Numeric Derivatives and Linearization
I. Numeric Derivatives
II. Marginal Cost
III. Linearization
IV. Differentials
V. Newton's Method
VI. Taylor Polynomials
A Numeric Derivative Example
y=f(x)
We wish to mtan
approximate the
slope of this (0.25, 5.7)
curve at the data
point (0.19, 4.2).
(0.19, 4.2)
(0.11, 2)
A Numeric Derivative Example
y=f(x)
y=f(x) mR y=f(x) msec
mL
(0.25, 5.7) (0.25, 5.7) (0.25, 5.7)
(0.19, 4.2) (0.19, 4.2) (0.19, 4.2)
(0.11, 2) (0.11, 2) (0.11, 2)
We can immediately calculate a left hand slope, a right
hand slope, or a secant slope. However, our data is not
evenly spaced, so we should average* the left and right
hand slopes.
y=f(x)
mL = 27.5
(0.25, 5.7)
mR = 25
ave of mL & mR = 26.25
(0.19, 4.2)
(0.11, 2) msec = 26.43
*Actually, we should do a weighted average of the
left and right hand slopes, but such a detail is beyond
the College Board's scope.
(0.11, 2) Δx = 0.08 (0.19, 4.2) Δx = 0.06 (0.25, 5.7)
Δx = 0.14
0.08 0.06
weighted ave = (1 - 0.14
)(27.5) + (1 - 0.14
)(25)
= 26.0714
Another Numeric Derivative Example
x f(x) j Plot the points given at
0 1.0 left on a coordinate grid.
0.1 1.2
0.2 1.4 k Approximate the values
0.3 1.9 of f '(x) at each interior
0.4 2.9 point of the data.
0.5 3.7
0.6 4.2 l When would you say the
0.7 4.6 function is increasing at
0.8 4.7 the greatest rate? At the
0.9 4.8 slowest rate?
m Now consider f '(x) as a function itself. On another grid,
graph the values of (x, f '(x)) that you obtained in part 2.
n Estimate at what point f '(x) seems to be increasing
most rapidly by doing similar calculations to get f ''(x)
as on the prior slide.
6 Is there an analagous relationship between a function's
maximum (minimum) value and what happens to its
derivative f '(x) near where the function reaches that value,
with the relationship between where a function increases
(or decreases) most rapidly and what happens to f ''(x) (the
second derivative) at that point?
f '(0.1) ≈ 2
Graph of f '(0.2) ≈ 3.5 f ''(0.2) ≈ 27.5
y = f(x)
f '(0.3) ≈ 7.5 f ''(0.3) ≈ 27.5
f '(0.4) ≈ 9 f ''(0.4) ≈ -5
f '(0.5) ≈ 6.5 f ''(0.5) ≈ -22.5
f '(0.6) ≈ 4.5 f ''(0.6) ≈ -20
f '(0.8) ≈ 2.5 f ''(0.8) ≈ -17.5
f '(0.9) ≈ 1
What skill do you need to walk away from
this with?
You need to be able to approximate a value
of a first derivative or of a second derivative
at an interior point of a set of data.
Marginal Cost
The approximate increase in cost accrued from
producing one additional item.
Example:
Consider a total cost function, TC = Q2 + 9Q
TC(10) = 190. We interpret this as the total
cost of producing 10 units is 190.
We would like to learn more about how costs evolve over
the production cycle, so let's calculate average cost, which
is total cost divided by the number of units produced:
Q2 + 9Q
Cave = =Q+9
Q
Therefore, when we produce 10 units of this good, the average
cost per unit is 19. This is somewhat deceptive, however,
because we still don't know how costs evolve or change as we
produce. For example, the first unit (Q = 1) costs 10 to
produce. Obviously, if the average ends up being 19, and the
first unit cost 10, then the cost of producing a unit must be
changing as we produce different units.
Alternatively, the change in total cost is not the same every
time we change Q. Let's define this change in total cost for a
given change in Q as the marginal cost.
d (TC)
MC = = 2Q + 9
dQ
So how is this the approximate increase in cost
accrued from producing one additional item?
y = C(x)
C(x0) is the known
cost of producing x0
many items C(x0+1) is the
known cost
of producing
(x0 , C(x0)) x0 + 1 many
items
x0 x0 + 1
ΔC = C(x0 + 1) - C(x0)
Δx = 1
dC ΔC C(x0 + 1) - C(x0)
≈ =
dx Δx 1
Simple Example:
A company estimates its cost of producing n many
items is given by C(n) = 1500 + 0.2n + 0.03n2.
1. What is the cost of producing 2000 items?
2. What is the cost of producing 2001 items?
Observation: to make 1 more item, there was an
increase in cost of approximately 120.
Example Continued:
A company estimates its cost of producing n many
items is given by C(n) = 1500 + 0.2n + 0.03n2.
3. Use marginal cost to estimate the increase in
cost from producing 2000 to 2001 items.
C'(n) = 0.2 + 0.06n
C'(2000) = 120.2
Approximating with Tangent Lines
Suppose you really really want to approximate 29 .
We can use the graph of y = x and tangent lines .
y= x
(25, 5) ( 29, 29 )
25 29 36
Linearization
y = L(x)
(x, L(x))
y = f(x)
(x, f(x))
We claim:
(a, f(a)) L(x)≈f (x) when x ≈ a.
Our Task: How can we use
a x this to approximate
f(x) using L(x) ?
y = L(x)
L(x) Δx
m Δx {
f(a)
Δy
y = f(x)
Δy
m=
Δx
so Δy = m Δx
a x
Linearization: L(x) = f(a) + f '(a) (x - a)
Here, you do one:
3
Without a calculator, approximate 120
≈ 74/15 = 4.9333...
True Value = 4.93242
% Error ≈ 0.018433%
Calculator Active
Consider the linearization, L(x) of the curve y = x
at (36, 6). What are the least and greatest values of x
for which this linearization is within 0.1 of the true
value of the curve?
y = L(x)
y = x + 0.1
y = x - 0.1
x
36
[xmin, xmax] = [24.0547, 50.3453]
The population N, of a small country is given in
millions, for three different years in the table below.
Use a linearization to approximate the population of
the country in 1984.
t 1980 1985 1990
N(t) 15.0 17.0 19.3
t 1980 1985 1990
N(t) 15.0 17.0 19.3
N(t) ≈ N(1985) + N'(1985)(t - 1985)
N(1984) ≈ 17 + 4.3 (1984 - 1985)
10
≈ 16.57 million
Differentials
change in y is proportional to change in x
dy dx
For small changes in x, change in y is
equal to f '(x) dx
Example A:
Let f(x) = x2 + 2 x + 2 . Using differentials,
estimate the change in y when x increases from
7 to 7.1.
dy = 1.43333 note: y = 1.44324
Example B:
The radius of a circle is to be increased from an initial
value of r0 = 10 cm by an amount dr = 0.05 cm.
Estimate the corresponding increase in the circle's
area A, by calculating dA.
dA = π cm2
Absolute and Relative Error
Absolute Error: | dy |
dy
Relative Error: y
Example C:
If y = 3x, show that the relative error in y is
exactly equal to the relative error in x.
Follow Up: How are the relative errors in x and y
related for y = x2?
Example D:
Mr. Beilin sized: so assume 0 feet tall.
A person stands 500 feet away from the base of a tall
tree. If the person must look up through an angle of
30 to see the top of the tree, how accurate must the
measurement of that angle be, in order for the
person's estimate of the tree's height to be within a
relative error of 3% ?
dθ < 0.0129 rads 0.744
Example E:
The cross section of a gear for a high precision milling
machine is in the shape of an ellipse whose major radius
is 25 mm and whose minor radius 16 mm. If the lengths
of the major and minor axes are measured within 0.02
and 0.01 mm of accuracy respectively, estimate the error
in the cross sectional area of the gear. The area of an
ellipse is given by, A = abπ where a and b are the
lengths of the two radii.
dA = 1.791 mm2
rel error = 0.1425%
Example P:
A cylindrical biscuit stick of radius 4 mm is dipped 162
mm deep into strawberry cream (with little strawberry
bits) to coat it in a layer 0.2 mm thick. A hemispherical
droplet of cream forms at the end of the stick when it is
pulled out. Use differentials to estimate the total
volume of cream used to coat this yummy cookie stick.
And what would you name this delicious snack if
you were going to mass market and sell it?
dV = 814.300 + 155.169 = 969.469 mm3
Homework
Linearization and Differentials
Pg. 256 257: 2, 9, 17, 33, 34, 38, 40, 43, 44
Newton's Method
A method of successive linearizations
y = f(x)
(x0, f(x0))
f(r) = 0
r x0
The General Formula for Newton's Method
f(xn)
xn+1 = xn -
f '(xn)
xn+1 xn
Example: Solve: x = cos x
[Calculator Active]
f(x) = x - cos x
x0 = 1
x1 = 0.750364
x2 = 0.739113
x3 = 0.739085
Yours: Use Newton's Method to approximate
45 within 0.001
2
f(x) = x - 45 x0 = 7
x1 = 6.7142857
x2 = 6.7082067
x3 = 6.7082039
(Actually at this point our error is 7.45 x 1012 %)
Sample Multiple Choice Newton's Method
Newton's Method is used to approximate a root of the graph of
y = f(x) as shown below. If an initial guess x0 is used in Newton's
Method, which lettered choice indicates the approximate location
of x1, the subsequent guess?
y = f(x)
A B
C D x0 E
Cauchy Convergence
When is "good enough"... good enough?
For all ε > 0, there exists a natural number N such that
n, m > N implies | sn - sm | < ε
In every day terms, this says that when a
sequence converges, the terms get closer
and closer not only to some limiting value,
but they get closer and closer to each other.
Calculus BC
Starter #26: 2/14/22
Use Newton's Method to
approximate the nonzero
solution to x2 = tan1x to
four decimal places of x
accuracy.
y = x2 y = tan1x
Starting with an initial guess of x0 = 1.0000
x1 = 0.856932109
x2 = 0.834221768
x3 = 0.833606476
x4 = 0.833606194
x5 = 0.833606194
When Potato Salad Goes Bad
When Good Formulas Go Bad
x2
x1
|x+1| x+1
y=
x+1
Newton's Method can oscillate between two guesses
Newton's Method
can diverge from
guesses that land
near horizontal
x1 tangent lines
x0
whoops...
Newton's Method can converge,
but not to the root you were
expecting.
x0 r x2 x1
Newton's Method on the
4
polynomial, z - 1 = 0.
Colors represent to which
root an initial guess a + bi
converges, while color
intensity represents the
number of iterations
required to converge
within a given tolerance.
Click to Play: 3Blue1Brown video that used to be titled "Why
don't high school Calculus teachers talk about this stuff?"
(You mean the stuff I've been talking about since 2004?
Yeah, that stuff)
Homework
Newton's Method
Pg. 348 350: #4, 11, 30, 31, 33, 34, 40