Multivariable Calculus

Don Shimamoto

The book was last updated on July 10, 2024.

There are two pdf versions of the book. In one, the hyperlinks are enabled, and, in the
other, they are not. The first may be a more suitable format for onscreen reading, the second
more suitable for printing.

Main changes since the November 2, 2023 version

(not including changes in punctuation or minor changes in wording)

The chain rule: Chapter 6, Sections 3 and 4. The conclusion of the chain rule
really has two parts, namely, if f is differentiable at a and g is differentiable at f (a), then
(1) g ◦ f is differentiable at a and (2) D(g ◦ f )(a) = Dg(f (a)) · Df (a). The discussion in
the text focuses almost exclusively on (2) and does not really address (1). To correct (and
acknowledge) this, the wording has been changed in a couple of places and a new exercise
covering a complete proof of the chain rule has been added at the end of the chapter. Here
are the specifics.

• p. 162, third paragraph.

The references to the letter ϵ and to a discussion of the proof of the chain rule in the
next section have been deleted. In their place is a reference to the new exercise. Here
is the revised paragraph:

The line of reasoning we used to obtain the chain rule is natural and rea-
sonably straightforward, but a proper proof requires greater attention to
the various approximations involved. The arguments are a little like those
needed for a correct proof of the Little Chain Rule back in Exercise 5.4 of
Chapter 4, though the details here are more complicated. We leave a full
proof of the chain rule to the exercises, namely, Exercise 4.13 at the very end
of this chapter.

• p. 165, second complete paragraph.

This paragraph, which talks about using the Little Chain Rule to prove the general
chain rule, has been replaced by the following:

This approach provides another way of coming up with the matrix formula
for D(g ◦ f )(a), but we should point out that it does not address the part of
the chain rule that states that g ◦ f is differentiable at a in the first place.
For that, see Exercise 4.13.

• pp. 171–172, Exercise 4.13 (new).

The full text of the exercise is reproduced below. It’s pretty long.
4.13 This exercise supplies a proof of the chain rule (Theorem 6.9). So let U and
V be open sets in Rn and Rm , respectively; let f : U → V and g : V → Rp
be functions; and let a be a point of U . The chain rule states that, if f is
differentiable at a and g is diffferentiable at f (a), then g ◦ f is differentiable at a
and D(g◦f )(a) = Dg(f (a))·Df (a). To prove this, we show that Dg(f (a))·Df (a)
satisfies the condition required of D(g ◦ f )(a) in the definition of differentiability.
That is, we show that:

g(f (x)) − g(f (a)) − Dg(f (a)) · Df (a) · (x − a)

lim = 0. (6.11)
x→a ∥x − a∥

(See (4.4) in Chapter 4 and the discussion that follows it for an explanation of
why any matrix that satisfies the definition must necessarily be D(g ◦ f )(a).)
(a) If x ̸= a, show that:

∥g(f (x)) − g(f (a)) − Dg(f (a)) · Df (a) · (x − a)∥

≤ A(x) + B(x), (6.12)
∥x − a∥

∥g(f (x)) − g(f (a)) − Dg(f (a)) · (f (x) − f (a))∥

where A(x) =
∥x − a∥
∥Dg(f (a)) · (f (x) − f (a)) − Dg(f (a)) · Df (a) · (x − a)∥
and B(x) = .
∥x − a∥

(Hint: Add zero, and use the triangle inequality.)

The left side of inequality (6.12) is the norm of the quotient in equation (6.11).
If we could prove that limx→a A(x) = 0 and limx→a B(x) = 0, then (6.12)
would imply that the limit in (6.11) is 0 as well. In other words, we would
have proven the chain rule. This strategy is carried out as follows: the limit
involving B(x) is part (b) below, and the limit involving A(x) is part (f).
(b) Show that limx→a B(x) = 0. (Hint: Use the differentiability of f at a.)
Next, let b = f (a). Then g is differentiable at b by assumption. Define a
function Q : V → Rp by:
(
g(y)−g(b)−Dg(b)·(y−b)
∥y−b∥
if y ̸= b
Q(y) =
0 if y = b.

Note that the quotient is the one that appears in the definition of differentia-
bility of g at b.
(c) If x ̸= a, show that A(x) = ∥Q(f (x))∥ · ∥f∥x−a∥
(x)−b∥
.
The next step, seemingly unrelated, shall be used to obtain an estimate of
the factor ∥f∥x−a∥
(x)−b∥
= ∥f (x)−f
∥x−a∥
(a)∥
that appears in part (c).

2
 (d) Show that there exists a constant M such that:

∥Df (a) · v∥ ≤ M ∥v∥ for all v in Rn .

 ∇f (a)·v 
1
∇f2 (a)·v
(Hint: Start by explaining why Df (a) · v =  .. , where f (x) =
.
∇fn (a)·v
(f1 (x), f2 (x), . . . , fn (x)). Then let M = ∥∇f1 (a)∥2 + ∥∇f2 (a)∥2 + · · · +

1/2
∥∇fn (a)∥2 . The Cauchy-Schwarz inequality may come in handy along the
way.) In fact, the argument can be modified to show that in general, given
any m by n matrix A, there exists a constant M such that ∥Av∥ ≤ M ∥v∥
for all v in Rn .
(e) If x ̸= a, show that:
 
∥f (x) − f (a) − Df (a) · (x − a)∥
A(x) ≤ ∥Q(f (x))∥ · +M .
∥x − a∥
(f ) Show that limx→a A(x) = 0.
As observed above, parts (b) and (f), together with (6.12), imply (6.11). This
completes the proof of the chain rule.

Main changes since the November 9, 2021, May 2, 2022, and November 14,
2022 versions
(not including changes in punctuation or minor changes in wording)

• p. 83, second paragraph, second sentence. The wording has been tightened up
to read: “Here, it turns out that the limit may fail to exist . . . .” In previous versions,
the word “may” did not appear.

• p. 95, footnote. The URL for the map of isotherms no longer works. Instead, the
map can now be found at:

https://www.noaa.gov/jetstream/surface-temperature-map.

Main changes since the January 4, 2021 and June 7, 2021 versions
(not including changes in punctuation or minor changes in wording)

• p. 41, Proposition 2.23. The cross product is defined only for vectors in R3 , so part
2 of the proposition should include a stipulation that α and β are paths in R3 :

...
2. (Cross product) For paths α, β in R3 , (α × β)′ = α′ × β + α × β ′ .
...

3
• p. 283, remarks before Exercise 4.19. There is a typographical error three lines up
from where the exercise begins. According to our notational convention, the integral
of the differential form f dy ∧ dx should have one integral Rsign, not two, so the correct
statement is: “ . .RR. as an integral of a differential form, D f dy ∧ dx is equal to the
Riemann integral D f (x, y) dx dy.”

Main changes since the October 15, 2020 version

(not including changes in punctuation or minor changes in wording)
• p. 32, Figure 2.10. A footnote has been added to the caption:
1
From now on, we won’t always state that a vector has been translated.
Any vector that starts at a point other than the origin should be understood
to be a translated copy whether there is an explicit note to that effect or
not.
This increases the numerical labels of all the other footnotes by 1.
• p. 153, Exercise 6.4. The wording has been clarified to indicate that the xy-plane
itself, z = 0, is also one of the surfaces that bounds W :
. . . bounded by the surface y = x2 and the planes z = y, y = 1, and z = 0.
• p. 187, equation (7.5). p The lower endpoint ofpthe innermost integral has been
corrected. It should be − 1 − x21 − x22 − x23 , not − 1 − x22 − x22 − x23 :
Z Z Z Z √1−x21 −x22 −x23 
Vol (W ) = √ 1 dx4 dx1 dx2 dx3 (7.5)
B − 1−x21 −x22 −x23

= ...

• p. 218, Example 9.12. In the first set of displayed equations, the middle equation
should have two integral signs:
···
ZZ


= y − (−2y) dx dy
D
···

• p. 285, answer to Section 1.2, Exercise 2.2. The answer begins with a typo:
“(y1 + y2 )” should be “(y1 , y2 ).” Perhaps it’s better anyway to write the solution as
follows instead:
2.2 T (x + y) = T (x1 + x2 , y1 + y2 ) = (x1 + y1 , x2 + y2 , 0) = (x1 , x2 , 0) + (y1 , y2 , 0) =
T (x) + T (y)
T (cx) = T (cx1 , cx2 ) = (cx1 , cx2 , 0) = c(x1 , x2 , 0) = c T (x)

4
 Main changes since the July 20, 2020 version
(not including changes in punctuation or minor changes in wording)

• p. 65, Example 3.16. In the definition of the set K, the plainface x should be a
boldface x: “Let K = {x ∈ R2 : ∥x∥ ≤ 1}. . . . ”

• p. 104, first paragraph. The operator ∇·v is better written as v ·∇, so the notation
has been changed:

. . . each additional derivative of g corresponds to applying the “operator”

∂ ∂ ∂ ∂
v · ∇ = (h, k) · ( ∂x , ∂y ) = h ∂x + k ∂y . ...

• p. 118, Exercise 11.3, second paragraph. As in the previous item, ∇ · v has been
∂ ∂
changed to v · ∇: “ . . . the powers of the operator v · ∇ = h ∂x + k ∂y ...”

• p. 266, Exercise 3.5, parts (a) and (b). The exercise has been reworded to resolve
potential ambiguity about the names of the coordinates in R3 , namely:

3.5 (a) Find an example of a smooth vector field F = (F1 , F2 , F3 ) defined on an open
set U of R3 such that:
• its mixed partials are equal, i.e., ∂F
∂y
1
= ∂F
∂x
2
, ∂F
∂z
1
= ∂F
∂x
3
, and ∂F
∂z
2
= ∂F
∂y
3
,
and
• thereRexists a piecewise smooth oriented simple closed curve C in U such
that C F1 dx + F2 dy + F3 dz ̸= 0.
(b) On the other hand, show that, if F is any smooth R vector field on an open set
3
U in R whose mixed partials are equal, then C F1 dx + F2 dy + F3 dz = 0 for
any piecewise smooth oriented simple closed curve C that is the boundary of
an oriented surface S contained in U .
(c) . . .

Main changes since the May 4, 2020 version

(not including changes in punctuation or minor changes in wording)

• p. 151, Exercise 4.1(c) and p. 169, Exercise 4.3. In these exercises, the spherical
coordinate ϕ was typeset as “φ,” which is inconsistent with the rest of the book. This
has been corrected.

• p. 252, definition of closed surface. The definition now includes the assumption
that the surface is path-connected. That is:

Definition. Let S be a piecewise smooth surface in Rn such that every pair

of points in S can be joined by a piecewise smooth curve in S. Then S is
called closed if ∂S = ∅ (the empty set).

5
 Main changes since the March 4, 2020 version
(not including changes in punctuation or minor changes in wording)

• p. 68, top line. The expression for f (x, 0) has been corrected: “ . . . f (x, 0) = x·0
x2 +02
=
0 ...”

• p. 87, equation (4.6). There are no cross-references to the equation, so its label has
been removed. As a result, the labeled equations that follow in the rest of the chapter
have their numbers decreased by 1, up to (4.30) on p. 117.

• p. 101, Figure 4.9 and p. 115, Figure 4.14. The viewpoint has been rotated so
that x and y increase in the usual directions.

• p. 139, sentence right before Example 5.16. For the spherical coordinate ϕ, we
use the interval 0 ≤ ϕ ≤ π to describe all of R3 , but the value ϕ = π was also included
among the unnecessary redundant values. The corrected sentence is: “This is why
values of ϕ in the interval π < ϕ ≤ 2π are not needed—they would duplicate points
already covered.”

• p. 247, equation (10.7). This is another labeled equation to which there are no
cross-references, so the label has been removed. The labeled equations that follow in
the rest of the chapter have their numbers decreased by 1, up to (10.31) on p. 269.

• p. 292, answer to Chapter 3, Exercise 2.7. For the surface in the middle, the
viewpoint has been rotated so that x and y increase in the usual directions.

Main changes since the January 8, 2020 version

(not including changes in punctuation or minor changes in wording)

• p. 27, Example 2.4. The direction vector v should be nonzero: “Given a point a
and a nonzero vector v in Rn , . . . ”

• p. 77, Exercise 6.1. Change to a function of two variables: “Let a = (1, 2). Show
that the function f : R2 → R given by f (x) = a · x is continuous.”
The answer on p. 293 must be modified, too: “f (x, y) = (1, 2) · (x, y) = x + 2y. We
know that the projections x and y are continuous, so, as an algebraic combination of
continuous functions, f is continuous as well.”

• p. 79, Exercise 8.3. This exercise is new.

8.3 (Sandwich principles.) Let U be an open set in Rn , and let a be a point of U .

(a) Let f, g, h : U → R be real-valued functions such that f (x) ≤ g(x) ≤ h(x) for
all x in U . If f (a) = h(a)—let’s call the common value c—and if f and h are
continuous at a, prove that g(a) = c and that g is continuous at a as well.

6
 (b) Let f, g, h be real-valued functions defined on U , except possibly at the point
a, such that f (x) ≤ g(x) ≤ h(x) for all x in U , except possibly when x = a.
If limx→a f (x) = limx→a h(x) = L, prove that limx→a g(x) = L, too.

• p. 98, Section 4.9, last paragraph. The discussion of differentiability, smooth-

ness, and continuity when the domain D is not open is meant to apply to real-valued
functions. The same comments remain valid for vector-valued functions f : D → Rm ,
but that would be getting ahead of ourselves at this point of the book. Also, the
accommodation in the case of continuity is spelled out in the more customary way.

. . . If D is a subset of Rn , not necessarily open, we say that a function f : D →

R is differentiable at a point a in D if it agrees with a differentiable function
near a, that is, there is an open ball B containing a and a function g : B → R
such that g(x) = f (x) for all x in D∩B and g is differentiable at a in the sense
previously defined. We make the analogous modification for smoothness.
For continuity, we keep the definition as before but restrict our attention
to points where f is defined: f is continuous at a if, given any open ball
B(f (a), ϵ) about f (a), there exists an open ball B(a, δ) about a such that
f (B(a, δ) ∩ D) ⊂ B(f (a), ϵ). . . .

The changes have the effect of pushing some of the material onto the next page, but,
by p. 101, everything is back to normal.

• p. 109, Exercise 1.14. Change the point of interest from (1, 0) to (1, π): “. . . Eval-
uate ∂f
∂y
(1, π). . . . ”

• p. 117, Exercise 10.10. There are at least two data points: “Let (x1 , y1 ), (x2 , y2 ), . . . ,
(xn , yn ) be a collection of n distinct points in R2 , where n ≥ 2. . . . ”

• p. 119, Exercise 11.5(b). Add a hint: “. . . Hint: In the case that C = 0, too,
consider the expansions of (a ± b)2 .”

• p. 123, Example 5.1. In the sequence of displayed equations in the middle of the
page, the third equation should be with respect to y, not x:
..
.
Z 4
9 1

= (12 − + 6y) − (4 − + 2y) dy
2 2 2
..
.

• p. 167, Exercise 2.8. Spherical coordinates were meant to be written consistently in

the order (ρ, ϕ, θ):

2.8 Let f : R3 → R3 be the spherical coordinate transformation from ρϕθ-space to

xyz-space: f (ρ, ϕ, θ) = (ρ sin ϕ cos θ, ρ sin ϕ sin θ, ρ cos ϕ).
(a) Find Df (ρ, ϕ, θ). . . .

7
• p. 171, Exercise 4.12. f should be a function of two variables, not three: “Let
f (x, y) = x2 + y 2 + xy. . . . ”
• p. 207, equation (9.4). A footnote has been added elaborating on reparametrization:
We assume here that g is smooth. Indeed, equation (9.3) may be the best
way to define what it means in the first place for α and β to parametrize the
same curve C, namely, that there exists a smooth function g : [c, d] → [a, b]
such that β(u) = α(g(u)) and g ′ (u) ̸= 0 for all u in [c, d]. Then, we say that
α and β traverse C in the same direction if g ′ > 0 and in opposite directions
if g ′ < 0.
• p. 218, first displayed equation. In the double integral on the right, a missing dy
has been added:
Z ZZ
2 ∂ ∂
(x − y 2 ) dx dy . . .


(x − y ) dx + xy dy = (xy) −
C=∂D D ∂x ∂y

• p. 221, first displayed equation. This equation has been given a number.
Z
y x
− 2 2
dx + 2 dy = 2π. (9.10)
Ca x +y x + y2
There are cross-references to it in the appropriate places on p. 222 (below the second
displayed equation) and p. 224 (on the next-to-last line). Also, the numbered equations
that follow in the rest of the chapter all have their numbers increased by 1, up to (9.14)
on p. 226.
• p. 222, last sentence of Section 9.5. To include the possibility of curves oriented
clockwise, the sentence should be replaced with the following:
. . . If C is oriented clockwise instead, then the change in orientation reverses
the sign of the integral, so the corresponding values in the two cases are 0 and
−2π, respectively. That there are exactly three possible values of the integral
over all piecewise smooth oriented simple closed curves in R2 − {(0, 0)} may
be somewhat surprising.
• p. 227, Exercise 1.11. In the 1-form that
R is being integrated, xy and xπ have been

changed to x + y and π , respectively: “ C e dx − ln cos2 (x + y) + π x y 1,000,000 dy,
x 8xy

where . . . ”
• p. 228, Exercise 1.13. In the second paragraph, the assumption that g is smooth is
stated explicitly: “ . . . As in equation (9.3), there is a function g : J → I, assumed to
be smooth, such that β(u) = α(g(u)) . . . ”
Also, the first couple of sentences in the last paragraph have been rewritten:
In other words, if x = β(u) = α(t) is any point of C other than an endpoint,
then κβ (u) = κα (t). We denote this common value by κ(x), i.e., κ(x) is
defined to be κα (t) for any smooth parametrization α of C, where t is the
value of the parameter such that α(t) = x. . . .

8
• p. 261, first sentence of Section 10.8. The assumption in the definition of the
surface integral that ∂σ
∂s
× ∂σ
∂t
is nonzero has been added as part of the setup:

Finally, we address the long-standing question

RR of the extentRRto which our
original definition of the surface integral, S F · dS = ± D F(σ(s, t)) ·
∂σ ∂σ


∂s
× ∂t
ds dt, depends on the smooth parametrization σ : D → R3 of S,
where D is a subset of the st-plane and, by assumption, ∂σ
∂s
× ∂σ
∂t
̸= 0, except
possibly on the boundary of D. . . .

• p. 264, Exercise 1.3. The vector field F should be assumed to be continuous: “Let
F be a continuous vector field on R3 . . . ”

• p. 264, Exercise 1.4. The vector field F may be assumed to be continuous: “. . . Let
F be a continuous vector field on U . . . . ”