Documente Academic
Documente Profesional
Documente Cultură
Introduction
1. Introduction to Mathematics
Part A 1
for Mechanical Engineering
Ramin S. Esfandiari
Complex numbers, variables and functions are the main x = Re(z) = −1 and y = Im(z) = 2. A complex num-
focus of this section. We will begin with complex num- ber with zero real part is known as pure imaginary, e.g.,
bers, their representations, as well as properties. The z = 4i. Two complex numbers are said to be equal if
idea is then extended to complex variables and their and only if their respective real and imaginary parts
functions. are equal. Addition of complex numbers is performed
component-wise, that is, if z 1 = x1 + iy1 and z 2 = x2 +
1.1.1 Complex Numbers iy2 , then
A complex number z appears in the rectangular form z 1 + z 2 = (x1 + iy1 ) + (x2 + iy2 )
= (x1 + x2 ) + i(y1 + y2 ) . (1.2)
z = x + iy ,
√ Multiplication of two complex numbers is performed in
i = −1 = imaginary number , (1.1) the same way as two binomials with the provision that
i2 = −1, i3 = −i, i4 = 1, etc. need be taken into account,
where x and y are real numbers, called the real and that is,
imaginary parts of z, respectively, and denoted by
x = Re(z), y = Im(z). For example, if z = −1 + 2i, then z 1 z 2 = (x1 + iy1 )(x2 + iy2 )
= x1 x2 + iy1 x2 + ix1 y2 + i2 y1 y2
Imaginary axis = (x1 x2 − y1 y2 ) + i(x1 y2 + x2 y1 ) . (1.3)
Complex Plane
Since complex numbers consist of a real part and an
imaginary part, they have a two-dimensional character,
and hence may be represented geometrically as points
z = x + iy
y in a Cartesian coordinate system, known as the complex
plane. The x-axis of the complex plane is the real axis,
and its y-axis is called the imaginary axis, (Fig. 1.1).
Noting that z = x + iy is uniquely identified by an or-
dered pair (x, y) of real numbers, we can represent z as
a two-dimensional (2-D) vector in the complex plane,
0 x Real axis with initial point 0 and terminal point z = x + iy; in
other words, the position vector of the point z. The
Fig. 1.1 Geometrical representation of complex numbers –
imaginary number i, for instance, can be identified by
the complex plane
(0, 1). So, the concept of vector addition also applies to
the addition of complex numbers. For that, let us con-
Imaginary axis sider z 1 = −2 + 3i and z 2 = 3 + i in Fig. 1.2. It is then
evident that their sum, z 1 + z 2 = 1 + 4i, is exactly what
4i 1 + 4i
we would obtain by adding the corresponding position
vectors of z 1 and z 2 .
3i –2 + 3i The magnitude of a complex number z = x + iy is
defined as
2i
|z| = x 2 + y2 . (1.4)
we conclude that
Part A 1.1
y
1
x = Re(z) = (z + z̄) ,
z2 2
1
y = Im(z) = (z − z̄) . (1.7)
z1 – z2
2i
and x
z − z̄ = (x + iy) − (x − iy) = 2iy Fig. 1.4 A complex number and its conjugate
6 Part A Fundamentals of Mechanical Engineering
Polar Representation of Complex Numbers The angle θ is measured from the positive real axis and,
Part A 1.1
Although the standard rectangular form is suitable in by convention, is regarded as positive in the sense of
certain instances, it is quite inconvenient in most oth- the counterclockwise (ccw) direction. It is measured in
ers. For example, imagine the simplification of (−2 + radians (rad) and is determined in terms of integer mul-
3i)10 . Situations of this type require a special form tiples of 2π. The specific value of θ that lies in the
that simplifies the complex algebra. The polar form of interval (−π, π] is called the principal value of arg z
a complex number, as suggested by its name, uses the and is denoted by arg z. In engineering analysis, it is
polar coordinates to represent a complex number in the also common to express the polar form of z as
complex plane. Recall that any point in the plane can
be determined by a radial coordinate r and an angular z = r θ (1.14)
coordinate θ. So, the same holds for a complex number where denotes the angle.
z = x + iy = 0 in the complex plane, (Fig. 1.5). The rela-
tionship between the rectangular and polar coordinates Example 1.4: Phase via location
is given by Express z = 2
in polar form.
−1+i
x = r cos θ , y = r sin θ . (1.9)
We first introduce Euler’s formula, Solution. First, express z in standard rectangular form,
as
e = cos θ + i sin θ .
iθ
(1.10)
2 −1 − i −2 − 2i
Then, (1.9) and (1.10) yield z= = = −1 − i ,
−1 + i −1 − i 2
z = x + iy = r cos θ + i (r sin θ) = r eiθ indicating that z is located in the third quadrant of the
In summary, complex plane. Next, we use (1.13) to find
z = r eiθ , (1.11) −1 π
θ = tan−1 = 45◦ = rad .
which is called the polar form of the complex number z. −1 4
Here, the magnitude (or modulus) of z is defined by However, the only information this provides is
√ that the (smallest) angle between OA and the real
r = |z| = x 2 + y2 = z z̄ (1.12)
axis (Fig. 1.6) is 45◦ . Since z is in the third quad-
and the phase (or argument) of z is rant, its actual phase is then 180 + 45 = 225◦ (π +
Im(z) π/4 = 5π/4 rad) if measured in the ccw direction, or
θ = arg z = tan−1 −135◦ (−3π/4 rad) in the clockwise (cw) direction. So,
Re(z)
the polar form of z can be written as
−1 y
= tan . (1.13)
x √ √ 5π
z = −1 − i = 2 ei(5π/4) or z= 2 .
4
Imaginary axis
Multiplication and Division in Polar Form. As cited
earlier, polar form substantially reduces complex alge-
y
y z = x + iy
0
r 45°
y = r sin θ –135°
θ
x A
–i
Real axis z = –1– i
x = r cos θ
–1 0 1 x
Fig. 1.5 Relation between the rectangular and polar forms
of a complex number Fig. 1.6 Example 1.4
Introduction to Mathematics for Mechanical Engineering 1.1 Complex Analysis 7
Part A 1.1
y
two complex numbers z 1 = r1 eiθ1 and z 2 = r2 eiθ2 .
Subsequently, z = r e iθ
z1 r1 r1 0 x
= ei(θ1 −θ2 ) or (θ1 − θ2 ) . (1.17)
z2 r2 r2
Fig. 1.7 A complex number and its conjugate in polar form
so that
z 1 r1 |z 1 |
= Roots of a Complex Number √
z r = |z | and In real calculus, if a is a real number then n a has a sin-
2 2 2
z1 gle value. On the contrary, given a complex number
arg = θ1 − θ2 = arg(z 1 ) − arg(z 2 ) (1.18) z = 0, and
z2 √ a positive integer n, then the nth root of z,
written n √z, is multivalued. In fact, there are n different
values of n z, corresponding to each value of z = 0. For
Complex Conjugation in Polar Form. Given the polar a known z = r eiθ , it can be shown that [1.1, 2]
form of a complex number, z = r eiθ , its conjugate is
obtained as √ √ θ + 2kπ θ + 2kπ
n
z = n r cos + i sin ,
n n
z̄ = x − iy = r cos θ − i(r sin θ)
k = 0, 1, · · · , n − 1 . (1.21)
= r(cos θ − i sin θ)
Geometrically, these n values are described as follows:
r e−iθ .
Euler’s formula
= r[cos(−θ) + i sin(−θ)] =
(1.19) 1. they all lie √
on a circle centered at the origin with
a radius of n r, and
This result makes sense geometrically, since a complex 2. they are the n vertices of an n-sided regular polygon.
number and its conjugate are reflections of one another
through the real axis. Hence, they are equidistant from
Example 1.5: Fourth
√ roots of unity
the origin, that is, |z| = |z̄| = r, and the phase of one
We are seeking 4 z, where z = 1. Noting that z = 1 is
is the negative of the phase of the other, i. e., arg(z) =
on the positive real axis, one unit from the origin, we
− arg(z̄); Fig. 1.7. The important property of complex
conclude that r = 1 and θ = 0, hence z = 1 = 1 ei(0) .
conjugation (1.6) can now be confirmed in polar form,
Following (1.21), we find the four roots to be 1, i, −1,
as
and −i; Fig.
√ 1.8. Note that all four roots lie on a circle
z z̄ = (r eiθ )(r e−iθ ) = r 2 = |z|2 . of radius 4 1 = 1 centered at the origin (the so-called
unit circle), and are the vertices of a regular four-sided
polygon, as asserted.
Integer Powers of a Complex Number
The effectiveness of the polar form may further be
demonstrated when raising a complex number to an 1.1.2 Complex Variables and Functions
integer power. Letting z = r eiθ , then
If x or y or both vary, then z = x + iy is referred to
z n = (r eiθ )n = r n einθ as a complex variable. The most well-known complex
Euler’s formula n
= r (cos nθ + i sin nθ) , (1.20)
variable is the Laplace variable (Sect. 1.3). Letting S be
a set of complex numbers, a function f defined on S
so that Re(z n ) = r n cos nθ and Im(z n ) = r n sin nθ. is a rule, which assigns a complex number w to each
8 Part A Fundamentals of Mechanical Engineering
y
if it is analytic at all points of that domain. Analytic
i functions arise in such areas as fluid flow and complex
i
potentials.
Fig. 1.8 Locations of the fourth roots of unity Consequently, if f (z) is analytic in some domain R,
then the Cauchy–Riemann equations hold at every point
z ∈ S. The notation is w = f (z) and the set S is called of R.
the domain of definition of f . As an example, the do-
Example 1.6: Cauchy–Riemann equations
main of the function w = z/(3 − z) is any region that
Decide whether f (z) = (i − 2)z 2 − 2iz + i is analytic.
does not contain the point z = 3. Because z assumes dif-
ferent values from S, it is clearly a complex variable.
Solution. Inserting z = x + iy, we find
Since w is complex, it must have a real part u and an
imaginary part v, or w = u + iv. Also w = f (z) implies u(x, y) = −2xy − 2x 2 + 2y2 + 2y ,
that w is dependent on z = x + iy. Therefore, w depends
on x and y, which means u and v depend on x and y, or v(x, y) = x 2 − y2 − 4xy − 2x + 1 .
Part A 1.2
Mathematical models of dynamic systems – mechani- 1.2.1 First-Order Ordinary
cal, electrical, electromechanical, liquid-level, etc. – are Differential Equations
represented by differential equations [1.3]. Therefore,
it is imperative to have a thorough knowledge of their First-order ODEs generally appear in the implicit form
basic properties and solution techniques. In this sec-
tion we will discuss the fundamentals of differential F(x, y, y ) = 0 . (1.26)
equations, specifically, ordinary differential equations
(ODEs), and present analytical and numerical methods For example, y + y2 = cos x can be expressed in the
to solve them. Differential equations are divided into above form with F(x, y, y ) = y + y2 − cos x. In other
two general categories: ordinary differential equations cases, the equation may be written explicitly as
and partial differential equations (PDEs). An equation
involving an unknown function and one or more of its y = f (x, y) . (1.27)
derivatives is called a differential equation. When there
An example would be y + 2y = ex where f (x, y)
is only one independent variable, the equation is called
= ex − 2y. A function y = s(x) is a solution of the first-
an ordinary differential equation (ODE). For example,
order ODE in (1.26) on a specified (open) interval if
y + 2y = ex is an ODE involving the unknown func-
it has a derivative y = s (x) and satisfies (1.26) for all
tion y(x), its first derivative y = dy/ dx, as well as
values of x in the given interval. If the solution is in
a given function ex . Similarly, xy − yy = sin x is an
the form y = s(x), then it is called an explicit solu-
ODE relating y(x) and its first and second derivatives
tion. Otherwise, it is in the form S(x, y) = 0, which is
with respect to x, as well as the function sin x. While
known as an implicit solution. For example, y = 4 e−x/2
dealing with time-varying functions – as in many phys-
is an explicit solution of 2y + y = 0. It turns out that
ical applications – the independent variable x will be
a single formula y = k e−x/2 involving a constant k = 0
replaced by t, representing time. In that case, the rate
generates all solutions of this ODE. Such formula is
of change of the quantity y = y(t) with respect to the
referred to as a general solution, and the constant is
independent variable t is denoted by ẏ = dy/ dt. If the
known as the parameter. When a specific value is
unknown function is a function of more than one inde-
assigned to the parameter, a particular solution is ob-
pendent variable, e.g., u(x, y), the equation is referred
tained.
to as a partial differential equation. The derivative of the
highest order of the unknown function y(x) with respect
Initial-Value Problem (IVP)
to x is the order of the ODE; for instance, y + 2y = ex
A first-order initial-value problem (IVP) appears in the
is of order one and xy − yy = sin x is of order two.
form
Consider an nth-order ordinary differential equation in
the form y = f (x, y) , y(x0 ) = y0 , (1.28)
an y (n)
+ an−1 y (n−1)
+ · · · + a1 y + a0 y = g(x) ,
where y(x0 ) = y0 , is called the initial condition.
(1.25)
Separable First-Order Ordinary Differential f (x) ≡ 0, then the ODE is called homogeneous, other-
Part A 1.2
Solution. The ODE is separable and treated as Solution. Noting that t is now the independent vari-
able, we first rewrite the ODE to agree with the form
dy of (1.31), as
ex = y2
dx
1 1 1
Provided that y = 0
⇒ dy = dx ẏ + y = 2 e2t
y 2 ex 2
1 so that
⇒ − = − e−x + c
y 1
g = , f = 2 e2t .
(c = const.) 2
Solve for y 1
⇒ y(x) = −x , With h = g(t) dt = 12 dt = 12 t, the general solution is
e −c given by (1.32),
which is the general solution to the original differential
−t/2
equation. The specific value of c is determined via the y(t) = e e · 2 e dt + c
t/2 2t
given initial condition, as
⎫Initial condition 4
= e−t/2 2 e5t/2 dt + c = e2t + c e−t/2 .
y(0) = 1 ⎪ ⎬ 5
1
⇒ =1⇒c=0.
⎪
1 ⎭ 1 − c Applying the initial condition, we find y(0) = 45 +
y(0) = 1−c By gen. solution c = 1 ⇒ c = 15 . The particular solution is y(t) = 45 e2t +
1 −t/2
Substitution into the general solution yields the particu- 5e .
lar solution y(x) = ex .
1.2.2 Numerical Solution of First-Order
Linear First-Order Ordinary Ordinary Differential Equations
Differential Equations
A differential equation that can be expressed in the form Recall that a first-order ODE can appear in an implicit
form F(x, y, y ) = 0 or an explicit form y = f (x, y).
y + g(x)y = f (x) , (1.31) We will consider the latter, and assume that it is subject
where g and f are given functions of x, is called a lin- to a prescribed initial condition, that is,
ear first-order ordinary ODE. This of course agrees with y = f (x, y) , y(x0 ) = y0 , x0 ≤ x ≤ x N . (1.33)
what was discussed in (1.25) with slight changes in
notation. If f (x) = 0 for every x in the interval under If finding a closed-form solution of (1.33) is difficult
consideration, that is, if f is identically zero, denoted or impossible, we resort to a numerical solution. What
Introduction to Mathematics for Mechanical Engineering 1.2 Differential Equations 11
this means is that we find approximate values for the Further inspection reveals that RK4 produces the exact
Part A 1.2
solution y(x) at several points values (at least to five-decimal place accuracy) of the
x1 = x0 + h , x2 = x0 + 2h · · · xn solution at the mesh points.
= x0 + nh , · · · , x N = x0 + Nh
1.2.3 Second- and Higher-Order, Ordinary
known as mesh points, where h is called the step size. Differential Equations
Note that the mesh points are equally spaced. Among
many numerical methods to solve (1.33), the fourth- The application of basic laws such as Newton’s second
order Runge–Kutta method is most commonly used in law and Kirchhoff’s voltage law (KVL) leads to math-
practice. The difference equation for the fourth-order ematical models that are described by second-order
Runge–Kutta method (RK4) is derived as [1.5, 6] ODEs [1.3]. Although it is quite possible that the sys-
1 tem models contain nonlinear elements, in this section
yn+1 = yn + (q1 + 2q2 + 2q3 + q4 ) , (1.34) we will mainly focus on linear second-order differen-
6
n = 0, 1, · · · , N − 1 , tial equations. Nonlinear systems may be treated via
numerical techniques such as the fourth-order Runge–
where Kutta method (Sect. 1.2), or via linearization [1.3]. In
q1 = h f (xn , yn ) agreement with (1.25), a second-order ODE is said to
be linear if it can be expressed in the form
h q1
q2 = h f xn + , yn + ,
2 2 y + g(x)y + h(x)y = f (x) , (1.35)
h q2 where f , g, and h are given functions of x. Otherwise,
q3 = h f xn + , yn + ,
2 2 it is called nonlinear.
q4 = h f (xn + h, yn + q3 ) .
Homogeneous Linear Second-Order ODEs
If y1 and y2 are two solutions of the homogeneous linear
Example 1.11: Fourth-order Runge–Kutta method
ODE
Apply RK4 with step size h = 0.1 to solve y + y= 2x 2 ,
y(0) = 3, 0 ≤ x ≤ 1. y + g(x)y + h(x)y = 0 (1.36)
on some open interval, their linear combination
Solution. Knowing that f (xn , yn ) = −yn + 2xn2 , the y = c1 y1 + c2 y2 (c1 , c2 constants) is also a solution on
four function evaluations/step of the RK4 are the same interval. This is known as the principle of
q1 = h − yn + 2xn2 , superposition.
1 1 2 General Solution of Linear Second-Order
q2 = h − yn + q1 + 2 xn + h ,
2 2 ODEs – Linear Independence
1 1 2 A general solution of (1.36) is based on the idea of lin-
q3 = h − yn + q2 + 2 xn + h , ear independence of functions, which involves what is
2 2
known as the Wronskian. We first mention that a 2 × 2
q4 = h − (yn + q3 ) + 2(xn + h)2 . determinant (Sect. 1.5.1) is evaluated as
Upon completion of each step, yn+1 is calculated p q
= ps − qr .
by (1.34). So, we start with n = 0, corresponding to r s
x0 = 0 and y0 = 3, and continue the process up to
If each of the functions y1 (x) and y2 (x) has at least
n = 10. Numerical results are generated as
a first derivative, then their Wronskian is denoted by
y(0) = 3 , W(y1 , y2 ) and is defined as the 2 × 2 determinant
y(0.1) = 2.7152 ,
y1 y2
y(0.2) = 2.4613 , W(y1 , y2 ) = = y1 y2 − y2 y1 . (1.37)
y1 y2
y(0.3) = 2.2392 , · · · ,
If there exists a point x ∗ ∈ (a, b) where W = 0, then y1
y(0.9) = 1.6134 , and y2 are linearly independent on the entire interval
y(1) = 1.6321 . (a, b).
12 Part A Fundamentals of Mechanical Engineering
Example 1.12: Independent solutions – the Wronskian Since eλx = 0 for any finite values of x and λ, then
Part A 1.2
Part A 1.2
1 Nonhomogeneous second-order ODEs appear in the
σ = a1 , form
2
1
ω= 4a2 − a12 . (1.43) y + g(x)y + h(x)y = f (x) ,
2
The two independent solutions are y1 = e−σ x cos ωx f (x) ≡ 0 . (1.46)
and y2 = e−σ x sin ωx, and a general solution of (1.39) A general solution for this equation is then obtained as
is obtained as
y(x) = yh (x) + yp (x) . (1.47)
y(x) = e−σ x (c1 cos ωx + c2 sin ωx) . Homogeneous solution Particular solution
General solution — λ1 =λ̄2 , complex conjugates
(1.44)
Homogeneous Solution yh (x). yh (x) is a general so-
Example 1.15: Case (3) lution of the homogeneous equation (1.36), and as
Solve y + 2y + 2y = 0, y(0) = 1, y (0) = 0. previously discussed, it is given by
aid of Table 1.1. This choice involves unknown coeffi- a homogeneous solution associated with a double root.
Part A 1.2
cients, which will be determined by substituting yp and Therefore, by special case II the modified choice is
its derivatives into (1.48). The details, as well as special Kx 2 e−x . Consequently, the particular solution is in the
cases that may occur, are given below. form
yp (x) = K 1 x + K 0 + K x 2 e−x .
Procedure. First term Second term
Step 1: Homogeneous Solution yh (x). Solve the homo- Substitution of yp and its derivatives into the nonhomo-
geneous equation y + a1 y + a2 y = 0 to find the two geneous ODE, and collecting terms, results in
independent solutions y1 and y2 , and the general solu-
tion yh (x) = c1 y1 (x) + c2 y2 (x). 2K e−x + K 1 x + K 0 + 2K 1 = x + 1 + 3 e−x .
Equating the coefficients of like terms, we have
Step 2: Particular Solution yp (x). For each term in f (x)
2K = 3 K = 32
choose a proper yp as suggested by Table 1.1. For in-
stance, if f (x) = x + 2 ex then pick yp = K 1 x + K 2 + K1 = 1 ⇒ K1 = 1
K ex . Note that, if instead of x we had 3x − 2, for ex- K 0 + 2K 1 = 1 K 0 = −1
ample, the choice of yp would still be the same because 3
they both represent first-degree polynomials. We then ⇒ yp (x) = x − 1 + x 2 e−x .
2
substitute our choice of yp , along with its derivatives,
into the original ODE to find the undetermined coeffi-
cients. Step 3: General Solution. The general solution is then
found as
3
Special cases. y(x) = (c1 + c2 x) e−x + x − 1 + x 2 e−x .
I. Suppose a term in our choice of yp coincides with 2
a solution (y1 or y2 ) of the homogeneous equation,
Step 4: Initial Conditions. Applying the initial condi-
and that this solution is associated with a simple
tions, we obtain c1 = 2 and c2 = 1. Finally, the solution
(i. e., nonrepeated) characteristic value. Then, make
to the IVP is
the modification by multiplying yp by x.
3
II. If a term in the choice of yp coincides with a so- y(x) = (2 + x) e−x + x − 1 + x 2 e−x .
lution of the homogeneous equation, and that this 2
solution is associated with a repeated characteristic
value, modify by multiplying yp by x 2 . Higher-Order Ordinary Differential Equations
Many of the techniques for the treatment of differential
Example 1.16: Special case II equations of order three or higher are merely extensions
Solve of those applied to second-order equations. Here we will
only discuss nth-order, linear nonhomogeneous ODEs
y + 2y + y = x + 1 + 3 e−x , y(0) = 1 , with constant coefficients, that is,
y (0) = 0 . y(n) + an−1 y(n−1) + · · · + a1 y + a0 y = f (x) ,
(1.49)
Step 1: Homogeneous Solution. The characteristic
equation (λ + 1)2 = 0 yields a double root λ = −1. This where a0 , a1 , · · · , an−1 are constants. As in the case
means y1 = e−x and y2 = x e−x , so that the homoge- of second-order ODEs, a general solution consists of
neous solution is yh (x) = (c1 + c2 x) e−x . the homogeneous solution and the particular solution.
For cases when f (x) is of a special type, the particu-
Step 2: Particular Solution. The right-hand side of the lar solution is obtained via the method of undetermined
ODE consists of two functions, coefficients.
an independent homogeneous solution, then no modi- Step 2: Particular Solution. Noting that f (x) = 1 + 12x
Part A 1.3
fication is necessary. Otherwise, the following special is a first-degree polynomial, we pick yp = K 1 x + K 0 .
cases need be taken into account. But x happens to be a homogeneous solution associated
with a double root (λ = 0). Hence, the modification is
Special Cases. yp = (K 1 x + K 0 )x 2 . Substituting this and its derivatives
1. If a term in our choice of yp coincides with a homo- into the original ODE, and simplifying, we arrive at
geneous solution, which corresponds to a simple
(6K 1 − 8K 0 ) − 24K 1 x = 1 + 12x
(nonrepeated) characteristic value, then we make
the modification by multiplying yp by x. 6K 1 − 8K 0 = 1 K = − 12
⇒ ⇒ 1
2. If a term in yp coincides with a solution of the −24K 1 = 12 K 0 = − 12
homogeneous equation, and this solution is associ- 1
ated with a characteristic value of multiplicity m, we ⇒ yp = − (x + 1)x 2
modify by multiplying yp by x m . 2
1.3.1 Inverse Laplace Transform and a1 and a2 are constant scalars, then
Part A 1.3
0
∞ ∞ Laplace
e−(s+a)t
= e−(s+a)t dt = transformation
−(s + a) t=0 Time domain
X(s) = L [x(t)]
s domain
0
Algebraic
1 ODE in x(t)
= equation
s+a in X(s)
for s + a > 0. x(t) = L–1 [X(s)]
Inverse Laplace
Linearity of Laplace transformation
and Inverse Laplace Transforms
The Laplace transform operator L is linear, that is, if the Fig. 1.9 Operations involved in the Laplace transformation
Laplace transforms of functions f 1 (t) and f 2 (t) exist, method
Introduction to Mathematics for Mechanical Engineering 1.3 Laplace Transformation 17
Table 1.2 Laplace transform pairs Table 1.2 Laplace transform pairs, continued
Part A 1.3
No. f (t) F(s) No. f (t) F(s)
√
1 Unit impulse δ(t) 1 1
π
32 1
e−at + 2 e 2 at sin 2 at − 6
3 1
2 1, unit step u s (t) 1/s 3a2 s 3 −a3
√
3 t, unit ramp u r (t) 1/s2 1
π
33 1
− e−at + 2 e 2 at sin 2 at + 6
3 s
4 δ(t − a) e−as
3a s 3 −a3
√
5 u(t − a) e−as /s 1
34 1
eat − 2 e− 2 at sin 3 π
2 at + 6
1
6 t n−1 , n = 1, 2, . . . (n − 1)!/sn 3a2 s 3 −a3
√
7 t a−1 , a > 0 Γ (a)/sa 1
35 1
e−at + 2 e− 2 at sin 3 π
2 at − 6
s
3a s 3 −a3
8 e−at 1
s+a
36 1
(cosh at sin at − sinh at cos at) 1
9 t e−at 1 4a3 s 4 +4a4
(s+a)2 1 s
37 sinh at sin at
10 t n e−at , n = 1, 2, . . . n! 2a2 s 4 +4a4
(s+a)n+1
−at
38 1
(sinh at − sin at) 1
11 1
b−a ( e − e−bt ) , a = b 1
(s+a)(s+b)
2a3 s 4 −a4
39 1
(cosh at − cos at) s
12 1 −at − b e−bt ) , a = b s 2a2 s 4 −a4
a−b (a e (s+a)(s+b)
13 1
ab 1 + a−b
1
(b e−at − a e−bt ) 1
s(s+a)(s+b) See Fig. 1.10. Alternatively, in terms of the inverse
14 1
(−1 + at + e−at ) 1 Laplace transform,
a2 s 2 (s+a)
15 1
(1 − e−at − at e−at ) 1 L−1 [F(s + a)] = e−at f (t) . (1.54)
a2 s(s+a)2
18 e−σt sin ωt ω Solution. Let f (t) = cos t so that F(s) = s/(s 2 + 1); see
(s+σ)2 +ω2 Table 1.2. Then, by (1.53) with a = −3,
ω2
19 e−σt cos ωt f (t)=cos t s−3
s(s 2 +ω2 )
L[ e3t cos t] = F(s − 3) =
20 1 − cos ωt s+σ a=−3 (s − 3)2 + 1
(s+σ)2 +ω2
ω3
21 ωt − sin ωt Differentiation and Integration
s 2 (s 2 +ω2 )
24 1
(sin ωt − ωt cos ωt) 1
2ω3 (s 2 +ω2 )2
s2 F (s)
2ω (sin ωt + ωt cos ωt)
1
25
(s 2 +ω2 )2
(Assuming a > 0)
26 1 1
ω2 sin ω2 t − ω1 sin ω1 t , 1
ω22 −ω21 1 s 2 +ω21 s 2 +ω22 a
ω21 = ω22
27 1
(cos ω1 t − cos ω2 t) , s
ω22 −ω21 s 2 +ω21 s 2 +ω22
a = b 0
s
31 1
[cosh at − cosh bt] , a = b s
a2 −b2 (s 2 −a2 )(s 2 −b2 )
Fig. 1.10 Shift on the s-axis (Theorem 1.1)
18 Part A Fundamentals of Mechanical Engineering
we assume that f (t) is such that F(s) = L[ f (t)] is Example 1.23: Theorem 1.3
Part A 1.3
either known directly from Table 1.2 or can be de- Show that
termined by other means. Either way, once F(s) is
sin ωt s
available, the two transforms labeled (1) and (2) will L = cot−1 .
be obtained in terms of the derivative and integral of t ω
F(s), respectively. Before presenting two key results
pertaining to these situations we make the follow- Solution. Comparing with (1.58), f (t) = sin ωt so that
ing definition. If a transform function is in the form F(s) = ω/(s2 + ω2 ). Subsequently,
F(s) = N(s)/D(s), then each value of s for which
D(s) = 0 is called a pole of F(s). A pole with a mul- ∞
sin ωt ω
tiplicity (number of occurrences) of one is known as L = dσ
a simple pole. t σ 2 + ω2
s
∞
Theorem 1.2: Differentiation of Laplace Transforms. 1 dσ
=
If L[ f (t)] = F(s) exists, then at any point except at the 1 + (σ /ω)2 ω
s
poles of F(s), we have σ ∞
d = tan−1
L[t f (t)] = − F(s) = −F (s) (1.55) ω σ=s
ds π s s
= − tan−1 = cot−1 .
or alternatively, 2 ω ω
Part A 1.3
u (t)
L[ f (t)] exists, then
L[ f (t − a)u(t − a)] = e−as F(s) , (1.63)
1 or, alternatively,
L−1 [ e−as F(s)] = f (t − a)u(t − a) . (1.64)
When the unit step function occurs at some time a = Physically, this models a signal that changes linearly
0 (Fig. 1.12), it is denoted by u(t − a), and with a unit rate. By (1.50),
⎧ ∞
⎪
⎪ if t > a
⎨1 L[u r (t)]
Notation
= Ur (s) = t e−st dt
u(t − a) = 0 if t < a . (1.62)
⎪
⎪ 0
⎩ ∞ ∞ −st
undefined (finite) if t = a
e−st e
= t − dt
As before, if the magnitude happens to be A = 1, the −s t=0 −s
0
notation is modified to Au(t − a). To find the Laplace −st ∞
transform of u(t − a), we first need to discuss the shift e 1
= = 2. (1.66)
on the t-axis, see Theorem 1.4 below. −s t=0 s
2
u(t - a) ur (t)
1
1
1
0 0
0 a t 0 t
Fig. 1.12 The unit -step function occurring at t = a Fig. 1.13 The unit ramp function u r (t)
20 Part A Fundamentals of Mechanical Engineering
Note that u r (t) = tu(t). When the rate is A = 1, the sig- If the area is A = 1, the signal is called a pulse, written
Part A 1.3
Unit Pulse up (t). The unit pulse function (Fig. 1.14) is Unit Impulse (Dirac Delta) δ(t). Consider the unit
defined as ⎧ pulse of Fig. 1.14 and let t1 → 0; Fig. 1.15. In this
⎨1/t if 0 < t < t1 limit, the rectangular-shaped signal occupies a re-
1
u p (t) = gion with an infinitesimally small width and a large
⎩0 if t < 0 and t > t1 . height (Fig. 1.16). The area, however, remains unity
The word ‘unit’ signifies that the signal occupies an area throughout the process. This limiting signal is known
of unity. Its Laplace transform is derived as as the unit impulse (or Dirac delta), denoted by
t1 δ(t). If the area is A = 1, it is an impulse, de-
Notation 1 −st
L[u p (t)] = Up (s) = e dt noted by Aδ(t). If an external disturbance (such as
t1 an applied force or voltage) is a pulse with very
0
large magnitude and applied for a very short period
1 − e−st1 of time, then it can be approximated as an im-
= . (1.67)
st1 pulse. Since δ(t) is the limit of u p (t) as t1 → 0, we
up (t) δ(t)
1/t1
Area = 1
Area = 1
0 0
0 t1 t 0 t
Fig. 1.14 The unit pulse u p (t) Fig. 1.16 The unit impulse δ(t)
up (t) δ(t - τ)
1/t1
Area = 1
Area = 1
0 0
0 t1 t 0 t=τ t
Fig. 1.15 The unit pulse as t1 → 0 Fig. 1.17 The unit impulse occurring at t = τ
Introduction to Mathematics for Mechanical Engineering 1.3 Laplace Transformation 21
have In general,
Part A 1.3
1 − e−st1
Notation
L[δ(t)] = Δ(s) = lim L[ f (n) (t)] = sn F(s) − sn−1 f (0) − sn−2 f˙(0)
t1 →0 st1
−st1 − · · · − f (n−1) (0) . (1.73)
L’Hôspital’s rule se
= lim =1. (1.68)
t1 →0 s
If the unit impulse occurs at some time t = τ (Fig. 1.17) Theorem 1.6: Laplace transform of integrals
it is represented by δ(t − τ), and If F(s) = L[ f (t)], then
t
L[δ(t − τ)] = e−τs . (1.69) 1
L f (τ) dτ = F(s) . (1.74)
s
This signal has the property ∞δ(t − τ) = 0 for t = τ, 0
δ(t − τ) = ∞ for t = τ, and −∞ δ(t − τ) dt = 1. It also
has the filtering property, Alternatively,
∞ t
−1 1
f (τ)δ(t − τ) dτ = f (t) . (1.70) L F(s) = f (τ) dτ . (1.75)
s
−∞ 0
1.3.3 Laplace Transform of Derivatives Solving Initial-Value Problems. The role of the
and Integrals Laplace transforms of derivatives and integrals of time-
varying functions is most significant when solving
Since engineering systems are generally modeled by an initial-value problem. Schematically, the solution
differential equations of various orders, we need to method is as in Fig. 1.18.
have knowledge of the Laplace transform of deriva-
Example 1.24: Second-order IVP
tives of different orders. In other occasions, the system
may be described by an equation that contains not only Solve ẍ + 2ẋ + x = 0, x(0) = 1, ẋ(0) = 1.
derivatives, but also integrals; for instance, a circuit in-
Solution. Laplace transformation results in
volving a resistor, an inductor, and a capacitor (RLC
circuit) [1.3]. We will also present a systematic ap- [s2 X(s) − sx(0) − ẋ(0)] + 2[sX(s) − x(0)] + X(s)
proach for solving initial-value problems.
Solve for X(s) s+3
=0 ⇒ X(s) = .
Theorem 1.5: Laplace transform of derivatives (s + 1)2
If F(s) = L[ f (t)], then Before inversion, we rewrite this last expression as
L[ f˙(t)] = sF(s) − f (0) (1.71) s+3 (s + 1) + 2
X(s) = =
and (s + 1)2 (s + 1)2
1 2
L[ f¨(t)] = s2 F(s) − s f (0) − f˙(0) . = + .
(1.72) s + 1 (s + 1)2
Algebraic equation
Initial-value problem Laplace transformation using initial conditions
in terms of
x(t) = dependent variable X(s) = transform of x(t)
where the constants Ak , · · · , A1 are determined as in Theorem 1.7: Convolution. Let G(s) = L[g(t)], H(s) =
Part A 1.3
case 1. As an example, we first write L[h(t)], and F(s) = G(s)H(s). Then,
4s + 7 L−1 [F(s)] = f (t) = (g ∗ h)(t)
X(s) =
(s + 1)2 (s + 4) t
A2 A1 A3
= + + = g(τ)h(t − τ) dτ
(s + 1)2 s + 1 s+4
Double pole; case 2 Simple pole; case 1 0
Case 4: Repeated Irreducible Polynomial Physical systems are often subjected to external distur-
(s2 + as + b)k . The fractions are formed as bances that exhibit repeated behavior over long periods
Bk s + Ck B2 s + C2 B1 s + C1
+· · ·+ 2 + .
(s2 + as + b)k (s + as + b)2 s2 + as + b f (t)
P= 2
Convolution Method
1
In systems analysis, the problem of determining the
time history of a function often comes down to
L−1 [G(s)H(s)], where the inverse Laplace transforms
of G(s) and H(s) are known. The convolution method 0
allows us to determine L−1 [G(s)H(s)] using knowledge
of g(t) and h(t). 0 1 2 3 4 t
Notation
Notation: L−1 [G(s)H(s)] = (g ∗ h)(t) is read
“convolution of g and h” Fig. 1.19 Periodic function of Example 1.27
24 Part A Fundamentals of Mechanical Engineering
of time. A function f (t) is called periodic with period Solution. It is evident that the period is P = 2. With
Part A 1.4
P > 0 if it is defined for all t > 0, and f (t + P) = f (t) this, the integral in (1.78) is
for all t > 0.
2
It can then be shown [1.1] that the Laplace transform
of this function is e−st f (t) dt
0
P
1 1
F(s) = e−st f (t) dt . (1.78) f (t)=1 for 0<t<1
1 − e−Ps = e−st dt = (1 − e−s )/s .
0 f (t)=0 for 1<t<2
0
Example 1.27: Periodic signal Then, by (1.78), F(s) = (1 − e−s )/(s(1 − e−2s )). Not-
Find the Laplace transform of the periodic function in ing that1 − e−2s = 1 − ( e−s )2 = (1 − e−s )(1 + e−s ), the
Fig. 1.19. above expression reduces to F(s) = 1/(s(1 + e−s )).
P/2
2
a0 = f (x) dx , (1.80)
P
−P/2 0
P/2
2 2nπx
an = f (x) cos dx ,
P P –1 0 1 x
−P/2
n = 1, 2, 3, · · · , (1.81) Fig. 1.20 A periodic function with period P = 2
Introduction to Mathematics for Mechanical Engineering 1.4 Fourier Analysis 25
Example 1.28: Fourier series Collect terms 1 2 1
Part A 1.4
Find the Fourier series representation of the periodic = − cos πx + cos 3πx + · · ·
4 π2 9
function whose description in one period is shown in
1 1 1
Fig. 1.20. + sin πx − sin 2πx + sin 3πx − · · · .
π 2 3
Solution
1 1 The third and ninth partial sums, together with the orig-
By (1.80), we have a0 = f (x) dx = x dx = 1/2. For inal function, are shown in Fig. 1.21. As mentioned
−1 0 earlier, at the points of discontinuity of f (x) the partial
n = 1, 2, 3, · · ·, (1.81) and (1.82) yield
sums assume the average value of the left- and right-
1 1 hand limits, that is, 1/2.
an = f (x) cos nπx dx = x cos nπx dx
−1 0
1.4.2 Fourier Transformation
1
= (cos nπ − 1) In Sect. 1.3 the notation F(s) was used to represent the
(nπ)2 ⎧ Laplace transform of f (t), i. e., F(s) = L[ f (t)]. Simi-
1 ⎨0 if n = even larly, fˆ(ω) is used to denote the Fourier transform of
= [(−1) n
− 1] = f (t), that is, fˆ(ω) = F[ f (t)]. Since ω is complex in gen-
(nπ)2 ⎩− 2
if n = odd ,
(nπ)2 eral, fˆ(ω) is expected to be complex-valued as well.
1 1 With this, we then write f (t) = F−1 [ fˆ(ω)] describing
bn = f (x) sin nπx dx = x sin nπx dx the inverse Fourier transform of fˆ(ω). We define the
Fourier transform pair as
−1 0
1 1 ∞
1
= − [x cos nπ]10 + [sin nπx]10 ˆf (ω) = √ f (τ) e−iωτ dτ , (1.83)
nπ (nπ)2 2π
1 −∞
= (−1)n+1 . ∞
nπ 1
Equation (1.79) gives the Fourier series of f (x), as f (t) = √ fˆ(ω) eiωt dω . (1.84)
2π
1 2 1 1 −∞
− 2 cos πx + sin πx − sin 2πx Fourier transformation can be thought of as a map-
4 π π 2π
2 1 ping that assigns to a given function of time t an
− 2 cos 3πx + sin 3πx + · · · integral function of frequency ω. In general, any trans-
9π 3π
f (x)
1.2
1.0
9th partial sum
0.8
0.6
0.4
3rd partial sum
0.2
– 0.2
–1.5 – 1.0 – 0.5 0 0.5 1.0 1.5
x
Fig. 1.21 Example 1.28
26 Part A Fundamentals of Mechanical Engineering
Table 1.3 Fourier transform pairs formation with this type of property is known as an
Part A 1.5
⎧
⎨ eat b < t < b Solution. By (1.83),
1 2 (a−iω)b2 − e(a−iω)b1
5 √1 e
⎩0 otherwise
2π a−iω
∞
1
6 e−a|t| , a > 0 2 a fˆ(ω) = √ f (τ) e−iωτ dτ
π ω2 +a2 2π
⎧ −∞
⎨− e−at t < 0
7 ,a < 0 2 −iω ∞
⎩ eat π ω2 +a2 1
t>0 =√ e−aτ e−iωτ dτ
⎧ 2π
⎨ eiat −b < t < b 0
8
⎩0
2 sin(ω−a)b
π ω−a 1 −1 −(a+iω)τ ∞
otherwise =√ e
⎧ 2π a + iω 0
⎨ eiat b < t < b i(a−ω)b1 − ei(a−ω)b2 1 1
9
1 2
√i e =√ .
⎩0 otherwise
2π a−ω
2π a + iω
π e−a|ω|
1
,a > 0
10
a2 +t 2
⎧
2 a Using fˆ(ω) above in (1.84), we find
⎨t 0 < t < b −ibω (1+ibω)
11 √1 −1+ e ∞
⎩0 otherwise 2π ω2
1 1 1
⎧ f (t) = √ √ eiωt dω
⎪
⎪ 0<t<b 2π 2π a + iω
⎨t ibω − e−2ibω
−∞
12 2t − b b < t < 2b √1 −1+2 e ∞
⎪
⎪ 2π ω2
1 1
⎩
0 otherwise = eiωt dω .
2π a + iω
13
2
e−at , a > 0
2
√1 e−ω /(4a) −∞
2a
2 /(4a) √ −aω2
14 e−t ,a > 0 2a e This is known as the complex Fourier integral represen-
tation of the function under consideration.
1.5.1 Vectors and Matrices of A. These diagonal elements form the main diagonal
Part A 1.5
of A. The diagonal directly below the main is known
An n-dimensional vector v is an ordered set of n as the subdiagonal and the one above the main is called
scalars, and is written as v = (v1 , v2 , · · · , vn ). Each vi the superdiagonal. Two matrices A = [aij ] and B = [bij ]
(i = 1, 2, · · · , n) is called a component of the vector v. are said to be equal if they have the same size, and the
In a general sense, an n-dimensional vector helps us lo- same entries in the respective locations. If some rows
cate a point in an n-dimensional space, regardless of or columns (or possibly both) of A are deleted, the out-
the nature of its components or perhaps what physical come is a submatrix of A. If no rows or columns of A are
quantity each may represent. omitted, we have A as a submatrix of itself. Submatri-
ces play important roles in such areas of matrix analysis
Matrices as determinants and rank.
A collection of numbers (real or complex) or possibly
functions, arranged in a rectangular array and enclosed Matrix Operations
by brackets, is referred to as a matrix. Each of the ele- Matrices of the same size can be added. The result,
ments in a matrix is called an entry (or element) of the or the sum, is a matrix of the same size. If A = [aij ]
matrix. The horizontal and vertical lines are referred to and B = [bij ] are m × n, their sum C = [cij ] is also
as rows and columns of the matrix, respectively. A ma- m × n. Matrix addition is performed entry-wise, that is,
trix is called a row vector if it consists of one row only, the entry of C in the (i, j) slot is the sum of the entries
and a column vector if it has one column only. The num- of A and B in that same slot. The m × n zero matrix, de-
ber of rows and columns of a matrix determine the size noted by 0m×n , is an m × n matrix all of whose entries
of that matrix. If a matrix A has m rows and n columns, are zero. If A = [aij ] is m × n and k is a scalar, then kA
then it is said to be of size m × n. If the number of rows is an m × n matrix whose entries are those of A mul-
and columns are the same, we speak of a square matrix, tiplied by k in every slot, that is, kA = [kaij ]m×n . Let
otherwise, a rectangular matrix. We denote matrices by A = [aij ]m×n and B = [bij ]n× p . It is important to note
bold-faced capital letters, such as A. The abbreviated that the number of columns of A is n, which is equal to
form of an m × n matrix is the number of rows of B. Then, their product C = AB
is m × p whose entries are obtained as
A = [aij ]m×n , i = 1, 2, · · · , m , j = 1, 2, · · · , n ,
where aij is known as the (i, j) entry of A, located "
n
cij = aik bk j , i = 1, 2, · · · , m , j = 1, 2, . . . , p .
at the intersection of the ith row and the jth column
k=1
of A so that a12 , for instance, occupies the entry at
(1.85)
which the first row and the second column meet. In the
event that A is a square matrix (m = n), the elements This is shown schematically in Fig. 1.22. If the number
a11 , a22 , · · · ,ann are referred to as the diagonal entries of columns of A does not match the number of rows of
b21 b22 b2 j b2 p
=
ith row ai1 ai 2 ain ci1 ci2 cij cip
(i, j ) entry
bn1 bn2 bnj bnp
B, the product is undefined. If the product is defined, Example 1.31: Special matrices
Part A 1.5
then to get the (i, j) entry of C, we proceed as follows: Matrices U, L, and D are upper triangular, lower trian-
the ith row of A is clearly a 1 × n vector. The jth column gular, and diagonal, respectively:
of B is an n × 1 vector, hence these two vectors have the ⎛ ⎞ ⎛ ⎞
same number of components, n. In these two vectors, −2 1 2 1 0 0
⎜ ⎟ ⎜ ⎟
multiply the first components, the second components, U=⎝ 0 5 0⎠ , L=⎝2 0 0⎠ ,
etc., up to the nth components. Then add the individual 0 0 3 4 7 −1
products together. The result is cij . ⎛ ⎞
3 0 0
⎜ ⎟
D = ⎝ 0 −4 0 ⎠ .
Example 1.30: Matrix Multiplication
Find 0 0 1
⎛
⎞
# $ −2 −1 4 Note that in U and L zeros are allowed along the main
1 −2 3 ⎜ ⎟ diagonal. In fact, the main diagonal may consist of all
AB = ⎝ 1 2 0⎠ .
0 1 4 2×3 zeros. On the other hand, D may have one or more zero
3 5 1 3×3 diagonal elements, as long as they are not all zeros. In
the event that all entries of an n × n matrix are zeros, it
Solution. We first note that the operation is valid be- is called the n × n zero matrix 0n×n .
cause A has three columns and B has three rows. And,
AB will be 2 × 3. Following the strategy outlined above,
Determinant
we find the product as
The determinant of a square matrix A = [aij ]n×n is a real
scalar denoted by |A| or det(A). For the most trivial case
AB = 1·(−2)+(−2)·1+3·3 1·(−1)+(−2)·2+3·5
0·(−2)+1·1+4·3
0·(−1)+1·2+4·5 of n = 1, A = [a11 ], and we define the determinant sim-
1·4+(−2)·0+3·1 ply as |A| = a11 . For n ≥ 2, the determinant is defined
0·4+1·0+4·1
# $ as
5 10 7 using the i-th row
= .
13 22 4 2×3 "
n
|A| = aik (−1)i+k Mik , i = 1, 2, · · · , n (1.89)
k=1
Matrix Transpose
or
Given an m × n matrix A, its transpose, denoted by AT ,
using the j-th column
is an n × m matrix with the property that its first row
is the first column of A, its second row is the second "
n
column of A, and so on. Given that all matrix operations |A| = ak j (−1)k+ j Mk j , j = 1, 2, · · · , n (1.90)
are valid, k=1
Here Mik is the minor of the entry aik , defined as the
(A + B)T = AT + BT (1.86)
determinant of the (n − 1) × (n − 1) submatrix of A ob-
(kA) = kA ,
T T
scalar k (1.87) tained by deleting the ith row and the kth column of A.
(AB)T = BT AT . (1.88) The quantity (−1)i+k Mik is known as the cofactor of
aik and is denoted by Cik . Also note that (−1)i+k is re-
sponsible for whether a term is multiplied by +1 or −1.
Special Matrices Equations (1.89) and (1.90) suggest that the determinant
A square matrix A is symmetric if AT = A and skew- of a square matrix can be calculated using any row or
symmetric if AT = −A. A square matrix An×n = [aij ] any column of the matrix. However, for all practical pur-
is called upper-triangular if aij = 0 for all i > j, that poses, it is wise to use the row (or column) containing
is, every entry below the main diagonal is zero, lower- the most number of zeros, or if none, the one with the
triangular if aij = 0 for all i < j, that is, all elements smallest entries. A square matrix with a nonzero deter-
above the main diagonal are zeros, and diagonal if minant is known as a nonsingular matrix. Otherwise, it
aij = 0 for all i = j. The n × n identity matrix is a di- is called singular. The rank of any matrix A, denoted by
agonal matrix whose diagonal entries are all equal to 1, rank(A), is the size of the largest nonsingular submatrix
and is denoted by I. of A. If |An×n | = 0, we conclude that rank (A) < n.
Introduction to Mathematics for Mechanical Engineering 1.5 Linear Algebra 29
Part A 1.5
Find the determinant of diagonal matrix as a square matrix partitioned such
⎛ ⎞
1 2 −3 that its diagonal elements are square matrices, while
⎜ ⎟ all other elements are zeros; see Fig. 1.23a. Similarly,
A = ⎝ 4 −1 1 ⎠ .
a block-triangular matrix is a square matrix partitioned
2 0 1 so that its diagonal elements are square blocks, while all
entries either above or below this main block diagonal
are zeros; see Fig. 1.23b,c.
Solution. We will use the third row because it happens
Many properties of these special block matrices are
to contain a zero. Following (1.89),
basically extensions of those of diagonal and triangu-
|A| = 2 · (−1)3+1 M31 + 0 + 1 · (−1)3+3 M33 lar matrices. In particular, the determinant of each of
these matrices is equal to the product of the individ-
2 −3 1 2 ual determinants of the blocks along the main diagonal.
= 2 +
−1 1 4 −1 Consequently, a block diagonal (or triangular) matrix is
singular if and only if one of the blocks along the main
= 2(2 − 3) + (−1 − 8) = −11 .
diagonal is singular.
Properties of Determinant. The determinant of a ma- Inverse of a Matrix. Given a square matrix An×n , its
trix possesses a number of important properties, some inverse is denoted by A−1 with the property that
of which are listed below [1.1]:
AA−1 = I = A−1 A , (1.91)
• A square matrix A and its transpose
have the same where I denotes the n × n identity matrix. If A−1 exists,
determinant, that is, |A| = AT . then it is unique. A square matrix has an inverse if and
• The determinant of diagonal, upper-triangular and only if it is nonsingular. Equivalently, An×n has an in-
lower-triangular matrices is the product of the diag- verse if and only if rank (A) = n. A square matrix with
onal entries. an inverse is called invertible. An immediate applica-
• If an entire row (or column) of a square matrix A is tion of the inverse is in the solution process of a linear
zero, then |A| = 0. system Ax = b. Multiplying this equation from the left,
• If A is n × n and k is scalar, then |kA| = kn |A|. known as premultiplication, by A−1 , yields
• If any two rows (or columns) of A are interchanged,
the determinant of the resulting matrix is − |A|. A−1 (Ax) = A−1 b
• The determinant of the product of two matrices ⇒ (A−1 A)x = A−1 b
obeys |AB| = |A| |B|. ⇒ Ix = A−1 b
• Any square matrix with any number of linearly de-
pendent rows (or columns) is singular. ⇒ x = A−1 b .
a) b) c)
0 0
*
0 0 *
Fig. 1.23 (a) Block-diagonal matrix. (b) Block-upper-triangular matrix. (c) Block-lower-triangular matrix
30 Part A Fundamentals of Mechanical Engineering
Inverse via the Adjoint Matrix. The inverse of an invert- the original matrix. The inverse of an upper-triangular
Part A 1.5
ible matrix A = [aij ]n×n is determined using the adjoint matrix is upper-triangular. The diagonal elements of
of A, denoted by adj(A) and defined as [1.1] the inverse are the reciprocals of the diagonal entries
of the original matrix, while the off-diagonal entries
adj(A) do not obey any pattern. A similar result holds for
⎛ ⎞
(−1)1+1 M11 (−1)2+1 M21 · · · (−1)n+1 Mn1 lower-triangular matrices. Furthermore, it turns out that
⎜ ⎟a block-diagonal matrix and its inverse have exactly the
⎜ (−1)1+2 M12 (−1)2+2 M22 · · · (−1)n+2 Mn2 ⎟
=⎜⎜ .. .. .. ⎟
⎟
same structure.
⎝ . . . ⎠
(−1)1+n M1n (−1)2+n M2n · · · (−1)n+n Mnn Properties of Inverse. Some important properties of the
⎛ ⎞ inverse [1.1, 8] are given below. The assumption is that
C11 C21 · · · Cn1 all listed inverses exist.
⎜ ⎟
⎜ C12 C22 · · · Cn2 ⎟
(1.92) • (A ) = A.
−1 −1
=⎜⎜ .. .. .. ⎟
⎟.
⎝ . . . ⎠ • (AB)−1 = B−1 A−1 .
C1n C2n · · · Cnn • (AT )−1 = (A−1 )T .
• The inverse of a symmetric matrix is symmetric.
Note that each minor Mij (or cofactor Cij ) occupies the • (A p )−1 = (A−1 ) p , where p is a positive integer.
( j, i) position in the adjoint matrix, the opposite of what • det(A−1 ) = 1/ det(A).
one would normally expect. Then, the inverse of A is
simply defined by
1.5.2 Eigenvalues and Eigenvectors
−1 1
A = adj(A) . (1.93)
|A| The fundamentals of linear algebra are now extended
to treat systems of differential equations, which are
of particular importance to us since they represent the
Example 1.33: Formula for the inverse of a 2 × 2 matrix
mathematical models of dynamic systems. In the anal-
Find a formula for the inverse of ysis of such systems, one frequently encounters the
# $
a11 a12 eigenvalue problem, solutions of which are eigenvalues
A= . and eigenvectors. This knowledge enables the analyst to
a21 a22
determine the natural frequencies and responses of sys-
tems. Let A be an n × n matrix, v a nonzero n × 1 vector,
Solution. Following the procedure outlined above, we and λ a number (complex in general). Consider
find Av = λv (1.95)
M11 = a22 , C11 = a22 , A number λ for which (1.95) has a nontrivial solution
M12 = a21 , C12 = −a21 , (v = 0n×1 ) is called an eigenvalue or characteristic value
M21 = a12 , C21 = −a12 , of matrix A. The corresponding solution v = 0 of (1.95)
is the eigenvector or characteristic vector of A corre-
M22 = a11 , C22 = a11 . sponding to λ. Eigenvalues, together with eigenvectors
Then, form the eigensystem of A. The problem of determin-
# $ ing eigenvalues and the corresponding eigenvectors of
−1 1 a22 −a12 A, described by (1.95), is called an eigenvalue problem.
A = , (1.94) The trace of a square matrix A = [aij ]n×n , denoted by
|A| −a21 a11
tr(A), is defined as the sum of the eigenvalues of A. It
which is a useful formula for 2 × 2 matrices, allowing us turns out that tr(A) is also the sum of the diagonal el-
to omit the intermediate steps. ements of A. A matrix and its transpose have the same
eigenvalues.
Inverses of Special Matrices. If the main diagonal en-
tries are all nonzero, the inverse of a diagonal matrix Solving the Eigenvalue Problem
is again diagonal. The diagonal elements of the inverse Let us consider (1.95), Av = λv. Because equations in
are simply the reciprocals of the diagonal elements of this form involve scalars, vectors, and matrices, it is im-
Introduction to Mathematics for Mechanical Engineering 1.5 Linear Algebra 31
perative that extra caution is taken while working with we apply suitable elementary row operations [1.1] to the
Part A 1.5
them. First, rewrite and manipulate (1.95) as augmented
# matrix
$ to reduce it to
Av − λv = 0n×1 ⇒ (A − λI)v = 0 , 0 1 0
(1.96) .
0 0 0
where we note that every term here is an n × 1 vector. The second row suggests that there is a free variable,
The identity matrix I = In has been inserted so that the implying that the two equations contained in (1.99) are
two terms in parentheses are compatible; otherwise we linearly dependent. From the first row, we have v21 =
would have A − λ, which is meaningless. This equa- 0 so that v21 cannot be the free variable, so v11 must
tion has a nontrivial solution (v = 0) if and only if the be. In this example, since we already have v21 = 0, then
coefficient matrix, A − λI, is singular. That means v11 = 0 because otherwise v1 = 0, which is not valid.
|A − λI| = 0 . (1.97) # $ let v11 = 1, so
For simplicity,
1
This is called the characteristic equation of A. The de- v1 = .
0
terminant |A − λI| is an nth-degree polynomial in λ
Similarly, the eigenvector corresponding to λ2 = 2 can
and is known as the characteristic polynomial of A
be shown to be v2 = [−1 1]T . The set (v1 , v2 ) is the
whose roots are precisely the eigenvalues of A. Once
basis of all eigenvectors of matrix A.
the eigenvalues have been identified, each eigenvector
corresponding to each of the eigenvalues is determined Special Matrices
by solving (1.96). The eigenvalues of triangular and diagonal matrices
Example 1.34: Eigenvalues and eigenvectors are the diagonal entries. The eigenvalues of block-
Find the eigenvalues and eigenvectors of triangular and diagonal matrices are the eigenvalues of
# $ the block matrices along the main diagonal. All eigen-
−1 −3 values of a symmetric matrix are real, while those of
A= . a skew-symmetric matrix are either zero or pure imagi-
0 2
nary.
Solution. To find the eigenvalues of A, we solve the Generalized Eigenvectors
characteristic equation, If λk is an eigenvalue of A occurring m k times,
then m k is the algebraic multiplicity of λk , denoted
|A − λI| = 0
by AM(λk ). The maximum number of linearly in-
dependent eigenvectors associated with λk is called
−1 − λ −3
⇒ =0 the geometric multiplicity of λk , GM(λk ). In gen-
0 2−λ
eral, GM(λk ) ≤ AM(λk ). In Example 1.34 the AM and
⇒ (λ + 1)(λ − 2) = 0 GM of each of the two eigenvalues was 1. When
⇒ λ1,2 = −1, 2 . GM(λk ) <AM(λk ), there are fewer eigenvectors than
one would expect. For instance, if AM(λ) = 2 and
Without losing any information, let us assign λ1 = −1. GM(λ) = 1, then only one independent eigenvector can
To find the eigenvector, solve (1.96) with λ = λ1 = −1, be found for λ, while one is missing; the missing one is
λ =−1 called a generalized eigenvector [1.1].
(A − λ1 I)v1 = 0 ⇒ (A + I)v1 = 0 ,
1
(1.98)
Similarity Transformation – Diagonalization
where v1 is the
# 2 ×$1 eigenvector corresponding to λ1 . Two matrices An×n and Bn×n are said to be sim-
v11 ilar if there exists a nonsingular matrix Sn×n such
Letting v1 = and using A in (1.98), we find
v21 that B = S−1 AS. We say that B is obtained from
# $# $ # $ A through a similarity transformation. Similar matri-
0 −3 v11 0 ces have the same eigenvalues. Suppose An×n has n
= . (1.99)
linearly independent eigenvectors v1 , v2 , · · · ,vn asso-
0 3 v21 0
ciated with eigenvalues λ1 , λ2 , · · · , λn . Form the n × n
As expected, this system has nontrivial solutions be- matrix P = [ v1 v2 . . vn ], known as the modal ma-
cause the coefficient matrix is singular. To solve (1.99), trix, whose columns are the eigenvectors of A. Then, P
32 Part A Fundamentals of Mechanical Engineering
Part A 1
Displacement
Numerical solution of the state-variable equations – x
such as that in (1.101) of Example 1.37 – is then ob- Spring k
tained via the extension of RK4 discussed in Sect. 1.2.2.
Consider a system in the form f (t)
m
ẋ(t) = f (t, x(t)) , x(a) = x0 , a≤t≤b, Applied force
(1.102)
Damper c Mass
where
⎛ ⎞ Fig. 1.24 A mechanical system
x1 (t)
⎜ ⎟
⎜ x2 (t) ⎟ N subintervals. The fourth-order Runge–Kutta method
x(t) = ⎜
⎜ .. ⎟ ,
⎟
⎝ . ⎠ (RK4) for a system of first-order ODEs is as fol-
lows [1.5]. Knowing the initial vector x0 , the solution
xn (t) vector xi at each of the subsequent mesh points ti is
⎛ ⎞
f 1 (t, x1 , x2 , · · · , xn ) obtained via
⎜ ⎟
⎜ f 2 (t, x1 , x2 , · · · , xn ) ⎟ 1
⎜
f (t, x(t)) = ⎜ .. ⎟, xi+1 = xi + [q1 + 2q2 + 2q3 + q4 ],
⎟ 6
⎝ . ⎠
i = 0, 1, 2, · · · , N − 1 ,
f n (t, x1 , x2 , · · · , xn )
⎛ ⎞ where
α1
⎜ ⎟
⎜ α2 ⎟ q1 = h f (ti , xi ) ,
⎜
x0 = x(a) = ⎜ . ⎟ ⎟.
⎝ .. ⎠ 1
q2 = h f ti + h, xi + q1 ,
1
2 2
αn
1 1
Define an integer N > 0 and let h = (b − a)/N be q3 = h f ti + h, xi + q2 ,
the step size. The mesh points ti = a + ih, i = 2 2
0, 1, · · · , N − 1, then partition the interval [a, b] into q4 = h f (ti + h, xi + q3 ) .
References
1.1 R.S. Esfandiari: Applied Mathematics for En- 1.5 J.H. Mathews: Numerical Methods for Computer Sci-
gineers, 4th edn. (Atlantis, Irvine, California ence, Engineering, and Mathematics (Prentice-Hall,
2008) New York 1987)
1.2 J.W. Brown, R.V. Churchill: Complex Variables and 1.6 R.L. Burden, J.D. Faires: Numerical Analysis, 3rd edn.
Applications, 7th edn. (McGraw-Hill, New York (Prindle, Boston 1985)
2003) 1.7 J.W. Brown, R.V. Churchill: Fourier Series and
1.3 H.V. Vu, R.S. Esfandiari: Dynamic Systems: Modeling Boundary Value Problems. 6th edn. (McGraw-Hill,
and Analysis (McGraw-Hill, New York 1997) New York 2001)
1.4 C.H. Edwards, D.E. Penney: Elementary Differential 1.8 G.H. Golub, C.F. Van Loan: Matrix Computations, 3rd
Equations with Boundary Value Problems, 4th edn. edn. (The Johns Hopkins University Press, London
(Prentice-Hall, New York 2000) 1996)