MA005 3.9 Linear Approximation
MA005 3.9 Linear Approximation
Newton's method used tangent lines to "point toward" a root of the function. In this section we examine and use
another geometric characteristic of tangent lines:
This idea is used to approximate the values of some commonly used functions
and to predict the "error" or uncertainty in a final calculation if we know the
"error" or uncertainty in our original data. Finally, we define and give some
examples of a related concept called the differential of a function.
Linear Approximation
Since this section uses tangent lines frequently, it is worthwhile to recall how we find the equation of the line
tangent to f at a point x = a. The line tangent to f at x = a goes through the point (a, f(a)) and has slope f '(a),
so, using the point–slope form y – y0 = m(x – x0) for linear equations, we have
If f is differentiable at x = a,
then the equation of the line L tangent to f at x = a is
L(x) = f(a) + f '(a).(x – a) .
Example 1: Find the equation of the line L(x) which is tangent to the graph of f(x) = x at the point
(9,3). Evaluate L(9.1) and L(8.88) to approximate 9.1 and 8.88 .
1/2 1 –1/2 1
Solution: f(x) = x = x and f '(x) = x = so f(9) = 3
2 2 x
1 1
and f '(9) = = . Then
2 9 6
1
L(x) = f(9) + f '(9).(x – 9) = 3 + (x – 9) . If x is close to 9, then the value of L(x) is a good approximation of
6
the value of x (Fig. 2) . The number 9.1 is close to 9 so
1
9.1 = f(9.1) ≈ L(9.1) = 3 + (9.1 – 9) = 3.016666.
6
Similarly,
1
8.88 = f(8.88) ≈ L(8.88) = 3 + (8.88 – 9) = 2.98. In fact, 9.1 ≈ 3.016621, so our estimate, using
6
L(9.1), is within 0.000045 of the exact answer. 8.88 ≈ 2.979933 (accurate to 6 decimal places) and our
estimate is within 0.00007 of the exact answer.
In each example, we got a good estimate of a square root with very little work. The graph indicates
the tangent line L(x) is slightly above f(x), and each estimate is slightly larger than the exact value.
Practice 1: Find the equation of the line L(x) tangent to the graph of
f(x) = x at the point (16,4) (Fig. 3). Evaluate L(16.1) and
L(15.92) to approximate 16.1 and 15.92 . Are your
approximations using L larger or smaller than the exact values of
the square roots?
Practice 2: Find the equation of the line L(x) tangent to the graph of
3
f(x) = x at the point (1,1) and use L(x) to approximate
3 3
(1.02) and (0.97) . Do you think your approximations
using L are larger or smaller than the exact values?
The process we have used to approximate square roots and cubics can be used to approximate any
differentiable function, and the main result about the linear approximation follows from the two statements in the
boxes. Putting these two statements together, we have the process for Linear Approximation.
then (geometrically) the graph of the tangent line L(x) is close to the
graph of f(x), and
(algebraically) the values of the tangent line function
L(x) = f(a) + f '(a).(x – a) approximate the values of f(x):
f(x) ≈ L(x) = f(a) + f '(a).(x – a) .
Sometimes we replace "x – a" with "∆x" in the last equation, and the
statement becomes f(x) ≈ f(a) + f '(a).∆x.
x x
Solution: f(x) = e so f '(x) = e . We need to pick a value a
near x = 0.1 for which we know the exact value of f(a) and f '(a), and a = 0 is the obvious choice.
Then
0.1
e = f(0.1) ≈ L(0.1) = f(0) + f '(0).(0.1 – 0)
0 0
= e + e .(0.1) = 1 + 1.(0.1) = 1.1 .
0.1
This approximation is within 0.0052 of the exact value of e .
4
Practice 3: Approximate the value of (1.06) , the amount $1 becomes after 4 years in a bank which
4
pays 6% interest compounded annually. (Take f(x) = x and a = 1.)
Practice 4: Use the linear approximation process and the values in the table to estimate the value
of f when x = 1.1, 1.23 and 1.38 .
x f(x) f '(x)
1 0.7854 0.5
1.208761 0.4098
1.4 0.9505 0.3378
Example 3: Find a linear approximation formula L(x) for 1 + x when x is small. Use your result to
approximate 1.1 and .96 .
1/2 1 –1/2 1
Solution: f(x) = 1 + x = (1 + x) so f '(x) =
(1 + x) = . Since "x is small" , we know that x is
2 2 1+x
1
close to 0 , and we can pick a = 0. Then f(a) = f(0) = 1 and f '(a) = f '(0) = so
2
1 x
1 + x ≈ L(x) = f(0) + f '(0).(x – 0) = 1 + x = 1 + .
2 2
x 0.1
If x is small, then 1+x ≈1 +2 . 1.1 = 1 + 0.1 ≈ 1 + 2 = 1.05 and
–.04
0.96 = 1 + (–.04) ≈ 1 + 2 = 0.98 . Use your calculator to determine by how much each estimate
Most scientific experiments involve using instruments to take measurements, but the instruments are not
perfect, and the measurements we get from them are only accurate up to a certain level. If we know the level of
accuracy of our instruments and measurements, we can use the idea of linear approximation to estimate the level of
accuracy of results we calculate from our measurements.
2
If we measure the side x of a square to be 8 inches, then, of course, we would calculate its area to be A(x) = x = 64
square inches. Suppose, as is reasonable in a real measurement, that our measuring instrument could only measure or
be read to the nearest 0.05 inches. Then our measurement of 8 inches would really mean some number between 8–0.05
= 7.95 inches and 8+0.05 = 8.05 inches, and the true area of the square would be between A(7.95) = 63.2025 and
A(8.05) = 64.8025 square inches. Our possible "error" or "uncertainty", because of the limitations of the instrument,
could be as much as 64.8025 – 64 = 0.8025 square inches so we could report the area of the square to be 64 ± 0.8025
square inches. We can also use the linear approximation method to estimate the "error" or uncertainty of the area. (For
a function as simple as the area of a square, this linear approximation method really isn't needed, but it is used to
illustrate the idea.)
2
For a square with side x, the area is A(x) = x and A '(x) = 2x . If ∆x represents the "error" or uncertainty of our
measurement of the side, then, using the linear approximation technique for A(x) ,
A(x) ≈ A(a) + A '(a).∆x so the uncertainty of our calculated area is A(x ) – A(a) ≈ A '(x).∆x. In this example, a = 8
inches and ∆x = 0.05 inches so
A(8 + 0.05) – A(8) ≈ A '(8).∆x = 2(8 inches)(0.05 inches) = 0.8 square inches.
Practice 5: If we measure the side of a cube to be 4 cm with an uncertainty of 0.1 cm, what is the
volume of the cube and the uncertainty of our calculation of the volume? (Use linear approximation.)
Solution: Our measured angle is x =1.1170 radians and ∆x = 0.0349 radians (all of our
trigonometric work is in radians), and the height of the rocket at an angle x is
f(x) = 3000.tan( x ) so f(1.1170) ≈ 6151 m. Our uncertainty in the height is
2 2
∆f(x) ≈ f '(x).∆x ≈ 3000.sec (x).∆x = 3000 sec (1.1170).(0.0349) = 545 m.
o o
If our measured angle of 64 can be in error by as much as 2 , then our calculated height of 6151 m can be in
error by as much as 545 m. The height is 6151 ± 545 meters.
o o
Practice 6: Suppose we measured the angle of inclination of in the previous Example to be 43 ± 1 .
Estimate the height of the rocket in the form "height ± error ."
In some scientific and engineering applications, the calculated result must be within some given specification. You
might need to determine how accurate the initial measurements must be in order to guarantee the final calculation is
within the specification. Added precision usually costs time and money, so it is important to choose a measuring
instrument which is good enough for the job but not too good or too expensive.
3
Example 5: Your company produces ball bearings (spheres) with a volume of 10 cm , and the volume must be
3
accurate to within 0.1 cm . What radius should the bearings have and what error can you tolerate in the radius
4 3
measurement to meet the accuracy specification for the volume? (V = π r )
3
4 3
Solution: Since we want V = 10, we can solve 10 = π r for r to get r = 1.3365 cm.
3
4 3 2
V(r) =
3
π r and V '(r) = 4π r so ∆V ≈ V '(r).∆r . In this case we have been given that
3 3
∆V = 0.1 cm , and we have calculated r = 1.3365 cm so 0.1 cm = V '(1.3365
2
cm).∆r = (22.45 cm ).∆r.
Solving for ∆r, we get ∆r ≈ 0.0045 cm. To meet the specifications for the
allowable error in the volume, we must allow no more than 0.0045 cm
variation in the radius. If we measure the diameter of the sphere rather than the
radius, then we want d = 2r = 2(1.3365 ± 0.0045) = 2.673 ± 0.009 cm .
Practice 7: You want to determine the height of the rocket to within 10 meters
when it is 4000 meters high (Fig. 6). How accurate must your angle
measurement be? (Do your calculations in radians.)
The "error" we have been examining is called the absolute error to distinguish it from two other commonly
used terms, the relative error and the percentage error which compare the absolute error with the magnitude of the
number being measured. An "error" of 6 inches in measuring the circumference of the earth would be extremely small,
but a 6 inch error in measuring your head for a hat would result in a very bad fit.
error of f ∆f
Definitions: The Relative Error of f is =
value of f f
∆f .
The Percentage Error of f is (100) .
f
Example 6: If the relative error in the calculation of the area of a circle must be less than 0.4 , then what
relative error can we tolerate in the measurement of the radius?
2
Solution: A(r) = π r so A '(r) = 2π r and ∆A ≈ A '(r) ∆r = 2π r ∆r. The Relative Error of A is
∆A 2π r ∆r ∆r ∆A
A
≈ 2
= 2
r
. We can guarantee that the Relative Error of A ,
A
, is less than 0.4
πr
∆r 1 ∆A 1
if the Relative Error of r , = , is less than ( 0.4) = 0.2 .
r 2 A 2
Practice 8: If you can measure the side of a cube with a percentage error less than 3%, then what will
the percentage error for your calculation of the surface area of the cube be?
The Differential of f
The differential of f represents the change in f , as x changes from x to x + dx, along the tangent line to the
graph of f at the point ( x, f(x) ) . If we take dx to be the number ∆x, then the differential is an approximation of
∆f: ∆f ≈ f '(x).∆x = f '(x).dx = df .
3 2
Example 7: Determine the differential df of each of f(x) = x – 7x , g(x) = sin(x), and h(r) = πr .
2
Solution: df = f '(x).dx = (3x – 7) dx, dg = g '(x).dx = cos(x) dx, and dh = h '(r) dr = 2πr dr .
We will do little with differentials for a while, but are used extensively in integral calculus.
An approximation is most valuable if we also have have some measure of the size of the "error", the distance
between the approximate value and the value being approximated,. Typically, we will not know the exact value of the
error (why not?), but it is useful to know that the error must be less than some number. For example, if one scale gives
the weight of a gold pendant as 10.64 grams with an error less than .3 grams (10.64 ± .3 grams) and another scale
gives the weight of the same pendant as 10.53 grams with an error less than .02 grams (10.53 ± .02 grams), then we
can have more faith in the second approximate weight because of the smaller "error" guarantee. Before finding a
guarantee on the size of the error of the linear approximation process, we will check how well the linear approximation
process approximates some functions we can compute exactly. Then we will prove one bound on the possible error
and state a somewhat stronger bound.
2
Example 8: Let f(x) = x . Evaluate f(2+∆x), L(2+∆x) and | f(2+∆x) – L(2+∆x) | for
∆x = 0.1, 0.05, 0.01, 0.001 and for a general value of ∆x .
2 2 2 2
Solution: f(2+∆x) = (2+∆x) = 2 + 4∆x + (∆x) and L(2+∆x) = f(2) + f '(2).∆x = 2 + 4.∆x . Then
Cutting the value of ∆x in half makes the error 1/4 as large. Cutting ∆x to 1/10 as large makes the error 1/100
2 2 2 2
as large. In general, | f(2+∆x) – L(2+∆x) | = | (2 + 4.∆x + (∆x) ) – (2 + 4.∆x) | = (∆x) .
This function and error also have a nice geometric interpretation (Fig. 8):
2
f(x) = x is the area of a square of side x so f(2+∆x) is the area of a square of
2
side 2+∆x, and that area is the sum of the pieces with areas 2 , 2.∆x, 2.∆x, and
2 2
(∆x) . The linear approximation L(2+∆x) = 2 + 4.∆x to the area of the square
2
includes the 3 largest pieces 2 , 2.∆x and 2.∆x , but it omits the small square with
2 2
area (∆x) so the approximation is in error by the amount (∆x) .
3
Practice 10: Let f(x) = x . Evaluate
f(4+∆x), L(4+∆x) and
| f(4+∆x) – L(4+∆x) | for ∆x = 0.1, 0.05, 0.01, 0.001 and for a general
value of ∆x. Use Fig. 9 to give a geometric interpretation of f(4+∆x),
L(4+∆x) and | f(4+∆x) – L(4+∆x) | .
In both the example and practice problem, the error | f(a+∆x) – L(a+∆x) |
2
turned out to be very small, proportional to (∆x) , when ∆x was small.
In general, the error approaches 0 as ∆x approaches 0.
Proof: | f(a+∆x) – L(a+∆x) | = | f(a+∆x) – f(a) – f '(a).∆x | = { f(a+∆x)∆x– f(a) – f '(a) }. ∆x. But f is differentiable
f (a x) f (a) f (a x) f (a)
at x=a so lim = f '(a) and lim f '(a) 0 .
x 0 x x 0
x
f (a x) f (a)
Then lim | f (a x) L(a x) | = lim f '(a) lim x 0 0 0
x 0 x 0
x x0
Not only does the difference f(a+∆x) – L(a+∆x) approach 0, but this difference approaches 0 so fast that we
can divide it by ∆x, another quantity approaching 0, and the quotient still approaches 0.
In the next chapter we can prove that the error of the linear approximation process is
2
proportional to (∆x) . For now we just state the result.
Problems
1. Fig. 10 shows the tangent line to a function g at the point (2,2) and
a line segment ∆x units long. On the figure, label the locations of
(a) 2 + ∆x on the x–axis, (b) the point ( 2 + ∆x, g(2 + ∆x) ), and
(c) the point (2 + ∆x, g(2) + g '(2) .∆x ).
(d) How large is the "error", ( g(2) + g '(2) .∆x ) – ( g(2+∆x) )?
2. In Fig. 11 is the linear approximation L(a + ∆x) larger or smaller than the
value of f(a + ∆x) when
(a) a = 1 and ∆x = 0.2? (b) a = 2 and ∆x = –0.1? (c) a = 3 and ∆x = 0.1?
(d) a = 4 and ∆x = 0.2? and (e) a = 4 and ∆x = –0.2?
In problems 3 and 4, find the equation of the tangent line L to the given function f at the given point
( a, f(a) ). Use the value L(a + ∆x) to approximate the value of f(a + ∆x).
3. (a) f(x) = x , a = 4, ∆x = 0.2 (b) f(x) = x , a = 81, ∆x = –1 (c) f(x) = sin(x) , a = 0, ∆x = 0.3
x 5
4. (a) f(x) = ln(x), a = 1, ∆x = 0.3 (b) f(x) = e , a = 0, ∆x = 0.1 (c) f(x) = x , a = 1, ∆x = 0.03
n
5. Show that (1 + x) ≈ 1 + nx if x is "close to" 0. (Suggestion: Put f(x) = (1+x)n , a = 0, and ∆x = x.)
In problems 6 and 7, use the Linear Approximation Process to derive each approximation formula
for x "close to" 0.
n x
6. (a) (1 – x) ≈ 1 – nx (b) sin(x) ≈ x (c) e ≈1+x
9. A rectangle has one side on the x–axis, one side on the y–axis, and a corner
2
on the graph of y = x + 1 (Fig. 13).
(a) Use Linear Approximation of the area formula to estimate the increase in
the area of the rectangle if the base grows from 2 to 2.3 inches.
(b) Calculate exactly the increase in the area of the rectangle as the
base grows from 2 to 2.3 inches.
10. You can measure the diameter of a circle to within 0.3 cm.
(a) How large is the "error" in the calculated area of a circle with a measured
diameter of 7.4 cm?
(b) How large is the "error" in the calculated area of a circle with a measured diameter of 13.6 cm?
(c) How large is the percentage error in the calculated area of a circle with a measured diameter of d?
3
11. You are minting gold coins which must have a volume of 47.3 ± 0.1 cm . If you can manufacture
the coins to be exactly 2 cm high, how much variation can you allow for the radius?
5700 .
12. If F is the fraction of carbon–14 remaining in a plant sample Y years after it died, then Y = ln(F).
ln(0.5)
(a) Estimate the age of a plant sample in which 83±2 % (0.83 ± 0.02) of the carbon–14 remains.
(b) Estimate the age of a plant sample in which 13±2 % (0.13 ± 0.02) of the carbon–14 remains.
3
13. Your company is making dice (cubes) and the specifications require that their volume be 87 ± 2 cm .
How long should each side be and how much variance can a side have in order to meet the specifications?
2
14. If the specifications require a cube with a surface area of 43 ± 0.2 cm , how long should each side be
and how much variance can a side have in order to meet the specifications?
15. The period P, in seconds, for a pendulum to make one complete swing and return to the release point is P = 2π
2
L/g where L is the length of the pendulum in feet and g is 32 feet/sec .
(a) If L = 2 feet, what is the period of the pendulum?
(b) If P = 1 second, how long is the pendulum?
(c) Estimate the change in P if L increases from 2 feet to 2.1 feet.
(d) The length of a 24 foot pendulum is increasing 2 inches per hour. Is the period getting longer or shorter?
How fast is the period changing?
2
v .
16. A ball thrown at an angle with an initial velocity v will land sin(2) feet from the thrower.
g
(a) How far away will the ball land if = π/4 and v = 80 feet/second?
(b) Which will result in a greater change in the distance: a 5% error in the angle or a 5% error in the initial
velocity v?
17. For the function in Fig. 14, estimate the value of df when
(a) x = 2 and dx = 1 (b) x = 4 and dx = –1
(c) x = 3 and dx = 2
18. For the function in Fig. 15, estimate the value of df when
1/2 1
Practice 1: f(x) = x so f '(x) = . At the point (16,4) on the graph of f, the slope of the tangent line
2 x
1 1
is f '(16) = = . The equation of the tangent line is
2 16 8
1 1 1
y – 4 = (x – 16) or y = x + 2 : L(x) = x + 2 . Then
8 8 8
1 1
16.1 ≈ L(16.1) = (16.1) + 2 = 4.0125 and 15.92 ≈ L(15.92) = (15.92) + 2 = 3.99
8 8
3 2
Practice 2: f(x) = x so f '(x) = 3x . At (1,1), the slope of the tangent line is f '(1) = 3. The equation of the
tangent line is y – 1 = 3(x – 1) or y = 3x – 2: L(x) = 3x – 2. Then
3
(1.02) ≈ L(1.02) = 3(1.02) – 2 = 1.06 and (0.97)3 ≈ L(0.97) = 3(0.97) – 2 = 0.91 .
4 3
Practice 3: f(x) = x so f '(x) = 4x . a = 1 and ∆x = 0.06.
4 4 3
(1.06) = f(1.06) ≈ L(1.06) = f(1) + f '(1).(0.06) = 1 + 4(1) .(0.06) = 1.24 .
3 2 3 3
Practice 5: x = 4 cm and ∆x = 0.1 cm. f(x) = x so f '(x) = 3x and f(4) = 4 = 64 cm . Then
2 2 3
"error" ∆f ≈ f '(x)∆x = 3x .∆x. When x = 4 and ∆x = 0.1, "error" ∆f ≈ 3(4) (0.1) = 4.8 cm .
o
Practice 6: 43 ± 1 is 0.75049 ± 0.01745 radians. f(x) = 3000 .tan(x) so
2
f(0.75049) = 3000.tan(0.75049) ≈ 2797.5 m. f '(x) = 3000 sec (x) so
2 2
∆f(x) ≈ f '(x).∆x = 3000.sec (x).∆x = 3000.sec (0.75049).(0.01745) = 97.9 m .
The height of the rocket is 2797.5 ± 97.9 m .
2
Practice 7: f() = 2000.tan() so f '() = 2000.sec (). We know 4000 = 2000.tan() so
2 2
tan() = 2 and ≈ 1.10715 (radians). f '() = 2000.sec () so f '(1.10715) = 2000.sec (1.10715) ≈ 10,000.
Finally, "error" ∆f ≈ f '().∆ so 10 ≈ 10,000.∆ and ∆ ≈ 10/10,000 = 0.001 (radians) ≈ 0.057o.
2
Practice 8: A(r) = 6r so A '(r) = 12r and ∆A ≈ A '(r).∆r = 12r.∆r . We are also told that ∆r/r < 0.03 .
∆A . 12r.∆r . 2.∆r .
Percentage error is 100 = 2 100 = 100 < 200.(0.03) = 6 .
A 6r r
1
Practice 9: f(x) = ln(x) df = f '(x).dx = dx
x
du . –3 .dx
u = 1 – 3x du = dx =
dx 2 1 – 3x
dr .
r = 3cos() dr = d = –3 sin() .d
d
3 2 3 2
Practice 10: f(x) = x , f '(x) = 3x , and L(4 + ∆x) = f(4) + f '(4)∆x = 4 + 3(4) ∆x = 64 + 48.∆x .
f(4+∆x) is the actual volume of the cube with side length 4+∆x.
L(4+∆x) is the volume of the cube with side length 4 (v = 64) plus the volume of
2
the 3 "slabs" (v = 3.4 .∆x)
2
| f(4+∆x)–L(4+∆x) | is the volume of the "leftover" pieces from L: the 3 "rods" (v = 3 .4.(∆x) )
3
and the tiny cube (v = (∆x) ).