CHAPTER 4 DERIVATIVES BY THE CHAIN RULE

4.1 The Chain Rule (page 158)

z = f (g(z)) comes from z = f (y) and y = g(x). At z = 2 the chain (z2 - 1)3equals 3' = 27. Its inside
function is y = x2 - 1, its outside function is z = ys. Then dzldx equals Sy2dy/dx. The first factor is
evaluated at y = x2 - 1(not at y = z) . For z = sin(z4 - 1)the derivative is 4uS cos(x4 - 1).The triple chain
z = cos(x + 1)2 has a shift and a s q u a r e and a cosine. Then dz/dz = 2 cos(x I)(- sin(x 1)). + +
The proof of the chain rule begins with Az/Az = (Az/Ay)(Ay/Ax) and ends with dz/dx= ( d ~ / d y )(dy/dx).
Changing letters, y = cosu(z) has dy/dz = -sin u ( x ) g . The power rule for y = [u(z)ln is the chain rule
dy/dz = nun-' du. The slope of 5g(z) is 5gt(x) and the slope of g(5z) is 5gt(5x). When f = cosine and g =
i?
sine and z = 0, t e numbers f (g(z)) and g(f (2)) and f (z)g(z) are 1 and sin 1and 0.

1z=y3,y=z2-3,z'=6z(z2-3)2 ~z=co~y,~=z~,z'=-3z~sinz~
5 z = f i , y = sinz, z' = c o s z / 2 d Z E 7 z = tany+ (l/tanz), y = l/z,zt = (9)sec2($) - ( t a n ~ ) - ~ s e c ' z
9z=cosy,y=z2+z+1,z'=-(2z+1)sin(z2+z+1) 1117cos17z lJsin(cosz)sinz
15z2cosz+2zsinz 1 7 ( c o s d ~ ) ~ ( ~ + 1 ) - ' 1 1~9 ~ ( 1 + s i n ~ ) - ~ ~ ~ ( c o s2z1)c 0 s ( & ) ( ; ~ 7 ~ )
2 3 8 ~ ' = 2 ( ~ ~ ) ~ ( 2 ~ ~ ) (225 z2 () z + 1 ) + ~ 0 s ( z + 7 r ) = 2 z + 2 - ~ o s z
27 (z2+ 1)2+ 1; sin U from 0 to sin 1; U(sin z) is 1and 0 with period 27r; R from 0 to z; R(sin z) is half-waves.
29 g(z) = z + 2, h(z) = z2 + 2; k(z) = 3 31 f t ( f (2))ft(z); no; ( - ~ / ( l / z ) ~ ) ( - l / z ~=) 1and f (f =z (2))

Js!j(iz+8)+8;$~+14;& 35f(g(z))=z,g(f(y))=y
37 f (g(z)) = &,g(f (2)) = 1- f ' f (f (2)) = Z = 9(9(z)), 9(f (g(z))) = = f (9(f (2)))
39 f (y) = y - 1,g(z) = 1 + +
4 3 2 cos(z2 1) - 4z2 sin(z2 1); -(z2 - I)-~/'; -(cos + (sin &)/4z3I2
45 ft(u(t))u'(t) 4 7 (cos2 u(z) - sin2 u ( z ) ) e 4 9 2zu(z) +x 2 e 51 1 / 4 d E d G
5 3 dfldt 55 f'(g(z))g'(z) = 4(z3)33z2 = 12z1' 5 7 3600; i;18 59 3; $

2 f (y) = y2; g(z) = z3 - 3

9
a
dx
= 6z2(z3 - 3) 4 f (y) = tan y; g(z) = 22; 2 = 2 sec2 22
6 f(y) = sin y; g(z) = &;2 = 2 6 8 f (y) = sin y; g(z) = cos z; 2 = - sin x cos(cos z)
f (y) = &;g(z) = z2; 2 = (1)(2z)
2 f i
= 1 1 2 2 = sec2(z + 1) 1 4 2 = 3z2 16 a a,
= 27
T6=
& = ~08(2+1) 20 2 = 22 2 = 4z(sin z2)(cos z2)
dz 24s- 2J;E
2 = s(sx)~(J) or = 27z3 and e = 8 l X 2 26 L
2 k - c O s x s i n ~- c o s x or z = sinz and 2 = c o s x
dx-2JCZiG-
28 f ( y ) = y + l ; h ( y ) = * ; k ( y ) ~ 1 3 0 f(y) = f i , g ( z ) = 1 - z 2 ; f ( y ) = d = , g ( x ) = z 2
32 (a) 22 (b) 4 f'(5) (c) 8 (d) 4 3 4 C = 16 because this solves C = +C 8 (fixed point) +
36 ~ ( Y I g, (z), ~ f ( g ( z )-) 91 < E
38 For g(g(z)) = z the graph of g should be symmetric across t h e 45' line: If the point (z, y)
.
is on the graph so is (y, z) Examples: g(z) = - or -x or ! q s .
4.2 Implieit Differentiation and Related Rates (page 163)

40 False (The chain rule produces -1 : so derivatives of even functions are odd functions)
False (The derivative of f (z) = z is f' (z) = 1)False (The derivative of f ( l l x ) is f ' ( l l z ) times - 1/z2)
T N e (The factor from the chain rule is 1) False (see equation (8)).
42 Fkom z = go up to y = sin :. Then go across to the parabola 2 = y2. Read off z = d.;lf on the
horizontal z axis.
44 This is the chain rule applied to (a function of y). Its z derivative is its y derivative ($) times 2.
If z = y2 and y = z3 then 2
= 2y and $2 = 2(3x2). Check another way: = 2z3 and d(&)
d x dy
= 6 x2 .
46 2= (3u2)(3%') = 9z8 48 = 2 *% so g = 22zf'(zZ) + 2f(z)%
52 2 = -nu(t)-n-l% 54 $ = - 5: 56 cos(sin z) cos z
58 (a) 53 (sum rule for derivatives) (b) 60 (chain rule)
60 Note that G' = cos(sin z) cos z and G" = - cos(sin z) sin z - sin(sin z)cos2 z. We were told that
H(z) = cos(cos z) should be included too.

4.2 Implicit Differentiation and Related Rates (page 163)

+
For z3 y3 = 2 the derivative dy/dz comes from implicit differentiation. We don't have to solve for y.
+ 2
Term by term the derivative is 3z2 sy2 = 0. Solving for dyldz gives -x2 /y2. At z = y = 1 this slope is
-1. The equation of the tangent line is y - 1 = -l(x - 1).

A second example is y2 = z. The z derivative of this equation is 2 yd 2 = 1. Therefore dyldz = 1/2y.

Replacing y by fi this is dyldz = 1/2+.

In related rates, we are given dg/dt and we want df /dt. We need a relation between f and g. If f = g2, then
+
(dfldt) = 2g(dg/dt). If f 2 92 = 1, then dfldt = - f % . If the sides of a cube grow by dsldt = 2, then its
volume grows by dV/dt = 3s2(2) = 6s2. To find a number (8 is wrong), you also need to know s.

1 - 9 - 1 /Yn-' 3 2 = 1 5&=& 7 (y2 - 2zy)/(z2 - 2zy) or 1 1 1

dx F1(y) 9Z50riT7
11 First 2 = -ji, second 2= 1 3 Faster, faster 15 222' = 2yy' + 2' = Ey' = y1sin9
17sec28=& 1 9 5 0 0 $ - ; ~ o o J ~ 21 $=-$;$=-2J;j;oothen0
23 V=Tr2h.dh
d t
I av - -- in/sec
= --
41rdt
A 25 A = i a b sin 8, % =7 27 1.6 m/sec; 9 m/sec; 12.8 rn/sec

+ 2 y g = 0 gives 2 = 1at (1,l); 22 + 2(y - 2 ) g = 0 also gives 2 = 1.

1 2 2(x - 2)

14 2 + 2 y 9 + ~ ( 2 ~ 9= - j-
= 0)yields
 4.3 Inverse f i n ctions and Their Derivatives (page 170)

16 y catches up to z as 9 increases to t . So y' should be larger than d. 18 y' approaches 200sc/200a = c

2 0 z is a constant (fixed at 7) and therefore a c h a n g e A x is not allowed
2 2 z 2 + y 2 = 1 0 2 s 0 2 z ~ + 2 y=Oand
~ 2
=-'dl = -2 '= -c when z = i c y . This means ( i c y ) 2+y2 = lo2
Y dt J
or y =

2 4 Distance to you is d m , rate of change is -2 with 2 = 560. (a) Distance = 16 and z = 8&
and rate is e ( 5 6 0 ) = 2 8 0 6 ; (b) z = 8 and rate is L(560)
,/iGiP = 280fi; (c) z = 0 and rate is zero.
2 6 10c(t - 3) = 8t divided by c(t - 3) = 4 gives 10 = 2t. So t = 5 and c = 2. The z and y distances between
ball and receiver are 2t - 10 and 12t - 60. The derivative of d ( 2 t - + (12t - 60)2
= , m i l t - 51 is -,hz.
2 8 Volume = $xr3 has = 47rr2$. If this equals twice the surface area 4ar2 (with minus for evaporation)
than 2 = -2.
SO $ = 4 a radianslsecond; 0 = 22% - 6 cos 9 2 + 6 2 sin 9 g; at 9 = :, z = 3& and 6fig + 18&%
gives 2 = - 1 2 ~ ;at B = T, z = 0 and 2 = 0.

4.3 Inverse Functions and Their Derivatives (page 170)

The functions g (x) = z - 4 and f (y) = y + 4 are inverse functions, because f (g(2)) = x. Also g (f (y)) = y.
The notation is f = g-' and g = f - l . The composition o f f a n d f - l is the identity function. By definition
z = gml(y) if and only if y = g(x). When y is in the range of g, it is in the domain of g-'. Similarly z is in
the domain of g when it is in the range of g-'. If g has an inverse then g(zl)#g(zz) at any two points. The
function g must be steadily increasing or steadily decreasing.

The chain rule applied to f (g(z)) = z gives (df/dy)(dg/dx) = 1. The slope of g-' times the slope of g
equals 1. More directly dxldy = l / ( d y / d x ) . For y = 22 1 and z = + i(y
- I), the slopes are dyldx = 2 and
1
dzldy = 3 . For y = z2 and z = a,
the slopes are dyldz = 2 x and dz/dy = 1/2&. Substituting x2 for y
gives d z l d y = 1/2x. Then (dz/dy)(dy/dx) = 1.

The graph of y = g(x) is also the graph of z = g-l(y), but with x across and y up. For an ordinary graph
of g-', take the reflection in the line y = z. If (3,8) is on the graph of g, then its mirror image (8,S) is on the
graph of g-'. Those particular points satisfy 8 = Z3 and 3 = log2 8.

The inverse of the chain z = h(g(z)) is the chain z = g-l(h-l(z)). If g(z) = 32 and h(y) = y3 then
z = (Sx)' = 27xS. Its inverse is z = ;z1/', which is the composition of g-l(y) = and h-'(z) = z1lS. iy
1s = Y+6
3
3x= ( z unrestricted -+ no inverse) 5 z = - s+ I 7 z = (1 + y) ' I 3
9 ( z unrestricted + no inverse) 11y = 1
2-0 1S2<f-'(z)<3 15fgoesupanddown
1 7 f(z)g(z) and &J 19 m # O ; m L O ; Iml 1 21 2 = 5 z 4 , z = ky-'/'
2S!&=3z2.dl=$(l+y)-2/3
dz ' dY 25$=
(=-')a
-1 dl=
dY
-1
( ~ - 1 ) ~ 2 7 y; y2 C +
2 9 f (g(z)) = -1/3z3; g-'(y) = $;g(ge'(z)) =z 39 2/& 4 1 1/6cos 9
4.4 In verses of Digonometric h n c tions (page 175)

4 3 Decreasing; 2 = &<0 45 F; T ; F 47 g(z) = z m , f (y) = y" ,z = ( z ' / " ) ' / ~

4 9 g(z) = z3, f (y) = y + 6, z = (z - 6)'13 51 g(z) = loz, f (y) = logy, z = log(lO9) = y
53 y = z3, y" = 62, bdz/d# = -$ y-5/3; m / see2, sec /m2 55 p = - 1; 0 < y 5 1
fi
5 7 max = G = 2 'I3 G' = 59 y2/100
8y 9 2y

2 z= 9 4 z = &(f-' matches f ) 6 no inverse 8z= { Y

1
jY
YLO
10 z = Y5

1 2 The graph is a hyperbola, s y m m e t r i c a c r o s s the 45' line; 2 = -&; 2 = -i(z - 1)2 (or -*).( Y 1)
1 4 f-' does not exist because f (3) is the same as f (5).
16 No two x's give the same y. 18 y = 5 and y = 2 - z (functions of z + y and zy lead t o suitable f )
2 0 The inverse of a piecewise linear function is piecewise l i n e a r (if the inverse exists).
2 2 !g = -- dZL=- . 1 (- 1 ) 2 4 5k=-,.&- 1 -413 2 6 2=dd%.dZ- a d - b c
,r 9 dy - - %Y (cx+d)' ' d~ - ((ey-a)2
a
(x-1)" dY y2 '

28 2 = y. 30 jumps a t 0, yl ,y2 to heights z l ,z2, z3; a piecewise const ant function has no inverse.
3 2 Hyperbola centered a t (-1,O) : shift the standard hyperbola zy = 1.
3 4 y = -3s for z < 0;y = -z for z 2 0. 36 The graph is the first quarter of the unit circle.
38 The graph starts a t (0,l) and increases with vertical asymptote a t z = 1.
4 0 1= see2 z g so dz
dY
= cos2 z = 1 42 2
= 1- cos z = 0 so $
= CQ. (The derivative does not exist.)
4 4 F i r s t p r o o f Suppose y = f (z). We are given that y > x. This is the same as y > f
S e c o n d p r o o f The graph of f (z) is above the 45' line, because f (z) > x. The mirror image is below
the 45' line so f-'(y) < y.
4 6 g(z) = z - 4 , f ( y ) = 5 ~ , g - ' ( ~=) y + 4 , f-'(2) = t , x = gl z + 4 .
4 8 g(z) = z + 6 , f ( y ) = y3, g-'(y) = y - 6, f-'(2) = G ; x = G- 6
50 g(z) = $ z + 4, f (y) = g(y), 9-'(y) = 2y - 8, f-'(2) = 9 - ' ( 2 ) ; ~ = 2(2z - 8) - 8 = 4 z - 24.
5 2 z* = f-'(0)
5 4 f-'(0) f-'(y) + (%)(0 - y) is the same as z * r; z + &(0 f (z)),which gives Newton's method.
56 dY
=f
fi

t o f(f-'(y))
+ Ys-' s.
- F'( f ( y))
= y. This leaves the first term = f-'(y).
-
The second term cancels the third because F'( f
G i s t h e a n t i d e r i v a t i v e o f f-'
is equal
if F' = f .
58 To maximize yz - F ( z ) set the z derivative to zero: y = = f (z) or z = f-'(y). Substitute this
z into zy - F ( x ) : the maximum value is exactly G(y) from Problem 56. Now maximize zy - G(y).
The y derivative gives z = or by Problem 56 z = f -'( y). Substitute y = f (z) into z y - G(y)
to find that the maximum value is z f ( z ) - G ( f ( z ) ) = z f ( z ) - [ f ( z ) z - F ( f - ' ( f ( z ) ) ] = F ( z ) .
N o t e : This is the L e g e n d r e t r a n s f o r m between F ( z ) and G(y) - important but not well known.
Since is increasing (then f -'exists), the function F ( x ) is convex (concave up). So is G(y).

4.4 Inverses of Trigonometric F'unctions (page 175)

The relation z = sin-' y means that y is the sine of x. Thus z is the angle whose sine is y. The number y
lies between -1 and 1. The angle z lies between -7r/2 and n/2. (If we want the inverse to exist, there cannot
be two angles with the same sine.) The cosine of the angle sin-' y is 43. The derivative of z = sin-' y is
 4.4 Inverses of 'Rigonometric finctions (page 175)

The relation z = cos-' y means that y equals cos x. Again the number y lies between -1 and 1. This time
the angle z lies between 0 and a (so that each y comes from only one angle z). The sum sin-' +cos-' y = r/2.
(The angles are called complementary, and they add to a right angle.) Therefore the derivative of x = cos- y
is dzldy = -1143, the same as for sin-' y except for a minus sign.

The relation x = tan-' y means that y = t a n x. The number y lies between -oo and oo. The angle z lies
be tween -a/2 and a/2. The derivative is dzldy = 1/(1+ y2). Since tan-' y cot-' y = a/2, the derivative of +
cot-' y is the same except for a minus sign.

The relation z = sec-' y means that y = sec x. The number y never lies between -1 and 1. The angle x lies
between O and a, but never at x = */a. The derivative of x = sec-' y is dxldy = 1/ I y I J y G .

1O , t , O

lSy=O:l,-1,1;y=
21 2
- 2sin-I z
S ;,0,
gsinx=Jl-yal-ya;Jwandl
:
1:0,0,?
5 a is outside

sin-'
l l

15F;F;T;T; F;F
25 d5 =
1-5, $1
~dy c

1
17 &
dx
o

= -- 1
7 y = -&/2
s ~ =

,/C? l 9 2 = 3
~
dsr
and &/2
+
-
~ -
con x -
J
,* - -

,/- a$ l- jqar dy IU+~I\/Y'+~~

27 u = 1so =0 31 secx =

33 $, 1,$ SS -y/Jw 3 7 $ sec tan 39 nxn-' 4 l $ = 1L+ ~ 1

4 3 $ = .kr 47 u = 4sin-' y 49 a 5 1 -u/4

2 sin-'(-1) =- 5;cos-' (-1) = a; tan-'(-1) = - f . Note that - 5,a, -:

are in the required ranges.
4 sin-' & doesn't & doesn't exist; tan-' fi = t .
exist; cos-'
6 The range of sin-'(y) is -:
5 x 5 5. Note that sin 2x = 0 but 2x is not sin-' 0.
a&- 1 - 1 . The graph goes from y = -T to y = a.
dy - - 4s
1 0 The sides of the triangle are y, J-, and 1. The tangent is --x.
6 P
12 (sin x) cos x equals cos x = 1as required.
14 vlx=o
-lx=, = 1;
-
1 sin1 x
;,-
= d(tan-'
dy y) IX=O = -1~=1
d sin-
= &;
-I.=, = a* 1
1 6 cos-' (sin x) is the complementary angle 5 - x. The tangent of that angle is SlIL X = cot x.
d. = .&qd2)==*
18 a 2 2 0 g =d+(-sinz)
1 (cmx)
= -1. Check: a = ? - z so 2 = -1.
22 dl = - l ( s i n - ' z ) - 2 A
dx 24 3 = 2x tan-' z + (1+ x2) & = 2 x tan-' x + 1.
,/GP*
26u=x2sog=&. 28&=-
1+y2
dg
. The range of this function is 0 5 y 5 5.
30 The right tria~glehas far side y and near side 1. Then the near angle is tan-' y. That angle is also cot-'(;).
34 The requirement is u' = &. TOsatisfy this requirement take u = tan-%.
36 u = tan-'y has $= 1 2 1
4.4 Inverses of ?tigonometric Functions (page 175)

40 By the chain rule 2 = ,1

x- (sec2 z).
42 By the product rule 2 = (cos %)(sin-' z) + ( s i n z ) e . Note that a # z and 2 # 1.
1-22
44 d' = cos(cos-' z)(*) + sin(sin-' z)(--) =
1
= 0.
46 Domain lyl 2 1;range -: 5 z 5 5 with z = 0 deleted.
48 u(z) = 3 t a n - l 2 x (need 4 to cancel 2 from the chain rule).
60 u(x) = has 2 = = F5. Then $ tan-' u(z) = ' d" = 2 -
x+1 x+ 1) z 1+(*)"(2+1)' -
2
(x+l)'+(x-1)' -
-x +. +1 This is also the derivative of t a n - l x ! So tan-' u(z) minus tan-' z is a constant.
52 Problem 5 1 finds u(0) = - 1 and tan-' u(0) = - and tan-' 0 = 0 and therefore tan-' u(z) - tan-' z should
have the constant value - f - 0. But as z -+ -00 we now find u -+ 1and tan-' u -+ f and the difference
is - (- ); = y.T h e "constant"has changed! It happened when z passed -1 and u became
infinite and the angle tan-' u jumped.
MIT OpenCourseWare
https://ocw.mit.edu

Resource: Calculus
Gilbert Strang

The following may not correspond to a particular course on MIT OpenCourseWare, but has been
provided by the author as an individual learning resource.

For information about citing these materials or our Terms of Use, visit: https://ocw.mit.edu/terms.