Instructor Solution Manual

Mathematical Modeling
Instructor’s Solutions Manual

Mathematical Modeling

Stefan Heinz
University of Wyoming

© Springer-Verlag Berlin Heidelberg 2011

 Preface
In this manual there are solutions to all exercises of Mathematical Modeling,
First Edition, by Stefan Heinz. The notation "Exercise 4.2.4" refers to the fourth
problem related to Sect. 4.2. There are 570 exercise questions organized in 220
problems. In many cases, it is possible to create additional homework by minor
modifications of the suggested problems. It is hoped that these solutions will be
useful for instructors.

Stefan Heinz
University of Wyoming
[email protected]

Contents
1 Deterministic Analysis of Observations 1
2 Stochastic Analysis of Observations 11
3 Deterministic States 19
4 Stochastic States 29
5 Deterministic Changes 41
6 Stochastic Changes 49
7 Deterministic Evolution 61
8 Stochastic Evolution 73
9 Deterministic Multivariate Evolution 87
10 Stochastic Multivariate Evolution 103
Chapter 1
Section 1.2
Exercise 1.2.1
a) See figures (a) and (b).
b) See figure (c). The model is given by y = 100 / x2.
c) The absolute value of the relative error is smaller
than 8%: see figure (d).
d) We have y(2.5) = 16. Therefore, the value of y at
x = 2.5 is reduced by 25% at x = 2.8868.

Exercise 1.2.2
a) See figures (a) and (b).
b) See figure (c). The model is given by y = 2 x4.
c) The absolute value of the relative error is smaller
than 9.4%: see figure (d).
d) We have y(2.5) = 78.125. Therefore, the value of
y at x = 2.5 is increased by 150% at x = 3.1436.
2 Chapter 1

Exercise 1.2.3
a) See figures (a) and (b).
b) See figure (c). The model reads y = 760 e0.5 x.
c) The absolute value of the relative error is smaller
than 1%: see figure (d).
d) At x = 9.2103 we have y = y(x = 0) / 100.

Exercise 1.2.4
a) See figure (a). The first data point (0, 100) is not considered.
b) See figure (b). The model is given by y = 100 + x3.45.
c) The absolute value of the relative error is smaller than 3%: see figure (c).
d) The value of y(x = 0) is doubled at x = 3.8.

Exercise 1.2.5
a) See figure (a). The first data point (0, 30) is not considered.
b) See figure (b). The model is given by y = 30 + 50 x e0.9x.
c) The absolute value of the relative error is smaller than 3.5%: see figure (c).
d) The maximum y = 9.5623 is found at x = 1 / 0.9.
 Section 1.3 3

Exercise 1.2.6
The parameters (x0, y0) in y = y0 + a (exp[b(xx0)]  1) are provided by any initial
data. The parameter a can be specified by using y = y0 + a (exp[b(xx0)]  1) for
another point (x1, y1) z (x0, y0), i.e.
y1  y0
a .
eb ( x1  x0 )  1
A condition for b can be obtained by considering y at a third point (y2, x2),
y y
y2 y0  1
b ( x1  x 0 )
0
eb ( x 2  x 0 )  1 .
e 1
To make use of this condition for the calculation of b it is convenient to introduce
the function F(b) by
eb ( x 2  x0 )  1
F (b) y2  y0  ( y1  y0 ) .
eb ( x1  x0 )  1
The numerical calculation of the zero of F(b) then provides the parameter b.

Section 1.3
Exercise 1.3.1
a) See figures (a) and (b).
b) See figure (c). The quadratic function defined by the data at v = 20, 50, and 80.
The quadratic function is given by D / v = (0.98 v2 + 13 v + 3,128) / 1,800. The
linear model is given by D / v = 1 + v / 20.
c) See figure (d): the solid line refers to the quadratic model, and the dashed line
refers to the linear model for D / v. The absolute values of the relative errors
related to the quadratic and linear models for D / v are smaller than 0.3% and
16%, respectively: see figure (e).
d) The most appropriate model is P = (1 + v / 20) v because of its simplicity and
reasonable accuracy regarding the question considered. Examples for the use of
this model are given by P(20) = 40, P(40) = 120, P(80) = 400.
4 Chapter 1

Exercise 1.3.2
a) See figure (a). This model is inappropriate because it predicts a nonzero D(0).
b) The sixth-order polynomial model can be significantly improved by adding the
point (0, 0) and removing, for example, the point (50, 173). The performance
of the resulting model is illustrated in figure (b).
c) A reasonable quadratic model can be found by defining the second-order poly-
nomial via the points (0, 0), (40, 116), and (80, 646): see figure (c).
d) The quadratic polynomial obtained in this way is given by D = 0.0725 v2. This
expression can be simplified to D = 0.07 v2.
 Section 1.3 5

Exercise 1.3.3
a) See figure (a). This model does not agree with the data trend for the interval
considered.
b) See figure (b). This model does not allow reasonable predictions.
c) See figure (c). This model may be expected to provide reasonable predictions.
The comparison with the projected values at 2025 and 2050 shows a relatively
small disagreement.
d) See figure (d). This model may be expected to provide reasonable predictions.
The comparison with the projected values at 2025 and 2050 shows a relatively
good agreement. Because of its simplicity, the linear model is the best model
among the models considered here.

Exercise 1.3.4
a) See figure (a). This model does not allow reasonable predictions.
b), c), and d) See figures (b), (c), and (d). The performance of these models is the
same (P3 and P2 do hardly deviate from a linear function). Due to its simplicity,
the linear function represents the best model among the models considered. A
disadvantage of the linear model is given by its way of construction. The model
passes exactly two data points. Thus, the deviations between the model and the
data are not minimized.
6 Chapter 1

Section 1.4
Exercise 1.4.1
a) The linear relation between redefined variables reads
§a ·
ln¨  1¸ ln b  c t .
©P ¹
b) The data support the choice a = 120 as a reasonable estimate for t o f. Figure
(a) demonstrates the suitability of the linearity assumption by comparison with
the function
§ 120 ·
ln¨  1¸ 0.5 (t  2) .
© P ¹
c) The model is given by
120
P .
1  e  0.5 (t  2 )
The graphs are shown in figures (b) and (c). The absolute value of the relative
error is smaller than 5.1%.
d) The initial population is given by P(0) = 120 / (1 + e). By using the linearized
relation between redefined variables we have the condition
§1 e ·
ln¨  1¸ 1  0.5 t .
© 2 ¹
 Section 1.4 7

Hence, the time at which the initial population at t = 0 is increased by a factor

of 2 is given by t = 2 ln[(e  1) / (2 e)] = 2.3036 days.

Exercise 1.4.2
a) The linear relation between redefined variables reads
§a ·
ln¨  1¸ ln b  c t .
©P ¹
b) The data suggest a = 100 as a reasonable estimate for t o f. Figure (a) shows
the suitability of the linearity assumption by comparison with the function
§ 100 ·
ln¨  1¸ 0.8 (t  4) .
© P ¹
c) The model is given by
100
P .
1  e 0.8 (t  4)
The graphs are shown in figures (b) and (c). The absolute value of the relative
error is smaller than 3.8%.
d) In terms of the linear relation between redefined variables we find the condition
ln(100  1) = 3.2 + 0.8 t. Hence, we find P = 1 at t = 9.7439 days.
8 Chapter 1

Exercise 1.4.3
a) The given observations lead to the conclusion that a = 10 and d = 1. The linear
relation between redefined variables then reads
§ 10 ·
ln¨  1¸ ln b  c t .
© P 1 ¹
b) Figure (a) shows the suitability of the linearity assumption by comparison with
the function (the data point P(1804) = 1 is disregarded)
§ 10 · t  1999
ln¨  1¸  .
© P 1 ¹ 30.65
c) The model is given by
10
P 1 .
1  e  (t 1999 ) / 30.65
The graphs are shown in figures (b) and (c). The absolute value of the relative
error is smaller than 6.9%.
d) The derivatives of P are given by
dP 10 e  (t 1999 ) / 30.65 ( P  1) 2 § 10 · P 1 § P 1 ·
¨  1¸ ¨1  ¸,
dt 30.65 1  e  (t 1999 ) / 30.65 2
306.5 © P  1 ¹ 30.65 © 10 ¹
d 2P 1 § P  1 · dP
¨1  ¸ ,
dt 2 30.65 © 5 ¹ dt
2
d 3P 1 1 § dP · 1 § P 1 · d 2P
 ¨ ¸  ¨1  ¸ .
dt 3 30.65 5 © dt ¹ 30.65 © 5 ¹ dt 2
The first expression for dP / dt shows that dP / dt > 0 except for t o rf. A
critical point of dP / dt is then given at 1 = (P  1) / 5, for which d 2P / dt 2 = 0. In
particular, we have a local maximum because d 3P / dt 3 < 0 there. By using the
definition of P and 1 = (P  1) / 5 we see that dP / dt has a maximum at t = 1999,
t  1999 § 10 ·
 ln¨  1¸  ln 2  1 0.
30.65 © P  1 ¹
 Section 1.5 9

Section 1.5
Exercise 1.5.1
a) See the figure.
b) This function is given by
CO2  334
T r 0.1.
110
c) The advantages of this linear model are its sim-
plicity and its relatively low error: the variation of
model parameters is only 3.3%. The disadvantage
of this model is the empirical model development
without making an attempt to minimize the devia-
tions between the model and data.
d) The 10% CO2 concentration increase will result in
1.1 ˜ 384  334
T 0.8 .
110
Thus, the temperature increase compared to T = 0.4ºC is T = 0.4ºC.

Exercise 1.5.2
a) See the figures.
b) These functions are given by
t  2007
CO2 383.7  2 (t  2007), T  0.465 r 0.065 ,
56
c) This relation is given by
CO2  383.7 CO2  331.6
T  0.465 r 0.065 r 0.065 .
112 112
d) The 10% CO2 concentration increase will result in
1.1 ˜ 384  331.6
T 0.81.
112
Thus, the temperature increase compared to T = 0.4ºC is T = 0.41ºC.
10 Chapter 1

Exercise 1.5.3
a) See the figures.
b) These functions are given by
t  2007
CO2 383.7  2 (t  2007), T  0.483 r 0.036 ,
54
c) This relation is given by
CO2  383.7 CO2  331.5
T  0.483 r 0.065 r 0.065.
108 108
d) The 10% CO2 concentration increase will result in a temperature T = 0.84ºC.
Thus, the temperature increase compared to T = 0.4ºC is T = 0.44ºC.
Chapter 2
Section 2.2
Exercise 2.2.1

a) Least-squares error Chebyshev error Least-absolute-

deviations error
Data set (d) a = 0.9912 a = 0.9967 a = 0.9500
Data set (f) a = 1.0157 a = 1.0667 a = 1.0000

b) The optimal value is a = 1. First, all three concepts provide a reasonable result.
Nevertheless, the least-absolute-deviations error and Chebyshev error concepts
do not perform as well as the least-squares error concept regarding the (d) and
(f) data, respectively. Overall, the least-squares error concept is the best choice.

Exercise 2.2.2
a) Two other reasonable error concepts are given by
1 N 1 N
¦ Yi  yM ( X i ) , ¦ Yi  yM ( X i ) .
4
E1 E2
Ni1 Ni1

b) Least-squares error E1 error E2 error

a = 1.0157 a = 1.0000 a = 1.0474

c) The least-squares error concept is better than the E2 error concept. The E1 error
provides a slightly better result than the least-squares error, but the related
calculation of optimal model parameters is not comparably simple. Thus, the E1
error concept does not provide an alternative to the least-squares error concept.

Exercise 2.2.3
a) The curve passes the point (1950, 34.6) if b = 34.6  1950 a.

b) Least-squares error Chebyshev error Least-absolute-deviations error

a = 1.2796 a = 1.2572 a = 1.3220

c) The figure illustrates the performance of the least-

squares error (solid line), Chebyshev error (long
dashes), and least-absolute-deviations error (short
dashes). All the three error concepts perform well,
but the least-squares error concept works slightly
better than the other concepts: this error concept
provides a curve between the other two curves.
12 Chapter 2

Section 2.3
Exercise 2.3.1
a) The optimal b value, which minimizes the error E2, is the value that minimizes
the non-negative last term in the E2 formula, i.e., b = <Y>  a <X>.
b) We write the E2 formula for this b value as
§ ~ ~ ·2 ~~ 2
~ ~ ¨ XY ¸ X Y
E2 Y 2  X 2 ¨ a  ~ 2 ¸  ~ 2 .
¨ X ¸ X
© ¹
The optimal a value, for which the non-negative second term on the right-hand
side is equal to zero, is given by
~~ ~
a XY / X 2 .

Exercise 2.3.2
a) By introducing x = t  1959 we can write the function considered as

§ x · 0.7 2
CO2 316  0.7¨1  ¸ x 316  0.7 x  x .
© 47 ¹ 47
Next, we calculate the required integrals:
1 51ª 0.7 2 º
CO2 ³ «316  0.7 x  x dx
I
20 31¬ 47 »¼
1 ª 1640 0.7 102,860 º 261,014
«316 ˜ 20  0.7  ,
20 ¬ 2 47 3 »¼ 705

1 51ª 0.7 2 º
CO2 t I ³ 316  0.7 x  47 x » ( x  1959) dx
20 31«¬ ¼
1 51ª 0.7 3 º
³ «316 x  0.7 x 2  x dx  1959 CO2
20 31¬ 47 »¼ I

1 ª 1640 102,860 0.7 5,841,680 º

316  0.7  »  1959 CO2
20 «¬ 2 3 47 4 ¼
I

3010
2000 CO2 I  .
47
The model parameters a and b are given by (we have c = 1990 and d = 2010)
12 ª d  c º 12 3010 3 301 903
a 2 «
CO2 t I  CO2 I ,
(d  c) ¬ 2 »¼ 400 47 10 47 470
d c
b CO2 I  a CO2 I  2000 a.
2
 Section 2.3 13

Hence, the linear approximation is given by

( LIN )
CO2 at  b a t  CO2 I
 2000 a CO2 I
 a (t  2000),
where <CO2>I = 370.23 and a = 1.92.
b) The quadratic model (solid line) and optimal
linear CO2 model (dashed line) are shown in
the figure. It may be seen that the linear model
represents a very good approximation.

Exercise 2.3.3
a)-d) Regarding the questions a), b), c), and d) we find rXY = 0.9537, rXY = 0.9334,
rXY = 0.9990, and rXY = 0.9990 for the correlation coefficients, respectively.
e) The results reveal that the correlation coefficients are actually unaffected by
randomness. The lower correlation coefficients for a) and b) are a consequence
of the nonlinear dependence of P on t.

Exercise 2.3.4
a) We can write this function as ln yM = ln a + b ln x.
The optimal model parameters are b = 0.6994 and
ln a = 0.9162. Consequently, the optimal model
is given by yM = 0.4 x0.7.
b) See the figure.

Exercise 2.3.5
a) This function can be written ln (P  P0) = ln b + a
ln t, where P0 = 100. The optimal model param-
eters are a = 3.4493 and ln b = 0.0652. Here, the
first data point is disregarded to avoid the evalua-
tion of ln 0. Therefore, the optimal model is given
by P = 100 + 0.937 t 3.45.
b) See the figure.
14 Chapter 2

Exercise 2.3.6
a) The optimal model is given by
C 1.2318 t  2365.6600
91.7810  1.2318 (t  1995).
b) See the figure. The Table 1.1 data are presented by
the dots. The solid line shows the optimal linear
model. The dashed line shows the model obtained
in Sect. 1.2.1, C = 91 + 1.25 (t  1995).
c) The performance of these two models is the same.
The advantage of the optimal model is that there is no adjustment needed
between the model and the data.

Exercise 2.3.7
a) The rewritten model reads ln P = d t + ln c.
b) The optimal model parameters are given by the
values c = e55.9285 and d = 0.0282. Consequently,
the optimal model is given by

P e0.0282 t  55.9285 0.0043 e( t 1790 ) / 35.46 .

c) See the figure. The Table 1.5 data are shown as
dots. The solid line shows the optimal exponen-
tial model. The dashed line presents the model
P = 0.0039 exp[(t  1790) / 35].
d) There is hardly any observable difference between these models. Therefore, the
optimal model confirms the suitability of the model obtained in Sect. 1.4.1.

Exercise 2.3.8
a) The population model can be represented by the
relation ln(A / P  1) = C (t  B).
b) The optimal model parameters are given by the
values B = 1889.7899 and C = 1/28.8725. The
optimal model is given by
0.12
P .
1  e  (t 1889.7899 ) / 28.8725
c) See the figure. The dots show the Table 1.5 data.
The optimal model and the model developed in
Sect. 1.4.1 are shown by solid and dashed lines, respectively.
d) There is no observable difference between the two models. The advantage of
the optimal model is that it provides without any adjustment a model that is as
good as a model obtained by empirically adjusting the model parameters.
 Section 2.4 15

Exercise 2.3.9
a) The model T = 0.2800 + [(t  1840) / 184.3142]4
represents the optimal model.
b) See the figure.

Section 2.4
Exercise 2.4.1
a) The quadratic model considered is given by the function
ª 2 º
yM
~
Y2
1/ 2
«A x  X B
x X
 C ».
« ~ ~ 2 1/ 2 »
X2 X
«¬ »¼
Here, the optimal model parameters are given by the expressions
rXXY  m3 rXY Y
A , B rXY  m3 A, C  A.
m4  1  m3
2 ~ 2 1/ 2
Y
This model reduces to a linear model if A = 0. This condition is satisfied for a
data set with the properties rXXY = 0 and m3 = 0. There should be no constraint
on rXY, which is an essential ingredient of an optimal linear model.
b) The optimal quadratic model reads for rXXY = 0 and m3 = 0
ª º
yM
~
Y2
1/ 2
«r x  X  Y » ~
Y  Y2
1/ 2
rXY
x X
.
« XY ~ 2 1 / 2 ~ 1/ 2 » ~ 1/ 2
«¬ X Y2 »¼ X2

By accounting for the definition of rXY, this model is equal to the optimal linear
model (2.47) derived in section 2.3.2.

Exercise 2.4.2
b) The optimal model is given by yM = 1.425 x2 + 3.195 x + 3.945.

Exercise 2.4.3
a) The optimal model is given by D = 0.0888 v2  1.9631 v + 49.1071.
b) This model has two shortcomings. First, it predicts a non-zero stopping distance
for v = 0. Second, it does not provide an appropriate basis for a reduction of this
model to a linear model.
16 Chapter 2

Section 2.5
Exercise 2.5.1
a) This model is linear in xb. We can apply the results obtained in Sect. 2.3.2 by
replacing x by xb and b by y0. Thus, the optimal parameters are given by
~ ~
X bY
a ~ , y0 Y  a X b .
X 2b

b) The optimal model is given by yM = 0.0071 + 3.4837 x2.

Exercise 2.5.2
a) This model is linear in ln(x / x0). We can use the results obtained in Sect. 2.3.2
by replacing x by ln(x / x0) and b by y0. We apply the abbreviation U = ln(X / x0).
The optimal parameters are given by
~~
UY
a ~ , y0 Y  a U .
U2

b) The optimal model is given by yM =  0.0004 + 4.4686 ln x.

Exercise 2.5.3
a) We follow the approach presented in Sect. 2.5. We introduce the abbreviation
U = X b eX. According to the formulas presented in Sect. 2.5.1 we find the
optimal parameters y0 and a to be given by
~~ ~
y0 Y  a U , a UY / U2 .
The error E2 is given for the optimal parameters y0 and a by
~ 2
E 2 Y 2 (1  rUY ).
b) The optimal parameter b can be found as the b value that minimizes E2. Then,
we can use the formulas for y0 and a to calculate the optimal a and y0 values.

Exercise 2.5.4
a) Yes. Let us say we have two sets of parameters for two choices of x0 given by
x0 and x0 + ',
yM y0  a (eb ( x  x0 )  1) y0  a  a e b ( x  x 0 ) ,
y 'M y '0  a' (eb '( x  x0  ' )  1) y '0 a ' a' e  b ' ' eb '( x  x0 ) .
Both functions are equal if the following conditions apply:
b' b, a' e b '' a, y '0  a' y0  a .
b) No, this is impossible.
 Section 2.5 17

Exercise 2.5.5
b) The function yM = 1.9984 + 0.3882 (exp[0.1317 x ]  1) represents the optimal
exponential model.

Exercise 2.5.6
b) The optimal power model is given by yM = 0.9787 + 8.0212 x2.9678.

Exercise 2.5.7
a) The optimal power model is given by yM = 0.8381 + 2.3832 (x  0.5)1.7760.
b) The optimal power model is given by yM = 0.5136 + 1.7633 (x  0.8)0.9763.
c) The maximum relative errors for the models using x0 = 0, x0 = 0.5, and x0 = 0.8
are |e| = 1.1%, |e| = 0.5%, and |e| = 2.3%, respectively. Thus, the model that
applies x0 = 0.5 is the best model among the three models considered.

Exercise 2.5.8
a) A better model can be obtained by varying x0 and choosing the model with the
smallest maximum relative error. Another possibility is to use the least absolute
deviations criterion to find an optimal x0.
b) A better model is given if we use x0 = 0.3772: the maximum relative error of
this model is |e| = 0.0007%.
Chapter 3
Section 3.2
Exercise 3.2.1
a) The dimension of FL is M L / T2. The relative aircraft velocity v (dimension L /
T) and total surface area A (dimension L2) of the tops of the wings cannot bring
in the mass dependence of FL. The latter can be achieved by involving the air
density U (dimension M L3).
b) This formula is given by
M L2 2
FL ~ L.
L3 T 2
By introducing the nondimensional coefficient CL we can write this formula as
CL
FL U v 2 A.
2

Exercise 3.2.2
a) Due to dimensional reason we have
s a  b t  c t 2 s0  v0 t  a0 t 2 / 2,
which means that a = s0 (dimension L), b = v0 (dimension L / T), and c = a0 / 2
(dimension L / T2).
b) The constants s0, v0, and a0 represent the initial position, the initial velocity, and
the initial acceleration, respectively, at t = 0.

Exercise 3.2.3
a) The dimension of Fd is M L / T2. Thus, the dimension of the parameter a is M / T,
and the dimension of the parameter b is M / L.
b) The parameter a = m /W can be written in terms of the mass m of the moving
body and the characteristic time scale W for the damping of the body motion.
The parameter b = m / " can be written in terms of the body mass m and the
characteristic length scale " for the damping of the body motion.

Exercise 3.2.4
a) The dependence of W on r and Q is given by
r2
W cW .
Q
Here, cW is a nondimensional coefficient.
b) This relation is dimensionally correct but physically wrong: the damping time
has to become smaller for an increasing particle radius: see Eq. (3.78).
20 Chapter 3

Exercise 3.2.5
a) The ratios v / g and v 2 / g have the dimensions T and L, respectively.
b) The best choice for the problem considered is to involve the raindrop radius r.
c) This nondimensional product is given by
v cv r g .
Here, cv is a nondimensional coefficient.

Exercise 3.2.6
a) The dimension of J is M / T. The dimension of k is M / T2.
b) The nondimensional time t* is given by t* = t (k / m)1/2.
c) This equation is given by
d2y J dy
2
 y 0.
dt* k m dt*

Exercise 3.2.7
a) The function f(P2, P3, P4) is unknown. Thus, the variable of interest given in the
nondimensional product P1 cannot be calculated.
b) The function f(P2, P3, P4) has to be calculated in dependence on its arguments.
c) The design conditions for a completely similar model system are given by
M R M R M R
P2 P2 , P3 P3 , P4 P4 .
d) A relatively similar model system can be designed by making sure that the most
relevant nondimensional products are equal in the real and the model system.

Exercise 3.2.8
a) The exponents related to v, d2, and U are used as dependent variables. All the
other exponents are independent variables.
b) The variables v and d2 can be used for finding a nondimensional product that
involves g. This nondimensional product is given by g d2 / v2.
c) The result of nondimensional analysis is FL = CL U v2 d22 / 2, where
§ d v P gd ·
CL CL ¨¨T , 1 , , , 2 2 ¸¸.
© d2 s U d2 v v ¹

Exercise 3.2.9
a) The conditions for a completely similar model system are given by
d1
M
d1
R
vM vR QM QR d2
M
d2
R
TM T R, M R
, , , .
d2 d2 sM sR M
d2 vM
R
d2 vR (v M )2 (v R )2
Here, the kinematic viscosity Q = P / U is used to simplify the writing.
 Section 3.3 21

b) For d2M = D d2R the design conditions read

TM d1
M
sM QM vM
1, D, D, D 3/ 2, D.
TR d1
R
sR QR vR
c) For the case that the same fluid is used in both systems we have the conditions
TM d1
M
vM
1, D, 1 D, 1 D 3/ 2, D.
TR d1
R
vR
Hence, the model system is equal to the real system because D = 1.

Exercise 3.2.10
a) We can calculate the coefficient cP = TP / [2S (r / g)1/2] by measuring the non-
dimensional product TP / [2S (r / g)1/2] as a function of D0 and P*.
b) There are two design conditions given by
M R
PM r3 PR r3
D0 M
D0 , M
.
mM g
M
mR g
R

The gravity acceleration g is constant, and we have the condition rM = rR / 2.

Thus, the design conditions read
PM PR
D0M D0M , 8 .
mM mR
c) The design conditions for the same fluid are given by
mR
D0M D0M , mM .
8

Section 3.3
Exercise 3.3.1
a) The use of dimensional analysis leads to
P a d b mc v d c1 ,
a d
§M L 2
· b c§L·
¨ 3 ¸ L M ¨ ¸ L2 a  b  d M a  c T  3a  d c2 .
¨ T ¸ ©T ¹
© ¹
The exponent of M implies that c = a. The exponent of T implies d = 3 a.
The exponent of L implies that b = 2 a  d = a. Hence, we find
a
§ Pd ·
P a d a m  a v  3a ¨¨ ¸
3 ¸
c1.
© mv ¹
Consequently, P = cP m v3 / d, where cP is a dimensionless constant.
22 Chapter 3

b) In terms of the kinetic energy K = m v2 / 2 and characteristic time scale W = d / v

we obtain P = 2 cP K / W.

Exercise 3.3.2
a) The use of dimensional analysis leads to
P a hb g c U d c1 ,
a c d
§ M · b§ L · § M ·
¨¨ ¸ L¨ 2¸ ¨ 3¸
2 ¸
L a  b  c  3d M a  d T  2 a  2 c c2 .
© LT ¹ © T ¹ © L ¹
The exponent of M implies that d = a. The exponent of T implies c = a. The
exponent of L implies that b = a  c + 3d = a. Hence, we find
a
§ P ·
P a h  a g a U a ¨¨ ¸¸ c1.
© Uhg ¹
Thus, P = cP U h g, where cP is a dimensionless constant.
b) In terms of the gravity force Fg = m g, P can be written as P = cP U h Fg / m. By
introducing the area A = m / (U h) we can write the pressure as P = cP Fg / A.

Exercise 3.3.3
a) The use of dimensional analysis leads to
b
§ dp ·
qa ¨ ¸ r c P d c1 ,
© dz ¹
a b d
§ L3 · § M · c§ M ·
¨ ¸ ¨¨ 2 2 ¸¸ L ¨¨ ¸¸ L3a  2b  c  d M b  d T  a  2b  d c2 .
¨T ¸
© ¹ ©L T ¹ © LT ¹
The exponent of M implies that b = d. The exponent of T implies d = a, where
b = d is used. Hence, we have b = a. The exponent of L implies that c = 3 a
+ 2 b + d = 4 a. Hence, we find
a a
§ dp · § qP ·
q a ¨ ¸ r 4 a P a ¨¨ 4 ¸¸ c1.
© dz ¹ © r dp / dz ¹
Thus, q = cq r4 (dp / dz) / P, where cq is a dimensionless constant.
b) The flow rate q becomes 16 times larger.

Exercise 3.3.4
a) The use of dimensional analysis leads to
Z a rb U c Gd c1 ,
c d
1 b § M · § L3 ·
L ¨ ¸ ¨ ¸ Lb  3c  3d M c  d T  a  2 d c2 .
T a © L3 ¹ ¨© M T 2 ¸¹
 Section 3.3 23

This constraint implies that c = d = a / 2 and b = 3 (c  d) = 0. Hence, we find

a
§ Z ·
Z r U
a 0 a / 2
G a / 2 ¨ ¸ c1.
¨ GU ¸
© ¹
Consequently, Z = cZ (G U)1/2, where cZ is a dimensionless constant.
b) The consequence of this result is that Z is independent of the star’s radius r.

Exercise 3.3.5
a) The dimension of K is M L T2 Ln = M L1+n T2 .
b) The use of dimensional analysis leads to
P a vb mc K d c1 ,
a b d
§ M · §L· § M L1 n ·
¨¨ ¸¸ ¨ ¸ M c ¨¨ 2
¸
¸ L a  b  (1 n ) d M a  c  d T  a  b  2 d c2 .
© LT ¹ ©T ¹ © T ¹
The sum of the L and T exponents implies that 2 a + (n  1) d = 0, this means
2
d a.
n 1
The use of this d expression in the M and T exponents leads to the conditions
2a a ( 2  n  1) n 1
c a  d a    a.
n 1 n 1 n 1
4a a (4  n  1) n3
b  a  2d a    a.
n 1 n 1 n 1
Hence, we obtain
a
§ P K 2 / ( n 1) ·
P v a  a ( n  3) / ( n 1)
m  a ( n 1) / ( n 1)
K 2 a / ( n 1)
¨ ( n  3) / ( n 1) ( n 1) / ( n 1) ¸ c1.
¨v m ¸
© ¹
Consequently, we find the relation
1 / ( n 1)
§ m n 1 v n  3 ·
P cP ¨¨ 2
¸
¸ ,
© K ¹
where cP is a dimensionless constant.
c) The use of v = (k T / m)1/2 leads to
1 / ( n 1) 2n  2 n 3 1 / ( n 1) n3
§ m n 1 k T / m ( n  3) / 2
· § k ( n  3) / 2 ·
P cP ¨ ¸ cP m 2 ( n 1)
¨
¨ K2 ¸
¸ T 2 ( n 1)
¨ K2 ¸
© ¹ © ¹
1 / ( n 1) n 1 4 1 / ( n 1)
§ k ( n  3) / 2 · 1/ 2 § k
( n  3) / 2
· 1 2

cP m1 / 2 ¨¨ 2
¸
¸ T 2 ( n 1)
c P m ¨
¨ K 2
¸
¸ T 2 n 1
.
© K ¹ © ¹
For 5 d n d f the temperature dependence of P varies between T and T1/2.
24 Chapter 3

Exercise 3.3.6
a) The use of dimensional analysis leads to
v a r b Fg P d
c
c1 ,
a c d
§ L · b§ M L · § M ·
¨ ¸ L¨ 2 ¸ ¨¨ ¸¸ La  b  c  d M c  d T  a  2 c  d c2 .
©T ¹ © T ¹ © LT ¹
The exponent of M implies d = c. By using this condition, the exponent of T
leads to c = a, which implies d = a. The exponent of L implies b =  a  c + d
= a. Hence, we find
a
§vr P ·
v r Fg
a a a
P a ¨ ¸ c1.
¨ Fg ¸
© ¹
Therefore, v = cv Fg / (r P), where cv is a dimensionless constant.
b) We consider 0 = Fg + Fd = P r v / cv 6 S P r v, where v = cv Fg / (r P) is used.
The balance of these terms requires cv = 1 / (6 S).
c) For the case that the effect of the fluid density can by neglected, Stokes’ Law
for the velocity of a small sphere falling under gravity in a viscous fluid reads
1 Fg 4S r 3 U s g 2 Us g r 2
v .
6S r P 18S r P 9 P

Section 3.4
Exercise 3.4.1
a) The use of dimensional analysis leads to
va r b g c U s P e
d
c1 ,
a c d e
§L· b§ L · §M · § M ·
¨ ¸ L ¨ 2 ¸ ¨ 3 ¸ ¨¨ ¸¸ La  b  c  3d  e M d  e T  a  2 c  e c2 .
©T ¹ © T ¹ © L ¹ © LT ¹
The exponents c and e are chosen as independent variables. The exponent of M
implies that d = e. The exponent of T implies that a = 2 c  e. The exponent
of L implies b =  a  c + 3 d + e = 2 c + e  c  3 e + e = c  e. Hence, we find
c e
e §rg· § P ·
v  2c  e r c  e g c U s P e ¨ 2 ¸ ¨ ¸ c1.
©v ¹ ¨ U vr ¸
© s ¹
Hence, the Reynolds number is given by Re = Us v r / P, and the Froude number
is given by Fr = v2 / (r g).
b) We have Fr / Re = v P / (g Us r2). Thus, v is given by v = c Us g r2 / P. Here, c is a
constant, which is equal to 2 / 9, see exercise 3.3.6.
 Section 3.4 25

Exercise 3.4.2
a) The velocity v is now given by the relation v = Us g r2 / P f(U / Us), where f(U / Us)
is any function of U / Us.
b) This assumption implies that v = c (Us  U) g r2 / P. Here, c is a constant, which is
equal to 2/9, see exercise 3.3.6.

Exercise 3.4.3
a) The use of dimensional analysis leads to
r a v0 g c D d
b
c1 ,
b c
§L· § L · d
La ¨ ¸ ¨ 2 ¸ L0 M 0 T 0 La  b  c T  b  2 c c2 .
©T ¹ ©T ¹
The exponent of T implies b = 2 c, and the exponent of L implies that c = a.
Hence, we find
a
§rg· d
r v0
a  2c
g D a d ¨ 2¸ D c1.
¨v ¸
© 0 ¹
Therefore, r = cr(D) v02 / g.
b) The constraint r(D = S / 2) = 0 suggests cr(D) = c sin(2 D), where c is a constant.
Consequently, we find r = c sin(2 D) v2 / g. Calculus provides the constant c = 1
(see exercise 7.3.1).

Exercise 3.4.4
a) The use of dimensional analysis leads to
P a d b vc U d P e Z f c1 ,
a c d e f
§ M L2 · b § L · § M · § M · §1·
¨ 3 ¸ L ¨ ¸ ¨ 3 ¸ ¨¨ ¸¸ ¨ ¸ L2 a  b  c  3d  e M a  d  e T  3a  c  e  f c2 .
¨ T ¸ © T ¹ © L ¹ © LT
© ¹ ¹ ©T ¹
The independent parameters are given by a, e, and f. The exponents of M and T
imply d = (a + e) and c = (3 a + e + f). In terms of these relations for d and c,
the exponent of L reads 2 a + b  (3 a + e + f) + 3 (a + e)  e = 2 a + b + e  f.
Thus, we find the constraint b = 2 a + f e. Hence, we find
a e f
§ P · § P · § dZ ·
a
P d  2a  f e
v  3a  e  f
U a e
P Z
e f
¨¨ ¸
2 3 ¸
¨¨ ¸¸ ¨ ¸ c1.
©Ud v ¹ © Udv¹ © v ¹
Therefore, P = cP U d 2 v3, where cP = cP[P / (U d v), d Z / v].
b) For the same fluid, the design conditions are given by

d M vM d Rv R , d M Z M vM d R Z R vR .

For d M = d R / q we obtain vM = q vR and ZM = q (vM / vR) ZR = q2 ZR.

26 Chapter 3

Exercise 3.4.5
a) The use of dimensional analysis leads to
'p a d b D c v d U e P f c1 ,
a d e f
§ M · b c§L· §M · § M ·
¨¨ ¸ L L¨ ¸ ¨ 3¸
2 ¸
¨¨ ¸¸ L a  b  c  d  3e  f M a  e  f T  2 a  d  f c2 .
© LT ¹ ©T ¹ © L ¹ © LT ¹
The independent parameters are given by a, b, and f. The exponents of M and T
imply e = (a + f) and d = 2a  f. In terms of these relations for e and d, the
exponent of L reads a + b + c 2a  f + 3 (a + f)  f = b + c + f. Therefore, we
obtain the constraint c =  b  f. These relations imply
a b f
§ 'p · § d · § P ·
'p d Da b b  f
v  2a  f
U a  f
P f
¨¨ ¸
2 ¸ ¨ ¸ ¨¨ ¸¸ c1.
© Uv ¹ © D ¹ © U Dv ¹
Therefore, we find 'p = cp U v2, where cp = cp[P / (U d v), d / D].
b) For the same fluid, the design conditions are given by
dM dR
, DM vM D Rv R .
DM DR
For DM = DR / q we obtain d M = d R / q and vM = q vR.

Section 3.5
Exercise 3.5.1
a) The design condition is T M = T R.
b) This ratio is given by FLM / FLR = (d2M / d2R )2 = 1 / q2.

Exercise 3.5.2
a) The use of dimensional analysis leads to
h a mb vc V d g e U f P h c1 ,
c d e f h
§L· §M · § L · §M · § M ·
L M ¨ ¸ ¨ 2 ¸ ¨ 2 ¸ ¨ 3 ¸ ¨¨
a b
¸¸ L a  c  e 3 f  h M b  d  f  h T  c  2d  2e  h c2 .
© T ¹ © T ¹ © T ¹ © L ¹ © LT ¹
The independent parameters are a, d, e, and h. The exponent of T does imply
that c = 2 d  2 e  h. The sum of L and T exponents is a 2 d  e  3 f  2 h.
Therefore, we find the constraint 3 f = a 2 d  e  2 h. The use of the latter
relation in the M coefficient does imply that b =  d  h  (a 2 d  e  2 h) / 3
=  (a + d  e + h) / 3. Hence, we find

h a m  ( a  d  e  h ) / 3 v 2 d  2 e  h V d g e U ( a  2 d  e  2 h ) / 3 P h c1.
 Section 3.5 27

The latter condition can be also written

a d e h
§ h · § V · § hU g · § P ·
¨ ¸ ¨ ¸ ¨¨ 2 ¸¸ ¨ ¸ c1 ,
¨ hU ¸ ¨ U hU v 2 ¸ © v ¹ ¨ U hU v ¸
© ¹ © ¹ © ¹
where hU = (m / U)1/3 is used. The nondimensional products involved are given
by the expressions
h V v2 U hU v
H , T , Fr , Re .
hU U hU v 2
hU g P
b) The height h of the column is given by h = ch hU , where ch = ch(T, Fr, Re).

Exercise 3.5.3
a) The use of dimensional analysis leads to
U a pb d c D d Z e U f P g c1 ,
a b e f g
§ M · §1· §M · §M ·
¨¨ ¸
2 ¸
Lc  d ¨ ¸ ¨ 3 ¸ ¨¨ ¸¸ L  a  b  c  d  3 f  g M a  b  f  g T  2 a  2b  e  g c2 .
© LT ¹ ©T ¹ © L ¹ © LT ¹
The independent parameters are given by a, c, e, and g. The exponent of T
implies that b =  a  (e + g) / 2. The use of this relation in the M exponent
leads to the condition f = (e + g) / 2  g = (e  g) / 2. The sum of the L and M
exponents reads c + d  2 f. In terms of the relation for f we obtain d = e  g  c.
Hence, we find
a e g
§ U · § d · §¨ D Z ·¸ §¨ P ·¸
c

U pa a e / 2 g / 2 c
d D ec  g
Z U
e e / 2 g / 2
P g
¨¨ ¸¸ ¨ ¸ c1 ,
© p ¹ © D ¹ ¨© vU ¸¹ ¨© U vU D ¸¹
where vU = (p /U)1/2 is used. The nondimensional products involved are given by
the expressions
U d DZ P
U* , d* , Z* , P* .
p D vU U D vU
b) The intensity of sound energy is given by U = cU p, where cU = cU (d*, Z*, P*).
Chapter 4
Section 4.2
Exercise 4.2.1
a) The required conditions are: (i) f(t) integrates to one, (ii) f(t) o f for t = t0, and
(iii) f(t) = 0 for t z t0.
b) First, we show that f(t) integrates to one,
t0
1  |t  t 0 | / W 1 (t  t 0 ) / W 1 f  (t  t 0 ) / W
³ f (t ) dt ³e dt ³e dt  ³e dt
2W 2W f 2W t 0

> @ > @
1 (t  t0 ) / W t0 1  (t  t 0 ) / W f 1 1
e f
 e t0
 1.
2 2 2 2
The function f(t) has the properties
­1 /(2W ) o f if t t0
lim f (t ) ® .
W of
¯0 if t z t0
Consequently, f(t) o G(t  t0) for W o 0.

Exercise 4.2.2
a) This is possible. The condition is b  a o 0.
b) There is no difference between these delta functions.

Exercise 4.2.3
Consider the relation
dg ( x) G ( x  x0 )
³ dx > g ( x) G ( x  x0 )@ f f 0
dx
dg dG ( x  x0 )
³ dx G ( x  x0 ) dx  ³ g ( x)
dx
dx.

The delta function is zero at positive and negative infinity. Therefore, the bracket
term disappears at positive and negative infinity. The latter relation can be written
dG ( x  x0 ) dg dg dg
³ g ( x) dx ³ ( x) G ( x  x0 ) dx ³ ( x0 ) G ( x  x0 ) dx  ( x0 ).
dx dx dx dx
The last two expressions arise from the use of the sifting property and normaliza-
tion property of delta functions.

Exercise 4.2.4
a) The probability is given by
0
1 2
P(3 d X d 0) ³ dx 5 40%.
2 5
30 Chapter 4

b) For this case we obtain

3
1 3
P (0 d X d f ) ³ 5 dx 5 60%.
0

Exercise 4.2.5
a) The passenger will wait less than 2 minutes if he arrives between 7:13 and 7:15,
or if he arrives between 7:28 and 7:30. The probability for these cases is
15
dx 30
dx 2 2
P(13 d X d 15)  P(28 d X d 30) ³  ³ 30 2 30 15 13.3%.
13 30 28

b) The passenger will wait more than 10 minutes if he arrives between 7:00 and
7:05, or if he arrives between 7:15 and 7:20. The probability for these cases is
5
dx 20
dx 5 1
P(0 d X d 5)  P(15 d X d 20) ³ 30  ³ 30 2 30 33.3%.
0 15 3

Section 4.3
Exercise 4.3.1
a) The entropy SNormal of the normal PDF is given by
ª 2º

 ³ f ( x) ln f ( x) dx « § 2S X~ 2 ·  x  X » dx
S Normal ³ f ( x ) ln
« ©
¨ ¸ ~
¹ 2 X2 »
¬ ¼
1 ~2 1 ~2 ~2
ln 2S X  ln(e) ln 2 e S X ln 2 e S X .
2 2
b) The corresponding entropy SUniform of the uniform PDF is given by
~
SUniform ln(b  a) ln 12 X 2 ,
~
where b  a = 121/2 < X 2 >1/2 is used to rewrite b  a.
c) We find SNormal > SUniform because 2 e S = 17.08 > 12.
d) Because the uniform PDF makes an assumption about the PDF shape.

Exercise 4.3.2
a) This probability is given by
P(80 d X d 120) erf ( B)  erf ( A) 2 0.8178.
Here, A = (80  100) / (2 15) = 0.9428 and B = (120  100) / (21/2 15) = 0.9428.
1/2

The error function approximation (4.84) was used to calculate the final value.
a) This probability is given by
P(150 d X  f) 1  erf ( A) 2 0.0004.
Here, A = (150  100) / (2 15) = 2.3570.
1/2
 Section 4.3 31

Exercise 4.3.3
a) This probability is given by
P(1.5 d X d 1.7) erf ( B)  erf ( A) 2 0.2373.
Here, A = (1.5  1.75) / (2 0.07) = 2.5254 and B = (1.7  1.75) / (21/2 0.07) =
1/2

0.5051. The error function approximation (4.84) was used to obtain the result.
a) This probability is
P(1.95 d X  f) 1  erf ( A) 2 0.0021.
Here, A = (1.95  1.75) / (2 0.07) = 2.0203. Hence, 0.21% of the adult male
1/2

population is more than 1.95 m tall.

Exercise 4.3.4
a) This probability is given by
P(265 d X d 271) erf ( B)  erf ( A) 2 0.1585.
Here, A = (265  268) / (2 15) = 0.1414 and B = (271  268) / (21/215) = 0.1414.
1/2

The error function approximation (4.84) was used to obtain the result.
b) This probability is given by
P P(f  X d 240)  P(290 d X  f) 1  P(240 d X d 290).
We have
P(240 d X d 290) erf ( B)  erf ( A) 2 0.8981,
where A = (240  268) / (2 15) = 1.3199 and B = (290  268) / (21/2 15) =
1/2

1.0371. Hence, the probability P = 0.1019.

Exercise 4.3.5
a) The parameter a is determined by (we have c t 0)
2
f 2 2
f 2 e c [b /( 2 c )] f
 y2
e a ³e
b x c x
dx ec [b /( 2 c )] ³e
 c [ x  b /( 2 c )]
dx ³e dy ,
0 0 c c b /( 2 c )

where the variable y = c1/2 [x + b / (2 c)] is introduced. This expression can be

rewritten in terms of the error function definition,

a eb
2
ª 0  y2
/ ( 4c ) f
 y2
º eb
2
/ ( 4c )
S ª § b ·º
e « ³ e dy  ³ e dy » «erf (f)  erf ¨¨ ¸¸»
c «¬b / ( 2 c ) 0 »¼ c 2 «¬ ©2 c ¹»¼
1 S b 2 / ( 4c ) ª § b ·º
e «1  erf ¨¨ ¸¸» .
2 c ¬« © 2 c ¹¼»
b) The mean is defined by
f
X ³ x f ( x) dx .
0
32 Chapter 4

We differentiate f(x) to obtain

df
 f ^b  2 c x`.
dx
By dividing this relation by 2 c we find
1 § df ·
xf  ¨  b f ¸.
2c © dx ¹
The use of this relation in the definition of <X> results in
f (f)  f (0) b ea  b
X   .
2c 2c 2c
c) The variance is defined by
f
~ 2 1 f § df · 2
X2 ³ x f ( x) dx  X  ³ x¨ b f ¸ dx  X
2

0 2c 0 © dx ¹
1 f df b 2
 ³ x dx  X  X ,
2c 0 dx 2c
where the relation for x f obtained in b) is used. The integral involving df / dx
can be rewritten,
f f
df ª º dxf dx
³ x dx dx ³ « dx  dx f » dx 1.
0 0¬ ¼
Hence, we obtain
~ 1 b X 2
X2  X .
2c

Exercise 4.3.6
a) The parameter a is determined by

³ e dx  b >e @0
f
1 f 1
e a b x b x
.
0 b
Hence, the PDF reads f(x) = b ebx. The mean and variance are determined by
f
f ª e b x º b 1
X b ³ x e  b x dx b«  b x 1 » .
¬ (b)
2
0 ¼0 b2 b
f
~ f
2 ª § x2 2x 2 ·º 1 1
X2 b ³ x 2 e  b x dx  X b «e  b x ¨¨   ¸  2
3 ¸»
.
0 ¬« ©  b ( b ) 2
( b ) ¹¼» 0 b b2
 Section 4.3 33

b) The consequence is that both the mean and standard

deviation are equal to 1/b. The idea that the standard
deviation increases linearly with the mean can be an
appropriate model for some cases, but such a model
will be not always appropriate. A closer look at the
properties of the PDF model for arbitrary c values
shows that this is a characteristic feature of the PDF
model considered: the standard deviation has about
the same value as the mean. This fact is illustrated in
terms of the figure showing the mean and standard deviation multiplied with b
as functions of y = 2 c1/2 / b. The corresponding functions follow from the results
of exercise 4.3.5. We have
2 § 1 · ~ 2 2
b X ¨  a  1¸¸ ,
2 ¨
b2 X 2 1  b X  b2 X ,
y © be ¹ y2
where b ea = S1/2 erfce(1 / y) / y. Here, the exponentially scaled complementary
error function is defined by erfce(1 / y) = exp(1 / y2) [1  erf(1 / y)]. The figure
~
also shows that < X 2 >1/2 d <X>. In consistency with the results obtained in part
a) it may be seen that the mean and standard deviation have an equal maximum
of 1/b at c = 0.

Exercise 4.3.7
a) The PDF has the structure of a second-order SML PDF. This PDF represents a
SML PDF if the PDF parameters are chosen so that the PDF f(x) agrees with
any given mean P and any given variance V2.
b) The conditions for the calculation of b and c are
ea  b 1 b P
P , V2  P2 ,
2c 2c
where ea is a function of b and c according to
1 S b 2 / ( 4c ) ª § b ·º
e a e «1  erf ¨¨ ¸¸» .
2 c «¬ ©2 c ¹»¼
For any given b value, c can be calculated by means of the variance relation,
1 1 b P
c .
2 V 2  P2
For the given b and c values, ea can be obtained via the ea relation above. Then,
the b value has to be varied until the relation for the mean is satisfied, i.e., until
ea  b
P .
2c
34 Chapter 4

Exercise 4.3.8
b) See the figure. The b and c values for these cases
V = 0.25, V = 0.5, and V = 1 are given by (b, c) =
(15.9908, 7.9957), (3.1478, 1.6591), and (1, 0),
respectively. The b and c values for the case V = P
= 1 follow from the results of exercise 4.3.6.
c) The PDF features are very similar to the features
of the gamma PDF for the corresponding values of
the mean and variance. However, the gamma PDFs
for V < 1 are slightly more skewed. A significant difference is given by the fact
that the SML PDF cannot produce PDFs for which V > 1 because the standard
deviation of the SML PDF cannot be larger than the mean: see exercise 4.3.6.

Exercise 4.3.9
a) The central moments of third and fourth order are given by the expressions
~ 3 2 3
X3 X X X3 3 X2 X 3 X X  X ,
~ 4 2 3 4
X4 X X X4 4 X3 X 6 X2 X 4 X X  X .

According to relation (4.96) we find

3 2
~ X X 3
X3 (a  2) ( a  1) a 3
 3 ( a  1) a 2
X 2 X
a a
3 3
X 2 X 2 ~ 3/ 2
2
a  3a  2  3 (a  a)  2a
2 2 2
1/ 2 3/ 2 1/ 2
X2 ,
a a a a
4
~ 2 4 X
X4 X4 4 X3 X 6 X2 X 3 X (a  3) (a  2) ( a  1) a
a4
3 2
X X 2 4
 4 (a  2) (a  1) a X  6 ( a  1) a X 3 X
a3 a2
4
X
a 3  3a 2  2a  3a 2  9a  6  4 (a 3  3a 2  2a)  6a 3  6a 2  3a 3
a3
4 4
a2 X
X a  2 ~2 2
3 a2 3
3 2
3 X .
a a a a
~
Here, < X 2 >1/2 = <X> / a1/2 is used to obtain the last expressions. Thus, we find
2 § 2·
m3 , m4 3 ¨1  ¸ .
a © a¹
b) For a o f we have the case that the skewness and flatness values are equal to
the values m3 = 0 and m4 = 3 of a normal PDF.
 Section 4.3 35

~
c) The parameter a is defined by a = <X>2 / < X 2 >. The case a o f corresponds
~2
to the condition that < X > o 0.
d) This PDF is a delta function with the peak at <X>.

Exercise 4.3.10
a) The moments of f(x) can be calculated by (k = 1, 2, }),
1
1 1 a  k 1 B ( a  k , b)
Xk k
³ x f ( x) dx ³x (1  x)b 1 dx .
0 B ( a , b) 0 B ( a, b)
The definition (4.108) of the beta function is used to obtain the last expression.
According to relation (4.109), the ratio of beta function can be written
B ( a  k , b) * (a  k )* (b) * (a  b) * ( a  k ) * ( a  b)
B ( a, b ) * (a  b  k ) * (a)* (b) * ( a ) * (a  b  k )
(a  k  1)(a  k  2) a
,
(a  b  k  1)(a  b  k  2) (a  b)
where Eq. (4.95) is used for *(a + k) / *(a) and *(a + b + k) / *(a + b). Hence,
the moment of order k of f(x) reads
(a  k  1)(a  k  2)  a
Xk .
(a  b  k  1)(a  b  k  2) (a  b)
b) According to the last result, the mean and variance are given by
a ( a  1) a
X , X2 .
ab (a  b  1) (a  b)
The combination of these expressions for <X> and <X2> then implies
~ a ª a 1
2 a º
X2 X 2
 X 
a  b ¬ a  b  1 a  b »¼
«
a (a  1)(a  b)  a(a  b  1) a b
.
ab (a  b  1)(a  b) a  b (a  b  1)(a  b)
Next, we use the product between <X> = a / (a + b) and 1  <X> = b / (a + b)
to replace a b / (a + b)2. Then, the variance relation reads
X 1 X
a  b 1 ~ .
X2
Next, we replace a + b by a + b = a / <X> and 1  <X> = b / (a + b), respective-
ly, according to <X> = a / (a + b). This leads to the relations
ª X 1 X º ª X 1 X º
a X « ~  1», b 1 X « ~  1» .
« X2 » « X2 »
¬ ¼ ¬ ¼
36 Chapter 4

Exercise 4.3.11
a) For the case a = b (this means for <X> = 0.5), we have m3 = 0 as given for the
normal PDF and m4 = 3 (2a + 1) / (2a + 3). Hence, the values m3 = 0 and m4 = 3
of a normal PDF are recovered for a = b o f.
~
b) The case a = b o f corresponds to the condition that < X 2 > o 0.
c) This PDF is a delta function with a peak at <X> = 0.5.

Exercise 4.3.12
a) By making use of the abbreviation a = 2 / a0, R is found to be given by
f
4 f
3 2 r / a 4 ª e ar 3 3 º 24 3
R 3 ³ r e 0 dr 3 « 4
a r  3a 2 r 2  6ar  6 » 3 4
a0 .
a0 0 a0 ¬ a ¼0 a0 a 2

b) The second-order moment is given by

f
4 f
4 2 r / a 4 ª e ar 4 4 º
r2 3 ³ r e 0 dr 3 « 5
a r  4a 3 r 3  12a 2 r 2  24ar  24 »
a0 0 a0 ¬ a ¼0
4 ˜ 24 2
 3 5
3 a0 .
a0 a
Hence, the variance reads
9 2 3 2
V2 r 2  R2
2
3 a0  a0 a0 .
4 4
c) We adjust the general gamma PDF to the case considered,
a 1
a § ar · § ar ·
p(r ) ¨ ¸ exp¨  ¸.
* (a) R © R ¹ © R ¹
The parameter a is given by a = R2 / V2 = (9 a02 / 4) / (3 a02 / 4) = 3. Therefore, we
have *(3) = 2. Consequently, the gamma PDF can be written
2 2
3 § 3r · § 3r · 1 § 2r · § 2r · 4r2 § 2r ·
p(r ) ¨ ¸ exp¨  ¸ ¨ ¸ exp¨  ¸ exp¨¨  ¸.
2R © R ¹ © R¹ ¨ ¸
a0 © a0 ¹ ¨ a ¸ 3 ¸
© 0 ¹ a0 © a0 ¹
This is the PDF considered. Hence, the PDF considered is a gamma PDF.
d) The first-order and the second-order derivatives of p(r) read

dp §1 1 ·
2 p ¨¨  ¸¸,
dr © r a0 ¹
2
d2p §1 1 · 2 p
4 p ¨¨  ¸¸  2 .
dr 2 © r a0 ¹ r
Hence, p(r) has a maximum at r = a0 (we have d 2p / dr2 < 0 for this value).
 Section 4.4 37

e) This probability is given by

3a 0
4 3a 0
 2 r / a0 4 ª e ar 2 2 º
P 3 ³r e
2
dr 3 « 3
a r  2ar  2 »
a0 0 a0 ¬ a ¼0
4 § e3 a a 0 2· 4 6
¨ 3 9 a 2 a0 2  6 a a0  2  3 ¸
3 ¨
e 36  12  2  2
a0 © a a ¸¹ 8
1  25 e  6 0.938.

Section 4.4
Exercise 4.4.1
a) We rewrite V2Mean in the following way,
2 2

V 2
Mean >P N  X @ 2 ª1 N
«N ¦
º
Xi  X »
1 ªN
2 «¦
N ¬i 1
º
Xi  X » .
¬ i1 ¼ ¼
Next, we use of the fact that Xi  <X> are independent random numbers: the
mean of the squared sum is equal to the sum of quadratic means because only
quadratic terms of Xi  <X> provide nonzero contributions. This leads to
1 N 2 N 2 V2
V 2 Mean ¦ Xi  X X X .
N2 i 1 N2 N
b) The two relations are very similar. The difference is that the conclusion VMean =
V / N 1/2 is independent of the consideration of normally distributed numbers,
and the relation VMean = V / N 1/2 explains the scaling of H. Therefore, the relation
VMean = V / N 1/2 is more general.

Exercise 4.4.2
a) To show the validity of <V2N> = V2 we consider the mean value of V2N,

V 2N
1 N
¦ X i  X  PN  X
N 1 i 1
> 2
@ > @
1 N 2
> @
¦ X i  X  2 X i  X PN  X  PN  X
N 1 i 1
> @> @ > @ 2

1 §N
¨¦ Xi  X
N 1 © i 1
2
>
 2N PN  X
2
@
 N PN  X > @ > @ 2 ·
¸
¹
1 §N
¨¦ Xi  X
N 1 © i 1
2
>
 N PN  X
2 ·
¸ @ > @
¹
1
N V 2  V 2 V 2.
N 1
38 Chapter 4

The last expression was obtained here by using the definitions of V2 and V2Mean
= V2 / N (see exercise 4.4.1),
V2
¦ >X i  X @ >P @
N 2 2
N V 2, N  X .
i 1 N
b) The relevance of Bessel’s correction does depend on the factor N / (N  1). The
relative error of approximating N / (N  1) by one is given by 1 / (N  1). The
error 1 / (N  1) d 1% for N > 101.

Exercise 4.4.3
The first-order and second-order derivatives of g(x) are given by
dg d 2g
2 X  x 2 X  x , 2.
dx dx 2
Therefore, g(x) has a minimum at x = <X>. Hence, the best characterization of X is
given by the mean <X> of X.

Section 4.5
Exercise 4.5.1
a) The neglect of O3 and O4 results in O1 = 0 and O2 = 1.
b) For a data set with m3 = 0 and m4 = 3 we find O3 = O4 = O1 = 0 and O2 = 1. The
latter set of model parameters corresponds to a second-order SML PDF model
(a normal PDF).

Exercise 4.5.2
a) The parameters m5 and m6 are defined as integrals over the fourth-order SML
PDF, which is a function of O1, O2, O3, and O4. Therefore, m5 and m6 represent
functions of O1, O2, O3, and O4. Thus, the equation system (4.136) is a nonlinear
equation system for O1, O2, O3, and O4.
b) We combine given values for m3 and m4 with appropriate initial values for m5
and m6. The model parameters O1, O2, O3, and O4 are calculated by the relations
(4.136). The PDF is used to calculate new m3, m4, m5, and m6. We continue in
this way with the calculation of O1, O2, O3, and O4 until the first four moments
provided by the PDF agree with the known values.

Exercise 4.5.3
a) The figure shows this PDF for 'T = 0.1, 'T = 0.2, and 'T = 0.3 in (a), (b), and
(c), respectively. The positions at which the PDF is calculated are separated by
0.02. The setting 'T = 0.2 is the optimal choice: the noise effects are not too
strong, and there is a relatively minor reduction of the peak value.
 Section 4.5 39

b) See the figure: the PDFs f1 and f2 refer to the 1850–

1930 data and 1930–2008 data, respectively.
c) The PDF for the next eighty years may have a peak
at higher temperatures. The variance may be larger
than the f2 variance corresponding to the potential
appearance of large temperature fluctuations.
Chapter 5
Section 5.2
Exercise 5.2.1
a) We consider yn for n = 1, 2, }
y1  y0  1, y2  y1  2 ( y0  1)  2 y0  1,
y3  y2  3 ( y0  1)  3  y0  2, y4  y3  4 ( y0  2)  4 y0  2,
y5  y4  5 ( y0  2)  5  y0  3, y6  y5  6 ( y0  3)  6 y0  3.
b) We may guess the general solution to be given by (n = 0, 1, 2, })
n 1
yn (1) n y0   1  (1) n .
2 4

Exercise 5.2.2
a) For n = 0 we obtain y0 = y0.
b) The use of the solution in the equation yn = a yn1 + b n + c leads to
b ª 1  an º 1  an
a n y0  «n  a »c
1 a ¬ 1 a ¼ 1 a
­° b ª 1  a n 1 º 1  a n 1 ½°
a ®a n 1 y0  «n  1  a »c ¾bn  c
°̄ 1 a ¬ 1 a ¼ 1  a °¿
b ª a  an º a  an 1  a
a n y0  «a (n  1)  a  n (1  a)»  c
1 a ¬ 1 a ¼ 1 a
b ª a 1 1  an º 1  an
a n y0  «n  a  a »c .
1 a ¬ 1 a ¼ 1 a
The last two lines represent rewritings of the second line. It may be seen that
we have the same expressions on both sides of the difference equation. Hence,
the solution considered does indeed solve the equation.

Exercise 5.2.3
a) For n = 0, n = 1, and n = 2 we have 0 = 0, 1 = 1, and a + 2 = a + 2, respectively.
b) For n = 4 and a = 7 we have 466 = 466.

Exercise 5.2.4
a) For a = 1 we find
1 ª 1  an º n (n  1)
lim « na » 1  2  3    (n  1)  n .
a o1 1  a 1 a ¼ 2
¬
42 Chapter 5

b) The use of both identities for the case a o 1 leads to

1 ª 1  an º 1  an n (n  1)
lim yn y0  b lim « na »  c lim y0  b  c n.
a o1 a o1 1  a 1 a ¼ a o1 1  a 2
¬

Exercise 5.2.5
a) The asymptotic trends are given by
­ b
°a  1 : n
° 1  a
°° n ( n  1)
lim yn ®a 1 : b .
nof
° 2
° n§ ab c ·
°a ! 1 : a ¨¨ y0   ¸
°¯ © ( 1  a ) 2
1  a ¸¹

b) We have to consider the case a > 1 to model the observations. The condition is
§ ab c ·
a n ¨¨ y0   ¸¸ 2e3n .
© (1  a ) 2
1  a ¹
This condition can be satisfied by setting
ab c
a e3 , y0   2.
(1  a ) 1  a
2

Exercise 5.2.6
a) The condition is given by
12 1
§ 0.04 · § p·
¨1  ¸ S0 ¨1  ¸ S 0 .
© 12 ¹ © 1 ¹
Hence, we have p = (1 + 0.04 / 12)12  1 = 0.0407.
b) Here, the condition is given by
20
§ p·
¨1  ¸ S 0 2 S 0 .
© 4 ¹
Hence, we have p = 4 (21/20  1) = 0.1411.

Exercise 5.2.7

a) Years A B C
5 1.3 S0 1.2210 S0 1.2202 S0
10 1.6 S0 1.4908 S0 1.4889 S0
15 1.9 S0 1.8203 S0 1.8167 S0
20 2.2 S0 2.2226 S0 2.2167 S0
 Section 5.2 43

b) The interest offer of the company A, A, A, and B maximizes the amount on

deposit after 5, 10, 15, and 20 years, respectively.

Exercise 5.2.8
a) For Dn = 0, the loan repayment formula provides
p 1 0.05 / 12
R n
D0 360
200,000 1073.6432.
r 1 1/ 1  p / r 1  1 / 1  0.05 / 12
b) The total amount paid back on the loan is $386,511.5684.

Exercise 5.2.9
a) After three conversion periods the amount in the fund is given by
S1 (1  p / r ) S0  S0 , S2 (1  p / r ) S1  S0 , S3 (1  p / r ) S 2  S0 .
The difference equation for this case reads (n = 1, 2, })
Sn (1  p / r ) S n 1  S0 .
This equation is solved by

§ p·
n
1  (1  p / r ) n ­°§ p·
n
r ª§ p·
n
º ½°
Sn ¨1  ¸ S 0  S 0 S0 ®¨1  ¸  «¨1  ¸  1» ¾.
© r¹ 1  (1  p / r ) °̄© r¹ p «¬© r¹ »¼ °¿
b) For p / r << 1 we have (1 + p / r)n = 1 + n p / r. Hence, the solution simplifies to
­ p rª p º½ ­ § p ·½
Sn S0 ®1  n  «1  n  1» ¾ S0 ®1  n¨1  ¸¾.
¯ r p¬ r ¼¿ ¯ © r ¹¿

Exercise 5.2.10
a) The difference equation reads for this case (n = 1, 2, })
Sn (1  p / r ) S n 1  D.

This equation is solved by (S0 = 1000, p = 0.05, r = 12)

r ª§ º
n n n
§ p· 1  (1  p / r ) n
§ p· p·
Sn ¨1  ¸ S 0  D ¨ 1  ¸ 0 S  D «¨ 1  ¸  1».
© r¹ 1  (1  p / r ) © r¹ p «¬© r¹ »¼
b) D can be calculated in the following way:
n 36
§ p· § 0.05 ·
S  ¨1  ¸ S 0 30,000  ¨1  ¸ 1000
p n © r¹ 0.05 © 12 ¹
D n 36
744.156.
r § p· 12 § 0.05 ·
¨1  ¸  1 ¨1  ¸ 1
© r¹ © 12 ¹
44 Chapter 5

Section 5.3
Exercise 5.3.1
a) The eigenvalues of this difference equation are r1 = i / 2 and r2 = i / 2. Hence,
we have rS = 0 and rD* = 1/2. The solution parameters are given by N = 1/2 and
M = arctan(f) = S /2. Hence, the solution reads
2 1
yn  sin[(n  1) S / 2] 2  sin[(n  1) S / 2].
2 n 1 2n 1
b) The use of sin(z1 + z2) = sin z1 cos z2 + cos z1 sin z2 allows the rewriting
sin[(n  3) S / 2] sin[(n  1) S / 2] cos(S )  cos[(n  1) S / 2] sin( S )
 sin[(n  1) S / 2].
Hence, we have
1 4
yn  2  n 3
sin[( n  3) S / 2] n 1
sin[(n  1) S / 2] 4 yn 1 ,
2 2
which proves that this function satisfies the difference equation. We also have
1 1 1
y0  1 sin[(1) S / 2] 1
2, y1  n 1 sin[0] 0.
2 2 2
Thus, the function is indeed the solution.
c) The asymptotic solution for n o f is yn = 0.

Exercise 5.3.2
a) The eigenvalues of this difference equation are given by r1 = 21/2 and r2 = 21/2.
The stationary solution is s = 1. Thus, the solution reads for this case

2 n / 2  (1) n 2 n / 2 2( n 1) / 2  (1) ( n 1) 2( n 1) / 2

yn 1 ( 1  1)  2 (1  1)
2 21 / 2 2 21 / 2
1  2( n 1) / 2  (1) n 2( n 1) / 2 1  2( n 1) / 2 1  ( 1) n .
b) The eigenvalues of this difference equation are given by r1 = r2 = 1/2. The sta-
tionary solution is s = 1. Thus, the solution reads for this case
1 1 n
yn 1 n n 1
(1  1)  (n  1) n
(1  1) 1  n  2 .
2 2 2
c) The eigenvalues of this equation are r1 = 1 + i and r2 = 1  i. Thus, we have rS =
1 and rD* = 1. The solution parameters are given by N = 21/2 and M = arctan(1) =
S / 4. The stationary solution is s = 1. Hence, the solution reads for this case
yn 1  2n / 2 sin[ n S / 4] ( 1  1)  2( n 1) / 2 sin[( n  1) S / 4] (1  1)
1  2( n  3) / 2 sin[(n  1) S / 4].
 Section 5.3 45

Exercise 5.3.3
a) The particular solution is given by Yn = 2 n2.
b) By using xn = yn  Yn in the yn equation we obtain
xn  2 n 2 2 [ xn 1  2 (n  1) 2 ]  [ xn  2  2 (n  2) 2 ]  4
2[ xn 1  2 (n 2  2 n  1)]  [ xn  2  2 (n 2  4 n  4)]  4
2 xn 1  xn  2  2 n 2 .
Hence, the xn equation is given by xn = 2 xn1  xn2. The initial data are given by
x0 = 1 and x1 = 2.
c) The eigenvalues of this equation are r1 = r2 = 1. The solution for this case reads
xn n1n 1 2  (n  1)1n 1 2 n  n  1 n  1.
d) We have yn = 1 + n 2 n2, which results in y0 = 1 and y1 = 0. This solution does
also satisfy the yn equation,
1  n  2 n2 2 [1  n  1  2 (n  1) 2 ]  [1  n  2  2 (n  2) 2 ]  4
4 n 2  10 n  4  [2 n 2  9 n  9]  4 2 n 2  n  1.

Exercise 5.3.4
a) The numbers of the Fibonacci sequence are given in the following table:

n 0 1 2 3 4 5
yn 1 1 2 3 5 8

b) The difference equation is given by (n = 2, 3, })

yn yn 1  yn  2 .
We also know that y0 = 1 and y1 = 1.
c) The eigenvalues of this equation are r1 = r and r2 = 1  r, where r = (1 + 51/2) / 2.
The variable r is called the golden ratio. The solution for this case reads
r n  (1  r ) n 1  5 r n 1  (1  r ) n 1 r n  (1  r ) n  r n 1  (1  r ) n 1
yn 1 1 .
2 r 1 4 2 r 1 2 r 1
d) After 12 months there will be 233 pairs of rabbits.

Exercise 5.3.5
a) This equation is given by
Rn (1  f ) Rn 1  K .
This model is a first-order difference equation. The solution reads
K § K·
yn  1 f
n
¨¨ y0  ¸¸.
f © f ¹
46 Chapter 5

b) The solution is shown in the figure. The model is

not acceptable: it predicts a steady increase of the
number of red blood cells.

Exercise 5.3.6
a) The given solution Rn = A cos(Z n  G) implies that
Rn 1 A cos[Z (n  1)  G ] A cos(Z n  G ) cos(Z )  sin(Z n  G ) sin(Z )
A cos(Z n  G ) cos(Z )  sin(Z n  G ) sin(Z ) ,
Rn  2 A cos[Z ( n  2)  G ] A cos(Z n  G ) cos(2Z )  sin(Z n  G ) sin(2Z )
A cos(Z n  G ) cos(2Z )  sin(Z n  G ) sin(2Z ) .
The equation Rn = a Rn1 + b Rn2 then implies
0 A cos(Z n  G ) > 1  a cos Z  b cos 2Z @  A sin(Z n  G ) >a sin Z  b sin 2Z @.
This equation is satisfied for all n if a = 2 cos Z and b = 1.
b) The shortcoming of the expressions a = 1  f and b = f used in Sect. 5.3.4 is the
assumption that b is related to a. According to b = 1, a better assumption is to
consider b as independent of a, this means as a constant.

Section 5.4
Exercise 5.4.1
a) There are two equilibrium solutions: ye1 = 1.618 and ye2 = 0.618. We find that
|df(ye) / dye| = 1 / (1 + ye)2. The solution at ye1 is unstable because |df(ye) / dye| > 1.
The solution at ye2 is stable because |df(ye) / dye| < 1.
b) There is one equilibrium solution ye = 0. We find that |df(ye) / dye| = 1 / (1 + ye)2.
The solution at ye is unstable because |df(ye) / dye| = 1.
c) There are three equilibrium solutions: ye1 = 0, ye2 = 1 and ye3 = 1. We have
|df(ye) / dye| = |1  3 ye2|. The solution at all three equilibrium points is unstable
because |df(ye) / dye| t 1.

Exercise 5.4.2
a) The nonzero equilibrium solution is determined by 1 = D exp(E ye). Therefore,
we have ye = (ln D) / E.
b) We have |df(ye) / dye| = |(D  D E ye ) exp(E ye)| = |(D  D ln D ) exp(ln D)|, i.e.
|df(ye) / dye| = |1  ln D |. Thus, the solution at the equilibrium point is stable
under the condition |1  ln D | < 1.
 Section 5.5 47

Exercise 5.4.3
a) The nonzero equilibrium solution is determined by the relation (1 + ye)E = D.
Thus, we find ye = D1/E  1.
b) We find

df ( ye ) D D E ye D D E (D 1 / E  1)
E
 1 E 1/ E E

dye (1  ye ) (1  ye ) (1  D  1) (1  D 1 / E  1)1 E
E (D 1 / E  1)
1 1  E (1  D 1 / E ) .
D 1/ E
Hence, the solution at the equilibrium point is stable under the condition
1  E (1  D 1 / E )  1.

Section 5.5
Exercise 5.5.1
a) The discrete derivative is given by
yn  yn 1 a § y ·
yn 1 ¨1  n 1 ¸.
't 't © K ¹
The right-hand side is independent of 't if a = r 't, where r is a constant that is
independent of 't.
b) The corresponding differential equation for y(t) reads
dy § y·
r y ¨1  ¸.
dt © K ¹

Exercise 5.5.2
a) The discrete derivative is given by
yn  yn 1 yn 1 A  E y n1
e 1 .
't 't
We set A = C 't and E = B 't. Here, C and B are constants that are independ-
ent of 't. In the limit 't o 0 we obtain
yn  yn 1 yn 1 C 't  B 't y n1
lim lim e 1 yn 1 C  B yn 1 .
't o 0 't 't o 0 't
b) The corresponding differential equation for y(t) reads
dy § B ·
C y ¨1  y ¸.
dt © C ¹
This equation is the logistic differential equation.
48 Chapter 5

Exercise 5.5.3
a) The discrete derivative is given by

yn 1 §¨ § A · ·
E
yn  yn 1
¨ ¸  1¸ .
't ¨ ¸
't ¨ © 1  yn 1 ¹ ¸
© ¹
We introduce E = B 't, where B is a constant that is independent of 't. In the
limit 't o 0 we obtain
y §§ A · ·
B 't
y  yn 1 y
lim n lim n 1 ¨ ¨¨ ¸  1¸ lim n 1 e B 't ln[ A / (1 y n1 )]  1
¸
't o 0 't 't o 0 't ¨ 1  y ¸ 't o 0 't
©© n 1 ¹
¹
§ A ·
B yn 1 ln¨¨ ¸.
¸
© 1  yn 1 ¹
b) The corresponding differential equation for y(t) reads
dy § A ·
B y ln¨¨ ¸¸.
dt © 1 y ¹
Chapter 6
Section 6.2
Exercise 6.2.1
a) The equation for <yn> follows from averaging the equation yn = a yn1 + b + r Hn1,

yn a yn 1  b.

By replacing n and n  1 by the subscript e, we obtain <ye> = b / (1  a) for the

equilibrium solution of this equation. The equilibrium solution can be realized
if |a| < 1.
b) The equation for fluctuations can be obtained by taking the difference between
yn = a yn1 + b + r Hn1 and the equation for means,
~
yn a~
yn 1  r H n 1.

By squaring and averaging this equation we find

~
yn
2
a2 ~
2
yn 1  r 2 .

The term < ~yn 1H n 1 > = 0 according to relation (6.12). By replacing n and n  1
by the subscript e we find the equilibrium variance to be < ~ye 2 > = r2 / (1  a2).
The equilibrium solution can be realized if |a| < 1.
c) The variance equation explains why the equilibrium variance of yn is larger than
the variance r2 of the noise process r Hk: the equilibrium variance is the sum of
previously accumulated noise (given by the first term on the right-hand side)
and the variance r2 of the noise process r Hk added by the last time step.

Exercise 6.2.2
a) The mean value is unaffected by the noise. Therefore, we have
1 an
yn a n y0  b .
1 a
b) The first few fluctuations are the following ones,
~y H ,
0 0
~
y1 a~
y0  r H 0 H 0 (a  r ),
~
y a~
y  rH H 0 (a 2  a r  r ),
2 1 0
~
y3 a~
y2  r H 0 H 0 ( a 3  a 2 r  a r  r ).
The difference to the correct calculation of fluctuations is that always the same
noise contribution is added to each realization.
50 Chapter 6

Therefore, we find
~ § 1  an ·
yn H 0 a n  r (a n 1    a1  a 0 ) H 0 ¨¨ a n  r ¸.
© 1  a ¸¹
Consequently, the standard deviation is given by
~ 2
1/ 2 1 an
yn an  r .
1 a
c) By using the expression for ~
yn we find
~ 1  an
y0 ~
yn an  r .
1 a
d) For 0 < a < 1 and n o f we have
~ r2 ~ r
y0 ~
2
yn , yn
(1  a) 2 1 a
instead of the correct expressions
~ r2 r2 ~
y0 ~ an ~
2 2
yn , yn y0 .
1  a2 (1  a)(1  a)
Thus, the incorrect calculation provides a variance that is too large (because of
the missing balance between positive and negative noise contributions) and a
correlation that does not decay (because of the correlated noise).

Exercise 6.2.3
a) See figure a).
b) and c) See figure b). The solid line shows the theoretical result.
d) and e) See figure c). The solid line shows the theoretical result.

Exercise 6.2.4
a) and b) See figure a). The solid line shows the theoretical result.
c) and d) See figure b). The solid line shows the theoretical result.
e) The reason for the difference is given by the minor difference of means and the
significant difference of standard deviations: see the results of exercise 6.2.3.
 Section 6.3 51

Exercise 6.2.5
a) and b) See the figure: the solid line shows the the-
oretical result.

Section 6.3
Exercise 6.3.1
Expression (6.12) shows for the case a = 1 considered that
~
y ~
y  r (H  H    H  H )
n 0 0 1 n2 n 1
~
yn  m ~
y0  r (H 0  H1    H n  2  H n 1  H n    H n  m 1 )
~ ~
y  y r (H    H ).
nm n n n  m 1

Due to the properties of Hk we find, therefore, < ~

yn ( ~
yn  m  ~
yn ) > = 0.

Exercise 6.3.2
a) The variance of xn can be calculated by using (6.15) in combination with a = 1
and r = d 't. In terms of (6.14) for the case a = 1 we obtain
~
yn
2 ~ 2
y0  n r 2 ~ 2
y0  n d 2 ('t ) 2 .

b) In terms of t = n 't, we can write the variance

~
yn
2 ~ 2
y0  't d 2 t .

The problem with the use of this relation is that the effective coefficient d 2 't
changes with the 't applied, which is not a desired feature of equations.
52 Chapter 6

Exercise 6.3.3
We have to show that (i) f(y) integrates to one, (ii), f(y) o f for y = y0, and (iii)
f(y) = 0 for y z y0. The function considered represents a normal PDF. Therefore, it
integrates to one. The function f(y) has the properties
­°1 / 2 S V 2 o f if y y0
lim f ( y ) ® .
V of
°̄0 if y z y0
Hence, f(y) o G(y  y0) for V o 0.

Exercise 6.3.4
a) We know that M = 0.1 kg, y0 = 10 m, U = 10 m / s, and D = U y0. The position
at which the concentration has a maximum is x*n = D xn / (U y02) = 1. Hence, the
maximum position is at x = U y02 / D = y0 = 10 m.
b) The maximum normalized concentration at this position is determined by the
relation C*n = Cn y0 / M = 1 / (2 S e)1/2. Therefore, the maximum concentration is
C = M / [y0 (2 S e)1/2] = 2.42 g / m.

Exercise 6.3.5
a) The mean velocity over the time step 't is v = (yn1  yn) / 't.
b) We have t1 = yn1 / v and t2 = 't  t1.
c) The new position is yn+1 = t2 v = ('t  yn1 / v) v = 't v  yn1 =  yn.

Exercise 6.3.6
a) We integrate Cn,
f
M ­°f ­° y  y0 2
½° f ­° y  y0 2
½° °½
³ Cn dy ® ³ exp® ¾dy  ³ exp® ¾dy ¾
2V n 2V n
2 2
0 2S V n °̄ 0 °̄ °¿ 0 °̄ °¿ °¿

M ­° f  r 2 ½ M ­° f  r 2 ½
f y 0 / 2V n 2
s2 ° r 2 °
® ³ e dr  ³ e ds ¾ ® ³ e dr  ³ e dr ¾
S °̄ y 0 / 2V n 2 y 0 / 2V n 2 °¿ S °̄ y 0 / 2V n 2 f °¿
y 0 / 2V n 2
M 2
³ e  r dr ,
S  y 0 / 2V n 2

where r = (y  y0)/(21/2Vn) and s = (y + y0)/(21/2Vn) are used.

b) For this case we have
2 2
y0 y0 y0 U y0 1
.
2V n
2
2V n
2 2 D xn 2 xn 2
 Section 6.4 53

Therefore, we find

f y 0 / 2V n 2
M 2 0.1 1/ 2 2 0.2 1 / 2 2
³ Cn dy ³ e  s ds s
³ e ds
s
³ e ds 0.1erf (1 / 2 )
0 S  y 0 / 2V n 2
S 1 / 2 S 0

0.0683 kg.

Exercise 6.3.7
a) This formula is given by

M ­° ­° y  y 2
½° ­° y  y0 2
½°½°
®exp® ¾  p exp®
0
Cn ¾¾.
2V n 2V n
2 2
2S V n °̄ °̄ °¿ °̄ °¿°¿

Here, the parameter p = 1, 0, 1 for a totally reflecting boundary, no boundary,

and a totally absorbing boundary, respectively.
b) For a boundary with 50% reflection we have

M ­° ­° y  y 2 °½ 1 ­° y  y0 2 °½½°
®exp® ¾  exp®
0
Cn ¾¾.
2 V n °¿ 2 2 V n °¿°¿
2 2
2S V n °̄ °̄ °̄

The ground concentration at y = 0 is given for this case by

3 M ­° y 2 ½°
Cn exp® 0 2 ¾.
2 2S V n °̄ 2 V n °¿

Hence, we obtain relation (6.55) multiplied with the factor 1.5. Therefore, the
concentration has a maximum at x*n = D xn / (U y02) = 1, and the maximum value
of C*n is given by C*n = Cn y0 / M = 1.5 / (2 S e)1/2.

Section 6.4
Exercise 6.4.1
a) This identity follows from

(1  a n 1 ) 2  (1  a n  2 ) 2    (1  a 2 ) 2  (1  a) 2
n  1  2 a  a 2   a n 1  a 2  a 4    a 2 n  2
§ 1  a n · 1  a 2n 1  a n 1  a 2n
n  1  2¨¨  1¸¸  1 n  2  .
© 1 a ¹ 1 a 1  a 1  a2
2

The first rewriting results from distributing the squared terms, and the second
rewriting makes use of the relations
1 an 1  a 2n
a n 1  a n  2    a1  a 0 , a 2n  2  a 2n    a 2  a 0 .
1 a 1 a2
54 Chapter 6

b) This identity follows from

a n 1 (1  a n 1 )  a n  2 (1  a n  2 )    a 2 (1  a 2 )  a (1  a )
a  a 2    a n 1  a 2  a 4    a 2 n  2
1  an § 1  a 2n · 1  a n 1  a 2n
 1  ¨¨  1¸¸  .
1 a © 1 a
2
¹ 1 a 1 a2

Exercise 6.4.2
a) The xn variance can be calculated by squaring and averaging expression (6.71a)
for fluctuations,
(1  a n ) 2
~ x0  v~0 ('t ) 2
~
2 2 2
xn 
(1  a) 2
> @
 D 't (1  a n 1 ) 2  (1  a n  2 ) 2    (1  a 2 ) 2  (1  a) 2 ,
where the correlation properties of ~
x0 , v~0 and Hk are applied. By replacing the
bracket term by the corresponding identity shown in exercise 6.4.1 we find

(1  a n ) 2 ª 1  a n 1  a 2n º
~ x0  v~0 ( 't ) 2
~
2 2 2
xn  D 't « n  2  »
(1  a ) 2 ¬ 1 a 1 a2 ¼
(1  a n ) 2 ª 1 an 't 1  a 2 n º
x0  v~0 ( 't ) 2
~ 2 2
 D « n 't  2 't  ».
(1  a) 2 ¬ 1 a 1 a 1 a ¼
This finding agrees with the result that has to be shown if we use 't / (1  a) = W.
b) The xn-vn covariance can be found by multiplying and averaging the relations
(6.71) for xn and vn fluctuations,
1 a n n
xnv~n
~ v~0 't
2
a 
1 a

D 't n 1
W
> @
a (1  a n 1 )  a n  2 (1  a n  2 )    a 2 (1  a 2 )  a (1  a) .

By replacing the bracket term by the corresponding identity shown in exercise

6.4.1 we find
1  a n n D 't ª 1  a n 1  a 2 n º
xnv~n
~ v~0 't
2
a  «  »
1 a W ¬ 1 a 1 a2 ¼
(6.112)
1  a n n D 't ª 1  a 2n º
v~0 't
2
a  « 1  a n
 ».
1 a W 1 a ¬ 1 a ¼

This finding agrees with the result that has to be shown if we use 't / (1  a) = W.
 Section 6.4 55

c) The vn variance can be derived by squaring and averaging (6.71b),

D 't
v~n a 2 n v~0  2 a 2 n  2    a 2  1
2 2

W
D 't 1  a 2 n D 't 1  a 2 n
a 2 n v~0  2 a 2 n v~0  2
2 2
.
W 1 a2 W 1 a 1 a
This finding agrees with the result that has to be shown if we use 't / (1  a) = W.

Exercise 6.4.3
a) As shown in Sect. 6.4.1, the Brownian motion model reduces to the diffusion
model (6.63) if 't = W, this means if a = 0. For n t 1, the expressions (6.70) and
(6.72) read
xn x0  W v0 , vn 0,
~ x0  W 2 v~0  D >n 't  W @
~ x0  ('t ) 2 v~0  (n  1) D 't ,
~
2 2 2 2 2
xn
D D
xnv~n
~ v~n
2
0, .
W 't
b) The consequences of these relations are <v0> = 0 and < v~0 > = D / 't.
2

c) By using these initial values we find

xn x0 , vn 0,
~
xn
2 ~ 2
x0  n D 't ,
D D
xnv~n
~ v~n
2
0, .
W 't

Exercise 6.4.4
a) The velocity model (6.59b) is a specific case of the linear first-order difference
equation (6.5). Therefore, according to the correlation function (6.24) implied
by (6.5), we obtain in combination with a = 1  't / W
m
§ 't · ~ 2
v~nv~n  m¨1  ¸ vn .
© W ¹
b) By using t = n 't and s = m 't, we find in the limit 't o 0
v~ (t ) v~ (t  s ) e  s / W v~ (t ) 2 ,

where relation (6.73) is applied.

c) The integral over the normalized correlation function reads
f v~ (t ) v~ (t  s ) f
s /W
³ ds ³ e ds W .
0 v~ (t ) 2 0
56 Chapter 6

d) This result means that the modeling parameter W represents the characteristic
lifetime of velocity correlations.

Exercise 6.4.5
a) The validity of this identity can be seen in the following way,
(1  a n 1 ) (1  a m a n 1 )  (1  a n  2 ) (1  a m a n  2 )   
 (1  a 2 ) (1  a m a 2 )  (1  a ) (1  a m a )
n  1  (1  a m ) a  a 2    a n 1  a m a 2  a 4    a 2 n  2
§ 1  a n · m § 1  a 2n · 1  an 1  a 2n
n  1  (1  a m )¨¨  1¸¸  a ¨¨  1¸¸ n  (1  a m )  am .
© 1 a © 1 a 1 a 1 a2
2
¹ ¹
b) We need the following expressions, which follow from (6.71a),
~
x ~nx  W v~ (1  a n ) 
0 0

>
 D 't H 0 (1  a n 1 )  H1 (1  a n  2 )    H n  3 (1  a 2 )  H n  2 (1  a ) . @
~
xn  m x0  W v~0 (1  a n  m ) 
~

>
 D 't H 0 (1  a n  m 1 )  H1 (1  a n  m  2 )    H n  m  3 (1  a 2 )  H n  m  2 (1  a) . @
By multiplying and averaging these fluctuations we find
~
xn ~
xn  m ~ 2 2
>
x0  W 2 v~0 (1  a n ) (1  a n  m )  D 't (1  a n 1 ) (1  a m a n 1 ) 
@
 (1  a n  2 ) (1  a m a n  2 )    (1  a 2 ) (1  a m a 2 )  (1  a ) (1  a m a) .
The use of the identity given in a) then leads to
~
xn ~ x0  W 2 v~0 (1  a n ) (1  a n  m )
~ 2 2
xn  m
ª 1 an 1  a 2n º
 D 't «n  (1  a m )  am ».
¬ 1 a 1 a2 ¼
In terms of 't / (1  a) = W we find
~
xn ~ x0  W 2 v~0 (1  a n ) (1  a n  m )
~ 2 2
xn  m
ª 1  a 2n º
 D «n 't  W (1  a m ) (1  a n )  W a m ».
¬ 1 a ¼
c) By using t = n 't and s = m 't we find in the limit 't o 0
~
x (t ) ~ x0  W 2 v~0 (1  e t / W ) (1  e  t / W e  s / W )
~ 2 2
x (t  s )
ªt 1  e  2t / W º
 D W «  (1  e  s / W ) (1  e  t / W )  e  s / W ».
¬W 2 ¼
 Section 6.5 57

d) For t o f we find the result

~
x (t ) ~
x (t  s ) D t .
e) The latter result is the expression obtained for < ~
x 2 (t ) > in the limit t o f. Thus
we recover in this way the finding that the normalized correlation function is
equal to one: see expression (6.32).

Section 6.5
Exercise 6.5.1
a) For the equilibrium case n o f we obtain (the subscript f is not written here)
P P2 ,

0 2P P 2 1  P  (V 2  P 2 't ) P 2 1  P
2
.
b) For the case 't o 0 we find
P P2 ,

2P P 2 1  P  V 2 P 2 1  P
2
0 .

c) In terms of <P> = <P2> we find

~ 1/ 2 2
P2 P2  P P 1 P .

Exercise 6.5.2
a) See figure 6.9.
b) See the figure: in a), b), and c) we used V = 0.5, 2, and 7.5 respectively. Figure
(b) is equal to Fig. 6.9a.
c) The higher V, the sooner the equilibrium value will be realized. For small V we
see only little modifications of the behavior of the deterministic logistic model:
see figure (a). For large V an overshooting of equilibrium values is possible: see
figure (c).
58 Chapter 6

Exercise 6.5.3
a) See figure 6.11.
~
b) According to exercise 6.5.1 c), we find Sn = < Pn 2 > / [<Pn> (1  <Pn>)] = 1.

Exercise 6.5.4
a) The beta PDF provides the relations
(a  2) (a  1) a
X3 ,
(a  b  2) (a  b  1) (a  b)

(a  3) (a  2) (a  1) a
X4 .
(a  b  3) (a  b  2) (a  b  1) (a  b)
Here, a and b are related to the mean and variance according to the relations
given in exercise 4.3.10,
ª X 1 X º ª X 1 X º
a X « ~  1», b 1  X «
~  1».
« X2 » « X2 »
¬ ¼ ¬ ¼
b) The consequences of this assumption are illustrated in the figure below, where
N = 105 realizations are used. The solid line shows the result of the population
model, and the dashed line shows the result that uses the closure via the beta
function. Attention is needed regarding the treatment of moments at t = 0. The
model parameters a and b are not defined for this case: the variance is zero due
to the lack of initial fluctuations. This problem can be overcome by calculating
the moments of third-order and fourth-order directly in terms of the given sharp
initial values.
c) It may be seen that the approximate solution of the equations for the mean and
variance works well: the mean value prediction is reasonable, and the standard
deviation is predicted very well.
 Section 6.5 59

Exercise 6.5.5
a) See figure (c).
b) See the figure. Here L = 102, L = 104, and L = 108 in (a), (b), and (c), respec-
tively.
c) The effect of L is reasonable: the time to extinction becomes smaller if realiza-
tions that are close to zero are sooner counted.
Chapter 7
Section 7.2
Exercise 7.2.1
We know that y0 = 90qC and ye = 21qC. According to equation (7.14), the coffee
temperature is then described by the equation
y 21  69 e  t / T .
We also know that the coffee temperature is 85qC after one minute, this means
85 21  69 e 1 / T .
The latter relation can be used to solve for the characteristic time scale T,
1 § 85  21 ·
 ln¨ ¸.
T © 69 ¹
Consequently, T = 13.3 minutes. By using this value for T we find for the time at
which the coffee temperature is 60ºC the relation
60 21  69 e t / 13.3 .
By solving for t,
t § 60  21 ·
 ln¨ ¸,
13.3 © 69 ¹
we find that it needs t = 7.6 minutes before it is possible to drink the coffee.

Exercise 7.2.2
a) We know that ye = 15qC. According to Eq. (7.14), the coffee temperature is then
described by the equation
y 15  e t / T ( y0  15).
We also know that the coffee temperature is 80qC after one minute and 72qC
after two minutes. Therefore, we have the relations
80 15  e 1 / T ( y0  15), 72 15  e 2 / T ( y0  15).
We replace y0  15 in the first relation according to the second relation,
65 57 e 1 / T e 2 / T 57 e1 / T .
By solving for T we find T = 1 / ln (65 / 57) = 7.6 minutes. The use of this value
in one of the relations involving y0 leads to the conclusion that the initial coffee
temperature was y0 = 89.1qC.
62 Chapter 7

b) According to a), the coffee temperature at any time is given by

y 15  e t / 7.6 (89.1  15) 15  74.1e  t / 7.6 .
Hence, the coffee temperature after five minutes is 53.4qC.

Exercise 7.2.3
a) This equation is given by
dy
 k ( y  ye ) 5 / 4 ,
dt
where k is a model parameter.
b) We write this equation
dy
 k dt.
( y  ye ) 5 / 4
By integrating both sides we find
4 4
 k t  C k t  ,
( y  ye )1 / 4 ( y0  ye )1 / 4
where the initial condition was used to determine C. We can write this equation
1
( y  ye )1 / 4 .
k t / 4  1 /( y0  ye )1 / 4
Hence, the solution reads
1
y ye  .
>k t / 4  1/( y 0  ye )1 / 4 @ 4

c) For y0 o ye we find
1 1
y ye  o ye  ye  y 0  ye y0 .
>k t / 4  1/( y 0  ye ) 1/ 4 4
@ >1/( y 0  ye )1 / 4 @ 4

Thus, the initial temperature will never change.

Exercise 7.2.4
We know that y0 = 90qC and ye = 21qC. According to the results of the exercise
7.2.3, the coffee temperature is then described by the equation
1 1
y 21  21  .
>k t / 4  1/(90  21) @ 1/ 4 4
>k t / 4  69 @
1 / 4 4

We also know that the coffee temperature is 85qC after one minute, this means

21  >k / 4  0.347@ .
4
85
 Section 7.2 63

This relation implies for k the value k = 0.0262. By using this value we find for
the time at which the coffee temperature is 60ºC the relation
1
60 21  .
>k t / 4  0.347@ 4
This condition can be written
1
k t / 4  0.347 .
391 / 4
Hence, it needs t = 8.1 minutes before it is possible to drink the coffee.

Exercise 7.2.5
a) We introduce the non-dimensional temperature T* = T / E and non-dimensional
time t* = E3 k t.
b) Then, the equation considered can be written
dT*
4
dt* .
1  T*
By integrating both sides we find
1 § 1  T* · 1 1 § 1  T*0 · 1
ln¨ ¸  arctan(T* ) t*  C t*  ln¨¨ ¸  arctan(T*0 ).
4 ¨© 1  T* ¸¹ 2 4 © 1  T*0 ¸¹ 2
Here, the initial condition was used to determine C. This relation can be written
1 § 1  T* 1  T*0 · 1 § T T ·
t* ln¨¨ ¸  arctan¨ * *0 ¸.
¸ ¨ 1 T T ¸
4 © 1  T*0 1  T* ¹ 2 © * *0 ¹

c) The dependence of T* on t* can be graphically illustrated by using this formula

for plotting t* in dependence on T*.

Exercise 7.2.6
a) We use equation (7.27) for the case considered,
dQ r
 (Q  V cin ).
dt V
Here, V = 50 m3, r = 0.02 m3 / min, cin = 0.1 C / V, and Q0 = C.
b) According to Eq. (7.28), the solution is given by
Q V cin  e  r t / V Q0  V cin .

c) For this case we have

Q 0.1C  e 0.02120 / 50 C  0.1C 0.1C  0.8578 C 0.9578 C.
64 Chapter 7

d) The chlorine mass in the pool is 50% of the initial mass if

0 .5 C 0.1C  e 0.02 t / 50 C  0.1C .
By solving this equation for t we find t = 2027.33 min. This value corresponds
to 33.79 hours.

Exercise 7.2.7
a) The equation for this case reads
dQ r
 Q.
dt V
Here, V = 50 m3, r = 0.02 m3 / min, and Q0 = 5 kg.
b) According to Eq. (7.28), the solution of this equation is given by

Q e  r t / V Q0 .

c) The concentration is given by

Q Q0
e r t / V .
V V
For this case we have the condition
5000
5 e  0.02 t / 50 .
50
This condition is satisfied if t = 7489.3 min. This value corresponds to 124.82
hours.

Section 7.3
Exercise 7.3.1
a) Newton’s Second Law provides m dv /dt = F =  m g j, this means dv /dt = g j.
The integration of this relation provides v =  g t j + v0, where the constant v0
represents the initial velocity vector.
b) The position vector satisfies the equation dr / dt = v =  g t j + v0. The integra-
tion of this relation provides r =  g t 2 /2 j + t v0 + r(0) =  g t 2 /2 j + t v0. The
last expression accounts for r(0) = 0. The application of the given expression
for v0 leads then to r = t v0 cos D i + (t v0 sin D  g t 2 /2) j.
c) The horizontal distance H traveled by the projectile is given by the component i
at the time tH for which the component j of r is zero. We consider tH > 0. Then,
tH is found to be given by tH = 2 v0 sin D / g. The corresponding i component
reads H = 2 v02 cos D sin D / g = v02 sin(2D) / g.
d) H has a maximum for D = S / 4. For this case we find sin(2D) = 1, which implies
H = v02 / g.
 Section 7.3 65

Exercise 7.3.2
a) The characteristic equation reads for this case
r 2  ( A  4) r  A  3 0.
This equation is solved by
A 4 1 A 4 1
r1, 2  r ( A  4) 2  4( A  3)  r A2  8 A  16  4 A  12
2 2 2 2
A4 1 A 4 1 A 4 A 2
 r A2  4 A  4  r A2  r .
2 2 2 2 2 2
The roots of this equation are given by r1 = 1 and r2 =  3  A. According to
Eq. (7.57), the solution is then given by
y '0  ( A  3) y0  t y '0  y0  ( A  3) t A  3 t 1  ( A  3) t
y (t ) e  e e  e .
A2 A2 A 2 A 2
b) The solutions are bounded as t o f if  3 < A.
c) The first-order derivative of y(t) is given by
dy A  3  t A  3  ( A  3) t A  3  ( A  3) t  t
 e  e e e .
dt A 2 A 2 A 2
Hence, the critical t value is t = 0.
d) The second-order derivative of y(t) is given by
d2y A3
 ( A  3) e  ( A  3) t  e  t .
dt 2 A 2
At t = 0 the second-order derivative is given by
d2y A3
 ( A  2) ( A  3) .
dt 2 A 2
For the range 0 < A + 3 considered we have, therefore, a local maximum at the
critical point t = 0.

Exercise 7.3.3
a) The characteristic equation reads for this case r2 + 4 r + 4 = (r + 2)2 = 0. This
equation is solved by r1 = r2 = 2. According to the Eq. (7.64), the solution for
this case is given by
y (t ) t e 2 t y '0 (1  2 t ) e 2 t y0 e 2 t A t  1  2 t e 2 t 1  (2  A) t .

b) The first-order derivative of y(t) is given by

dy
e  2 t  2 [1  (2  A) t ]  2  A e  2 t A  2(2  A) t .
dt
Hence, the critical t value is t = A / [2 (A + 2)].
66 Chapter 7

c) The second-order derivative of y(t) is given by

d2y
2e  2 t A  2(2  A) t  2  A 4 e  2 t A  1  (2  A) t .
dt 2
At t = A / [2 (A + 2)] the second-order derivative reads
d2y
4 e  A /( A  2 ) A  1  A / 2 4 e  A /( A  2) A / 2  1 .
dt 2
Therefore, for A < 2 we find a local minimum, for A > 2 we find a local
maximum, and for A = 2 we do not have an extreme value: see the expression
obtained for y(t).

Exercise 7.3.4
We know the roots r1 = 3 and r2 = 1 of the characteristic equation, which can be
written (r  r1) (r  r2) = 0. Therefore, the characteristic equation is given by the
relation (r 3) (r + 1) = r2 2 r 3 = 0. The differential equation related to this
characteristic equation reads
d2y dy
2
 2  3y 0.
dt dt

Exercise 7.3.5
a) The characteristic equation reads for this case
r2  4r  8 0.
This equation is solved by
r1, 2 2 r 4  8 2 r 2 i.
Hence, we have rS = b / (2a) = 2 and rD* = 2. The solution (7.68) then reads
ª sin(2t ) º  2 t
y (t ) «cos(2t )  2 2 » e y0 >cos(2t )  sin(2t )@e 2 t y0 .
¬ ¼
b) The latter solution can be also written as
cos(2t  S / 4) cos(2t  S / 4)  2 t
y (t ) e  2 t y0 e y0 .
2 /2 cos(S / 4)

Exercise 7.3.6
a) The function cos(rD* t  G) has the same value for the two arguments rD* t  G
and rD* t  G + 2 S = rD* (t + 2 S / rD*)  G. Thus, the time between successive
maxima is W = 2 S / rD*.
b) The ratio R of the displacements of any two successive maxima is given by R =
exp(rS t) / exp[rS (t + W)] = exp(rS W).
 Section 7.4 67

c) The combination of the expression for R with the definition rS =  J / (2 m) leads

to ln R = J W / (2 m). Hence, J can be calculated via J = 2 m ln R / W.

Exercise 7.3.7
a) The equation considered represents a second-order linear differential equation
of the type (7.45), where a = L, b = R, and c = 1 / C. For the given data values
we find b2  4 a c = R2  4 L / C = 202  8 / 0.01 F = 400. The solution for this
case is given by (7.68),
sin(5 t )  5 t ª sin(5 t ) º  5 t
Q(t ) e Q'0  «cos(5 t )  5» e Q0 5>cos(5 t )  sin(5 t )@e  5 t .
5 ¬ 5 ¼
Here, we applied rD* = (400)1/2 / 4 = 5 and rS = b / (2a) = 20 / 4 = 5.
b) The circuit is critically damped for R2 = 4 L / C = 800, i.e., for R = (800)1/2.
c) The solution for the critical damping case follows from (7.64),

Q (t ) t e 50 t
Q'0  1  50 t e  50 t
Q0 5 1  50 t e  50 t
,

where rS = (800)1/2 / (2a) = (800)1/2 / 4 = (50)1/2 is used.

Section 7.4
Exercise 7.4.1
a) The mean of the PODF (7.106) can be determined by means of the formula
f
1 f t  tc f
x
³ (t  tc ) g (t ) dt ³ dt W ³ dx 0.
f 4W  f § t  tc · 2
 f cosh ( x )
cosh 2 ¨ ¸
© 2W ¹
The integral is rewritten by introducing x = (t  tc) /(2W). By accounting for the
normalization property of g(t), this relation shows that tc is the mean of g(t).
b) The variance of the PODF (7.106) can be determined by means of the formula
f
1 f (t  tc ) 2 f
x2 2W 2 S 2 W 2S 2
³ (t  tc ) g (t ) dt
2
³ dt 2W 2 ³ dx .
f 4W  f 2 § t  tc ·
2
 f cosh ( x ) 6 3
cosh ¨ ¸
© 2W ¹
Here, we used x = (t  tc) /(2W) and the integral given in the exercise. This result
shows that the standard deviation is given by W S /31/2.
c) The skewness is zero, because the central moment of third order is zero,
f
1 f (t  tc )3 f
x3
³ (t  tc ) g (t ) dt
3
³ dt 4W 3 ³ dx 0.
f 4W  f 2 § t  tc ·
2
 f cosh ( x )
cosh ¨ ¸
© 2W ¹
68 Chapter 7

The central moment of fourth order is given by

f
1 f (t  tc ) 4 f
x4 7 4
³ (t  tc ) g (t ) dt
4
³ dt 8W 4 ³ dx 8W 4 S .
f 4W  f § t  tc · 2
 f cosh ( x ) 120
cosh 2 ¨ ¸
© 2W ¹
Therefore, the flatness is given by
8W 4 7S 4 / 120 7 ˜8 ˜9 21
m3 4.2.
W 4S 4 /9 120 5

Exercise 7.4.2
a) To calculate the mean of g we consider
f
1 f
t  tc 2W f
x
³ (t  tc ) g (t ) dt ³
2S W  f
dt
S
³ dx,
f 1  x
2 2
f § t  tc ·
1 ¨ ¸
© 2W ¹
where x = (t  tc) / (2W) is applied. We may split the latter integral into two
terms: the integral from negative infinity to zero, and the integral from zero to
infinity. These two integrals do not exist because the integral value ln(1 + x2) / 2
diverges for large x. Therefore, the mean of the Cauchy PODF does not exist.
The variance or higher-order central moments take reference to the mean. Thus,
these moments are not defined, too.
b) However, the second moment about the center tc can be calculated,
f
1 f
(t  tc ) 2 4W 2 f
x2
³ (t  tc ) g (t ) dt
2
³
2S W  f 2
dt ³
S f 1  x2
dx
f § t  tc ·
1 ¨ ¸
© 2W ¹
4W 2 f § 1 · 4W 2 f
³ ¨1 
S f © 1  x2 ¹
¸ dx
S
[ x  arctan( x) ] f o f .

It turns out that the second moment about tc is infinite.

Exercise 7.4.3
a) This model can be written
dP 1§ P · 1§ P· 1W E § P ·
¨1  ¸ P  E P ¨1  W E  ¸ P ¨1  ¸ P.
dt W© K¹ W© K¹ W ¨© K (1  W E ) ¸¹
This model corresponds to the structure of the logistic model. Instead of W and
K in the logistic model we have here W / (1  W E) and K (1  W E), respectively.
b) Equation (7.102) taken for L = 0 shows that the logistic model is solved by
K
P .
1  ( K / P0  1) e  (t  t 0 ) / W
 Section 7.4 69

By replacing W and K by W / (1  W E) and K (1  W E), respectively, we find the

solution of Schaefer's model to be given by
(1  W E ) K
P .
1  [(1  W E ) K / P0  1] e  (1W E ) (t  t 0 ) / W
c) A nonzero value of E decreases the carrying capacity K by the factor 1  W E,
and it results in a growth rate 1/W  E, which is smaller than the growth rate 1/W
of the logistic model. These two effects are plausible: the carrying capacity and
growth rate have to become smaller due to predation.

Exercise 7.4.4
a) In terms of the abbreviations P* = P / K, t* = r t, and D = 1 / 4  H / (r K) we find
2 2
dP* § 1· 1 H § 1·
¨ P*  ¸   ¨ P*  ¸  D .
dt* © 2¹ 4 rK © 2¹
b) The cases H > r K / 4, H = r K / 4, and H < r K / 4 correspond to D < 0, D = 0, and
D > 0, respectively. For D < 0, an equilibrium state cannot be established: the
solution will always decrease. For D = 0, there is one equilibrium state P* = 0.5.
For D > 0, there are two potential equilibrium states P* = 0.5 r D1/2.
c) The non-dimensional equation can be written
dP* § 1 ·§ 1 · §1 ·§ 1 ·
¨ P*   D 1 / 2 ¸ ¨ P*   D 1 / 2 ¸ ¨  D 1 / 2  P* ¸ ¨  D 1 / 2  P* ¸.
dt* © 2 ¹© 2 ¹ ©2 ¹© 2 ¹
We may have the following cases:
­ 0 if P*0  0.5  D 1 / 2
dP* °°
®! 0 if 0.5  D  P*0  0.5  D .
1/ 2 1/ 2

dt* °
°¯ 0 if 0.5  D  P*0
1/ 2

Hence, the equilibrium point P* = 0.5  D1/2 will not be realized (except for P*0
= 0.5  D1/2), but the equilibrium point P* = 0.5 + D1/2 can be realized.

Exercise 7.4.5
a) By differentiating y = ln(P / K) and applying the Gompertz equation we find for
y the equation
dy K 1 dP 1 §K· 1 §P· y
ln¨ ¸  ln¨ ¸  .
dt P K dt W ©P¹ W ©K¹ W
t/W
This equation is solved by y = y0 e . Hence, in terms of P we find the relation
§P· §P ·
ln¨ ¸ ln¨ 0 ¸ e  t / W .
©K¹ ©K¹
70 Chapter 7

This relation leads to the following solution to the Gompertz equation,

P K exp e t / W ln( P0 / K ) .
b) The equilibrium solution is given by P = K.
c) We have to show that
dP 1 §K· dP 1§ P·
ln¨ ¸ P t ¨1  ¸ P.
dt W P
© ¹ dt W© K¹
This relation is valid if
§K· P
ln¨ ¸ t 1  .
©P¹ K
By introducing x = P /K and f(x) = 1 + ln x  x, we have to show that f(x) d 0 for
x > 0. For doing this we calculate the first and second order derivative of f(x),
df 1 d2 f 1
 1,  .
dx x dx 2 x2
Hence, f(x) has a local maximum at x = 1. The local maximum is zero, which
means that f(x) d 0.

Exercise 7.4.6
a) By integrating the equation we find
ln P  Pc D ln t  tc  C.
This relation can be written
D
P  Pc C1 t  tc ,
or
D D
P  Pc rC1 t  tc C2 t  t c .
The constant C2 can be found in terms of the initial condition,
P0  Pc
C2 D
.
t0  t c
Therefore, the solution to this differential equation is given by
P0  Pc D
P Pc  D
t  tc .
t 0  tc
b) This conclusion would be incorrect. The assumption of self-similarity used for
the derivation of the model considered here does not have a stronger theoretical
background than the exponential growth model and the logistic model: in each
case we have a model assumption that may apply to a certain process or not.
 Section 7.5 71

Section 7.5
Exercise 7.5.1
a) The limit WM o 0 leads to the following equation,
dy 1 t y (t )  ye t
dt
 ³ lim P (t  s)> y ( s)  ye @ds
T f W M o0

T
³ W lim
o0
P (t  s) ds
f M

y (t )  ye
 .
T
The fact that y(s)  ye can be replaced by y(t)  ye can be seen in terms of the
properties of the memory function: P(t  s) is always zero for WM o 0 except at
s = t (i.e., P(t  s) behaves like a delta function). The last expression arises from
the normalization property (7.121), which applies to all WM.
b) We differentiate the differential equation considered,
1 t dP (t  s )
d2y
 ³ >y( s)  ye @ds  P (0) y(t )  ye
dt 2 T f dt T
1 t
1 y (t )  ye
³ P (t  s )> y ( s )  ye @ds 
T W M f WM T
1 dy 1 y  ye
  .
W M dt W M T
c) By multiplying this equation with WM and taking the limit WM o 0 we find
dy y (t )  ye
 .
dt T

Exercise 7.5.2
a) This equation has the structure of the second-order linear differential equation
(7.45) with constant coefficients. In particular, we have here a = 1, b = 1 / WM,
and c = 1 / (WM T). According to Eq. (7.57), where we account for the constant
ye, we find the solution to be given by
y '0  r2 ( y0  ye ) r1 t y '0  r1 ( y0  ye ) r 2 t
y (t )  ye e  e .
r1  r2 r1  r2
According to the relations (7.50), the eigenvalues are given by

1ª 1 1 4 º 1 ª WM º
r1 «   » « 1  1  4 »,
2 ¬« W M W M W M T ¼»
2
2W M ¬« T ¼»
1ª 1 1 4 º 1 ª WM º
r2 «   » « 1  1  4 ».
2 «¬ W M WM 2
W M T »¼ 2W M ¬« T ¼»
72 Chapter 7

b) For WM << T the eigenvalues become

1 ª WM º 1 ª § WM ·º 1
r1 « 1  1  4 » «  1  ¨1  2 ¸»  ,
2W M ¬« T »¼ 2W M ¬ © T ¹¼ T

1 ª WM º 1 ª § WM ·º 1
r2 « 1  1  4 » «  1  ¨1  2 ¸»  ,
2W M ¬« T ¼» 2W M ¬ © T ¹¼ WM

where (1 r x)1/2 = 1 r x / 2 is used.

c) For WM o 0 we find the solution
y (t )  ye ( y 0  ye ) e  t / T ,
where we accounted for r2 o f and r1 = 1 / T.

Exercise 7.5.3
We consider the memory function P(t  s) = G(t  s  r). For this case, the differ-
ential equation becomes
dy 1 t 1 t
 ³ P (t  s )> y ( s )  ye @ds  ³ G (t  s  r )> y ( s )  ye @ds
dt T f T f
y (t  r )  ye t
y (t  r )  ye
 ³ G (t  s  r ) ds  ,
T f T
where we applied the properties of delta functions. Therefore, the choice P(t  s) =
G(t  s  r) leads to the differential equation with the discrete delay r.
Chapter 8
Section 8.2
Exercise 8.2.1
To show the consistency of the Fokker-Planck equation (8.21) we integrate it over
the sample space x,
f
wf ( x, t ) w D (1) ( x, t ) f ( x, t )
f f
w 2 D ( 2 ) ( x, t ) f ( x, t )
³ dx ³ dx  ³ dx.
f wt f wx f wx 2
By writing the time derivative in front of the integral and using the fundamental
theorem of calculus, this equation can be written
f
w f f § w D ( 2 ) ( x, t ) f ( x, t ) ·
³ f ( x, t ) dx  D (1) ( x, t ) f ( x, t )  f  ¨¨ ¸ .
¸
wt  f © wx ¹f
The left-hand side vanishes because f(x, t) is normalized. The terms on the right-
hand side do also vanish because f(x, t) and its change are zero at infinity.

Exercise 8.2.2
a) The asymptotic solution is determined by
w ª w D ( 2 ) ( x) f ( x) º
«  D ( x ) f ( x, t ) 
(1)
» 0.
wx ¬ wx ¼
The coefficients have to be independent of t in this limit (otherwise, the PDF
would depend on t). The bracket term has to be equal to a constant C,
w D ( 2) ( x) f ( x)
 D (1) ( x) f ( x, t )  C.
wx
To evaluate C we may consider this relation for |x| o f. The PDF f(x) and its
change are equal to zero for |x| o f, which shows that C = 0. Hence, we have
d D ( 2) ( x) f ( x)
D (1) ( x) f ( x).
dx
We replaced the partial by a regular derivative. This equation can be written
d f ( x) ª (1) d D ( 2 ) ( x) º
D ( 2) ( x) « D ( x )  » f ( x),
dx ¬ dx ¼
or,
d ln[ f ( x)] 1 ª (1) d D ( 2) ( x) º
« D ( x )  ».
dx D ( 2) ( x) ¬ dx ¼
74 Chapter 8

The integration of this relation provides

1 ª d D ( 2) ( x) º
³ D ( 2 ) ( x) « D ( x)  » dx  C ,
(1)
ln[ f ( x)]
¬ dx ¼
where C is a constant. Hence, the PDF is given by
­° 1 ª (1) d D ( 2 ) ( x) º ½°
f ( x) C0 exp®³ ( 2 ) « D ( x)  » dx ¾,
°̄ D ( x) ¬ dx ¼ °¿
where C0 is another constant.
b) We consider the case D(1) = a x and D(2) = b, where a and b are constants. The
corresponding PDF
­ a ½ ­ a 2½
f ( x) C0 exp® ³ x dx ¾ C1 exp® x ¾
¯ b ¿ ¯ 2b ¿
approaches zero for |x| o f. C1 is another constant.
c) We consider the case D(1) = a x2 and D(2) = b, where a and b are constants. The
corresponding PDF
­ a ½ ­ a 3½
f ( x) C0 exp® ³ x 2 dx ¾ C1 exp® x ¾
¯ b ¿ ¯ 3b ¿
diverges (depending on a / b) for x o f or x o f. C1 is another constant.

Exercise 8.2.3
The PDF has the derivatives
df xP d2 f ª ( x  P )2 1 º
 f, «  2» f.
dx V 2
dx 2 ¬ V
4
V ¼
The use of these derivatives in the asymptotic Fokker-Planck equation leads to
w D (1) ( x) w f ( x) w 2 f ( x)
0  f ( x)  D (1) ( x)  D ( 2)
wx wx wx 2
­° w D (1) ( x) xP ( 2) ª ( x  P )
2
1 º ½°
®   D (1)
( x )  D «  2 » ¾ f ( x)
°̄ wx V 2
¬ V
4
V ¼ °¿
­ w D (1) ( x) D (1) ( x) x  P ( x  P ) 2 1 ½ ( 2 )
®    2 ¾ D f ( x).
¯ wx D
( 2)
D ( 2) V 2 V4 V ¿
To balance the quadratic term we set
D (1) ( x) xP
 .
D ( 2) V2
This condition does also lead to the balance of terms that are independent of x.
 Section 8.2 75

Hence, this condition ensures that the normal PDF is an asymptotic solution of the
Fokker-Planck equation. There is no condition on the constant D(2).

Exercise 8.2.4
For D(1)(x, t) = a x2, Eq. (8.25) reads
d X
a X 2 .
dt
This equation is unclosed. Therefore, a solution cannot be found.

Exercise 8.2.5
a) For D(1)(x, t) = a x, where a and D(2) are constants, Eq. (8.32) reads
~
d X2 ~
2a X 2  2 D ( 2 ) .
dt
This equation can be written
d § ~2 D ( 2) · § ~ D ( 2) ·
¨ X  ¸ 2a ¨¨ X 2  ¸.
dt ¨© a ¸¹ © a ¸¹
The solution of this equation is given by
~ D ( 2) § ~ 2 D ( 2) ·  2 at
X2  ¨¨ X (0)  ¸e .
a © a ¸¹
b) The asymptotic variance is then given by
~ D ( 2)
X2 .
a
c) By using measurements for the equilibrium variance we obtain a linear relation
between the model parameters D(2) and a.

Exercise 8.2.6
a) We multiply the Fokker-Planck equation (8.21) with x3 and integrate it,
wf x, t wD (1) x, t f x, t w 2 D ( 2 ) x, t f x, t
³x
3
dx ³ x3 dx  ³ x 3 dx .
wt wx wx 2
This relation can be written
w 3 wx 3 D (1) x, t f x, t wx 3 (1)
³ x f x, t dx ³ dx  ³ D x, t f x, t dx
wt wx wx
w ª wD ( 2 ) x, t f x, t º wx 3 wD ( 2 ) x, t f x, t
 ³ « x3 » dx  ³ wx dx .
wx ¬ wx ¼ wx
76 Chapter 8

The integral on the left-hand side is equal to <X 3>. The first and the third term
on the right-hand side disappear. Thus, we obtain
w X3 wx 2 D ( 2 ) x, t f x, t wx 2 ( 2 )
3 X 2 D (1)  3³ dx  3³ D x, t f x, t dx
wt wx wx
3 X 2 D (1)  6 X D ( 2 ) .
The central moment of third-order is given by
~ 3 2 3
X3 X X X3 3 X2 X 3 X X  X
3
X3 3 X 2 X 2 X .

By differentiating this expressions and making use of the evolution equations

for <X 3>, <X 2>, and <X> we find (the partial derivative by t can be replaced by
the regular derivative)
~
d X3 d X3 d X2 d X 2 d X
3 X 3 X 2 6 X
dt dt dt dt dt
2
3 X 2 D (1)  6 X D ( 2 )  6 X X D (1)  D ( 2 )  3 X 2 D (1)  6 X D (1)
~~ 2
6 X D ( 2)  3 X 2 D (1)  X 2 D (1)  2 X X D (1)  2 X D (1)
~~ ~ ~~
6 X D ( 2)  3 X 2  X 2 D (1)  2 X X D (1)
~~ ~ ~
6 X D ( 2)  3 X 2  2 X X  X 2 D (1)
~~ ~ ~~ ~ ~
6 X D ( 2)  3 X 2  2 X X  X 2 D (1) 6 X D ( 2 )  3 X 2 D (1) .

b) For D(1)(x, t) = a x and a constant D(2) this equation reads

~
d X3 ~
3a X 3 .
dt
The solution is given by
~ ~
X3 X 3 (0) e 3at .

Section 8.3
Exercise 8.3.1
a) By using the two-point PDF definition f(x, t; x', t') = <G[x  X(t)] G[x'  X(t')]>,
the one-point PDF can be written
f ( x, t ) ³ f ( x, t; x' , t ' ) dx'.
 Section 8.3 77

This relation can be proven by using the definition of the two-point PDF. We
may replace the one-point PDF in the PDF equation (8.21) by its relation to the
two-point PDF,

ª wf ( x, t ; x' , t ' ) w D (1) ( x' , t ' ) f ( x, t ; x' , t ' ) w 2 D ( 2 ) ( x' , t ' ) f ( x, t ; x' , t ' ) º
³«   » dx 0.
¬ wt ' wx' wx'2 ¼
This equation can only be satisfied if the bracket term is equal to zero. Hence,
the two-point PDF satisfies the Fokker-Planck equation (8.21),
wf ( x, t ; x' , t ' ) w D (1) ( x' , t ' ) f ( x, t ; x' , t ' ) w 2 D ( 2 ) ( x' , t ' ) f ( x, t ; x' , t ' )
  0.
wt ' wx' wx'2
b) By using the definition f(x, t; x', t') = <G[x  X(t)] G[x'  X(t')]> we find

³ ³ x  X (t ) x' X (t ' ) f ( x, t; x' , t ' ) dx dx'

³ ³ x  X (t ) x' X (t ' ) G x  X (t ) G x' X (t ' ) dx dx'

³ ³ X (t )  X (t ) X (t ' )  X (t ' ) G x  X (t ) G x' X (t ' ) dx dx'

~ ~
X (t )  X (t ) X (t ' )  X (t ' ) X (t ) X (t ' ) .

c) We integrate the Fokker-Planck equation multiplied with the variables required

for the calculation of the correlation function,
­ wf ( x, t ; x' , t  r )
0 ³ ³ x  X (t ) x' X (t  r ) ® wr

¯
w D ( x ' , t  r ) f ( x, t ; x ' , t  r ) w 2 D ( 2 ) ( x ' , t  r ) f ( x, t ; x ' , t  r ) ½
(1)
  ¾ dx dx'.
wx' wx'2 ¿
This relation can be written 0 = I1 + I2 + I3, where
w
I1
wr
³ ³ x  X (t ) x' X (t  r ) f ( x, t ; x' , t  r ) dx dx'

w x  X (t ) x' X (t  r )
³³ f ( x, t ; x' , t  r ) dx dx'
wr
w x  X (t ) x' X (t  r ) D (1) ( x' , t  r ) f ( x, t ; x' , t  r )
I2 ³³ wx'
dx dx'

w x  X (t ) x' X (t  r )
³³ D (1) ( x' , t  r ) f ( x, t ; x' , t  r ) dx dx'
wx'
78 Chapter 8

I3 ³ ³
w x  X (t ) > @
x' X (t  r ) w D ( 2 ) ( x' , t  r ) f ( x, t ; x' , t  r ) / wx
dx dx'
wx'
w x  X (t )
x' X (t  r ) wD ( 2 ) ( x' , t  r ) f ( x, t ; x' , t  r )
³³ dx dx'.
wx' wx'
All the integrals vanish except the first integral of I1 and the last integral of I2.
Thus, we obtain
d ~ ~
X (t ) X (t  r ) ³ ³ x  X (t ) D ( x' , t  r ) f ( x, t; x' , t  r ) dx dx'
(1)

dr

³ ³ x  X (t ) D ( x' , t  r ) G x  X (t ) G x' X (t  r ) dx dx'

(1)

³ ³ X (t )  X (t ) D X (t  r ), t  r G x  X (t ) G x' X (t  r ) dx dx'
(1)

~ ~ ~
X (t ) D (1) X (t  r ), t  r X (t ) D (1) X (t  r ), t  r .

Exercise 8.3.2
This equation represents a specific case of the PDF equation (8.21). By following
the solution of exercise 8.2.2 we find
­ G ½ ­G 2 ½
f ( x) C0 exp®³ x  X dx ¾ C1 exp® x X ¾,
¯ D ¿ ¯ 2D ¿
where C1 is another constant. By choosing C1 such that the PDF integrates to one
we see that the PDF represents a normal PDF with mean <X> and variance D / G.

Exercise 8.3.3
The validity of these initial conditions can be seen by using D(t0) = x0 and E(t0) = 0
for the conditional PDF f(x, t | x0, t0),
1 ­ ( x  x0 ) 2 ½
f ( x, t0 x0 , t0 ) lim exp® ¾ G x  x0 .
E o0 2S E ¯ 2E ¿

Exercise 8.3.4
a) The consistency with the initial conditions can be seen by setting t = t0. The fact
that the given expressions satisfy Eqs. (8.48) can be proven by means of the
Leibnitz integral rule

d b( x) b( x)
wg ( x, t ) db da
³ g ( x, t ) dt ³ dt  g x, b( x)  g x, a ( x ) .
dx a ( x ) a( x) wx dx dx
 Section 8.3 79

By adopting this rule we find

dD ­t  s  t 0 ½
> @
t
G (t ) ³ exp® ³ G (r ) dr ¾ G ( s ) x0  X  F ( s ) ds
dt t0 ¯ t0 ¿
­t 0 ½
>
 exp® ³ G (r ) dr ¾ G (t ) x0  X  F (t ) @
¯t 0 ¿
> @
G (t ) D  x0  G (t ) x0  X  F (t ) G (t ) D  X  F (t ). > @
dE t
­ t  s  t0 ½ ­ t0 ½
2 G (t ) ³ exp®2 ³ G ( r ) dr ¾ D( s ) ds  2 exp®2 ³ G (r ) dr ¾ D(t )
dt t0 ¯ t0 ¿ ¯ t0 ¿
2 G (t ) E  2 D (t ).
b) For the case that G and D are independent of t and F = 0 we find
t
D x0  G x0  G X ³ eG (t  s ) ds x0  x0  X 1  eG ( t  t 0 ) .
t0

t
D
E 2 D ³ e 2G (t  s ) ds  1  e 2G ( t  t 0 ) .
t0 G

Exercise 8.3.5
a) The parameter E satisfies the equation
dE
2 G (t ) E ,
dt
combined with the initial condition E(t0) = 0. The solution of this equation is
E(t) = 0.
b) The solution for this case is given by Eq. (8.43),

1 ­ ( x  D )2 ½
f ( x, t ) ³ exp® ¾ f ( x0 , t0 ) dx0 .
2S E ¯ 2E ¿

For E o 0, this solution becomes

1 ­ ( x  D )2 ½
f ( x, t ) lim ³ exp®
2E ¿
¾ f ( x0 , t0 ) dx0 ³ G x  D f ( x0 , t0 ) dx0 .
E o0 2S E ¯
Here, D satisfies the equation
dD
G (t ) D  X  F (t )
dt
combined with the initial condition D(t0) = x0.
80 Chapter 8

c) For f(x0, t0) = G(x0  D0) we find

f ( x, t ) G x  D ³ f ( x0 , t0 ) dx0 G x D .
Here, D satisfies the equation
dD
G (t ) D  X  F (t )
dt
combined with the initial condition D(t0) = D0 (i.e., D is independent of x0).
d) The process X(t) is given in this case by the non-random process X(t) = D(t).
Here, D satisfies the ordinary differential equation
dD
G (t ) D  X  F (t )
dt
combined with the initial condition D(t0) = D0.

Exercise 8.3.6
a) The initial concentration is given by C(y0, 0) = M G(y0  H).
b) The solution of the concentration equation reads

1 ­ ( y  D )2 ½
C ( y, t ) ³ exp® ¾ C ( y0 , t0 ) dy0 .
2S E ¯ 2E ¿

According to Eqs. (8.48), D and E are given here by D = y0 and E = 2 D t. With

these values for D and E and the initial concentration we obtain

M ­ ( y  y0 ) 2 ½ M ­ ( y  H )2 ½
C ( y, t ) ³ exp® ¾ G ( y0  H ) dy0 exp® ¾.
2S E ¯ 2E ¿ 4S D t ¯ 4 Dt ¿

Section 8.4
Exercise 8.4.1
a) By differentiating (8.87) by r we obtain
~ ~
d X (t ) X (t  r ) 1 ~ ~
 X (t ) X (t  r ) .
dr W
This equation is the correlation equation (8.33) for D(1) = a = (x  <X>) / W.
b) By differentiating (8.88) we find
~
d X 2 (t ) 2§ ~ 2 D · W
2t
2§ ~ D· 2 ~ D
 ¨¨ X 0  ¸¸ e  ¨¨ X 2  ¸¸  X 2  2 .
dt W© 2W ¹ W© 2W ¹ W W
The latter equation is equivalent to the variance equation (8.32) combined with
D(1) = a = (x  <X>) / W and D(2) = b2 / 2 = D / [2 W2].
 Section 8.4 81

Exercise 8.4.2
a) One way is to use the variance expression
D § ~2 D · W
2t
~
X 2 (t )  ¨¨ X 0  ¸¸ e .
2W © 2W ¹
~
The equilibrium variance < X 2 >e = D / (2W) provides one relation between D
and W. The time scale W can be chosen such that the evolution of the modeled
variance agrees with the measured variance evolution.
b) First of all, it is needed to show that the process considered has an equilibrium
PDF that is a normal PDF. In addition, the normalized correlation function has
to be an exponential function according to (8.87).

Exercise 8.4.3
a) We can use the variance equation (8.32) combined with D(1) = a = (I  <I>) / W
and D(2) = b2 / 2 = c2 / 2. The variance equation is then found to be given by
~ ~
d I2 I2 2§ ~ W c2 ·
2  c 2  ¨¨ I 2  ¸.
dt W W© 2 ¸¹
The solution of this equation reads
~ § ~
W c2 W c 2 ·  2t / W
I2  ¨¨ I 2 (0)  ¸e .
2 © 2 ¸¹
~
b) The equation for ) = (I  <I>) / < I 2 >1/2 reads
~ ~ ~2 ~
 I / W  c dW / dt I  2 I / W  c
2
d) c2 I c dW
   .
dt ~ 1/ 2 ~ 3/ 2 ~2 3/ 2 ~ 2 1 / 2 dt
I2 2 I2 2I I

c) The problem with using a zero c is that we have d) / dt in this case, this means
)(t) = )(0). This means that the shape of the PDF of I will never change (it is
given by the initial PDF), only the variance of I changes as given in a).
d) The problem with a nonzero c is that the noise term may lead to the appearance
of I values that are not inside the allowed region 0 d I d 1 of I.

Exercise 8.4.4
a) According to (8.40) and (8.41), the evolution equation for f(T, t| T ', t ') reads

wf (T , t T ' , t ' ) 1 wT  I f (T , t T ' , t ' ) c 2 w f (T , t T ' , t ' )

2
  0.
wt W wT 2 wT 2
The initial condition is given by f(T, t '| T ', t ') = G(T  T ').
82 Chapter 8

b) According to (8.43), the solution of this equation is given by

1 ­ (T  D ) 2 ½
f (T , t ) ³ exp® ¾ f (T ' , t ' ) dT '.
2S E ¯ 2E ¿

The parameters D and E satisfy Eqs. (8.48),

dD 1 dE 2
 D I ,  E  c2 .
dt W dt W
The initial conditions are given by D(t ') = T ' and E(t ') = 0. The equations for D
and E are solved by the expressions
W c 2 W c 2  2t / W
D I  T ' I e  t / W , E  e .
2 2
c) For the case f(T ', t ') = G(T '  <I>) we have D(t) = <I>. Hence, f(T, t ') reads

1 ­° (T  I ) 2 °½
f (T , t ) exp® ¾.
2S E °̄ 2E °¿

d) This is a normal PDF. In particular, it is a delta function at t = 0. The mean <I>

is constant. The variance changes from zero to the equilibrium value W c2 / 2.

Exercise 8.4.5
a) According to (8.25) and (8.32) combined with D(1) = a = P P (1  P) and D(2) =
b2 / 2 = [V P (1  P)]2 / 2, the equations for the mean and variance of P read
d P
P P(1  P) .
dt
~
d P2 ~
2 P P P (1  P)  P (1  P)  V 2 P 2 (1  P ) 2
dt
~
2P P P(1  P)  V 2 P 2 (1  P) 2

b) The equations (6.94) and (6.96) for the mean and variance are given by

1  P 't Pn 1  P 't Pn 1 .
2
Pn

Pn 1  2 P 't Pn 1 1  Pn 1  P 2 ('t ) 2 Pn 1 1  Pn 1
2 2 2 2 2
Pn

 V 2 't Pn 1 1  Pn 1
2 2
.
For 't o 0 the equation for the mean is given by
d P
P P(1  P) .
dt
 Section 8.5 83

This equation agrees with the equation obtained in a). For 't o 0 the equation
for the second-order moment is given by
d P2
2 P P 2 (1  P)  V 2 P 2 (1  P ) 2 .
dt
This equation implies the variance equation
~
d P2 d P2 d P
2
d P
 2 P P 2 (1  P)  V 2 P 2 (1  P ) 2  2 P
dt dt dt dt
2 P P 2 (1  P)  V 2 P 2 (1  P ) 2  2P P P(1  P)
~
2 P P P (1  P )  V 2 P 2 (1  P) 2 .

This equation agrees with the equation obtained in a).

c) According to Eq. (8.21) combined with D(1) = a = P P (1  P) and D(2) = b2 / 2 =
[V P (1  P)]2 / 2, the equation for the PDF f(p, t) is given by

wf ( p, t )
P 
> @
w > p (1  p)@ f ( p, t ) V 2 w 2 p 2 (1  p) 2 f ( p, t )
.
wt wp 2 wp 2
d) This equation cannot be solved analytically because D(1) and D(2) are nonlinear
functions of p: the solution technique applied in Sect. 8.3 cannot be used.

Section 8.5
Exercise 8.5.1
a) According to (8.89), the acceleration fluctuation is given by

v~ (t ) 4e dW
a~ (t )   (t ) .
W 3W dt
Hence, the acceleration correlation is given for any positive or negative r by
§ v~ (t ) 4e dW ·¸§¨ v~ (t  r ) 4e dW ·
a~ (t ) a~ (t  r ) ¨  (t )   (t  r ) ¸
¨ W 3W dt ¸¨ W 3W dt ¸
© ¹© ¹
1 ~ ~ 4e dW dW 1 4e
v (t ) v (t  r )  (t ) (t  r )  S (t , r ),
W2 3W dt dt W 3W
|r |
2e  W 4e 1 4e
e  G (r )  S (t , r ),
3W 2
3W W 3W
where the abbreviation S(t, r) refers to
dW dW
S (t , r ) v~ (t ) (t  r )  v~ (t  r ) (t ) .
dt dt
84 Chapter 8

We used here the equilibrium velocity correlation (8.100) and the properties of
dW / dt. S(t, r) can be evaluated in the following way,
S (t , r )
t r
4e t  (t  s ) / W dW dW 4e dW
 (t  r  s ) / W dW
³ e (s) (t  r ) ds  ³e (s) (t ) ds
3W 0 ds dt 3W 0 ds dt
4e t  (t  s ) / W 4e t  r  (t  r  s ) / W
³ e G ( s  t  r ) ds  ³e G ( s  t ) ds
3W 0 3W 0
4e r / W t 4e  r / W t  r
e ³ G ( s  t  r ) ds  e ³ G ( s  t ) ds
3W 0 3W 0

4e r / W 4e  r / W
e >T ( r )  T (t  r )@  e >T (r )  T (t )@
3W 3W
4e r / W 4e  r / W 4e | r | / W
e T (r )  e T (r ) e .
3W 3W 3W

Here, we used the formal solution of the velocity model,

4e t  ( t  s ) / W dW
v~ (t ) v~ (0) e  t / W  ³e ( s ) ds,
3W 0 ds
and we made use of the fact that dW / dt is uncorrelated to the initial velocity
fluctuation. The equilibrium correlation is independent of t, i.e., we can assume
that t o f. This leads to the neglect of T(t  r) in addition to the neglect of
T(t). The final result is a consequence of considering the case that r is positive
and negative, respectively. Hence, the acceleration correlation reads
|r |
2e  W 4e 4e 2e  | r | / W 4e
a~ (t ) a~ (t  r ) e  G (r )  2 e | r | / W  e  G (r ).
3W 2 3W 3W 3W 2 3W
b) This property can be seen in the following way,

³ a (t ) a (t  r ) dr ³ « 3W 2 e  3W G (r )» dr 3W >e @ 0  3W
f
~ ~
f ª
r /W 2e 4e
r /W º 2e f 2e
0.
0 0¬ ¼

Exercise 8.5.2
This integral can be written
~
~ d v (t  r ) dr
f f
~ ~
³ a (t ) a (t  r ) dr ³ a (t ) dr a~ (t ) v~ (f)  a~ (t ) v~ (t )
0 0

d ~ ~ 1 d ~2
v (t ) v (f)  v (t ) 0.
dt 2 dt
 Section 8.5 85

Exercise 8.5.3
We expand the square root in the definitions (8.120) of r1 and r2 for small Wf /W,

1 ª Wf º 1 ª § Wf ·º 1
r1 « 1  1  4 » « 1  ¨¨1  2 ¸¸»  ,
2W f « W »¼ 2W f ¬« © W ¹¼» W
¬

1 ª Wf º 1 ª § Wf ·º 1
r2 « 1  1  4 » « 1  ¨¨1  2 ¸¸»  .
2W f « W »¼ 2W f ¬« © W ¹¼» Wf
¬
In the limit Wf o 0 the velocity correlation (8.109) then becomes
r |r | r |r |
2e r e 2  r2 e 1
lim v~ (t )v~ (t  r ) lim 1
W f o0 3 W f o0 r1  r2
|r |
W f W f 
|r |

ª 1 W |r | |r | º  e e W |r |
2e 1 1 W 2e W 2e  W
lim « e f  e » lim e .
3 W f o0  1  1 « W Wf » 3 W f o0 W 3
¬ ¼  f 1
W Wf W
Thus, we recover Cv(t, r) = exp(|r| /W) in this way

Exercise 8.5.4
We use the expressions for r1 and r2 obtained in exercise 8.5.4 in the limit Wf o 0.
The acceleration correlation (8.114) then reads
|r | |r |

1 Wf 1 
 e  e W
2e r2 e
r 2 |r |
 r1 e 1
r |r |
2e Wf W
lim a~ (t ) a~ (t  r ) lim lim
W f o0 W f o0 3W W f r2  r1 W f o0 3W W f 
1

1
Wf W
|r |
W f  |Wr |  W f 
|r |
e e |r | W |r |
2e W 2e  4e e f 2e  W 4e
 lim  e W
 lim  e  G (r ).
W f o 0 3W W W 3W 2 3W W f o 0 2W f 3W 2 3W
f
1 f
W
This result recovers (8.107). The representation of the delta function involved was
proven in exercise 4.2.1.

Exercise 8.5.5
a) We take the derivative of dx / dt,
d v~ 1ª 4e t  ( t  s ) / W dW º 4e dW v~ 4e dW
 «v~ (0) e  t / W  ³ e ( s ) ds »    .
dt W ¬« 3W 0 ds ¼» 3W dt W 3W dt
This equation recovers the velocity model (8.89).
86 Chapter 8

b) This is a non-Markovian model because the position evolution is not complete-

ly determined by information at any t: all the history has to be accounted for.
c) F(t) represents the fluctuating velocity of the Markovian velocity model. Thus,
| t  t '|
2e  W
F (t ) 0, F (t ) F (t ' ) e .
3
d) By accounting for e = 3 Q / W and by neglecting the first term, the force F(t) is
given by
t
e  (t  s ) / W dW
F (t ) 4Q ³ ( s ) ds.
0 W ds
The exponential function divided by 2 W becomes equal to a delta function in
the limit W o 0 (see exercise 4.2.1),
|t  s |
1 
lim e W
G (t  s ).
W o 0 2W

Hence, F(t) is given in this limit by

t
e  (t  s ) / W dW t
dW
F (t ) 4Q 2 ³ ( s ) ds 4Q 2 ³ G (t  s) ( s) ds
0 2W ds 0 ds
dW t
dW
4Q 2 (t ) ³ G (t  s ) ds 4Q (t ).
dt 0 dt
The resulting position equation is given by
dx dW
v  4Q (t ).
dt dt
This equation is a Markovian equation. Therefore, the non-Markovian position
model reduces to a Markovian position model in the limit W o 0.
Chapter 9
Section 9.2
Exercise 9.2.1
a) The eigenvalues satisfy Eq. (9.21), which reads for the case considered
1  r 2
0 (1  r )(1  r )  8 1  r 2  8 r 2  9.
4 1 r
Therefore, we find r1 = 3 and r2 = 3. According to Eqs. (9.10) combined with
the relations y'10 = a11 y10 + a12 y20 and y'20 = a21 y10 + a22 y20, the solution is found
to be given by
(a11  r2 ) y10  a12 y20 r1 t (a11  r1 ) y10  a12 y20 r 2 t
y1 e  e ,
r1  r2 r1  r2
a21 y10  (a22  r2 ) y20 r1 t a21 y10  (a22  r1 ) y20 r 2 t
y2 e  e .
r1  r2 r1  r2
For the data considered we find
2 y10  2 y20 3 t 4 y10  2 y20  3 t y10  y20 3 t 2 y10  y20  3 t
y1 e  e e  e ,
6 6 3 3
4 y10  4 y20 3 t 4 y10  2 y20  3 t y10  y20 3 t 2 y10  y20  3 t
y2 e  e 2 e  e .
6 6 3 3
b) The required condition is y10 = y20. The corresponding linear function is given
by y2 = y1.
c) The required condition is y10 = y20 / 2. The corresponding linear function is
given by y2 = 2 y1.

Exercise 9.2.2
a) The eigenvalues satisfy Eq. (9.21), which reads for the case considered
1  r  A(1  A)
0 (1  r )( A  r )  2 A(1  A) r 2  (1  A)r  A  2 A(1  A).
2 Ar
This equation is solved by
1 A 1 1 A 1
r  r (1  A) 2  4 A  8 A 2  r 1  6 A  9 A2
2 2 2 2
1 A 1 3A
 r .
2 2
88 Chapter 9

Thus, we find r1 = A and r2 = 1  2A. According to Eqs. (9.10) combined with

Eqs. (9.7), the solution is given by
1  1  2 A  A(1  A) A t 1  A  A(1  A)  (1 2 A) t
y1 e  e
1 3A 1  3A
2  3 A  A2 A t 1  A2  (1 2 A) t
e  e ,
1 3A 1 3A
2  A  1  2 A A t 2  A  A  (1 2 A) t 1 A At 2
y2 e  e 3 e  e  (1 2 A) t .
1  3A 1  3A 1  3A 1 3A
b) The solutions become zero as t o f if 0.5 < A < 0.

Exercise 9.2.3
a) The equilibrium points satisfy the equation system
2
0 y1  y2 ,
0 y1  y1 y2 .
The use of the first equation in the second equation implies
2 3 2
0 y2  y2 y2 (1  y2 ).
This equation is solved by y2 = 0 and y2 = 1. According to the first equation we
have corresponding values y1 = 0 and y1 = 1. Hence, the two equilibrium points
are given by
( y1 , y2 ) (0, 0), ( y1 , y2 ) (1,1).
b) The linear equation system (9.69) reads for the equations considered
d § y1  Y1 · § 1  2Y2 ·§ y1  Y1 ·
¨ ¸ ¨¨ ¸¨ ¸.
dt ¨© y2  Y2 ¸¹ ©1  Y2  Y1 ¸¹¨© y2  Y2 ¸¹
At the first equilibrium point (0, 0) we find the characteristic equation
1 r 0
0 r (r  1).
1 r
Hence, we have r1 = 0 and r2 = 1, which means that the solution is unstable at
this equilibrium point. At the second equilibrium point (1, 1) we find the char-
acteristic equation
1 r 2
0 (r  1)(r  1).
0 1 r
Consequently, we have r1 = 1 and r2 = 1, which means that the solution is
unstable at this equilibrium point. Hence, none of the equilibrium points will be
realized.
 Section 9.2 89

Exercise 9.2.4
a) The equilibrium points satisfy the equation system
2 2
0 y1  y2 ,
0 2 y1 y2 .
The only equilibrium point of these equations is given by (0, 0).
b) The linear equation system (9.69) reads for the equations considered
d § y1  Y1 · § 2Y1 2Y2 ·§ y1  Y1 ·
¨ ¸ ¨¨ ¸¨ ¸.
dt ¨© y2  Y2 ¸¹ © 2Y2 2Y1 ¸¹¨© y2  Y2 ¸¹
At the equilibrium point (0, 0) we find the characteristic equation
r 0
0 r 2.
0 r
Hence, we have r1 = r2 = 0, which means that the stability of the equilibrium
point cannot be determined in this way.
c) The equation system can be used for deriving equations for y1 + y2 and y1  y2,
d ( y1  y2 ) 2 2
y1  2 y1 y2  y2 ( y1  y2 ) 2 ,
dt
d ( y1  y2 ) 2 2
y1  2 y1 y2  y2 ( y1  y2 ) 2 .
dt
It may be proven that these equations are solved by the expressions
1 1
y1  y2  , y1  y2  .
C1  t C2  t
Both y1 + y2 and y1  y2 approach zero as t o f, which means that y1 o 0 and
y2 o 0 asymptotically. Hence, the equilibrium point (0, 0) will be realized.

Exercise 9.2.5
a) The equilibrium points satisfy the equation system
0 y2 ,
2
0 y1 (1  y1 ).
Hence, there are three equilibrium points given by
( y1 , y2 ) (0, 0), ( y1 , y2 ) (1, 0), ( y1 , y2 ) (1, 0).
b) The linear equation system (9.69) reads for the equations considered
d § y1  Y1 · § 0 1 ·§ y1  Y1 ·
¨ ¸ ¨¨ ¸¨ ¸.
dt ¨© y2  Y2 ¸¹ 0 ¸¹¨© y2  Y2 ¸¹
2
©1  3Y1
90 Chapter 9

At the first equilibrium point (0, 0) we find the characteristic equation

r 1
0 r 2  1.
1 r
Hence, we have r1 = 1 and r2 = 1, which means that the solution is unstable at
the first equilibrium point. At the second and third equilibrium points (1, 0) and
(1, 0) we find
r 1
0 r 2  2.
2 r
Consequently, we have r1 = 21/2 i and r2 = 21/2 i, i.e., the solution is stable at
these equilibrium points. Therefore, we have centers at (1, 0) and (1, 0).
c) We consider the centers (Y1, 0), where Y1 = 1 and Y1 = 1, respectively. Then,
d ( y1  Y1 ) d ( y1  Y1 ) / dt 1 y2
 ,
dy2 dy2 / dt 2 y1  Y1
where the linear equation system is used. This equation can be written
1
( y1  Y1 ) d ( y1  Y1 )  y2 dy2 .
2
By integrating both sides we find
1 1 2
( y1  Y1 ) 2  y2 C,
2 4
where C is a constant. By accounting for the initial condition we can write this
equation
1 2 1 2
( y1  Y1 ) 2  y2 ( y10  Y1 ) 2  y20 .
2 2
d) The latter equation represents the equation for an ellipse,
2
( y1  Y1 ) 2 y2
2
 2
1.
( y10  Y1 ) 2  y20 / 2 2 [( y10  Y1 ) 2  y20 / 2]

Section 9.3
Exercise 9.3.1
a) According to the relations (9.62) we find for b2 = 0 three equilibrium points

§ 1 d 1 ·
(Y1 , Y2 ) (0,0), (Y1 , Y2 ) 1,0 , (Y1 , Y2 ) ¨ , 2 ¸.
©d d ¹
 Section 9.3 91

b) The linear equation system (9.69) reads for the equations considered
d § y1  Y1 · §1  2Y1  d Y2  d Y1 ·§ y1  Y1 ·
¨ ¸ ¨¨ ¸¨ ¸.
dt ¨© y2  Y2 ¸¹ ©  d Y2 1  d Y1 ¸¹¨© y2  Y2 ¸¹
At the first equilibrium point (0, 0) we find the characteristic equation
1 r 0
0 (1  r ) 2 .
0 1 r
Thus, we have r1 = r2 = 1, i.e., the solution is unstable at this equilibrium point.
At the second equilibrium point (1, 0) we find the characteristic equation
1 r d
0 (r  1)(r  d  1).
0 1 d  r
Therefore, we find r1 = 1 and r2 = 1  d. Depending on the value of d, the
eigenvalue r2 can be positive or negative, i.e., the solution near this equilibrium
point can be unstable or asymptotically stable. At the coexistence equilibrium
point (1 / d, [d  1] / d 2) we find the characteristic equation

1  2 / d  ( d  1) / d  r 1 1/ d  r 1 § 1· 1
0 r¨ r  ¸  1
 (d  1) / d r 1/ d  1 r © d ¹ d
r 1
r2    1.
d d
This equation is solved by
1 1 1 1 1 1 | 1  2d |
r  r 2
 1  r 1 4 d  4 d 2  r .
2d 4d d 2d 2d 2d 2d
Hence, we find the eigenvalues r1 = 1 and r2 = 1  1 / d = (d  1) / d. Depend-
ing on d, r2 can be positive or negative, i.e., the solution near this equilibrium
point can be unstable or asymptotically stable.
c) For d > 1 we find a non-negative coexistence equilibrium point.
d) The coexistence equilibrium point is unstable for d > 1. The second equilibrium
point (1, 0) is asymptotially stable. Consequently, (1, 0) will be asymptotically
realized.

Exercise 9.3.2
The coexistence of species is characterized by
C f A f
y1 , y2 .
D B
Therefore, a nonzero f increases the prey and decreases the predator density. The
condition f < A is a requirement to calculate a positive y2.
92 Chapter 9

Exercise 9.3.3
a) We use N = S + I to obtain
dI E § I ·
I (N  I ) E ¨1  ¸ I.
dt N © N¹
This equation is a logistic equation: see Eq. (7.88).
b) According to Eq. (7.102), the solution of this equation is given by
N
I .
1  N / I 0  1 e  E (t  t 0 )
c) This formula shows that I = N for large values of t. According to N = S + I we
find that S = 0 for large t. Hence, all individuals will be finally infected.

Exercise 9.3.4
a) The model parameter J describes the recovery rate: J reduces the number of
infected individuals and increases the number of individuals who may become
infected (i.e., individuals may become infected again after one infection).
b) We use N = S + I to obtain
dI E § EI· § E ·
(N  I ) I  J I ¨E J  ¸I ( E  J ) ¨¨1  I ¸¸ I .
dt N © N ¹ © (E  J ) N ¹
This equation is a logistic equation: see Eq. (7.88).
c) By adopting the solution obtained in exercise 9.3.3 we find
(1  J / E ) N
I .
1  (1  J / E ) N / I 0  1 e  ( E  J ) (t  t 0 )
d) For the case E > J we find the asymptotic values I = (1  J / E) N and S = J / E N.
For the case E d J we find I = 0 and S = N.

Exercise 9.3.5
a) The difference is that the reduction of the number of infected individuals does
not increase the number of individuals who may become infected: individuals
cannot become reinfected.
b) The model equations for I and S can be used to obtain
dI dI / dt E S I / N J I J N
1  .
dS dS / dt E SI/N E S
c) We integrate this equation to obtain
J S
I (S )  I 0 S0  S  N ln .
E S0
 Section 9.3 93

Here, I0 and S0 refer to the initial value of S, and we used the fact that S t 0. We
multiply this equation with E / (J N),
E (I  I0 ) E S0 § S JN S ·
¨1   ln ¸¸.
JN ¨
J N © S0 E S0 S0 ¹
In terms of y = E (I  I0) / (J N), x = 1  S / S0, and R = E S0 / (J N) we obtain then
the equation y R x  ln(1  x).
d) The first and second order derivatives of y are given by
dy 1 d2y 1
R ,  .
dx x 1 dx 2 ( x  1) 2
Thus, a critical point is given at x = 1  1 / R. Because of 0 d x d 1, this critical
point can only be realized if R t 1. The second order derivative shows that we
have a local maximum at x = 1  1 / R for R > 1. The maximum value of y is
given by
y R (1  1 / R )  ln(1 / R ) R  1  ln R.
e) Due to 0 d x d 1 we find
dy 1
f  R d R  1.
dx x 1
Hence, for R < 1 we have the case that y is a decreasing function of x. For R = 1
we have
dy 1 x
1  .
dx x 1 1 x
This derivative is negative except at x = 0.

Exercise 9.3.6
a) We apply the relation R(t) = S*(t)  I(t) to obtain a closed equation system for
S*(t) = N  S(t) and I,
dI E E §¨ § J ··
(N  S*) I  J I  I ¨ S *  N ¨¨1  ¸¸ ¸,
¸
dt N N © © E ¹¹
dS * E § E ·
( N  S * ) I Q ( S *  I ) ¨ E  Q  S * ¸ I Q S * .
dt N © N ¹
b) An obvious first equilibrium point is given by (0, 0). Another equilibrium point
is found by setting the parenthesis term in the first equation equal to zero,
§ J ·
S* N ¨¨1  ¸¸.
© E ¹
94 Chapter 9

The use of this condition in the second equation leads to

§ § J ·· § J · § J ·
0 ¨¨ E  Q  E ¨¨1  ¸¸ ¸¸ I  Q N ¨¨1  ¸¸ (Q  J ) I  Q N ¨¨1  ¸¸.
© © E ¹¹ © E ¹ © E¹
Hence, we have
1 J / E
I N .
1  J /Q
Consequently, we find the two equilibrium points
§ 1 J / E ·
, N 1 J / E
* *
( I1 , S1 ) (0, 0), ( I 2 , S2 ) ¨¨ N ¸¸.
© 1  J /Q ¹
c) The requirement to find positive I2 and S2 values is E > J.
d) The linear equation system (9.69) reads for the equations considered

§ E * E ·
d § I  I e · ¨  N Se  E  J N
I e ¸§ I  I ·
¨ *¸
¨ ¸¨ * e * ¸.
dt ¨© S *  S e ¸¹ ¨ E  Q  E S * E ¨ ¸
 I e  Q ¸¸© S  S e ¹
¨ e
© N N ¹
At the first equilibrium point (0, 0) we find the characteristic equation
E J  r 0
0 ( r  J  E )(r  Q ).
E Q Q  r
Hence, we have r1 = E  J and r2 = Q. Because of the condition E > J derived in
c) we find that the solution is unstable. At the second equilibrium point we find
the characteristic equation

E
r  I2 E E
N § ·
0 r ¨ r  I 2  Q ¸  (J Q ) I 2 .
E © N ¹ N
J Q  I 2 Q  r
N
This equation can be written
§E · E
0 r 2  ¨ I 2 Q ¸ r  (J  Q ) I 2 .
©N ¹ N
This equation is solved by
2
1§ E · 1 §E · E
r  ¨ I 2 Q ¸ r ¨ I 2 Q ¸  4(J Q ) I 2 .
2© N ¹ 2 ©N ¹ N

The two eigenvalues have negative real parts. Consequently, the solution at this
equilibrium point is asymptotically stable.
 Section 9.4 95

Section 9.4
Exercise 9.4.1
By differentiating the total energy E we find
dE dD ª g d 2D º
m r2 « sin D  2 » 0.
dt dt ¬ r dt ¼
The bracket term has to be zero. The resulting equation is the differential equation
(9.111) for the undamped pendulum.

Exercise 9.4.2
a) By using (1  x)1/2 = 1 + x / 2 and the integral ³ sin2 x dx = x / 2  (1/4) sin(2x) we
obtain
S /2§
r S /2
dT r sin 2 (D 0 / 2) 2 ·
TP 4
g
³ 4
g
³ ¨¨1  2
sin T ¸¸ dT
0 1  sin 2 (D 0 / 2) sin 2 T 0 © ¹
S /2
r ª sin 2 (D 0 / 2) § T sin 2T ·º r § sin 2 (D 0 / 2) ·
4 «T  ¨  ¸» 2S ¨1  ¸.
g 2 ©2 4 ¹¼ 0 g ¨ 4 ¸
¬ © ¹
b) The influence of D0 on TP is smaller than 1% if |D0| d 23º.

Exercise 9.4.3
a) We can use Eq. (9.140) at any initial values,
1 2
1  cos y10  y20 E* .
2
The use of this expression in Eq. (9.142) results in
2
y2 r 2(cos y1  cos y10 )  y20 .

b) The consistency of the latter formula with the initial conditions requires the use
of the positive sign for a positive value of y20 and the use of the negative sign
for a negative value of y20.
c) The requirement for finding open curves is E* > 2, this means
1 2
1  cos y10  y20 ! 2.
2
This relation can be written
2
y20 ! 2 (1  cos y10 ).
The maximum of the right hand side is four. Thus, we do we always find open
curves in the y1-y2 phase plane if |y20| > 2.
96 Chapter 9

Exercise 9.4.4
a) The equations considered imply d 2z / dt 2 = 0. The latter equation is solved by z
= z'0 t + z0. Here, z0 is the initial value of z and z'0 is the initial value of dz / dt.
b) The equation system considered implies
d2y d 2 x2 d 2 x1 k k § 1 1 ·
 2  y y ¨¨  ¸ k y.
¸
dt 2 dt 2 dt m1 m2 © m1 m2 ¹
c) This equation represents an undamped spring-mass equation d 2y / dt 2 + k y / m* =
0, where m* is determined by
1 1 1
 .
m* m1 m2
This means, y describes the oscillations of a mass m* attached to a spring that is
fixed at one end: see the illustration in Fig. 7.3.
d) According to Eq. (7.76), the solution of the y equation is given by

cos k / m* t  G
y y0 ,
cos G
where G = arctan(y'0 / [(k / m*)1/2 y0]).

Exercise 9.4.5
a) The definitions of y and z provide a linear equation system for x1 and x2,
m1 x1  m2 x2 (m1  m2 ) z ,
 x1  x2 y  L.
The solutions of this equation system read
( m1  m2 ) z  m2 ( y  L) (m1  m2 ) z  m1 ( y  L)
x1 , x2 .
m1  m2 m1  m2
In terms of the solutions for z and y derived in exercise 9.4.4 we find
m2 m1
x1  cos k / m* t  L , x2 cos k / m* t  L .
m1  m2 m1  m2
Here, we accounted for z'0 = y'0 = 0 and z0 = 0 and y0 = 1.
b) For the case m1 o f we find
x1 0, x2 cos k / m2 t  L ,

where we accounted for m* = m2. This result makes sense: m2 cannot affect m1.
c) For the case m1 o 0 we find

x1  cos k / m1 t  L , x2 0,

where we accounted for m* = m1. This result makes sense: m1 cannot affect m2.
 Section 9.5 97

Section 9.5
Exercise 9.5.1
a) By following the explanations in Sect. 9.5.2 we find three equilibrium points,

P1 (0, 0, 0), P2 b ( R  1) , b ( R  1) , R  1 ,
P3  b ( R  1) ,  b ( R  1) , R  1 .
b) Instead of the condition (9.155) we have now the condition
 Pr  r Pr 0
0 1 1 r  s b ( R  1) .
s b ( R  1) s b ( R  1) br

The expansion of the determinant provides the condition

0 ( Pr  r )>(1  r )(b  r )  b ( R  1)@  Pr >(b  r )  b ( R  1)@

(b  r )>( Pr  r )(1  r )  Pr @  b ( R  1)>Pr  ( Pr  r )@
r (b  r )>r  Pr  1@  b ( R  1)>2 Pr  r @
r r 2  (b  Pr  1) r  b ( Pr  1)  b ( R  1)  2 Pr b ( R  1)
r 3  (b  Pr  1) r 2  b ( R  Pr ) r  2 Pr b ( R  1) .
R2 is determined by the condition c = a b, this means
2 Pr b ( R2  1) (b  Pr  1) b ( R2  Pr ) .
This equation can be written
R2b 2 Pr  (b  Pr  1) Pr b b  Pr  1  2 .
Therefore, we find
Pr  b  3 Pr  b  1  2b  4 Pr (b  2)
R2 Pr Pr Pr  2 .
Pr  b  1 Pr  b  1 Pr  b  1

Exercise 9.5.2
a) R2 will increase with growing values of b because of
dR2 Pr  b  1  Pr  b  3 2 Pr  2
Pr Pr ! 0.
db ( Pr  b  1) 2 ( Pr  b  1) 2
b) R2 is positive if Pr > b + 1.
c) The first two derivatives of R2 by Pr are given by
dR2 (b  2) ( Pr  b  1)  Pr (b  2) (b  2) (b  1)
1 2 1 2 .
dPr ( Pr  b  1) 2 ( Pr  b  1) 2
98 Chapter 9

d 2 R2 (b  2) (b  1)
4 .
dPr 2 ( Pr  b  1)3
Hence, R2 has a minimum at

Prc b  1  2 (b  2) (b  1) .
The minimum of R2 is given by
§ b2 · § b  2 ·¸
R2 Prc ¨1  2 ¸ Prc ¨1  2 .
¨ 2 (b  2) (b  1) ¸ ¨ b  1 ¸¹
© ¹ ©

Exercise 9.5.3
a) The equilibrium points satisfy the equations (a = 0.2)

0 Y2  Y3 , 0 Y1  0.2 Y2 , 0 0.2  Y3 (Y1  b).

The first two equations imply that Y2 = Y3 and Y1 = 0.2 Y2 = 0.2 Y3. Hence,
the last equation can be written
2
0 0.2  Y3 (0.2 Y3  b) 0.2 Y3  5 b Y3  1 .

This equation is solved by

5 §¨ 4 ·¸
Y3 b r b2  .
2 ¨© 25 ¸¹
Therefore, we find two potential equilibrium points P = (0.2 Y3, Y3, Y3).
b) Correspondingly, we find no equilibrium point, one equilibrium point, and two
equilibrium points if b < 0.4, b = 0.4, and b > 0.4, respectively.
c) The solutions will go to infinity.

Exercise 9.5.4
a) According to the equation system (9.149), the linear equation system near the
equilibrium points is given by
§ y Y · §0 1  1 · § y1  Y1 ·
d ¨ y 1  Y1 ¸ ¨ ¸¨ ¸
¨1 0.2 0 ¸ ¨ y2  Y2 ¸.
dt ¨¨
2 2¸
¸ ¨Y ¨ ¸
y
© 3 3¹ Y © 3 0 Y1  b ¸¹ © y3  Y3 ¹
b) The corresponding characteristic equation reads
r 1 1
0 1 0.2  r 0 r (0.2  r ) (Y1  b  r )  (0.2  r ) Y3  Y1  b  r.
Y3 0 Y1  b  r
 Section 9.5 99

By accounting for Y1 = 0.2 Y3, we can write this equation

0 r 3  0.2Y3  b  0.2 r 2  0.2 0.2Y3  b r  Y3  1 r  0.2Y3  0.2Y3  b.
The latter equation can be reduced to
0 r 3  0.2Y3  b  0.2 r 2  1.04Y3  0.2 b  1 r  0.4Y3  b.
By using the definition of Y3 we can write this equation
§ 4 · § 4 · 4
0 r 3  ¨ 0.5 b # 0.5 b 2   0.2 ¸ r 2  ¨ 2.4 b r 2.6 b 2   1¸ r # b 2  ,
¨ 25 ¸ ¨ 25 ¹ ¸ 25
© ¹ ©
which means that
§ 4 ·¸ 2 §¨ 4 ·¸ 4
0 r 3  0.5 ¨ b  0.4 # b 2  r  2.4 b  1 r 2.6 b 2  r # b2  .
¨ ¸
25 ¹ ¨ ¸
25 ¹ 25
© ©
c) The critical value of b is determined by the condition c = a b related to equation
(9.157). For the case considered this condition reads
§ 4 ·¸ §¨ 4 ·¸ 4
0.5 ¨ b  0.4 # b 2  2.4 b  1 r 2.6 b 2  # b2  .
¨ 25 ¸¹ ¨© 25 ¸¹ 25
©
We distribute the left-hand side to obtain
4 4
0 1.2 b 2  0.5 b r 1.3 b b 2   0.48 b  0.2 # 0.52 b 2 
25 25
4 4 § 4 · 4
# 1.2 b b 2  # 0.5 b 2   1.3 ¨ b 2  ¸ r b 2 
25 25 © 25 ¹ 25
4 4
0.1b 2 r 0.1b b 2  # 0.02 b 2   0.02 b  0.008
25 25
4 4
b2 # b b2  r 0.2 b 2   0.2 b  0.08
25 25
4
# (b  0.2) b 2   (b  0.4)(b  0.2)
25
b  0.4 (b  0.2) b  0.4 # (b  0.2) b  0.4 .

b  0.4 §¨ (b  0.4)(b 2  0.4b  0.04) # (b  0.4)(b 2  0.4b  0.04) ·¸

© ¹
b  0.4 §¨ b3  0.4b 2  0.04b  0.4b 2  0.16b  0.016
©
# b 3  0.4b 2  0.04b  0.4b 2  0.16b  0.016 ·¸
¹
b  0.4 §¨ b (b 2  0.12)  0.016 # b (b 2  0.12)  0.016 ·¸ .
© ¹
100 Chapter 9

d) The parenthesis term cannot become zero. Therefore, the condition c = a b is

only satisfied for b = 0.4.
e) This is not the case. Only the range b t 0.4 can be considered here regarding the
eigenvalue analysis, i.e., the sign of eigenvalues cannot change at b = 0.4.

Exercise 9.5.5
a) There exists one equilibrium point P = (0.2, 1, 1). The characteristic equation
reads for this case 0 = r (r2 + 1.96). Consequently, the roots are given by r1 = 0,
r2 = 1.4 i, and r3 = 1.4 i. The solution behavior near this equilibrium point is
stable: the trajectories are found in closed regions that surround the center. The
system trajectories do not approach the center.
b) See the figure.

Exercise 9.5.6
a) The first equilibrium point is P = (0.4, 2, 2). The related characteristic equation
is solved by r1 = 0.1007, r2 = 0.0337 + 1.7263 i, and r3 = 0.0337  1.7263 i.
Hence, the solution behavior near this equilibrium point is unstable.
b) See the figure.

c) The second equilibrium point is given by P = (0.1, 0.5, 0.5). The roots of the
related characteristic equation are given by r1 = 0.2109, r2 = 0.0721 + 1.1926 i,
and r3 = 0.0721  1.1926 i. This solution is unstable.
 Section 9.5 101

d) See the figure. The structures in these figures become larger in time.
Chapter 10
Section 10.2
Exercise 10.2.1
a) This property arises from the fact that f(x) is the mean value of a non-negative
variable.
b) This property is caused by the fact that the delta function vanishes at infinity.
c) In terms of the definition of the joint PDF we find

³³ f ( x, y ) dx dy ³³ G ( x  X ) G ( y  Y ) dx dy 1 1.

d) The validity of this property can be seen as follows,

³³ g ( x, y ) f ( x, y ) dx dy ³³ g ( x, y ) G ( x  X ) G ( y  Y ) dx dy
³³ g ( x, y ) G ( x  X ) G ( y  Y ) dx dy
g ( X , Y ) ³³ G ( x  X ) G ( y  Y ) dx dy
g ( X ,Y ) .
The averaging can be used for all the integral because it does only apply to the
random variables X and Y. The next lines apply the sifting property and nor-
malization of delta functions.

Exercise 10.2.2
For independent variables X and Y we find
1 1
g ( X ,Y ) | x g ( X ,Y )G ( x  X ) g ( x, Y ) G ( x  X )
f ( x) f ( x)
1 f ( x)
g ( x, Y ) G ( x  X ) g ( x, Y ) g ( x, Y ) .
f ( x) f ( x)

Exercise 10.2.3
a) The joint PDF f(x, y) can be used to calculate the marginal PDF f(x) and the
conditional PDF f(y | x) via the relations
f ( x, y )
f ( x) ³ f ( x, y ) dy, f ( y | x)
f ( x)
.

According to Eq. (10.17), <Y | x> can be calculated then in terms of f(y | x) by
Y|x ³ y f ( y | x) dy.
b) No assumption about <Y | x> has to be made here in difference to the approach
applied in Chap. 2 (where we assumed linear or quadratic functions).
104 Chapter 10

Exercise 10.2.4
~ ~~
a) The stochastic model provides for < Y 2 > and < X Y > the relations
~ 2 ~ ~~ ~ 1/ 2 ~ 1/ 2
Y2 rXY Y 2 , XY rXY Y 2 X2 .
~~ ~
b) The relation for < X Y > represents the definition of rXY. The relation for < Y 2 >
is only valid for rXY = r1. Consequently, the model for Y can only represent a
reasonable model for rXY = r1.

Section 10.3
Exercise 10.3.1
a) By integrating the last line of (10.39) over y we obtain

1 ­° ( yˆ  rXY xˆ ) 2  (1  rXY 2 ) xˆ 2 ½°
³ f ( x, y ) dy ~ ~ ³ ®
exp 2 ¾ dy
2S (1  rXY ) X 2 Y 2
2 °̄ 2 (1  rXY ) °¿

1 ­ xˆ 2 ½ ­° ( yˆ  rXY xˆ ) 2 ½°
exp® ¾³ exp® ¾ dyˆ .
~ °̄ 2 (1  rXY 2 ) °¿
2S (1  rXY ) X 2
2
¯ 2¿

Here, the integration over y was replaced by an integration over ŷ by using the
definition of ŷ . The integration can be performed by introducing the variable z
= ( yˆ  rXY xˆ ) / [2 (1  rXY2)]1/2. In terms of z we obtain

1 ­ xˆ 2 ½ 2 1 ­ xˆ 2 ½
³ f ( x, y) dy exp® ¾³ e  z dz exp® ¾ f ( x).
~ 2 ~
S 2 X2 ¯ ¿ 2S X 2 ¯ 2¿

The integral over exp(z2) is equal to S1/2 according to relation (4.70).

b) By replacing x by y and y by x, the latter result can be written ³ f(y, x) dx = f(y).
Due to the symmetry f(x, y) = f(y, x) we find then ³ f(x, y) dx = f(y).

Exercise 10.3.2
The integral over the conditional PDF provides

1 ­° ( yˆ  rXY xˆ ) 2 ½°
³ f ( y | x) dy exp®
~ ³ ¾ dy
°̄ 2 (1  rXY 2 ) °¿
2S (1  rXY ) Y 2
2

1 ­° ( yˆ  rXY xˆ ) 2 ½°
³ ®°̄
exp 2 ¾
dyˆ 1.
2S (1  rXY )
2 2 (1  rXY ) °¿

The integral is equal to one because we have an integral over a normal PDF.
 Section 10.3 105

Exercise 10.3.3
a) The conditional variance can be written

>Y  Y | x @ 2
|x
1
f ( x)
>
Y Y|x x X
@ G (x  X ) .
2

Here, the definition (10.20) of a conditional mean is used, and <Y | x>, which is
a function of x, is taken at x = X. In terms of this relation we find

³ >Y  Y | x @ 2
| x f ( x) dx >Y  Y | x x X
@ 2
.

b) The normal structure of f(y | x) implies that

>Y  Y | x @ 2
|x
~ 2
Y 2 (1  rXY ).

The global variance is then found to be given by

>Y  Y | x x X
@ 2
³ >Y  Y | x @ 2
| x f ( x) dx
~ 2
Y 2 (1  rXY ).

c) The conditional variance is equal to <[Y  <Y | x>x=X ]2> because the statistics of
Y  <Y | x> are independent of x.

Exercise 10.3.4
For the case rXY = r1 we find for the parameters
­f if rXY o 1 ­ln 1 0 if rXY o 1
ao® , bo® .
¯ln 1 0 if rXY o 1 ¯f if rXY o 1
Hence, the ellipse becomes a line: for rXY o +1 we have a line along the x' axis,
and for rXY o 1 we have a line along the y' axis.

Exercise 10.3.5
a) The stable case is basically characterized by horizontal motions. The vertical
velocity component w is hardly correlated to u, v, and T, which indicates that
there is hardly a vertical motion present.
b) The unstable case indicates significant upward motions of warm air because rwT
has a much higher value than the other correlation coefficients.
c) For the stable case we definitely need u, v, and T. For the neutral case we need
u, v, w, and T. For the unstable case we need w and T.

Exercise 10.3.6
a) The use of (10.39) for f(x, y) in
f
f ( z) ³ f ( x, z  x) dx
f
106 Chapter 10

leads to the integral

1 ­° ( yˆ  rXY xˆ ) 2  (1  rXY 2 ) xˆ 2 ½°
exp®
~2 ³
f ( z) ¾ dx
~ °̄ 2 (1  rXY )
2
°¿
2S (1  rXY ) X 2 Y
2

1 ­° ( yˆ  rXY xˆ ) 2  (1  rXY ) xˆ 2 ½° 2

~ ³ ®
exp 2 ¾ dxˆ.
2S (1  rXY ) Y 2
2 °̄ 2 (1  rXY ) °¿

The second line applies the x̂ definition. According to the definitions (10.40)
of x̂ and ŷ combined with y = z  x we use here the abbreviations
x X zx Y
xˆ , yˆ .
~ 1/ 2 ~ 1/ 2
X2 Y2

The last expression shows that ŷ is a function of x. This dependence can be

made explicit by the rewriting
~ 1/ 2
z X2 xˆ  X  Y
yˆ zˆ  R xˆ.
~ 1/ 2
Y2
~
Here, we introduced the abbreviations ẑ = (z  <X>  <Y>) / < Y 2 >1/2 and R =
~ 2 1/2 ~ 2 1/2
< X > / < Y > . Correspondingly, the numerator in the braces term of f(z)
can be written
2 2 2
( yˆ  rXY xˆ ) 2  (1  rXY ) xˆ 2 zˆ  ( R  rXY ) xˆ  (1  rXY ) xˆ 2
2
zˆ 2  2 ( R  rXY ) zˆ xˆ  ( R  rXY ) 2 xˆ 2  (1  rXY ) xˆ 2
(1  R 2  2 R rXY ) xˆ 2  2 ( R  rXY ) zˆ xˆ  zˆ 2
2
§ ( R  rXY ) zˆ · § ( R  rXY ) 2 ·
(1  R 2  2 R rXY ) ¨¨ xˆ  ¸  zˆ 2 ¨1  ¸
¸ ¨ 1 R2  2 R r ¸
© 1  R  2 R rXY
2
¹ © XY ¹
2
§ ( R  rXY ) zˆ · § 1  rXY
2
·
(1  R 2  2 R rXY ) ¨¨ xˆ  ¸  zˆ 2 ¨ ¸.
¸ ¨ 1 R2  2 R r ¸
© 1  R  2 R rXY
2
¹ © XY ¹
In terms of this expression we can write f(z) as

1 ­ zˆ 2 ½
f ( z) exp® ¾
~
¯ 2(1  R  2 R rXY ) ¿
2
2S (1  rXY ) Y 2
2

­° 1  R 2  2 R r § ( R  rXY ) zˆ ·
2
½°
˜ ³ exp® XY
¨ xˆ  ¸ ¾ dxˆ.
2
2 (1  rXY ) ¨ 1 R2  2 R r ¸
°̄ © XY ¹ °¿
 Section 10.3 107

To calculate the integral we introduced the variable

1  R 2  2 R rXY § ( R  rXY ) zˆ ·
s ¨ xˆ  ¸.
2
2 (1  rXY ) ¨ 1 R2  2 R r ¸
© XY ¹
Then, f(z) reads
­ zˆ 2 ½
exp® ¾
¯ 2 (1  R 2
 2 R r )
XY ¿
2
2 (1  rXY ) s2
f ( z)
~2 1  R 2
 2 R r
³ e ds.
2S (1  rXY ) Y
2
XY

The integral is equal to S1/2. Thus, we obtain

1 ­ zˆ 2 ½
f ( z) exp® ¾.
~
¯ 2(1  R  2 R rXY ) ¿
2
2 S Y 2 1  R 2  2 R rXY
~
Finally, we apply the abbreviations ẑ = (z  <X>  <Y>) / < Y 2 >1/2 and R =
~ ~
< X 2 >1/2 / < Y 2 >1/2 to obtain
­ z X  Y
2 ½
1 ° °
f ( z) exp® ~2 ~~ ~2 ¾.
~ ~~ ~
2S ( Y 2 2 XY  X2 ) °̄ 2 ( Y  2 X Y  X ) °¿
b) By using Z = X + Y we can write f(z) as

­ z Z 2 ½
1 ° °
f ( z) exp® ~2 ¾.
~
2S Z 2 °̄ 2 Z °¿
~
Thus, f(z) is a normal PDF with mean <Z> and variance < Z 2 >.
c) X and Y may balance each other, which reduces the variance of the sum. For
example, for Y = X we have a zero variance of the sum.

Exercise 10.3.7
a) The PDF f(z) of the sum Z = X + Y of any two independent random variables X
and Y is given according to Eq. (10.66) by f(z) = ³ f(x) f(z  x) dx. For the case
considered we have
1 1
f ( z) ³ f ( x) f ( z  x) dx ³ f ( z  x) dx.
0 0

By following the explanations in Sect. 4.2.3 we can write f(z  x) as

f ( z  x) T 1  ( z  x)  T 0  ( z  x) T z  x T z 1 x .
108 Chapter 10

By using this expression we obtain

1
f ( z) ³ >T z  x  T z  1  x @dx.
0

The evaluation of this integral shows that f(z) is a triangular PDF,

­ z

°0 d z d 1 : ³ dx z
° 0

° 1 z 1
f ( z ) ®1  z d 2 : ³ dx  ³ dx 1  ( z  1) 2  z .
° 0 0
°otherwise : 0
°
¯
b) According to Sect. 4.4.3, see Fig. 4.14, the PDF of the sum of independent and
uniformly distributed random variables approaches a normal PDF if the number
of variables involved becomes large.

Section 10.4
Exercise 10.4.1
a) The principal invariants are given in principal axes by the expressions
I O1  O2  O3 ,

II
1
2
>
(O1  O2  O3 ) 2  O1  O2  O3
2 2
@
2

1 2
2
>
O1  2O1O2  O2 2  2(O1  O2 )O3  O32  O12  O2 2  O32 @
O1O2  O1O3  O2O3 ,
1 1 1 3
(O1  O2  O3 )3  (O1  O2  O3 )(O1  O2  O3 )  (O1  O2  O3 )
2 2 2 3 3
III
6 2 3
1 1 3
(O1  O2  O3 ) (O1  O2  O3 ) 2  3(O1  O2  O3 )  (O1  O2  O3 )
2 2 2 3 3

6 3
1 1 3
(O1  O2  O3 )(O1O2  O1O3  O2 O3  (O1  O2  O3 ))  (O1  O2  O3 )
2 2 2 3 3

3 3
1 2
3
>
O1O2O3  O1 (O2  O3 )  O1 (O2  O3 )  O2 (O1  O3 )  O2 (O12  O32 )
2 2 2

2 2
@
 O3 (O1  O2 )  O3 (O1  O2 ) O1O2O3 .
2

Thus the principal invariants are positive for positive eigenvalues O1, O2, and O3.
b) The characteristic equation can be written O (O2 + II ) = I O2 + III. The right-hand
side and the parenthesis term are positive if I, II, and III are positive. Therefore,
all eigenvalues O have to be positive.
 Section 10.4 109

Exercise 10.4.2
We consider the identity
E kn E 1nj G kj .
By taking the derivative by t we obtain
dE kn 1 dE 1nj
E nj  E kn 0.
dt dt
By multiplying this expression with E1ik we obtain
dE 1ij dE kn 1
 E 1ik E nj .
dt dt

Exercise 10.4.3
The partial derivative of f(x, t) by t reads
wf ( x, t ) / wt 1 w det( E ) 1 wE 1ij
  xi  D i x j  D j
f ( x, t ) 2 det( E ) wt 2 wt
1 wD i 1 wD j
 E 1ij x j  D j  E 1ij xi  D i
2 wt 2 wt
1 w det( E ) 1 wE ij1
wD i
  xi  D i x j  D j  E 1ij xj D j .
2 det( E ) wt 2 wt wt
We multiply this equation with f(x, t) and integrate it over the sample space x. The
integral over wf(x, t) / wt is zero because of the normalization constraint of f(x, t).
The integral over (xi  Di) (xj  Dj) f(x, t) is equal to the variance Eij, and the integral
over the last term of the right-hand side is zero because the integral is the mean of
a fluctuation. Hence, we obtain
1 w det( E ) 1 wE 1ij
0   E ji .
2 det( E ) wt 2 wt
In terms of the result obtained in exercise 10.4.2 we can write this relation
1 w det(E ) wE 1ij wE kn 1 wE ki
 E ji E 1ik E nj E ji E 1ik .
det( E ) wt wt wt wt
Then, the partial derivative by t can be replaced by the regular derivative because
Eij does only depend on t,
1 d det( E ) dE ki
E 1ik .
det( E ) dt dt
110 Chapter 10

Exercise 10.4.4
a) First, wf(x, t | x', t')/wxm is given by
wf ( x , t x' , t ' ) / wxm 1 1
 E 1mj x j  D j  E 1im xi  D i  E 1mj x j  D j .
f ( x , t x' , t ' ) 2 2
This result is obtained by replacing i by j in the previous expression and using
the symmetry of E1ij. The derivative w2f(x, t | x', t') / wxn wxm is given by
w 2 f ( x, t x' , t ' ) / wxn wxm
 E 1mn  E 1nl xl  D l E 1mj x j  D j .
f ( x, t x' , t ' )
Here, it is relevant to use different subscripts for the terms xl  Dl and xj  Dj.
The time derivative wf(x, t | x', t')/ wt of f(x, t | x', t') is given by
wf ( x, t x' , t ' ) / wt 1 w det( E ) 1 wE 1ij
  xi  D i x j  D j
f ( x , t x' , t ' ) 2 det( E ) wt 2 wt
1 1 wD i 1 wD j
E ij
 x j  D j  E 1ij xi  D i
2 wt 2 wt
1 w det( E ) 1 wE ij
1
wD i
  xi  D i x j  D j  E 1ij xj D j .
2 det( E ) wt 2 wt wt
The last expression is obtained by combining the last two terms of the previous
line and using E1ij = E1ji. It is helpful to rewrite wE1ij / wt and wdet(E) / wt. In
terms of the relations shown in exercises 10.4.2 and 10.4.3 we obtain
wf ( x , t x' , t ' ) / wt 1 wE wD i
 E 1ik ki  E 1ij xj D j
f ( x , t x' , t ' ) 2 wt wt
1 1 wE kn 1
 E ik E nj xi  D i x j  D j .
2 wt
b) It is convenient to write the Fokker-Planck equation (10.107) as follows,
wf ( x , t x' , t ' ) / wt
>
Gmm  Gm  Gmk ^x k  Xk ` @ wf ( x, t x' , t ' ) / wxm
f ( x , t x' , t ' ) f ( x, t x' , t ' )
w 2 f ( x , t x' , t ' ) / wxn wxm
 Dnm .
f ( x , t x' , t ' )
We use the partial derivatives obtained in a) in this equation,
1 wE wD i 1 wE
 E 1ik ki  E 1ij x j  D j  E 1ik kn E 1nj xi  D i x j  D j
2 wt wt 2 wt
>
Gmm  Gm  Gmk ^ xk  D k  D k  X k ` E 1mj x j  D j @
 Dnm  E > 1
mn E 1
nl xl  D l E 1
mj xj D j . @
 Section 10.4 111

Here, Dk  Dk is added in the first bracket term on the right-hand side. The only
way to satisfy this relation for all xj  Dj involved is to choose the coefficients
such that terms that are independent of xj  Dj, linear and quadratic in xj  Dj are
equal to zero. Hence, we obtain three conditions,
1 1 wE nm
0 E mn  Gmm  Dnm E 1mn , (i)
2 wt
wD i
0
­
x j  D j ® E 1ij
wt
>
 Gm  Gmk D k  X k @E 1
mj
½
¾, (ii)
¯ ¿
­ 1 wE ½
0 xi  D i x j  D j ® E 1ik kn E 1nj  Gmi E 1mj  Dnm E 1ni E 1mj ¾ . (iii)
¯ 2 wt ¿
Here, the subscripts were renamed to enable the combination of these terms or
to write the terms in a more convenient way. Relation (iii) can be written

­ 1 wE ½
0 xi  D i x j  D j ® E 1in nm E 1mj  Gmk E kn E 1ni E 1mj  Dnm E 1ni E 1mj ¾ .
¯ 2 wt ¿
By combining terms that involve E1ij we obtain
1 ­ wE ½
0 xi  D i x j  D j E 1ni E 1mj ® nm  2Gmk E kn  2 Dnm ¾ .
2 ¯ wt ¿
The latter relation is satisfied if Enm satisfies the equation
wE nm
Gnk E km  Gmk E kn  2 Dnm . (A)
wt
The partial derivative by t can be replaced by the regular derivative because the
Enm are only functions of t. The reason for the writing of the first two terms on
the right-hand side is to ensure that these two terms are symmetric like Enm and
Dnm. The validity of equation (A) can be seen by using this expression for the
time derivative of Enm in the previous equation,
1
0 xi  D i x j  D j E 1ni E 1mj ^ Gnk E km  Gmk E kn  2 Dnm  2Gmk E kn  2 Dnm `
2
1 1
 xi  D i x j  D j E 1ni E 1mj Gnk E km  xi  D i x j  D j E 1ni E 1mj Gmk E kn
2 2
1 1
 xi  D i x j  D j E 1ni Gnj  xi  D i x j  D j E 1mj Gmi 0.
2 2
The conclusion that the two terms add to zero can be obtained by replacing in
the last term m by n, i by j, and j by i. Equation (ii) can be written
wD i
0
­
x j  D j ® E 1ij
wt
>
 Gi  Gik D k  X k @E 1
ij
½
¾.
¯ ¿
112 Chapter 10

By combining terms that involve E1ij we obtain

­ wD ½
0 x j  D j E 1ij ® i  Gi  Gik D k  X k ¾ .
¯ wt ¿
This relation is satisfied if Di satisfies the equation
wD i
Gi  Gik D k  X k . (B)
wt
The partial derivative by t can be replaced by the regular derivative. Finally,
there is the question of whether relation (i) is satisfied. The latter is the case, as
it may be seen by using (A),
1 1 wE nm
0 E mn  Gmm  Dnm E 1mn
2 wt
1 1
E mn Gnk E km  Gmk E kn  2 Dnm  Gmm  Dnm E 1mn
2
1
Gnn  Gmm  Gmm 0.
2
c) Equations (A) and (B) require initial conditions for Dk and Ekn. These conditions
are given by
D k (t ' ) x 'k , (C)
E kn (t ' ) 0. (D)

By using the properties of delta functions one may prove that the conditional
PDF f(x, t | x', t') recovers the delta function G(x  x') if Dk and Ekn have the
values given by (C) and (D).

Exercise 10.4.5
a) The Fokker-Planck equation implies equation (10.94) for means,
d Xk
Dk .
dt
By using the relationship Dk = ak we find the equation for means that is implied
by the stochastic differential equation (10.115) to be given by
d Xk
ak X(t ), t .
dt
b) The Fokker-Planck equation implies the equation (10.102) for variances
~ ~
d Xk Xn ~ ~ ~ ~
X n Dk  X k Dn  2 Dnk .
dt
 Section 10.5 113

By using Dk = ak and Dnk = bnm bkm / 2 we find the equation for variances that is
implied by the stochastic differential equation (10.115) to be given by
~ ~
d Xk Xn ~ ~
X n t a~k X(t ), t  X k t a~n X(t ), t  bnm X(t ), t bkm X(t ), t .
dt

Exercise 10.4.6
a) The stochastic differential equation (10.115) is helpful because of two reasons.
First, models are usually developed in terms of stochastic differential equations
that describe the mechanism of process changes. Second, stochastic differential
equations are helpful for the numerical solution of Fokker-Planck equations by
means of Monte Carlo simulation. The numerical solution of such partial dif-
ferential equations represents a significant problem if many variables have to
be considered.
b) The Fokker-Planck equation (10.85) is helpful for the derivation of evolution
equations for statistics of processes (equations for means and variances). Such
equations may be solvable. Fokker-Planck equations are particularly helpful if
it is possible to determine the conditional PDF analytically (see the discussion
in Sect. 10.4.3).

Section 10.5
Exercise 10.5.1
a) According to equation (10.102) combined with Dij = bin bjn / 2 we find for the
stochastic velocity model (10.125) the variance equations
~* ~ *
d Vi (t )Vk (t ) 2 ~* ~ * 4e
 Vi (t )Vk (t )  G ik .
dt W 3W
b) These equations are solved by
~* ~ * 2e § ~ ~ 2e ·
Vi (t )Vk (t ) G ik  ¨ Vi* (0)Vk * (0)  G ik ¸ e  2t / W .
3 © 3 ¹
c) The asymptotic velocity variances are given by
~* ~ * 2e
Vi (t )Vk (t ) G ik .
3
Therefore, cross variances between different velocity components are equal to
zero, and all asymptotic auto-variances (for i = k = 1, 2, 3) have the same value
2 e / 3.
114 Chapter 10

Exercise 10.5.2
a) According to Eq. (10.124) we find for the stochastic velocity model (10.125)
the correlation equations
~* ~ *
d Vi (t )Vk (t  r ) 1 ~* ~ *
 Vi (t )Vk (t  r ) .
dr W
b) These equations are solved by
~* ~ * ~* ~ *
Vi (t )Vk (t  r ) Vi (t )Vk (t ) e  r / W .

c) According to the results of exercise 10.5.1 we find for t o f

~* ~ * 2e
Vi (t )Vk (t  r ) G ik e  r / W .
3
Hence, the correlation function of unequal components is equal to zero for all r.

Exercise 10.5.3
The reason for the development of anisotropy is the appearance of spatial gradi-
ents of mean velocities and triple correlations, which lead to couplings between
velocity components. For the case that there are no such spatial gradients, i.e., for
the case that the total variance derivative is the only nonzero term on the left-hand
side of the variance evolution equation (10.165), we find that initially isotropic
variances will remain isotropic for all time (there will be never couplings between
different velocity components).

Exercise 10.5.4
a) First, it cannot be expected that 50,400 available sample values can provide a
very accurate picture: much more samples are needed to derive very accurate
conclusions: see the discussion in Sect. 4.4.2. This fact matters, in particular, if
we have to deal with rather small moments, as given, e.g., for the <u3 v> case.
There is no obvious explanation for the high <u3 v> / <u3 v>N value. It may be
the case that this value is a consequence of the limited number of samples.
b) The corresponding <> / <>N mean value is given by 1.0667.
c) Overall, the data provide support for the normal parametrization (10.129). The
mean over the <> / <>N values indicates deviations of 7% from the theoretical
value, which is a reasonable finding regarding the limited number of samples
applied. Another fact is that the measurements were performed on a very large
scale compared to the molecular scale. One may expect that the parametrization
(10.129) is more accurate for the molecular scale because the assumption of a
local equilibrium state is better justified than for the scale considered in the
experiments.
 Section 10.5 115

Exercise 10.5.5
a) For this case we have
§U · §0 I · §0 0 ·
a ¨¨ ¸¸, G ¨¨ ¸¸, b ¨ ¸.
¨0 4e /(3W ) I ¸¹
©0¹ © 0  1/W I ¹ ©
Here, 0 and I represent 3 u 3 matrices.
b) The requirement is that U, e and W are independent of the position.

Exercise 10.5.6
a) According to (10.109), the conditional PDF f(x, t | x', t ') is given by the normal
PDF
1 ­ 1 ½
f ( x, t x' , t ' ) exp® E 1ij xi  D i x j  D j ¾.
2S det( E )
3/ 2
¯ 2 ¿

Equations (10.110) for the parameters Di and Eij read for the case considered
dD i dE ij 4
Ui , eW G ij .
dt dt 3
By accounting for the initial conditions (10.111) we find the solutions of these
equations to be given by
4
D i xi '  U it , E ij eW t G ij .
3
We use the expressions obtained for Di and Eij in the conditional PDF,
1 ­ 3 ½
f ( x, t x' , t ' ) exp® xi  xi '  U i t xi  xi '  U i t ¾.
8 S eW t / 3 ¯ 8 eW t
3/ 2
¿
b) The asymptotic conditional PDF reads then
1 ­ 3 ½
f ( x , t x' , t ' ) exp® xi  U i t xi  U i t ¾.
8 S eW t / 3 ¯ 8 eW t
3/ 2
¿
c) The asymptotic conditional PDF is independent of x'. Hence,
1 ­ 3 ½
f ( x, t ) f ( x , t x' , t ' ) exp® xi  U i t xi  U i t ¾.
8 S eW t / 3 ¯ 8 eW t
3/ 2
¿