is contained in N(x, r) and

7. (a) Exercise lb.

sublimit = (-1, 0); or they become 0, and we apply similar reasoning

(b) The coordinates look like x = sins/s -+ 1 (as s -+ 0), y = (1 -

5. Yes, under our convention that R 1 is not the same as R. In R 1 , Xi := [_I]i i

8. (a) A sublimit is the limit of some subsequence. If the parent sequence is

9. => Assume that x is a sublimit of (Xi) and N a neighborhood of x. By

{= Assume that every neighborhood of x has infinitely many terms. Then

(b) Let 2k + 1 ~ i ~ 2k+1. We know that Hi 2: 0, and Hi is constantly 1

3. Nowhere. The upper horizontals for each group (graphs number 2k + 1 to

4. (a) gl (x) == 0 and g2(X) == 1. If g(a) = 0 and g(b) = 1, then (intermedi-

5. (a) See the solution to Exercise 7b in Section 2.3.

~ Iff(x) - fey) = A[x - y] for all vectors, then f(x) - Ax = fey) - Ay =

2. (a) g\(x, y) := (1/[x 2 + y2], 1/[x 2 + y2], 1/[(x - 1)2 + (y - 1)2]) is

4. ae - bd -:j:. O. The hypothesis says that

5. Yes. The question becomes whether the system represented by F(x, y, z) =

6. Leti=(j,k, ... ,p),j=(l,m, ... ,q).

(a) (axi)(,Bxj) := (ax/xl ... x/)(,Bx/ x 2m ... x nq ) = (a,B)x/+ 1x 2k+ m

(b) A polynomial is a sum of terms, and the term ax/ x 2k ... x/ is

(c) g(x, y, z) = x 2y 3Z4 + 5x 6y 7Z8 has degree 21. In general, ax i + ... +

3. (a) No, because G(i, i) = i2 ~ 00, while G(i, 0) = 0 ~ O.

(d) Yes. ,JG is undefined in quadrants II and IV, but if Xi ~ 0 in the

4. (a) At 0, limit is 00, because JXi Yi ~ 0+, as in Exercise 3d. At 00, no

5. Let c be a closure point of S. By definition, some sequence (Xi) from S has

6. (a) =} Assume that d is an accumulation point of S. The neighborhood

{= Trivial; the hypothesis is stronger than the definition of accumula-

7. (a) Along y = ax, H(x, y) = a 2 x 2 /x = a 2 x. Since (x, y) -+ (0,0)

8. (a) No. If Xi := (_I)i Ii, then Xi E R# and Xi -+ 0, but K(Xi) = (-I)i

2. Compare (1): (x, y) -+ (rr, 0) => x -+ rr, y -+ 0 (Theorem 3.7) => xy -+ 0