10 Numerical Methods PDF

10 Numerical
Methods
10.1 Gaussian Elimination with Partial Pivoting
10.2 Iterative Methods for Solving Linear Systems
10.3 Power Method for Approximating Eigenvalues
10.4 Applications of Numerical Methods
Car Crash Simulations (p. 521)
Laboratory Experiment
Probabilities (p. 520)
Google’s PageRank Algorithm (p. 515)
Illustration Software (p. 508)
Pharmacokinetics (p. 501)

Clockwise from top left, Evgeny Murtola/Shutterstock.com; lculig/Shutterstock.com;
Leah-Anne Thompson,2009/Shutterstock.com; Perov Stanislav/Shutterstock.com; Annette Shaff/Shutterstock.com
495
9781305658004_1001.indd 495 12/15/15 11:27 AM

496 Chapter 10 Numerical Methods
10.1 Gaussian Elimination with Partial Pivoting

Express a real number in floating point form, and determine the
stored value of a real number rounded to a specified number of
significant digits.
Compare exact solutions with rounded solutions obtained by
Gaussian elimination with partial pivoting, and observe rounding
error with ill-conditioned systems.
Stored Value and Rounding Error

In Chapter 1, you studied two methods of solving a system of n linear equations in n
variables—Gaussian elimination and Gauss-Jordan elimination. When either of these
methods is used with a digital computer, it introduces a problem that has not yet been
discussed in this text—rounding error.
Digital computers store real numbers in floating point form,
± M × 10 k
where k is an integer and the mantissa M satisfies the inequality 0.1 ≤ M < 1.
For example, the floating point forms of some real numbers are listed below.
Real Number Floating Point Form
527 0.527 × 103
−3.81623 −0.381623 × 101
0.00045 0.45 × 10−3
827,000 0.827 × 106
1 0.1 × 101
63.61 0.6361 × 10 2
−2200 −0.22 × 10 4
The number of decimal places that can be stored in the mantissa depends on the
computer. If n places are stored, then it is said that the computer stores n significant
digits. Numbers with additional digits are either truncated or rounded. When a number
is truncated to n significant digits, all digits after the first n significant digits are simply
omitted. For example, truncated to two significant digits, the number 0.1251 becomes
0.12.
If a number is rounded to n significant digits, then the last retained digit is
increased by 1 when the discarded portion is greater than half a digit, and the last
retained digit is not changed when the discarded portion is less than half a digit. For
the special case in which the discarded portion is precisely half a digit, round so that
the last retained digit is even. For example, the numbers below are rounded to two
significant digits.
Number Rounded Number
0.1249 0.12
0.125 0.12
0.1251 0.13
0.1349 0.13
0.135 0.14
0.1351 0.14
Most computers store numbers in binary form (base 2) rather than decimal form (base
10). Although rounding occurs in both systems, this discussion is restricted to the more
familiar base 10. When a computer truncates or rounds a number, it introduces a rounding
error that can affect subsequent calculations. The result after rounding or truncating is
called the stored value.
9781305658004_1001.indd 496 12/15/15 11:27 AM

10.1 Gaussian Elimination with Partial Pivoting 497
Finding Stored Values
Determine the stored value of each real number in a computer that rounds to three
significant digits.
a. 54.7 b. 0.1134 c. −8.2256 d. 0.08335 e. 0.08345
solution
Number Floating Point Form Stored Value
a. 54.7 0.547 × 102 0.547 × 102
b. 0.1134 0.1134 × 100 0.113 × 100
c. −8.2256 −0.82256 × 101 −0.823 × 101
d. 0.08335 0.8335 × 10−1 0.834 × 10−1
e. 0.08345 0.8345 × 10−1 0.834 × 10−1
Note in parts (d) and (e) that when the discarded portion of a decimal is precisely half
a digit, the number is rounded so that the stored value ends in an even digit.
Rounding error tends to propagate when using arithmetic operations. The next
example illustrates this phenomenon.
Propagation of Rounding Error
Evaluate the determinant of the matrix
A= [0.12
0.12
0.23
0.12]
rounding each intermediate calculation to two significant digits. Then find the exact
value, rounded to two significant digits, and compare the two results.
solution
Rounding each intermediate calculation to two significant digits produces
∣A∣ = (0.12)(0.12) − (0.12)(0.23)

= 0.0144 − 0.0276
≈ 0.014 − 0.028 Round to two significant digits.
= −0.014.
∣∣
The exact value is A = 0.0144 − 0.0276 = −0.0132. So, to two significant digits,
the determinant is −0.013. Note that rounding at the intermediate steps produced a
determinant that is not correct to two significant digits, even though each arithmetic
operation was performed with two significant digits of accuracy. This is what is meant
when it is said that arithmetic operations tend to propagate rounding error.
In Example 2, rounding at the intermediate steps introduced a rounding error of

−0.0132 − (−0.014) = 0.0008. Rounding error
Although this error may seem slight, it represents a percentage error of

0.0008
≈ 0.061 = 6.1%. Percentage error
0.0132
In most practical applications, a percentage error of this magnitude would be intolerable.
Keep in mind that this percentage error arose with only a few arithmetic steps. When
the number of arithmetic steps increases, the likelihood of a large percentage error
also increases.
9781305658004_1001.indd 497 12/15/15 11:27 AM

Gaussian Elimination with Partial Pivoting

For large systems of linear equations, Gaussian elimination can involve hundreds
of arithmetic computations, each of which can produce rounding error. Example 3
illustrates the potential magnitude of this problem.
Gaussian Elimination and Rounding Error
Use Gaussian elimination to solve the system below.

0.143x1 + 0.357x2 + 2.01x3 = −5.173
−1.31x1 + 0.911x2 + 1.99x3 = −5.458
11.2x1 − 4.30x2 − 0.605x3 = 4.415
After each intermediate calculation, round the result to three significant digits.
solution
The augmented matrix for this system is
[ ]
0.143 0.357 2.01 −5.17
−1.31 0.911 1.99 −5.46 .
11.2 −4.30 −0.605 4.42
Applying Gaussian elimination produces
[ ]
1.00 2.50 14.1 −36.2 Dividing the first row
−1.31 0.911 1.99 −5.46 by 0.143 produces a
11.2 −4.30 −0.605 4.42 new first row.
[ ]
1.00 2.50 14.1 −36.2 Adding 1.31 times the
0.00 4.19 20.5 −52.9 first row to the second row
11.2 −4.30 −0.605 4.42 produces a new second row.
[ ]
1.00 2.50 14.1 −36.2 Adding −11.2 times the
0.00 4.19 20.5 −52.9 first row to the third row
0.00 −32.3 −159 409 produces a new third row.
[ ]
1.00 2.50 14.1 −36.2 Dividing the second row
0.00 1.00 4.89 −12.6 by 4.19 produces a new
0.00 −32.3 −159 409 second row
[ ]
1.00 2.50 14.1 −36.2 Adding 32.3 times the
0.00 1.00 4.89 −12.6 second row to the third row
0.00 0.00 −1.00 2.00 produces a new third row.
[ ]
1.00 2.50 14.1 −36.2 Multiplying the third row
0.00 1.00 4.89 −12.6 . by −1 produces a new
0.00 0.00 1.00 −2.00 third row.
Using back-substitution, you obtain
x3 = −2.00
x2 = −2.82
x1 = −0.900.
Check this “solution” in the original system of equations to see that it is not correct.
Verify that the correct solution is
x1 = 1, x2 = 2, x3 = −3.
9781305658004_1001.indd 498 12/15/15 11:27 AM

In Example 3, rounding after each intermediate calculation resulted in percentage

errors of 190% for x1, 241% for x2, and 33.3% for x3 (check these), which are significant.
Part of the problem is that the original augmented matrix contains entries that
differ in orders of magnitude. For example, the first column of the matrix
[ ]
0.143 0.357 2.01 −5.17
−1.31 0.911 1.99 −5.46
11.2 −4.30 −0.605 4.42
has entries whose absolute values increase roughly by powers of 10 as you move down
the column. In subsequent elementary row operations, you multiplied the first row by
1.31 and −11.2, and you multiplied the second row by 32.3. When using floating point
arithmetic, such large row multipliers tend to propagate rounding error. For example,
notice what happened during Gaussian elimination when you multiplied the first row
by 1.31 and added to the second row:
[ ]
1.00 2.50 14.1 −36.2
−1.31 0.911 1.99 −5.46
11.2 −4.30 −0.605 4.42
[ ]
1.00 2.50 14.1 −36.2 Adding 1.31 times the
0.00 4.19 20.5 −52.9 first row to the second row
11.2 −4.30 −0.605 4.42 produces a new second row.
The second, third, and fourth entries in the second row each lost one decimal place of
accuracy. Also, notice what happened when you multiplied the first row by −11.2 and
added to the third row:
[ ]
1.00 2.50 14.1 −36.2
0.00 4.19 20.5 −52.9
11.2 −4.30 −0.605 4.42
[ ]
1.00 2.50 14.1 −36.2 Adding −11.2 times the
0.00 −32.3 −159 409 produces a new third row.
The second entry in the third row lost one decimal place of accuracy, the third entry in
the third row lost three decimal places of accuracy, and the fourth entry in the third row
lost two decimal places of accuracy.
This type of error propagation can be lessened by appropriate row interchanges
that produce smaller multipliers. One way to restrict the size of the multipliers is to use
Gaussian elimination with partial pivoting.

1. Find the entry in the first column with the largest absolute value. This entry
is called the pivot.
2. Perform a row interchange, if necessary, so that the pivot is in the first row.
3. Divide the first row by the pivot.
4. Use elementary row operations to reduce the remaining entries in the first
column to 0.
The completion of these four steps is a pass. After performing the first pass,
ignore the first row and first column and repeat the four steps on the remaining
submatrix. Continue this process until the matrix is in row-echelon form.
Example 4 on the next page shows what happens when this partial pivoting
technique is used on the system of linear equations from Example 3.
9781305658004_1001.indd 499 12/15/15 11:27 AM

See LarsonLinearAlgebra.com for an interactive version of this type of example.
Use Gaussian elimination with partial pivoting to solve the system from Example 3.
After each intermediate calculation, round the result to three significant digits.
solution
As in Example 3, the augmented matrix for this system is
[ ]
0.143 0.357 2.01 −5.17
−1.31 0.911 1.99 −5.46 .
11.2 −4.30 −0.605 4.42
Pivot
In the first column, 11.2 is the pivot because it has the largest absolute value. So,
interchange the first and third rows and apply elementary row operations.
[ ]
11.2 −4.30 −0.605 4.42 Interchange the
−1.31 0.911 1.99 −5.46 first and third
0.143 0.357 2.01 −5.17 rows.
[ ]
1.00 −0.384 −0.0540 0.395 Dividing the first row
−1.31 0.911 1.99 −5.46 by 11.2 produces a new
0.143 0.357 2.01 −5.17 first row.
[ ]
1.00 −0.384 −0.0540 0.395 Adding 1.31 times the first
0.00 0.408 1.92 −4.94 row to the second row
0.143 0.357 2.01 −5.17 produces a new second row.
[ ]
1.00 −0.384 −0.0540 0.395 Adding −0.143 times the
0.00 0.412 2.02 −5.23 produces a new third row.
Pivot
This completes the first pass. For the second pass, consider the submatrix formed by
deleting the first row and first column. In this matrix the pivot is 0.412, so interchange
the second and third rows and proceed with Gaussian elimination.
[ ]
1.00 −0.384 −0.0540 0.395 Interchange the
0.00 0.412 2.02 −5.23 second and third
0.00 0.408 1.92 −4.94 rows.
REMARK
[ ]
1.00 −0.384 −0.0540 0.395 Dividing the second row
Note that the row multipliers 0.00 1.00 4.90 −12.7 by 0.412 produces a new
used in Example 4 are 1.31, 0.00 0.408 1.92 −4.94 second row.
−0.143, and −0.408, whereas
[ ]
the multipliers used in Example 3 1.00 −0.384 −0.0540 0.395 Adding −0.408 times the
are 1.31, −11.2, and 32.3. 0.00 1.00 4.90 −12.7 second row to the third row
0.00 0.00 −0.0800 0.240 produces a new third row.
This completes the second pass, and the entire procedure can be completed by dividing
the third row by −0.0800.
[ ]
1.00 −0.384 −0.0540 0.395
0.00 1.00 4.90 −12.7
0.00 0.00 1.00 −3.00
So, x3 = −3.00, and back-substitution produces x2 = 2.00 and x1 = 1.00, which
agrees with the exact solution of x1 = 1, x2 = 2, and x3 = −3.
9781305658004_1001.indd 500 12/15/15 11:27 AM

The term partial in partial pivoting refers to the fact that in each pivot search, you
consider only entries in the first column of the coefficient matrix or submatrix. This
search can be extended to include every entry in the coefficient matrix or submatrix;
the resulting technique is called Gaussian elimination with complete pivoting.
Unfortunately, neither complete pivoting nor partial pivoting solves all problems of
rounding error. Some systems of linear equations, called ill-conditioned systems, are
extremely sensitive to numerical errors. For such systems, pivoting is not much help.
A common type of system of linear equations that tends to be ill-conditioned is one
for which the determinant of the coefficient matrix is nearly zero. The next example
illustrates this problem.
An ill-Conditioned System of Linear Equations
Use Gaussian elimination to solve the system of linear equations.

x+ y= 0
401
x+ y = 20
400
Round each intermediate calculation to four significant digits.
solution
Gaussian elimination with rational arithmetic produces the exact solution x = −8000
and y = 8000. But rounding 401400 = 1.0025 to four significant digits introduces a
large rounding error, as shown below.
[11 1
1.002
0
20 ]
[10 1
0.002
0
20 ]
[10 1 0
1.00 10,000 ]
So, y = 10,000 and back-substitution produces
x = −y
= −10,000.
This “solution” represents a percentage error of 25% for both the x-value and the
y-value. Note that this error was caused by a rounding error of only 0.0005, when
1.0025 was rounded to 1.002.
Pharmacokinetics is the study of the movement of a drug

Linear into, out of, and through the body. Pharmacokinetics
Algebra researchers study the absorption, distribution, bioavailability,
Applied metabolism, and excretion of drugs. They analyze the
credibility of data, account for environmental effects,
and determine subpopulations of subjects who have the
same characteristics. Their work often involves the use of
mathematical formulas that relate the onset, intensity, and
duration of a drug’s effect. The results of pharmacokinetics
research are used in the development of dosage regimens
for patients. So, it is vital that computations are not skewed
by factors such as rounding errors. To help minimize these
errors, researchers rely on software programs that use
numerical methods similar to those discussed in this chapter.
luchschen/Shutterstock.com
9781305658004_1001.indd 501 12/15/15 11:27 AM

10.1 Exercises
Floating Point Form In Exercises 1–8, express the real Solving an ill-Conditioned System In Exercises 25
number in floating point form. and 26, use Gaussian elimination to solve the
1. 1824 2. 321.61 3. −2.62 4. −21.001 ill-conditioned system of linear equations, rounding each
1 intermediate calculation to three significant digits. Then
5. −0.00121 6. 0.00026 7. 8 8. 61 12
compare this solution with the exact solution provided.
Finding Stored Values In Exercises 9–16, determine 25. x + y = 3
the stored value of the real number in a computer 31
x + 30 y = −50
that rounds to (a) three significant digits and (b) four
significant digits. (Exact: x = 1593, y = −1590)
9. 4413 10. 21.4 11. −92.646 12. 216.964 26. x − 800
801 y = 10

7
13. 16 7
14. 32 15. 17 16. 16 −x + y = 50
(Exact: x = 48,010, y = 48,060)
Propagation of Rounding Error In Exercises 17 and
18, evaluate the determinant of the matrix, rounding 27. Comparing ill-Conditioned Systems Solve each
each intermediate calculation to three significant digits. ill-conditioned system and compare the solutions.
Then find the exact value and compare the two results. (a) x + y = 2 (b) x + y= 2

17. [66.00
1.24 56.00
1.02
] 18. [2.12
1.07 ]
4.22
2.12
x + 1.0001y = 2 x + 1.0001y = 2.0001
Gaussian Elimination and Rounding Error In 28. C

APSTONE By inspection, determine whether
Exercises 19 and 20, use Gaussian elimination to solve the system below is more likely to be solvable by
the system. After each intermediate calculation, round Gaussian elimination with partial pivoting, or is
the result to three significant digits. Then find the exact more likely to be ill-conditioned. Explain.
solution and compare the two results. −0.216x1 + 0.044x2 − 8.15x3 = 7.762
19. 1.21x + 16.7y = 28.8 20. 14.4x − 17.1y = 31.5 2.05x1 − 1.02x2 + 0.937x3 = 4.183
23.1x1 + 6.05x2 − 0.021x3 = 52.271
4.66x + 64.4y = 111.0 81.6x − 97.4y = 179.0
Gaussian Elimination with Partial Pivoting In

29. The Hilbert matrix of size n × n is the n × n symmetric
Exercises 21–24, (a) use Gaussian elimination without
matrix Hn = [aij], where aij = 1(i + j − 1). As n
partial pivoting to solve the system, rounding to three
increases, the Hilbert matrix becomes more and more
significant digits after each intermediate calculation,
ill-conditioned. Use Gaussian elimination to solve the
(b) use Gaussian elimination with partial pivoting to solve
system of linear equations H3 x = b, where b = [1 1 1]T
the same system, again rounding to three significant digits
(shown below), rounding to two significant digits after
after each intermediate calculation, and (c) compare both
each intermediate calculation. Compare this solution
solutions with the exact solution provided.
with the exact solution (x1 = 3, x2 = −24, and
21. x + 1.04y = 2.04 x3 = 30).

6x + 6.20y = 12.20
x1 + 12 x2 + 13 x3 = 1
(Exact: x = 1, y = 1)
1
22. 0.51x + 92.6y = 97.7 2 x1 + 13 x2 + 14 x3 = 1
1
99.00x − 449.0y = 541.0 3 x1 + 14 x2 + 15 x3 = 1
(Exact: x = 10, y = 1) 30. Repeat Exercise 29 for H4x = b, where
23. x + 4.01y + 0.00445z = 0.00 b = [1 1 1 1]T, rounding to four significant
−x − 4.00y + 0.00600z = 0.21 digits. Compare this solution with the exact solution
2x − 4.05y + 0.05000z = −0.385 (x1 = −4, x2 = 60, x3 = −180, and x4 = 140).
(Exact: x = −0.49, y = 0.1, z = 20) 31. The inverse of the n × n Hilbert matrix Hn has integer
entries. Use a software program or a graphing utility
24. 0.007x + 61.20y + 0.093z = 61.3
to calculate the inverses of the Hilbert matrices Hn for
4.810x − 5.92y + 1.110z = 0.0
n = 4, 5, 6, and 7. For what values of n do the inverses
81.400x + 1.12y + 1.180z = 83.7 appear to be accurate?
(Exact: x = 1, y = 1, z = 1)
9781305658004_1001.indd 502 12/15/15 11:27 AM

10.2 Iterative Methods for Solving Linear Systems 503
10.2 Iterative Methods for Solving Linear Systems

Use the Jacobi method to approximate the solution of a system of
linear equations.
Use the Gauss-Seidel method to approximate the solution of a
system of linear equations.
Determine whether an iterative method converges or diverges,
and use a strictly diagonally dominant coefficient matrix to attain
convergence.
The Jacobi Method

As a numerical technique, Gaussian elimination with partial pivoting is rather unusual
because it is direct. That is, you obtain a solution after a single application of this
technique. Once you obtain a solution, this technique offers no method of refinement.
That can be a problem because, as the preceding section shows, Gaussian elimination
with partial pivoting is sensitive to rounding error.
Numerical techniques more commonly involve an iterative method. For instance, in
calculus, Newton’s method is an iterative way to approximate the zeros of a differentiable
function. This section introduces two iterative methods for approximating the solution of
a system of n linear equations in n variables.
The first iterative technique discussed is the Jacobi method, named after German
mathematician Carl Gustav Jacob Jacobi (1804–1851). This method makes two
assumptions: (1) that the system
a11x1 + a12 x2 + a13x3 + . . . + a1nxn = b1
a21x1 + a22 x2 + a23x3 + . . . + a2nxn = b2
⋮
an1x1 + an2 x2 + an3 x3 + . . . + ann xn = bn
has a unique solution and (2) that the coefficient matrix A has no zeros on its main
diagonal. If any of the diagonal entries a11, a22, . . . , ann are zero, then interchange
rows or columns to obtain a coefficient matrix that has all nonzero entries on the main
diagonal.
To begin the Jacobi method, solve the first equation for x1, solve the second
equation for x2, and so on, as shown below.
1
x1 = (b − a12x2 − a13x3 − . . . − a1nxn)
a11 1
1
x2 = (b − a21x1 − a23x3 − . . . − a2nxn)
a22 2
⋮
1
xn = (b − an1x1 − an2x2 − . . . − an,n−1xn−1)
ann n
Then make an initial approximation of the solution
(x1, x2, x3, . . . , xn) Initial approximation
and substitute these values of xi on the right-hand sides of the rewritten equations
to obtain the first approximation. After obtaining the first approximation, you have
performed one iteration. In the same way, form the second approximation by substituting
the first approximation’s x-values on the right-hand sides of the rewritten equations. By
repeated iterations, you form a sequence of approximations that often converges to the
actual solution. Example 1 illustrates the use of the Jacobi method.
9781305658004_1002.indd 503 12/15/15 11:29 AM

Applying the Jacobi Method
Use the Jacobi method to approximate the solution of the system of linear equations below.
5x1 − 2x2 + 3x3 = −1
−3x1 + 9x2 + x3 = 2
2x1 − x2 − 7x3 = 3
Continue the iterations until two successive approximations are identical when rounded
to three significant digits.
solution
To begin, rewrite the system in the form
x1 = − 15 + 25 x2 − 35 x3
2
x2 = 9 + 39 x1 − 19 x3
x3 = − 37 + 27 x1 − 17x2.
You do not yet know the solution, so choose
x1 = 0, x2 = 0, x3 = 0 Initial approximation
as a convenient initial approximation. By substituting these values on the right-hand sides

of the equations in the rewritten system, you determine the first approximation to be
x1 = − 15 + 25 (0) − 35 (0) = −0.200
2
x2 = 9 + 39 (0) − 19 (0) ≈ 0.222
− 37 2 1
x3 = + 7 (0) − 7 (0) ≈ −0.429.
Next, substitute these x-values to obtain the second approximation.
x1 = − 15 + 2
5 (0.222) − 35 (−0.429) ≈ 0.146
2 3 1
x2 = 9 + 9 (−0.200) − 9 (−0.429) ≈ 0.203
− 37 2 1
x3 = + 7 (−0.200) − 7 (0.222) ≈ −0.517
Continuing this procedure, you obtain the sequence of approximations shown in the
table.
n 0 1 2 3 4 5 6 7
x1 0.000 −0.200 0.146 0.191 0.181 0.185 0.186 0.186
x2 0.000 0.222 0.203 0.328 0.332 0.329 0.331 0.331
x3 0.000 −0.429 −0.517 −0.416 −0.421 −0.424 −0.423 −0.423
The last two columns in the table are identical, so you can conclude that to three
significant digits the solution is
x1 = 0.186, x2 = 0.331, x3 = −0.423.
For the system of linear equations in Example 1, the Jacobi method converges.
That is, repeated iterations result in approximations that are identical to a specified
number of significant digits. As is generally true for iterative methods, greater accuracy
would require more iterations.
9781305658004_1002.indd 504 12/15/15 11:29 AM

The Gauss-Seidel Method

You will now look at a modification of the Jacobi method called the Gauss-Seidel
method, named after German mathematicians Carl Friedrich Gauss (1777–1855) and
Philipp L. Seidel (1821–1896). This modification is no more difficult to use than the Jacobi
method, and it often requires fewer iterations to produce the same degree of accuracy.
With the Jacobi method, the values of xi obtained in the nth approximation remain
unchanged until after you determine the entire (n + 1)th approximation. On the other
hand, with the Gauss-Seidel method you use the new values of each xi as soon as you
know them. That is, once you have determined x1 from the first equation, you then use
its value in the second equation to obtain the new x2. Similarly, you use the new x1
and x2 in the third equation to obtain the new x3, and so on. Example 2 demonstrates
this procedure.
Applying the Gauss-Seidel Method
Use the Gauss-Seidel iteration method to approximate the solution of the system of
equations in Example 1.
solution
As in Example 1, use the system of equations rewritten in the form
x1 = − 15 + 25 x 2 − 35 x 3
2 3 1
x 2 = 9 + 9 x1 − 9 x 3
x 3 = − 37 + 27 x1 − 17 x 2.
The first computation is identical to that in Example 1. That is, using (x1, x2, x3) = (0, 0, 0)
as the initial approximation, you obtain the new value of x1.
x1 = − 15 + 25 (0) − 35 (0) = −0.200
Now that you have a new value of x1, use it to compute a new value of x2. That is,
x2 = 29 + 39 (−0.200) − 19 (0) ≈ 0.156.
Similarly, use x1 = −0.200 and x2 = 0.156 to compute a new value of x3. That is,
x3 = − 37 + 27 (−0.200) − 17 (0.156) = −0.508.
So, the first approximation is x1 = −0.200, x2 = 0.156, and x3 = −0.508. Now,
performing the next iteration produces
x1 = − 15 + 25 (0.156) − 35 (−0.508) ≈ 0.167
2 3 1
x2 = 9 + 9 (0.167) − 9 (−0.508) ≈ 0.334
3 2 1
x3 = −7 + 7 (0.167) − 7 (0.334) ≈ −0.429.
Continued iterations produce the sequence of approximations shown in the table.
n 0 1 2 3 4 5 6
x1 0.000 −0.200 0.167 0.191 0.187 0.186 0.186
x2 0.000 0.156 0.334 0.334 0.331 0.331 0.331
x3 0.000 −0.508 −0.429 −0.422 −0.422 −0.423 −0.423
Note that after only six iterations of the Gauss-Seidel method, you achieved the
same accuracy as was obtained with seven iterations of the Jacobi method in Example 1.

9781305658004_1002.indd 505 12/15/15 11:29 AM

Convergence and Divergence

The iterative methods presented in this section do not always converge. That is, it
is possible to apply the Jacobi method or the Gauss-Seidel method to a system of
linear equations and obtain a divergent sequence of approximations. In such cases, the
method diverges.
An Example of Divergence
Apply the Jacobi method to the system

x1 − 5x2 = −4
7x1 − x2 = 6
using the initial approximation (x1, x2) = (0, 0), and show that the method diverges.
solution
As usual, begin by rewriting the system in the form
x1 = −4 + 5x2
x2 = −6 + 7x1.
Then the initial approximation (0, 0) produces the first approximation
x1 = −4 + 5(0) = −4
x2 = −6 + 7(0) = −6.
The first approximation (−4, −6) produces the second approximation
x1 = −4 + 5(−6) = −34
x2 = −6 + 7(−4) = −34.
Repeated iterations produce the sequence of approximations shown in the table.
n 0 1 2 3 4 5 6 7
x1 0 −4 −34 −174 −1224 −6124 −42,874 −214,374
x2 0 −6 −34 −244 −1224 −8574 −42,874 −300,124
The solution of this system of linear equations is x1 = 1 and x2 = 1. (Check this.)

So you can see from the table that the approximations obtained using the Jacobi
method become progressively worse instead of better, and you can conclude that the
method diverges.
You cannot resolve the problem of divergence in Example 3 by using the

Gauss-Seidel method rather than the Jacobi method. In fact, for the system in Example 3,
the Gauss-Seidel method diverges even more rapidly, as shown in the table.
n 0 1 2 3 4 5
x1 0 −4 −174 −6124 −214,374 −7,503,124
x2 0 −34 −1224 −42,874 −1,500,624 −52,521,874
You will now look at a special type of coefficient matrix A, called a strictly
diagonally dominant matrix, for which it is guaranteed that both the Jacobi method
and the Gauss-Seidel method will converge.
9781305658004_1002.indd 506 12/15/15 11:29 AM

Definition of a Strictly Diagonally Dominant Matrix

An n × n matrix A is strictly diagonally dominant when the absolute value of
each entry on the main diagonal is greater than the sum of the absolute values of
the other entries in the same row. That is,
∣ 11∣ ∣ 12∣ ∣ 13∣

a > a + a +. . .+ a ∣ 1n∣
∣a22∣ > ∣a21∣ + ∣a23∣ + . ∣ ∣
. . + a2n
⋮
∣ ∣ ∣ ∣ ∣ ∣ ∣
ann > an1 + an2 + . . . + an, n−1 . ∣
Strictly Diagonally Dominant Matrices
Which of the systems of linear equations shown below has a strictly diagonally
dominant coefficient matrix?
a. 3x1 − x2 = −4 b. 4x1 + 2x2 − x3 = −1
2x1 + 5x2 = 2 x1 + 2x3 = −4
3x1 − 5x2 + x3 = 3
solution
a. The coefficient matrix
−1
A= [32 5 ]
∣∣ ∣ ∣ ∣∣ ∣∣
is strictly diagonally dominant because 3 > −1 and 5 > 2 .
b. The coefficient matrix
[ ]
4 2 −1
A= 1 0 2
3 −5 1
is not strictly diagonally dominant because the entries in the second and third rows
do not conform to the definition. For example, in the second row, a21 = 1, a22 = 0,
∣ ∣ ∣ ∣ ∣ ∣
and a23 = 2, and it is not true that a22 > a21 + a23 . Interchanging the second
and third rows in the original system of linear equations, however, produces the
coefficient matrix
[ ]
4 2 −1
A′ = 3 −5 1 ,
1 0 2
which is strictly diagonally dominant.
The next theorem, stated without proof, tells you that a strictly diagonally
dominant coefficient matrix assures the convergence of either the Jacobi method or the
Gauss-Seidel method.
THEOREM 10.1 Convergence of the Jacobi

and Gauss-Seidel Methods
If A is strictly diagonally dominant, then the system of linear equations Ax = b
has a unique solution to which the Jacobi method and the Gauss-Seidel method
will converge for any initial approximation.
9781305658004_1002.indd 507 12/15/15 11:29 AM

In Example 3, you looked at a system of linear equations for which the Jacobi and
Gauss-Seidel methods diverge. The next example shows that by interchanging the rows
of the system in Example 3, you obtain a coefficient matrix that is strictly diagonally
dominant. After this interchange, convergence is assured by Theorem 10.1.
Interchanging Rows to Attain Convergence
Interchange the rows of the system in Example 3 to obtain a system with a strictly
diagonally dominant coefficient matrix. Then apply the Gauss-Seidel method to
approximate the solution to four significant digits.
solution
Begin by interchanging the two rows of the system to obtain
7x1 − x2 = 6
x1 − 5x2 = −4.
Note that the coefficient matrix of this system is strictly diagonally dominant. Then
solve for x1 and x2, as shown below.
x1 = 67 + 17 x2
x2 = 45 + 15 x1
Using the initial approximation (x1, x2) = (0, 0), obtain the sequence of approximations
shown in the table.
REMARK
Do not conclude from
Theorem 10.1 that strict n 0 1 2 3 4 5
diagonal dominance is a x1 0.0000 0.8571 0.9959 0.9999 1.000 1.000
necessary condition for
convergence of the Jacobi or x2 0.0000 0.9714 0.9992 1.000 1.000 1.000
Gauss-Seidel methods (see
Exercise 25 and 26).
So, the solution is x1 = 1 and x2 = 1.
The Jacobi and Gauss-Seidel methods can be applied to

Linear boundary value problems to analyze a region having a
Algebra quantity distributed across it. A grid, or mesh, is drawn
Applied over the region and values are estimated at the grid points.
Some illustration software programs have a gradient mesh
feature that distributes color in a region between and
around user-defined grid points. The diagram below
shows one way to do this. A grid is superimposed over
a triangular region, and the percentages of blue at the
exterior points of the grid are specified. The value of color
at each interior grid point is approximately equal to the
mean of the percentages at surrounding grid points.
c1 = 0.25(80 + 80 + c3 + 60) 100
c2 = 0.25(60 + c3 + 40 + 40)
c3 = 0.25(c1 + 70 + 50 + c2) 80 90
As the mesh spacing decreases, the number

of equations and unknowns in the system 60 c1 80
increases. Numerical methods allow
the illustration software to determine 40 c2 c3 70
the color at each point in the
mesh, resulting in a smooth
distribution of color. 20 30 40 50 60
Slaven/Shutterstock.com
9781305658004_1002.indd 508 12/15/15 11:29 AM

10.2 Exercises 509
10.2 Exercises
The Jacobi Method In Exercises 1–6, apply the Jacobi Interchanging Rows to Attain Convergence In
method to the system of linear equations, using the Exercises 21–24, interchange the rows of the system of
initial approximation (x1, x2, . . . , xn) = (0, 0, . . . 0). linear equations in the stated exercise to obtain a system
Continue performing iterations until two successive with a strictly diagonally dominant coefficient matrix.
approximations are identical when rounded to three Then apply the Gauss-Seidel method to approximate the
significant digits. solution to two significant digits.
1. 3x − x2 = 2 2. −4x1 + 2x2 = −6 21. Exercise 13 22. Exercise 14
1
x1 + 4x2 = 5 3x1 − 5x2 = 1 23. Exercise 15 24. Exercise 16
3. −5x1 + x2 = −14 4. 7x + 4x2 = −4
1 Showing Convergence In Exercises 25 and 26, the
x1 − 12x2 = 8 4x1 + 7x2 = 7
coefficient matrix of the system of linear equations is
5. 2x1 − x2 = 2 not strictly diagonally dominant. Show that the Jacobi
x1 − 3x2 + x3 = −2 and Gauss-Seidel methods converge using an initial
−x1 + x2 − 3x3 = −6 approximation of (x1, x2, . . . , xn) = (0, 0, . . . , 0).
6. 4x1 + x2 + x3 = 7 25. −4x1 + 5x2 = 1 26. 4x1 + 2x2 − 2x3 = 0
x1 − 7x2 + 2x3 = −2 x1 + 2x2 = 3 x1 − 3x2 − x3 = 7
3x1 + 4x3 = 11 3x1 − x2 + 4x3 = 5
The Gauss-Seidel Method In Exercises 7–12, apply the True or False? In Exercises 27–29, determine whether
Gauss-Seidel method to the system of linear equations in each statement is true or false. If a statement is true, give
the stated exercise. a reason or cite an appropriate statement from the text.
7. Exercise 1 8. Exercise 2 If a statement is false, provide an example that shows the
9. Exercise 3 10. Exercise 4 statement is not true in all cases or cite an appropriate
statement from the text.
11. Exercise 5 12. Exercise 6
27. The Jacobi method converges when it produces repeated
Showing Divergence In Exercises 13–16, show iterations that are identical to a specified number of
that the Jacobi and Gauss-Seidel methods diverge significant digits.
for the system using the initial approximation 28. If the Jacobi method or the Gauss-Seidel method
(x1, x2, . . . , xn ) = (0, 0, . . . , 0). diverges, then the system has no solution.
13. x1 − 2x2 = −1 29. If a matrix A is strictly diagonally dominant, then the

2x1 + x2 = 3 system of linear equations Ax = b has no unique solution.
14. −x1 + 4x2 = 1

3x1 − 2x2 = 2 30. CAPSTONE Consider the system
15. 2x1 − 3x2 = −7 ax1 − 5x2 + 2x3 = 19
x1 + 3x2 − 10x3 = 9 3x1 + bx2 − x3 = −1
3x1 + x3 = 13 −2x1 + x2 + cx3 = 9.
16. x1 + 3x2 − x3 = 5
(a) Describe all the values of a, b, and c that will allow
3x1 − x2 =5
you to use the Jacobi method to approximate the
x2 + 2x3 = 1
solution of this system.
Strictly Diagonally Dominant Matrices In Exercises (b) Describe all the values of a, b, and c that will
17–20, determine whether the matrix is strictly guarantee that the Jacobi and Gauss-Seidel
diagonally dominant. methods converge.
−1 −2
[ ] [ ]
2 1
17. 18.
3 5 0 1 31. In Exercise 30, let a = 8, b = 1, and c = 4. What can
[ ] [ ]
12 6 0 7 5 −1 you determine about the convergence or divergence of
19. 2 −3 2 20. 1 −4 1 the Jacobi and Gauss-Seidel methods? Explain.
0 6 13 0 2 −3
9781305658004_1002.indd 509 12/15/15 11:29 AM

10.3 Power Method for Approximating Eigenvalues

Find a dominant eigenvalue and the corresponding dominant
eigenvector.
Use the power method with scaling to approximate a dominant
eigenvector of a matrix, and use the Rayleigh quotient to compute
its dominant eigenvalue.
Dominant Eigenvalues and Dominant Eigenvectors

In Chapter 7, you obtained the eigenvalues of an n × n matrix A by solving its
characteristic equation
λn + cn−1λn−1 + cn−2λn−2 + . . . + c0 = 0.
For large values of n, polynomial equations such as this are difficult and time
consuming to solve. Moreover, numerical techniques for approximating roots of
polynomial equations of high degree are sensitive to rounding errors. This section
introduces an alternative method for approximating eigenvalues. As presented here, the
method can be used only to find the eigenvalue of A that is largest in absolute value—
this eigenvalue is the dominant eigenvalue of A. Although this may seem restrictive,
dominant eigenvalues are of primary interest in many physical applications.
Definitions of Dominant Eigenvalue and Dominant Eigenvector

Let λ1, λ2, . . . , λn be the eigenvalues of an n × n matrix A. λi is the
dominant eigenvalue of A when
∣λi∣ > ∣λj∣, i ≠ j.

The eigenvectors corresponding to λi are the dominant eigenvectors of A.
REMARK
Note that the eigenvalues of
A are λ1 = 1 and λ2 = −1, and Not every matrix has a dominant eigenvalue, such as matrices A and B below.
the eigenvalues of B are λ1 = 2,
[ ]
λ2 = 2, and λ3 = 1. 2 0 0
A= [ 1 0
0 −1 ]
B = 0 2 0
0 0 1
Finding a Dominant Eigenvalue
Find the dominant eigenvalue and corresponding eigenvectors of the matrix
A= [21 −12
−5]
.
solution
From Example 4 in Section 7.1, the characteristic polynomial of A is
λ2 + 3λ + 2 = (λ + 1)(λ + 2). So, the eigenvalues of A are λ1 = −1 and λ2 = −2,
of which the dominant one is λ2 = −2. From the same example, the dominant
eigenvectors of A (those corresponding to λ2 = −2) are of the form
x=t [31], t ≠ 0.
9781305658004_1003.indd 510 12/15/15 11:31 AM

10.3 Power Method for Approximating Eigenvalues 511
The Power Method

As with the Jacobi and Gauss-Seidel methods, the power method for approximating
eigenvalues is iterative. First assume that the matrix A has a dominant eigenvalue with
corresponding dominant eigenvectors. Then choose an initial approximation x0 of one
of the dominant eigenvectors of A. This initial approximation must be a nonzero vector
in Rn. Finally, form the sequence
x1 = Ax0
x2 = Ax1 = A(Ax0) = A2x0
x3 = Ax2 = A(A2x0) = A3x0
⋮
xk = Axk−1 = A(Ak−1x0) = Akx0.
For large values of k, and by properly scaling this sequence, you obtain a good
approximation of the dominant eigenvector of A. Example 2 illustrates this.
Approximating a Dominant Eigenvector

by the Power Method
Complete six iterations of the power method to approximate a dominant eigenvector of
A= [21 −12
−5]
.
solution
Begin with an initial nonzero approximation of
x0 = [11].
Then obtain the approximations shown below.
Iteration “Scaled” Approximation
2 −12 −10
x1 = Ax0 =
1 −5 [ ][ ] [
1
1
=
−4] −4 [2.50
1.00]
x2 = Ax1 = [21 −12 −10

−5][ −4] [ 10]
=
28
10 [2.80
1.00]
x3 = Ax2 = [21 −12 28

=
−64
−5][ 10] [ −22]
−22 [2.91
1.00]
x4 = Ax3 = [21 −12 −64

−5][ −22] [ 46]
=
136
46 [2.96
1.00]
x5 = Ax4 = [21 −12

−5][ 136
46] =[
−280
−94]
−94 [2.98
1.00]
x6 = Ax5 = [21 −12 −280

−5][ −94] [ 190]
=
568
190 [2.99
1.00]

Note that the approximations in Example 2 are approaching scalar multiples of
[31]
which is a dominant eigenvector of the matrix A from Example 1. In this case, the
dominant eigenvalue of the matrix A is known to be λ = −2. When the dominant
eigenvalue of A is unknown, however, you can use the next theorem, which gives a
formula for determining the eigenvalue corresponding to an eigenvector. This theorem
is credited to the English physicist John William Rayleigh (1842–1919).
9781305658004_1003.indd 511 12/15/15 11:31 AM

THEOREM 10.2 Determining an Eigenvalue

from an Eigenvector
If x is an eigenvector of a matrix A, then its corresponding eigenvalue is
Ax ∙ x
λ= .
x∙x
This quotient is called the Rayleigh quotient.
proof
x is an eigenvector of A, so it follows that Ax = λx and
Ax ∙ x λx ∙ x λ(x ∙ x)
= = = λ.
x∙x x∙x x∙x
In cases for which the power method generates a good approximation of a dominant
eigenvector, the Rayleigh quotient provides a correspondingly good approximation of the
dominant eigenvalue. Example 3 demonstrates the use of the Rayleigh quotient.
Approximating a Dominant Eigenvalue
Use the result of Example 2 to approximate the dominant eigenvalue of the matrix
A= [21 −12
−5]
.
solution
In Example 2, the sixth iteration of the power method produced
x6 = [568
190] ≈ 190[
1.00]
2.99
.
With x = (2.99, 1) as the approximation of a dominant eigenvector of A, use the

Rayleigh quotient to obtain an approximation of the dominant eigenvalue of A. First
find the product Ax.
Ax = [21 −12 2.99

=
−6.02
−5][ 1.00] [ −2.01]
Then find the dot products Ax ∙ x and x ∙ x.
Ax ∙ x = (−6.02)(2.99) + (−2.01)(1.00) ≈ −20.0
x ∙ x = (2.99)(2.99) + (1.00)(1.00) ≈ 9.94
The Rayleigh quotient is
Ax ∙ x −20.0
λ= ≈ ≈ −2.01
x∙x 9.94
which is a good approximation of the dominant eigenvalue λ = −2.
As shown in Example 2, the power method tends to produce approximations

REMARK with large entries. In practice it is best to “scale down” each approximation before
Other scaling techniques are proceeding to the next iteration. One way to accomplish this scaling is to determine the
possible. For an example, see component of Axi that has the largest absolute value and multiply the vector Axi by the
Exercises 23 and 24. reciprocal of this component. The resulting vector then has components whose absolute
values are less than or equal to 1.
9781305658004_1003.indd 512 12/15/15 11:31 AM

The Power Method with Scaling
Complete six iterations of the power method with scaling to approximate a dominant
eigenvector, and corresponding dominant eigenvalue, of the matrix
[ ]
1 2 0
A = −2 1 2 .
1 3 1
Use
[]
1
x0 = 1
1
as the initial approximation of the dominant eigenvector.
solution
One iteration of the power method produces
[ ][ ] [ ]
1 2 0 1 3
Ax0 = −2 1 2 1 = 1
1 3 1 1 5
and by scaling you obtain the approximation
[] [ ]
3 0.60
1
x1 = 1 = 0.20 .
5
5 1.00
REMARK
A second iteration yields
Note that the scaling factors
[ ][ ] [ ]
used to obtain the vectors in 1 2 0 0.60 1.00
the table Ax1 = −2 1 2 0.20 = 1.00
x1 5.00 1 3 1 1.00 2.20
x2 2.20 and
x3 2.80
[ ] [ ]
1.00 0.45
x4 3.13 1
x2 = 1.00 ≈ 0.45 .
x5 3.03 2.20
2.20 1.00
x6 3.00
Continuing this process produces the sequence of approximations shown in the table.
approach the dominant
eigenvalue λ = 3.
x0 x1 x2 x3 x4 x5 x6
[ ] [ ] [ ] [ ] [ ] [ ][ ]
1.00 0.60 0.45 0.48 0.50 0.50 0.50
1.00 0.20 0.45 0.55 0.51 0.50 0.50
1.00 1.00 1.00 1.00 1.00 1.00 1.00
From the table, the approximate dominant eigenvector of A is
[ ]
0.50
x = 0.50 .
1.00
Then use the Rayleigh quotient to approximate the dominant eigenvalue of A to be
λ = 3. (For this example, check that the approximations of x and λ are exact.)
9781305658004_1003.indd 513 12/15/15 11:31 AM

In Example 4, the power method with scaling converges to a dominant eigenvector.

The next theorem states that a sufficient condition for convergence of the power method
is that the matrix A be diagonalizable (and have a dominant eigenvalue).
THEOREM 10.3 Convergence of the Power Method

If A is an n × n diagonalizable matrix with a dominant eigenvalue, then there
exists a nonzero vector x0 such that the sequence of vectors
Ax0, A2x0, A3x0, A4x0, . . . , Akx0, . . .
approaches a multiple of the dominant eigenvector of A.
proof
A is diagonalizable, so from Theorem 7.5 it has n linearly independent eigenvectors
x1, x2, . . . , xn with corresponding eigenvalues of λ1, λ2, . . . , λn. Assume that these
eigenvalues are ordered so that λ1 is the dominant eigenvalue (with a corresponding
eigenvector of x1). The n eigenvectors x1, x2, . . . , xn are linearly independent, so they
must form a basis for Rn. For the initial approximation x0, choose a nonzero vector such
that the linear combination x0 = c1x1 + c2x2 + . . . + cnxn has a nonzero leading
coefficient c1. (When c1 = 0, the power method may not converge to the dominant
eigenvector, and a different x0 must be used as the initial approximation. See Exercises
19 and 20.) Now, multiplying both sides of this equation by A produces
Ax0 = A(c1x1 + c2x2 + . . . + cnxn)
= c1(Ax1) + c2(Ax2) + . . . + cn(Axn )
= c1(λ1x1) + c2(λ2x2) + . . . + cn(λnxn).
Repeated multiplication of both sides of this equation by A produces
Akx0 = c1(λk1x1) + c2(λk2x2) + . . . + cn(λknxn)
which implies that
λ2 λ k
[ (λ ) x ( ) ]
k
Akx0 = λ1k c1x1 + c2 + . . . + cn n xn .
1
2
λ1
Now, from the original assumption that λ1 is larger in absolute value than the other
eigenvalues, it follows that each of the fractions
λ2 λ3 λ
, , . . . , n
λ1 λ1 λ1
is less than 1 in absolute value. So, as k approaches infinity, each of the factors
λ2 k λ3 k λn
( ) ( ) ( )
k
, , . . . ,
λ1 λ1 λ1
must approach 0. This implies that the approximation Akx0 ≈ λk1c1x1, c1 ≠ 0, improves
as k increases. x1 is a dominant eigenvector, so it follows that any scalar multiple of x1
is also a dominant eigenvector, which shows that Akx0 approaches a multiple of the
dominant eigenvector of A.
The proof of Theorem 10.3 provides some insight into the rate of convergence of
the power method. That is, if the eigenvalues of A are ordered so that
∣λ1∣ > ∣λ2∣ ≥ ∣λ3∣ ≥ . . . ≥ λ

n ∣ ∣
∣ ∣∣ ∣
then the power method converges quickly when λ2 λ1 is small, and slowly when
∣ ∣∣ ∣
λ2 λ1 is close to 1. Example 5 illustrates this principle.
9781305658004_1003.indd 514 12/15/15 11:31 AM

Rate of Convergence of the Power Method
a. The matrix
A= [46 5
5 ]
∣ ∣∣ ∣
has eigenvalues of λ1 = 10 and λ2 = −1. (Check this.) So the ratio λ2 λ1 is 0.1.
For this matrix, it only takes four iterations to obtain successive approximations that
agree when rounded to three significant digits.
x0 x1 x2 x3 x4
REMARK [1.000
1.000] [ 1.000] [ 1.000] [ 1.000] [ 1.000]
0.818 0.835 0.833 0.833
The approximations shown in

this example are rounded to
three significant digits, but the b. The matrix
intermediate calculations
performed to obtain these
approximations have a higher
A= [−47 10
5 ]
degree of accuracy to minimize has eigenvalues of λ1 = 10 and λ2 = −9. (Check this.) For this matrix, the ratio
rounding error. You may
get slightly different results,
∣ ∣∣ ∣
λ2 λ1 is 0.9, and it takes 68 iterations to obtain successive approximations that
agree to three significant digits.
depending on your rounding
method or the method used
x0 x1 x2 x66 x67 x68
by your software program or
graphing utility. . . .
[ 1.000
1.000 ][ 0.500
1.000 ] [ 0.941
1.000 ] . . . [0.715
1.000] [ 0.714
1.000] [ 1.000]
0.714
This section used the power method to approximate the dominant eigenvalue of
a matrix. This method can be modified to approximate other eigenvalues through use
of a procedure called deflation. Moreover, the power method is only one of several
techniques that can be used to approximate the eigenvalues of a matrix. Another
popular method is called the QR algorithm. This is the method used in most programs
and calculators for finding eigenvalues and eigenvectors. The QR algorithm uses the
QR-factorization of the matrix, as presented in Chapter 5. Discussions of the deflation
method and the QR algorithm are in most texts on numerical methods.
Recall from Section 2.5 the application of Markov chains to

Linear Google’s PageRank algorithm. For a search set that contains
Algebra n web pages, define an n × n matrix A such that aij = 1
Applied when page j references page i and aij = 0 otherwise. Adjust
A to account for web pages without external references,
scale each column of A so that A is stochastic, and call this
matrix B. Then define
1−p
M = pB + E
n
where p is the probability that a user follows a link on a
page, 1 − p is the probability that the user goes to any page
at random, and E is an n × n matrix whose entries are all 1.
The Markov chain whose matrix of transition probabilities is
M converges to a unique steady state matrix, which gives
an estimage of page ranks. This steady state matrix is a
dominant eigenvector of M, so using the power method
on M produces an estimate of the steady state matrix.
D8nn/Shutterstock.com
9781305658004_1003.indd 515 12/15/15 11:31 AM

10.3 Exercises
Finding a Dominant Eigenvector In Exercises 1–4, 17. Rate of Convergence of the Power Method
use the techniques presented in Chapter 7 to find the (a) Find the eigenvalues of
eigenvalues of the matrix A. When A has a dominant
eigenvalue, find a corresponding dominant eigenvector. A=
2
1 [1
2
and B = ] 2
1 [
3
4
. ]
−1 −4 −2
1. A =
−2[ 1
]2. A =
3 −6
0
[ ] (b) Apply four iterations of the power method with
scaling to each matrix in part (a), starting with
[ ] [ ]
1 3 0 −5 0 0 x0 = [−1 2]T.
3. A = 2 1 4. A =
2 3 5 0 (c) Compute the ratios λ2λ1 for A and B. For which
1 1 0 4 −2 3 matrix do you expect faster convergence?
Using the Rayleigh Quotient In Exercises 5 and 6,
use the Rayleigh quotient to find the eigenvalue λ of the 18. C
APSTONE Rework Example 2 using the
matrix A corresponding to the eigenvector x. power method with scaling. Compare your answer
4 −5 with the one found in Example 2.
5. A = [2 −3
,x=] []5
2
Other Eigenvectors In Exercises 19 and 20, (a) find the
6. A = [ ],x=[ ]
5 0 4
6 −3 3 eigenvalues and corresponding eigenvectors of A, (b) use
the initial approximation x0 to complete two iterations
Eigenvectors and Eigenvalues In Exercises 7–10, of the power method with scaling, and (c) explain why
use the power method with scaling to approximate a the method does not seem to converge to a dominant
dominant eigenvector of the matrix A. Start with eigenvector.
x0 = [1 1]T and complete five iterations. Then use x5 to −1
approximate the dominant eigenvalue of A. 19. A = [−23 4 ]
, x0 =
1
1 []
[10 −95] [−11 0
]
[ ] []
7. A = 8. A = −3 0 2 1
6 20. A = 0 −1 0 , x0 = 1
1 −4 −3
9. A = [
8]
10. A = [ ]
6 0 1 −2 1

−2 −2 1
Other Eigenvalues In Exercises 21 and 22, observe
Eigenvectors and Eigenvalues In Exercises 11–14, that Ax = λx implies that A−1x = (1λ)x. Apply five
use the power method with scaling to approximate iterations of the power method with scaling on A−1 to
a dominant eigenvector of the matrix A. Start with approximate the eigenvalue of A with the smallest
x0 = [1 1 1]T and complete four iterations. Then use magnitude.
x4 to approximate the dominant eigenvalue of A.
[ ]
2 3 1
2 −12
[ ]
[ ] [ ]
3 0 0 1 2 0 21. A = 22. A = 0 −1 2
11. A = 1 −1 0 12. A = 0 −7 1 1 −5
0 0 3
0 2 8 0 0 0
Another Scaling Technique In Exercises 23 and
[ ] [ ]
−1 −6 0 0 6 0
24, apply four iterations of the power method with
13. A = 2 7 0 14. A = 0 −4 0
another scaling technique to approximate the dominant
1 2 −1 2 1 1
eigenvalue of the matrix: after each iteration, scale the
The Power Method with Scaling In Exercises 15 and approximation by dividing by the length so that the
16, the matrix A does not have a dominant eigenvalue. resulting approximation is a unit vector.
[ ]
Apply the power method with scaling, starting with 7 −4 2
[ ]
5 6
x0 = [1 1 1]T, and observe the results of the first four 23. A = 24. A = 16 −9 6
iterations. 4 3
8 −4 5
[ ] [ ]
1 1 0 1 2 −2
25.
Use the proof of Theorem 10.3 to show that
15. A = 3 −1 0 16. A = −2 5 −2
A(Akx0) ≈ λ1(Akx0) for large values of k. That is, show
0 0 −2 −6 6 −3
that the scale factors obtained by the power method
approach the dominant eigenvalue.
9781305658004_1003.indd 516 12/15/15 11:31 AM

10.4 Applications of Numerical Methods 517
10.4 Applications of Numerical Methods

Find a least squares regression polynomial for a set of data.
Apply the Gauss-Seidel method to determine the probability of an
event.
Model population growth using an age transition matrix and use
the power method to find a stable age distribution vector.
Applications of Gaussian Elimination with

Partial PivotinG
In Sections 2.6 and 5.4, you used least squares regression analysis to find mathematical
models that best fit a set of n points in the plane. This procedure can be generalized as
shown below.
Regression Analysis for Polynomials

The least squares regression polynomial of degree m for the points
{(x1, y1), (x2, y2), . . . , (xn, yn)}
is
y = a0 + a1x + a2x 2 + . . . + am−1x m−1 + am x m
where the coefficients are determined by the system of m + 1 linear equations
below.
na0 + (∑xi)a1 + (∑xi2)a2 + . . . + (∑xim)am = ∑yi
(∑xi)a0 + (∑xi2)a1 + (∑xi3)a2 + . . . + (∑xim+1)am = ∑xi yi
(∑x2i )a0 + (∑xi3)a1 + (∑xi4)a2 + . . . + (∑xim+2)am = ∑x2i yi
⋮
(∑xim)a0 + (∑xim+1)a1 + (∑xim+2)a2 + . . . + (∑xi2m)am = ∑ximyi
Note that when m = 1, this system of equations reduces to
(∑ x )a = ∑ y
na0 + i 1 i
(∑ x )a + (∑ x )a = ∑ x y
i 0 i
2
1 i i
which has a solution of

n∑xi yi − (∑xi )(∑yi )
a1 =
n∑xi2 − (∑xi )2
∑yi ∑x
a0 = − a1 i.
n n
In Exercise 22, you are asked to show that this formula is equivalent to the matrix
formula for linear regression presented in Section 2.6.
Example 1 illustrates the use of regression analysis to find a second-degree
polynomial model.
9781305658004_1004.indd 517 12/15/15 11:33 AM

Least Squares Regression Analysis
The table shows the world populations (in billions) for selected years from 1985
through 2015. (Source: U.S. Census Bureau)
Year 1985 1990 1995 2000 2005 2010 2015

Population 4.86 5.29 5.70 6.09 6.47 6.87 7.26
Find the second-degree least squares regression polynomial for the data and use the
resulting model to predict the world populations in 2020 and 2025.
solution
Begin by letting x = 0 represent 1985, letting x = 1 represent 1990, and so on. So,
the collection of points is {(0, 4.86), (1, 5.29), (2, 5.70), (3, 6.09), (4, 6.47), (5, 6.87),
(6, 7.26)}, which yields
7 7 7
n = 7, ∑ x = 21,
i=1
i ∑x
i=1
i
2
= 91, ∑x
i=1
i
3
= 441,
7 7 7 7
∑x
i=1
i
4
= 2275, ∑ y = 42.54, ∑ x y = 138.75, ∑ x y = 619.53.
i=1
i
i=1
i i
i=1
2
i i
The system of linear equations giving the coefficients of the quadratic model
y = a0 + a1x + a2x2
is
7a0 + 21a1 + 91a2 = 42.54
21a0 + 91a1 + 441a2 = 138.75
91a0 + 441a1 + 2275a2 = 619.53.
Gaussian elimination with partial pivoting on the matrix
[ ]
7 21 91 42.54
21 91 441 138.75
91 441 2275 619.53
y and rounding to four decimal places after each intermediate calculation produces
[ ]
9 2020 1.0000 4.8462 25.0000 6.8080
8
2025 0.0000 1.0000 6.4998 0.3959 .
7 Predicted
6 points
0.0000 0.0000 1.0000 −0.0033
5 So, back-substitution produces the solution
4
3 a2 ≈ −0.0033, a1 ≈ 0.4173, a0 ≈ 4.8682
2
1 and the regression quadratic model is
x
−1 1 2 3 4 5 6 7 8 9 y = 4.8682 + 0.4173x − 0.0033x 2.
Figure 10.1 Figure 10.1 compares this model with the collecion of points. To predict the world
population in 2020, let x = 7, and obtain
y = 4.8682 + 0.4173(7) − 0.0033(72) ≈ 7.63 billion.
Similarly, the prediction for 2025 (x = 8) is
y = 4.8682 + 0.4173(8) − 0.0033(82) ≈ 8.00 billion.
9781305658004_1004.indd 518 12/15/15 11:33 AM

Least Squares Regression Analysis
Find the third-degree least squares regression polynomial

y = a0 + a1 x + a2 x 2 + a3 x3
for the points {(0, 0), (1, 2), (2, 3), (3, 2), (4, 1), (5, 2), (6, 4)}.
solution
For this set of points, the linear system
na0 + (∑ xi)a1 + (∑ xi2)a2 + (∑ xi3)a3 = ∑ yi
(∑ xi)a0 + (∑ xi2)a1 + (∑ xi3)a2 + (∑ xi4)a3 = ∑ xi yi
(∑ xi2)a0 + (∑ xi3)a1 + (∑ xi4)a2 + (∑ xi5)a3 = ∑ xi2 yi
(∑ xi3)a0 + (∑ xi4)a1 + (∑ xi5)a2 + (∑ xi6)a3 = ∑ xi3yi
becomes
7a0 + 21a1 + 91a2 + 441a3 = 14
21a0 + 91a1 + 441a2 + 2275a3 = 52
91a0 + 441a1 + 2275a2 + 12,201a3 = 242
441a0 + 2275a1 + 12,201a2 + 67,171a3 = 1258.
Using Gaussian elimination with partial pivoting on the matrix
[ ]
7 21 91 441 14
21 91 441 2275 52
91 441 2275 12,201 242
441 2275 12,201 67,171 1258
and rounding to four decimal places after each intermediate calculation produces
[ ]
1.0000 5.1587 27.6667 152.3152 2.8526
0.0000 1.0000 8.5322 58.3539 0.6183
0.0000 0.0000 1.0000 9.7697 0.1285
0.0000 0.0000 0.0000 1.0000 0.1670
which implies
a3 ≈ 0.1670, a2 ≈ −1.5030, a1 ≈ 3.6971, a0 ≈ −0.0731.
So, the cubic model is
y = −0.0731 + 3.6971x − 1.5030x2 + 0.1670x3.
The figure below compares this model with the points.
y
4
(6, 4)
(2, 3)
3
(3, 2)
2 (5, 2)
(1, 2)
1
(4, 1)
(0, 0)
x
1 2 3 4 5 6
9781305658004_1004.indd 519 12/15/15 11:33 AM

Applications of the Gauss-Seidel Method
An Application to Probability
Figure 10.2 is a diagram of a maze used in a laboratory experiment. The experiment

begins by placing a mouse at one of the ten interior intersections of the maze. Once
the mouse emerges in the outer corridor, it cannot return to the maze. When the mouse
1 is at an interior intersection, its choice of paths is assumed to be random. What is the
2 3 probability that the mouse emerges in the “food corridor” when it begins at the ith
intersection?
4 5 6
solution
7 8 9 10
Let the probability of winning (getting food) when starting at the ith intersection be
Food represented by pi. Then form a linear equation involving pi and the probabilities
associated with the intersections bordering the ith intersection. For example, at the
Figure 10.2 first intersection the mouse has a probability of 14 of choosing the upper right path and
losing, a probability of 14 of choosing the upper left path and losing, a probability of 14
of choosing the lower right path (at which point it has a probability of p3 of winning),
and a probability of 14 of choosing the lower left path (at which point it has a probability
of p2 of winning). So
p1 = 14 (0) + 14 (0) + 14 p3 + 14 p2.
Upper Upper Lower Lower

right left right left
The other nine probabilities can be represented using similar reasoning. So, you have
the ten equations below.
p1 = 14 (0) + 14 (0) + 14 p3 + 14 p2
p2 = 15 (0) + 15 p1 + 15 p3 + 15 p4 + 15 p5
p3 = 15 (0) + 15 p1 + 15 p2 + 15 p5 + 15 p6
p4 = 15 (0) + 15 p2 + 15 p5 + 15 p7 + 15 p8
p5 = 16 p2 + 16 p3 + 16 p4 + 16 p6 + 16 p8 + 16 p9
p6 = 15 (0) + 15 p3 + 15 p5 + 15 p9 + 15 p10
p7 = 14 (0) + 14 (1) + 14 p4 + 14 p8
p8 = 15 (1) + 15 p4 + 15 p5 + 15 p7 + 15 p9
p9 = 15 (1) + 15 p5 + 15 p6 + 15 p8 + 15 p10
p10 = 14 (0) + 14 (1) + 14 p6 + 14 p9
Rewriting these equations produces the system of ten linear equations in ten variables
below.
4p1 − p2 − p3 =0
−p1 + 5p2 − p3 − p4 − p5 =0
−p1 − p2 + 5p3 − p5 − p6 =0
− p2 + 5p4 − p5 − p7 − p8 =0
− p2 − p3 − p4 + 6p5 − p6 − p8 − p9 =0
− p3 − p5 + 5p6 − p9 − p10 =0
− p4 + 4p7 − p8 =1
− p4 − p5 − p7 + 5p8 − p9 =1
− p5 − p6 − p8 + 5p9 − p10 =1
− p6 − p9 + 4p10 =1
9781305658004_1004.indd 520 12/15/15 11:33 AM

The augmented matrix for this system is

4 −1 −1 0 0 0 0 0 0 0 0
−1 5 −1 −1 −1 0 0 0 0 0 0
−1 −1 5 0 −1 −1 0 0 0 0 0
0 −1 0 5 −1 0 −1 −1 0 0 0
0 −1 −1 −1 6 −1 0 −1 −1 0 0
.
0 0 −1 0 −1 5 0 0 −1 −1 0
0 0 0 −1 0 0 4 −1 0 0 1
0 0 0 −1 −1 0 −1 5 −1 0 1
0 0 0 0 −1 −1 0 −1 5 −1 1
0 0 0 0 0 −1 0 0 −1 4 1
Using the Gauss-Seidel method with an initial approximation of p1 = p2 = . . . = p10 = 0
produces (after 18 iterations) an approximation of
p1 = 0.090, p2 = 0.180
p3 = 0.180, p4 = 0.298
p5 = 0.333, p6 = 0.298
p7 = 0.455, p8 = 0.522
p9 = 0.522, p10 = 0.455.
The structure of the problem described in Example 3 is related to a technique called

finite element analysis, which is used in many engineering problems.
Note that the matrix in Example 3 has mostly zero entries. Such matrices are called
sparse. For solving systems of equations with sparse coefficient matrices, the Jacobi
and Gauss-Seidel methods are much more efficient than Gaussian elimination.
One engineering application of finite element analysis is

Linear in the simulation of how a car will deform in a crash. This
Algebra allows engineers to identify potential problem areas in a
Applied car design prior to manufacturing. The car is modeled by a
complex mesh of nodes in order to analyze the distribution
of stresses and displacements. Consider a simplified
example of a beam with two nodes, as illustrated below.
Node Node x
u2 u4
u1 u3
The nodes at the ends of the beam can have displacements

in the x- and y-directions, denoted by u1, u2, u3, and u4,
and the stresses imposed on the beam during displacement
result in internal forces F1, F2, F3, and F4 . Each Fi is related
to uj by a stiffness coefficient kij. These coefficients are
used as elements of a 4 × 4 stiffness matrix for the beam.
(Recall working with stiffness matrices in Section 2.3.) The
matrix is used to determine the displacements of the nodes
for different force vectors F. For complex three-dimensional
structures, the analysis takes into consideration other
material and structural properties besides stiffness, which
determine how the structure will react in various conditions.
iStockphoto.com/rkankaro
9781305658004_1004.indd 521 12/15/15 11:33 AM

Applications of the Power Method

Section 7.4 introduced the idea of an age transition matrix as a model for population
growth. Recall that this model was developed by grouping the population into n age
classes of equal duration. So, for a maximum life span of M years, the intervals listed
below represent the age classes.
[0, Mn) First age class
[ Mn, 2Mn) Second age class
[ 2Mn, 3Mn) Third age class
[ 3Mn, 4Mn) Fourth age class
⋮ ⋮
[ (n −n1)M, M] nth age class
The age distribution vector x represents the number of population members in each age
class, where
[]
x1 Number in first age class
x2 Number in second age class
x= .
⋮ ⋮
xn Number in nth age class
Over a period of Mn years, the probability that a member of the ith age class will
survive to become a member of the (i + 1)th age class is pi, where
0 ≤ pi ≤ 1, i = 1, 2, . . . , n − 1.
The average number of offspring produced by a member of the ith age class is bi, where
0 ≤ bi, i = 1, 2, . . . , n.
These numbers can be written in matrix form as shown below.
[ ]
b1 b2 b3 . . . bn−1 bn
p1 0 0 . . . 0 0
L = 0 p2 0 . . . 0 0
⋮ ⋮ ⋮ ⋮ ⋮
0 0 0 . . . pn−1 0
Multiplying this age transition matrix L (also known as a Leslie matrix) by the age
distribution vector for a specific time period produces the age distribution vector for
the next time period. That is,
Lxj = xj+1.
In Section 7.4 you saw that the growth pattern for a population is stable when the same
percentage of the total population is in each age class each year. That is,
Lxj = xj+1 = λxj.
For populations with many age classes, the solution of this eigenvalue problem can be
found using the power method with scaling, as illustrated in Example 4.
9781305658004_1004.indd 522 12/15/15 11:33 AM

A Population Growth Model
Assume that a population of human females has the characteristics listed at the left. The
Age Class Female
table shows the age class (in years), the average number of female children born to the
(in years) Children Probability
members of each age class, and the probability of surviving to the next age class. Find
[0, 10) 0.000 0.985 a stable age distribution vector for this population.
[10, 20) 0.174 0.996 solution
[20, 30) 0.782 0.994 The age transition matrix for this population is
[30, 40) 0.263 0.990 0.000 0.174 0.782 0.263 0.022 0.000 0.000 0.000 0.000 0.000
0.985 0 0 0 0 0 0 0 0 0
[40, 50) 0.022 0.975 0 0.996 0 0 0 0 0 0 0 0
[50, 60) 0.000 0.940 0 0 0.994 0 0 0 0 0 0 0
0 0 0 0.990 0 0 0 0 0 0
[60, 70) 0.000 0.866 A= .
0 0 0 0 0.975 0 0 0 0 0
[70, 80) 0.000 0.680 0 0 0 0 0 0.940 0 0 0 0
[80, 90) 0.000 0.361 0 0 0 0 0 0 0.866 0 0 0
0 0 0 0 0 0 0 0.680 0 0
[90, 100] 0.000 0.000
0 0 0 0 0 0 0 0 0.361 0
To apply the power method with scaling to find an eigenvector for this matrix, use
an initial approximation of x0 = [1 1 1 1 1 1 1 1 1 1]T. An approximation
for an eigenvector of A, with the percentage of each age in the total population, is
shown below.
Percentage in
Eigenvector Age Class Age Class
1.000 [0, 10) 15.27
0.925 [10, 20) 14.13
0.864 [20, 30) 13.20
0.806 [30, 40) 12.31
0.749 [40, 50) 11.44
x=
0.686 [50, 60) 10.48
0.605 [60, 70) 9.24
0.492 [70, 80) 7.51
0.314 [80, 90) 4.80
0.106 [90, 100] 1.62
The eigenvalue corresponding to the eigenvector x in Example 4 is λ ≈ 1.065.

That is,
1.000 1.065 1.000
0.925 0.985 0.925
0.864 0.921 0.864
0.806 0.859 0.806
0.749 0.798 0.749
Ax = A ≈ ≈ 1.065 .
0.686 0.730 0.686
0.605 0.645 0.605
0.492 0.524 0.492
0.314 0.335 0.314
0.106 0.113 0.106
This means that the population in Example 4 increases by about 6.5% every 10 years.
9781305658004_1004.indd 523 12/15/15 11:33 AM

10.4 Exercises .
Least Squares Regression Analysis In Exercises 1–6, 14. Health Expenditures The table shows the total
find the second-degree least squares regression polynomial national health expenditures (in trillions of dollars) in the
for the data. Then graphically compare the model with the United States from 2006 through 2013. (Source: U.S.
data. Census Bureau)
1. (−2, 1), (−1, 0), (0, 0), (1, 1), (3, 2)
Year 2006 2007 2008 2009
2. (0, 4), (1, 2), (2, −1), (3, 0), (4, 1), (5, 4)
3. (−2, 1), (−1, 2), (0, 6), (1, 3), (2, 0), (3, −1) Health
Expenditures 2.167 2.304 2.414 2.506
4. (1, 1), (2, 1), (3, 0), (4, −1), (5, −4)
5. (0, 0.44), (1, 0.78), (2, 1.43), (3, 2.92), (4, 5.08)
Year 2010 2011 2012 2013
6. (−2, −3.52), (−1, 0.09), (0, 3.84), (1, 6.53), (2, 9.06)
Health
Expenditures 2.604 2.705 2.817 2.919
Least Squares Regression Analysis In Exercises
7–12, find the third-degree least squares regression
polynomial for the data. Then graphically compare the (a)
Find the second-degree least squares regression
model with the data. polynomial for the data. Let x = 6 correspond to
2006.
7. (0, 0), (1, 2), (2, 4), (3, 1), (4, 0), (5, 1)
(b) Use the model found in part (a) to predict the
8. (1, 1), (2, 4), (3, 4), (5, 1), (6, 2)
expenditures for the years 2018 through 2020.
9. (−3, 4), (−1, 1), (0, 0), (1, 2), (2, 5)
15. Population The table shows the total numbers of
10. (−7, 2), (−3, 0), (1, −1), (2, 3), (4, 6) people (in millions) in the United States 65 years of age
11. (−1, 0), (0, 3), (1, 2), (2, 0), (3, −2), (4, −3) or older from 2009 through 2013. (Source: U.S. Census
12. (0, 0), (2, 10), (4, 12), (6, 0), (8, −8) Bureau)
Year 2009 2010 2011 2012 2013
13. Decompression Sickness The table shows
recommended diving depth and time limits for Number
recreational divers to reduce the probability of of People 39.5 40.4 41.4 43.1 44.7
acquiring decompression sickness. (Source: U.S. Navy)
(a)
Find the third-degree least squares regression

Depth (in feet) 35 40 50 60 70 polynomial for the data. Let x = 9 correspond to
2009.
Recommended
Maximum Time 310 200 100 60 50 (b) Use the model found in part (a) to predict the total
(in minutes) numbers of people in the United States 65 years of
age or older in the years 2018 through 2020.
16. Food Stamp Benefits The table shows the numbers
Depth (in feet) 80 90 100 110 of households (in millions) receiving food stamp and
Recommended supplemental nutrition assistance in the United States
Maximum Time 40 30 25 20 from 2009 through 2013. (Source: U.S. Census Bureau)
(in minutes)
Year 2009 2010 2011 2012 2013
(a) Find the second-degree least squares regression Amount
Spent 11.7 13.6 14.9 15.8 15.7
polynomial for the data.
(b) Sketch the graph of the model found in part (a). (a)
Find the third-degree least squares regression
(c) Use the model found in part (a) to approximate the polynomial for the data. Let x = 9 correspond
maximum number of minutes a diver should stay at to 2009.
a depth of 120 feet. (b)
Use the model found in part (a) to predict the
(d) Compare your answer to part (c) with the numbers of households receiving such assistance in
recommended time of 15 minutes. Is the model 2018 and 2019. Do your answers seem reasonable?
found in part (a) accurate? Explain. Explain.
9781305658004_1004.indd 524 12/15/15 11:33 AM

10.4 Exercises 525
17. Trigonometry Find the second-degree least squares 24.

regression polynomial for the points
(− π2 , 0), (− π3 , 12), (0, 1), (π3 , 12), (π2 , 0). 1 2 3
Then use the result to approximate cos(π4). Compare 4 5 6

the approximation with the exact value.
18. Trigonometry Rework Exercise 17 but with a Food
fourth-degree least squares regression polynomial.
19. Trigonometry Find the third-degree least squares Temperature In Exercises 25 and 26, the figure shows an
regression polynomial for the points insulated thin metal plate. The steady-state temperature
at an interior junction is approximately equal to the mean
(− π3 , −√3), (− π4 , −1), (0, 0), (π4 , 1), (π3 , √3). of the temperatures at the four surrounding junctions.
Use a system of equations and the Gauss-Seidel method
Then use the result to approximate tan(π6). Compare to approximate the interior temperatures for each set of
the approximation with the exact value. boundary temperatures w, x, y, and z (in degrees Celsius).
20. Trigonometry Rework Exercise 19 but with a least Then explain how the boundary temperature changes
squares regression line. from (a) to (b) affect the interior temperatures.
21. Least Squares Regression Line Find the least 25. x° x° x°
squares regression line for the population data from
Example 1. Then use the model to predict the world
populations in 2020 and 2025, and compare the results w° T1 T2 T3 y°
with the predictions obtained in Example 1.
22. Least Squares Regression Line Show that the
w° T4 T5 T6 y°
formula for the least squares regression line presented
in Section 2.6 is equivalent to the formula presented in
this section. That is, if T7 T8 T9
w° y°
[] [ ]
y1 1 x1
Y=
y2
⋮
, X =
1
⋮
x2
⋮
, A = [aa ]
0
1
z° z° z°
yn 1 xn
(a) w = x = 100, y = z = 0
then the matrix equation A = (X T X)−1X T Y is equivalent to (b) w = x = 110, y = z = 10
n∑ xi yi − (∑ xi)(∑ yi) ∑ yi ∑x 26.
a1 = and a0 = − a1 i. x° x° x° x°
− (∑ xi) n n
2
n∑ xi2
Probability In Exercises 23 and 24, a researcher w° T1 T2 T3 T4 y°

performs the experiment in Example 3 with the maze
shown in the figure. Find the probability that the mouse
emerges in the food corridor when it begins at the ith w° T5 T6 T7 T8 y°
intersection.
23.
w° T9 T10 T11 T12 y°
1 2 3
z° z° z° z°
4 5 6
w = 80, x = 120, y = 40, z = 0
(a)
w = 70, x = 110, y = 30, z = −10
(b)
7 8 9
Food
9781305658004_1004.indd 525 12/15/15 11:33 AM

Stable Age Distribution In Exercises 27–36, the 39. Television Watching A college dormitory houses
matrix represents the age transition matrix for a 200 students. Those who watch an hour or more of
population. Use the power method with scaling to find a television on any day always watch for less than an
stable age distribution vector. hour the next day. One-fourth of those who watch
television for less than an hour one day will watch an
[ ]
1 4
27. A = 1
hour or more the next day. Half of the students watched
2 0 television for an hour or more today. Use the power
[ ]
1 2 method to approximate a dominant eigenvector for the
28. A = 1 corresponding stochastic matrix.
4 0
40. Smokers and Nonsmokers In a population of
[ ]
1 3 10,000, there are 5000 nonsmokers, 2500 smokers of
29. A = 1
0 one pack or less per day, and 2500 smokers of more
5
than one pack per day. During any month, there is a
[ ]
1 5 5% probability that a nonsmoker will begin smoking
30. A = 1
3 0 a pack or less per day, and a 2% probability that a
nonsmoker will begin smoking more than a pack
[ ]
1 2 2
1 per day. For smokers who smoke a pack or less per
31. A = 3 0 0
1 day, there is a 10% probability of quitting and a 10%
0 3 0 probability of increasing to more than a pack per
[ ]
0 1 2 day. For smokers who smoke more than a pack per
1
32. A = 2 0 0 day, there is a 5% probability of quitting and a 10%
1 probability of dropping to a pack or less per day. Use the
0 4 0
power method to approximate a dominant eigenvector
[ ]
0 2 2
1 for the corresponding stochastic matrix.
33. A = 2 0 0
0 1 0 41. Writing In Example 2 in Section 2.5, the stochastic
matrix
[ ]
1 4 2
[ ]
34. A = 3
0 0 0.70 0.15 0.15
4
0 1
0 P = 0.20 0.80 0.15
4
0.10 0.05 0.70
[ ]
1 7 20
35. A = 0.2 0 0 represents the transition probabilities for a consumer
0 0.3 0 preference model. Use the power method to approximate
a dominant eigenvector for this matrix. How does the
[ ]
0 7 10
36. A = 0.5 0 0 approximation relate to the steady state matrix described
in the discussion following Example 3 in Section 2.5?
0 0.5 0
37. Population Growth Model A population has the 42. CAPSTONE

characteristics below.
(a) Explain how to use Gaussian elimination with
(a) A total of 75% of the population survives its first pivoting to find a least squares regression polynomial.
year. Of that 75%, 25% survives its second year.
(b) Explain how to use finite element analysis to
The maximum life span is 3 years.
determine the probability of an event.
(b) The average number of offspring for each member
(c) Explain how to model population growth with an
of the population is 2 the first year, 4 the second
age distribution matrix and use the power method
year, and 2 the third year.
to find a stable age distribution vector.
Find a stable age distribution vector for this population.
38. Population Growth Model A population has the 43. Fibonacci Sequence Apply the power method to
characteristics below.
(a) A total of 80% of the population survives the first
year. Of that 80%, 25% survives the second year.
A= [11 1
0 ]
The maximum life span is 3 years. which was discussed in Project 2 of Chapter 7 (The
(b) The average number of offspring for each member Fibonacci Sequence). Use the power method to
of the population is 3 the first year, 6 the second approximate the dominant eigenvalue of A.
year, and 3 the third year.
Find a stable age distribution vector for this population.
9781305658004_1004.indd 526 12/15/15 11:33 AM

Review Exercises 527
10 Review
Floating Point Form In Exercises 1–6, express the real Solving an ill-Conditioned System In Exercises 21 and
number in floating point form. 22, use Gaussian elimination to solve the ill-conditioned
1. 528.6 2. 475.2 system of linear equations, rounding each intermediate
calculation to three significant digits. Then compare this
3. −4.85 4. −22.5
solution with the exact solution provided.
5. 3 2 6. 4 78
1
21. x + y = −1
Finding Stored Values In Exercises 7–12, determine 999 4001
x+ y=
the stored value of the real number in a computer that (a) 1000 1000
rounds to three significant digits and (b) rounds to four x = 5000, y = −5001)
(Exact:
significant digits. 22. x−y= −1
7. 25.2 8. −41.2 99 20,101
−
x+y=
9. −250.231 10. 628.742 100 100
5
11. 12 3
12. 16 x = 20,001, y = 20,002)
(Exact:
Propagation of Rounding Error In Exercises 13 and The Jacobi Method In Exercises 23 and 24, apply the
14, evaluate the determinant of the matrix, rounding Jacobi method to the system of linear equations, using
each intermediate calculation to three significant digits. the initial approximation
Then find the exact value and compare the two results. (x1, x2, x3, . . . , xn) = (0, 0, 0, . . . , 0).
13. [20.24
12.5 2.5
6.25] 14. [10.25
8.5
]
3.2
8.1
Continue performing iterations until two successive
approximations are identical when rounded to three
significant digits.
Gaussian Elimination and Rounding Error In
Exercises 15 and 16, use Gaussian elimination to solve 23. 2x1 − x2 = −1 24. x1 + 4x2 = −3

the system. After each intermediate calculation, round x1 + 2x2 = 7 2x1 + x2 = 1
the result to three significant digits. Then find the exact
The Gauss-Seidel Method In Exercises 25 and 26,
solution and compare the two results.
apply the Gauss-Seidel method to the system of linear
15. 2.53x + 8.5y = 29.65 16. 12.5x − 18.2y = 56.8 equations in the stated exercise.

2.33x + 16.1y = 43.85 3.2x − 15.1y = 4.1
25. Exercise 23 26. Exercise 24
Gaussian Elimination with Partial Pivoting In
Strictly Diagonally Dominant Matrices In Exercises
Exercises 17–20, (a) use Gaussian elimination without
27–30, determine whether the matrix is strictly
partial pivoting to solve the system, rounding to three
diagonally dominant.
significant digits after each intermediate calculation, (b)
use Gaussian elimination with partial pivoting to solve the −2
same system, again rounding to three significant digits
27. [40
2
−3
] 28. [−11 −3]
after each intermediate calculation, and (c) compare
[ ] [ ]
4 0 2 4 2 −1
both solutions with the exact solution provided. 29. 10 12 −2 30. 0 −2 −1
17. 2.15x + 7.25y = 13.7 18. 4.25x + 6.3y = 16.85 1 −2 0 1 1 −1

3.12x + 6.3y = 15.66 6.32x + 2.14y = 10.6
(Exact: x = 3, y = 1) (Exact: x = 1, y = 2) Interchanging Rows to Obtain Convergence In
19.2.54x + 4.98y + 5.77z = 24.73 Exercises 31–34, interchange the rows of the system
of linear equations to obtain a system with a strictly
1.67x − 6.03y − 12.15z = −50.18

diagonally dominant coefficient matrix. Then apply the
−8.61x − 3.86y + 15.38z = 47.03
Gauss-Seidel method to approximate the solution to four
(Exact: x = −1, y = 2, z = 3)
significant digits.
20. −4.11x + 2.35y − 7.80z = −30.12 31. x1 + 2x2 = −5 32. x1 + 4x2 = −4
5.44x − 4.59y + 6.01z = 44.43
5x1 − x2 = 8 2x1 + x2 = 6
19.28x + 8.56y − 13.47z = −26.27
33. 2x1 + 4x2 + x3 = −2 34. x1 + 3x2 − x3 = 2
(Exact: x = 2, y = −6, z = 1)
4x1 + x2 + x3 = 1 x1 + x2 + 3x3 = −1
x1 − x2 − 4x3 = 2 3x1 + x2 + x3 = 1
9781305658004_100R.indd 527 12/15/15 11:34 AM

Finding a Dominant Eigenvector In Exercises 35–38, 53. Hospital Care The table shows the amounts spent
use the techniques presented in Chapter 7 to find the for hospital care (in billions of dollars) in the United
eigenvalues of the matrix A. When A has a dominant States from 2009 through 2013. (Source: Centers for
eigenvalue, find a corresponding dominant eigenvector. Medicare and Medicaid Services)
35. A = [11 1
1
] 36. A = [20 ]
1
4

Year 2009 2010 2011 2012 2013
[ ] [ ]
−2 2 −3 1 2 −3 Amount
776.8 814.9 849.9 898.5 936.9
37. A = 2 1 −6 38. A = 0 5 1 Spent
−1 −2 0 0 0 4
Find the second-degree least squares regression
Using the Rayleigh Quotient In Exercises 39–44, use polynomial for the data. Let x = 9 correspond to 2009.
the Rayleigh quotient to find the eigenvalue λ of the Then use a software program or a graphing utility to find
matrix A corresponding to the eigenvector x. a second-degree least squares regression polynomial.
39. A = [21 −12
−5]
,x=[ ]
3
1
Compare the results.
54. Dormitory Costs The table shows the average costs
6 −3
40. A = [
1]
,x=[ ]
3 (in dollars) of a college dormitory room from 2009
−2 −1 through 2013. (Source: Digest of Education Statistics)
[ ] []
2 0 1 0
41. A = 0 3 4 ,x= 1 Year 2009 2010 2011 2012 2013
0 0 1 0 Cost 4446 4657 4874 5095 5296
[ ] []
1 2 −2 1
42. A = −2 5 −2 , x = 1 Find the third-degree least squares regression polynomial
−6 6 −3 3 for the data. Let x = 9 correspond to 2009. Then use
a software program or a graphing utility to find a
[ ] [ ]
0 −1 1 −1
third-degree regression polynomial. Compare the results.
43. A = 2 4 2 ,x= 5
1 1 0 1 55. Probability A researcher performs the experiment in
Example 3 of Section 10.4 with the maze shown in the
[ ] []
3 2 −3 3 figure. Find the probability that the mouse emerges in
44. A = −3 −4 9 ,x= 0 the food corridor when its begins at the ith intersection.
−1 −2 5 1

Eigenvectors and Eigenvalues In Exercises 45–48,
use the power method with scaling to approximate a
1 2 3 4
dominant eigenvector of the matrix A. Start with
x0 = [1 1]T and complete four iterations. Then use x4 to
approximate the dominant eigenvalue of A. 5 6 7 8
−3 10
45. A =
7
2 [ 2
4
] 46. A = [
5 2 ] Food
47. A =
2
0 [ 1
−4
] 48. A =
6
[
3 −3
0
] 56. Probability Rework Exercise 55 assuming that the
upper corridor is also a food corridor.
Least Squares Regression Analysis In Exercises
49–52, find the second-degree least squares regression Stable Age Distribution In Exercises 57–60, the matrix
polynomial for the data. Then use a software program or represents the age transition matrix for a population.
a graphing utility to find a second-degree least squares Use the power method with scaling to find a stable age
regression polynomial. Compare the results. distribution vector.
[ ] [ ]
49. (−2, 0), (−1, 2), (0, 3), (1, 2), (3, 0) 1 2 1 3
57. A = 1 58. A = 1
50. (−2, 2), (−1, 1), (0, −1), (1, −1), (3, 0) 2 0 3 0
51. (0, −5), (3, −4), (5, −1), (7, 3), (11, 9)
[ ]
0 2 4
[ ]
52. (0, 12), (2, 11), (4, 7), (6, 4), (8, −3) 1 5 1
59. A = 1 60. A = 4 0 0
4 0 3
0 4 0
9781305658004_100R.indd 528 12/15/15 11:34 AM

Chapter 10 Projects 529
10 Projects
1 The Successive Over-Relaxation (SOR) Method
In Section 10.2, you studied two iterative methods for solving linear systems, the
Jacobi method and the Gauss-Seidel method. A third method, known as the successive
over-relaxation (SOR) method, uses extrapolations of the Gauss-Seidel method. These
extrapolations take on the form of a weighted average of the preceding iteration X(k )
i
(k)
and the corresponding Gauss-Seidel iteration xi ,
Xi(k+1) = ωxi(k+1) + (1 − ω )Xi(k)

where the weight ω ∈ (0, 2) is chosen to maximize the rate of conergence and is
generally not known in advance. Note that when ω = 1, this method is equivalent to
the Gauss-Seidel method.
1. Use the Internet or some other reference source to write an algorithm for the SOR
method that is specific to your software program or graphing utility.
2. Use your algorithm and software program or graphing utility to rework Example 2
of Section 10.2 and Example 3 of Section 10.4 using the SOR method. Note the
values of ω you found and how many iterations you “saved” to solve each problem
using the SOR method in comparison to the Gauss-Seidel method.
2 United States Population

The table shows the populations (in millions) of the United States by decade from 1900
to 2010. (Source: U.S. Census Bureau)
Year 1900 1910 1920 1930 1940 1950

Population 76.1 92.4 106.5 123.1 132.1 152.3
Year 1960 1970 1980 1990 2000 2010

Population 180.7 205.1 227.2 249.5 282.2 309.3
1. Use a software program or a graphing utility to create a scatter plot of the data. Let
x = 0 correspond to 1900, x = 1 correspond to 1910, and so on.
2. Using the scatter plot, describe any patterns in the data. Do the data appear to be
linear, quadratic, or cubic in nature? Explain.
3. Use the techniques presented in this chapter to find (a) a linear least squares
regression equation, (b) a second-degree least squares regression equation, and
(c) a third-degree least squares regression equation to fit the data.
4. Graph each equation with the data. Briefly describe which of the regression
Year Population equations best fits the data.
5. Use each model to predict the populations of the United States for the years 2020,
2020 334.5 2030, 2040, and 2050. Which of the regression equations appears to be the best
2030 359.4 model for predicting future populations? Explain.
2040 380.2 6. The U.S. Census Bureau projects the populations of the United States for the years
2020, 2030, 2040, and 2050 as shown in the table at the left. Do any of your models
2050 398.3 produce the same projections? Explain any possible differences between your
projections and the U.S. Census Bureau projections.
East/Shutterstock.com
9781305658004_100R.indd 529 12/15/15 11:34 AM

10 Numerical Methods PDF

Cargado por

Información del documento

Descripción original:

Título original

Derechos de autor

Formatos disponibles

Compartir este documento

Compartir o incrustar documentos

Opciones para compartir

¿Le pareció útil este documento?

¿Este contenido es inapropiado?

Copyright:

Formatos disponibles

10 Numerical Methods PDF

Cargado por

Copyright:

Formatos disponibles

10 Numerical

Car Crash Simulations (p. 521)

Google’s PageRank Algorithm (p. 515)

Illustration Software (p. 508)

Pharmacokinetics (p. 501)

9781305658004_1001.indd 495 12/15/15 11:27 AM

10.1 Gaussian Elimination with Partial Pivoting

Stored Value and Rounding Error

9781305658004_1001.indd 496 12/15/15 11:27 AM

Finding Stored Values

Propagation of Rounding Error

Evaluate the determinant of the matrix

∣A∣ = (0.12)(0.12) − (0.12)(0.23)

In Example 2, rounding at the intermediate steps introduced a rounding error of

Although this error may seem slight, it represents a percentage error of

9781305658004_1001.indd 497 12/15/15 11:27 AM

Gaussian Elimination with Partial Pivoting

Gaussian Elimination and Rounding Error

Use Gaussian elimination to solve the system below.

9781305658004_1001.indd 498 12/15/15 11:27 AM

In Example 3, rounding after each intermediate calculation resulted in percentage

Gaussian Elimination with Partial Pivoting

9781305658004_1001.indd 499 12/15/15 11:27 AM

Gaussian Elimination with Partial Pivoting

See LarsonLinearAlgebra.com for an interactive version of this type of example.

9781305658004_1001.indd 500 12/15/15 11:27 AM

An ill-Conditioned System of Linear Equations

Use Gaussian elimination to solve the system of linear equations.

Pharmacokinetics is the study of the movement of a drug

9781305658004_1001.indd 501 12/15/15 11:27 AM

Gaussian Elimination and Rounding Error In 28. C

Gaussian Elimination with Partial Pivoting In

9781305658004_1001.indd 502 12/15/15 11:27 AM

10.2 Iterative Methods for Solving Linear Systems

The Jacobi Method

9781305658004_1002.indd 503 12/15/15 11:29 AM

Applying the Jacobi Method

See LarsonLinearAlgebra.com for an interactive version of this type of example.

as a convenient initial approximation. By substituting these values on the right-hand sides

9781305658004_1002.indd 504 12/15/15 11:29 AM

The Gauss-Seidel Method

Applying the Gauss-Seidel Method

9781305658004_1002.indd 505 12/15/15 11:29 AM

Convergence and Divergence

Apply the Jacobi method to the system

The solution of this system of linear equations is x1 = 1 and x2 = 1. (Check this.)

You cannot resolve the problem of divergence in Example 3 by using the

9781305658004_1002.indd 506 12/15/15 11:29 AM

Definition of a Strictly Diagonally Dominant Matrix

∣ 11∣ ∣ 12∣ ∣ 13∣

Strictly Diagonally Dominant Matrices

THEOREM 10.1 Convergence of the Jacobi

9781305658004_1002.indd 507 12/15/15 11:29 AM

Interchanging Rows to Attain Convergence

The Jacobi and Gauss-Seidel methods can be applied to

c1 = 0.25(80 + 80 + c3 + 60) 100

As the mesh spacing decreases, the number

9781305658004_1002.indd 508 12/15/15 11:29 AM

9781305658004_1002.indd 509 12/15/15 11:29 AM

10.3 Power Method for Approximating Eigenvalues

Dominant Eigenvalues and Dominant Eigenvectors

Definitions of Dominant Eigenvalue and Dominant Eigenvector

THEOREM 10.1 Convergence of the Jacobi

x=t [31], t ≠ 0.

Approximating a Dominant Eigenvector

THEOREM 10.2 Determining an Eigenvalue

17. Trigonometry Find the second-degree least squares 24.

Then use the result to approximate cos(π4). Compare 4 5 6