Higher Mathematics for Students of

Chemistry and Physics

Joseph William Mellor 1869-1938
Chapter VII. How to Solve Dierential Equations
Preface to 2009 electronic edition
THE PRIMARY content of this e-book is chapter VII of the 1902 edition
of Higher Mathematics for Students of Chemistry and Physics (with spe-
cial reference to practical work) by Joseph William Mellor, D.Sc of New
Zealand Univerity (published by Longmans, Green, and Company of Lon-
don). Chapter VII of this text is an 80 page missive entitled, How to
Solve Dierential Equations. The text is primarily concerned with teach-
ing practical methods for arriving at solutions, and contains almost nothing
on the theory of dierential equations that is, there are very few theo-
rems and proofs. It covers a wide variety of solvable ordinary dierential
equations with over a hundred examples with solutions. The later sections
also provide an introductory overview of partial dierential equations.
Included also is a supplement I have prepared with many of the exam-
ples in the main text worked out step-by-step. Some of them I have worked
in more than one way. Students using this material are encouraged to at-
tempt each example in the main text on their own before referring to the
supplement for my working of that example. In addition the supplement
contains my comments on the main text, as well as descriptions of methods
of solution not covered in the main text. No student should expect to learn
the material here well without working the examples for himself or herself.
Prerequisite skills: Students using this material should be competent
in nding derivatives, including applying the product, quotient, and chain
rules, in methods for nding integrals, and in the manipulation of poly-
nomials, roots, logs, exponentials, and trigonometric functions. Students
should also have had at least some exposure to the dierential calculus of
functions of more than one variable and partial derivatives.
The text makes scattered use of complex numbers, as does the supple-
ment. So students should be familiar with the algebra of complex numbers.
They should also be familiar with Eulers formula relating exponentials of
a complex variable to the sine and cosine functions:
e
x
= cos x + sin x hence cos x =
e
x
+e
x
2
and sin x =
e
x
e
x
2
.
i
Note the usage of the symbol, , for

1. This notation is used through-

out the text. Most modern texts use the symbol, i, in lieu of for this pur-
pose. I have continued Mellors notation, using in the supplement in order
to remain consistent. Mellor uses the adjective, imaginary, to describe
any non-real complex number. This is at odds with modern nomenclature,
in which imaginary refers only to real multiples of . In the supplement, I
use the phrase, purely imaginary, to indicate such multiples and to make
the distinction from Mellors looser usage.
More on notation: Throughout the text and the supplement, the no-
tation, log (x), is always to be understood as logarithm to the base, e, oth-
erwise known as the natural log. Most modern texts use the notation, ln (x),
for natural log. The website, http://je560.tripod.com/functions.html, in-
dicates that the ln (x) notation was not introduced until 1893, so at the
time of publication of this text, that notation was still new and not yet
widely adopted. The log (x) notation, according to the same source, had
been in use since 1748. At any rate, the text uses the older notation, and
I have stuck with it in the supplement.
The text consistently uses the following hierarchy for grouping of terms:
[()]. Again I have stuck with that style in the supplement.
The original text used an unfamiliar notation for limits. Since its nota-
tion is no longer in common use, I have converted it to the modern notation,
e.g.,
lim
h0
sin (h)
h
= 1.
Errors in the original text: In preparing the electronic version of
the text from the original printed material (and especially when I prepared
the supplement), I discovered many errors in the text. Some of these were
type-setting errors, but others were the result of carelessness on the part
of the author. Where mistakes were obvious, I have simply corrected them
without comment. Where there were mistakes of carelessness in the exam-
ples and their solutions, I have, in many cases, left the mistake unaltered,
but I have tagged it in the right-hand margin with a left-facing arrow.
Where you encounter the arrow, you will nd my reasons for believing that
the text is in error in the supplement, together with what I believe the
text should have been. Serious students will attempt to nd the error on
their own before referring to the supplement. In many other places I have
also inserted comments in the right-hand margin, many of which refer the
student to the supplement for further clarication.
The page numbering of the original printed text is preserved in this
electronic version. The supplement uses its own page numbering, starting
with S-1, which follows page 359 of the main text. Following that are
various excerpts from other chapters of the main text that are referred to
in chapter VII.
The supplement includes three nal sections that are not specically
referenced from example problems in the main text. These are Resonant
systems revisited starting on page S-61, which describes how resonant
ii
systems respond to sinusoidal input, Dierential equation solution to
Keplerian orbits starting on page S-66, which develops the shape of a
gravitational orbit in a two body problem, and Introduction to Fouriers
heat equation and its solution starting on page S-67, which develops
methods for dealing with a one-dimensional heat equation. These have been
added because they are important applications of the methods taught in
this tutorial and may be of interest to some students.
The text by J.W. Mellor passes into public domain on the
rst day of the year 2009. My comments and supplement are oered
without charge to all under Creative Commons 3.0 license. See
http://creativecommons.org/licenses/by/3.0/ for details. I am often receiv-
ing emails from students living in Africa, the Middle East, or South Central
Asia, who are eager to learn mathematics but for whom the cost of mod-
ern textbooks is burdensome. One of the reasons I have gone to the eort
to prepare an electronic edition of this material is for them. If you are a
math educator working in those parts of the world and teaching dierential
equations, please consider using this material as part of your curriculum so
as to allow your students aordable access to the knowledge it provides.
L
A
T
E
X sources for this material will be made available, free of charge,
to anybody who can demonstrate that he or she will put them to good
use. Translation of the material into languages other than English will
always be considered to be a good use. Email me for details.
The probability that I have transcribed all of this material without error
is vanishingly close to zero. If you discover anything in this document that
appears to you to be in error, please notify me by email so that I may
correct it.
Karl Hahn, December 2008
iii
Mellors Preface to the 1902 edition
IT is almost impossible to follow the later developments of physical or
general chemistry without a working knowledge of higher mathematics. I
have found that the regular textbooks of mathematics rather perplex than
assist the chemical student who seeks a short road to this knowledge, for
it is not easy to discover the relation which the pure abstractions of formal
mathematics bear to the problems which every day confront the student of
Natures laws, and realize the complementary character of mathematical
and physical processes.
During the last ve years I have taken note of the chief diculties
met with in the application of the mathematicians x and y to physical
chemistry, and, as these notes have grown, I have sought to make clear
how experimental results lend themselves to mathematical treatment. I
have found by trial that it is possible to interest chemical students and to
give them a working knowledge of mathematics by manipulating the results
of physical or chemical observations.
I should have hesitated to proceed beyond this experimental stage if I
had not found at The Owens College a set of students eagerly pursuing
work in dierent branches of physical chemistry, and most of them looking
for help in the discussion of their results. When I told my plan to the
Professor of Chemistry he encouraged me to write this book. It has been
my aim to carry out his suggestion, so I quote his letter as giving the spirit
of the book, which I only wish I could have carried out to the letter.
THE OWENS COLLEGE,
MANCHESTER.
MY DEAR MELLOR,
If you will convert your ideas into words and write a book explaining the inwardness
of mathematical operations as applied to chemical results, I believe you will confer a
benet on many students of chemistry. We chemists, as a tribe, ght shy of any symbols
but our own. I know very well you have the power of winning new results in chemistry
and discussing them mathematically. Can you lead us up the high hill by gentle slopes?
Talk to us chemically to beguile the way? Dose us, if need be, with learning put lightly,
like powder in jam ? If you feel you have it in you to lead the way we will try to follow,
and perhaps some of the youngest of us may succeed. Wouldnt this be a triumph worth
working for? Try.
Yours very truly,
H.B. DIXON
THE OWENS COLLEGE,
MANCHESTER, May, 1901
iv
CHAPTER VII.
HOW TO SOLVE DIFFERENTIAL EQUATIONS.
Table of Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
117. Solution by Separation of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
118. What is a Dierential Equation? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
119. Exact Dierential Equations of First Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
120. How to nd Integrating Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
121. The First Law of Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
122. Linear Dierential Equations of First Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
123. Equations of First Order and Higher Degree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
124. Clairaults Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
125. Singular Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
126. Trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
127. Symbols of Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
128. The Linear Equation of nth Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
129. Linear Equations with Constant Coecients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
130. How to nd Particular Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
131. The Linear Equation with Variable Coecients . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
132. The Exact Linear Dierential Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
133. The Integration of Equations with Missing Terms . . . . . . . . . . . . . . . . . . . . . . . . . 319
134. Equations of Motion, Chiey Oscillatory Motion . . . . . . . . . . . . . . . . . . . . . . . . . . 322
135. The Velocity of Simultaneous and Dependent Chemical Reactions . . . . . . . . . 330
136. Simultaneous Dierential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
137. Partial Dierential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
138. What is the Solution of a Partial Dierential Equation? . . . . . . . . . . . . . . . . . . 341
139. The Solution of Partial Di. Equations of First Order . . . . . . . . . . . . . . . . . . . . 344
140. Partial Dierential Equations of nth Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
141. Linear Partial Di. Equations with Constant Coecients . . . . . . . . . . . . . . . . . 347
142. The Particular Integral of Linear Partial Di. Equations . . . . . . . . . . . . . . . . . . 351
143. The Linear Partial Equation with Variable Coecients . . . . . . . . . . . . . . . . . . . 354
144. The Integration of Dierential Equations in Series . . . . . . . . . . . . . . . . . . . . . . . . 355
145. Harmonic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
Supplemental material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . begins after page 359
The Principle of Superposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-24
The Method of Laplace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-31
Resonant Systems Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-61
Solution to Keplerian Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-66
Introduction to Fouriers Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S-67
Selected sections referenced in chapter VII . . . . . . . . . . . . . . begins after page S-69
20. Leibniz Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
22. Eulers Theorem of Homogeneous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
23. Successive Partial Dierentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
24. Exact Dierentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
25. Integrating Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
67. Envelopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
72. How to Find a Value for the Integration Constant . . . . . . . . . . . . . . . . . . . . . . . . . . 162
v
Public domain rationale: J.W. Mellor was a British subject and his works originally published by
British publishers. Hence under British copyright laws principle of life plus 70 years, the works of
J.W. Mellor passed into the public domain on 1-January-2009.
CHAPTER VII.
HOW TO SOLVE DIFFERENTIAL EQUATIONS.
THIS chapter may be looked upon as a sequel to that on the integral cal-
culus, but of a more advanced character. The methods of integration
already described will be found ample for most physico-chemical processes,
but chemists are proving every day that more powerful methods will soon
have to be brought in. As an illustration, I may refer to the set of dif-
ferential equations which Geitel encountered in his study of the velocity of
hydrolysis of the triglycerides by acetic acid (Journal f ur praktische Chemie
[2], 55, 429, 1897).
I have previously pointed out that in the eort to nd the relations
between phenomena, the attempt is made to prove that if a limited number
of hypotheses are prevised, the observed facts are a necessary consequence
of these assumptions. The modus operandi is as follows:
1. To anticipate Nature by means of a working hypothesis, which
is possibly nothing more than a convenient, ction.
From the practical point of view, says Professor Bucker (Presidential Address to
the B. A. meeting at Glasgow, September, 1901), it is a matter of secondary importance
whether our theories and assumptions are correct, if only they guide us to results in
accord with facts. . . . By their aid we can foresee the results of combinations of causes
which would otherwise elude us.
2. Thence to deduce an equation representing the momentary rate of
change of the two variables under investigation.
3. Then to integrate the equation so obtained in order to reproduce the
working hypothesis in a mathematical form suitable for experimental
verication (see '' 18, 69, 88, 89, and elsewhere).
So far as we are concerned this is the ultimate object of our integration.
By the process of integration we are said to solve the equation.
282
For the sake of convenience, any equation containing dierentials or dier-
ential coecients will, after this, be called a, dierential equation.
117. The Solution of a Dierential Equation by the Separa-
tion of the Variables. The dierent equations hitherto considered have
required but little preliminary arrangement before integration. For exam-
ple, when preparing the equations representing the velocity of a chemical
reaction of the general type:
dx
dt
= kf (t) , (1)
we have invariably collected all the xs to one side, the ts, to the other,
before proceeding to the integration.
This separation of the variables is nearly always attempted before re-
sorting to other artices for the solution of the dierential equation, because
the integration is then comparatively simple. The following examples will
serve to emphasise these remarks:
EXAMPLES. (1) Integrate the equation, y dx + xdy = 0. Rearrange the terms so
that
dx
x
+
dy
y
= 0; or,

dx
x
+

dy
y
= C,
by multiplying through with 1/xy. Ansr. log x + log y = C. Two or more apparently
dierent answers may be the same. Thus, the solution of the preceding equation may
also be written,
log xy = log e
C
, i.e., xy = e
C
; or log xy = log C

, i.e., xy = C

.
C and log C

are, of course, the arbitrary constants of integration.

(2) The equation for the rectilinear motion of a particle under the inuence of an
attractive force from a xed point is
v
dv
dx
+

x
2
= 0.
Solve. Ansr.
1
2
v
2
=

x
+C.
(3) Solve

1 +x
2

dy =

y dx. Ansr. 2

y tan
1
x = C.
(4) Solve y x
dy
dx
= a

y +
dy
dx

. Ansr. y = C (a +x)
1a
.
(5) In consequence of imperfect insulation, the charge on an electried body is
dissipated at a rate proportional to the magnitude R of the charge. Hence show that if
a is a constant depending on the nature of the body, and R, represents the magnitude
of the charge when t (time) = 0,
E = E
0
e
at
.
Hint. Compound interest law. Integrate by the separation of the variables. Interpret
your result.
(6) Abeggs formula for the relation between the dielectric constant (D) of a uid
and temperature , is

dD
d
=
D
190
.
283
Hence show that D = Ce
/190
, where C is a constant whose value is to be determined
from the conditions of the experiment. Put the answer into words.
(7) What curves have a slope x/y to the x-axis? Ansr. The rectangular hyperbolas
xy = C. Hint. Set up the proper dierential equation and solve.
(8) The relation between small changes of pressure and volume of a gas under
adiabatic conditions, is p dx +v dp = 0. Hence show that pv

= constant.
(9) A lecturer discussing the physical properties of substances at very low tem-
peratures, remarked it appears that the specic heat of a substance decreases with
decreasing temperatures at a rate proportional to the specic heat of the substance it-
self . Set up the dierential equation to represent this law and put your result in a
form suitable for experimental verication.
(10) Helmholtz equation for the strength of an electric current (C) at the time t, is
C =
E
R

L
R
dC
dt
,
where E represents the electromotive force in a circuit of resistance R and self-induction
L. If E, R, L, are constants, show that RC = E(1 e
Rt/L
) provided C = 0, when
t = 0.
A substitution will often enable an equation to be treated by this simple
method of solution.
EXAMPLE, Solve (x y
2
)dx + 2xy dy = 0. Ansr. xe
y
2
/x
= C. Hint, put y
2
= v,
divide by x
2
dx/x +d(v/x) = 0, etc.
If the equation is homogeneous in x and y, that is to say, if the
sum of the exponents of the variables in each term is of the same degree,
a preliminary substitution of x = ty, or y = tx, according to convenience,
will always enable variables to be separated. The rule for the substitution
is to treat the dierential coecient which involves the smallest number of
terms.
EXAMPLES, (1) x +y
dy
dx
2y = 0. Substitute y = tx,

t dt
(1 t)
2
+

dx
x
= C;
1
1 t
+ log x = C.
Ansr. (y x)e
x/(xy)
= C.
(2) If (y x)dy +ydx = 0, y = Ce
x/y
.
(3) If x
2
dy y
2
dx xy dx = 0, x = e
x/y
+C.
(4) (x
2
+y
2
)dx = 2xy dy, x
2
y
2
= Cx.
Non-homogeneous equations in x and y can be converted into the
homogeneous form by a suitable substitution.
The most general type of a non-homogeneous equation of the rst degree
is,
(ax +by +c) dx + (a

x +b

y +c

) dy = 0. (2)
284
To convert this into an homogeneous equation, assume that
x = v +h and y = w +k,
and substitute in the given equation (2). Thus, we obtain
av +bw + (ah +bk +c) dv +a

v +b

w + (a

h +b

k +c

) dw = 0. (3)
Find h and k so that
ah +bk +c = 0; a

h +b

k +c

= 0.
h =
b

c bc

b ab

; k =
ac

c
a

b ab

.
(4)
Substitute these values of h and k in (3). The resulting equation
(av +bw) dv + (a

v +b

w) dw = 0 (5)
is homogeneous and, therefore, may be solved as just indicated.
EXAMPLES, (1) Solve (3y 7x 7) dx + (7y 3x 3) dy = 0. Ansr.
(y x 1)
2
(y +x + 1)
5
= C. Hints. From (2), a = 7, b = 3, c = 7, a

= 3,
b

= 7, c

= 3. From (4), h = 1, k = 0. Hence from (3),

3wdv 7v dv + 7wdw 3v dv = 0.
To solve this homogeneous equation, substitute w = vt, as above, and separate the
variables.
7
dv
v
=
3 7t
t
2
1
dt; 7

dv
v
+

2dt
t 1
+

5dt
t + 1
= C.
7 log v + 2 log (t 1) + 5 log (t + 1) = C; or v
7
(t 1)
2
(t + 5)
5
= C.
But x = v +h v = x + 1; y = w +k y = w; t = w/v = y/ (x + 1), etc.
(2) If (2y x 1) dy + (2x y + 1) dx = 0; x
2
xy +y
2
+x y = C.
If in (3),
a : b = a

: b

= 1 : m (say),
h and k are indeterminate, since (2) then becomes,
(ax +by +c) dx +m(ax +by) +c

dy = 0.
The denominators in equations (4) also vanish. In this case put
z = ax +by
and eliminate y, thus, we obtain,
a +b
z +c
mz +c
+
dz
dx
= 0, (6)
an equation which allows the variables to be separated.
EXAMPLES, (1) Solve (2x + 3y 5) dy + (2x + 3y 1) dx = 0.
Ansr. x +y 4 log (2x + 3y + 7) = C.
(2) Solve (3y + 2x + 4) dx (4x + 6y 1) dy = 0.
Ansr. 9 log (21y + 14x + 22) /7 21 (2y x) = C.
When the variables cannot be separated in a satisfactory manner, spe-
cial artices must be adopted. We shall nd it the simplest plan to adopt
the routine method of referring each artice to the particular class of equa-
tion which it is best calculated to solve.
285
These special devices are sometimes far neater and quicker processes of
solution than the method just described.
We shall follow the conventional x and y rather more closely than in
the earlier part of this work. The reader will know, by this time, that his
x and ys, his p and vs and his s and ts are not to be kept in water-tight
compartments. It is perhaps necessary to make a few general remarks on
the nomenclature.
118. What is a Dierential Equation? We have seen that the
straight line,
y = mx +b, (1)
fulls two special conditions:
1. It cuts one of the coordinate axes at a distance b from the origin.
2. It makes an angle tana = m, with the x-axis.
By dierentiation.
dy
dx
= m. (2)
This equation has nothing at all to say about the constant b. That condition
has been eliminated. Equation (2), therefore, represents a straight line
fullling one condition, namely, that it makes an angle tan
1
m with the
x-axis. Now substitute (2) in (1), the resulting equation,
y =
dy
dx
x +b, (3)
in virtue of the constant b, satises only one denite condition, (3), there-
fore, is the equation of any straight line passing through b. Nothing is said
about the magnitude of the angle tan
1
m.
Dierentiate (2). The resulting equation,
d
2
y
dx
2
= 0, (4)
represents any straight line whatever. The special conditions imposed by
the constants m and b in (1), have been entirely eliminated. Equation (4) is
the most general equation of a straight line possible, for it may be applied
to any straight line that can be drawn in a plane.
Let us now nd a physical meaning for the dierential equation.
286
In ' 7, we have found that the third dierential coecient, d
3
s/dt
3
represents the rate of change of acceleration from moment to moment.
Suppose that the acceleration d
2
s/dt
2
, of a moving body does not change
or vary in any way. It is apparent that the rate of change of a constant
or uniform acceleration must be zero. In mathematical language, this is
written,
d
3
s
dt
3
= 0. (5)
Now integrate this equation once. We obtain
d
2
s
dt
2
= constant, say, g. (6)
Equation (6) tells us not only that the acceleration is constant, but it xes
that value to the denite magnitude g ft. per second.
But acceleration measures the rate of change of velocity. Integrate (6),
we get,
ds
dt
= gt +C
1
. (7)
From ' 72, we have learnt how to nd the meaning of C
1
. Put t = 0, then 72 is
included
in this
docu-
ment
following
the sup-
plement.
dx/dt = C
1
. This means that when we begin to reckon the velocity, the
body may have been moving with a denite velocity C
1
. Let C
1
= v
0
ft.
per second. Of course, if the body started from a position of rest, C
1
= 0.
Now integrate (7) and nd the value of C
2
in the result,
s =
1
2
gt
2
+v
0
t +C
2
, (8)
by putting t = 0. It is thus apparent that C
2
represents the space which
the body had traversed when we began to study its motion. Let C
2
= s
0
ft. The resulting equation
s =
1
2
gt
2
+v
0
t +s
0
, (9)
tells us three dierent things about the moving body at the instant we
began to take its motion into consideration.
l. It had traversed a distance of s
0
ft. To use a sporting phrase, if the
body is starting from scratch, s
0
= 0.
2. The body was moving with a velocity of v
0
ft. per second.
3. The velocity was increasing at the uniform rate of g ft. per second.
Equation (7) tells us the two latter facts about the moving body; equa-
tion (6) only tells us the third fact; equation (5) tells us nothing more than
that the acceleration is constant. (5), therefore, is true of the motion of
any body moving with a uniform acceleration.
EXAMPLE. If a body falls in the air, experiment shows that the retarding eect of
the resisting air is proportional to the square of the velocity of the moving body. Instead
of g, therefore, we must write g v
2
, where is the
287
variation constant of page 487. For the sake of simplicity, put = g/a
2
and show that
v = a
e
gt/a
e
gt/a
e
gt/a
+e
gt/a
; s =
a
2
g
log
e
gt/a
+e
gt/a
2
=
a
2
g
log cosh
gt
a
,
since v = 0 when t = 0, and s = 0 when t = 0.
Similar reasoning holds good from whatever sources we may draw our
illustrations. We are, therefore, able to say that a dierential equation,
freed from constants, is the most general way of expressing a, natural law.
Any equation can be freed from its constants by combining it with the
various equations obtained by dierentiation of the given equation as many
times as there are constants. The operation is called elimination.
EXAMPLES.(1) Eliminate the arbitrary constants a and b, from
y = ax +bx
2
.
Dierentiate twice and combine the results with the original equation. result,
x
2
d
2
y
dx
2
2x
dy
dx
+ 2y = 0,
is quite free from the arbitrary restrictions imposed in virtue of the presence of the
constants a and b in the original equation.
(2) Eliminate m from y
2
= 4mx. Ansr. y
2
= 2xdy/dx.
(3) Eliminate and from y = cos x + sin x. Ansr. d
2
y/dx
2
+y = 0.
(4) Eliminate and from y = e
ax
+e
bx
.
Ansr. d
2
y/dx
2
(a +b)dy/dz +aby = 0.
(5) Eliminate k from dx/dt = k(a x) of ' 69. What does the resulting equation
mean?
We always assume that every dierential equation has been obtained
by the elimination of constants from a given equation called the primitive.
In practical work we are not so much concerned with the building up of
a dierential equation by the elimination of constants from the primitive,
as with the reverse operation of nding the primitive from which the dif-
ferential equation has been derived. In other words, we have to nd some
relation between the variables which will satisfy the dierential equation.
Given an expression involving x, y, dy/dx, d
2
y/dx
2
,. . . , to nd an equa-
tion containing only x, y and constants which can be reconverted into the
original equation by the elimination of the constants.
This relation between the variables and constants which satises the
given dierential equation is called a general solution, or a complete
solution, or a complete integral of the dierential
288
equation. A solution obtained by giving particular values to the arbitrary
constants of the complete solution is a particular solution.
Thus y = mx is a complete solution of y = xdy/dx; p = xtan45

, is a
particular solution.
A dierential equation is ordinary or partial, according as there is
one or more than one independent variables present. Ordinary dierentia1
equations will be treated rst.
Equations like (2) and (8) above, are said to be of the rst order. For a
similar. reason (4) and (6) are of the second order, (5) of the third order.
The order of a dierential equation, therefore, is xed by that of the
highest dierential coecient it contains. The degree of a dierential
equation is the highest power of the highest order of dierential coeicient
it contains. Thus,
d
2
y
dx
2
+k

dy
dx

3
+x
4
= 0,
is of the second order and third degree.
It is not dicult to show that the complete integral of a dierential
equation, of the nth order, contains n and only n arbitrary constants.
We shall rst consider equations of the rst order.
119. Exact Dierential Equations of the First Order. The rea-
son many dierential equations are so dicult to solve is due to the fact
that they have been formed by the elimination of constants as well as by
the elision of some common factor from the primitive. Such an equation,
therefore, does not actually represent the complete or total dierential of
the original equation or primitive. The equation is then said to be inexact.
On the other hand, an exact dierential equation is one that has been
obtained by the dierentiation of a function of x and y and performing no
other operation involving x and y.
Easy tests were described in '' 24, 25, to determine whether any given
dierential equation is exact or inexact. It was pointed out that the dier-
ential equation,
M dx +N dy = 0, (1)
is the direct result of the dierentiation of any function u, provided,
M
y
=
N
x
. (2)
289
This last result was called the criterion of integrability, because, if an equa-
tion satises the test, the integration can be readily performed by a direct
process. This is not meant to imply that only such equations can be inte-
grated as satisfy the test, for many equations which do not satisfy the test
can be solved in other ways.
EXAMPLES.(1) Apply the test to the equations,
y dx +xdy = 0 and y dx xdy = 0.
In the former, M = y, N = x;
M/y = 1, N/x = 1; M/y = N/x.
The test is, therefore, satised and the equation is exact. In the other equation, M = y,
N = x,
M/y = 1, N/x = 1.
This does not satisfy the test. In consequence, the equation cannot be solved by the
method for exact dierential equations.
(2) Is the equation, (x + 2y) xdx +

x
2
y
2

dy = 0 exact? M = x(x + 2y),

N =

x
2
y
2

, M/dy = 2x, N/x = 2x. The condition is satised, the equation

is exact.
(3) Show that

a
2
y +x
2

dx +

b
3
+a
2
x

dy = 0, is exact.
(4) Show that (sin y +y cos x) dx + (sin x +xcos y) dy = 0, is exact.
To integrate an equation which satises the criterion of integrability, we
must remember that M is the dierential coecient of u with respect to x,
y being constant, and N is the dierential coecient of N with respect to
y, x being constant. Hence we may integrate M dx on the supposition that
y is constant and then treat N dy as if x were a constant. The complete
solution of the whole equation is obtained by equating the sum of these
two integrals to an undetermined constant. The complete integral is
u = C. (3)
EXAMPLES.(1) Integrate x(x + 2y) dx +

x
2
y
2

dy = 0, from the preceding set

of examples. Since the equation is exact,
M = x(x + 2y) ; N = x
2
y
2
;

M dx =

x(x + 2y) dx =
1
3
x
3
+x
2
y = Y,
where Y is the integration constant which may, or may not, contain y, because y has
here been regarded as a constant.
Now the result of dierentiating
1
3
x
3
+x
2
y = Y,
should be the original equation. On trial,
x
2
dx + 2xy dx +x
2
dy = dY.
On comparison with the original equation, it is apparent that
dY = y
2
dy; Y =
1
3
y
3
+C.
Substitute this in the preceding result. The complete solution is, therefore,
1
3
x
3
+x
2
y
1
3
y
3
= C.
290
To summarise: The method detailed in the example just given may be put into a
more practical shape.
To integrate an exact dierential equation, rst nd

M dx on the assumption that

y is constant and substitute the result in

M dx +

N

y

M dx

dy = C. (4)
With the old example, therefore, having found

M dx, we may write down at once

1
3
x
3
+x
2
y +

x
2
y
2

1
3
x
3
+x
2
y

dy = C.

1
3
x
3
+x
2
y +

x
2
y
2
x
2

dy = C.
And the old result follows directly. If we had wished we could have used

N dy +

M

x

N dy

dx = C,
in place of (4).
In practice it is often convenient to modify this procedure. If the equation satises
the criterion of integrability, we can easily pick out terms which make M dx+N dy = 0,
and get
M dx +Y and N dy +X,
where Y cannot contain x and X cannot contain y. Hence if we nd M dy and N dx,
the functions X and Y will be determined.
In the above equation, the only terms containing x and y are 2xy dx +x
2
dy, which
obviously have been derived from x
2
y. Hence integration of these and the omitted terms
gives the above result.
(2) Solve

x
2
4xy y
2

dx +

y
2
4xy 2x
2

dy = 0. Pick out terms in x and y,

we get

4xy + 2y
2

dx

4xy + 2x
2

dy = 0.
Integrate. 2x
2
y 2xy
2
= constant.
Pick out the omitted terms and integrate for the complete solution. We get,

x
2
dx +

y
2
dy 2x
2
y 2xy
2
= x
3
6x
2
y 6xy
2
+y
3
= constant.
(3) Show that the solution of

a
2
x +x
2

dx + b
3
+z
2
x)dy = 0 is
a
2
xy +b
3
y +
1
3
x
3
= C. Use (4).
(4) Solve

x
2
y
2

dx 2xy dy = 0. Ansr.
1
3
x
2
y
2
= C/x. Use (4).
Equations made exact by means of integrating factors. As just pointed
out, the reason any dierential equation does not satisfy the criterion of
exactness, is because the integrating factor has been cancelled out during
the genesis of the equation from its primitive.
If, therefore, the equation
M dx +N dy = 0,
does not satisfy the criterion of integrability, it will do so when the factor,
previously divided out, is restored. Thus, the preceding equation is made
exact by multiplying through with the integrating factor . Hence,
(M dx +N dy) = 0,
satises the criterion of exactness, and the solution can be obtained as
described above.
291
120. How to nd Integrating Factors. Sometimes integrating fac-
tors are so simple that they can be detected by simple inspection.
EXAMPLES.(1) y dx xdy = 0 is inexact. It becomes exact by multiplication with
either x
2
, x
1
y
1
, or y
2
.
(2) In (y x) dy+y dx = 0, the term containing y dxxdy is not exact, but becomes
so when multiplied as in the preceding example.

dy
y

xdy y dx
y
2
; or log y
y
x
= C.
For the general theorems concerning the properties of integrating factors, the reader
must consult some special treatise, say Booles A Treatise on Dierential Equations,
pages 55 et seq., 1865.
25 is
included
in this
docu-
ment
following
the sup-
plement.
We have already established, in ' 25, that an integrating factor always
exists which will make the equation
M dx +N dy = 0,
an exact dierential.
Moreover, there is also an innite number of such factors, for if the
equation is made exact when multiplied by , it will amain exact when
multiplied by any function of .
The dierent integrating factors correspond to the various forms in
which the solution of the equation may present itself. For instance, the
integrating factor x
1
y
1
, of y dx+xdy = 0, corresponds with the solution
log x + log y = C. The factor y
2
corresponds with the solution xy = C

.
Unfortunately, it is of no assistance to know that every dierential equa-
tion has an innite number of integrating factors. No general practical
method is known for nding them. Here are a few elementary rules appli-
cable to special cases.
Rule I. Since
d (x
m
y
n
) = x
m1
y
n1
(my dx +nxdy) ,
an expression of the type my dx+nxdy, has an integrating factor x
m1
y
n1
;
or, the expression
x

(my dx +nxdy) = 0, (1)

has an integrating factor
x
m1
y
n1
,
or more generally still,
x
km1
y
kn1
, (2)
where k may have any value whatever.
EXAMPLE. Find an integrating factor of y dx + xdy = 0. Here, = 0, = 0,
m = 1, n = 1 y
2
is an integrating factor of the given equation.
292
If the expression can be written
x

(my dx +nxdy) +x

(m

y dx +n

xdy) , (3)
the integrating factor can be readily obtained, for
x
km1
y
kn1
; and x
k

y
k

,
are integrating factors of the rst and second members respectively. In
order that these factors may he identical,
km1 = k

; kn 1 = k

.
Values of k and k

can be obtained to satisfy these two conditions by solving

these two equations. Thus,
k =
n

) m

)
mn

n
; k

=
n(

) n(

)
mn

n
. (4)
EXAMPLES.(1) Solve y
3
(y dx 2xdy) + x
4
(2y dx +xdy) = 0. Hints. Show that
= 0, = 3, m = 1, n = 2,

= 4,

= 0, m

= 2, n

= 1; x
k1
y
2k4
is
an integrating factor of the rst, x
2k

5
y
k

1
of the second member. Hence, from (4),
k = 2, k

= 1, x
3
is an integrating factor of the whole expression. Multiply
through and integrate for 2x
4
y y
4
= Cx
2
.
(2) Solve

y
3
2yx
2

dx +

2xy
2
x
3

dy. Ansr. x
2
y
2

y
2
x
2

= C. Integrating
factor deduced after rearranging the equation is xy.
Rule II. If the equation is homogeneous and of the form: M dx+N dy =
0, then (Mx +Ny)
1
an integrating factor.
Let the expression
M dx +N dy = 0,
be of the mth degree and . an integrating factor of the nth degree,
M dx +N dy = du, (5)
is of the (m+n)th degree, and the integral u is of the (m+n + 1)th degree. 22 is
included
in this
docu-
ment
following
the sup-
plement.
By Eulers theorem, ' 22,
Mx +Ny = (m+n + 1) u. (6)
Divide (5) by (6),
M dx +N dy
Mx +Ny
=
1
m+n + 1
du
u
The right side of this equation is a complete dierential, consequently, the
left side is also a complete dierential. Therefore, (Mx +Ny)
1
has made
M dx +N dy = 0 an exact dierential equation.
EXAMPLES.(1) Show that

x
3
y xy
3

1
is an integrating factor of

x
2
y +y
3

dx
2xy
2
dy = 0.
(2) Show that 1/

x
2
nyx +y
2

is an integrating factor of y dy +(x ny) dx = 0.

The method, of course, cannot be used if Mx + Ny is equal to zero. In this case,
we may write y = Cx, a solution.
293
Rule III. If the equation is of the form,
f
1
(x, y) y dx +f
2
(x, y) xdy = 0,
then (Mx Ny)
1
is an integrating factor.
EXAMPLE. Solve (1 +xy) y dx+(1 xy) xdy = 0. Hint. Show that the integrating
factor is 1/2xy. Divide out
1
2
.

M dx = 1/xy + log x. Ansr. x = Ce

1/xy

If Mx Ny = 0, the method fails and xy = C is then a solution of the equation.

E.g., (1 +xy) y dx + (1 +xy) xdy = 0.
Rule IV. If
1
N

M
y

N
x

is a function of x only, e

f(x)dx
is an inte-
grating factor. Or, if
1
M

N
x

M
y

= f (y), then e

f(y)dy
is an integrating
factor. These are important results.
EXAMPLES.(1) Solve

x
2
+y
2

dx 2xy dy = 0. Ansr. x
2
y
2
= Cx. Hint. Show
f (x) = 2/x. The integrating factor is, therefore,
e

2dx/x
= e
log 1/x
2
= 1/x
2
.
(Why?) Prove that this is an integrating factor, and solve as in the preceding section.
(2) Solve

y
4
+ 2y

dx +

xy
3
+ 2y
4
4x

dy = 0. Ansr. xy
3
y
4
+ 2x = Cy2.
(3) We may prove the rule for a special case in the following manner. The steps will
serve to recall some of the principles established in some earlier chapters.
Let
dy
dx
+Py = Q, (7)
where P and Q are either constant or functions of x. Let be an integrating factor
which makes
dy + (Py Q) dx = 0, (8)
an exact dierential.
dy +(Py Q) dx N dy +M dx.

N
x
=

x
;
M
y
= (Py q)

y
+P.

x
= (Py Q)

dx
+P.

x
dx = (Py Q)

y
dx +Pdx;
=

y
dy +Pdx.

x
dx +

y
dy = d = Pdx
P =
1

d
dx
;

P dx = log .
and since log
e
e = 1.

P dx

log e = log ; = e

P dx
. (9)
This result will be employed in dealing with linear equations, ' 122.
294
121. The First Law of Thermodynamics. According to the dis-
cussion at the end of the rst chapter, one way of stating the rst law of
thermodynamics is as follows:
dQ = dU +dW,
which means that when a quantity of heat, dQ, is added to a substance,
one part of the heat is spent in changing the internal energy, dU, of the
substance and another part, dW, is spent in doing work against external
forces. In the special case, when that work is expansion against atmospheric
pressure, dW = p dv, as shown in ' 91. See (11), page 524.
We know that the condition of a substance is completely dened by
any two of the three variables p, v, , because when any two of these three
variables is known, the third can be deduced from the relation
pv = R.
Hence it is assumed that the internal energy of the substance is completely
dened when any two of these variables are known.
Now let the substance pass from any state A to another state S (Fig.
111). The internal energy of the substance in the state B
Figure 111:
completely determined by the coordinates
of that point, because U is quite inde-
pendent of the nature of the transforma-
tion from the state A to the state B. It
makes no dierence to the magnitude of U
whether that path has been via APE or
AQB. In this case U is said to be a single-
valued function by the coordinates of the
point corresponding to any given state. In
other words, dU is a complete dierential.
Hence
dU =
U
x
dx +
U
y
dy,
is an exact dierential equation, where x and y represent any pair of the
variables p, v, .
On the other hand, the external work done durmg the transformation
from the one state to another, depends not only on the initial and -
nal states of the substance, but also on the nature of the path described
in passing from the state A to the state B. For example, the substance
may perform the work represented by the area AQBB

or by the area
APBB

, in its passage from the

295
state A to the state B. In fact the total work done in the passage from
A to B and back again, is represented by the area APBQ (page 183). In
order to know the work done during the passage from the state A. to the
state B, it is not only necessary to know the initial and nal states of the
substance as dened by the coordinates of the points A and B, but we
must know the nature of the path from the one state to the other.
Similarly, the quantity of heat supplied to the body in passing from one
state to the other, not only depends on the initial and nal states of the
substance but also on the nature of the transformation.
All this is implied when it is said that dW and dQ are not perfect
dierentials. Although we can write

2
U
xy
=

2
U
yx
,
we must put, in the case of W or Q,

2
Q
xy

2
Q
yx
; or

2
Q
xy
=

2
Q
yx
.
Therefore the partial dierentiation of x with respect to y, furnishes a
complete dierential equation only when we multiply through with the
integrating factor , so that
dQ =
Q
x
dx +
Q
y
dy,
where x and y may represent any pair of the variables p, v, .
The integrating factor is proved in thermodynamics to be equivalent to
the so-called Carnots function (see Prestons Theory of Heat).
To indicate that dW and dQ are not perfect dierentials, some writers
superscribe a comma to the top right-hand corner of the dierential sign.
The above equation would then be written,
d

Q = dU +d

W.
122. Linear Dierential Equations of the First Order. A linear
dierential equation of the rst order involves only the rst power of the
dependent variable y and of its rst dierential coecients. The general
type is,
dy
dx
+Py = Q, (1)
where P and Q may be functions of x, or constants.
296
We have just proved that e

P dx
is an integrating factor of (1), therefore
e

P dx
(dy +Py dy) = e

P dx
Qdx,
is an exact dierential equation. The general solution is,
ye

P dx
=

P dx
Qdx +C. (2)
The linear equation is one of the most important in applied mathematics.
In particular cases the integrating factor may assume a very simple form.
In the following examples, remember that e
log x
= x, if

P dx = log x,
e

P dx
= x.
EXAMPLES.(1) Solve

1 +x
2

dy = (m+xy) dx. Reduce to the form (1) and we

obtain
dy
dx

x
1 +x
2
y =
m
1 +x
2
.

P dx =

xdx
1 +x
2
=
1
2
log

1 +x
2

= log

(1 +x
2
).
Remembering log 1 = 0, log e = 1, the integrating factor is evidently,
log e

P dx
= log 1 log

1 +x
2
, or e

P dx
=
1

1 +x
2
.
Multiply the original equation with this integrating factor, and solve the resulting exact
equation as ' 119, (4), or, better still, by (2) above. The solution: y = mx+C

(1 +x
2
),
follows at once.
(2) Ohms law for a variable current owing in a circuit with a coecient of self-
induction L (henries), a resistance R (ohms), and a current of C (amperes) and an
electromotive force E (volts), is given by the equation,
E = RC +L
dC
dt
.
This equation has the standard linear form (1). If E is constant, show that the solution
is,
C = E/R +Be
Rt/L
,
where B is the arbitrary constant of integration (page 159). Show that C approximates
to E/B after the current has been owing some time (t). Hint for solution. Integrating
factor is e
Rt/L
.
(3) The equation of motion of a particle subject to a resistance varying directly as
the velocity and as some force which is a given function of the time, is
dv/dt +kv = f (t) .
Show that v = Ce
kt
+ e
kt

e
kt
f (t) dt.
If the force is gravitational, say g,
v = Ce
kt
+g/k
(4) Solve xdy + y dx = x
3
dx. Integrating factor= x. Ansr. y =
1
4
x
3
+C/x.
Many equations may be transformed into the linear type of equation,
by a change in the variable. Thus, in the so-called Bernoullis equation,
dy/dx +Py = Qy
n
. (3)
297
Divide by y
n
, multiply by (1 n) and substitute y
n1
= v, in the result.
Thus,
1 n
y
n
dy
dx
+ (1 n) Py
1n
= (1 n) Q,
and dy/dx + (1 n) Pv = Q(1 n),
which is linear in v. Hence, the solution is
ve
(1n)

P dx
= (1 n)

Qe
(1n)

P dx
dx +C.
y
1n
e
(1n)

P dx
= (1 n)

Qe
(1n)

P dx
dx +C.
EXAMPLES.(1) Solve dy/dx+y/x = y
2
. Treat as above, substituting v = 1/y. The
integration factor is e

dx/x
= e
log x
= 1/x.
Ansr. Cxy xy log x = 1.
(2) Solve dy/dx + xsin 2y = x
3
cos
2
y. Divide by cos
2
y. Put tan y = v. The
integration factor is e

2x dx
, i.e., e
x
2
. Ansr. e
x
2
tan y
1
2
e
x
2

x
2
1

= C. Hint to
solve ve
x
2
=

x
3
e
x
2
dx + C. Put x
2
= z, 2xdx = dz, and this integral becomes
1
2

ze
z
dz, or
1
2
e
z
(z 1), etc.
(3) Here is an instructive dierential equation, which Harcourt and Esson encoun-
tered during their work on chemical dynamics in 1866.
1
y
2
dy
dx
+
K
y

K
x
= 0.
I shall give a method of solution in full, so as to revise some preceding work. The
equation has the same form as Bernoullis. Therefore, substitute
v =
1
y
; i.e.,
dv
dx
=
1
y
2
dy
dx
.

dv
dx
Kv +
K
x
= 0,
an equation linear in v. The integrating factor is
e

P dx
, or, e
Kx
; Q, in (2), = K/x,
therefore, from (2) ve
Kx
=

K
x
e
Kx
dx + C.
From ' 108,
ve
Kx
= K

1
x

1 (Kx) +
(Kx)
2
1 2

(Kx)
3
1 2 3
+. . .

dx +C.
ve
Kx
= K

dx
x
K dx +
K
2
xdx
1 2

K
3
x
2
dx
1 2 3
+. . .

+C.
But v = 1/y. Multiply through with ye
Kx
, and integrate.
1 = Ke
Kx

C
1
log x +Kx
(Kx)
2
1 2
+
(Kx)
3
1 2 3
. . .

y.
We shall require this result on page 333.
123. Dierential Equations of the First Order and of the First
or Higher Degree. Solution by Dierentiation.
Case i. The equation can be split up into factors. If the dierential equation can
be resolved into n factors of the rst degree, equate each factor to zero and solve each
of the n equations separately. The n solutions may be left either distinct, or combined
into one.
298
EXAMPLES.(1) Solve x

dy
dx

2
= y. Resolve into factors of the rst degree,
dy
dx
=

y
x
.
Separate the variables and integrate,

y =

C,
which, on rationalisation, becomes
(x y)
2
2C (x +y) +C
2
= 0
Geometrically this equation represents a system of parabolic curves each of which
touches the axis at a distance C from the origin. The separate equations of the above
solution merely represent dierent branches of the same parabola.
(2) Solve

dy
dx

x
2
y
2

dy
dx
xy = 0. Ansr. xy = C, or x
2
y
2
= C. Hint.
Factors (xp +y) (yp x), where p
dy
dx
.
(3) Solve

dy
dx

2
7
dy
dx
+ 12 = 0. Ansr. y = 4x +C, or 3x +C.
See page
S-13 of
the sup-
plement
for more
details
on the
case ii
method.
KH
Case ii. The equation cannot be resolved into factors, but it can be solved for x,
y, dy/dx, or y/x. An equation which cannot be resolved into factors, can often be
expressed in terms of x, y, dy/dx, or y/x, according to circumstances. The dierential
coecient of the one variable with respect to the other may be then obtained by solving
for dy/dx and using the result to eliminate dy/dx from the given equation.
EXAMPLES. Solve dy/dx + 2xy = x
2
+y
2
. Since (x y)
2
= x
2
2xy +y
2
,
y = x +

dy/dx
Dierentiate
dy
dx
= 1 +
1
2

d
2
y
dx
2

dy
dx
1
2
.
Separate the variables x and p, where p dy/dx, and solve for dy/dx,
x =
1
2
log

p 1

p + 1
+ log C;

dy
dx
=
C +e
2x
C e
2x
.
Ansr. y = x +

C +e
2x

C e
2x

.
(2) Solve x(dy/dx)
2
2y (dy/dx) + ax = 0. Ansr. y =
1
2
C

x
2
+a/C

. Hint.
Substitute for p. Solve for y and dierentiate. Substitute p dx for dy, and clear of
fractions. The variables p and x can be separated. Integrate. p = xC. Substitute in the
given equation for the answer.
(3) Solve x(dy/dx)
2
+ 2y (dy/dx) y = 0. Ansr. (2x +C). Hint. Solve for x.
Dierentiate and substitute dy/p for dx, and proceed as in example (2). yp = C, etc.
Case iii. The equation cannot be resolved into factors, x or y is absent.
If x is absent solve for dy/dx or y according to convenience; if y is absent,
solve for dx/dy or x. Dierentiate the result with respect to the absent
letter if necessary and solve in the regular way.
299
EXAMPLES.(1) Solve (dy/dx)
2
+ x(dy/dx) + 1 = 0. For the sake of greater ease,
substitute p for dx/dy. The given equation thus reduces to
x = p + 1/p. (1)
Dierentiate with regard to the absent letter y, thus,
p =

1 1/p
2

dp/dy; or dy/dp = 1/p + 1/p

2
. (2)
Combining (1) and (2), we get the required solution.
(2) Solve dy/dx = y + 1/y. Ansr. y
2
= Ce
2x
1.
(3) Solve dy/dx = x + 1/x. Ansr. y =
1
2
x
2
+ log x +C.
124. Clairauts Equation. The general type of this equation is,
y = x
dy
dx
+f

dy
dx

; (1)
or, writing p = dy/dx, for the sake of convenience,
y = px +f (p) . (2)
Many equations of the rst degree in x and y can be reduced to this form
by a more or less obvious transformation of the variables, and solved in the
following way:
Dierentiate (2) with respect to x, and equate the result to zero.
p = p +x
dp
dx
+f

(p)
dp
dx
; or, x +f

(p)
dp
dx
= 0.
Hence either
dp
dx
= 0; or, x + f (p) = 0.
If in the former,
dp/dx = 0; p = C,
where C is an arbitrary constant. Hence,
dy = C dx; or, y = Cx +f (C) ,
is a solution of the given equation.
Again, p in x + f (p) may be a solution of the given equation. To nd
p, eliminate p between
y = px +f (p) , or, x +f

(p) = 0.
The resulting equation between x and y also satises the given equation. 67 is
included
in this
docu-
ment
following
the sup-
plement.
There are thus two classes of solutions to Clairauts equation.
EXAMPLES. Find both solutions. in the following equations:
(1) y = px +p
2
. Ansr. Cx +C
2
= y and x
2
+ 4y = 0.
(2) (y px) (p 1). Ansr. (y = Cx) (C 1) = C;

y +

x = 1. Read over ' 67.

300
125. Singular Solutions. Clairauts equation introduces us to a new
idea. Hitherto we have assumed that whenever a function of x and y
satises an equation, that function plus an arbitrary constant, represents
the complete or general solution. We now nd that functions of x and y
can sometimes be found to satisfy the given equation, which, unlike the
particular solution, are not included in the general solution.
This function must be considered a solution, because it satises the
given equation. But the existence of such a solution is quite an accidental
property conned to special equations, hence their cognomen, singular
solutions.
To take the simple illustration of page 149,
y = px +
a
p
. (1)
Remembering that y has been written for dy/dx, dierentiate with respect
to x, we get, on rearranging terms,

x
a
p
2

dp
dx
= 0.
where either x
a
p
2
= 0; or,
dp
dx
= 0.
If the latter,
p = C; or, y = Cx +
a
C
. (2)
If the former, p =

a/x, which gives,.when substituted in (1), the

solution,
y
2
= 4ax. (3)
This solution is not included in the general solution, but yet it satises the
given equation. (3) is the singular solution of (1).
Equation (2), the complete solution of (1), has been shown to represent
a system of straight lines which dier only in the value of the arbitrary
constant C; equation (3), containing no arbitrary constant, is an equation
to the common parabola. A point moving on this parabola has, at any
instant, the same value of dy/dx as if it were moving on the tangent of
the parabola, or on one of the straight lines of equation (2). The singular
solution of a dierential equation is geometrically equivalent to the envelope
of the family of curves represented by the general solution. The singular
solution is distinguished from the particular solution, in that the latter is
contained in the general solution, the former is not.
Again referring to Fig. 78, it will be noticed that for any point on the
envelope, there are two equal values of p or dy/dx, one for the parabola,
one for the straight line.
301
For ex-
panded
discus-
sion of
discrim-
inants,
see sup-
plement
page
S-17.
KH
In order that the quadratic
ax
2
+bx +c = 0,
may have equal roots, it is necessary (page 388) that
b
2
= 4ac; or, b
2
4ac = 0. (4)
This relation is called the discriminant. From (1), since
y = px +a/p; xp
2
yp +a = 0. (5)
In order that equation (5) may have equal roots,
y
2
= 4ax,
as in (4). This relation is the locus of all points for which two values of C
become equal, hence it is called the p-discriminant of (1).
In the same way if C be regarded as variable in the general solution (2),
y = Cx +a/C; or, xC
2
yC +a = 0.
The condition for equal roots, is that
y
2
= 4ax.
which is the locus of all points for which the value of C is the same. It is
called the C-discriminant.
Before applying these ideas to special cases, we may note that the en-
velope locus may be a single curve (Fig. 78) or several (Fig. 79). For
an exhaustive discussion of the properties of these discriminant relations
l must refer the reader to the numerous textbooks on the subject, or to
Cayley, Messenger of Mathematics, 2, 6, 1879. To summarise:
1. The envelope locus satises the original equation but is not in-
cluded in the general solution (see xx

, Fig. 112).
Figure 112: Nodal and Tac Loci
2. The tac locus is the locus passing through the several points where
two non-consecutive members of a family of curves touch. Such a locus is
represented by the line AB (Fig. 79), PQ (Fig. 112). The tac locus does
not satisfy the original equation, it appears in the p-discriminant, but not
in the C-discriminant.
302
3. The node locus is the locus passing through the dierent points
where each curve of a given family crosses itself (the point of intersection
node may be double, triple, etc.). The node locus does not satisfy
the original equation, it appears in the C-discriminant but not in the p-
discriminant. RS (Fig. 112) is a nodal locus passing through the nodes
A,. . . , B,. . . , C,. . . , M.
Figure 113: Cusp Locus
4. The cusp locus
passes through all the cusps
(page 186) formed by the
members of a family of
curves. The cusp locus does
not satisfy the original equa-
tion, it appears in the p- and
in the C-discriminants. It
is the line Ox in Fig. 118.
Sometimes the nodal or cusp loci coincides with the envelope locus.*
EXAMPLES.Find the singular solutions and the nature of the other loci in the
following equations:
(1) xp
2
2yp + ax = 0.
For equal roots y
2
= ax
2
. This satises the original equation and is not included in
the general solution: x
2
2Cy +aC
2
= 0. y
2
= ax
2
is thus the singular solution.
(2) 4xp
2
= (3x a)
2
. General solution: (y + C)
2
= x(x a)
2
.
For equal roots in p, 4x(8x a)
2
= 0, or x(3xa)
2
= 0 (p-discriminant). For equal
roots in C, dierentiate the general solution with respect to C. Therefore
(x +C) dx/dC = 0, or C = x. x(x a)
2
= 0 (C-discriminant) is the condi-
tion to be fullled when the C-discriminant has equal roots. x = 0 is common to the
two discriminants and satises the original equation (singular solution); x = a satis-
es the C-discriminant but not the p-discriminant and, since it is not a solution of the
original equation, x = a represents the node locus; x =
1
3
a satises the p- but not the
C-discriminant nor the original equation (tac locus).
(3) p
2
+ 2xp y = 0.
General solution:

2x
3
+ 3xy +C

2
= 4

x
2
+y

3
; p-discriminant: x
2
+ y = 0; C-
discriminant:

x
2
+y

3
= 0. The original equation is not satised by either of these
equations and, therefore, there is no singular solution. Since

x
2
+y

appears in both
discriminants, it represents a cusp locus.
(4) Show that the complete solution of the equation, y
2

p
2
+ 1

= a
2
, is
y
2
+ (x C)
2
= a
2
; that there are two singular solutions, y = a; that there is a
tac locus on the x-axis for y = 0 (Fig. 79, see also ' 138).
* The second part of van der Waals The Continuity of the Gaseous and Liquid States
of Aggregation Binary Mixtures (1900) has some examples of the preceding mathe-
matics.
303
126. Trajectories. This section will serve as an exercise on some
preceding work. A: trajectory is a curve which cuts another system of
curves at a constant angle. If this angle is 90

the curve is an orthogonal

trajectory.
EXAMPLES. (1) Let xy = C be a system of rectangular hyperbolas, to nd the
orthogonal trajectory, rst eliminate C by dierentiation with respect to x, thus we
obtain,
xdy/dx +y = 0
If two curves are at right angles

1
2
= 90

, then from (17), ' 32,

1
2
= (

), where (17) of
32 es-
tablishes
that
slopes of
m and
1/m
are
normal
to each
other.
KH
,

are the angles made by tangents to the curves at the point of intersection with the
x-axis. But by the same formula,
tan

1
2

= (tan

tan) / (1 + tantan

) .
Now tan
1
2
= and 1/= 0,
tan = cot

; or, dy/dx = dx/dy.*

The dierential equation of the one family is obtained from that of the other by
substituting dy/dx for dx/dy. Hence the equation to the orthogonal trajectory of
the system of rectangular hyperbolas is, xdx + y dy = 0, or x
2
y
2
= C, a system of
rectangular hyperbolas whose axes coincide with the asymptotes of the given system.
(2) For polar coordinates show that we must substitute dr/ (r d) for r d/dr.
(3) Find the orthogonal trajectories of the system of parabolas y
2
= 4ax. Ansr.
Ellipses, 2x
2
+y
2
= C
2
.
(4) Show that the orthogonal trajectories of the equipotential curves, 1/r1/r

= C,
are the magnetic curves cos + cos

= C.
127. Symbols of Operation. It will be found convenient to denote
the symbol of the operation dy/dx by the letter D. If we assume
that the innitesimal increments of the independent variable dx have the
same magnitude, whatever be the value of x, we can suppose D to have a
constant value. Thus
D, D
2
, D
3
, . . . , stand for
d
dx
,
d
2
dx
2
,
d
3
dx
3
, . . . ;
Dy, D
2
y, D
3
y, . . . , stand for
dy
dx
,
d
2
y
dx
2
,
d
3
y
dx
3
, . . .
The operations denoted by the symbols D, D
2
,. . . , satisfy the elementary
rules of algebra except that they are not commutativewith regard to the
variables. For example,
* No doubt the reader sees that in (18), ' 12, dx/dy is the cotangent of the angle
whose tangent is dy/dx.
The so-called fundamental laws of algebra are: I. The law of association: The
number of things in any group is independent of the order. II. The commutative law: (a)
Addition. The number of things in any number of groups is independent of the order.
(b) Multiplication. The product of two numbers is independent of the
304
D(u +v +. . . ) = Du +dv +. . . , (distributive law).
D(Cu) = CD(u) , (commutative law),
where C is a constant. We cannot write D(xy) = D(yx). But,
D
m
D
n
u = D
m+n
u (index law),
is true when m and n are positive integers. If
Du = v; u = D
1
v; or, u =
1
D
v;
v = D D
1
v; or, D D = 1;
that is to say, by operating with D upon D
1
v, we annul the eect of the D
1
operator.
It is necessary to remember later on, that if Dx = 1,
1
D
2
=
1
2
x
2
;
1
D
3
=
1
23
x
3
, . . .
In this notation, the equation
d
2
y
dx
2
( +)
dy
dx
+y = 0,
is written,

D
2
( +) D +

y = 0; or, (D) (D ) y = 0.
Now replace D with the original symbol, and operate on one factor with y, Thus,

d
dx

d
dx

y = 0;

d
dx
y

dy
dx
y

= 0.
By operating on the second factor with the rst, we get the original equation back again.
128. The Linear Equation of the nth Order.
(General Remarks.)
As a general rule the higher orders of dierential equations are more dicult of
solution than equations of the rst order. As with the latter, the more expeditious
mode of treatment will be to refer the given equation to a set of standard cases having
certain distinguishing characters. By far the most important class is the linear equation.
A linear equation of the nth order is one in which the dependent variable and
its n derivatives are all of the rst degree and are not multiplied together. The typical
form in which it appears is
d
n
y
dx
n
+X
1
d
n1
y
dx
n1
+ +X
n
y = X. (1)
order. III. The distributive law: (a) Multiplication. The multiplier may be distributed
over each term of the multiplicand, e.g., m(a+b) = ma+mb. (b) Division. (a+b)/m =
a/m + b/m. IV. The index law: (a) Multiplication. a
m
a
n
= a
m+n
. (5) Division.
a
m
/a
n
= a
mn
.
305
Or, in symbolic notation,
D
n
y +X
1
D
n1
y + +X
n
y = X,
where X, X
1
, . . . , X
n
, are either constant magnitudes, or functions of the
independent variable x. If the coecient of the highest derivative be other
than unity, the other terms of the equation can be divided by this coe-
cient. The equation will thus assume the typical form (1). We have studied
the linear equation of the rst order in ' 123. For the sake of xing our
ideas, the equation
d
2
y
dx
2
+P
dy
dx
+Qy = r, (2)
of the second order, will be taken as typical of the class. P, Q, R have the
meaning above attached to X
1
, X
2
, X.
The general solution of the linear equation is made my of two parts.
1. The complementary function which is the most general solution
of the left-hand side of equation (2) equated to zero, or,
d
2
y
dx
2
+P
dy
dx
+Qy = 0. (3)
The complementary function involves two arbitrary constants.
2. The particular integral which is any solution of the original equa-
tion (2), the simpler the better. In particular cases when the right-hand
side is zero, the particular integral does not occur.
To show that the general solution of (2) contains a general solution of
(3). Assume that the complete solution of (2) may be written,
y = u +v, (4)
where v is any function of x which satises (2), that is to say, v is the
particular integral* of (2), u is the general solution of (3), to be determined.
Substitute (4) in (2).
d
2
u
dx
2
+P
du
dx
+Qu +
d
2
v
dx
2
+P
dv
dx
+Qv = R.
But
d
2
v
dx
2
+ P
dv
dx
+Qv = R;
therefore,
d
2
u
dx
2
+P
du
dx
+Qu = 0.
Therefore, u must satisfy (3).
Given a particular solution* of the linear equation, to nd the
* Not to be confused with the particular solution of page 289.
306
complete solution. Let y = v be a particular solution of the following
equation,
d
2
y
dx
2
+P
dy
dx
+Qy = 0,
where P and Q are functions of x. Substitute y = uv, See page
S-21 of
the sup-
plement
for the
deriva-
tion
of this
equation.
KH
v
d
2
u
dx
2
+

2
dv
dx
+Pv

du
dx
= 0.
This equation is of the rst order and linear with du/dx as the dependent
variable. Put du/dx = z and
v
dz
dx
+

2
dv
dx
+Pv

z = 0;
dz
z
+ 2
dv
v
+P dx = 0;
log
du
dx
+ 2 log v +

P dx = 0; or,
du
dx
= C
1
1
v
2
e

P dx
.
u = C
1

1
v
2
e

P dx
+C
2
; or, C
1
v

1
v
2
e

P dx
+C
2
v,
where C
1
and C
2
are arbitrary constants.
EXAMPLES.(1) If y = e
ax
is a particular solution of d
2
y/dx
2
= a
2
y, show that the
complete solution is y = C
1
e
ax
+C
2
e
ax
.
(2) If y = x is a particular solution of

1 x
2

d
2
y/dx
2
xdy/dx + y = 0, the
complete solution is y = C
1

1 x
2
+C
2
x.
If a particular solution of the linear equation is known, the order of the
equation can be lowered by unity. This follows directly from the preceding
result. If y = v is a known solution, then, if y = tv be substituted in the
rst member of the equation, the coecient of t in the result, will be the
same as if t were constant and therefore zero. t being absent, the result
will be a linear equation in t but of an order less by unity than that of
the given equation. It follows directly, that if n particular solutions of the
equation are known, the order of the equation can be reduced n times.
For the description of a machine designed for solving (3), see Proceedings
of the Royal Society, 24, 269, 1876 (Lord Kelvin).
129. The Linear Equation with Constant Coecients. The in-
tegration of these equations obviously resolves itself into nding the com-
plementary function and the particular integral.
First, when the second member is zero, in other words, to nd the com-
plementary function of any linear equation with constant coecients. The
typical equation is,
d
2
y
dx
2
+P
dy
dx
+Qy = 0, (1)
307
where P and Q are constants. The particular integral does not appear in
the solution.
If the equation were of the rst order, its solution would be, y = Ce

mdx
.
On substituting e
mx
for y in (1), we obtain

m
2
+Pm+Q

e
mx
= 0,
provided m
2
+Pm+Q = 0. (2)
This equation is called the auxillary equation. If m
1
be one value of
m which satises (1), then y = e
m
1
x
, is an integral of (1). But we must go
further.
Case 1. When the auxillary equation has two unequal roots, say m
1
,
and m
2
the general solution of (1) may be written down without any further
trouble.
u = C
1
e
m
1
x
+C
2
e
m
2
x
. (3)
EXAMPLES.(1) Solve

D
2
+ 14D32

y = 0. Assume y = Ce
mx
is a solution. The
auxillary becomes, m
2
+ 14m32 = 0. The roots are m = 2, or m = 16. The required
solution is, therefore, y = C
1
e
2x
+C
2
e
16x
.
(2) Solve d
2
y/dx
2
m
2
y = 0. Ansr. y = C
1
e
mx
+C
2
e
mx
(see page 319).
(3) Show that y = C
1
e
3x
+C
2
e
x
is a complete solution of
d
2
y/dx
2
+ 4 dy/dx + 3y = 0.
Case 2. When the two roots of the auxillary are equal. if m
1
= m
2
in
(3), it is no good putting (C
1
+C
2
) e
m
1
x
as the solution, because C
1
+C
2
, is
really one constant. The solution would then contain one arbitrary constant
less than is required for the general solution. To nd the other particular
integral, it is usual to put
m
1
= m
2
+h,
where h is some nite quantity which will ultimately be made zero. With
this proviso, we write the solution,
y = lim
h0
C
1
e
m
1
x
+C
2
e
(m
2
+h)x
.
Hence, y = lim
h0
e
m
1
x

C
1
+C
2
e
hx

.
Now expand e
hx
by Maclaurins theorem (page 230). See page
S-22 of
the sup-
plement
for an al-
ternative
devel-
opment
of this
general
solution.
KH
y = lim
h0
e
m
1
x

C
1
+C
2

1 +hx +
1
2
h
2
x
2
+. . .

;
= lim
h0
e
m
1
x

C
1
+C
2
+C
2

hx +
1
2!
h
2
x
2
+. . .

;
= lim
h0
e
m
1
x

A+Bx +
1
2!
C
2
h
2
x
2
+C
2
R

,
where R denotes the remaining terms of the expansion of e
hx
, A = C
1
+C
2
,
B = C
2
h. Therefore, at the limit,
y = e
m
1
x
(A+Bx) . (4)
For the sake of uniformity, we shall still write the arbitrary integration
constants C
1
, C
2
, C
3
, . . .
308
For an equation of a still higher degree, the preceding result may be
written,
y = e
m
1
x

C
1
+C
2
x +C
3
x
2
+ +C
r2
x
r1

. (5)
where r denotes the number of equal roots.
EXAMPLES.(1) Solve d
3
y/dx
3
d
2
y/dx
2
dy/dx + u = 0. Assume y = e
mx
. The
auxillary equation is m
3
m
2
m + 1 = 0. The roots are 1, 1, 1. Hence the general
solution can be written down at sight:
y = C
1
e
x
+ (C
2
+C 3x) e
x
.
(2) Solve

D
3
3D
2
+ 4

y = 0. Ansr. e
2x
(C
1
+C
2
x) +C
3
e
x
.
See S-66
of the
supple-
ment for
how case
3 arises
in the
equa-
tions
for the
orbits of
planets.
KH
Case 3. When the auxillary equation has imaginary roots, all unequal.
Remembering that imaginary roots are always found in pairs in equations
with real coecients (page 386), let the two imaginary roots be
m
1
= +; and m
2
= .
Instead of substituting y = e
mx
in (3), we substitute these values of m in
(3) and get
y = C
1
e
(+)x
+C
2
e
()x
;
= e
x

C
1
e
x
+C
2
e
x

;
= e
x
C
1
(cos x + sin x) +C
2
(cos x sin x) . (6)
(See the chapter on Hyperbolic Functions.) Separate the real and imag-
inary parts, as in Ex. 3, p. 280,
y = e
x
(C
1
+C
2
) cos x + (C
1
C
2
) sin x;
if we put C
1
+C
2
= A, (C
1
C
2
) = B.
y = e
x
(Acos x +Bsin x) . (7)
In order that the constants A and B in (7) may be real, the constants C
1
and C
2
must include the imaginary parts.
EXAMPLES.(1) Show from (6) that
y = (coshx + sinhx) (A
1
cos x +B
2
sin x) .
(Exercise on Chapter VI.)
(2) Integrate d
2
y/dx
2
+ dy/dx + y = 0. The roots are =
1
2
and =
1
2

3;
y = e
x/2

Acos
1
2

3x +Bsin
1
2

3x

.
(3) The equation of a point vibrating under the inuence of a periodic force, is,
d
2
x
dt
2
+
2
x = a cos 2
t
T
.
Find the complementary function. The roots are . From (7)
x = Acos t +Bsin t.
(4) If

D
3
D
2
+D1

y = 0, y = C
1
cos x +C
2
sin x +Ce
x
.
Case 4. When some of the imaginary roots of the auxillary equation
are equal. If a pair of the imaginary roots are repeated,
309
we may proceed as in Case 2, since, when m
1
= m
2
, C
1
e
m
1
x
+ C
2
ee
m
2
x
is
replaced by (A +Bx) e
m
1
x
; similarly, when m
3
= m
4
, C
3
e
m
3
x
+C
3
e
m
4
x
may
be replace by (C +Dx) e
m
3
x
. If, therefore,
m
1
= m
2
= +; and m
3
= m
4
= ,
the solution
y = (C
1
+C
2
x) e
(+)x
+ (C
3
+C
4
x) e
()x
,
becomes y = e
x
(A+Bx) cos x + (C +Dx) sin x. (8)
EXAMPLES.(1) Solve

D
4
12D
3
+ 62D
2
156D + 169

y = 0. Given the roots

of the auxillary: 3 + 2, 3 + 2, 3 2, 3 2. Hence
y = e
3x
(C
1
+ C
2
x) sin2x + (C
3
+C
4
x) cos 2x .
(2) If

D
2
+ 1

2
(D1)
2
y = 0, (A +Bx) sinx + (C +Dx) cos x + (E +Fx) e
x
.
Second, when the second member is not zero, that is to say to nd
both the complementary function and the particular integral. The general
equation is,
d
2
y
dx
2
+P
dy
dx
+Qy = R, (9)
where P and Q are constant, R is a function of x. We have just shown
how to nd one part of the complete solution of the linear equation with
constant coecients, namely, by putting R, in (9), equal to zero. The
remaining problem is to nd a particular integral of this equation. The
more useful processes will be described in the next section.
In the symbolic notation, (9) may be written,
f (D) y = R (10)
The particular integral is, therefore,
y = f (D)
1
R; or y =
R
f (D)
. (11)
The right-hand side of either of equations (11), will be found to give a
satisfactory value for the particular integral in question.
Since the complementary function contains all the constants necessary
for the complete solution of the dierential equation, it follows that no
integration constant must be appended to the particular integral.
130. How to nd Particular Integrals. It will be found quickest to
proceed by rule:
Case 1 (General). When the operator f (D)
1
can be resolved into factors.
310
There
are other
(some-
times
easier)
methods
of nding
par-
ticular
integrals.
See the
supple-
ment
starting
at the
bottom
of page
S-23 for
details.
KH
We have seen that the linear dierential equation of the rst order,
dy/dx ay = R; or y = R/ (Da) , (1)
is solved by
y = e
ax

e
ax
Rdx. (2)
The term Ce
ax
in the solution of (1), belongs to the complementary func-
tion.
Suppose that in a linear equation of a higher order, say,
d
2
y/dx
2
5 dy/dx 6y = R,
the operator f (D)
1
can be factorised. The complementary function is
written down at sight from,

D
2
5D + 6

y = 0; or (D 3) (D 2) y = 0.
namely, y = C
1
e
3x
+C
2
e
2x
. (3)
The particular integral is
y
1
=
1
(D 3) (D 2)
R =

1
D 3

1
D 2

R;
= e
3x

e
3x
Rdx e
2x

e
2x
Rdx, (4)
from (2). The general solution is the sum of (3) and (4),
y = C
1
e
3x
+C
2
e
2x
+e
3x

e
3x
Rdx e
2x

e
2x
Rdx.
EXAMPLES.(1) In the preceding illustration, put R = e
4x
and show that the general
solution is, C
1
e
3x
+C
2
e
2x
+
1
2
e
4x
.
(2) If

D
2
4D + 3

y = 2e
3x
, y = C
1
e
x
+C
2
e
3x
+xe
3x
.
Case 2 (General). When the operator f (D)
1
can be resolved into
partial fractions with constant numerators. The way to proceed in this
case is illustrated in the rst example below.
EXAMPLES.(1) Solve d
2
y/dx
2
3 dy/dx +2y = e
3x
. In symbolic notation this will
appear in the form,
(D 1) (D2) y = e
3x.
The complementary function is y = C
1
e
x
+ C
2
e
3x
. The particular integral is obtained
by putting
y =
1
(D2) (D 1)
e
3x
;

1
D2

1
D 1

e
3x
,
according to the method of resolution into partial fractions. Operate with the rst
symbolic factor, as above,
y
1
= e
2x

e
2x
e
3x
dx e
x

e
x
e
3x
dx =
1
2
e
3x
,
The complete solution is, therefore, y = C
1
e
x
+C
2
e
2x
+
1
2
e
3x
.
(2) Solve (D 2)
2
y = e
x
. Ansr. y = C
1
e
2x
+C
2
xe
2x
+e
x
.
Case 3 (Special). When R is a rational function of x, say x
n
. This case
is comparatively rare. The procedure is to expand. f (D)
1
in ascending
powers of D as far as the highest power of x in R.
311
EXAMPLES.(1) Solve d
2
y/dx
2
4 dy/dx + 4y = x
2
. The complementary function
is y = e
2x
(A+Bx); the particular integral is:
(2 D)
2
x
2
=
1
4

1 + 2
D
2
+ 3
D
2
2

x
2
=
1
8

2x
2
+ 4x + 3

.
(2) If d
2
y/dx
2
y = 2 + 5x, y = C
1
e
x
+C
2
e
x
5x 2.
Case 4 (Special). When R contains an exponential factor, so that
R = e
ax
X,
where X may or may not be a function of x and a, has some constant value.
i. When X is a, function of x. Since D
n
e
ax
= a
n
e
ax
, where n is any positive integer
(page 38), we have (page 95)
D(e
ax
X) = e
ax
DX +ae
ax
X = e
ax
(D +a) X,
and generally, as in Leibnitz theorem (page 49), See page
S-25 of
the sup-
plement
for better
notation
of (5)
KH
D
n
e
ax
X = e
ax
(D +a)
n
X;

D
n
e
ax
(D +a)
n
= e
ax
X; and
e
ax
(D +a)
n
X =
e
ax
D
n
X.
(5)
The operation

D
1
e
ax
X is performed (when X is any function of x) by transplanting
e
ax
from the right- to the left-hand side of the operator f (D)
1
and replacing D by
(D +a). This will, perhaps, be better understood from the following examples:
EXAMPLES.(1) Solve d
2
y/dx
2
2 dy/dx + y = x
2
e
3x
. The complete solution by
page 308, is (C
1
+xC
2
) e
x
+ (D 2D + 1)
1
x
2
e
3x
. From (5),
1
D
2
2D + 1
x
2
e
3x
=
1
(D1) (D1)
x
2
e
3x
.
By rule: e
3x
may be transferred from the right to the left side of the operator provided
we replace D by D + 3.
e
3x
1
(D + 2)
2
x
2
.
We get e
3x

1
4
x
2

1
2
x +
3
8

,
as the value of the particular integral.
(2) Evaluate (D 1)
1
e
x
log x. Ansr. xe
x
log x/e.
ii. When X a constant. If X is constant, the operation (5) reduces to
1
f (D)
e
ax
=
1
f (a)
e
ax
. (6)
The operation f (D) e
ax
X is performed by replacing D +a by a.
EXAMPLES.(1) Find the particular integral in

D
2
3D + 2

y = e
3x
. Obviously,
1
D
2
3D + 2
e
3x
=
1
3
2
3 3 + 2
e
3x
=
1
2
e
3x
.
(2) Show that
1
4
e
x
, is a particular integral in
d
2
y/dx
2
+ 2 dy/dx + 1 = e
x
.
312
An anomalous case arises when a, is a root of f (D) = 0. By this
method, we should get for the particular integral of dy/dx y = e
x
.
1
D 1
e
x
=
e
x
1 1
= e
x
.
The diculty is evaded by using the method (5) instead of (6). Thus,
1
D 1
e
x
= e
x
1
D
1 = xe
x
The complete solution is, therefore, y = Ce
x
+xe
x
.
Another mode of treatment is the following: Since a is a root of f (D) = 0,
by hypothesis, D a is a factor of f (D) (page 386). Hence,
f (D) = (D a) f

(D) ;

1
f (D)
e
ax
=
1
(D a)

1
f

(D)
e
ax
=
1
(D a)

1
f

(a)
e
ax
=
xe
ax
f

(a)
.
(7)
If the root a occurs r times in f (D) = 0, then Da enters r times into
f (D). Therefore,
1
f (D)
e
ax
=
1
(D a)
r

1
f

(D)
e
ax
=
1
(D a)
r

1
f

(a)
e
ax
=
x
r
e
ax
rf

(a)
. (8)
EXAMPLES. Find the particular integrals in, (1) (D + 1)
3
y = e
x
. Ansr.
1
6
x
3
e
x
.
Hint. Replace D by D1. e
x
D
3
, etc. See page 312.
(2)

D
3
1

y = xe
x
. Ansr. e
x

1
6
x
2

1
3
x

. Hint. First get e

x
(D1)
1
x, then
e
x
(1 +D +. . . ) x, etc.
Case 5 (Special). When R contains sine or cosine factors. By the
successive dierentiation of sin (nx +a),

D
2

n
sin (nx +a) =

n
2

n
sin (nx +a) .
where n and a are constants.
f

D
2

sin (nx +a) = f

n
2

sin(nx +a).
1
f (D
2
)
sin (nx +a) =
1
f (n
2
)
sin (nx +a) . (9)
It can be shown in the same way that,
1
f (D
2
)
cos (nx +a) =
1
f (n
2
)
cos (nx +a) . (10)
EXAMPLES.(1) Find the particular integral of
d
3
y/dx
3
+d
2
y/dx
2
+dy/dx +y = sin 2x.
Here,
R
f (D)
=
1
D
3
+D
2
+D + 1
sin 2x =
1
(D
2
+ 1) +D(D
2
+ 1)
sin 2x.
* The proof resembles a well-known result in trigonometry, ' 19:
D(sin nx) = d (sin nx) /dx = ncos nx;
D
2
(sin nx) = d
2
(sin nx) /dx
2
= n
2
cos nx, etc.
313
Substitute for D
2
=

2
2

as in (9). We thus get

1
3
(D + 1)
1
sin2x. Multiply by
D 1 and again substitute D
2
=

2
2

in the result. Thus

1
15
(D 1) sin 2x, or
1
15
(2 cos 2x sin2x)
is the desired result.
(2) Solve d
2
y/dx
2
k
2
y = cos mx. Ansr. C
1
e
kx
+C
2
e
kx
(cos mx) /

m
2
+k
2

.
(3) If and are the roots of the auxillary equation derived from
d
2
y/dt
2
+mdy/dt +n
2
y = a sinnt,
(Helmholtzs equation for the vibrations of a tuning-fork) show that,
C
1
e
t
+C
2
e
t
(a cos nt) /mn.
is the complete solution.
An anomalous case arises when D
2
in D
2
+n
2
is equal to n
2
. For in-
stance, the complementary function of d
2
y/dx
2
+n
2
y = cos nx, is C
1
cos nx+
C
2
sin nx, the particular integral is (D
2
+n
2
) cos nx. If the attempt is made
to evaluate this, by substituting D
2
= n
2
, we get (cos nx) / (n
2
+n
2
) =
cos nx. We were confronted with a similar diculty on page 243. The
treatment is practically the same. We take the limit of (D
2
+n
2
) cos nx,
when n becomes n +h and h converges towards zero.
1
n
2
+n
2
cos nx; or
1
(n +h)
2
+ n
2
cos (n +h) x.
lim
h0
1
(n +h)
2
+n
2
cos (n +h) x = lim
h0
1
2nh h
2
cos (nx +hx) ;
= lim
h0

1
2nh +h
2
(cos nx cos hx sin nx sin hx) ;
= lim
h0

1
2nh +h
2

cos nx

1
h
2
x
2
2!
+. . .

sinnx(hx . . . )

;
= lim
h0

1
2n +h

cos nx
h
xsin nx + powers of h

.
But cos nx is contained in the complementary function and hence, when h = 0, we
obtain,
xsin nx
2n
+ (a term in the complementary function) .
This latter may be disregarded when the particular integral alone is under consideration.
The complete solution is, therefore,
y = C
1
cos nx +C
2
sin nx + (xsin nx) /2n.
EXAMPLES.(1) Show that
1
2
xcos x, is the particular integral of

D
2
+ 1

y = sin x
(2) Evaluate

D
2
+ 4

1
cos 2x. Ansr.
1
4
xsin 2x.
(3) Evaluate

D
2
+ 4

1
sin 2x. Ansr.
1
4
xcos 2x.
(4) Solve d
3
y/dx
3
y = xsin x. The particular integral consists of two parts,
1
2
(x 3) cos x xsin x. The complementary function is
C
1
e
x
+C
2
e
x/2
sin

1
2

3x

+C
3
e
x/2
cos

1
2

3x

.
But see next case.
Case 6. When R contains some power of x as a factor. Say,
R = nX,
314
where X is any function of x. The successive dierentiation of two products See page
S-31 of
the sup-
plement
for the
Laplace
method,
which is
a single
approach
to solv-
ing all
6 of the
cases
in this
section
and
more.
20 is
included
following
the sup-
plement.
KH
gives, ' 20,
D
n
xX = xD
n
X +nD
n1
X.
f (D) xX = xf (D) X +f

(D) X.
Substitute Y = f (D) X, where Y is any function of x. Operate with
f (D)
1
, we get
1
f (D)
xY =

x
1
f (D)
f

(D)

1
f (D)
Y. (11)
EXAMPLES.(1) Find the particular integral in d
3
y/dx
3
y = xe
2x
. From (11), the
particular integral is
=

x
1
D
3
1
3D
2

1
D
3
1
e
2x
=

x
1
7
3 4

1
7
e
2x
, etc.
(2) Show in this way, that the particular integral of

D
4
1

y = xsin x,
is
1
8

x
2
cos x 3xsinx

.
(3) Solve d
2
y/dx
2
y = xe
x
sin x.
Ansr. y = C
1
e
x
+ C
2
e
x

1
25
e
x
(10x + 2) cos x + (5x 14) sin x.
(4) Integrate d
2
y/dx
2
y = x
2
cos x. Ansr. y = C
1
e
x
+C
2
e
x
+xsinx+
1
2
cos x

1 x
2

.
131. The Linear Equation with Variable Coecients.
Case 1. The homogeneous linear dierential equation. The general
type of this equation is:
x
n
d
n
y
dx
n
+a
1
x
n1
d
n1
y
dx
n1
+ +a
n
y = X, (1)
where X is a function of x; a
1
, a
2
, . . . , a
n
are constants. This equation can
be transformed. into one with constant coecients by the substitution of
x = e
z
; or z = log x.
we then have,
dy/dx = e
z
and therefore, xdy/dx = dy/dz. (2)
Just as we have found it very convenient to employ the symbol, D to
denote the operation,
d
dx
, so we shall nd it even more convenient to
denote the operation, x
d
dx
, by the symbol, . is treated in ex-
actly the same manner as we have treated D* in ' 128 and subsequently.
* A little care is required in using this new notation. The operations of dierentia-
tion and multiplication by a variable are not commutative. The operation x
2
D
2
is not
the same as
2
, or as xD xD. But we must write,
xDy = y;
x
2
D
2
y = ( 1) y;
x
3
D
3
y = ( 1) ( 2) y;
. . . . . .
x
n
D
n
y = ( 1) ( 2) . . . ( n + 1) y.
315
EXAMPLES.(1) = xD = x
d
dx
=
d
dz
.
(2) Show that x
m
= mx
m
i. The complementary function. From the rst of equations (2), we
have ' 12, 9,
dy
dx
=
dy
dz

dz
dx
=
1
x

dy
dx
;
d
2
y
dx
=
1
x

d
2
y
dx
2

dy
dz

; etc.
Substitute these values in (1). The equation reduces to one with constant
coecients which may be treated by the methods described in the preceding
sections.
EXAMPLES.(1) Solve x
3
d
3
y/dx
3
+ 3x dy/dx 3y = x
2
+x.
( 1) ( 2) y + 3y 3y = e
2z
+e
z
.
( 1)

2
+ 3

y = e
2z
+e
z
.
y = C
1
e
z
+C
2
cos

+C
3
sin

+
1
7
e
2z
+
1
4
ze
z
.
y = C
1
x +C
2
cos

3 log x

+C
3
sin

3 log x

+
1
7
x +
1
4
xlog x.
(2) Solve x
2
d
2
y/dx
2
+x dy/dx +q
2
y = 0; i.e.,

2
+q
2

y = 0.
Ansr. y = C
1
sin(q log x) +C
2
cos (q log x).
The linear equation with variable coecients bears the same relation
to x, that the equation with constant coecients does to e
mx
. Hence if
x
m
be substituted for y, the factor x
m
will divide out from the result and
an equation in m will remain. The n roots of this latter equation will
determine the complementary function.
EXAMPLES.(1) Solve x d
2
y/dx
2
+ 2x dy/dx 2y = 0. Put y = x
m
. We get
m(m1) + 2 (m1) = 0; or (m + 2) (m1) = 0.
Hence from our preceding results, we can write down the complementary function at
sight, y = C
1
x +C
2
x
2
.
(2) Solve x
2
d
2
y/dx
2
+ 4x dy/dx + 2y = 0. Ansr. y = C
1
/x +C
2
/x
2
.
(3) Find the complementary function in ( 1) 3 + 4 y = x
3
.
Ansr. y = (C
1
+C
2
log x) x
2
.
(4) Integrate ( 1) 2 y = 0. Ansr. y = C
1
x
2
+C
2
/x.
ii. The particular integral. We may use the operator , to obtain the
particular internal of linear equations with variable coecients in the same
way that D was used to determine the particular integral of equations with
constant coecients.
The symbolic form of the particular integral is,
y =
R
f ()
.
The operator f ()
1
may be resolved into partial fractions or into factors
as in the case of D.
316
EXAMPLES.(1) Show that y = C
1
x
4
+ C
2
/x +
1
5
x
4
log x a complete solution of
x
2
d
2
y/dx
2
2x dy/dx 4y = x
4
.
(2) Find the value of
1

2
(3)
x
3
. Using the ordinary method just described we get
the indeterminate form
1
9

x
3
3
. In this case we must adopt the method of page 312 and
write
1
9
x
3
1
1
1 =
1
9
x
3

dx
x
=
1
9
x
3
log x.
(3) Solve x
3
d
2
y/dx
2
+ 7x dy/dx + 5y = x
5
. Write this
( 1) + 7 y = x
5
.
The particular integral is

2
+ 6 + 5

1
x
5
, or x
5
/60. complementary function is
C
1
x
1
+C
2
x
5
.
(4) Solve x
2
d
2
y/dx
2
+ 4x dy/dx + 2y = e
x
.
Ansr. y = C
1
/x +C
2
/x
2
+e
x
/x
2
.
(5) Solve x
3
d
3
y/dx
3
+ 2x
2
d
2
y/dx
2
x dy/dx +y = x +x
3
.
Ansr. y = C
1
/x +C
2
x +C
3
xlog x +
1
4
x(log x)
2
+x
3
/16.
(6) Solve x
3
d
3
y/dx
3
+ 2x
2
d
2
y/dx
2
+ 2y = 10x + 10/x.
Ansr. y = C
1
xcos (log x) +C
2
xsin (log x) + 5x +C
3
/x + (2 log x) /x.
(7) Find the particular integral of the third example in the last set. Ansr. x
3
.
(8) Equate example (2), of the preceding set, to 1/x instead of to zero, and show
that the particular integral is then (log x) /x.
Here the
latter
appears
to be
(3) and
the rst
example
but the
former
not to
have an
example
shown.
Other
than
choice of
symbols,
the two
substitu-
tions are
equiv-
alent.
KH
Case 2. Legendres Equation. Type:
(a +bx)
n
d
n
y
dx
n
+A
1
(a +bx)
n1
d
n1
y
dx
n1
+ +A
n
y = R, (3)
where A
1
, A
2
, . . . A
n
are constants, R is any function of x. This sort of
equation is easily transformed into the homogeneous equation and, there-
fore, into the linear equation with constant coecients. To make the former
transformation, substitute z = ax +b, for the latter, e = a +bx.
EXAMPLES.(1) Solve
(a +bx)
2
d
2
y/dx
2
+b (a +bx) dy/dx +c
2
y = 0.
Ansr. y = C
1
sin (c/b) log (a +bx) +C
2
cos (c/b) log (a +bx).
(2) Solve (x +a)
2
d
2
y/dx
2
4 (x +a) dy/dx + 6y = x.
Ansr. y = C
1
(x +a)
2
+C
2
(x +a)
2
+
1
6
(3x + 2a).
132. The Exact Linear Dierential Equation. A very simple
relation exists between the coecients of an exact dierential equation
which may be used to test whether the equation is exact or not. Take the
equation,
X
0
d
3
y
dx
3
+X
1
d
2
y
dx
2
+X
2
dy
dx
+X
3
y = R. (1)
317
where X
1
, X
2
, . . . , R are functions of x. Let their successive dierential
coecients be indicated by dashes, thus X

, X

, . . .
Since X
0
d
3
y/dx
3
has been obtained by the dierentiation of X
0
d
2
y/dx
2
,
this latter is necessarily the rst term of the integral of (1). But,
d
dx

X
0
d
2
y
dx
2

= X
0
d
3
y
dx
3
+X

0
d
2
y
dx
2
.
Subtract the right-hand side of this equation from (1),
(X
1
X

0
)
d
2
y
dx
2
+X
2
dy
dx
+X
3
y = R. (2)
Again, the rst term of this expression is a derivative of (X
1
X

0
) dy/dx.
This, therefore, is the second term of the integral of (1). Hence, by dier-
entiation and subtraction, as before,
(X
2
X

1
+X

0
)
dy
dx
+X
3
y = R. (3)
This equation may be deduced by the dierentiation of (X
2
X

1
+X

0
) y,
provided the rst dierential coecient of (X
2
X

1
+X

0
) with respect to
x, is equal to X
3
that is to say,
X

2
X

1
+X

0
= X
3
; or X
3
X

2
+X

1
X

0
= 0. (4)
But if this is really the origin of (3), the original equation (1) has been
reduced to a lower order, namely,
X
0
d
2
y
dx
2
+ (X
1
X

0
)
dy
dx
+ (X
2
X

1
+X

0
) y =

Rdx + C. (5)
This equation is called. the rst integral of (1), because the order of the See sup-
plement
page
S-39 for
clari-
cation
of how
the rst
integral
arises.
KH
original equation has been lowered unity, by a process of integration.
Condition (4) is a test of the exactness of a dierential equation.
If the rst integral is an exact equation, we can reduce it, in the same
way, to another rst integral of (1). The process of reduction may be
repeated until an inexact equation appears, or until y itself is obtained.
Hence, an exact equation of the nth order has n independent rst integrals.
EXAMPLES.(1) Is the equation
x
5
d
3
y/dx
3
+ 15x
4
d
2
y/dx
2
+ 60x
3
dy/dx + 60x
2
y = e
x
exact?
From (4), X
3
= 60x
2
; X

2
= 180x
2
; X

1
= 180x
2
; X

0
= 60x
2
. Therefore,
X
3
X

2
+X

1
X

0
= 0 and the equation is exact.
(2) Solve the equation, x d
3
y/dx
3
+

x
2
3

d
2
y/dx
2
+4x dy/dx +2y = 0, as far
as possible, by successive reduction. The process can be employed twice, the residue is
a linear equation of the rst order, not exact.
(3) Solve the equation given in example (1).
Ansr. x
5
y = e
x
+C
1
x
2
+C
2
x +C
3
.
318
There is another quick practical test for exact dierential equations
(Forsyth) which is not so general as the preceding. When the terms in
X are either in the form of ax
m
, or of the sum of expressions of this type,
x
m
d
n
y/dx
n
is a perfect dierential coecient, if m < n. This coecient
can then be integrated whatever be the value of y. If m = n or m > n,
the integration cannot be performed by the method for exact equations.
To apply the test, remove all the terms in which m is less than n, if the
remainder is a perfect dierential coecient, the equation is exact and the
integration may be performed.
EXAMPLES.(1) Apply the test to
x
3
d
4
y/dx
4
+x
2
d
3
y/dx
3
+x dy/dx +y = 0.
x dy/dx + y remains. This has evidently been formed by the operation D(xy), hence
the equation is a perfect dierential.
(2) Apply the test to

x
3
D
4
+x
2
D
3
+x
2
D + 2x

y = sin x.
x
2
dy/dx + 2xy remains. This is a perfect dierential, formed from D

x
2
y

. The
equation is exact.
e.g., if
between
the 1st
integral
and
the 1st
integral
of the 1st
integral
of the
equa-
tion...
KH
If two independent rst integrals are known the equation is sometimes
easily solved. The elimination of dy/dx between two rst integrals will give
the complete solution.
133. The Integration of Equations with Missing Terms. Dier-
ential equations with missing letters are common.
First, the independent variable is absent. Type:
d
2
y/dx
2
= qy; or d
2
y/dx
2
= q f (y) . (1)
This equation is, in general, neither linear nor exact.
Case 1. When f (y), in (1), is negative, so that
d
2
y
dx
2
+q
2
x = 0, (2)
where the academic x and y have given place to t and x respectively, in
order to give the equation the familiar form of the equation of the motion
of a particle under the inuence of a central attracting force.
Multiply both sides of the equation by 2dx/dt, and integrate with re-
spect to x,
2
dx
dt

d
2
x
dt
2
= 2q
2
x
dx
dt
; or

dx
dt

2
= q
2
x
2
+C.
Separate the variables and integrate again,
dx

2
x
2
= q dt; cos
1
x

= (qt +) ,
319
where is the integration constant and C = q
2

2
. The solution involves
two arbitrary constants and , which respectively denote the amplitude
and epoch of a simple harmonic motion, whose period of oscillation is 2/q.
Put C
1
= cos , and C
2
= sin .
x = C
1
cos qt +C
2
sin qt.
[Case 2.] When f (y) in (1) is positive, the solution assumes the form,
x = C
1
e
qt
+C
2
e
qt
; or x = Acosh qt +Bsinh qt,
as on page 309. All these results are important in connection with alter-
nating currents and other forms of harmonic motion.
Another way of treating equations of type (9), occurs with an equation
like
y
d
2
y
dx
2
+

dy
dx

2
2y
2
= 0, (3)
which has the form of the standard. equation for the small oscillations
of a pendulum in air. Under this condition, the resistance of the air is
negligible. Let
p = dy/dx, d
2
y/dx
2
= dp/dx = p dp/dy.
Substitute these results in the given equation, multiply through with 2/y.
py
dp
dy
+p
2
= 2y
2
; or 2p
dp
dy
+
2
y
p
2
= 4y.
Multiply by y
2
and
d (y
2
p
2
)
dy
= 4y
3
; or p
2
y
2
= y
4
+C
4
,
where C
4
is an arbitrary constant. The rest is obvious.
EXAMPLES.(1) The solution of equation (3) is sometimes written in the form
y
2
= C
2
1
sinh (2x +C
2
) .
Verify this.
(2) Solve d
2
x/dt
2
+ x + = 0. Put x = x
1
+ / and afterwards omit the sux.
Ansr. x = / +C
1
cos t

+C
2
sint

.
(3) If the term x in the preceding example had been of opposite sign, show that
the solution would have been, x = /+C
1
cosht

+C
2
sinh t

, where is negative.
(4) Solve d
2
y/dx
2
a (dy/dx)
2
= 0. Ansr. C
1
x +C
2
= e
ay
.
(5) Solve 1 + (dy/dx)
2
= y d
2
y/dx
2
. Ansr. y = a cosh(x/a +b).
(6) Fouriers equation for the propagation of heat in a cylindrical bar, is
d
2
V/dx
2

2
V = 0. Hence show that V = C
1
e
x
+C
2
e
x
.
Second, the dependent variable is absent. Type:
d
2
y/dx
2
= x; or d
2
y/dx
2
= f (x) . (4)
320
If these equations are exact, they may be solved by successive integration.
If the equation has the form
d
2
y/dx
2
+dy/dx +x = 0.
Say,
d
2
v
dr
2
+
1
r

dv
dr
=
P
l
,
a familiar equation in hydrodynamics, it is usually solved by substituting
y = dp/dx, dp/dx = d
2
p/dx
2
. The resulting equation is of the rst
order, integrable in the usual way.
EXAMPLES.(1) The above equation represents the motion of a uid in a cylindrical
tube of radius r and length l. The motion is supposed to be parallel to the axis of the
tube and the length of the tube very great in comparison with its radius r. P denotes
the dierence of the pressure at the two ends of the tube. If the liquid wets the walls of
the tube, the velocity is a maximum at the axis of the tube and gradually diminishes to
zero at the walls. This means that the velocity is a function of the distance (r
1
) of the
uid from the axis of the tube. Solve the equation, remembering that is a constant
depending on the nature of the uid.
Substitute p = dv/dr,
dp/dr +p/r = P/l;
r
dp
dr
+p =
P
l
r, is pr =
P
2l
r
2
+C
1
; (5)

dv
dr
=
P
2l
r +
C
1
r
; v =
P
4l
r
2
+C
1
log r + C
2
.
To evaluate C
1
in (5), note that at the axis of the tube r = 0. This means that if
C
1
, is a nite or an innite magnitude the velocity will be innite. This is obviously
impossible, therefore, C
1
, must be zero. To evaluate C
2
, note that when r = r
1
, v
vanishes and, therefore, we get the nal solution of the given equation in the form,
v =
1
4
P

r
2
1
r
2

/l, which represents the velocity of the uid at a distance r

1
, from
the axis.
(2) Solve ad
2
y/dx
2
=

1 + (dy/dx)
2
. Make the necessary substitutions and inte-
grate.
a dp/

(1 +p
2
); becomes x/a log

p +

p
2
+ 1

+C;
or, in the exponential form,
C
1
e
x/a
p =

(1 +p
2
); and dy/dx =
1
2
C
1
e
x/a
e
x/a
/2C
1
;
by squaring. On integration
y =
1
2
aC
1
e
x/a
+
1
2
ae
x/a
/C
1
+C
2
.
(3) Some expressions can be reduced to the standard form by an obvious transfor-
mation. Thus,
d
5
y/dx
5
d
3
y/dx
3
= x.
Substitute p for d
3
y/dx
3
and dierentiate p = d
3
y/dx
3
twice. Thus,
d
2
p/dx
2
p = x,
whence y can be obtained by successive integration as indicated above.
(4) Solve d
2
V/dr
2
+ 2dV/r dr = 0. This equation occurs in the theory of
321
potential. Put dV/dr for the independent variable and divide through. On integration
log (dV/dr) + 2 log r = log C
1
.
where log C
1
is an arbitrary constant. Integrate again
dV/dr = C
1
/r, becomes V = C + 2 C
1
/r.
(5) If x d
2
y/dx
2
= 1, show that y = xlog x +C
1
x +C
2
.
134. Equations of Motion, chiey Oscillatory Motion. By New-
tons second law, if a certain mass (m) of matter is subject to a constant
force (F
0
) for a certain time; we have, in rational units,
F
0
= (Mass) (Acceleration of the particle).
If the motion of the particle is subject to friction, we must regard the
friction as a force tending to oppose the motion generated by the impressed
force. But friction is proportional to the velocity (v) of the motion of the
particle, and equal to the product of the velocity and a constant called the
coecient of friction, written, say, . Let F
1
denote the total force acting
on the particle in the direction of its motion,
F
1
= F
0
v = md
2
s/dt
2
. (1)
If there is no friction, we have, for unit mass,
F
0
= d
2
s/dt
2
. (2)
The motion of a pendulum in a medium which oers no resistance to
its motion, is that of a material particle under the inuence of a central
force (F) attracting with an intensity which is proportional to the distance
of the particle away from the centre of attraction. That is (Fig. 7),
F = q
2
s. (3)
where q
2
is to be regarded as a positive constant which tends to restore the
particle to a position of equilibrium the so-called coecient of restitution.
It is written in the form of a power to avoid a root sign later on. The
negative sign shows that the attracting force (F) tends to diminish the
distance (s) of the particle away from the centre of attraction. If s = 1, q
2
represents the magnitude of the attracting force unit distance away. From
(2),
d
2
s
dt
2
= q
2
s (4)
This is a typical equation of harmonic motion, as will be shown directly.
One solution of (4) is
s = C cos (at +) . (5)
322
This equation is the simple harmonic motion of ' 50, C denotes the am- 50
reviews
the prop-
erties of
sinusoids
as they
can be
used to
describe
periodic
sinu-
soidal
motion.
plitude of the vibration. If = 0, we have the simpler equation,
s = C cos qt. (6)
When the particle is at its greatest distance from the central attract-
ing force, qt = , ' 50, page 112. For a complete to and fro motion,
2t = T
0
= period of oscillation, hence
T
0
= 2/q. (7)
Equation (4) represents the small oscillations of a pendulum; also the
undamped* oscillations of the magnetic needle of a galvanometer.
In the sine galvanometer, the restitutional force tending to restore the needle to
a position of equilibrium, is proportional to the sine of the angle of deection of the
needle. If J denotes the moment of inertia of the magnetic needle and G the directive
force exerted by the current on the magnet, the equation of motion of the magnet, when
there is no retarding force, is
J
d
2

dt
2
= Gsin. (8)
For small angles of displacement, and sin are approximately equal. Hence,
d
2

dt
2
=
G
J
. (9)
From (4), q =

J/G, and therefore, from (9),

T
0
= 2

J/G, (10)
a well-known relation showing that the period of oscillation of a magnet in the magnetic
eld, when there is no damping action exerted on the magnet, is proportional to the
square root of the moment of inertia of the magnetic needle, and inversely proportional
to the square root of the directive force exerted by the current on the magnet. See page
524.
In a similar manner, it can be shown that the period of the small oscillations of
a pendulum suspended freely by a string of length l, is 2

l/g, where g denotes the

acceleration of gravity.
Equation (4) takes no account of the resistance to which a particle is
subjected as it moves through such resisting media as
* When an electric current passes through a galvanometer, the needle is deected and
begins to oscillate about a new position of equilibrium. In order to make the needle
come to rest quickly, so that the observations may be made quickly, some resistance
is opposed to the free oscillations of the needle either by attaching mica or aluminum
vanes to the needle so as to increase the resistance of the air, or by bringing a mass of
copper close to the oscillating needle. The currents induced in the copper by the motion
of the magnetic needle, react on the moving needle, according to Lenzs law, so as to
retard its motion. Such a galvanometer is said to be damped. When the damping is
suciently great to prevent the needle oscillating at all, the galvanometer is said to be
dead beat and the motion of the needle is aperiodic. In ballistic galvanometers, there
is very much damping.
323
air, water, etc. This resistance is proportional to the velocity, and has a
negative value. To allow for this, equation (4) must have an additional
negative term. We thus get
d
2
s
dt
2
=
ds
dt
q
2
s,
where , is the coecient of friction. For greater convenience, we may write
this 2f,
d
2
s
dt
2
+ 2f
ds
dt
+q
2
s = 0. (11)
Before proceeding further, it will perhaps make things plainer to put
the meaning of this dierential equation into words. The manipulation of
the equations so far introduced, involves little more than an application of
common algebraic principles. Dexterity in solving comes by practice. Of
even greater importance than quick manipulation is the ability to form a
clear concept of the physical process symbolised by the dierential equation.
Some of the most important laws of Nature appear in the guise of an
unassuming dierential equation. The reader should spare no pains to
acquire familiarity with the art. The late Professor Tait has said that a
mathematical formula, however brief and elegant, is merely a step towards
knowledge, and an all but useless one until we can thoroughly read its
meaning.
In equation (11), the term d
2
s/dt
2
denotes the relative change of the
velocity of the motion of the particle in unit time, ' 7; 2f ds/dt shows
that this motion is opposed by a force which tends to restore the body to
a position of rest, the greater the velocity of the motion, the greater the
retardation; q
2
s represents another force tending to bring the moving body
to rest, this force also increases directly as the distance of the body from
the position of rest. To investigate this motion further, we cannot do better
than follow Professor Perrys graphic method.
The rst thing is to solve (11) for s. This is done by the method of
' 130. Put s = e
mt
and solve the auxillary quadratic equation. We thus
obtain
m = f

f
2
q
2
. (12)
And nally,
s = e
(+)t
; or rather s = C
1
e
t
+C
2
e
t
,
where = f +

f
2
+q
2
and = f

f
2
+q
2
. The solution of (11)
thus depends on the relative magnitudes of f and q.
Suppose that we know enough about the moving system to be able to
determine the integration constants. When t = 0, let v = v
0
and s = 0.
324
Case i. The roots of the auxillary equation are real and unequal. The condition for
real roots and , in (12), is that f is greater than q (page 388). In this case,
s = C
1
e
t
+C
2
e
t
, (13)
solves equation (11). To nd what this means, let us suppose that f = 3, q = 2, t = 0,
s = 0, v = v
0
. From (12), therefore,
m = 3

9 4 = 3 2.24 = 0.76 and 5.24.

Substitute these values in (18) and dierentiate for the velocity v or ds/dt. Thus,
s = C
1
e
5.24t
+C
2
e
0.76t
; ds/dt = 5.24C
1
e
5.24t
0.76C
2
e
0.76t
.
5.24C
1
0.76C
2
= 1.
From (18), when t = 0, s = 0 and C
1
+C
2
= 0, or C
1
= +C
2
=
1
6
.
s =
1
6

e
0.76t
e
5.24t

. (14)
Assign particular values to t, and plot the corresponding values of s by means of Tables
XXI and XXII. Curve No. 1 (Fig. 114) was obtained by plotting corresponding values
of s and t obtained in this way.
Figure 114: (after Perry).
Case ii. The roots of the auxillary equation are real and equal. The condition for
real and equal roots is that f = q.
s = (C
1
+C
2
t) e
ft
. (15)
As before, let f = 2, q = 2, t = 0, s = 0, v
0
= 1. The roots of the auxillary are 2 and
2. Hence
s = (C
1
+C
2
t)e
2t
; and ds/dt = C
2
e
2t
2 (C
1
+C
2
t) e
2t
.
C
2
2C
1
= 1, C
1
= 0 and C
2
= 1; or s = te
2t
. (16)
Plot (16) in the usual manner. Curve 2 (Fig. 114) was so obtained.
It would
appear
that the
text is
refer-
ring to
when the
roots are
purely
imag-
inary
(multi-
ples of
) and
opposite
sign. KH
Case iii. The roots of the auxillary equation are real, equal and of opposite sign.
For equal roots of opposite sign, say q, we must have f = 0. Then
s = C
1
sin qt +C
2
cos qt. (17)
Let t = 0, s = 0, v
0
= 1, q = 2, f = 0. Dierentiate (17),
ds/dt = qC
1
cos qt qC
2
sin qt.
325
Hence 1 = 2C
1
1 2 C
2
0, or C
1
=
1
2
; C
2
= 0. Hence the equation,
s =
1
2
sin 2t. (18)
A graph from this equation is shown in curve 4 (Fig. 114).
Case iv. The roots of the auxillary equation are imaginary. For imaginary roots,
f +

f
2
q
2
, or, say a b, it is necessary that f < q (page 388). In this case,
s = e
at
(C
1
sin bt +C
2
cos bt) . (19)
Let the coecient of friction, f = 1, q = 2, t = 0, s = 0, v
0
= 1. The roots of the
auxillary are m = 1

1 4 = 1

3 = 1 +1.7, where =

1. Hence a = 1,
b = 1.7. Dierentiate (19),
ds/dt = ae
at
(C
1
sin bt +C
2
cos bt) +be
at
(C
1
sin bt C
2
cos bt) .
From (19), C
2
= 0 and, therefore, C
1
= 1/b = 0.57. Therefore,
s = 0.57e
t
sin 1.7t. (20)
Curve 3 (Fig. 114) was plotted from equation (20) in the usual way.
There are several interesting feature about the motions represented by
these four solutions of (11), shown graphically in Fig. 114. Curves Nos. 3
and. 4 (Cases iv. and iii.) show the conditions under which the equation
of motion (11) is periodic or vibratory. The eects of increased friction
due to the viscosity of the medium, is shown very markedly by the lessened
amplitude and increased period of curve 3. The net result is a damped
vibration, which dies away at a rate depending on the resistance of the
medium (2f v) and on the magnitude of the oscillations (q
2
s). Such is the
motion of a magnetic or galvanometer needle aected by the viscosity of
the air and the electromagnetic action of currents induced in neighbouring
masses of metal by virtue of its motion; it also represents the natural
oscillations of a pendulum swinging in a medium whose resistance varies
as ihe velocity. Curve 4 represents an undamped oscillation, curve 3 a
damped oscillation.
Curves 1 and 2 (Cases i. and ii.) represent the motion when the retard-
ing forces are so great that the vibration cannot take place. The needle,
when removed from its position of equilibrium, returns to its position of
rest after the elapse of an innite time. (What does this statement mean?
Compare with page 329.) Raymond calls this an aperiodic motion.
To show that the period of oscillation is augmented by damping. From
equation (19) we can show that
s = e
at
A sin bt. (21)
' 50. The amplitude of this vibration corresponds to that value of t for
which s has a maximum or a minimum value. These values are obtained in
the usual way, by equating the rst dierential coecient to zero, hence
e
at
(b cos bt a sin bt) = 0. (22)
326
If we now dene the angle such that bt = , or
tan = a/b, (23)
, lying between 0 and
1
2
(i.e., 90

), becomes smaller as a increases in value. We have

just seen that the imaginary roots of f

f
2
q
2
are a + b, for values of f less
than q. Let
a
2
+b
2
= q
2
. (24)
The period of oscillation of an undamped oscillation is, by (7), T
0
= 2/q, of a
damped oscillation T = 2/b.
T
2
/T
2
0
= q
2
/b
2
=

a
2
+b
2

/b
2
= 1 +a
2
/b
2
,
T/T
0
=

a
2
+b
2
/b. (25)
which expresses the relation between the periods of oscillation of a damped and of an
undamped oscillation. The period of vibration is thus augmented on damping.
It is easy to show by plotting that tan , of (28), is a periodic function such that
tan = tan ( +) = tan ( + 2) = . . .
Hence , +, + 2, . . .
satisfy the above equation. It also follows that
bt
1
, bt
2
+, bt
3
+ 2, . . .
also satisfy the equation, where t
1
, t
2
, t
3
. . . are the successive values of the time. Hence
bt
2
= bt
1
+, bt
3
= bt
1
+ 2, . . . ;
t
2
= t
1
+
1
2
T, t
3
= t
1
+T, . . .
Substitute these values in (21) and put s
1
, s
2
, s
3
, . . . for the corresponding displacements,
s
1
= Ae
at1
sin bt
1
; s
2
= Ae
at2
sin bt
2
; . . .
where the negative sign indicates that the displacement is on the negative side. Hence
s
1
/s
2
= e
a(t1t2)
= e
aT/2
;
s
2
/s
3
= s
3
/s
4
= = e
aT/2
. (26)
The amplitude thus diminishes in a constant ratio. Plotting these successive
Figure 115: Damped Oscillation
values of s and t, we
get the curve shown in
Fig. 115. This ratio
is called the damping ra-
tio, by Kohlrausch (D amp-
fungsverhaltnis). It is writ-
ten k. The natural loga-
rithm of the damping ratio,
is Gauss logarithmic decre-
ment, written L (the ordi-
nary logarithm of k, is writ-
ten L). Hence
= log k = aT log e = aT = a/b, (27)
and from (25),
T
2
T
2
0
= 1 +

2

2
; or T = T
0

1 +
1
2


2

2
+. . .

. (28)
327
Hence, if the damping is small, the period of oscillation is augmented by a small quantity
of the second order.
The following table contains six observations of the amplitudes of a sequence of
damped oscillations:
Observed
Deection
k. . L.
69
1.438 0.3633 0.1578
48
1.434 0.3604 0.1565
33.5
1.426 0.3548 0.1541
23.5
1.425 0.3542 0.1538
16.5
1.435 0.3612 0.1569
11.5
1.438 0.3633 0.1578
8
Meyer, Maxwell, etc., have calculated the viscosity of gases from the rate at which
the small oscillations of a vibrating pendulum are damped.
When the motion represented by equation (11) is subject to some pe-
riodic impressed force which prevents the oscillations dying away, the re-
sulting motion is said to be a forced vibration. The equation representing
such an oscillation is
d
2
s
dt
2
+
ds
dt
+q
2
s = f (x) . (29)
When f (x) = 0, the equation refers to the natural oscillations of a vi-
brating electrical or mechanical system. The impressed force is, therefore,
mathematically represented by the particular integral of equation (29) (see
example (3) below).
The subjoined examplesprincipally refer to systems in harmonic mo-
tion.
Modern
texts
use the
symbol,
i, for
current
instead
of C.
The
latter
is cus-
tomarily
used for
electrical
capac-
itance
rather
than
current.
KH
EXAMPLES.(1) Ohms law for a constant current is E = RC; for a variable current
of C amperes owing in a circuit with a coecient of self-induction of L henries, with
a resistance of R ohms and an electromotive force of E volts, Ohms law is represented
by the equation,
E = RC = L dC/dt, (30)
where dC/dt evidently denotes the rate of increase of current per second, L is the
equivalent of an electromotive force tending to retard the current.
(i.) When E is constant, the solution of (30) has been obtained in a preceding set
of examples,
C = E/R +Be
Rt/L
,
where B is the constant of integration. To nd B, note that when t = 0, C = 0. Hence,
C = E

1 e
Rt/L

R. (31)
The second term is the so-called extra current at make, an evanescent
328
Modern
engineer-
ing texts
use the
word,
tran-
sient
rather
than
evanes-
cent.
KH
factor due to the starting conditions. The current, therefore, tends to assume the steady
condition: C = E/R, when t is very great.
(ii.) When C is an harmonic function of the time, say,
C = C
0
sin qt; dC/dt = C
0
q cos qt,
or, compounding these harmonic motions (' 50),
E = RC
0
sinqt +LC
0
q cos qt,
where = tan
1
(Lq/R), the so-called lag* of the current behind the electromotive force,
the expression

R
2
+ L
2
q
2
is the so-called impedance.
(iii.) When E is a function of the time, say f (t),
C = Be
Rt/L
=
1
L
e
Rt/L

e
Rt/L
f (t) dt,
The evanescent term e
Rt/L
may be omitted when the current has settled down into
the steady state. (Why?)
(v.) When E is zero,
C = Be
Rt/L
.
Evaluate the integration constant B by putting C = C
0
, when t = 0.
(2) The relation between the charge (q) and the electromotive force (E) of two plates
of a condensor of capacity C connected by a wire of resistance R, is
E = R dq/dt +q/C,
provided the self-induction is zero. Solve for q. Show that when These
equa-
tions
follow
from (2)
on page
311 or
from the
alter-
native
methods
detailed
in 130.
Review
that
section if
you have
diculty
arriving
at these
solu-
tions.
KH
E = f (t) , q =
1
R
e
t/RC

e
t/RC
f (t) dt +Be
t/RC
;
E = 0, q = Q
0
e
t/RC
; (Q
0
is the charge when t = 0).
E = constant, q = CE +Be
t/RC
;
E = E
0
sin t, q = Be
t/RC
+CE
0
(sin t +RC cos t) /

1 +R
2
C
2

.
(3) The equation of motion of a pendulum subject to a resistance which varies with
the velocity and which is acted upon by a force which is a simple harmonic function of
the time, is
d
2
x
dt
2
+ 2f
dx
dt
+q
2
x = cos (qt +) .
Show that the complementary function is
x = Acos (qt +) +Bsin (qt +) .
* An alternating (periodic) current is not always in phase (or, in step) with the
impressed (electromotive) force driving the current along the circuit. If there is self-
induction in the circuit, the current lags behind the electromotive force; if there is a
condensor in the circuit, the current in the condenser is greatest when the electromotive
force is changing most rapidly from a positive to a negative value, that is to say, the
maximum current is in advance of the electromotive force, there is then said to be a
lead in the phase of the current.
329
This
para-
graph
demon-
strates
the
method
of unde-
termined
coe-
cients.
Further
examples
of this
method
can be
found in
the sup-
plement
pages
S-24
through
S-29.
To solve the equation, assume that
x = Acos (qt +) +Bsin (qt +) ,
is a solution. Substitute in the given equation,
Aq
2
+ 2fBq +n
2
A = 1; Aq
2
+ 2fBq +n
2
B = 0.
A =
n
2
q
2
(n
2
q
2
) + 4f
2
q
2
; B =
2fq
(n
2
q
2
) + 4f
2
q
2
.
Put A = Rcos . B = Rsin .
The solution of the given equation is then
x = Rcos (qt +
1
) , (32)
where R = 1/

(n
2
q
2
) + 4f
2
q
2
; tan = 2fq/

n
2
q
2

.
The forced oscillations due to the impressed periodic force, are thus determined by
(32). The complementary function gives the natural vibrations superposed upon these.
(4) If the friction in the preceding example, is zero,
d
2
x
dt
2
+n
2
x = cos (qt + ) . (33)
A particular integral is x = f cos (qt +) /

n
2
q
2

. This fails when n = q. In this

case, assume that x = Ct sin (nt +) is a particular integral. (33) is satised. provided
C = f/2n. The physical meaning of this is that when the pendulum is acted on by
a periodic force in step with the oscillations of the pendulum, the amplitude of the
forced oscillations will increase proportionally with the time, until, when the amplitude
exceeds a certain limit, equation (33) no longer represents the motion of the pendulum.
(5) When an electric current, passing through an electrolytic cell, has assumed
the steady state, show that the ionic velocity is proportional to the impressed force
(electromotive force). By Newtons law, for a moving body,
(Impressed force) = (Mass) (Acceleration) .
Friction is to be regarded as a retarding force acting in an opposite direction to the
impressed force; this frictional force is proportional to the velocity of the body.
(Impressed force less friction) = (Mass) (Acceleration) .
Express these facts in symbolic language. See (1) above. Integrate the result and
evaluate the constant for v = 0, when t = 0.
v = F

1 e
t/m

. (34)
For ionic motion, m is very small, , is very great. When t is great, show that the
exponential term vanishes, and
F v.
135. The Velocity of Simultaneous and Dependent Chemical
Reactions. While investigating the rate of decomposition of phosphine
(' 88), we had occasion to point out that the action really takes place in
two stages:
STAGE I. PH
3
= P + 3H.
STAGE II. 4P = P
4
; 2H = H
2
.
330
The former change alone determines the velocity of the whole reaction.
The physical meaning of this is that the speed of the reaction which occurs
during stage II., is immeasurably faster than the speed of the rst. Exper-
iment quite fails to reveal the complex nature of the complete reaction.*
Suppose, for example, a substance A forms an intermediate compound
B, and this, in turn, forms a nal product C. If the speed of the reaction
A = B, is one gram per
1
100000
second,
when the speed of the reaction
B = C, is one gram per hour,
the observed order of the complete reaction
A = C,
will be xed by that of the slower reaction, B = C, because the meth-
ods used for measuring the rates of chemical reactions are not sensitive
to changes so rapid as the assumed rate of transformation of A into B.
Whenever the order of this latter reaction, B = C is alone accessible to
measurement. If, therefore, A = C is of the rst, second, or nth order, we
must understand. that one of the subsidiary reactions (A = B, or B = C)
is
(1) an immeasurably fast reaction, accompanied by
(2) a slower measurable change of the rst, second or nth order, accord-
ing to the particular system under investigation.
If, however, the velocities of the two reactions are of the same order
of magnitude, the order of the complete reaction will not fall under any
simple type ('' 88, 89), and, therefore, some changes will have to be made
in the dierential equations representing the course of the reaction. Let us
study some of the simpler cases.
Case i. In a given system, a substance A forms an intermediate sub-
stance B, which nally forms a third substance C.
Let one gram molecule of the substance A be taken. At the end of a certain time t,
the system contains x of A, y of B, z of C. The rate of diminution of x is evidently

dx
dt
= k
1
x, (1)
* Professor Walker illustrates this by the following analogy (Velocity of Graded
Reactions, Proc. Royal Soc. Edin., Dec., 1897): The time occupied in the transmission
of a telegraphic message depends both on the rate of transmission along the conducting
wire and on the rate of the messenger who delivers the telegram; but it is obviously this
last, slower rate that is of really practical importance in determining the total time of
transmission. . . .
331
where k
1
denotes the velocity constant of the transformation of A to B. The rate of
formation of C is
dz
dt
= k
2
y, (2)
where k
2
is the velocity constant of the transformation of B to C. Again, the rate at
which B accumulates in the system is evidently the dierence in the rate of diminution
of x and the rate of increase of z, or
dy
dt
= k
1
x k
2
y. (3)
The speed of the chemical reactions,
A = B = C,
is fully determined by this set of dierential equations. When the relations between a
set of variables involves a set of equations of this nature, the result is said to be a system
of simultaneous dierential equations.
In a great number of physical problems, the interrelations of the variables are rep-
resented in the form of a system of such equations. The simplest class occurs when each
of the dependent variables is a function of the independent variable.
The simultaneous equations are said to be solved when each variable is expressed
in terms of the independent variable, or else when a number of equations between the
dierent variables can be obtained free from dierential coecients.
To solve the present set of dierential equations, rst dierentiate (2),
d
2
z
dt
2
k
2
dy
dt
= 0;
Add and subtract k
1
k
2
y, substitute for dy/dt from (3) and for k
2
y from (2), we thus
obtain
d
2
z
dt
2
+ (k
1
+k
2
)
dz
dt
k
1
k
2
(x +y) = 0.
But from the conditions of the experiment,
x +y +z = 1, z 1 = (x +y) .
Hence, the last equation may be written,
d
2
(z 1)
dt
2
+ (k
1
+k
2
)
d (z 1)
dt
+k
1
k
2
(z 1) = 0. (4)
This linear equation of the second order with constant coecients, is to be solved for
z 1 in the usual manner (' 130). At sight, therefore,
z 1 = C
1
e
k1t
+C
2
e
k2t
. (5)
But z = 0, when t = 0,
C
1
+C
2
= 1. (6)
Dierentiate (5). From (2), dz/dt = 0, when t = 0. Therefore, making the necessary
substitutions,
C
1
k
1
C
2
k
2
= 0. (7)
From (6) and (7),
C
1
= k
1
/ (k
2
k
1
) ; C
2
= k
2
/ (k
2
k
1
) .
The nal result may therefore be written,
z 1 =
k
2
k
1
k
2
e
k2t
+
k
1
k
2
k
1
e
k1t
. (8)
332
Harcourt and Esson have studied the rate of reduction of potassium permanganate
by oxalic acid.
2KMnO
4
+ 3MnSO
4
+ 2H
2
O = K
2
SO
4
+ 2H
2
SO
4
+ 5MnO
2
;
MnO
2
+H
2
SO
4
+H
2
C
2
O
4
= MnSO
4
+ 2H
2
O + 2CO
2
.
By a suitable arrangement of the experimental conditions this reaction may be used
to test equations (5) or (8).
Let x, y, z, respectively denote the amounts of Mn
2
O
7
, MnO
2
, and MnO (in
combination) in the system. The above workers found that C
1
= 28.5; C
2
= 2.7;
e
k1
= 0.82; e
k2
= 0.98. The following table places the above suppositions beyond
doubt.
t min-
utes.
z 1.
t min-
utes.
z 1.
Found. Calculated. Found. Calculated.
0.5 25.85 25.9 3.0 10.45 10.4
1.0 21.55 21.4 8.5 8.95 9.0
1.5 17.9 17.8 4.0 7.7 7.8
2.0 14.9 14.9 4.5 6.65 6.6
2.5 12.55 12.5 5.0 5.7 5.8
Case ii. A solution contains a gram molecules of each of A and C, the substance A
gradually changes to B, which, in turn, reacts with C to form another compound D.
Let x denote the amount of A which remains untransformed after the elapse of an
interval of time t, y the amount of B, and z the amount of C present in the system after
the elapse of the same interval of time t. Hence show that

dx
dt
= k
1
x;
dz
dt
= k
2
yz. (9)
The rate of diminution of B is proportional to the product of its active mass y into the
amount of C present in the solution at the time t, but the velocity of increase of y is
equal to the velocity of diminution of x,

dy
dt
= k
1
x k
2
yz. (10)
If x, y, z, could be measured independently, it would be sucient to solve these equations
as in case i., but if x and y are determined together, we must proceed a little dierently.
Note z = x+y. From the rst of equations (9), and (10) by addition and the substitution
of dt = dx/k
1
x from (9), and of z x = y, we get
1
z
2

dz
dt
+
K
z

K
x
= 0. (11)
where K has been written in place of k
2
/k
1
. The solution of this equation has been
previously determined (page 298) in the form
Ke
Kx

C
1
log x +Kx
1
1.2
2
(Kx)
2
+. . .

z = 1. (12)
333
In some of Harcourt and Essons experiments, C
1
= 4.68; k
1
= 0.69; k
2
= 0.006864.
From the rst of equations (9), it is easy to show that x = aek
1
t. Where does a come
from? What does it mean? Hence verify the third column in the following table:
t min-
utes.
z.
Found. Calculated.
2 51.9 51.6
8 42.4 42.9
4 85.4 85.4
5 29.8 29.7
After the lapse of six minutes, the value of x was found to be negligibly very small.
The terms succeeding log x in (12) may, therefore, be omitted without committing any
sensible error. Substitute x = ae
k1t
in the remainder,
k
2
k
1
(C
1
log a +k
1
t) t; or (C

1
+t) z =
1
k

2
where C

1
= C
1
/k
1
(log a) /k
1
. Harcourt and Esson found that C

1
= 0.1, and
1/k
2
= 157. Hence, in continuation of the preceding table, these investigators ob-
tained the results shown in the following table. The agreement between the theoretical
and experimental numbers is remarkable.
t min-
utes.
z 1.
t min-
utes.
z 1.
Found. Calculated. Found. Calculated.
6 25.7 25.7 10 15.5 15.5
7 22.2 22.1 15 10.4 10.4
8 19.4 19.4 20 7.8 7.8
9 17.8 17.8 30 5.5 5.2
The theoretical numbers are based on the assumption that the chemical change
consists in the gradual formation of a substance which at the same time slowly disappears
by reason of its reaction with a proportional quantity of another substance.
This really means that the so-called initial disturbances in chemical reactions, are
due to the fact that the speed during one stage of the reaction, is faster than during the
other. The magnitude of the initial disturbances depends on the relative magnitudes of
k
1
and k
2
. The observed velocity in the steady state depends on the dierence between
the steady diminution dx/dt and the steady rise dz/dt. If k
2
is innitely great in
comparison with k
1
, (8) reduces to
z = a

1 e
k1t

.
334
which will be immediately recognised as another way of writing the familiar equation
k
1
=
1
t
log
a
a z
.
So far as practical work is concerned, it is necessary that the solutions of the dier-
ential equations shall not be so complex as to preclude the possibility of experimental
verication.
Case iii. In a given system A combines with R to form B, B combines with R to
form C, and C combines with R to form D.
In the hydrolysis of triaeetin,
C
3
H
5
A
3
+H OH = 3A H +C
3
H
5
(OH) 3,
(Triacetin) (Glycerol)
where A has been written for CH
3
COO, there is every reason to believe that the
reaction takes place in three stages:
C
3
H
5
A
3
+H OH = A H +C
3
H
5
A
2
OH (diacetin);
C
3
H
5
A
2
OH +H OH = A H +C
3
H
5
A (OH)
2
(monacetin);
C
3
H
5
A (OH)
2
+H OH = A H +C
3
H
5
(OH)
3
(glycerol).
These reactions are interdependent. The rate of formation of glycerol is conditioned by
the rate of formation of monacetin; the rate of monacetin depends, in turn, upon the
rate of formation of diacetin. There are, therefore, three simultaneous reactions of the
second order taking place in the system.
Let a denote the initial concentration (gram molecules per unit volume) of triacetin,
b the concentration of the water; let x, y, z, denote the number of molecules of mono-,
di- and triacetin hydrolysed at the end of t minutes. The system then contains a z
molecules of triaoetin, zy, of diacetin, yx, of monacetin, and b(x +y +z) molecules
of water. The rate of hydrolysis is therefore completely determined by the equations:
dx/dt = k
1
(y x) (b x y z) ; (13)
dy/dt = k
2
(z y) (b x y z) ; (14)
dz/dt = k
3
(a z) (b x y z) ; (15)
where k
1
, k
2
, k
3
, represent the velocity coecients (' 88) of the respective reactions.
Geitel tested the assumption: k
1
= k
2
= k
3
. Hence dividing (15) by (13) and by
(14), he obtained See
worked
solution
of this
equation
on page
S-42. KH
dz/dy = (a z) / (z y) ; dz/dx = (a z) / (y x) . (16)
From the rst of these equations,
dy +y
dz
a z
=
z dz
a z
.
which can be integrated as a linear equation of the rst degree. The constant is equated
by noting that if a = 1, z = 0, y = 0. The reader might do this as an exercise on ' 123.
The answer is
y = z + (a z) log (a z) . (17)
Now substitute (17) in the second of equations (16), rearrange terms and integrate as a
further exercise on linear equations of the rst order. The nal result is,
x = z + (a z) log (a z)
a z
2
log (a z)
2
. (18)
335
Geitel then assigned arbitrary numerical values to z (say from 0.1 to
1.0), calculated the corresponding amounts of z and y from (17) and (18)
and compared the results with his experimental numbers. For experimental
and other details the original memoir must be consulted (vide infra).
EXAMPLE. Calculate equations analogous to (17) and (18) on the supposition that
k
1
= k
2
= k
3
.
A study of the dierential equations representing the mutual conversion
of red into yellow, and yellow into red phosphorus, will be found in a paper
by Lemoine in the Annales de Chimie et de Physique [4], 27, 289, 1872.
There is also a series of interesting papers by Rud. Wegscheider bearing
on this subject in Zeit. f. phys. Chem., 30, 593, 1899; ib., 34, 290, 1900;
ib., 35, 513, 1900; Monatshefte f ur Chemie, 22, 749, 1901.
The preceding discussion is based upon papers by Harcourt and Esson,
Phil. Trans., 156, 198, 1866; Geitel, Journ. f ur prakt. Chem. [2], 55,
429, 1897; J. Walker, Proc. Roy. Soc. Adman., 22, 1897. It is somewhat
surprising that Harcourt and Essons investigation has not received more
attention from the point of view of simultaneous and dependent reactions.
The indispensable dierential equations, simple as they are, might perhaps
account for this. But chemists, in reality, have more to do with this type
of reaction than any other. The day is surely past when the study of a
particular reaction is abandoned simply because it wont go according to
the stereotyped velocity equations of ' 88.
136. Simultaneous Dierential Equations. By way of practice it
will be convenient to study a few more examples of simultaneous equations.
For a complete determination of each variable there must be the same
number of equations as there are independent variables. Quite an analogous
thing occurs with the simultaneous equations of ordinary algebra.
I. Simultaneous equations with constant coecients. The methods used
for the solution of these equations are analogous to those employed for
similar equations in algebra. The operations here involved are chiey pro-
cesses of elimination and substitution, supplemented by dierentiation or
integration at various stages of the computation. The use of the symbol D
often shortens the work. Most of the following examples are from results
proved in the regular textbooks on physics.
EXAMPLES.(1) Solve dx/dt + ay = 0, dy/dt + bx = 0. Dierentiate the rst,
multiply the second by a. Subtract and y disappears. Hence writing
x = C
1
e
mt
+C
2
e
mt
; or, y = C
2

b/a e
mt
C
1

b/a e
mt
.
336
We might have obtained an equation in y, and substituted it in the second. Thus four
constants appear in the result. But one pair of these constants can be expressed in terms
of the other two. Two of the constants, therefore, are not arbitrary and independent,
while the integration constant is arbitrary and independent. It is always best to avoid
an unnecessary multiplication of constants by deducing the other variables from the rst
without integration. The number of arbitrary constants is always equal to the sum of
the highest orders of the set of dierential equations under consideration.
(2) Solve dx/dt + y = 3x; dy/dt y = x. Dierentiate the rst. Subtract each
of the given equations from the result.

D
2
4D + 4

x remains. Solve as usual.

x = (C
1
+C
2
x) e
2t
. Substitute this value of x in the second of the given equations
and y = (C
1
C
2
+C
3
t) e
2t
.
(3) The equations of rotation of a particle in a rigid plane, are
dx/dt = y; dy/dt = x.
To solve these, dierentiate the rst, multiply the second by , etc. Finally
x = C
1
cos t + C
2
sin t; y = C

1
cos t + C

2
sin t. To nd the relation between these
constants, substitute these values in the rst equation and
= C
1
sin t +C
2
cos t = C

1
cos t +C

2
sint,
or C

1
= C
2
and C

2
= C
1
.
(4) Solve d
2
x/dt
2
= n
2
x; d
2
y/dt
2
= n
2
y.
x = C
1
cos nt +C
2
sin nt; y = C

1
cos nt +C

2
sin nt.
Eliminate t so that
(C

1
x C
1
y)
2
+ (C

2
x C
2
y)
2
= (C
1
C

2
C
2
C

1
)
2
, etc.
The result represents the motion of a particle in an elliptic path, subject to a central
gravitational force.
(5) Solve dx/dt + by + cz = 0; dy/dt + a
1
x + c
1
z = 0; dz/dt + a
2
x + b
2
z = 0.
Operate on the rst with D
2
b
2
c
1
on the second with b
2
c bD, on the third with
bc
1
cD. Add. The terms in y and z disappear. The remaining equation has the
integral,
x = C
1
e
t
+C
2
e
t
+C
3
e
t
,
where , , , are the roots of
z
3
+ (a
1
b +a
2
c +b
2
c
1
) z +a
1
b
2
c +a
2
bc
1
= 0.
The values of y and z are easily obtained from that of x by proper substitutions in the
other equations.
(6) If two adjacent circuits have currents i
1
and i
2
then, according to the theory of
electromagnetic induction,
M
di
1
dt
+L
2
di
2
dt
+R
2
i
2
= E
2
; M
di
2
dt
+L
1
di
1
dt
+R
1
i
1
= E
1
,
(see J. J. Thomsons Elements of Electricity and Magnetism, p. 882), where R
1
, R
2
denote the resistances of the two circuits, L
1
, L
2
the coecients of self-induction, E
1
,
E
2
the electromotive forces of the respective circuits and M the coecient of mutual
induction. All the coecients are supposed constant.
First, solve these equations on the assumption that E
1
= E
2
= 0. Assume that
i
1
= ae
mt
and i
2
= be
mt
,
337
satisfy the given equations. Dierentiate each of these variables with respect to t and
substitute in the original equation
aMm+b (L
2
m+R
2
) = 0; bMm+a (L
1
m +R
1
) = 0.
multiply these equations so that
(L
1
L
2
M) m
2
+ (L
1
R
2
+R
1
L
2
) m +R
1
R
2
= 0.
For physical reasons, the induction L
1
L
2
must always be greater than M; The roots of
this quadratic must, therefore, be negative and real (page 388), and
i
1
= a
1
e
m1t
, or a
2
e
m2t
; i
2
= b
1
e
m1t
, or b
2
e
m2t
.
Hence, from the preceding equation,
a
1
Mm
1
+b
1
L
2
m
1
+B
1
R
2
= 0; or a
1
/b
1
= (L
2
m
1
+R
2
) /Mm
1
;
similarly a
2
/b
2
= Mm
2
/ (L
1
M
2
+R
1
).
Combining the particular solutions for i
1
and i
2
, we get
i
1
= a
1
e
m1t
+a
2
e
m2t
; i
2
= b
1
e
m1t
+b
2
e
m2t
,
the required solutions.
Second, if E
1
and E
2
, have some constant value,
i
1
= E
1
/R
1
+a
1
e
m1t
+a
2
e
m2t
; i
2
= E
2
/R
2
+b
1
e
m1t
+b
2
e
m2t
,
are the required solutions.
II. Simultaneous equations with variable coecients. The general type
of simultaneous equations of the rst order, is
P
1
dx +Q
1
dy +R
1
dz;
P
2
dx +Q
2
dy +R
2
dz, . . .

(1)
where the coecients are functions of x, y, z. These equations can often
be expressed in the form
dx
P
=
dy
Q
=
dz
R
, (2)
which is to be looked upon as a typical set of simultaneous equations of
the rst order. If one of these equations involves only two dierentials, the
equation is to be solved in the usual way, and the result used to deduce
values for the other variables, as in the rst of the subjoined examples.
When the members of a set of equations are symmetrical, the solution
can often be simplied by taking advantage of a well-known theorem* in
algebra (ratio). According to this,
* Perhaps it is best to state the proof. Let
dx/P = dy/Q = dz/R = k, say; then,
dx = Pk; dy = Qk; dz = Rk;
or, l dx lPk; mdy mQk; ndz = nRk.
Add these results,
l dx +mdy +ndz = k (lP +mQ+nR) .

l dx +mdy +ndz
lP +mQ+nR
= k =
dx
P
=
dy
Q
=
dz
R
.
338
dx
P
=
dy
Q
=
dz
R
=
l dx +mdy +ndz
lP +mQ +nR
=
l

dx + m

dy +n

dz
l

P +m

Q+n

R
= (3)
where l, m, n, l

, m

, n

,. . . are sets of multipliers such that

lP +mQ +nR = 0; l

P +m

Q+n

R = 0, etc. (4)
hence, l dx +mdy +ndz = 0, etc. (5)
The same relations between x, y, z, that satisfy (5), satisfy (2).
If (4) be an exact dierential equation, equal to say du, direct integra-
tion gives the integral of the given system, viz.,
u = a, (6)
where a, denotes the constant of integration.
In the same way, if
l dx +mdy +ndz = 0,
is an exact dierential equation, equal to say dv, then, since dv is also equal
to zero,
v = b, (7)
is a second solution. These two solutions must be independent.
EXAMPLES.(1) Solve dx/y = dy/x = dz/z. The relation between dx and dy
contains x and y only, the integral, y
2
x
2
= C
1
, follows at once. Use this result to
eliminate x from the relation between dy and dz. The result is
dz/z = dy/

y
2
C
1
; or, y +

y
2
+C
1
= C
2
z.
These two equations, involving two constants of integration, constitute a complete so-
lution.
(2) Solve dx/ (mz ny) = dy/ (nz lz) = dz/ (ly mx). l, m, n and x, y, z form
a set of multipliers satisfying the above condition. Hence,
l dx +mdy +ndz = 0; xdx +y dy +z dz = 0.
The integrals of these equations are
u = lx +my +nz = C
1
; v = x
2
+y
2
+z
2
= C
2
,
which constitute a complete solution.
(3) Solve dx/

x
2
y
2
z
2

= dy/2xy = dz/2xz. From the two last equations

y = C
1
z. Substituting x, y, z for l, m, n, each of the given ratios is equal to
(xdx +y dy +z dz) /

x
2
+y
2
+z
2

. x
2
+y
2
+z
2
= c
2
z.
is another solution.
137. Partial Dierential Equations. Equations obtained by the
dierentiation of functions of three or more variables are of two kinds:
1. Those in which there is only one independent variable, such as
P dx +Qdy +Rdz = S dt
which involves four variables three dependent and one independent. These
are called total dierential equations.
339
2. Those in which there is only one dependent and two or more inde-
pendent variables, such as,
P
z
x
+Q
z
y
+R
z
t
= 0,
where z is the dependent variable, x, y, t the independent variables. These
equations are classed under the name partial dierential equations.
The former class of equations are rare, the latter very common. We shall
conne our attention to partial dierential equations.
In the study of ordinary dierential equations, we have always assumed
that the given equation has been obtained by the elimination of constants
from the original equation. In solving, we have sought to nd this primitive
equation.* Partial dierential equations, however, may be obtained by the
elimination of arbitrary functions of the variables as well as of constants. page
56 is
included
in this
docu-
ment
following
the sup-
plement.
It can be shown from Eulers theorem (page 56) that if
u = x
n
f

y
x
,
z
x
, . . .

,
be a homogeneous function,
x
u
x
+y
u
y
+z
u
z
+ = nx
m
f

y
x
,
z
x
, . . .

= nu,
where the arbitrary function has disappeared. Again, if
u = f

ax
3
+by
3

,
is an arbitrary function of x and y.

u
x
= af

ax
3
+by
3

;
u
y
= bf

ax
3
+by
3

; b
u
x
a
u
y
= 0.
* Physically, the dierential equation represents the relation between the dependent
and the independent variables corresponding to an innitely small change in each of the
independent variables.
The reader will, perhaps, have noticed that the term independent variable is
an equivocal phrase. (1) If u = f (z), u is a quantity whose magnitude changes when
the value of a changes. The two magnitudes u and z are mutually dependent. For
convenience, we x our attention on the eect which a variation in the value of z has
upon the magnitude of u. If need be we can reverse this and write z = f (u), so that u
now becomes the independent variable. (2) If v = f (x, y), x and y are independent
variables in that x and y are mutually independent of each other. Any variation in
the magnitude of the one has no eect on the magnitude of the other. x and y are
also independent variables with respect to v in the same sense that a has just been
supposed the independent variable with respect to u.
This is usually proved in the textbooks in the following manner:
Let u = x
n
f (y/x, z/x, . . . ). Put y/x = Y , z/x = Z,. . .
Y/x = y/x
2
, Z/x = z/x
2
. . . ; Y/y = 1/x, Z/y = 0, . . .
Let v = f (Y, Z, . . . ), for the sake of brevity, therefore, since u = x
n
v,
340
EXAMPLES.(1) If y bu = f (bx ay), b
u
x
+b
u
y
= 1.
(2) If 1/z 1/x = f (1/y 1/x), x
2
z/x +y
2
z/y = z
2
.
(3) If z = a (x +y), z/x z/y = 0.
For this reason an arbitrary function of the variables is added to the
result of the integration of a partial dierential equation instead of the
constant hitherto employed for ordinary dierential equations.
If the number of arbitrary constants to be eliminated is equal to the
number of independent variables, the resulting dierential equation is of
the rst order. The higher orders occur when the number of constants to
be eliminated, exceeds that of the independent variables.
If u = f (x, y), there will be two dierential coecients of the rst order,
namely, u/x and u/y; three of the second order, namely,
2
u/x
2
,

2
u/xy,
2
u/y
2
. . .
138. What is the Solution of a Partial Dierential Equation?
Ordinary dierential equations have two classes of solutions the complete
integral and the singular solution. Particular solutions are only varieties
of complete integral. Three classes of solutions can be obtained from some
partial dierential equations, still regarding the particular solution as a
special case of the complete integral. These are indicated in the following
example.
The equation of a sphere in three dimensions is,
x
2
+y
2
+z
2
= r
2
, (1)
when the centre of the sphere coincides with the origin of the coordinate
planes and r denotes the radius of the sphere. If the centre of the sphere lies
somewhere on the xy-plane at a point (a, b), the above equation becomes
u
x
= nx
n1
f (Y, Z, . . . ) +x
n

v
Y

Y
x
+
v
Z

Z
x
+. . .

;
by the method for the dierentiation of a function of a function, 6 and 9, ' 12. Therefore,
= nx
n1
f (Y, Z, . . . ) x
n2

y
v
Y
+z
v
Z
+. . .

.
u
y
= x
n
v
dY

Y
y
= x
n1
v
Y
;
u
z
= x
n
v
Z

Z
z
= x
n1
v
Z
; . . .
Now multiply by x, y, z,. . . respectively, and add,
x
u
x
+y
u
y
+z
u
z
+ = nx
n
f (Y, Z, . . . ) = nu.
341
(x a)
2
+ (y b)
2
+z
2
= r
2
. (2)
When a and b are arbitrary constants, each or both of which may have
any assigned magnitude, equation (2) may represent two innite systems
of spheres of radius r. The centre of any member of either of these two
innite systems (called a double innite system) must lie somewhere on the
xy-plane.
Dierentiate (2) with respect to x and y.
x a +z
z
x
= 0; y b +z
z
y
= 0. (3)
Substitute for x a and y b in (2). We obtain
z
2

z
x

2
+

z
y

2
+ 1

= r
2
. (4)
Equation (2), therefore, is the complete integral of (4). By assigning any
particular numerical value to a or b, a particular solution of (4) will be
obtained, such is
(x 1)
2
+ (y 79)
2
+z
2
= r
2
. (5)
If (2) be dierentiated with respect to a and b,

(x a)
2
+ (y b)
2
+z
2
= r
2

;

b

(x a)
2
+ (y b)
2
+z
2
= r
2

.
or, x a = 0, and y b = 0.
Eliminate a and b from (2),
z = r. (6)
This result satises equation (4), but, unlike the particular solution, is not
included in the complete internal (2). Such a solution of the dierential
equation is said to be a singular solution.
Geometrically, the singular solution represents two plane surfaces touched
by all the spheres represented by equation (2). The singular solution is thus
the envelope of all the spheres represented by the complete integral. If AB
(Fig. 79) represents a cross section of the xy-plane containing spheres of
radius a, CD and EF are cross sections of the plane surfaces represented
by the singular solution.
If the one constant is some function of the other, say,
a = b,
(2) may be written
(x a)
2
+ (y a)
2
+z
2
= r
2
. (7)
342
Dierentiate with respect to a. We nd
a =
1
2
(x +y) .
Eliminate a from (7). The resulting equation
x
2
+y
2
+ 2z
2
2xy = r
2
.
is called a general integral of the equation.
Geometrically, the general integral is the equation to the tubular en-
velope of a family of spheres of radius r and whose centres are along the
line x = y. This line corresponds with the axis of the tube envelope. The
general integral satises (4) and is also contained in the complete integral.
Instead of taking a = b as the particular form of the function connecting
a and b, we could have taken any other relation, say a =
1
2
. The enve-
lope of the general integral would then be like a tube surrounding all the
spheres of radius r whose centres were along the line x =
1
2
y. Had we put
a
2
b
2
= 1, the envelope would have been a tube whose axis was an hy-
perbola x
2
y
2
= 1.
A particular solution is one particular surface selected from the double
innite series represented by the complete solution. A general integral is the
envelope of one particular family of surfaces selected from those comprised
in the complete integral A singular solution is the complete envelope of
every surface included in the complete integral.*
Theoretically an equation is not supposed to be solved completely until
the complete integral, the general integral and the singular solution have
been indicated. In the ideal case, the complete integral is rst determined;
the singular solution obtained by the elimination of arbitrary constants as
indicated above; the general integral then determined by eliminating a and
f (a).
Practically, the complete integral is not always the direct object of
attack. It is usually sucient to deduce a number of particular solutions
to satisfy the conditions of the problem and afterwards to so combine these
solutions that the result will not only satisfy the given conditions but also
the dierential equation.
* The study of Gibbs Surfaces of Dissipated Energy, Surfaces of Dissociation,
Surfaces of Chemical Equilibrium, as well as van der Waals Surfaces, is the natural
sequence of ' 68, 126 and the present section. But to enlarge upon this subject would
now cause a greater digression than is here convenient. Airys little book, in Elementary
Treatise on Partial Dierential Equations, will repay careful study in connection with
the geometrical interpretation of the solutions of partial dierential equations.
343
Of course, the complete integral of a dierential equation applies to any
physical process represented by the dierential equation. This solution, Most
modern
texts
use the
phrase,
bound-
ary
condi-
tions
rather
than
limiting
condi-
tions.
KH
however, may be so general as to be of little practical use. To represent
any particular process, certain limitations called limiting conditions have
to be introduced. These exclude certain forms of the general solution as
impossible. See examples at the end of Chapter VIII.; also example (1)
last set ' 138, and elsewhere.
The more important varieties of partial dierential equations from the
point of view of this work are the linear equations of the second and higher
orders.
139. The Solution of Partial Dierential Equations of the First
Order. For the ingenious general methods of Lagrange, Charpit, etc., the
reader will have to consult the special textbooks, say, Forsyths A Treatise
on Dierential Equations (Macmillan & Co., 1888).
Type I. The variables do not appear directly. The general form is,
f (z/x, z/y) = 0. (I.)
The solution is
z = ax + by +C,
provided a and b satisfy the relation
f (a, b) = 0, or b = f (a) .
The complete integral is, therefore,
z = ax +yf (a) +C. (1)
EXAMPLES.(1) Solve (z/x)
2
+ (z/y)
2
= m
2
. The solution is
z = ax +by +C

,
provided a
2
+b
2
= m
2
. The solution is, therefore, ax+y

m
2
a
2
+C. For the general
integral, put C = f (a). Eliminate a between the two equations,
z = ax +y

m
2
a
2
+f (a) ; and x
ay

m
2
a
2
+f

(a) = 0.
in the usual way.
(2) Solve pq = 1. Ansr. z = ax +y/a +f (a).
NOTE.We shall sometimes write, for the sake of brevity,
z/x = p; z/y = q.
(3) Solve a (p +q) = z. Sometimes, as here, when the variables do appear in the
equation, the function of x, which occurs in the equation, may be associated with z/x,
or a function of y with z/y, by a change in the variables. We may write the given
equation ap/z + aq/z = 1. Put dz/z = dZ; dy/a = dY , dx/a = dX; hence, Z/Y +
Z/X = 1, the form required.
344
(4) Solve x
2
p
2
+y
2
q
2
= z
2
. Put X = log x, Y = log y, Z = log z. Proceed as before.
Ansr. z = Cx
a
y

1a
2
.
If it is not possib1e to remove the dependent variable z in this way, the
equation will possibly belong to the following class:
Type II. The independent variables x and y are absent. The general
form is,
f (z, z/x, z/y) = 0. (II.)
Assume as a trial solution, that
z/y = a z/x.
Let z/x be some function of z obtained from II., say y = (z). Substitute
these values in
dz = p dx +q dy.
We thus get an ordinary dierential equation which can be readily inte-
grated.
dz = (z) dz +a (z) dy.
x +ay =

dz/ (z) +C. (2)

EXAMPLES.(1) Solve p
2
z +q
2
= 4. Here,

a
2
+z

(dz/dx)
2
= 4.

a
2
+z dz/dx = 2,
x +C =

a
2
+z dz =
1
2

a
2
+z

3/2
. Ansr. 2

a
2
+z

3
= 3 (x +ay +C)
2
.
(2) Solve p (1 +q)
2
= q (z a). Ansr. 4 (bz ab 1) = (x +by +C)
2
.
If z does not appear directly in the equation, we may be able to refer
the equation to the next type.
Type III. z does not appear directly in the equation, but x and z/x
can be separated from y and z/y. The leading type is
f
1
(x, z/x) = f
2
(y, z/y) . (III.)
Assume as a trial solution, that each member is equal to an arbitrary
constant a, so that z/x and z/y can be obtained in the form,
z/x =
1
(x, a) ; z/y =
2
(y, a) .
dz = p dx +q dy.
then assumes the form
dz =
1
(x, a) dx +
2
(y, a) dy. (3)
EXAMPLES. Solve the following equations:
(1) q p = x y. Put z/x +x = z/y +y = a. Write
z/x = a x, etc.; z/y = a y.
Hence, z =
1
2
(a x)
2

1
2
(a y)
2
+C.
(2) q
2
+p
2
= x +y. Ansr. z =
2
3
(x +a)
3/2
+
2
3
(y a)
3/2
+C.
(3) q = 2yp
2
. Ansr. z = ax +a
2
y
2
+C.
345
Type IV. Analogous to Clairaults equation. The general type is
z = p x +q y +f (p, q) . (IV.)
The complete integral is See sup-
plement
page
S-49
for the
method
of nding
singular
solu-
tions to
Clairault
analog
equa-
tions.
KH
z = ax +by +f (a, b) (5)
EXAMPLES.Solve the following equations:
(1) z = px +qy +pq. Ansr. z = ax +by +ab. Singular solution z = xy.
(2) z = px +qy +r

1 +p
2
+q
2
. Ansr. z = ax +by +r

1 + a
2
+b
2
.
Singular solution, x
2
+y
2
+ z
2
= r
2
. The singular solution is, therefore, a sphere; r, of
course, is a constant.
(3) z = px +qy n
n

pq. Ansr. z = ax +by n

n

ab. Singular solution,

z = (2 n) (xy)
1/(2n)
.
140. Partial Dierential Equations of the nth Order. These are
the most important equations that occur in physical mathematics. There
are no general methods for their solution, and it is only possible to perform
the integration in special cases. The greatest advances in this direction have
been made with the linear equation. Before proceeding to this important
equation, it appears convenient to solve some simpler types.
EXAMPLES.Integrate the following equations.
(1)

2
z
xy
= a. If
z
x
= p and
z
y
= q. Integrate with regard to y and we get
p = ay +f

(x). It is very possible that f

(x) is a function of y. Integrate with respect

to x and z =

ay +f

(x) dx = axy +f
1
(x) +f
2
(y).
(2)

2
z
xy

x
y
= a. Ansr. z =
1
2
x
2
log y +f
1
(x) +f
2
(y).
(3)

2
z
xy
+
z
x
f (x) = (y). Ansr. z=

e
yf(x)

e
yf(x)
(y)dy +f
1
(x)

dx+f
2
(y).
There are many points of analogy between the partial and the ordinary
linear dierential equations. Indeed, it may almost be said that every or-
dinary dierential equation between two variables is analogous to a partial
dierential in the same class. The solution is in each case similar, but with
these dierences:
First, the arbitrary constant of integration in the solution of an ordinary
dierential equation is replaced by a function of a variable or variables.
Second, the exponential form, Ce
mx
, of the solution of the ordinary
linear dierential equation assumes the form e
mx

y
(y)
.
346
The expression, e
mx

y
(y)
, is known as the symbolic form of Taylors theorem.
Having had considerable practice in the use of the symbol of operation D for

x
, we
may now use D

to represent the operation

y
.
By Taylors theorem,
(y +mx) = (y) +mx
(y)
y
+
m
2
x
2
2!


2
(y)
y
2
+. . . ,
where x is regarded as constant.
(y +mx) =

1 +mx

y
+
m
2
x
2
2!

2
y
2
+. . .

(y) .
The term in brackets is clearly an exponential series (page 280) equivalent to e
mx

y
(y)
,
or, writing D

for

y
,
(y +mx) = e
mxD

(y)
(1)
The general form of the linear equation is,
A
0

2
z
x
2
+A
1

2
z
xy
+A
2

2
z
y
2
+A
3
z
x
+A
4
z
x
+A
5
z = A, (2)
where A
1
, A
2
,. . . , A, may be constants, or functions of x and y.
As with ordinary linear equations,
Complete Solution = Particular Integral + Complementary function.
The complementary function is obtained by solving the left-hand side
of equation (2), equated to zero. We may write (2) in symbolic form,

A
0
D
2
+A
1
DD

+A
2
D

2
+A
3
D +A
4
D

+A
5

z = 0. (3)
where D is written for

x
; D

for

y
; DD

for

2
xy
. Sometimes we under-
stand
F (D, D

) z = 0. (4)
in place of (2).
141. Linear Partial Equations with Constant Coecients.
A. Homogeneous equations. Type:
A
0

2
z
x
2
+A
1

2
z
xy
+A
2

2
z
y
2
= R, (5)
where R is a function of x.
To nd the complementary function, put R = 0, and. instead of as-
suming, as a trial solution, that y = e
mx
, as was the case with the ordinary
equation, suppose that
z = (y +mx) , (6)
347
is a trial solution. Dierentiate (6), with respect to x and p, we thus obtain,
z
x
= mf

(y +mx) ;
z
y
= f

(y +mx) ;

2
z
xy
= mf

(y +mx) ;

2
z
x
2
= m
2
f

(y +mx) ;

2
z
y
2
f

(y +mx) ;
Substitute these values in equation (5) equated to zero, and divide out the
factor f

(y +mx). The auxillary equation,

A
0
m
2
+A
1
m+A
2
= 0. (7)
remains. If m is a root of this equation, f

(y +mx) = 0, is a part of the

complementary function. If and are the roots of (7), then
z = e
xD

1
(y)
+e
xD

2
(y)
, (8)
as in (3), ' 130. From (6), therefore,
z = f
1
(y +x) +f
2
(y +x) (9)
since and are the roots of the auxillary equation (7), we can write (5)
in the form,
(D +D

) (D +D

) = 0. (10)
EXAMPLES.Solve the following equations:
(1)

2
z
x
2


2
z
y
2
= 0. Ansr. z = f
1
(x +y) + f
2
(y x).
(2)

2
z
x
2
4

2
z
xy
+ 4

2
z
y
2
= 0. Ansr. z = f
1
(y 2x) +f
2
(y 2x).
(3) 2

2
z
x
2
3

2
z
xy
2

2
z
y
2
= 0. Ansr. z = f
1
(2y x) +f
2
(y + 2x).
(4)

2
u
t
2
= a
2

2
u
x
2
. Ansr. u = f
1
(at +x) + f
2
(at x). This most important equa-
tion, sometimes called dAlemberts equation, represents the motion of vibrating strings,
the law for small oscillations of air in narrow tubes (organ pipes), etc.
We cannot say much about the undetermined functions f
1
(at +x) and f
2
(at x)
in the absence of data pertaining to some specic problem. Consider a vibrating harp
string, where no force is applied after the string has once been put in motion. Let x = l
denote the length of the string under a tension equal to the weight of a length L of
the same kind of string. In order to avoid a root sign later on, a
2
has been written in
place of gL, where g represents the constant of gravitation. Further, let u represent the
displacement of any part of the string we please, and let the ordinate of one end of the
string be zero. Then, whatever value we assign to the time t, the limiting conditions are
u = 0, when x = 0; and u = 0, when x = l.
f
1
(at) +f
2
(at) = 0; f
1
(at +l) +f
2
(at l) = 0,
are solutions of dAlemberts equation. From the former, it follows that
f
1
(at) must always be equal to f
2
(at) ;
f
1
(at +l) f
1
(at l) = 0.
348
But at may have any value we please. In order to x our ideas, suppose that at l = q,
at +l = q + 2l, where q has any value whatever.
f
1
(q + 2l) = f
1
(q) .
The physical meaning of this solution is that when q is increased or diminished by 2l, the
value of the function remains unaltered. Hence, when at is increased by 2l, or, what is
the same thing, when t is increased by 2l/a, the corresponding portions of the string will
have the same displacement. In Other words, the string performs at least one complete
vibration in the time 2l/a. Hence, we conclude that dAlemberts equation represents a
nite periodic motion, with a period of oscillation 2l/c.
EXAMPLE.Show that
1
2
f
1
(at +l) +f
1
(at l) = 0.
is a solution of dAlemberts equation, and interpret the result.
A further study of dAlemberts equation would require the introduction of Fouriers
series, Chapter VIII.
When two of the roots are equal, say = . We know that the solution
of
(Da)
2
z = 0, z = e
mx
(C
1
x +C
2
) , ' 130;
by analogy, the solution of
(DaD

)
2
z = 0; z = e
mx
xf
1
(y) +f
2
(y) ,
or, z =xf
1
(y +ax) +f
2
(y +ax) . (11)
EXAMPLES.Solve:
(1)

2
z
x
2
+ 2

2
z
xy
+

2
z
y
2
= 0. Ansr. z = xf
1
(y +x) +f
2
(y +x).
(2)

D
3
3D
2
D

+DD

2
+D

z = 0.
Ansr. z = xf
1
(y x) +f
2
(y x) +f
3
(y +x) .
The particular integral will be discussed after.
B. Non-homogeneous equations. Type:
A
0

2
z
x
2
+A
1

2
z
xy
+A
2

2
z
y
2
+A
3
z
x
+A
4
z
y
+A
5
z = 0. (12)
If the non-homogeneous equation can be separated into factors, the integral
is the sum of the integrals corresponding to each symbolic factor, so that
each factor of the form D mD

, appears in the solution as a function of

y +mx, and every factor of the form DmD

a, appears in the solution

in the form z = e
ax
f (y +mx).
EXAMPLES. (1) Solve

2
z
x
2


2
z
y
2
+
z
x
+
z
y
= 0.
Factors, (D +D

) (D D

+ 1) z = 0. Ansr. z = f
1
(y x) +e
x
f
2
(y +x).
(2) Solve

2
z
x
2


2
z
xy

z
x

z
y
= 0.
Factors, (D + 1) (D D

) z = 0. Ansr. z = e
x
f
1
(y) +f
2
(x y).
349
It is, however, not often possible to represent the solutions of these
equations in this manner. When this is so, it is customary to take the trial
solution,
x = e
x+y
, (13)
and substitute for z in the given equation (12). Then,
z
x
= z;
z
y
= z;

2
z
xy
= z;

2
z
x
2
=
2
z;

2
z
y
2
=
2
z.
Equate the resulting auxillary equation to zero. We thus obtain

A
0

2
+A
1
+A
2

2
+A
3
+A
4
+A
5

z = 0. (14)
This may be looked upon as containing a bracketed quadratic in and
. For any value of , we can nd the corresponding value of , or the
value of , for any assigned value of . There is thus an innite number of
particular solutions of this dierential equation.
If u
1
, u
2
, u
3
,. . . , use particular solutions of any partial dierential equa-
tion, each solution can be multiplied by an arbitrary constant and the re-
sulting products are also solution of the equation.
Similarly, it is not dicult to see that the sum of any member of par-
ticular solutions will also be a solution of the given equation.
It is usually not very dicult to nd particular solutions, even when
the general solution cannot be obtained. The chief diculty lies in the
combining of the particular solutions in such a way, that the conditions of
the problem under investigation are satised. Plenty of illustrations will
be found at the end of the next chapter.
If the above quadratic is solved for in terms of , and if the resulting
f (, ), is homogeneous, we shall have the roots in the form,
= m
1
; = m
2
; , . . . , = m
n
.
The equation will, therefore, be satised by any expression of the form, Where m
and C
are not
sets of
discrete
numbers,
this sum
becomes
an
integral
taken dm
where C
is any in-
tegrable
function
of m.
KH
z = Ce
(y+mx)
, (15)
where m has any value m
1
, m
2
,. . . and C may have any value C
1
,
C
2
,. . . The symbol indicates the sum of the innite series, obtained
by giving m and C all possible values.
The above solution (15), may be put in a simpler form when is a
linear function of , say, = a+b. This applies to equation (12). Again,
we can sometimes solve the equation z = e
x+y
= 0, for , in terms of .
In order to x these ideas, let us proceed to the following examples.
350
EXAMPLES.(1) Solve

D
2
D

z = 0. Hence
2
= 0. Put =
1
2
, = 1,
= 2,. . . and we get the particular solutions
e
1
4
(2x+y)
, e
x+y
, e
2x+4y
, . . .
Now the dierence between any two terms of the form e
x+y
, is included in the above
solution, it follows, therefore, that the rst dierential coecient of e
x+y
, is also an
integral, and, in the same way, the second, third and higher derivatives must be integrals.
Thus we have the following particular solutions:
De
x+
2
y
= (x + 2y) e
x+
2
y
.
D
2
e
x+
2
y
= (x + 2y) + 2y e
x+
2
y
.
D
3
e
x+
2
y
= (x + 2y) + 6y (x + 2y) , etc.
If = 0, we get the special case,
z = C
1
x +C
2

x
2
+ 2y

+C
3

x
3
+ 6xy

+. . .
(2) Solve

2
z
x
2


2
z
y
2
3
z
x
+ 3
z
y
= 0. Put z = Ce
x+y
. Note
( ) ( + + 3) = 0. = and = 3 .
Hence z = C
1
e
(x+y)
+e
3y
C
1
e
(xy)
; z = f
1
(y +x) +e
3y
f
2
(y x).
(3) Solve

2
z
xy
+a
z
x
+b
z
y
+abz = 0. Ansr. z = e
ay
f
1
(x) +e
bx
f
2
(y).
(4) Solve

D
2
D
2
+D + 3D

z = 0. Ansr. z = e
x
f
1
(y x) +e
2x
f
2
(y +x).
142. The Particular Integral of Linear Partial Equations. The
following methods for nding the particular integral of homogeneous or
non-homogeneous equations, are deduced by processes analogous to those
employed for the particular integrals of the ordinary equations.
The complementary function of the ordinary linear equation
(Dm) z = 0, is, Ce
mx
;
so, for the partial equation
(D mD

) z = 0, we have e
mxD

(y)
, or the equivalent (y +mx).
This analogy extends to the particular integrals. In the former case,
the particular integral of
(D m) z = R, is, z = (D m)
1
R;
while for F (D, D

) z = R, we have z = F (D, D

)
1
R.
Case 1 (General). When F (D, D) can be resolved into factors, so that,
z = (D mD

) f (x, y) . (16)
It is now necessary to nd a value for this symbol (16). First show that
De
x
R = (D +) R,
351
by putting mD in place of , and f (x, y), in place of R (Case 4, ' 131).

1
D mD

f (x, y) =
1
D mD

e
mxD

e
mxD

f (x, y) ;
= e
mxD
1
D
f (x, y mx) . (17)
The value sought. The particular integral may, therefore, be found by the
following series of operations:
(1) Subtract mr from p in the f (x, y), to be operated upon. (2) Inte-
grate the result with respect to dx. (3) Add mx to y, after the integration.
If there is a succession of factors, the rule is to be applied to each one
seriatim, beginning on the right.
EXAMPLES.Find particular integrals in the following examples. It is well to be
careful about the signs of the dierent terms added and subtracted. It is particularly
easy to err by want of attention to this.
(1)

2
z
x
2
a
2
2
z
y
2
= xy. The particular integral is
1
DaD

1
D+aD

xy. Now xy becomes

x(y ax). This, on integration with respect to dx, becomes
1
2
x
2
y
1
3
ax
3
, and nally
1
2
x
2
y
1
2
ax
3

1
3ax
3
. Hence,
1
DaD

1
D +aD

xy =
1
DaD

x
2
y
2
+
ax
3
6
.
Subtract ax from y, for
1
2
x
2
(y +ax) +
1
6
ax
3
. Integrate and add ax to the result.
1
6
x
3
y remains. This is the required result.
(2)

D
2
+ 3DD

+ 2D
2

1
x +y. Ansr.
1
2
x
2
y
1
3
x
3
.
Case 2 (Special). When R has the form f (ax +by). Multiply F (D, D

) z
by D
n
and get F (D, D

)
must be
homoge-
neous in
powers
in order
that
(D

/D)
will be a
function
of a
single
variable.
KH
D
m
(D

/D) z = f (ax +by) .

Operate on ax +by with D

and D respectively,
D

D
f (ax +by) =
b
a
.
As on page 313, the particular integral is
z =
1
D
n

1
(D

/D)
f (ax +by) .
=
1
(b/a)

. . .

f (ax +by) dx
n
.
How to use this formula will appear from the examples.
Number
of inte-
grations
is equal
to the
highest
power
of D in
F (D, D

).
KH EXAMPLES.Find particular integrals in, (1)

D
2
+DD

2D
2

z = sin (x + 2y).
The particular integral is
1
D
2
(1 +D

/D2D
2
/D
2
)
sin(x + 2y) =
1
1 + 2 8

sin (x + 2y) dx
2
;
=
1
5
sin(x + 2y) .
(2)

D
2
+ 5DD

+ 6D
2

z = 1/ (y + 2x). Ansr.
1
20
(y + 2x) log (y + 2x) y 2x.
352
The above process cannot be employed when F (D, D

), or F (a, b) has
the same form as R, because a vanishing factor then appears in the result.
In such a case, use the above method for all factors which do not vanish
when a is put for D, b for D. The solution is then completed by means of
the formula:
1
D mD

f (y +mx) = xf (y + mx) . (18)

For a
fully
worked
example
of this
type of
equation,
see the
top of
S-53 of
the sup-
plement.
KH
EXAMPLE.Evaluate the particular integral in
(D D

) (D + 2D

) z = x +y.
For the rst factor, use the above method and then
=
1
D D

1
3D
(x +y) =
1
3D
e
xD

e
xD

(x +y) ;
=
1
3D
e
xD
1
D
y =
1
3D
e
xD

xy =
1
9
x
3
+
1
6
x
2
y.
Case 3 (Special). When R has the form of sin (ax +by), or cos (ax +by).
Proceed. as on page 313, when
1
F (D
2
, DD

, D
2
)
sin (ax +by) =
1
(a
2
, ab, b
2
)
sin (ax +by) , (19)
and in the same way for the cosine.
EXAMPLES.Find the particular integrals
(1)

D
2
+DD

+D

z = sin (x + 2y).

1
D
2
+DD

+D

1
sin (x + 2y) =
1
1 2 +D

1
sin (x + 2y)
=
D

+ 4
D
2
16
sin (x + 2y) =
1
20
(D

+ 4) sin(x + 2y) =
1
10
cos (x + 2y) + 2 sin(x + 2y) .
(2) (D +DD

2D

) z = sin (x y) + sin(x +y). Find the particular integral for

This
example
is more
easily
solved
using the
method
of unde-
termined
coe-
cients.
See sup-
plement
S-53. KH
sin (x y), then for sin (x +y). Add the two results together.
Ansr.
1
2
sin (x y) +
1
3
cos (x +y).
For the anomalous case proceed as in ' 131.
Case 4 (Special). When R has the form e
ax+by
, proceed as directed on
page 312,
1
F (D, D

)
e
ax+by
=
1
F (a, b)
e
ax+by
, (20)
that is to say, put a for D and b for D

.
EXAMPLES.Find particular integrals in the following:
(1)

D
2
DD

2D
2
+ 2D + 2D

z = e
2x+3y
.
Ansr.

D
2
DD

2D
2
+ 2D + 2D

1
e
2x+3y
=
1
10
e
2x+3y
.
(2) (DD

+aD +bD

+ab) z = e
mx+ny
. Ansr. e
mx+ny
/ (m+a) (n +b).
If F (a, b) = 0, proceed as on page 313,
z =
1
F

a
(a, b)
e
ax+by
; or, z =
1
F

b
(a, b)
e
ax+by
, (21)
353
where F

a
or F

b
, denotes the rst dierential coecient with respect to the
subscript. The two results agree with each other.
EXAMPLE.Solve

D
2
D
2
3D + 3D

z = e
x+2y
.
Ansr. f
1
(x +y) +e
3y
f
2
(y x) ye
x+2y
.
Case 5 (Special). When R has the form x
r
y
s
, where r and s are positive
integers. Operate with F (D, D

)
1
on x
r
y
s
expanded in ascending powers
of r and s.
EXAMPLES.(1) Find the particular integral in:

D
2
+DD

+D1

z = x
2
y.
=

D
2
+DD

+D

1
=

1+

D
2
+DD

+D

D
2
+DD

+D

2
+. . .

x
2
y.
=

1 +D
2
+DD

+D

+ 2D
2
D

x
2
y = x
2
y 2y 2x x
2
4.
The expansion is not usually carried higher than the highest power of the highest power
in f (x, y). See S-54
of the
supple-
ment for
why the
above
state-
ment
fails to
apply to
example
2. KH
(2) Evaluate

D
2
D
2
3D + 3D

1
xy. Ansr.

1
18
x
3

1
9
x
2

2
27
x
1
9
xy
1
6
x
2
y.
(3)

D
2
a
2
D
2

z = x.
=
1
D
2

1 + a
2
D
2
D
2
+. . .

x =
1
D
2
x =

xdx
2
=
1
6
x
3
.
Case 6 (Special). When R has the form e
ax+by
X, where X is a function
of x or y. Use
F (D, D

)
1
e
ax+by
X = e
ax+by
F (D +a, D +b)
1
X, (22)
derived as on page 315.
Use the
method
in case 5
on this
page to
apply the
operator
to the
poly-
nomial
after
you have
applied
case 6
to this
example.
KH
EXAMPLE.Find the particular integrals in

2
z/x
2
z/y = xe
ax+a
2
y
.
Ansr. e
ax+a
2
y
1
D
2
+ 2aDD

x = e
ax+a
2
y
1
2a

1
D

1 +
D
2a

1
x
= e
ax+a
2
y 1
4

x
2
/a x/a
2

.
143. The Linear Partial Equation with Variable Coecients.
These may sometimes be solved by transforming them into a form with
constants. E.g.,
(i.) Any term x
r
y
s
r+s
z
x
r
y
s
, may be reduced to the form with constant
coecients, by substituting u = log x, v = log y.
EXAMPLES.Solve the equations:
(1) x
2

2
z
x
2
y
2

2
z
y
2
y
z
y
+x
z
x
= 0. This reduces to

2
z
u
2


2
z
v
2
= 0.
354
Hence the solution of this equation, z =
1
(u +v) +
2
(u v), must be re-converted
into the form in x and y, thus, z = f
1
(xy) +f
2
(x/y).
(2) x
2

2
z
x
2
+ 2xy

2
z
xy
+

2
z
y
2
= 0. Ansr. z = f
1
(y/x) +xf
2
(y/x).
(3) (x +y)

2
z
xy
a
z
x
= 0. Put z = vx. Ansr. z = f
1
(y)+

(x +y)
a
f

2
(x)dx.
(ii.) The transformation may be eected by substituting = x

x
and

= y

y
, and treating the result as for constant coecients.
EXAMPLES.(1) Solve the rst two examples of the preceding set in this way.
(2) x
2

2
z
x
2
+ 2xy

2
z
xy
+y
2

2
z
y
2
nx
z
x
ny
z
y
+nz = 0.
( 1) + 2

1) n n

+n z = 0.
Ansr. z = x
n
f
1
(y/x) +xf
2
(y/x).
144. The Integration of Dierential Equations in Series. When
a function can be developed in a series of converging terms, arranged in
powers of the independent variable, an approximate value for the dependent
variable can easily be obtained. The degree of approximation attained
obviously depends on the number of terms of the series included in the
calculation. The older mathematicians considered this an underhand way of
getting at the solution but, for practical work, it is invaluable. As a matter
of fact, solutions of the more advanced problems in physical mathematics
are nearly always represented in the form of an abbreviated innite series.
Finite solutions are the exception rather than the rule.
EXAMPLES.(1) Evaluate the integral in f (x) = 0. Assume that f (x) can be
developed in a converging series of ascending powers of x, that is to say,
f (x) = a
0
+a
1
x +a
2
x
2
+a
3
x
3
+. . . (1)
By integration

f (x) dx =

a
0
+a
1
x +a
2
x
2
+. . .

dx;
=

a
0
dx +

a
1
xdx +

a
2
x
2
dx +. . . ;
= a
0
x +
1
2
a
1
x
2
+
1
3
a
2
x
3
+. . . ;
= x

a
0
+
1
2
a
1
x +
1
3
a
2
x
2
+. . .

+C. (2)
(2) It is required to nd the solution of dy/dx = y, in series. Assume that y = f (x),
has the form (1) above, and substitute in the given equation.
(a
1
a
0
) + (2a
2
a
1
) x + (3a
3
a
2
) x
2
+ = 0. (3)
355
This equation would be satised, if
a
1
= a
0
; a
2
=
1
2
a
1
=
1
2
a
0
; a
3
=
1
3
a
2
=
1
3!
a
0
;. . .
Hence,
y = a
0
(x) ,
where (x) = 1 +x +
1
2!
x
2
+
1
3!
x
3
+ = e
x
.
Put a for the arbitrary function,
y = ae
x
.
That this is a complete solution, is proved by substitution in the original equation.
Write the original equation in the form
y = v(x) ,
where v is to be determined. Hence,

dy
dx
(x) +v

(x) (x) = 0,
since (x) satises the original equation,
dy
dx
= 0, or v is constant.
For equations of higher degree, we must proceed a little dierently. For example:
(3) Solve
d
2
y
dx
2
x
dy
dx
cy = x
2
. (4)
(i.) The complementary function. As a trial solution, put y = ax
m
. The auxillary
equation is
m(m1) a
0
x
m2
(m+c) x
m
= 0. (5)
This shows that the dierence between the successive exponents of x in the assumed
series, is 2. The required series is, therefore,
y = a
0
x
m
+a
1
x
2m2
+ +a
n1
x
m+2n2
+a
n
x
m+2n
,
which is more conveniently written
y =

0
a
n
x
m+2n
. (6)
In order to completely determine this series, we must know three things about it.
Namely, the rst term, the coecients of x and the dierent powers of x that make up
the series.
Substitute (6) in (4),

0
(m+ 2n) (m+ 2n 1) a
n
x
m+2n2
(m+ 2n +c) a
n
x
m+2n
= 0, (7)
where n, has all values from zero to innity. If x is a solution of (4), the coecient of
x
m+2n2
must vanish with respect to m. Hence by equating the coecient of x
m+2n2
to zero,*
(m+ 2n) (m+ 2n 1) a
n
(m+ 2n 2 +c) a
n1
= 0. (8)
if n = 0, m = 0, or m = 1.
When n is greater than zero,
a
n
=
m+ 2n 2 +c
(m + 2n) (m + 2n 1)
. (9)
This formula allows us to calculate the relation between the successive coecients of x
by giving n all integral values 1, 2, 3,. . .
* If we take the other part of the auxillary a diverging series is obtained, useless for our
purpose.
356
First, suppose m = 0, then we can easily calculate from (9),
a
1
=
c
1 2
a
0
; a
2
=
c
3 4
a
1
=
c (c + 2)
4!
a
0
; . . .
y

= a
0

1 +c
x
2
2!
+c (c + 2)
c
4
4!
+. . . (10)
Next put m = 1, and, to prevent confusion, write b, in (9), in place of a,
b
n
=
c + 2n 1
2n(2n + 1)
b
n1
;
proceed exactly as before to nd successively b
1
, b
2
, b
3
. . .
y

= b
0

x + (c + 1)
x
3
3!
+ (c + 1) (c + 3)
x
5
5!
+. . . (11)
The complete solution of the equation, is the sum of series (10) and (11), or if ay
1
= y

,
by
2
= y

,
y = ay
1
+by
2
,
which contains the two arbitrary constants a and b.
(ii.) The particular integral. By the above procedure we obtain the complementary
function. For the particular integral, we must follow a somewhat similar method. E.g.,
equate (8) to x
2
instead of to zero. The coecient of m2, in (5), becomes The
m2 = 2
equation
comes
from
equating
the ex-
ponents.
KH
m(m1)a
0
x
m2
= x
2
;
m2 = 2 and m(m1) a
0
= 1;
m = 4; a
0
=
1
2
.
From (9) a
n
=
c + 2n + 2
2 (n + 2) (2n + 3)
a
n1
.
Substitute successive values of n = 1, 2, 3, . . . in the assumed expansion, and we obtain
(particular integral) = a
0
x
m
+ a
1
x
m+2
+a
2
x
m+4
+. . . ,
where a
0
, a
1
, a
2
, . . . and m have been determined.
(4) Solve d
2
y/dx
2
+xy = 0.
Ansr. y = a

1
1
3!
x
3
+
1 4
6!
x
6
. . .

x
2
4!
x
4
+
2 5
7!
x
7
. . .

.
The so-called Riccatis equation,
dy
dx
+by
2
= cx
n
has attracted a lot of attention in the past. Otherwise it is of no particular interest
here. It is easily reduced to a linear form of the second order. Its solution appears as a
converging series, nite under certain conditions.
Forsyth (l.c.) or Johnson (l.c.) must be consulted for fuller details. A detailed
study of the more important series employed in physical mathematics follows naturally
from this point. These are mentioned in the next section along with the titles of special
textbooks devoted to their use.
145. Harmonic Analysis. One of the most important equations in physical
mathematics, is

2
V
x
2
+

2
V
y
2
+

2
V
z
2
=
1

V
t
. (1)
It has practically the same form for problems on the conduction of heat, the motion of
uids, the diusion of salts, the vibrations of elastic solids and
357
exible strings, the theory of potential, electric currents and numberless other phenom-
ena. x, y, z are the coordinates of a point in space, t denotes the time and V may denote
temperature, concentration of a solution, electric and magnetic potential, the Newto-
nian potential due to an attracting mass, etc., is a constant. If the second member is
zero, we have Laplaces equation, if the second number is equated to 4; where is a
function of x, y, z, the result is known as Poissons equation.

2
V
x
2
+

2
V
y
2
+

2
V
z
2
= 0, is Laplaces equation.

2
V
x
2
+

2
V
y
2
+

2
V
z
2
= 4, is Poissons equation.
The rst member is written
2
V by some writers,
2
V by others. The equation is Modern
texts
almost
univer-
sally use

2
V ,
which is
spoken
as del-
squared
vee. KH
often more convenient to use in polar coordinates, viz.,

2
V =

2
V
r
2
+
1
r
2


2
V

2
+
2
r

V
r
+
cot
r
2

V

+
1
r sin
2

, (2)
where the substitutions are indicated in (11), ' 48.*
Any homogeneous algebraic function of x, y, z, which satises equation (1), is said
to be a solid spherical harmonic. These functions are chief used for nding the potential
on the surface of a sphere, due to forces which are not circularly symmetrical.
Particular solutions of (1) give rise, under. special conditions, to the so-called surface
spherical harmonics, tesseral harmomics and toroidal harmonics. The series See S-58
for devel-
opment
of the
series ex-
pansion
of J
0
.
KH
x
n
2
n
(n + 1)

1
x
2
2
2
(n + 1)
+
x
4
2
4
2! (n + 1) (n + 2)
. . .

,
is called a Cylindrical Harmonic or a Bessels function of the nth order. The symbol
J
n
(x) is used for it. The series is a particular solution of Bessels equation.
d
2
y
dx
2
+
1
x

dy
dx
+

1
n
2
x
2

y = 0.
If n = 0, the series is symbolised by J
0
(x) and called a Bessels function, of the zeroth
order. These functions are employed in physical mathematics when dealing with certain
problems connected with equation (1). Another particular solution is Most
modern
texts use
Y rather
than
K for
Bessel
functions
of the
second
kind,
with K
reserved
for mod-
ied
Bessel
func-
tions.
KH
J
0
(x) log x +
x
2
2
2

x
4
2
2
4
2

1
1
+
1
2

+. . . ,
called a Bessels function of the second kind (of the zeroth order), symbolised by K
0
(x).
Similarly, the solution of Legendres equation

1 x
2

d
2
y
dx
2
2x
dy
dx
+m(m + 1) y = 0,
is the series
1
m(m+ 1)
2!
x
2
+
m(m2) (m + 1) (m+ 3)
4!
x
4
. . .
* This transformation is described in the regular textbooks. But possibly the reader
can do it for himself.
A point is said to be circularly symmetrical, when its value is not aected by
rotating it through an angle about the axis.
358
written, for brevity, P
m
(x). This furnishes the so-called Surface Zonal Harmonics,
Legendres coecients, or Legendrians. Another particular solution,
x
(m1) (m+ 2)
3!
x
3
+
(m1) (m3) (m+ 2)
5!
x
5
. . .
written Q
m
(x), gives rise to Surface Zonal Harmonics of the second kind. Both series
are extensively employed in physical problems connected with equation (1).
The equation

x
2
b
2

x
2
c
2

d
2
y
dx
2
+x

x
2
b
2
+x
2
c
2

dy
dx

m(m+ 1)

x
2
b
2
+c
2

y = 0,
called Lames equation, has series solution which furnishes Lames functions or
Ellipsoidal Harmonics, users in special problems connected with the ubiquitous equation

2
V +
1

dV
dt
The so-called hypergeometric or Gauss series,
1 +
ab
1!c
x +
a (a + 1) b (b + 1)
2!c (c + 1)
x
2
+. . . ,
appears as a solution of certain dierential equations of the second order, say,
x(1 x)
d
2
y
dx
2
+c (a +b + 1) x
dy
dx
aby = 0,
(Gauss equation), where a, b, c, are constants.
The application of these series to particular problems constitutes that branch of
mathematics known as Harmonic Analysis.
But we are getting beyond the scope of this work; for more practical details, the
reader will have to take up some special work such as Byerlys Fouriers Series and
Spherical Harmonics. Weber and Riemanns Die Partiellen Dierential-Gleichungen der
Mathematischen Physik is the textbook for more advanced work. Gray and Mathews
have A Treatise on Bessels Functions and their Application, to Physics (Macmillan &
Co., 1895).
359
Supplemental Notes, including
Worked Solutions to Selected Examples
These notes by Karl Hahn are furnished under Creative Commons 3.0 license. See
http://creativecommons.org/licenses/by/3.0/ for details. You may redistribute
these notes freely or incorporate them into other works provided Karl Hahn is attributed
as their author and the distributed copies reference Creative Commons 3.0 license.
Page 283 ( 117. Separable Equations), example 2: Problem: solve
v
dv
dx
+

x
2
= 0.
Separating variables: v dv +
dx
x
2
= 0. Integrating:
1
2
v
2

x
= C, from which the books
answer follows immediately.
example 3: Solve

1 +x
2

dy =

ydx. Separating variables: dy/

y = dx/

1 +x
2

or
equivalently dy/

y dx/

1 +x
2

= 0. Integrating both sides yields the book answer.

example 4: Solve y x dy/dx = a (y +dy/dx). Gathering terms:
(1 a) y = (x +a) dy/dx. Multiply by dx and divide by y (x + 1):
(1 a) dx/ (x + 1) = dy/y. Integrate: (1 a) log (x + 1) + C

= log y, which is the log

of the book answer if you allow that C = e
C

.
example 5: Word problem. Let a be the constant of proportionality. Since the charge
is being dissipated, it time-derivative is negative. So we say
dE
dt
= aE. Separating
variables: dE/E = a dt. Integrating: log E = at + C, or equivalently, E = E
0
e
at
,
where E
0
= e
C
.
Page 284 ( 117. Separable Equations continued), example 7: The book appears
to be mistaken on this one. Solving the problem stated, it sets up as:
dy
dx
=
x
y
Multiplying by y dx to separate variables yields y dy = xdx or equivalently
y dy +xdx = 0. Integrating gives the equation of a circle centered at the origin with an
arbitrary radius. To get the books answer, it would have to say, What curves have slope
y/x to the x-axis? That sets up to y/dy +x/dx = 0 or equivalently log y +log x = C

,
which is the log of the books answer if you allow C = e
C

.
Supplementary problem the cooling of an object: This is a common problem
assigned in classes on dierential equations. An object cools at a rate that is proportional
to the dierence between its temperature and the ambient temperature. Suppose such
an object starts at 100

C. After 5 minutes its temperature is 60

C and after 10 minutes

its temperature is 50

C. What is its temperature after 20 minutes? The dierential

equation is
dT
dt
= (T
a
T) k,
where T
a
is the constant ambient temperature, and k is the rate of proportionality.
Separating variables:
dT
T T
a
= k dt,
whose solution, when integrated, is log (T T
a
) = kt +C. Taking the antilog:
T = T
a
+ T
0
e
kt
,
where T
0
= e
C
is the initial dierence between temperature of the object and ambient
temperature. Observe that there are three unknowns in the above, T
a
, T
0
, and k. The
three conditions given in the problem correspond to
S-1
100

C = T
a
+ T
0
, (a)
60

C = T
a
+ T
0
u, (b)
50

C = T
a
+ T
0
u
2
, (c)
where u = e
k5 minutes
. Subtracting (a) from (b) and (a) from (c) eliminates T
a
.
40

C = T
0
(u 1) , (d)
50

C = T
0

u
2
1

.
Taking the quotient of these two eliminates T
0
.
4
5
=
u 1
u
2
1
,
which leads to the quadratic, 4u
2
5u + 1 = 0, hence u = 1 or u =
1
4
. The former
is impossible because it leads to a zero denominator when you try to solve (d) for
T
0
. The latter yields T
0
=
160
3

C. Further back-substitution yields T

a
=
140
3

C. At
t = 20 minutes you have e
kt
= u
4
=
1
256
. Hence
T (20 minutes) =
140
3

C +
160
768

C =
375
8

C = 46.875

C.
Page 284 ( 117. Equations with homogeneous powers), example 1: The book
does steps up to integrating the substituted equation to get
t
1t
+log (1 t)+log x = C.
Back-substituting t =
y
x
,
1
1
y
x
+ log

1
y
x

+ log x = C,
x
x y
+ log (x y) + log x log x = C,
which, after you cancel the log x terms, becomes the log of the books answer.
example 2: The equation is equivalent to (y x)
dy
dx
+y = 0. Substituting y = tx and
dy
dx
= t +x
dt
dx
(tx x)

t +x
dt
dx

+tx = 0,
(t 1)

t +x
dt
dx

+t = 0,
t
2
t + (t 1) x
dt
dx
+t = 0,

(t 1) dt
t
2
+

dx
x
= C

,
log t +
1
t
+ log x = C

.
Back-substituting t = y/x:
log ylog x +
x
y
+ log x = C

,
log y = C

x
y
,
which is the log of the books answer, if you let C = e
C

.
exmple 3: Convert the equation to x
2 dy
dx
xy y
2
= 0. Making the same substitution
as in the last example:
S-2
x
2

t +x
dt
dx

x
2
t x
2
t
2
= 0,

t +x
dt
dx

t t
2
= 0,
x
dt
dx
t
2
= 0,
dt
t
2

dx
x
= 0,

1
t
log x = C

.
Back-substituting and rearranging:
log x = C

x
y
,
If you take the antilog of the above, you will see that it disagrees with the books answer.
Yet if you take the dierential of the above, you do indeed get back the original equation,
so the error seem to be in the book. The nal answer should be x = Ce
x/y
, where
C = e
C

.
example 4: Dividing by dx then making the substitution and integrating
x
2
+t
2
x
2
= 2x
2
t

t +x
dt
dx

,
1 +t
2
= 2t

t +x
dt
dx

,
1 +t
2
= 2t
2
+ 2tx
dt
dx
,
1 t
2
= 2tx
dt
dx
,
dx
x
=
2t dt
1 t
2
,
log x + log

1 t
2

= C

.
Back-substituting and simplifying
log x+log

1
y
2
x
2

= C

,
log x+log

x
2
y
2

2 log x = C

,
log

x
2
y
2

log x = C

,
x
2
y
2
x
= C,
from which the books answer follows immediately.
Page 285 ( 117. Equations that can be made homogeneous in powers),
example 2: This equation is exact, so using the method outlined in this paragraph is
very much the long road to its solution. Using the method explained in ' 119, you can
solve this one on inspection. Nevertheless we shall do it the hard way here. We assume
we have h and k such that x = v +h and y = w +k yields the equation, (2v w) dv +
(2w v) dw = 0. The sequence to the solution to this (using the substitution, w = tv)
is
S-3
(2v w) + (2w v)
dw
dv
= 0,
(2v tv) + (2tv v)

t +v
dt
dv

= 0,
2 t + (2t 1)

t +v
dt
dv

= 0,
2 t + 2t
2
t + (2t 1)

v
dt
dv

= 0,
2 2t + 2t
2
+ (2t 1)

v
dt
dv

= 0,
dv
v
+
1
2
(2t 1) dt
t
2
t + 1
= 0,
log v +
1
2
log

t
2
t + 1

= C

,
Back-substituting and simplifying
log v +
1
2
log

w
2
/v
2
w/v + 1

= C

,
1
2
log

w
2
wv +v
2

= C

,
w
2
wv +v
2
= C,
where C = e
2C

. Replacing v with x h and w with y k, this becomes our solution

equation,
(y k)
2
(y k) (x h) + (x h)
2
= C,
whose dierential is
2 (y k) dy (y k) dx (x h) dy + 2 (x h) dx = 0.
For h and k to be such that this is equivalent to the original equation,
(2y x 1) dy + (2x y + 1) dx, you must have 2h k = 1 from the dx terms and
2k h = 1 from the dy terms. Solving yields h = 1 and k = 1. Replacing h and k
with these values in the solution equation produces the books answer.
Page 285 ( 117. Equations where h and k cannot be established), second
example 1: Substituting z = 2x+3y, you also have dz/dx = 2+3dy/dx or equivalently
dy/dx =
1
3
(dz/dx 2).
(2x + 3y 5)
dy
dx
+ (2x + 3y 1) = 0,
1
3
(z 5)

dz
dx
2

+z 1 = 0,
(z 5)

dz
dx
2

+ 3z 3 = 0,
(z 5)
dz
dx
2z + 10 + 3z 3 = 0,
(z 5)
dz
dx
+z + 7 = 0,
(z 5) dz
z + 7
+dx = 0.
Since
z5
z+7
= 1
12
z+7
, integrating followed by back-substitution yields
z 12 log (z + 7) +x = C,
2x + 3y 12 log (2x + 3y + 7) +x = C,
3x + 3y 12 log (2x + 3y + 7) = C.
which, when divided by 3, is the books answer.
S-4
Page 285, second example 2: The substitution is again z = 2x + 3y, so again,
dy/dx =
1
3
(dz/dx 2).
(3y + 2x + 4) (4x + 6y + 5)
dy
dx
= 0,
(z + 4)
1
3
(2z + 5)

dz
dx
2

= 0,
3z + 12 (2z + 5)

dz
dx
2

= 0,
3z + 12 (2z + 5)
dz
dx
+ 4z + 10 = 0,
7z + 22 (2z + 5)
dz
dx
= 0,
dx
(2z + 5) dz
7z + 22
= 0.
Since
2z+5
7z+22
=
2
7

9
7
1
7z+22
, integration and back-substitution is
x
2
7
z +
9
49
log (7z + 22) = C,
7x 2 (2x + 3y) +
9
7
log (14x + 21y + 22) = C,
3x 6y +
9
7
log (14x + 21y + 22) = C.
The above disagrees with the books answer by a scalar of the terms outside the log.
Yet taking the dierential of the above does produce the original equation, so it appears
that the book is incorrect.
Page 288 ( 118. Eliminating parameters), example 1: Divide by x, then dier-
entiate twice
y
x
= a +bx,
1
x
dy
dx

y
x
2
= b,
1
x
d
2
y
dx
2

1
x
2
dy
dx

1
x
2
dy
dx
+ 2
y
x
3
= 0.
Multiply by x
3
, then gather like terms to arrive at the books answer.
example 2: Divide by x and dierentiate
y
2
x
= 4m,
2y
x
dy
dx

y
2
x
2
= 0.
Multiplying by x
2
and dividing by y yields 2x dy/dx y = 0. This disagrees with the
books answer. Yet if you substitute y =

4mx and dy/dx =

m/x into this solution,

it works. So the book appears to be in error.
example 3:
y = cos x + sinx
+y

= cos x sinx
y +y

= 0
S-5
example 4: First eliminate :
y

= ae
ax
+be
bx
ay = ae
ax
ae
bx
y

ay = (b a) e
bx
Then use that result to eliminate :
y

ay

= b (b a) e
bx
by

+aby = b (b a) e
bx
y

(a +b) y

+aby = 0
example 5: Divide by a x and dierentiate
dx
dt
1
a x
= k,
d
2
x
dt
2
1
a x
+

dx
dt

2
1
(a x)
2
= 0,
(a x)
d
2
x
dt
2
+

dx
dt

2
= 0. (a)
The general solution to the original equation is x = a +Ce
kt
. The implication is that
given arbitrary k and C, this function always satises (a). Observe that x(t) is, among
other things, the position of the end of a relaxing spring that pulls on a damper, where
the fully relaxed position is at x = a. In this application, our result indicates that the
square of the spring-ends velocity is equal to the product of its acceleration times its
distance from the fully relaxed position, regardless of the spring constant or damping
constant. Similar relationships exist for analogous systems, such as the current in an
electric circuit having a resistor and inductor in series, or the temperature of an object
cooling to ambient temperature.
Page 290 ( 119. Exact Equations) example 3: M = a
2
y +x
2
and N = b
3
+a
2
x.
Hence
M
y
= a
2
and
N
x
= a
2
.
example 4: M = siny +y cos x and N = sin x +xcos y. Hence
M
y
= cos y + cos x and
N
x
= cos x + cos y.
Page 291, example 3: M = a
2
y +x
2
and N = b
3
+a
2
x. Hence

M dx = a
2
xy +
1
3
x
3
+g (y) and

N dy = b
3
y +a
2
xy +h(x) .
Assuming the two integrals are equal, we infer g (y) = b
3
y and h(x) =
1
3
x
3
. Either way
you end up with the books answer.
example 4: M =

x
2
y
2

and N = 2xy. Hence

M dx =
1
3
x
3
xy
2
+g (y) and

N dy = xy
2
+h(x) .
Assuming the two integrals are equal, we infer g (y) = 0 and h(x) =
1
3
x
3
. This gives
the solution,
1
3
x
3
xy
2
= C, from which the books answer follows immediately.
S-6
Page 293 ( 120. Finding integrating factors, Rule I) example 2: Multiplying
this out and extracting factors of xdy and y dx,
y
2
(y dx) x
2
(2y dx) +y
2
(2xdy) x
2
(xdy) = 0.
Rearranging according to the factors outside the parentheses,
x
2
(2y dx xdy) +y
2
(y dx + 2xdy) = 0.
In this arrangement, = 2, = 0, m = 2, n = 1,

= 0,

= 2, m

= 1, and
n

= 2. Hence the equations for k and k

are
2k 1 2 = k

1 and k 1 = 2k

1 2.
Multiply the second by 2 and subtract
2k 3 = k

1
2k 2 = 4k

6
1 = 3k

+ 5
Hence k

= 2 and k = 2. Using the formulae for nding the exponents of x and y in

the integrating factor: km1 = 4 1 2 = 1 and kn 1 = 2 1 = 1. A quick
check nds that using the primed symbols you get the same exponents, as expected.
Recasting the original equation using the integrating factor, xy

xy
4
2y
2
x
3

dx +

2x
2
y
3
x
4
y

dy = 0,
which is indeed exact.

M dx =
1
2
x
2
y
4

1
2
y
2
x
4
+g (y) and

N dy =
1
2
x
2
y
4

1
2
x
4
y
2
+h(x) .
Clearly g (y) = 0 and h(x) = 0. The solution, x
2
y
4
x
4
y
2
= C, that follows is equivalent
to the books answer. Observe that in other such equations, it is not necessary for k or
k

(or m, m

, n, n

, , etc.) to be integers.
The original equation in this example,

y
3
2yx
2

dx +

2xy
2
x
3

dy = 0, is homo-
geneous in powers, so it can also be solved using the method introduced in ' 117.
Substituting y = tx and dy/dx = t +xdt/dx:

t
3
x
3
2tx
3

2t
2
x
3
x
3

t +x
dt
dx

= 0,

t
3
2t

2t
2
1

t +x
dt
dx

= 0,
t
3
2t
2t
2
1
+t +x
dt
dx
= 0,
3t
3
3t
2t
2
1
+x
dt
dx
= 0,

2t
2
1

dt
3t
3
3t
+
dx
x
= 0,

1
t
+
t
t
2
1

dt
3
+
dx
x
= 0,
log t
3
+
log

t
2
1

6
+ log x = C,
2 log
y
x
+ log

y
2
x
2
1

+ 6 log x = C,
2 log y 2 log x + log

y
2
x
2

2 log x + 6 log x = C,
y
2

y
2
x
2

x
2
= C,
which is also equivalent to the books answer.
S-7
Page 293 ( 120. Finding integrating factors, Rule II) example 1: Note that
by supposition, all equations susceptible to rule II are homogeneous in powers and
therefore can also be solved by the method introduced in ' 117. In this example (which
is equivalent to page 284, example 4), M = x
2
y +y
3
and N = 2xy
2
. Hence
Mx +Ny = x
3
y +xy
3
2xy
3
= x
3
y xy
3
.
If the reciprocal of the above is an integrating factor, as the book suggests, then
x
2
y +y
3
x
3
y xy
3
dx
2xy
2
x
3
y xy
3
dy = 0
is exact. Cancelling common factors
x
2
+y
2
x
3
xy
2
dx
2y
x
2
y
2
dy = 0.
Letting M

be the left-hand function of the above and N

be the right, then

M

y
=

x
3
xy
2

(2y)

x
2
+y
2

(2xy)
(x
3
xy
2
)
2
=
2x
3
y

2xy
3
+ 2x
3
y +

2xy
3
x
2
(x
2
y
2
)
2
=
4x
3
y
&
&
x
2
(x
2
y
2
)
2
,
N

x
= (2y)
2x
(x
2
y
2
)
2
,
proving that the equation has indeed been made exact, hence =

x
3
y xy
3

1
is an
integrating factor of the original equation. Integrating

dx = log

x
2
y
2

log x +g (y) and

dy = log

y
2
x
2

+h(x) .
So g (y) = 0, h(x) = log x, and the solution is x
2
y
2
= Cx.
Page 293, example 2: In this equation, M = x ny and N = y. So the integrating
factor is
= (Mx +Ny)
1
=
1
x
2
nxy +y
2
.
This leads to the modied equation:
M dx +N dy =
x ny
x
2
nxy +y
2
dx +
y
x
2
nxy +y
2
dy = 0.
Letting M

= M and N

= N, we have
M

y
=

x
2
nxy +y
2

(n) (x ny) (nx + 2y)

(x
2
nxy +y
2
)
2
,
=
$
$
$
nx
2
+

n
2
xy ny
2
+

nx
2
2xy

n
2
xy + 2ny
2
(x
2
nxy +y
2
)
2
.
N

x
=
2xy +ny
2
(x
2
nxy +y
2
)
2
,
Hence the modied equation is indeed exact and =

x
2
nxy +y
2

1
is an inte-
grating factor for the original equation. Showing that

dx = f (x, y) + g (y) and

N

dy = f (x, y) +h(x), with the same f (x, y) for both, is messy and dicult for arbi-
trary n, but the reader is encouraged to try it for the easy case when n = 2. Use the
substitution method in ' 117 to show that

n
2
+

x
y

n
2

1
2
= Cx
is a solution for n = 2, where
2
=
1
4
n
2
1.
S-8
Page 294 ( 120. Rule III) example: From the problem we have M = (1 +xy) y
and (1 xy) x. Then
Mx Ny =

xy +x
2
y
2

xy +x
2
y
2
= 2x
2
y
2
, hence =
1
2x
2
y
2
.
The modied equation is
(1 +xy) y
2x
2
y
2
dx +
(1 xy) x
2x
2
y
2
= 0.
Again letting M

= M and N

= N, we nd that
M

y
=
2x
3
y (1 +xy)

2x
2

4x
4
y
2
=

2x
3
y 2x
2

2x
3
y
4x
4
y
2
,
N

x
=
2xy
3
(1 xy)

2y
2

4x
2
y
4
=
$
$
$$
2xy
3
2y
2
+

2xy
3
4x
2
y
4
,
conrming that the modied equation is exact and that = 1/2x
2
y
2
is an integrating
factor.

dx =

dx
2x
2
y
+

xy
2x
2
y
dx =
1
2xy
+
1
2
log x +g (y) ,

dy =

dy
2xy
2

xy
2xy
2
dy =
1
2xy

1
2
log y +h(x) ,
Clearly g (y) =
1
2
log y and h(x) =
1
2
log x. So the solution is

1
2xy

1
2
log y +
1
2
log x = C. (a)
Multiplying by 2, then moving everything except the log x term to the right-hand side
and taking the exponential of both sides,
x = Cye
1
xy
.
This solution diers from the books answer by the sign of the exponent. Yet if you take
the dierential of (a), you can readily return to the original equation. So it appears that
the book is in error.
Page 294 ( 120. Rule IV) example 1: In this equation, M = x
2
+ y
2
and
N = 2xy. Hence
1
N

M
y

N
x

=
1
2xy
(2y (2y)) =
2
x
.
So the integrating factor is
= e
2 log x
= x
2
.
The modied equation is
M dx +N dy =

1 +
y
2
x
2

dx
2y
x
dy = 0,
which is easily veried as being exact. Integrating

M dx = x
y
2
x
+g (y) and

N dy =
y
2
x
+g (x) .
After you infer g and h, the solution is x
y
2
x
= C, from which the books answer follows
immediately.
S-9
Page 294, example 3: In this equation, M = y
4
+2y and N = xy
3
+2y
4
4x. Hence
1
M

N
x

M
y

=
y
3
4 4y
3
2
y
4
+ 2y
=
3y
3
6
y
4
+ 2y
=
3
y
.
So the integrating factor is
= e
3 log y
= y
3
.
The modied equation is
M dx +N dy =

y +
2
y
2

dx +

x + 2y
4x
y
3

dy = 0,
which is easily veried as being exact. Integrating

M dx = xy +
2x
y
2
+g (y) and

N dy = xy +y
2
+
2x
y
2
+h(x) .
After you infer g and h, the solution is xy +
2x
y
2
+y
2
= C, from which the books answer
follows when you multiply through by y
2
.
Page 297 ( 122. 1st Order Linear Equations), example 1: The book already
demonstrates the origin of the integrating factor,
=
1

1 +x
2
Applying equation (2) of ' 122 yields
y =

mdx
1 +x
2
+C =
y

1 +x
2
=

mdx
(1 +x
2
)
3
2
+C.
The remaining integral can be found by substituting tan u = x, which yields
y

1 +x
2
=
mx

1 +x
2
+C.
from which the books answer follows.
Page 297, example 2: The symbology used in this problem is likely to be very
confusing to students of electrical engineering. Virtually all modern text books use
the symbol, i, to represent electrical current rather than C. So a modern text would
represent the equation to be solved here as E = iR+L
di
dt
. I will use todays symbology
in the solution of this example.
i
R
L
+
di
dt
=
E
L
.
Hence P =
R
L
and Q =
E
L
. The integrating factor is
= e

R
L
dt
= e
Rt
L
Applying equation (2) of ' 122,
ie
Rt
L
=

E
L
e
Rt
L
dt +B,
ie
Rt
L
=
E
R
e
Rt
L
+B,
i =
E
R
+Be

Rt
L
,
which is equivalent to the books answer.
S-10
Page 297, example 4: The equation is equivalent to
dy
dx
+
y
x
= x
2
So P =
1
x
and Q = x
2
. The integrating factor is given by = e

dx/x
= x. So according
to equation (2) of ' 122
xy =

x
3
dx +C,
xy =
1
4
x
4
+C,
y =
1
4
x
3
+
C
x
,
which agrees with the books answer.
Page 298 ( 122. Bernoullis and similar equations), example 1: Letting
v = 1/y, as the book suggests, we have
dv
dt
=
1
y
2
dy
dx
.
Dividing the equation by y
2
and multiplying by 1 yields the modied equation,

1
y
2
dy
dx

1
xy
= 1.
Substituting yields
dv
dx

v
x
= 1.
This equation is rst-order linear, with P = x
1
and Q = 1. Hence the integrating
factor is = e

dx/x
= x
1
. Applying equation (2) of ' 122,
v
x
=

dx
x
+C,
v
x
= log x +C,
v = log x +Cx,
1
y
= xlog x +Cx,
1 = xy log x +Cxy,
which is equivalent to the books answer.
Page 298, example 2: The original text had xsin
2
x rather than xsin 2x as the second
term. Although the books misprint is not obvious on rst glance, it becomes obvious
when you begin following the procedure the book gives for solving it. Using the original
version of the equation in the book, you end up with a nonlinear equation when you make
the substitution an equation that is not susceptible by any of the methods covered
so far. So my rendering of the text makes the correction I have already indicated. By
trigonometric identity, the equation becomes
dy
dx
+ 2xsin (y) cos (y) = x
3
cos
2
y.
Dividing by cos
2
y, as the book suggests, and then applying more trig identities,
dy
dx

1 + tan
2
y

+ 2xtany = x
3
.
The book also suggests the substitution, v = tan y, whose derivative is
dv
dx
=

1 + tan
2
y

dy
dx
.
The equation becomes
dv
dx
+ 2xv = x
3
.
S-11
Hence P = 2x and Q = x
3
. The integrating factor is = e

2x dx
= e
x
2
. Applying
equation (2) of ' 122 yields
e
x
2
v =

x
3
e
x
2
dx +C.
To integrate the remaining integral, let u = x
2
and du = 2xdx, then integrate by parts.
e
x
2
v =
1
2

ue
u
du +C,
e
x
2
v =
1
2
(u 1) e
u
+C,
e
x
2
v =
1
2

x
2
1

e
x
2
+C,
e
x
2
tan y =
1
2

x
2
1

e
x
2
+C,
which is equivalent to the books answer. If you really needed y as a function of x, you
could rearrange this into y = tan
1

1
2

x
2
1

+Ce
x
2

.
Comments on page 298, example 3: All the steps to solution are shown in the text,
arriving at
e
Kx
y
= K

e
Kx
x
dx +C.
Handbook of Mathematical Functions, edited by Milton Abramowitz and Irene Stegun
(U.S. Department of Commerce, National Bureau of Standards, Applied Mathematics
Series 55), denes the function
E
1
(x) =

x
e
t
t
dt
in chapter 5 on the exponential integral function (by Walter Gautschi and William F.
Cahill). According to this denition, example 3 is solved by
e
Kx
y
= K E
1
(Kx) +C.
This is the rst example we have encountered so far in which the solution cannot be
expressed in closed form. In this example there is no function that can be expressed as
any nite combination of powers, roots, logs, exponentials, trig functions and inverse trig
functions whose derivative is e
Kx
/x. The E
1
(x) function can only be approximated
using power series or other convergent algorithms. In general, we say we have solved a
dierential equation if we can nd a function solving the equation whose derivative is
expressible in elementary functions, even if the solution itself cannot be.
Page 299 ( 123. Solution by Dierentiation), case i, example 2: Substituting
the symbol, p, for dy/dx, the equation becomes xyp
2

x
2
y
2

pxy = 0. This factors

into (xp +y) (yp x) = 0. This results in factored dierential equation

dy
dx
+
y
x

dy
dx

x
y

= 0.
Hence either
dy
dx
=
y
x
or
dy
dx
=
x
y
.
Both of these are separable resulting in
dy
y
=
dx
x
or y dy = xdx.
Integrating yields log y = log x +C or
1
2
y
2
=
1
2
x
2
+C. The books answer follows from
these.
Page 299, case i, example 3: Substituting p for dy/dx, the equation is p
2
7p+12 = 0.
This factors into (p 4) (p 3) = 0. Back-substitute and solve the two factors sepa-
rately, and the books answer follows immediately.
S-12
Page 299 case ii, comments: The book is overly brief on explaining this method of
solution. The steps are to isolate either y or x and then dierentiate with respect to the
other independent variable. This eliminates the isolated variable. Now replace dy/dx
or dx/dy with p and the second derivatives of y or x with the rst derivative of p. In
each of the examples, the resulting rst order dierential equation for p turns out to
be separable. Solving that equation results in a solution containing an undetermined
constant, C. Once you have that solution for p, you integrate that function to nd
the y or x function that is the solution to the original equation. This integration picks
up an additional constant of integration, K. But unlike C, this second constant is not
undetermined. In the earlier step where you took the derivative of the equation to
eliminate one of the variables, you also lost a part of the equation. The constant, K,
is the recovery of that lost part. So the nal step of this method is to solve for K. To
do this, substitute your solution function, including its added K, back into the original
equation. You will nd that not every value of K will make it work. Solving for the
value or values of K that does make it work completes your solution.
Page 299, case ii, example 1: The book shows the step of isolating y and taking the
derivative of the resulting equation with respect to x. Making the substitution of p for
dy/dx
p = 1 +
1
2
dp
dx

p
,
(p 1)dx =
1
2
dp

p
,
dx =
1
2
dp
p
3
2
p
1
2
,
which can be integrated by substituting u
2
= p and 2u du = dp. The integral is shown
in the book as x being a function of p, which then must be manipulated into p as a
function of x.
x =
1
2
log

p 1

p + 1

+ log C,
e
2x
= C

p 1

p + 1

,
e
2x
(

p + 1) = C (

p 1) ,

e
2x
C

= C e
2x
,

dy
dx
=
C +e
2x
C e
2x
,
dy
dx
=

C +e
2x

2
(C e
2x
)
2
. (a)
This function can be integrated by substituting v = e
2x
and dv/(2v) = dx, then decom-
posing the resulting integrand into partial fractions.
y =

C +e
2x

2
(C e
2x
)
2
dx =
2C
C e
2x
+x +K. (b)
We solve for K by substituting this expression for y into the original equation. To keep
things tidy, we start by letting y = y
0
+K. Then the substituted original equation is
S-13
d
dx
(y
0
+K) + 2x(y
0
+K) = x
2
+ (y
0
+K)
2
,
dy
dx
+ 2xy
0
+ 2xK = x
2
+y
2
0
+ 2Ky
0
+K
2
.
Now replace y
0
with
2C
Ce
2x
+x and dy/dx with the right-hand side of (a).

C +e
2x

2
(C e
2x
)
2
+ 2x

2C
C e
2x
+x

+ 2Kx =
x
2
+

2C
C e
2x
+x

2
+ 2K

2C
C e
2x
+x

+K
2
.
Carefully multiplying all of this out and taking the cancellations
C
2
+ 2Ce
2x
+e
4x
(C e
2x
)
2
+

4Cx
C e
2x
+

2x
2
+$
$$
2Kx =

x
2
+
4C
2
(C e
2x
)
2
+

4Cx
C e
2x
+

x
2
+
4KC
C e
2x
+$
$$
2Kx +K
2
.
Now multiply what remains by

C e
2x

2
C
2
+ 2Ce
2x
+e
4x
= 4C
2
+ 4KC

C e
2x

+K
2

C e
2x

2
,
C
2
+ 2Ce
2x
+e
4x
= 4C
2
+ 4KC
2
4KCe
2x
+K
2
C
2
2K
2
Ce
2x
+K
2
e
4x
. (c)
Clearly e
4x
= K
2
e
4x
, hence K = 1 or K = 1. Trying these values of K on the
remaining terms of (c) quickly reveals that only K = 1 works for the entire equation.
Now replace K with 1 in (b)
y =
2C
C e
2x
+x 1.
y =
2C
C e
2x
+x
C e
2x
C e
2x
,
y =
C +e
2x
C e
2x
+x,
which is the books answer.
Page 299, case ii, example 2: First substitute p for dy/dx and solve for y
xp
2
2yp +ax = 0,
x
2

p +
a
p

= y.
Taking the derivative using the product rule (remembering that p is the derivative of y)
1
2

p +
a
p

+
x
2

1
a
p
2

dp
dx
= p.
Before solving, subtract p from both sides and divide out the common factor
1
2

p +
a
p

+
x
2

1
a
p
2

dp
dx
= 0,
p
2

1 +
a
p
2

+
x
2

1
a
p
2

dp
dx
= 0,
p +x
dp
dx
= 0,
log p = log x + log C.
S-14
Hence p = Cx and
y =

p dx = C

xdx =
1
2
Cx
2
+ K.
We substitute this solution for y back into the original equation and solve for K.
x(Cx)
2
2

1
2
Cx
2
+K

(Cx) +ax = 0,
C
2
x
3
C
2
x
3
2CKx +ax = 0,
K =
a
2C
.
The resulting solution, y =
1
2

Cx
2
+a/C

, does not exactly match the books answer,

but this solution does work when substituted into the original equation, and the books
does not. Furthermore, page 303 lists a solution for this same equation that is equivalent
to the solution shown here (with the solution given on page 303, solve for y and replace
C with 1/C

).
Page 299, case ii, example 3: The book instructs to solve for x and take the derivative
with respect to y.
2x
dy
dx
= y

dy
dx

,
x =
y
2

dx
dy

dy
dx

, (d)
x =
y
2

p
1
p

,
where p is the derivative of x with respect to y. Taking the derivative with respect to y
of the above eliminates x
p =
1
2

p
1
p

+
y
2

1 +
1
p
2

dp
dy
,
1
2

p +
1
p

=
y
2

1 +
1
p
2

dp
dy
,
p = y
dp
dy
,
log p =log y +C,
p =
dx
dy
= Cy.
Integrating the last line of the above to get x as a function of y
x =

p dy = C

y dy =
1
2
Cy
2
+K.
Substituting this solution into (d)
1
2
Cy
2
+K =
y
2

Cy
1
Cy

,
K =
1
2C
.
So the solution is x =
1
2
Cy
2

1
2C
. Letting C

= 1/C
2x =
y
2
C

,
2x +C

=
y
2
C

,
C

(2x +C

) = y
2
.
which is equivalent to the books answer.
S-15
Page 300, case iii, example 2: Both this example and the next succumb readily to
the method of separation of variables (see ' 117). Indeed using the method explained
in ' 123, case iii to do these seems almost an exercise in masochism, as it involves an
order of magnitude more work. But to illustrate the method, here is example 2 done
both ways. First by separation of variables.
dy
y +
1
y
= dx,
y dy
y
2
+ 1
= dx,
1
2
log

y
2
+ 1

= x +
1
2
log C,
y
2
+ 1 = Ce
2x
.
from which the books answer follows immediately. But case iii would have you solve
for y then dierentiate. Letting p = dy/dx,
p =y +
1
y
, (a)
py =y
2
+ 1,
y
2
py + 1 = 0,
y =
1
2

p
2
4

,
dy
dx
= p =
1
2
dp
dx

p
dp
dx
2

p
2
4
.
This last equation is separable. Divide both sides by p, multiply by 2dx to get
2

dx =

dp
p

dp

p
2
1
,
2x + log C = log p log

p +

p
2
4

.
I will use the dierence-solution of the and leave it up to the reader to try the sum-
solution to conrm that it also arrives at the same nal solution.
Ce
2x
=
p
p +

p
2
4
,

p +

p
2
4

Ce
2x
= p,
Ce
2x

p
2
4 =

1 Ce
2x

p,
C
2
e
4x

p
2
4

1 2Ce
2x
+$
$
$
C
2
e
4x

p
2
,
4C
2
e
4x
=

1 2Ce
2x

p
2
, (b)
2Ce
2x

2Ce
2x
1
= p =
dy
dx
. (c)
Integrate (c) by substituting u = 2Ce
2x
and
du
2u
= dx to get

2Ce
2x
1 + K = y.
Letting K = 0 (which you can conrm is correct by substituting y =

2Ce
2x
1 + K
into the original equation), squaring both sides, and replacing 2C with C

yields an
answer equivalent to the books. You could also have substituted (a) into (b), solved
for y, and obtained the same result, but doing that algebra is more dicult than taking
integral of (c).
S-16
Page 300 ( Clairauts equation), example 1: The equation given is already in
Clairaut form. With p used to represent dy/dx, take the derivative of both sides:

p =

p +x
dp
dx
+ 2p
dp
dx
,
which factors into
(x + 2p)
dx
dp
= 0.
Either x = 2p or dp/dx = 0. If the former, then substitute p = x/2 into the original
equation.
y =
x
2
2
+
x
2
4
=
x
2
4
,
from which one of the books answers follows. From dp/dx = 0 we have p = C and
y = Cx + K. Substituting that into the original equation yields K = C
2
. Hence the
other solution is y = Cx +C
2
.
Page 300 ( 124), example 2: To put this equation into Clairaut form, divide by
p 1 and then add px to both sides.
y = px +
p
p 1
. (a)
Taking the derivative

p =

p + x
dp
dx

1
(p 1)
2
dp
dx
,
0 =

x
1
(p 1)
2

dp
dx
.
If the rst factor is zero, then
1

x
+ 1 = p.
Substitute that into (a)
y =

x
+ 1

x +
1

x
+ 1
1

x
=

x +x + 1 +

x = x + 2

x + 1 =

x + 1

2
.
Hence

y = (

x + 1). Regardless of which sign you choose from the , it still diers
from the books answer by a sign. The solution, y = x+2

x+1 works when substituted

into the original equation. The books answer does not. The solution that results from
the factor, dy/dx = 0, is again y = Cx+K. Substituting that into (a) yields K =
C
C1
,
from which the books other solution follows.
Page 302 ( 125. Singular solutions), Comments on p-discriminants and C-
discriminants: My opinion is that Mellor does not expand enough on the relationship
between p-discriminants, C-discriminant, and singular solutions. For one thing, he sug-
gests that these discriminants exist only when the dierential equation is a quadratic in
p or when the solution is a quadratic in C. Chapter A2, section 10 of Ordinary Dieren-
tial Equations and their Solutions by George M. Murphy (van Nostrand, 1960) provides
a much broader method for nding discriminants. A p-discriminant occurs whenever x
and y in the dierential equation can conspire to cause p to have a double root. You nd
such double roots in the general case by taking the partial derivative of the equation
with respect to p, solving the result for p, and substituting that expression for p back
into the original equation. Mellor uses the example, xp
2
yp + a = 0, and, using the
quadratic formula, arrives at the p-discriminant of y
2
= 4ax.
S-17
Solving it Murphys way, we have

p
(xp
2
yp +a) = 0,
2xp y = 0,
p =
y
2x
.
Substituting p into the original equation,
x

y
2x

y
2
2x
+a = 0,

y
2
4x
+a = 0,
y
2
= 4ax.
Observe that Murphys method exploits the fact that a function has a double root
only when both the function itself and its rst derivative are simultaneously zero. The
same method will also nd the C-discriminant when applied to this equations general
solution, xC
2
yC +a = 0.
The existence of a p-discriminant or a C-discriminant is a necessary but not su-
cient condition for the existence of a singular solution. Both p-discriminants and C-
discriminants of more complicated equations (an equation that is cubic in p for example)
may be factorable, in which case there is a possibility for multiple singular solutions
that is each factor may be a singular solution. Murphy states without proof that if a
factor of a p-discriminant is a singular solution, then there will be an identical factor
of the C-discriminant, and vice versa. Mellor seems to suggest the same, but never
states it outright. Both books are clear, however, that when a discriminant (or factor
of a discriminant) is found, it must be tested by substituting it back into the original
equation to see whether or not it is indeed a singular solution.
Graph showing lines, y = Cx+C
2
,
enveloping the parabola, y =
1
4
x
2
Envelopment of singular solutions: As an
example, the equation, px y + p
2
= 0, has as
its general solution, y = Cx + C
2
. This solution
has a C-discriminant of y =
1
4
x
2
. The gure
shows how the family of curves specied by the
general solution forms an envelope for the par-
ticular solution (general solution is shown with
C = 0.2, 0.4, . . . 2.0). To nd an envelope
point, we nd the general solution for C and also
for C + C, then we nd where those two lines
intersect:
y = (C +C) x + (C +C)
2
y = Cx C
2
0 = Cx + 2C C +C
2
Dividing out C from the dierence gives
0 = x + 2C + C. Taking the limit as C goes to zero gives x = 2C. Substituting
for x into the general solution gives y = C
2
. So any point of the form,

2C, C
2

,
is a point on the envelope. Observe that the relationship between the expressions for x
and y is precisely y =
1
4
x
2
, that is, the singular solution to the original equation. You
could also have solved the equation, x = 2C for C and substituted into the general
solution for C to obtain the same result. Observe also that solving for the intersection
of neighboring lines in the family as we have done here is no dierent from Murphys
method for nding the C-discriminant. By subtracting the neighbors from each other,
S-18
dividing out C, then taking the limit as C goes to zero, we have eectively taken the
partial derivative of the general solution with respect to C. After that we substituted
that into the general solution, which is the same as the last step of Murphys method.
This shows, at least for C-discriminants, how nding the conditions for Cs double root
is identical to nding points on the curve that the general solution envelopes.
Page 303 (Singular solutions), example 2: You arrive at the general solution by
solving the dierential equation for p, and integrating both sides. Use the substitution,
u = x
2
and du = 2xdx to integrate. Apply the constant of integration, C, to the
y-side of the result, then square both sides. The book provides details for nding the
p-discriminant and C-discriminant using the equal quadratic roots method. Applying
Murphys method to nding the p-discriminant, you have 4xp
2
(3x a)
2
= 0. Taking
the partial with respect to p gives 8xp = 0. Hence either x = 0 or p = 0. Note that
this p-discriminant is factorable. Substituting that into the original equation yields
(3x a)
2
= 0. So p-discriminants are x = 0 and x =
1
3
a. Using Murphys method
to nd the C discriminant, we take the partial of the general solution with respect to
C and get 2(x + C) = 0, hence C = x. Substituting into the general solution gives
0 = x(x a)
2
. Again the discriminant is factorable. x = 0 is the only common factor
of both the p-discriminant and C-discriminant. It satises the original equation in that
if you divide the original equation by p
2
(3x a)
2
and allow 1/p = dx/dy = 0 for the
line dened by x = 0.
Page 303 (Singular solutions), example 4: The general solution emerges as follows:
y
2

p
2
+ 1

= a
2
,
y =
a

p
2
+ 1
.
Take the derivative of both sides to eliminate y:
p =
pa
(p
2
+ 1)
3
2
dp
dx
,
dx =
a
(p
2
+ 1)
3
2
dp,
integrating yields x C =
ap

p
2
+ 1
,
a
2
p
2
= (x C)
2

p
2
+ 1

,
0 =

u
2
a
2

p
2
+u
2
,
where u = x C and du = dx. Hence
p
2
=
u
2
u
2
a
2
,
p =
u

a
2
u
2
.
Integrating to establish y,
y =

a
2
u
2
+K =

a
2
(x C)
2
+K.
Substituting back into the original equation yields K = 0. Hence y
2
= a
2
(x C)
2
,
from which the books general solution follows.
Rearrange the original equation into y
2
p
2
+ y
2
a
2
= 0. Using the quadratic method,
the p-discriminant is y
2
(y
2
a
2
) = 0. This discriminant factors into y = 0 or y = a.
Rearrange the general solution into C
2
2xC +x
2
+y
2
a
2
= 0. Using the quadratic
S-19
method, the C-discriminant is 4x
2
= 4x
2
+4y
2
4a
2
, or equivalently, y = a, which is
in common with a p-discriminant factor. When y = a, then p = dy/dx = 0. So this
factor works as a singular solution.
Page 304 ( 126. Trajectories), example 2: The transformation between Cartesian
and polar coordinates is x = r cos and y = r sin . Taking the derivatives of these with
respect to r,
dx
dr
= cos sin

r
d
dr

and
dy
dr
= sin + cos

r
d
dr

.
By the chain rule we have
dy
dx
=

dy
dr

dx
dr
=
sin + cos

r
d
dr

cos sin

r
d
dr
. (a)
In Cartesian coordinates we nd orthogonal trajectories by replacing dy/dx with dx/dy.
Hence

dx
dy
=
cos + sin

r
d
dr

sin + cos

r
d
dr
. (b)
Replacing r d/dr with dr/ (r d), the right-hand side of (a) becomes
sin cos

1
r
dr
d

cos + sin

1
r
dr
d
.
Graph showing orthogonal trajec-
tories to y
2
= 4ax
Multiplying top and bottom by r d/dr yields the
same expression as we got in (b) for dx/dy, which
completes the proof.
Page 304 ( 126. Trajectories), example 3:
The curve family is dened by y
2
= 4ax. Solving
this for a gives
y
2
4x
= a.
Taking the derivative using the quotient rule,
8xy
dy
dx
4y
2
16x
2
= 0.
Replacing dy/dx with dx/dy and multiplying
through by 4x
2
,
2xy
dx
dy
y
2
= 0.
An equivalent dierential equation is
2xdx + y dy = 0. This is immediately integrable
to x
2
+
1
2
y
2
= C
2
, which is equivalent the books
answer.
Page 304 ( 126. Trajectories), example 4:
Place two point charges of q and q on the hor-
izontal axis symmetrically about the origin at a
distance of L apart, with the positive charge on
the right. Allowing Lq to remain constant, we let L go to zero and q go to innity to
create an idealized dipole. Then the potential function, V , in polar coordinates is given
by
V =
kLq cos
r
2
,
where k is Coulombs constant. You could take the derivative with respect to r of this
function and apply the transformation indicated in example 3 above, and you would
arrive at the correct trajectory. You are encouraged to try that. You would nd that,
in the end, you would have to take the exponential of a function expressed in logs. So
S-20
here I will pass into logs before taking the derivative and solve it that way. The above
equation becomes
log V = log (kLq) + log (cos ) 2 log r.
Taking the derivative with respect to r,
0 = tan
d
dr

2
r
,
0 = tan

r
d
dr

2,
Equipotentials and trajectories of
an ideal dipole.
Now make the transformation from example 3:
0 = tan

1
r
dr
d

2,
0 =
dr
r
2 cot d,
0 = 2 cot d
dr
r
.
Integrating the above,
log C = 2 log (sin) log r,
C =
sin
2

r
.
Page 307 ( 128. Linear equations, when a particular solution is known),
derivation of order reduction: We begin with the assumption that v is a known
solution to the equation,
d
2
y
dx
2
+P
dy
dx
+Qy = 0.
where P and Q are functions of x. We let y = uv. Then, by the product rule,
dy
dx
= u
dv
dx
+v
du
dx
and
d
2
y
dx
2
= u
d
2
v
dx
2
+ 2
du
dx
dv
dx
+v
d
2
u
dx
2
.
Substituting
u
d
2
v
dx
2
+ 2
du
dx
dv
dx
+v
d
2
u
dx
2
+P

u
dv
dx
+v
du
dx

+Quv = 0. (a)
But because v is assumed to solve the original equation,
d
2
v
dx
2
+P
dv
dx
+Qv = 0 = u
d
2
v
dx
2
+Pu
dv
dx
+Quv. (b)
Removing from (a) the terms that occur to the right of the second equal in (b), what
remains is
v
d
2
u
dx
2
+

2
dv
dx
+Pv

du
dx
= 0.
Page 307 ( 128. Linear equations), example 1: y = e
ax
is assumed to be a
solution to
d
2
y
dx
2
= a
2
y, or equivalently,
d
2
y
dx
2
a
2
y = 0.
Hence P = 0, Q = a
2
and v = e
ax
. Substituting these into the formula on page 307,
log
du
dx
+ 2 log (e
ax
) = C,
du
dx
= Ce
2ax
,
S-21
u =
C
2a
e
2ax
+C
2
,
y = uv =
C
2a
e
ax
+C
2
e
ax
.
Letting C
1
=
C
2a
yields the books answer.
Page 307 ( 128. Linear equations), example 2: y = x is assumed to be a solution
to

1 x
2

d
2
y
dx
2
x
dy
dx
+y = 0, or equivalently,
d
2
y
dx
2

x
1 x
2
dy
dx
+
y
1 x
2
= 0.
Hence P =
x
1x
2
, Q =
1
1x
2
, and v = x. We also have

P dx =
1
2
log

1 x
2

. Using
the formula on page 307,
log
du
dx
+ 2 log x +
1
2
log

1 x
2

= log C,
du
dx
=
C
x
2

1 x
2
,
u =
C

1 x
2
x
+C
2
.
Because v = x, we have y = uv = ux = C

1 x
2
+ C
2
x. Letting C
1
= C gives
books answer.
Page 308 ( 129. Linear equations with constant coecients), example 2:
The auxillary equation is D
2
m
2
= (D +m) (D m) = 0, which has roots at m.
The books answer follows.
Page 308 ( 129. Linear equations with constant coecients), example 3:
The auxillary equation is D
2
+ 4D + 3 = (D + 3) (D + 1) = 0, which has roots at 3
and 1. The books answer is missing minus signs in the exponents.
Page 308 ( 129. Linear equations with constant coecients), case 2 com-
ments: Rather than using the Maclaurin series to establish the general solution for
the case of a double root, you can also use the formula on page 307 that establishes a
general solution if a particular solution is known. The general case of a second order
linear equation with constant coecients that has a double root is
d
2
y
dx
2
2a
dy
dx
+a
2
y = 0,
whose auxillary equation has its double root at a. Because a is a root, we know that
y = e
ax
is a particular solution. So P = 2a, Q = a
2
and v = e
ax
. We also have

P dx = 2ax. Applying the formula

log
du
dx
+ 2 log (e
ax
) 2ax = log C
1
,
log
du
dx
+

2ax

2ax = log C
1
,
log
du
dx
= log C
1
,
du
dx
= C
1
,
u =C
1
x +C
2
,
y = uv = (C
1
x +C
2
) e
ax
.
S-22
Page 309 ( 129. linear equations with constant coecients, case 3, nonreal
roots) example 1: All thats needed is to prove that e
x
= coshx + sinh x. By
denition
coshx =
e
x
+e
x
2
and sinh x =
e
x
e
x
2
,
hence
coshx + sinh x =
e
x
+
$
$$
e
x
+e
x

$
$$
e
x
2
= e
x
.
Page 309 ( 129. linear equations with constant coecients, case 3, nonreal
roots) example 2: The auxillary equation is D
2
+D+1 = 0. Its roots are
1
2

3
2
.
This means that y = C
1
e
(1+

3)x/2
+C
2
e
(1

3)x/2
is the general solution. To turn
this into a real-valued function, let C
1
=
1
2
(A B) and C
2
=
1
2
(A+B) be complex
conjugates of each other. Then
y =
1
2
(A B) e
(1+

3)x/2
+
1
2
(A +B) e
(1

3)x/2
,
y =
1
2
(A B) e
x/2
e

3x/2
+
1
2
(A +B) e
x/2
e

3x/2
, and by Eulers formula,
y =

A
2

B
2

e
x/2

cos

3x
2
+ sin

3x
2

A
2
+
B
2

e
x/2

cos

3x
2
sin

3x
2

,
y =
A
2
e
x/2
cos

3x
2
+
$
$
$
$
$
$
$
$

A
2
e
x/2
sin

3x
2

$
$
$
$
$
$
$
$

B
2
e
x/2
cos

3x
2
+
B
2
e
x/2
sin

3x
2
+
A
2
e
x/2
cos

3x
2

$
$
$
$
$
$
$
$

A
2
e
x/2
sin

3x
2
+
$
$
$
$
$
$
$
$

B
2
e
x/2
cos

3x
2
+
B
2
e
x/2
sin

3x
2
,
y =Ae
x/2
cos

3x
2
+Be
x/2
sin

3x
2
.
Page 309 ( 129. linear equations with constant coecients, case 3, nonreal
roots) example 4: The auxillary equation factors into (D1)

D
2
+ 1

= 0. This
has roots at 1 and . The books answer follow immediately from these roots.
Page 311 ( 130. Finding particular solutions) case 1, example 1: The illus-
tration has you nd the particular integral, p (x), by integrating (see equation (4) on
page 311)
p (x) = e
3x

e
3x
e
4x
dx e
2x

e
2x
e
4x
dx = e
3x

e
x
dx e
2x

e
2x
dx.
The result is p (x) = e
4x

1
2
e
4x
, which, when simplied and combined with the comple-
mentary solution, yields the books answer.
Page 311 ( 130. Finding particular solutions) case 1, example 2: Auxillary
equation factors into (D 3) (D1) = 0. The decomposition into partial fractions is
1
(D3) (D1)
=
1
2
D3

1
2
D 1
.
Integrating using equation (4) on page 311,
p (x) =
1
2
e
3x

2e
3x
e
3x
dx
1
2
e
x

2e
x
e
3x
dx = e
3x

dx e
x

e
2x
dx.
This yields p (x) = xe
3x

1
2
e
3x
. The second summand is linearly dependent upon
complementary solution, so we can eliminate it from the particular solution, leaving
p (x) = xe
3x
. Combining that with the complementary solution, C
1
e
x
+ C
2
e
3x
, yields
the books answer.
Page 311 ( 130. Finding particular solutions) case 2, example 2: In this
case, R = e
x
. In problems of nding particular solutions, the expression in the R-
position is commonly referred to as the forcing function. When the forcing function is
S-23
an exponential that is linearly independent of the complementary solution, there is an
easy way to nd the particular solution. When these conditions exist, the particular
solution must be a constant multiple of the forcing function. So we let p (x) = R. In
this case that means p (x) = e
x
. We then substitute that into the original equation
and solve for . In this case, where the original equation is

D
2
4D + 4

y = e
x
,
e
x
$
$$
4e
x
+$
$$
4e
x
= e
x
.
Hence = 1 and p (x) = e
x
. So the general solution is y = C
1
e
2x
+C
2
xe
2x
+e
x
.
This method also works when the forcing function is the sum of exponentials, provided
that none of the exponentials in that sum are linearly dependent upon the complemen-
tary solution. So, for example, if we were to solve

D
2
4D + 4

y = e
3x
2e
3x
, then
p(x) =
1
e
3x
+
2
e
3x
. Substituting into the original equation,
9
1
e
3x
+ 9
2
e
3x
12
1
e
3x
+ 12
2
e
3x
+ 4
1
e
3x
+ 4
2
e
3x
= e
3x
2e
3x
.
By classifying terms, this separates into
9
1
12
1
+ 4
1
=
1
= 1 and 9
2
+ 12
2
+ 4
2
= 25
2
= 2.
Hence
1
= 1 and
2
=
2
25
and p (x) = e
3x

2
25
e
3x
.
The procedures shown above for nding particular solutions for both R = e
x
and
R = e
3x
2e
3x
are examples of the method of undetermined coecients.
The principle of superposition: In the example above, we had a forcing function
of R = R
1
+ R
2
, where R
1
= e
3x
and R
2
= e
3x
. If you solve the particular integral
for just R
1
by itself to get p
1
(x) and for just R
2
by itself to get p
2
(x), you will nd
that p (x) = p
1
(x) + p
2
(x) is the particular integral we got for R = R
1
+ R
2
. This
is the principle of superposition. It asserts that whenever the forcing function is a
sum of functions, you can nd the particular integral for each summand separately,
and the particular integral for the sum will be the sum of those particular integrals.
In addition, for any constant scalar multiplier, , the particular integral for a forcing
function, R, will be p (x), where p (x) is the particular integral for R by itself. The
principle of superposition of forcing functions and the assertion on scalar multiples of
forcing functions are both generally true for all linear dierential equations.
Page 312 ( 130. Finding particular solutions) case 3, example 1: This example
too can be done using the method of undetermined coecients. We observe that if the
forcing function, R, is a polynomial of degree n, then the particular integral must also
be a polynomial of degree n. In the example, n = 2. So we let p (x) =
2
x
2
+
1
x +
0
and substitute into into the original equation.

D
2
4D + 4

2
x
2
+
1
x +
0

= 2
2
8
2
x + 4
2
x
2
4
1
+ 4
1
x + 4
0
= x
2
.
Separating terms according to powers of x,
4
2
x
2
= x
2
,
8
2
x + 4
1
x = 0,
2
2
4
1
+ 4
0
= 0.
Dividing out the powers of x from the rst two lines and solving the linear system yields

0
=
3
8
,
1
=
1
2
, and
2
=
1
4
, which is the same as the books answer.
Page 312 ( 130. Finding particular solutions) case 3, example 2: This one
can be done by inspection using the method of undetermined coecients. The forcing
function, R, is the rst degree polynomial, 2 + 5x. Since the second derivative of any
rst degree polynomial is zero, then if p (x) is the particular integral, substituting it into
the original equation gives p (x) = 2 +5x, or equivalently, p (x) = 2 5x. Summing
that particular solution with the complementary solution yields the books answer.
S-24
Were you to do it by the method in the book, you would rst resolve the reciprocal
operator into partial fractions.
1
D
2
1
=
1
(D + 1) (D 1)
=
1
2
1
D1

1
2
1
D + 1
.
Hence the integral problem to be solved is
p (x) =
1
2
e
x

(2 + 5x) e
x
dx
1
2
e
x

(2 + 5x) e
x
dx,
which, after your expending a lot more work than you would using the rst method,
yields the same p (x).
Page 312 ( 130. Particular solutions) case 4, clarication of notation: The
rst line of (5) on page 312 would be clearer as
D
n
(e
ax
X) = e
ax
(D +a)
n
(X) ,
so as to show the extent of application of the dierential operators. The nal result of
(5) would be clearer as
D
n
(e
ax
X) = e
ax
(D +a)
n
(X) .
This is simply assertion that the rst formula works even when n is negative.
Page 312 ( 130. Particular solutions) case 4, example 2: By (5) on page 312,
(D 1)
1
(e
x
log x) = e
ax
(D + 1 1)
1
(log x) = e
ax
D
1
(log x) .
The D
1
operator is simply the taking of the antiderivative. The antiderivative of log x
is x(log (x) 1) = xlog (x/e). The books answer follows from this.
Page 313 ( 130. Particular solutions) case 4, example 1: By (5) on page 312,
(D + 1)
3

e
x

= e
x
(D 1 + 1)
3
1 = e
x
D
3
1.
The D
3
operator tells us to take the antiderivative of 1 three times. The rst time
gives x, the second,
1
2
x
2
, and the third,
1
6
x
3
. The books answer follows from that.
Page 313 ( 130. Particular solutions) case 4, example 2: First factor the
operator:

D
3
1

= (D 1)

D
2
+D + 1

. By (5) on page 312,

(D 1)
1

D
2
+ D + 1

1
xe
x
= e
x
D
1

D
2
+ 3D + 3

1
x.
Applying D
1
to x gives
1
2
x
2
. We then apply

D
2
+ 3D + 3

1
to
1
2
x
2
. If p (x) is the
function that solves this, then

D
2
+ 3D + 3

(p) =
1
2
x
2
. (a)
We can apply the method of undetermined coecients to this. Let p =
2
x
2
+
1
x+
0
.
Applying the dierential operator in (a) to p,
2
2
+ 6
2
x + 3
1
+ 3
2
x
2
+ 3
1
x +
0
=
1
2
x
2
.
Sorting terms by powers of x gives the linear system,
3
2
x
2
=
1
2
x
2
,
6
2
x + 3
1
x = 0,
2
2
+ 3
1
+ 3
0
= 0,
yielding
2
=
1
6
,
1
=
1
3
, and
0
=
2
9
. Hence p (x) =

1
6
x
2

1
3
x +
2
9

e
x
. The books
answer is missing the constant term inside the parentheses. You are invited, though, to
demonstrate for yourself that
d
3
p
dx
3
=

1
6
x
2
+
2
3
x +
2
9

e
x
,
which means that

D
3
1

p = xe
x
, as the problem requires. The books answer does
not, however, solve this equation.
S-25
Page 313 ( 130. Particular solutions) case 5, example 1: This one factors
into p (x) =

D
2
+ 1

1
(D + 1)
1
sin 2x. Since n = 2, we have n
2
+ 1 = 3.
So according to (9) on page 313, we have p (x) =
1
3
(D + 1)
1
sin 2x, or equiva-
lently, (D + 1) p (x) =
1
3
sin 2x. When you multiply both sides by (D1), you
get

D
2
1

p (x) =
1
3
(D1) sin 2x. On the left we replace D
2
with n
2
to get
5p (x) =
1
3
(D1) sin 2x. Now apply the remaining dierential operator to sin 2x
and divide both sides by 5 to get the books answer.
Page 314 ( 130. Particular solutions) case 5, example 2: In the D-form, this
equation is

D
2
k
2

y = cos mx. Clearly the roots for the formation of the comple-
mentary solution are k. We set up
p (x) =

D
2
k
2

1
cos mx.
Here n = m, so we replace D
2
with m
2
, so according to (10) on page 313
p (x) =
1
m
2
k
2
cos mx.
The books answer follows from this.
Page 314 ( 130. Particular solutions) case 5, example 3: Setting this up in
D-form,

D
2
+mD +n
2

y = a sin nt.
So for the particular solution, p (x), we have
p (x) =

D
2
+mD +n
2

1
a sinnt.
Replacing D
2
with n
2
(in accordance with (9) on page 313) results in a cancellation
inside the parentheses to give
p (x) =
1
m
D
1
a sinnt.
Again the D
1
operator is simply taking the antiderivative. Taking the antiderivative
of sin nt gives
1
n
cos nt. The books answer follows from that and from the given
information that the roots for forming the complementary solution are and .
In many engineering books, this equation is given in the form of
d
2
y
dt
2
+ 2
dy
dt
+
2
= a sin t.
Here is known as the resonant frequency of the system and is known as the damping See page
S-61 for
further
develop-
ment of
resonant
systems.
factor. The reciprocal of 2 is known as the Q-factor. You are encouraged to solve this
problem in this form as well. What happens to the roots of the auxillary equation when
0 < < 1? How does that aect the complementary solution? As approaches zero,
how does this aect the particular solution? This equation represents what engineers
call a resonant system.
Page 314 ( 130. Particular solutions) case 5, example 1: That
1
2
xcos x
results from

D
2
+ 1

1
sin x can be proved in the same way as is shown on page 314
for the identical operator on cos x. But with a little a priori knowledge, we can apply
the method of undetermined coecients to this too. The a priori knowledge is that if

D
2
+n
2

is a factor of the operator, and a sin nx+b cos nx is the forcing function, then
the particular integral will be in the form of p (x) =
s
xsin nx+
c
xcos nx. In this case,
a = 1, b = 0 and n = 1.

D
2
+ 1

(
s
xsin x +
c
xcos x) = sin x.
Hence

s
(2 cos x $
$
$
xsin x +$
$
$
xsin x) +
c
(2 sinx
$
$
$$
xcos x +
$
$
$$
xcos x) = sin x.
Clearly
s
= 0 and
c
=
1
2
. The books answer follows from that.
S-26
Similarly you can nd that for arbitrary n, where

D
2
+n
2

is a factor and sinnx is

the forcing function,

2ncos nx @
@
@
@@
n
2
xsin nx +@
@
@
@@
n
2
xsin nx

+
c

2nsinx $
$
$
$$
n
2
xcos x +$
$
$
$$
n
2
xcos x

= sin nx.
In this case,
s
= 0 and
c
=
1
2n
.
Page 314 ( 130. Particular solutions) case 5, example 4: In the D-form, this is

D
3
1

y = xsin x, which factors into (D 1)

D
2
+D + 1

y = xsin x. The roots of

the auxillary equation are 1 and
1
2

1
2

3. These roots account for the complementary

solution in accordance with case 3 on page 309.
The book oers no satisfactory strategy for solving the particular integral on this one
until case 6, so this may be regarded as a challenge problem. Here are two approaches
you might take. The rst is to observe that according to Eulers formula
p (x) =

D
3
1

1
xsin x =

D
3
3

1
x

e
x
e
x
2

,
where =

1. By the principle of superposition

p (x) =

D
3
1

1
x
e
x
2

D
3
1

1
x
e
x
2

.
By case 4 on page 312, this is the same as
p (x) =

e
x

D
3
+3D
2
3D 1

1 x
2

e
x

D
3
3D
2
3D + 1

1 x
2

.
So if p (x) = p
1
(x) p
2
(x) where
p
1
(x) =e
x

D
3
+3D
2
3D 1

1 x
2
and
p
2
(x) =e
x

D
3
3D
2
3D + 1

1 x
2
,
then we can solve for p
1
and p
2
separately using the method of undetermined coecients,
and then recombine the solutions to form p. If p
1
= e
x
(
0
+
1
x), then we need only
concern ourselves with the (3D 1) part of the dierential operator, as higher
derivatives of (
0
+
1
x) are zero. So for p
1
we must solve
(3D 1) (
0
+
1
x) =
x
2
for
0
and
1
. Expanding this,
( 1)
1
x =
x
2
,
3
1
+ ( 1)
0
= 0.
Hence
1
=
1
4
(1 +) and
0
=
3
4
, from which we can write p
1
(x). Using the same
methodology we also solve
(3D + 1) (
2
+
3
x) =
x
2
.
The result is
3
=
1
4
(1 +) and
2
=
3
4
. This allows us to write p
2
(x) as well
as p
1
(x).
p (x) = p
1
(x) p
2
(x) =
1
4
e
x
3 + (1 +) x
1
4
e
x
3 + (1 +) x ,
p (x) =
3
4

e
x
+e
x

+
1
4

e
x
+e
x

x +
1
4

e
x
e
x

x,
and by Eulers formula
p (x) =
3
2
cos x +
1
2
xcos x
1
2
xsin x,
which is the same as the books answer.
S-27
The second way to solve this is to use the method of undetermined coecients from
beginning to end. When the forcing function is x times sine or cosine of x (and D
2
+ 1
is not a factor of the operator), then the particular integral will be in the form of
p (x) =
c0
cos x +
c1
xcos x +
s0
sin x +
s1
xsin x. (a)
By applying Leibnizs rule we nd that
d
3
p
dx
3
=
c0
sin x +
c1
xsin x 3
c1
cos x
s0
cos x
s1
xcos x 3
s1
sin x.
When you take the dierence,
d
3
p
dx
3
p = xsin x, then pick the result apart into separate
equations according to whether the coecient is multiplied by cos x or xcos x or sin x
or xsin x, it gives the linear system,

c1

s1
= 1,

c1

s1
= 0,

c0
3
c1

s0
= 0,

c0

s0
3
s1
= 0.
Solving the above yields
c0
=
3
2
,
c1
=
1
2
,
s0
= 0, and
s1
=
1
2
. Using these
coecients in (a) gives the books answer.
Page 315 ( 130. Particular solutions) case 6, example 2: If you tried the
method given on the top half of page 315 directly, you probably ran into diculty. This
is because

D
2
+ 1

is a factor of

D
4
1

, making sinx a part of the complementary

solution. A way to do this is to solve it in two phases.

D
4
1

1
xsin x =

D
2
+ 1

D
2
1

1
xsin x.
Applying (11) on page 315 to the

D
2
1

factor,

D
2
+ 1

D
2
1

1
xsin x =

D
2
+ 1

D
2
1

1
(2D)

D
2
1

1
sin x.
By (9) on page 313 this is

D
2
+ 1

D
2
1

1
xsin x =
1
2

D
2
+ 1

D
2
1

1
(2D)

sin x,
=
1
2

D
2
+ 1

xsin x 2

D
2
1

1
cos x

.
By (10) on page 313

D
2
+ 1

D
2
1

1
xsin x =
1
2

D
2
+ 1

1
xsin x + cos x .
By superposition

D
2
+ 1

D
2
1

1
xsin x =
1
2

D
2
+ 1

1
xsin x

1
2

D
2
+ 1

1
cos x

.
By the development on page 314

D
2
+ 1

D
2
1

1
xsin x =
1
2

D
2
+ 1

1
xsin x

1
4
xsin x. (a)
For the second phase, it remains to nd

D
2
+ 1

1
xsin x. Unfortunately this is the
exceptional case of (11) on page 315 that does not immediately yield to this method. You
end up with an expression in which you must, once again, evaluate

D
2
+ 1

1
xsin x.
To continue you need to apply a trick. Here then is that trick.

D
2
+ 1

1
xsin x =

D
2
+ 1

1
(2D)

D
2
+ 1

1
sin x,
=

D
2
+ 1

1
(2D)

1
2

xcos x,
=
1
2
x
2
cos x +

D
2
+ 1

1
D(xcos x) ,
=
1
2
x
2
cos x +

D
2
+ 1

1
(xsin x + cos x) .
S-28
Now add

D
2
+ 1

1
xsin x to both sides of the above:
2

D
2
+ 1

1
xsin x =
1
2
x
2
cos x +

D
2
+ 1

1
cos x,
=
1
2
x
2
cos x +
1
2
xsin x,
from which we see that

D
2
+ 1

1
xsin x =
1
4
x
2
cos x +
1
4
xsin x.
We can also use the method of undetermined coecients to solve this. In this case,
rather than set it up formally, I will use the inspection method (that is making educated
guesses and rening them). Because
1
2
xcos x results when you apply

D
2
+ 1

1
to
sin x, we should suspect that when we apply the same to xsin x, one of the terms of the
result ought to be a multiple of x
2
cos x. So we apply the operator to that expression.

D
2
+ 1

x
2
cos x =
$
$
$
$$
x
2
cos x 4xsinx + 2 cos x +$
$
$
$
x
2
cos x.
We need to add another term that will cancel the 2 cos x term in the above. Based upon
experience, we suspect that term will be a multiple of xsin x.

D
2
+ 1

x
2
cos x +xsin x

=
$
$
$
$$
x
2
cos x 4xsinx + 2 cos x $
$
$
$
xsin x + 2cos x +$
$
$
$
x
2
cos x +$
$
$
$
xsin x.
Clearly if = 1, then

D
2
+ 1

x
2
cos x xsin x

= 4xsinx. From this we conclude

again that
D
2
+ 1

1
xsin x =
1
4
x
2
cos x +
1
4
xsin x.
We substitute this into (a) on the previous page to get

D
2
+ 1

D
2
1

1
xsin x =
1
8
x
2
cos x
1
8
xsin x
1
4
xsin x.
Gathering terms in the above yields the books answer.
Page 315 ( 130. Particular solutions) case 6, example 3: To solve this
one it is useful rst to evaluate

D
2
1

e
x
sin x and

D
2
1

e
x
cos x. This is done
easily using the method of undetermined coecients. In each case we assume that
p (x) =
c
e
x
cos x +
s
e
x
sin x. We nd

D
2
1

p (x) = (2
s

c
) e
x
cos x + (2
c

s
) e
x
sin x.
Separating terms according multiples of e
x
cos x and e
x
sin x and putting it all into
matrix form we have

c

s

1 2
2 1

1
0

row for e
x
cos x
row for e
x
sin x

for nding

D
2
1

1
e
x
cos x.
By Cramers rule,
c
=
1
5
and
s
=
2
5
. Hence

D
2
1

1
e
x
cos x =
1
5
e
x
cos x +
2
5
e
x
sinx. (a)
Also

c

s

1 2
2 1

0
1

row for e
x
cos x
row for e
x
sin x

for nding

D
2
1

1
e
x
sin x.
By Cramers rule,
c
=
2
5
and
s
=
1
5
. Hence

D
2
1

1
e
x
sin x =
2
5
e
x
cos x
1
5
e
x
sin x. (b)
The problem is to solve

D
2
1

y = xe
x
sinx. Clearly the complementary solution is
y
c
= C
1
e
x
+C
2
e
x
. To nd the particular solution, p (x), we apply (11) on page 315.
p (x) =

D
2
1

1
e
x
sin x =

D
2
1

1
(2D)

D
2
1

1
e
x
sin x.
S-29
By (b) above we have
p (x) =

D
2
1

1
(2D)

2
5
e
x
cos x
1
5
e
x
sinx

=
2
5
xe
x
cos x
1
5
xe
x
sinx 2

D
2
1

3
5
e
x
cos x +
1
5
e
x
sin x

.
Finally by superposition and by (a) and (b) above we have
p (x) =
2
5
xe
x
cos x
1
5
xe
x
sinx
2
25
e
x
cos x +
14
25
e
x
sin x.
Factoring out
1
25
e
x
and gathering terms from the above yields the books answer.
Page 315 ( 130. Particular solutions) case 6, example 4: To solve this one
it is useful rst to evaluate

D
2
1

1
xcos x and

D
2
1

1
xsin x. Each of them
succumb easily to combining (11) on page 315 with (9) and (10) on page 313.

D
2
1

1
xcos x =

D
2
1

1
(2D)

D
2
1

1
cos x,
=

D
2
1

1
(2D)

1
2
cos x

,
=
1
2
xcos x +

D
2
1

1
Dcos x,
=
1
2
xcos x

D
2
1

1
sin x,
=
1
2
xcos x +
1
2
sin x. (a)
Likewise

D
2
1

1
xsin x =

D
2
1

1
(2D)

D
2
1

1
sinx,
=

D
2
1

1
(2D)

1
2
sinx

,
=
1
2
xsin x +

D
2
1

1
Dsin x,
=
1
2
xsin x +

D
2
1

1
cos x,
=
1
2
xsin x
1
2
cos x. (b)
Now apply (11) on page 315 then (a) above and nally (b) above to the problem at
hand.

D
2
1

1
x
2
cos x =

D
2
1

1
(2D)

D
2
1

1
xcos x,
=

D
2
1

1
(2D)

1
2
xcos x +
1
2
sin x

,
=
1
2
x
2
cos x +
1
2
xsin x +

D
2
1

1
D(xcos x sin x) ,
=
1
2
x
2
cos x +
1
2
xsin x

D
2
1

1
xsin x,
=
1
2
x
2
cos x +
1
2
xsin x

1
2
xsin x
1
2
cos x

,
=
1
2
x
2
cos x +xsin x +
1
2
cos x,
which is the books answer.
S-30
The Method of Laplace. Section ' 130 details a variety of artices you can employ
to determine the particular integral when you have a linear dierential equation with
constant coecients together with a forcing function. I have also described, here in the
supplement, the additional method of undetermined coecients, which Mellor does not
cover, as well as stating the principle of superposition. By applying these methods, you
can nd the particular integral for any such equation whenever the forcing function is any
sum and/or product of exponentials, sines, cosines, and polynomials. The only diculty
is that the method you need to bring to bear depends on the type of forcing function.
Here I will describe the method of Laplace, with which you can nd the particular
integral in any problem using only this single method. The method of Laplace has
the additional advantage in that it can deal with initial conditions and it can easily be
applied to simultaneous equations. It is, to quote Tolkein, the one ring to rule them all.
You may have noticed that there is a correspondence between functions of the equations
independent variable (usually x or t), and the functions that result from the dierential
operator, D. For example, if you have
dy
dx
ay = e
bx
hence y =
1
D a
e
bx
,
You know that the complementary function will have an e
ax
term in it. If we were to
nd the particular integral by integrating, the integral expression to be solved would be
p (x) = e
ax

e
bx
e
ax
dx.
We see that there is a correspondence between seeing the (Da)
1
as an operator and
the function, e
ax
. This correspondence works in both directions. So we can observe that
since the forcing function is e
bx
in the example, we should be able to replace it with
(D b)
1
. We end up solving for y as a function of D instead of as a function of x.
Y (D) =
1
(D a) (Db)
=
1
a b

1
Da

1
Db

.
Notice that when you go from a function of x (or t) to a function of D, as we did with y,
it is customary notation to use the upper case symbol for the function of D (e.g., Y (D)).
Once the D-function has been expanded into partial fractions, it is immediately evident
what the nal solution to this problem is. We simply translate from the D-function
back to a function of x. We know that (Da)
1
corresponds with e
ax
, which is a part
of the complementary function. So that part of the D-function becomes Ce
ax
. The
(D b)
1
portion translates back to e
bx
. Since it is a part of the particular integral,
we must pay attention to its constant multiplier and carry that into the nal solution.
Hence the solution to the example is
y = Ce
ax

1
a b
e
bx
.
The table on the following page shows the pairings of various functions of x with their
corresponding functions of D. It also shows some rules by which you can infer the
corresponding D-function of more complicated functions of x. If you have a function
of x, then its corresponding D-function is known as its Laplace transform. Many if
not most modern textbooks use the symbol, s (or sometimes p), in place of D, as the
independent variable of a functions Laplace transform. This is a matter of nomenclature
only, and it does not change the essence of the method one bit. Because Mellor uses
the symbol, D, throughout, I will stick with it here in the further explanation of the
Laplace method.
What the Laplace method does is to convert the problem of nding a particular integral
into nothing more than a partial fractions problem. Take, for example, the problem
d
2
y
dx
2
+ 9y = x
2
, or equivalently,

D
2
+ 9

y = x
2
.
(text continues on page S-33).
S-31
f (x) F (D)
1
1
D
e
ax
1
D a
sinbx
b
D
2
+b
2
cos bx
D
D
2
+b
2
x
n
, where n 0
n!
D
n+1
x
n
e
ax
, where n 0
n!
(Da)
n+1
Rule: e
ax
g (x) G(Da)
Hence e
ax
sin bx
b
(Da)
2
+b
2
and e
ax
cos bx
D a
(Da)
2
+b
2
Rule: x
n
g (x)
(1)
n
d
n
G
dD
n
Hence xe
ax
sin bx
2b (Da)

(Da)
2
+b
2

2
and xe
ax
cos bx
(Da)
2
b
2

(Da)
2
+b
2

2
Rule: g (ax)
1
a
G

D
a

Rule:
dg
dx
(D) G(D)
Rule:

x
0
g (u) du
1
D
G(D)
Rule: g (x) + h(x) G(D) +H (D)
sinh bx
b
D
2
b
2
coshbx
D
D
2
b
2
Short table of Laplace Transforms.
S-32
Using the table to look up the forcing function and solving for Y (D), it becomes
Y (D) =
2
(D
2
+ 9) D
3
=
2D
81 (D
2
+ 9)

2
81D
+
2
9D
3
.
We know that the complementary function for the original equation is
C
1
sin 3x + C
2
cos 3x. Once expanded into partial fractions, the rst term of Y (D)
translates back to a multiple of cos 3x, so it is a part of the complementary function.
The remaining terms are a part of the particular integral. Using the table to translate
them back to a function of x, we nd the full solution is
y = C
1
sin3x +C
2
cos 3x
2
81
+
1
9
x
2
.
A type of example that tends to trip up beginners is where the forcing function is linearly
dependent upon a term of the complementary function. The Laplace method handles
these seamlessly. For example
d
2
y
dx
2
+b
2
y = cos bx, or equivalently,

D
2
+b
2

y = cos bx.
The complementary function is C
1
sin bx + C
2
cos bx, so clearly the forcing function is
linearly dependent upon a term of the complementary function. Using the table to solve
this one for Y (D),
Y (D) =
D
(D
2
+b
2
)
2
.
Using the table to translate back we see that the particular integral in this case is
p (x) =
1
2b
xsin bx, which is consistent with the result on page 314 of the text (note that
the table entry for xe
ax
sinbx gives xsin bx when a = 0). If we had started with a forcing
function of sinbx, which is also linearly dependent upon the complementary function,
the solution for Y (D) would have been
Y (D) =
b
(D
2
+b
2
)
2
,
=
1
b
b
2
(D
2
+b
2
)
2
,
=
1
b
1
2
b
2

1
2
D
2
+
1
2
D
2
+
1
2
b
2
(D
2
+b
2
)
2
,
=
1
2b

D
2
b
2
(D
2
+b
2
)
2

1
D
2
+b
2

.
The second term inside the brackets is a part of the complementary function. The rst
term corresponds in the table to xcos bx, so the particular solution is p (x) =
1
2b
xcos bx,
which is consistent with example 1 on page 314.
As stated before, the method of Laplace can also determine the undetermined con-
stants in the complementary function given initial conditions at x (or t) equal to
zero. If y is a function of x (or t), then the Laplace transform of the rst deriva-
tive of y is DY (D) y (0). The Laplace transform of the second derivative of y is
D
2
Y (D) Dy (0) y

(0), where y (0) and y

(0) are initial conditions at x = 0 (or

t = 0). Higher derivatives follow the same pattern. As an example, we solve the follow-
ing equation with a stipultated initial condition:
dy
dt
+ay = sin t, where y (0) = 1.
In the Laplace transform version, this equation is
(D +a) Y (D) 1 =

D
2
+
2
.
Make sure you understand how the initial condition results in the 1 just to the left of
the equal.
S-33
Solving for Y ,
Y (D) =

(D
2
+
2
) (D +a)
+
1
D +a
,
=

a
2
+
2

a D
D
2
+
2
+
1
D +a

+
1
D +a
. (a)
Letting A =

a
2
+
2
, we nd that
y (t) = A
a

sin t Acos t + (A + 1) e
at
.
The solution above derives from looking up the terms of (a) on the table. As an exercise,
you are encouraged to conrm that y (0) = 1 and to verify that this solution satises the
original equation. Observe that there are no undetermined constants in this solution.
This is because we stipulated the initial condition, and doing so xes the multiplier of
the complementary function.
Applying the Laplace method to simultaneous equations. Here is an example.
dy
dx
+ z = 0, where y (0) = 0,
dz
dx
4y = x, where z (0) = 3.
Translating this into Laplace form,
DY + Z = 0,
DZ 4Y =
1
D
2
3.
In matrix form this is
Y Z

D 1
4 D

0
1
D
2
3

The determinant of this matrix is D

2
+ 4. Cramers rule yields
Y (D) =
3
D
2
+ 4

1
D
2
(D
2
+ 4)
=
3
D
2
+ 4

1
4

1
D
2

1
D
2
+ 4

,
Z (D) =
1
D(D
2
+ 4)

3D
D
2
+ 4
=
1
4

1
D

D
D
2
+ 4

3D
D
2
+ 4
.
Gathering terms and then translating back using the table gives
y =
13
8
sin2x
1
4
x,
z =
1
4

13
4
cos 2x.
Again it is left as an exercise for you to verify that this solution satises both the original
system of equations and the initial conditions.
In engineering, the problem of simultaneous linear equations occurs over and over again
in analog electronics and in feedback and control theory. In those elds, Laplace is the
preferred method for solving such problems. Indeed manufacturers of analog electronic
parts and electro-mechanical transducers frequently characterize their products by citing
product responses in terms of Laplace functions. The power of the Laplace method is
that it converts calculus problems into algebra problems. You are urged to review the
example above to see how solving for Y and Z became, in the Laplace domain, is simply
a problem of solving a system of linear algebra equations. Decomposing that solution
into partial fractions is again just an algebra problem. The nal step of translating back
is nothing more than a matter of looking it up on the table.
S-34
Page 315 (Linear equations with variable coecients) case 1, info: The funny-
looking symbol, , that the text uses throughout this case is called vartheta, and is a
cursive form of the Greek letter, .
Page 316 ( 131. Linear equations with variable coecients) case 1, rst
example 1: Page 315 establishes that z = log x. Hence
dz
dx
=
1
x
, so by the chain rule,
1
x
d
dz
=
dz
dx
d
dz
=
d
dx
.
Multiplying through by x completes the proof.
Page 316 ( 131. Linear equations with variable coecients) case 1, rst
example 2: The operation consists of rst taking the derivative, then multiplying
by x.
d
dx
(x
m
) = mx
m1
; Now multiplying by x, x
d
dx
(x
m
) = mx
m
.
Alternatively, from page 315, x = e
z
, so x
m
= e
mz
. By example 1,
(x
m
) =
d
dz
(e
mz
) = me
mz
= mx
m
.
Page 316 ( 131. Linear equations with variable coecients) case 1, second
example 2: Using the operator according to identities at the bottom of page 315,

y +
&
& y +q
2
y = 0,
2
y +q
2
y = 0.
Per page 315, we make the substitution, z = log x, and apply the identity from ex-
ample 1. So with the substitution and D =
d
dz
, you have

D
2
+q
2

y = 0. Hence
y = C
1
sin qz +C
2
cos qz = C
1
sin (q log x) +C
2
cos (q log x), which is the books answer.
Page 316 (Linear equations with variable coecients) case 1, third
example 2: Substituting x
m
for y,
m(m1) x
m
+ 4mx
m
+ 2x
m
= 0.
Dividing out x
m
and solving for m:
m
2
+ 3m+ 2 = 0, hence m = 1 or m = 2.
The solution, then, is y = C
1
x
1
+C
2
x
2
, which is is the books answer.
Observe that we could also have applied this method to the second example 2 (that is the
previous example shown on this page), but m would have been imaginary in that case.
For that example we would have gotten m = q, for a solution of y = A
1
x
q
+A
2
x
q
.
As an exercise, apply Eulers formula to show that this solution is equivalent to the
solution already shown of y = C
1
sin (q log x) +C
2
cos (q log x). What is the relationship
between the undetermined constant sets, A
1
, A
2
and C
1
, C
2
?
Page 316 ( 131. Linear equations with variable coecients) case 1, example
3: Replacing x with e
z
, we have:
D(D1) 3D + 4 y = e
3z
,
(D 2)
2
y = e
3z
. (a)
The complementary function is y = C
1
e
2z
+ C
2
ze
2z
. Replacing e
z
with x and z with
log x yields y = C
1
x
2
+C
2
x
2
log x, which is equivalent to the books answer. In addition,
we can apply the method of undetermined coecients to (a) to establish the particular
integral. Let p (x) = e
3z
. Then
(D 2)
2
p (x) =

D
2
4D + 4

p (x) = (9 12 + 4) e
3x
= e
3x
,
hence = 1. So p (x) = e
3x
= x
3
.
S-35
Page 316 ( 131. Linear equations with variable coecients) case 1, example
4: The auxillary equation for x
m
is m
2
m2 = 0. So m = 2 or m = 1. That leads
to y = C
1
x
2
+C
2
x
1
, which is the books answer.
Page 317 ( 131. Linear equations with variable coecients) case 1, example
1: The equation is equivalent to ( 1) 2 4 y = x
4
. This factors into
( + 1) ( 4) y = x
4
.
The roots are 4 and 1, which establishes the complementary solution as C
1
x
4
+C
2
x
1
.
Substituting x = e
z
and with D, we can nd the particular integral by solving
p (z) = (D + 1)
1
(D4)
1
e
4z
.
According to (5) on page 312, this is the same as
p (z) = e
4z
(D + 5)
1
D
1
1.
Applying the D
1
rst,
p (z) = e
4z
(D + 5)
1
z.
From this you get p (z) = e
4z

1
5
z
1
25

. But e
4z
= x
4
is a part of the complementary
solution. So we can drop the
1
25
e
4z
term. Substituting z = log x gives p (x) =
1
5
x
4
log x.
Page 317 ( 131. Linear equations with variable coecients) case 1, example
4: This one is equivalent to
( 1) + 4 + 2 y = e
x
,

2
+ 3 + 2

y = e
x
,
( + 1) ( + 2) y = e
x
.
Hence the complementary function is C
1
x
1
+C
2
x
2
. The book oers no direct method
for nding the particular integral of this one. But we can cobble together a procedure
out of various methods weve learned so far. First we replace x with e
z
and with D,
as we have done in other examples.
p (z) = (D + 2)
1
(D + 1)
1
e
(e
z
)
.
Then we apply integral in (4) from page 311 to the (D + 1)
1
operator.
p (z) = (D + 2)
1
e
z

e
z
e
(e
z
)
dz = (D + 2)
1
e
z
e
(e
z
)
.
Now do the same with the (D + 2)
1
operator.
p (z) = (D + 2)
1
e
z
e
(e
z
)
= e
2z

e
2z
e
z
e
(e
z
)
dz = e
2z
e
(e
z
)
.
Back-substitute x for e
z
, and you get the books answer.
Taking a more general view of the procedure we used on this example, we solve
p (x) = ( +a)
1
R(x) by solving p (z) = (D +a) R(e
z
). We do this by applying
the integral in (4) on page 311.
p (z) = e
az

e
az
R(e
z
) dz.
Since log x = z, we also have dx/x = dz. Back-substituting this integral gives the
formula,
p (x) = ( +a)
1
R(x) = x
a

x
a1
R(x) dx, (a)
which is a general formula for such problems. If you apply ( + 1)
1
to R(x) = e
x
using (a), then apply ( + 2)
1
to the result, you will end up with the same particular
integral for example 4 as we arrived at above.
S-36
Suppose we do the same example, but this time we rst apply ( + 2)
1
to R(x) = e
x
,
then apply ( + 1)
1
to the result. We expect to end up with the same function as
before.
( + 2)
1
e
x
= x
2

xe
x
dx = x
2
(x 1) e
x
=
e
x
x

e
x
x
2
.
Before we apply the second operator to this result, we must deal with the fact that we
will end up with integrals that cannot be expressed in closed form. Let
E
1
(x) =

e
x
x
dx and E
2
(x) =

e
x
x
2
dx.
Using integration by parts (letting u = e
x
, du = e
x
dx, dv = dx/x
2
, and v = 1/x),
E
2
(x) =

u dv = uv

v du =
e
x
x
+

e
x
x
dx = E
1
(x)
e
x
x
.
We now apply ( + 1)
1
to e
x
/x e
x
/x
2
, using (a) from the previous page and the
above identity to substitute for E
2
(x).
( + 1)
1

e
x
x

e
x
x
2

= x
1

x
0

e
x
x

e
x
x
2

dx = x
1

$
$
$
E
1
(x)
$
$
$
E
1
(x) +
e
x
x

,
demonstrating that we get the same result no matter in which order we apply the
operators.
The more common class of forcing function for problems like this is a polynomial. If we
apply (a) from the previous page to R = x
m
, we get
p (x) = ( +a)
1
x
m
= x
a

x
a+m1
dx =
x
m
a +m
for a +m = 0. (b)
When a + m = 0, you have p (x) = x
a
log x. The formula (b) also works for negative
and noninteger exponents. This formula is applicable to the remaining examples in this
case (5 through 8).
Page 317 ( 131. Linear equations with variable coecients) case 1, example
5: Equation becomes
( 1)( 2) + 2( 1) + 1 y = x +x
3
,

3
3
2
+
&
& 2 + 2
2

&
& 2 + 1

y = x +x
3
,

2
+ 1

y = x +x
3
,
( 1)
2
( + 1) y = x +x
3
.
Observe that in general, a repeated root, ( r)
n
, produces terms in the complementary
function of C
k
x
r
(log x)
k
for k = 0 to k = n1. So in this example, the complementary
function is C
0
x + C
1
xlog x + C
2
x
1
, which is equivalent to the books complementary
function. To nd the particular integral, we apply (b) on this page, and where necessary,
(a) on the previous page to
p (x) = ( 1)
2
( + 1)
1

x +x
3

,
= ( 1)
2

x
2
+
x
3
4

by (b),
= ( 1)
1

x
2
log x +
x
3
8

by (b),
=
x
2

log x
x
dx +
x
3
16
by (a) on the rst term and (b) on the second,
=
x
4
(log x)
2
+
x
3
16
, which is the books answer.
S-37
Page 317 ( 131. Linear equations with variable coecients) case 1, example
6: Equation becomes
( 1)( 2) + 2( 1) + 2 y = 10x +
10
x
,

3
3
2
+
&
& 2 + 2
2

&
& 2 + 2

y = 10x +
10
x
,

2
+ 2

y = 10x +
10
x
,
( + 1)

2
2 + 2

y = 10x +
10
x
.
The second factor above has no real roots. Its nonreal roots are = 1 . So the
complementary function is C
1
x
1
+ C
2
x
1+
+ C
3
x
1
. To turn the second and third
terms into real form,
C
2
x
1+
+ C
3
x
1
= C
2
xe
log x
+C
3
xe
log x
, and then by Eulers formula
= C
2
xcos (log x) + sin (log x) +C
3
xcos (log x) sin (log x) .
If you stipulate that C
2
and C
3
be complex complements of each other, then the above
is equal to C

2
xcos (log x) +C

3
xsin (log x), where both C

2
and C

3
are both real. Putting
the pieces together we nd that we now have the same complementary function as the
book. To nd the particular integral, the setup is
p (x) = ( + 1)
1
( 1 )
1
( 1 +)
1

10x +
10
x

.
Formula (b) on the previous page works even when the -factor is nonreal. Applying
the second and third factor to each of the forcing function terms using (b) yields
p (x) = ( + 1)
1

10x
() ()
+
10x
1
(2 ) (2 +)

= ( + 1)
1

10x + 2x
1

.
Finally using (b) to apply ( + 1)
1
to each of the terms that result,
p (x) = ( + 1)
1

10x + 2x
1

= 5x + 2x
1
log x.
Putting this together with the complementary function yields the books answer.
Page 317 ( 131. Linear equations with variable coecients) case 1, example
8: The third example 2 on page 316 uses the operator,

2
+ 3 + 2

= ( + 1) ( + 2).
So the setup for nding the particular integral when R = x
1
is
p (x) = ( + 1)
1
( + 2)
1
x
1
, and by (b),
= ( + 1)
1
x
1
2 1
= ( + 1)
1
x
1
. by (b) again,
= x
1
log x, which is the books answer.
Page 317 (Linear equations with variable coecients) case 2, example 1: A
better substitution than the one suggested in the book for this is z = a/b + x. This is
because all derivatives of y with respect to z are equal to the corresponding derivatives
with respect to x. Equation becomes
b
2
z
2
d
2
y
dz
2
+b
2
z
dy
dz
+c
2
y = 0

b
2
( 1) +b
2
+c
2

y =

b
2

2
c
2

y = 0.
By the methods given for case 1, the solution to this is C

1
sin (c/b log z)+C

2
cos (c/b log z).
See page 316, second example 2, which is worked on page S-35. Back-substitute for z to
obtain at C

1
sin (c/b log (a/b +x)) +C

2
cos (c/b log (a/b +x)), which is the same as
C

1
sin

c
b
log (a +bx)
c
b
log b

+ C

2
cos

c
b
log (a +bx)
c
b
log b

.
Use trig identities for sin(u v) and for cos (u v) to arrive at the books answer.
S-38
Page 317 ( 131. Linear equations with variable coecients) case 2, example
2: Substitute z = x +a. Equation becomes
z
2
d
2
y
dz
2
4z
dy
dz
+ 6y = x ( 1) 4 + 6 y = ( 2) ( 3) y = z a.
Clearly the complementary function for this is C
1
z
2
+C
2
z
3
= C
1
(x +a)
2
+C
2
(x +a)
3
.
The setup for nding the particular integral is
p (z) = ( 2)
1
( 3)
1
(z a) , which by (b) on S-37 is
= ( 2)
1

z
(3 + 1)

a
(3 + 0)

= ( 2)
1

z
2
+
a
3

; by (b) again,
=
z
2 (2 + 1)
+
a
3 (2 + 0)
=
z
2

a
6
=
x +a
2

a
6
=
x
2
+
a
3
,
which, when combined with the complementary function, is the books answer.
Page 318 ( 132. Exact equations) clarication: You might have found books
development of how to deal with higher order exact equations to be confusing. The easy
way to understand it is to work it backwards from the rst integral. If you begin with
X
0
d
2
y
dx
2
+ (X
1
X

0
)
dy
dx
+ (X
2
X

1
+X

0
) y =

Rdx +C,
then take its derivative using the product rule, you get
X
0
d
3
y
dx
3
+ X

0
d
2
y
dx
2
+
(X
1
X

0
)
d
2
y
dx
2
+ (X

1
X

0
)
dy
dx
+
(X
2
X

1
+X

0
)
dy
dx
+(X

2
X

1
+X

0
) y = R,
X
0
d
3
y
dx
3
+ X
1
d
2
y
dx
2
+ X
2
dy
dx
+ X
3
y = R,
which holds provided that X
3
= X

2
X

1
+X

0
. So you can see that by dierentiating
(5) on page 318, we have recovered (1) on page 317. At the same time we have also
evolved the exactness criterion shown in (4) of page 318. This method can easily be
extended to determine exactness and rst integrals of equations of order higher than 3.
Page 318 ( 132. Exact equations) example 2: We have X
3
= 2, X

2
= 4, X

1
= 2,
and X

0
= 0. Hence X
3
X

2
+X

1
X

0
= 2 4 +2 0 = 0, so the equation is exact.
By (5) on page 318, the rst integral is
x
d
2
y
dx
2
+

x
2
3 1

dy
dx
+ (4x 2x) y = C
1
.
If you allow the coecient functions of the above to be Y
0
= x, Y
1
= x
2
4, and Y
2
= 2x,
then Y
2
Y

1
+ Y

0
= 2x 2x = 0. So this new equation is also exact. We apply (5)
again to get
x
dy
dx
+

x
2
4 1

y = C
1
x +C
2

dy
dx
+

x
5
x

y = C
1
+
C
2
x
.
Using the method on page 294, the solution to the above is
y = x
5
e
x
2
/2

C
3
+

C
1
x
5
+C
2
x
6

e
x
2
/2
dx

.
The integral in the expression above cannot be expressed in elementary functions, but
can be described using something called the incomplete gamma function. You can nd
more about incomplete gamma in Handbook of Mathematical Functions by Abramowitz
and Stegun (U.S. Department of Commerce, National Bureau of Standards, Applied
Mathematics Series 55).
S-39
Page 318 ( 132. Exact equations) example 3: Example 1 has already demon-
strated that this equation is exact. So we apply (5) on page 318 to nd the rst integral.
x
5
d
2
y
dx
2
+ (15 5) x
4
dy
dx
+ (60 60 + 20) x
3
y = x
5
d
2
y
dx
2
+ 10x
4
dy
dx
+ 20x
3
y = e
x
+C.
If we let Y
0
= x
5
, Y
1
= 15x
4
, and Y
2
= 20x
3
, then Y
2
= Y

1
Y

0
, hence this equation is
also exact. So we apply (5) again to get the second integral.
x
5
dy
dx
+ (10 5) x
4
y = x
5
dy
dx
+ 5x
4
y = e
x
+Cx +C
2
.
This equation is also exact. Its left side is a perfect dierential.
d

x
5
y

dx
= e
x
+Cx +C
2
x
5
y = e
x
+C
1
x
2
+C
2
x +C
3
,
with C
1
=
1
2
C, which is the books answer.
Observe that if you divide the original equation by x
2
, the resulting equation is homo-
geneous in powers,
x
3
d
3
y
dx
3
+ 15x
2
d
2
y
dx
2
+ 60x
dy
dx
+ 60y =
e
x
x
2
,
and can be solved using the method from ' 131.
( 1) ( 2) + 15( 1) + 60 + 60 y =
e
x
x
2
,

3
+ 12
2
+ 47 + 60

y =
e
x
x
2
,
( + 5) ( + 4) ( + 3) y =
e
x
x
2
.
The complementary function is C
1
x
3
+C
2
x
4
+C
3
x
5
. To nd the particular integral,
we need to solve the following by repeated application of (a) on page S-36:
( + 5) ( + 4) ( + 3) p (x) =
e
x
x
2
,
( + 5) ( + 4) p (x) = x
3

x
2
e
x
x
2
dx =
e
x
x
3
,
( + 5) p (x) = x
4

x
3
e
x
x
3
dx =
e
x
x
4
,
p (x) = x
5

x
4
e
x
x
4
dx =
e
x
x
5
.
Combining this result with the complementary function, we get a solution for y that is
equivalent to the books answer.
Page 319 ( 132 Exact equations) Comments on Forsyths elimination of
terms: That x
m
d
n
y/dx
n
is a perfect dierential whenever m < n is proved by taking
the integral of this expression by parts:

x
m
d
n
y
dx
n
dx = x
m
d
n1
y
dx
n1
m

x
m1
d
n1
y
dx
n1
dx.
Note that to the right of the equal both the exponent and the order of the derivative of
the expression under the integral have been reduced by unity. So by repeating the process
over and over, we can see that the exponent reaches zero before the order of the derivative
does, provided m < n. At that point only one more integration is required, and since
we would be integrating only a derivative of y (not multiplied by any polynomial of x)
we would simply get the next lower derivative (or y itself if m = n1). An easy way to
justify eliminating such terms from the exactness criterion, (4) on page 318, is that the
order of the derivative wed have to take of X
k
would exceed its highest exponenent of
x, and so that derivative would be zero and contribute nothing to the exactness criterion
formula.
S-40
Page 320 ( 133. Equations with missing terms) example 1: Back-substituting
p with dy/dx into the texts result, p
2
y
2
= y
4
+C
4
,
y
2

dy
dx

2
= y
4
+C
4
hence y
dy
dx
=

y
4
+C
4
.
Now let u = y
2
, and
1
2
du = y dy, the above becomes
1
2
du
dx
=

u
2
+C
4
, which by separation of variables is
du

u
2
+C
4
= 2dx.
Integrating,
sinh
1

u
C
2

= 2x +C
2
hence
u
C
2
= sinh (2x +C
2
) .
Back-substituting y
2
for u, multiplying by C, and changing the symbol, C, to C
1
results
in the books answer.
Page 320 ( 133. Equations with missing terms) example 2: This equation is
linear with constant coecients and constant forcing function, and as such it can be
solved by methods developed in ' 129 and ' 130. Doing it using the method suggested
in the example, we divide through by and make the substitution, observing that
d
2
x
1
/dt
2
= d
2
x/dt
2
. Equation becomes
1

d
2
x
1
dt
2
+x
1
= 0 hence
d
2
x
1
dt
2
+x
1
= 0.
You can solve this by inspection: x
1
= C
1
cos

+ C
2
sin

. Back-substitute
x
1
= x / to arrive at the books answer.
Page 320 ( 133. Equations with missing terms) example 4: Let p = dy/dx.
Equation becomes
dp
dx
ap
2
= 0, which by separation of variables is
dp
p
2
= a dx.
Integrating then back-substituting,

1
p
= ax +C hence
dy
dx
=
1
ax +C
.
Integrating again gives y = (1/a) log (ax +C) +C

. Multiply both sides by a, then

take the exponential function of both sides to get,
e
ay
= e
aC

(ax +C) = C
1
x +C
2
where C
1
= ae
aC

and C
2
= e
aC

C,
which is the books answer.
Page 320 ( 133. Equations with missing terms) example 5: Substitute p for
dy/dx and p dp/dy for d
2
y/dx
2
, and the equation becomes
1 +p
2
= yp
dp
dy
, which by separating variables is
dy
y
=
p dp
1 +p
2
.
Integrating,
log y +C =
1
2
log

1 +p
2

hence C
1
y =

1 +p
2
where C
1
= e
C
.
Back-substitute dy/dx for p,
C
1
y =

1 +

dy
dx

2
hence

C
2
1
y
2
1 =
dy
dx
.
Separating variables and integrating,

dy

C
2
1
y
2
1
=

dx hence
cosh
1
(C
1
y)
C
1
= x +C
2
.
Letting a = C
1
and b = C
1
C
2
, and multiplying by a, then taking cosh of both sides,
then dividing both sides by a, the above results in the books answer.
S-41
Page 320 ( 133. Equations with missing terms) example 6: The books answer
is evident by inspection using methods from earlier sections. But doing it using the
method from this section, we substitute p for dV/dx and p dp/dV for d
2
V/dx
2
. Equation
becomes
p
dp
dV

2
V = 0 or equivalently p dp =
2
V dV.
Integrating and cancelling the common factor of
1
2
,
p
2
=
2
V
2
+C hence p =

V
2
+C

where C

=
C

.
Back-substituting and separating variables,
dV
dx
=

V
2
+C

hence
dV

V
2
+C

= dx.
Integrating,
sinh
1

= x +C

hence V =

sinh(x +C

) .
Using the exponential identity to expand sinh,
V =
1
2

e
x+C

e
xC

.
Let C
1
=
1
2

e
C

and C
2
=
1
2

e
C

to obtain the books answer. The question

then arises, can you choose C

and C

in such a way as to arrive at an arbitrary C

1
and
C
2
? This would be true provided that the ratio of e
C

to e
C

can be made to be any

desired value, positive or negative, by proper choice of C

. So if for arbitrary A,
e
C

e
C

= e
2C

= A, then C

=
log A
2
.
If we allow nonreal values for C

(such as C

=
1
2
log 2 +
1
2
for A = 2 for example)
then by Eulers formula, we can solve this given any A. Also by allowing C

to be
imaginary when desired, we can satisfy the cases where we would want

to be
negative.
Page 322 ( 133. Equations with missing terms) example 5: Letting p = dy/dx,
the equation becomes
x
dp
dx
= 1, or equivalently

dp =

dx
x
.
Integrating yields p = log x +C. Back-subsituting for p,
dy
dx
= log x +C, or equivalently

dy =

log x +C dx.
By integrating the above and letting C
1
= C 1 you get the books answer.
Page 335 ( 135. Chemical Reaction Rates) solving (16): The equation to be
solved is
dz
dy
=
a z
z y
, or equivalently
dy
dz
=
z y
a z
or
dy
dz
+
y
a z
=
z
a z
.
The last version is in the form given by (1) of ' 122 (page 296). Hence the integrating
factor for this is = e
log(az)
=
1
az
. The exact equation is
dy
a z
+
y z
(a z)
2
dz = 0.
Integrating the second term dz yields the general solution,
y
a z

a
a z
log (a z) = C.
Substituting the conditions on page 335, a = 1, y = 0, and z = 0, into the above forces
C = 1. The solution, (17) on page 335, follows from that.
S-42
Page 335 ( 135. Chemical Reaction Rates) solving (18): The equation to be
solved is
dz
dx
=
a z
y x
, or equivalently
dx
dz
=
y x
a z
or
dy
dz
+
x
a z
=
y
a z
,
where y = z + (a z) log (a z). Like the last equation, this is rst-order linear, with
the integrating factor being = e
log(az)
=
1
az
. Multiplying by the integrating factor
and substituting for y,
dx
a z
+
xdz
(a z)
2
=
z + (a z) log (a z)
(a z)
2
dz,
=

z
(a z)
2
+
log (a z)
a z

dz,
which is exact. Integrating yields
x
a z
= log (a z) +
a
a z

1
2
log (a z)
2
+C.
Using the analogous initial conditions as were stipulated for (16), that is a = 1, x = 0,
and z = 0, we nd again that C = 1. Substitute 1 for C, then multiply the above
through by a z and take the cancellation to arrive at the books answer.
Page 336 ( 135. Chemical Reaction Rates) example: Taking the quotient of
(14) and (15) on page 335 gives
dy
dz
=
k
2
k
3
z y
a z
=
z y
a z
where =
k
2
k
3
.
Once again this equation is rst-order linear. The integrating factor is
= e
log(az)
=
1
(az)
. So the exact equation is
dy
(a z)

+
y dz
(a z)
+1
=
z dz
(a z)
+1
.
Integrating,
y
(a z)

z dz
(a z)
+1
.
To integrate the expression to the right of the equal, let u = a z, a u = z, and
du = dz. Integral becomes

a u
u
+1
du =

a
u
+1
du +

du
u

=
a
u

1
1
u
1
+C.
Back-substituting and multiplying through by (a z)

,
y = a

1
(a z) +C (a z)

=
a
1
+

1
z +C (a z)

.
Applying the same conditions as indicated on page 335 of a = 1, z = 0, and y = 0, we
nd C =
1
1
.
To nd the equation for x in terms of z, we let = k
1
/k
3
and solve it the same way as
we did for y, except that replaces .
dx
dz
=
y x
a z
.
Replace y with the solution we got above for y. The integrating factor will be =
1
(az)
.
You get
x
(a z)

a
1
+

1
z +C (a z)

(a z)
+1
dz +C

.
Taking of the integral above is left to the reader.
S-43
Page 336 ( 136. Simultaneous equations) example 1: This one can also be done
using the Laplace method (see supplement pages S-31 through S-34). The equations are
Dx +ay = 0, and Dy +bx = 0.
In matrix form we have
x y

D a
b D

.
The determinant of this matrix is D
2
ab =

ab

D +

ab

. Because there
are no initial conditions given and forcing functions are zero, we can determine only the
complementary function from this system of equations. That function is determined by
the roots of the determinant. If we apply these roots to x, the complementary function
is x = C
1
e

abt
+ C
2
e

abt
. Substitute the derivative of this for dx/dt into the rst
equation to solve for ys complementary function.
Page 337 ( 136. Simultaneous equations) example 2: Doing it using the books
method, we take the derivative of the rst equation,
d
2
x
dt
2
+
dy
dt
= 3
dx
dt
.
Subtracting the rst equation from this gives
d
2
x
dt
2

dx
dt
+
dy
dt
y = 3
dx
dt
3x.
Subtracting the second from the above gives
d
2
x
dt
2

dx
dt
= 3
dx
dt
4x.
As indicated in the text, this eliminates y and its derivative and we are left with

D
2
4D + 4

x = (D2)
2
x = 0. This gives a complementary function for x of
C
1
e
2t
+ C
2
te
2t
. Taking the derivative of this gives dx/dt = (2C
1
+C
2
) e
2t
+ 2C
2
te
2t
.
Substituting both of these into the rst equation,
(2C
1
+C
2
) e
2t
+ 2C
2
te
2t
+y = 3

C
1
e
2t
+C
2
te
2t

.
Solve for y to get the books complementary function for y.
Using the Laplace method we subtract 3x from both sides of the rst equation and
x from both sides of the second. This puts the equation in standard form with all
dependent variables to the left of the equal. Forming the matrix from those equations
yields
x y

D 3 1
1 D 1

.
The determinant of this matrix is (D3) (D 1) + 1 = D
2
4D + 4. You can see
how the same dierential operator results from this method as from the books method.
Apply this operator to x to get its complementary function, then use the substitution
method shown above to nd ys complementary function from that of x.
Page 337 ( 136. Simultaneous equations) example 3: First using the books
method.
d
2
x
dt
2
=
dy
dt
, derivative of 1st equation

2
x =
dy
dt
2nd equation times , but reversed
d
2
x
dt
2
+
2
x = 0 dierence eliminates y.
S-44
Which is the same as

D
2
+
2

x = 0. Observe the same operator evolves from using

the matrix method. We rewrite the equations as dx/dt x = 0 and dy/dt + x = 0.
The matrix is
x y

D
D

,
whose determinant is D
2
+
2
. If we ascribe to x the complementary solution,
x = C
1
cos t + C
2
sin t, then dx/dt = C
1
sin t + C
2
cos t. Substituting into
the rst equation,
C
1
sint +C
2
cos t = y,
hence y = C
1
sin t + C
2
cos t, and the relationships between the x-coecients and
the y-coecients are as the book indicates.
Page 337 ( 136. Simultaneous equations) example 5: In the case where there
are more than two dependent variables, the matrix method shows its superiority to the
dierentiate-and-substitute method. All three equations, as given, are in standard form
that is all of the dependent variables are to the left of the equal. The matrix, then, is
x y z

D b c
a
1
D c
1
a
2
b
2
D

,
whose determinant is D
3
(a
1
b +a
2
c +b
2
c
1
) D+a
1
b
2
c +a
2
bc
1
. This cubic is the same
as the one shown in the book for this example, except using a dierent independent
variable. Note that the book carelessly reused the symbol, z, as an independent variable
for the cubic after it had already used it as one of the dependent variables in the original
equation set. The two usages are not equivalent, and the book should have chosen a
dierent symbol for the cubic to avoid confusion.
Page 337 ( 136. Simultaneous equations) example 6: In a modern electrical en-
gineering class, this problem would almost certainly be solved using the Laplace method.
We shall do it that way here. The matrix is
i
1
i
2

MD L
2
D +R
2
L
1
D +R
1
MD

,
the determinant of which is
M
2
D
2
(L
1
D +R
1
) (L
2
D +R
2
)
=

M
2
L
1
L
2

D
2
(L
1
R
2
+L
2
R
1
) D R
1
R
2
.
Observe that the above quadratic is equivalent to the one given on page 338. The
remainder of the problem that is nding the particular solutions given that E
1
and
E
2
are constants is shown in the book.
Page 339 ( 136. Simultaneous equations, variable coecients) example 3: To
make the second solution work out, you have to multiply the equations by an integrating
factor of x.
xdx
x
2
y
2
z
2
=
dy
2y
=
dz
2z
.
Hence P =

x
2
y
2
z
2

/x, Q = 2y, and R = 2z. Multiplying these respectively by

x, y, and z, and then taking the sum yields x
2
+y
2
+z
2
. Hence by (3) on page 339,
xdx
x
2
y
2
z
2
=
dy
2y
=
dz
2z
=
xdx +y dy +z dz
x
2
+y
2
+z
2
.
S-45
The last expression is a perfect dierential of
1
2
du/u, where u = x
2
+ y
2
+ z
2
. Setting
this equal to either of latter two expressions in the modied original equations (well
choose dz/2z) and integrating yields
1
2
log

x
2
+y
2
+z
2

=
1
2
log z +C,
from which the books answer follows. Observe that you could have chosen dy/2y and
had
1
2
log

x
2
+y
2
+z
2

=
1
2
log y +C,
which would also be correct.
Supplemental problem: If you simplify example 3 of page 339 by dropping z, you
have the dierential equation,
dx
x
2
y
2

dy
2xy
= 0. (a)
Based upon the solution for example 3 of page 339, we can infer that the solution to
(a) is x
2
+ y
2
= Cy. Note that (a) is not exact. The problem then is, based upon the
solution given here, to infer an integrating factor, (x, y), that makes (a) become exact.
First we work the solution to (a) into a form in which taking the derivative eliminates
the undetermined constant, C.
log

x
2
+y
2

= log y + log C, whose derivative is

2xdx + 2y dy
x
2
+y
2
=
dy
y
,
Rearranging the derivative,
2xdx
x
2
+y
2
+

2y
x
2
+y
2

1
y

dy = 0,
which is exact because we arrived at it by taking the derivative of (a)s solution. Sum-
ming the dy-fractions over a common denominator,
2xdx
x
2
+y
2
+
y
2
x
2
y (x
2
+y
2
)
dy = 0,
which remains exact. In this form, it becomes clear how to form the integrating factor.
To go from (a) to the above, you would have to divide by x
2
+ y
2
, and multiply by
2x

x
2
y
2

. Hence
(x, y) =
2x

x
2
y
2

x
2
+y
2
.
S-46
Page 340, apparent error in book: If u = f

ax
3
+by
3

, then applying the chain

rule to take the partial derivatives of u with respect to x and y results in
u
x
= 3ax
2
f

ax
3
+by
3

and
u
y
= 3by
2
f

ax
3
+by
3

,
which diers from the partial derivatives shown in the book for this function. The books
derivatives appear to be incorrect. Multiplying the partial (above) with respect to x by
by
2
and the partial with respect to y by ax
2
, we nd that the right-hand side of both
equations become equal. Hence
by
2
u
x
ax
2
u
y
= 0,
which also diers from the books conclusion on page 340.
Page 341 (partial dierential equations), example 1: Taking the partial deriva-
tives with respect to x and y respectively,
b
u
x
= b f

(bx ay) and 1 b

u
y
= af

(bx ay) .
Substituting for f

(bx ay) from the rst equation into the second,

1 b
u
y
= a
u
x
, hence a
u
x
+b
u
y
= 1.
Page 341 (partial dierential equations), example 2: Taking the partial deriva-
tives with respect to x and y respectively,

1
z
2
z
x
+
1
x
2
=
1
x
2
f

1
y

1
x

and
1
z
2
z
y
=
1
y
2
f

1
y

1
x

.
Using the second equation to substitute into the rst for f

1
y

1
x

1
z
2
z
x
+
1
x
2
=
y
2
x
2
z
2
z
y
, hence x
2
z
x
+z
2
= y
2
z
y
,
from which the books answer follows.
Page 341 (partial dierential equations), example 3: This one is trivial compared
with the preceding two examples. Both partial derivatives (with respect to x and with
respect to y) are identically equal to a

(x +y), and therefore equal to each other. So

clearly the dierence between the two partials must be zero, as asserted by the books
answer.
Page 344 (partial dierential equations), example 3: The equation to solve is
a

z
x
+
z
y

= z or equivalently
a
z
z
x
+
a
z
z
y
= 1.
Make the substitutions, Z = log z, X = x/a, and Y = y/a. Then
a
z
z
x
=
Z
X
and
a
z
z
x
=
Z
Y
; equation becomes
Z
X
+
Z
Y
= 1.
In this case we have f (, ) = + 1 based upon the modied equation. Hence
for f (, ) = 0, we have = 1 . The solution to the modied equation then is
Z = X + (1 ) Y +C. Converting back to the original variables,
Z = log z =
x
a
+ (1 )
y
a
+C, hence z = Ce
1
a
{x+(1)y}
.
S-47
Page 345 (partial dierential equations), example 4: The equation to be solved
is
x
2

z
x

2
+y
2

z
y

= z
2
or equivalently

x
z
z
x

2
+

y
z
z
y

2
= 1.
If Z = log z, then dZ = dz/z, hence that substitution yields

x
Z
x

2
+

y
Z
y

2
= 1.
Similarly substituting X = log x and Y = log y, the equation becomes

Z
X

2
+

Z
Y

2
= 1.
The auxillary function for this modied equation is f (a, b) = a
2
+ b
2
1 = 0. So
b =

1 a
2
. By (1) on page 344, the solution is
Z = aX +

1 a
2
Y +C hence log z = a log x +

1 a
2
log y +C.
Taking the antilog of the latter equation results in the books answer.
Page 345 (partial dierential equations, type II), example 1: The book has
denite errors on this problem. The equation,

a
2
+z dz/dx = 2, is correct. From (2)
on page 345, the solution is
x +ay +C =
1
2

a
2
+z dz =
1
3

a
2
+z
3
2
.
Squaring both sides results in the solution,
(x +ay +C)
2
=
1
9

a
2
+z

3
, (b)
which diers from the books answer (because the book misplaced a scalar). To show
that this solution is correct, we substitute it back into the equation. Taking /x
of (b),
2 (x +ay +C) =
1
3

a
2
+z

2 z
x
, hence 6
x +ay +C
(a
2
+z)
2
=
z
x
.
By design, z/y = a z/x, which is easily conrmed by taking /y of (b). Hence

z
x

2
z +

z
y

2
=

z
x

z +a
2

= 36

x +ay +C
(a
2
+z)
2

z +a
2

.
Substituting from (b) for (x +ay +C)
2
,
36
1
9

a
2
+z

3
(a
2
+z)
4

z +a
2

= 4,
which veries that the solution is correct. Indeed if you replace
1
9
in (b) with any other
value, the above test fails.
Page 345 (partial dierential equations, type II), example 2: Replacing q with
bp, the equation becomes p

1 +b
2
p
2

= bp (z a), which simplies to

1 + b
2
p
2
= b (z a). Solving for p we have p = (z) =
1
b

bz ab 1. By (2) on
page 345, the solution is
x +by +C =

dz
(z)
= b

dz

bz ab 1
= 2

bz ab 1.
The books answer follows from this. From that answer its easy to isolate z as a function
of x and y.
z =
1
4b
(x +by +C)
2
+a +
1
b
, hence
z
x
=
1
2b
(x +by +C) .
S-48
Substituting these results into 1 +b
2
p
2
= b (z a), we have
1 +b
2

z
x

2
= 1 +
1
4
(x +by +C)
2
and b (z a) =
1
4
(x +by + C)
2
+ 1,
which veries that the solution is correct.
Page 345 (partial dierential equations, type III), example 2: Rearranging this
we have p
2
x = y q
2
= a. Now separating, it becomes p
2
= x + a and q
2
= y a.
By (3) on page 345, the integral is
z =

x +a dx +

y a dy.
The books answer follows from taking the integrals. Observe that you could also have
separated it into p
2
= x a and q
2
= y + a and obtained a slightly dierent solution.
Explain how it is that these two solutions are, in fact, equivalent.
Page 345 (partial dierential equations, type III), example 3: Rearranging we
have q/y = 2p
2
= 2a
2
. Notice that we set this to 2a
2
rather than a in order to avoid
a radical in the solution. You could set it to a and arrive at an equivalent solution.
Separating it becomes q = 2a
2
y and p = a. By (3) on page 345,
z =

a dx +

2a
2
y dy = ax +a
2
y
2
+C,
which is the books answer.
Page 346 (equations analogous to Clairaults), how to nd singular solutions:
The book oers no clue on how to do this. In this supplement beginning on page S-17,
we developed the method that can be found in Ordinary Dierential Equations and their
Solutions by George M. Murphy (van Nostrand, 1960) for nding singular solutions to
ordinary dierential equations. That same method can be extended to apply to partial
dierential equations such as the ones found at the top of page 346. Murphy suggests
that if p = dy/dx, then we take /p of the entire equation, solve for p, then substitute
that back into the original equation to obtain the singular solution. We now have the
situation where p = z/x and q = z/y. The extended method is to take both /p
and /q of the original equation, solve for p and q simultaneously, then substitute
those solutions back into the original equations. Here is that method applied to the rst
example on page 346:

p
(z = px + qy +pq) gives 0 = x +q, and

q
(z = px + qy +pq) gives 0 = y +p.
Clearly p = y and q = x. Substituting for p and q into the original equation, we
have z = xy xy +xy = xy, which is the singular solution given in the book.
Applying the same method to the third example,

z = px +qy n(pq)
1
n

gives 0 = x q (pq)
1
n
1
, and

z = px +qy n(pq)
1
n

gives 0 = y p (pq)
1
n
1
.
Solving for q in the rst equation, q = x
n
p
n1
. Substituting q into the second equation,
y = p

x
n
p
n1

1
n
1
= p (x
n
p
n
)
1
n
1
= p (xp)
1n
, hence p = y
1
2n
x
n1
2n
.
By symmetry we also have q = x
1
2n
y
n1
2n
, and therefore pq = (xy)
n
2n
. The books
singular solution follows by substituting these into the original equation.
S-49
The second example on page 346 is by far the most dicult on which to apply this
method, but its singular solution also emerges in the same way.

z = px +qy +r

1 +p
2
+q
2

gives 0 = x +
rp

1 +p
2
+q
2
, (a)

z = px +qy +r

1 +p
2
+q
2

gives 0 = y +
rq

1 +p
2
+q
2
. (b)
From (a) we have
x
2
=
r
2
p
2
1 +p
2
+q
2
, hence x
2
+ p
2
x
2
+q
2
x
2
= r
2
p
2
and x
2
+q
2
x
2
=

r
2
x
2

p
2
.
Solving for p
2
,
p
2
=
x
2

1 +q
2

r
2
x
2
; substituting p
2
into (b), y
2
=
r
2
q
2
1 +
x
2
(1+q
2
)
r
2
x
2
+q
2
.
Now multiply top and bottom of the nasty fraction by r
2
x
2
, expand the denominator,
and then take the cancellations:
y
2
=
r
2
q
2

r
2
x
2

r
2
&
&
x
2
+&
&
x
2
+

x
2
q
2
+q
2
r
2

q
2
x
2
=
q
2

r
2
x
2

1 +q
2
,
from which
y
2

1 +q
2

= q
2

r
2
x
2

, hence y
2
= q
2

r
2
x
2
y
2

.
We easily nd q from this, and by the symmetry of (a) and (b) we can nd p as well:
q =
y

r
2
x
2
y
2
and p =
x

r
2
x
2
y
2
.
When these expressions are substituted for p and q in the original equation, we have
(choosing the minus of the ):
z =
x
2
+y
2

r
2
x
2
y
2
+r

1 +
x
2
+y
2
r
2
x
2
y
2
.
Multiplying by

r
2
x
2
y
2
and taking the cancellations,
z

r
2
x
2
y
2
= x
2
y
2
+r
2
,
from which the books singular solutions follows.
Page 348 ( 141. Linear Partial Equations with Constant Coecients) exam-
ple 2: The books solution is in error. The operator equation is

D
2
4DD

+ 4D
2

z = 0.
This factors into (D2D

)
2
z = 0. If you substitute f (mx +y) into the equation, the
auxillary equation is m
2
4m + 4 = (m2)
2
= 0. Hence m = 2. By (9) on page 348,
the solution is f (2x +y). Substituting this into the original equation veries it as a
solution, whereas the books answer fails to verify. Furthermore, the auxillary equation
has a double root. We cant just say that z = f
1
(2x +y) + f
2
(2x +y) is the general
solution, since the f
1
and f
2
terms are equivalent. Equation (11) on page 349 shows
how to deal with repeated roots. Accordingly the complete solution is
z = f
1
(2x +y) +xf
2
(2x +y) .
If you think about it, there is no reason that yf
3
(2x +y) shouldnt also be a solution.
And indeed, if you substitute it into the original equation, you will see that it is. So
why dont we include it as a third term in the general solution? Because it is implicit
in the existing two terms of that general solution. Consider that (2x +y) f (2x +y) is
a single function of 2x + y, and is therefore a solution by being equivalent to the rst
term of the general solution. But the 2xf (2x +y) part of this is also a solution by being
equivalent to the second term of the general solution. As such it becomes zero when
S-50
the equations operator is applied to it. That means that the yf (2x +y) part must also
become zero when the operator is applied to it, so it must also be a solution. Hence its
being a solution is implied by the general solution given above. So yf (2x +y), though
a solution, is not independent of the general solution given above.
Page 349 ( 141. Linear Partial Equations with Constant Coecients) ex-
ample 1: Again the book gets the sign wrong. The auxillary equation factors into
(m + 1)
2
. The double root is m = 1. The solution, z = f
1
(x y) +xf
2
(x y) follows
from (11) on page 349. Observe that z = f
1
(x y) +yf
2
(x y) would also have been
a suitable general solution.
Page 349 ( 141. Linear Partial Equations with Constant Coecients) exam-
ple 2: The answer given in the book implies an auxillary equation of (m + 1)
2
(m1) = 0.
This expands into m
3
+ m
2
m 1 = 0. The partial dierential equation that goes
with this auxillary equation is

D
3
+D
2
D 1

z = 0. You will nd that the books

solution solves this equation and fails to solve the equation given in the example (al-
though the f (x +y) term by itself does solve the books equation, as m = 1 is a root of
m
3
3m
2
+m+ 1 = 0). The books equation factors into

m1

m1 +

m1

= 0.
Hence the general solution to the books equation (which has no repeated roots) is
z = f
1
(x +y) +f
2

x +y

+f
3

1 +

x +y

.
Page 351 ( 141. Non-homogeneous linear partial equations) example 3:
Using the trial solution, z = e
x+y
results in the auxillary equation,
+a +b +ab = 0, which factors into ( +b) ( +a) = 0.
The books solution follows from the paragraph following (12) on page 349.
Page 351 ( 141. Non-homogeneous linear partial equations) example 4:
The equation factors into
(D +D

1) (DD

+ 2) z = 0.
Again the books solution follows from the paragraph following (12) on page 349.
Page 351 ( 141. Non-homogeneous non-factorable linear partial equa-
tions) supplemental example: Some equations cannot be factored at all. This ex-
ample is not in the book:

2
z
x
2


2
z
y
2
+z = 0.
The trial solution, z = e
x+y
yields the auxillary equation,
2

2
+ 1 = 0, or
equivalently, =

2
+ 1. Substituting for and inserting the undetermined scalars
for each possible , we have a solution of
z =

2
+1 x+y
according to (15) on page 350. We can modify this solution by allowing to assume
purely imaginary values. So if = , then for 1,
z =

sin

2
1 x +y +

is also a solution in accordance with Eulers formula.

Example 1 on page 351 is also not factorable. Here the prototype solution was
z = e
x+
2
y
. This one becomes of special interest when you allow to assume the
purely imaginary value of .
S-51
Its general solution becomes
z =

2
y
sin (x +

) .
If you replace y with kt, where k is a heat conductivity rate, then this equation represents See
S-67 for
more on
Fouriers
heat
equation
and its
solution.
the temperature over time of a long uniform bar. Observe that if the temperature at
t = 0 varies sinusoidally over the length of the bar, then the magnitude of that sinusoid
decays in time at a rate in proportion to the square of its spacial frequency, . Diusion
of uids in a long narrow tube complies with this same equation.
Page 352 (Particular integrals of homogeneous partial equations) rst exam-
ple 2: The problem is to solve

D
2
+ 3DD

D
2

1
(x +y) = (D + 2D

)
1
(D +D

)
1
(x +y) . (a)
Applying the right-hand factor rst, we follow the procedure and let m = 1. Subtract-
ing mx from y of the forcing function gives x+y +x = 2x+y. Integrating that dx gives
x
2
+ xy. Now adding mx = x to y of this result gives x
2
+x(y x) = xy. So at this
point weve reduced the problem to nding
(D + 2D

)
1
(xy) .
We repeat the same procedure, but this time with m = 2. Subtracting mx = 2x from
y gives x(y + 2x) = xy + 2x
2
. Integrating that gives
1
2
x
2
y +
2
3
x
3
. Now add mx = 2x
to y of this result to get
1
2
x
2
(y 2x) +
2
3
x
3
=
1
2
x
2
y x
3
+
2
3
x
3
, which is equal to the
answer in the book.
Alternatively we could have integrated (a) starting with the left-hand factor rst. Then
we have m = 2. Subtracting mx = 2x from y in (a) gives x + y + 2x = 3x + y.
Integrating dx gives
3
2
x
2
+xy. Now adding mx to y we have
3
2
x
3
+x(y 2x) = xy
1
2
x
2
.
At this point weve reduced the problem to nding
(D +D

)
1

xy
1
2
x
2

.
We repeat the procedure, this time with m = 1. So subtract mx = x from y to get
x(y +x)
1
2
x
2
= xy +
1
2
x
2
. Integrating this dx gives
1
2
x
2
y +
1
6
x
3
. Add mx = x to
y in this result gives
1
2
x
2
(y x) +
1
6
x
3
=
1
2
x
2
y
1
2
x
3
+
1
6
x
3
, which is also equal to the
books answer.
Page 352 (Particular integrals of homogeneous partial equations) second ex-
ample 2: Here F (D, D

) = D
2
+ 5DD

+ 6D
2
, which has two factors, hence we will
integrate twice. Dividing by D
2
gives (D

/D) = 1 +5D

/D+6D
2
/D
2
. We also have
b/a =
1
2
. So (b/a) = 1 +
5
2
+
6
4
= 5. So the particular integral will be
p (x, y) =
1
5

dx
2
y + 2x
,
=
1
10

log (y + 2x) dx,

=
1
20
(y + 2x) log (y + 2x) y 2x .
S-52
The 1902 edition of the book originally gave this example as

D
2
+ 5DD

+ 6D
2

z =
1
(y 2x)
.
In this case b/a is a root of (b/a). So the problem falls into the category explained at
the top of page 353. Factoring this one, it becomes
(D + 2D

) (D + 3D

) z =
1
y 2x
,
hence p (x, y) = (D + 2D

)
1
(D + 3D

)
1

1
y 2x

.
The factor that has the problem with b/a =
1
2
is the left-hand factor. So we integrate
just once using the right-hand factor. For that factor alone we have (b/a) = 1 +
3b
a
hence (b/a) = 1
3
2
=
1
2
. So we take
p (x, y) = (D + 2DD

)
1

dx
y 2x

,
= (D + 2DD

)
1
log (y 2x) .
Now applying (18) on page 353 to the vanishing factor, we have p (x, y) = xlog (y 2x).
You are encouraged to substitute this expression for p into

D
2
+ 5DD

+ 6D
2

p and
apply the operator to conrm that you get back 1/ (y 2x).
Page 353 (Particular integrals, case 3) rst example 2: As the book suggests, we
solve (D +DD

2D

) z = sin (x y) and (D +DD

2D

) z = sin (x +y) separately,

and then by the principle of superposition (see page S-24) we add the results to get
the composite solution. Instead of using the method given on page 353, which, in this
example, is prone to mistakes, we apply the method of undetermined coecients (also
introduced on S-24). In the case of the forcing function of sin (x y), the particular
integral must be a linear combination of sin (x y) and cos (x y). Hence
(D +DD

2D

)
c
cos (x y) +
s
sin (x y) = sin (x y) .
Applying the operator to the expression in the curly-brackets,

c
sin (x y) + cos (x y)) 2 sin(x y) +

s
cos (x y) + sin (x y) + 2 cos (x y) = sin(x y) .
Separating terms and dividing out the sines and cosines yields
3
c
+
s
= 1,

c
+ 3
s
= 0,
hence
c
=
3
10
and
s
=
1
10
.
Likewise we solve for the particular integral of sin(x +y) by solving
(D +DD

2D

)
c
cos (x +y) +
s
sin (x +y) = sin (x +y) .
Applying the operator to the expression in the curly-brackets,

c
sin (x +y) cos (x +y)) + 2 sin(x +y) +

s
cos (x +y) sin (x +y) 2 cos (x +y) = sin(x y) .
Separating terms and dividing out the sines and cosines yields

s
= 1,

s
= 0.
Hence
c
=
1
2
and
s
=
1
2
. So the complete particular integral is
p (x, y) =
3
10
cos (x y) +
1
10
sin (x y) +
1
2
cos (x +y)
1
2
sin (x +y) ,
which diers from the books answer. And by the way, did you try to nd the complemen-
S-53
tary solution to this one? If z = e
x+y
, then the auxillary equation is + 2 = 0.
A few more algebraic steps gets you to
z =

e
2x+(1+)y

1+
.
Now that we have solved the more dicult problem, Ill reveal that I believe that the
typographical error in the book was not in the answer given, but in the original equation,
which should have been

D
2
+DD

2D
2

z = sin (x y)+sin(x +y). The particular

integral for this one is easily solved using the method in case 3 of page 353.
1
D
2
+DD

2D
2
sin (x y) =
1
1 1 + 2
sin (x y) .
Observe that by using the notation, we solve both cases at the same time. In the
case of the , we immediately see that we get
1
2
sin (x y). In the + case, the
denominator vanishes. But the operator factors into
D +DD

2D
2
= (D + 2D

) (DD

) .
So by case 2 and (18) on page 353, with a = 1 and b = 1, we have
(D + 2D

)
1
(D D

)
1
sin (x +y) =
1
1 + 2
x

sin(x +y) dx =
1
3
xcos (x +y) .
Combining the two results still diers from the books answer by the sign of the co-
sine term. You are encouraged to apply the original operator to the expression above
to see which is the right answer. The complementary solution to this equation is
f
1
(2x y) +f
2
(x +y) in accordance with (9) on page 348. So the complete solution is
z = f
1
(2x y) +f
2
(x +y) +
1
2
sin (x y)
1
3
xcos (x +y) .
Page 353 (Particular integrals, case 4) second example 1: We have
a = 2 and b = 3. Accordingly if F (D, D

) = D
2
DD

2D
2
+ 2D + 2D, then
F (a, b) = a
2
ab 2b
2
+2a+2b = 4 6 18 +4 +6 = 10. The books answer follows.
Page 353 (Particular integrals, case 4) second example 2: We have
F (n, m) = nm+an +bm+ab = (n +b) (m+a). The books answer follows.
Page 354 (Particular integrals, case 5) second example 2: First divide out 3D
from the operator:
1
3D

D
3

D
2
3D
1 +
D

1
xy =
1
3D

D
3

D
2
3D
+
D

1
xy.
No sooner does the book state, The expansion is not usually carried higher than the
highest power . . . in f (x, y), than it gives an example that requires carrying out the
expansion higher than that. In this case we need to take the expansion out to the cubed
term. The reason is the D that appears in the denominator of terms that will be in the
expansion. You have to use your careful judgement on this. Look at how the powers
expand. Any D or D

to the squared or higher power will cause xy to become zero.

Hence the D
2
/3D term in the above contributes nothing to the solution and can be
ignored throughout your analysis. There is another very sneaky aspect to this problem
that I will show as we analyze the expansion. The expansion out to the cubed term is
then

1
3D

1 +

D
3

D
2
3D
+
D

D
3

D
2
3D
+
D

2
+

D
3

D
2
3D
+
D

3
+. . .

xy.
Going through term by term, we see that

1
3D
xy contributes
1
6
x
2
y.
S-54
Next, recalling that the D
2
/D contributes nothing, we have,

1
3D

D
3

D
2
3D
+
D

xy which contributes
1
9
xy
1
18
x
3
.
Next, again ignoring the D
2
/D term,

D
3

D
2
3D
+
D

2
expands to
D
2
9
+
2D

3
+
D
2
D
2
,
Only the middle term has powers of D or D

less than squared, so it is the only term

that contributes. Hence

1
3D
2D

3
xy contributes
1
9
x
2
.
Finally, again ignoring D
2
/D,

D
3

D
2
3D
+
D

3
expands to
D
3
27
+
DD

3
+
D
2
D
+
D
3
D
3
,
of which only the DD

/3 term will make any contribution. This expansion is in accor-

dance with the binomial theorem, but here is where there is a trap for you. The binomial
theorem assumes multiplication to be commutative. But in this case that assumption
leads to a wrong answer. Letting a = D/3 and b = D

/D, we have,
(a +b) (a +b) (a +b) = a
3
+aab +abb +aba +baa +bab +bba +b
3
.
But notice that if you apply aa to xy before applying b to it, you get zero. The order
of application here does matter. Hence only two out of the three terms containing two
as and one b contribute anything. The one that applies both as before applying the b
contributes nothing. So the real contribution of the cubed expansion is

1
3D
2DD

9
xy which contributes
2
27
x.
You should conrm for yourself that expanding powers higher than cubed leads to all
of the terms containing a factor of D
n
or D
n
where n 2. So no higher powers are
needed. Gathering all of the contributions shown above and summing them yields the
books answer. You are encouraged to substitute that answer back into the original
equation to conrm that it does work.
Note that you could have started by dividing by 3D

rather than 3D. Using the

same procedure as above, you would have arrived at
1
6
xy
2
+
1
18
y
3
+
1
9
xy+
1
9
y
2
+
2
27
y. This
solution also works. If p is the books answer and q is this answer, then Cp + (1 C) q
also solves the particular integral.
Page 355 (nonconstant coecients) rst example 2: Making the substitution,
the equation becomes

2
z
u
2

z
u
+ 2

2
z
uv
+

2
z
v
2

z
v
= 0.
Letting D and D

be partial derivatives with respect to u and v respectively, we have

D
2
D + 2DD

+D
2
D

z = 0.
The auxillary equation is

2
+ 2 +
2
= ( +)
2
( +) = 0.
Solving we nd that either + = 0 or + = 1. The rst root leads to the
solution, z = g
1
(v u), in accordance with (9) on page 348. The second root leads
to the solution, z = e
u
g
2
(v u) in accordance with B on page 349. So the complete
solution is
z = g
1
(v u) +e
u
g
2
(v u) .
Back-substituting x = e
u
and y = e
v
, and allowing f
k
(s) = g
k
(e
s
) for both functions
yields the books answer.
S-55
Page 355 (nonconstant coecients) example 3: Making the suggested substitu-
tion, we have
v =
z
x
and

2
z
xy
=
dv
dy
.
Equation becomes
(x +y)
dv
dy
av = 0.
Separating variables
a
dy
x +y
=
dv
v
.
Integrating gives
a log (x +y) +g (x) = log v or equivalently f
1
(x) (x +y)
a
= v,
where f
1
= e
g
. Back-substitute for v and we have
f
1
(x) (x +y)
a
=
z
x
.
Integrating dx yields the books answer.
Page 355 (nonconstant coecients) second example 2: Expanding the books
formulation in ,

2
(1 +n) + 2

+
2
(1 +n)

+n = 0,
or equivalently
( +

)
2
(1 +n) ( +

) +n = 0.
This is a quadratic in +

, and is easily factored into

( +

1) ( +

n) = 0.
Allowing u = log x and v = log y, then by B on page 349,
z = e
u
g
1
(v u) +e
nu
g
2
(v u) ,
from which the books answer follows by back-substituting x = e
u
, y = e
v
, and
f
k
(s) = g
k
(e
s
).
Page 356 (Solutions in power series) Jumbled equation: In the uniqueness
proof, it should go like this: If y = (x) is the power series solution to dy/dx = y, and
y = v(x) were a dierent solution, then
dy
dx
y = v

(x) +v

(x) v(x) = 0.
But since (x) solves the dierential equation, that means

(x) = (x). Hence the

above equation collapses into
dy
dx
y = v

(x) = 0,
requiring either v

or (x) to be identically zero. So if is not zero, v must be a

constant.
This same strategy can demonstrate uniqueness of the general solution to any rst order
linear equation that has no forcing function. If y = (x) solves the equation,
dy
dx
+p (x) y = 0. (a)
We suppose that y = v(x) also solves (a). Then
v
d
dx
+v

(x) +p (x) v(x) = 0. (b)

But since (x) solves the original equation, we have
d
dx
= p (x) (x) .
Substituting the above into (b) yields v

(x) = 0, which is the same result as before.

S-56
Page 357 (solutions in power series) example 4: The method given in the book
for determining series solutions is also known as the method of Frobenius, after math-
ematician, Ferdinand Georg Frobenius (1849-1917). The modern book, Mathematical
Methods in the Physical Sciences, 2
nd
edition, by Mary L. Boas, (1983, Wiley and Sons)
provides an especially lucid presentation of this method in chapter 12. The problem at
hand is to develop a series for the solution to
d
2
y
dx
2
+xy = 0.
The table below breaks the series solution into its elements according to the terms of
the above equation.
x
m
x
m+1
x
m+n
d
2
y
dx
2
(m+ 2) (m+ 1) a
2
(m+ 3) (m + 2) a
3
(n +m + 2) (n +m+ 1) a
n+2
xy a
0
a
n1
Each row of the above table is the contribution of one of the terms of the dierential
equation. Each column shows the contributions attributed to each power of x by various
terms of the equation. The quantity, m, is the exponent of the lowest nonzero term of
the series solution (note that in general m is not necessarily positive and not necessarily
an integer).
We use the right-most column of the table to determine the recursion relationship among
the a
n
s. The sum of that column must be zero in order to satisfy the original equation.
Hence
a
n1
+(n+m+2) (n+m+1) a
n+2
= 0 consequently a
n+2
=
a
n1
(n +m+ 2) (n +m + 1)
.
Replacing n with n + 1, we have
a
n
= (n +m+ 3) (n +m + 2) a
n+3
consequently a
n+3
=
a
n
(n +m+ 3) (n + m+ 2)
.
To determine legitimate values for m, we use the left-hand equation above to run the
the recursion backward. At the point where n < 0, we need to get a zero result in order
to prevent the series from continuing indenitely into negative indices. Note that the
recursion relationship requires that n m be a multiple of 3. This means that when
n m = 3, we have 0 = m(m1). Hence m = 0 or m = 1.
Running the recursion forward with m = 0 and a
0
= 1, we see that
a
0
= 1; a
3
=
1
2 3
; a
6
=
1
(2 3) (5 6)
; a
9
=
1
(2 3) (5 6) (8 9)
; . . .
Doing the same with m = 1 and a
1
= 1, we see that
a
1
= 1; a
4
=
1
3 4
; a
7
=
1
(3 4) (6 7)
; a
10
=
1
(3 4) (6 7) (9 10)
; . . .
You should be able to see that these coecients are equal to the ones given by the books
answer. The two series (where m = 0 and where m = 1) are independent of each other,
so the general solution is a linear combination of those two series.
It is useful to see what happens when we allow m = 2. That means that
a
2
, a
5
, a
8
, . . . would all be nonzero. Hence there is a nonzero term in the series, a
2
x
2
.
The second derivative of this term is the constant, 2a
2
. That means that some other
term of the series times x must cancel 2a
2
. But that term would have to come from
a
1
x
1
, which, by assumption, is zero. So if we were to allow nonzero a
2
, then the
series would continue with nonzero terms indenitely into negative indices. You are
encouraged to experiment for yourself with applying the dierential equation to a series
consisting of

a
3n+2
x
3n+2
to see for yourself why it can never work unless all of these
terms are zero.
S-57
It is fairly easy to prove that both the m = 0 and the m = 1 series converge for all x.
Each nonzero term of either series is arrived at by multiplying the last nonzero term by
x
3
and dividing it by (n 1) n or by n(n + 1), where n is the index of the new term.
Clearly you can always nd large enough n that the divisor will exceed x
3
, no matter
large x is. So for large enough n, the series will always pass the ratio test and therefore
must converge. This convergence analysis applies to this example only. There are other
dierential equations for which the series solutions have nite radii of convergence.
Returning now to the problem of solving for possible values of m, there is an abbreviated
method for doing this once you have established the table. Usually all non-blank rows
the left-most column of the table will have only a single index for a. If so, take the sum
of all rows in that column. In the case of this example, only the rst row of the left-most
column is non-blank. It shows an expression involving a
2
, namely (m+ 2) (m + 1) a
2
.
Let k be the index of a, in this case, k = 2. The procedure is to replace m with mk,
drop the a, and set the resulting expression to zero. In this case you would have
(m + 2 k) (m+ 1 k) = m(m1) = 0. (c)
Equation (c) is known as the indicial equation. Simply solve it for m. In this case it
is clear that m = 0 or m = 1. For this example the solution is easy and results in m
having two nonnegative integer values. More complicated examples may have more than
one row of the left-most column of the table containing nonzero entries. Solving for m
will then involve nding the roots of an indicial polynomial whose degree is equal to the
order of the original equation. Those roots will not necessarily be integer or positive.
Indeed they could even be non-real complex complements. Whatever they be, each such
solution for m will be the basis for an independent series solution for the dierential
equation in question, and will appear as a summand in each exponent. As such, it will
always be possible to factor x
m
from the series solution to arrive at a solution in the
form of
y =

n=0
a
n
x
m+n
= x
m

n=0
a
n
x
n
.
Page 358 (Expansion of J
0
into a power series): Substituting n = 0 into the
Bessel equation given on page 358 gives
d
2
y
dx
2
+
1
x
dy
dx
+y = 0 or equivalently x
d
2
y
dx
2
+
dy
dx
+xy = 0. (d)
The equation on the left is a canonical form, d
2
y/dx
2
+P (x) dy/dx +Q(x) y = 0, that
is useful in the analysis of second order linear equations. We see that in the case in
which we are interested, P (x) = 1/x and Q(x) = 1. We shall come back to that later,
but for the purposes of nding a series solution, J
0
(x), to this, we shall start with the
right-hand form of (d). We could just go ahead and assemble the table from that form,
which would yield the correct series, but doing it that way would hide an important
piece of information. Instead we observe that
x
d
2
y
dx
2
+
dy
dx
=
d
dx

x
dy
dx

.
Making that substitution, (d) becomes
d
dx

x
dy
dx

+xy = 0. (e)
This is the form out of which we build the table. If y
n
(x) = a
n
x
n
, then
xy
n
(x) = a
n
x
n+1
, and
d
dx

x
dy
n
dx

= a
n
n
2
x
n1
hence
d
dx

x
dy
n+1
dx

= a
n+1
(n + 1)
2
x
n
.
S-58
Based upon those facts, we ll out the table as follows:
x
m
x
m+1
x
m+n
xy a
0
a
n1
d
dx

x
dy
dx

a
1
(m+ 1)
2
a
2
(m + 2)
2
a
n+1
(m+n + 1)
2
Using the method we used on the previous example, the indicial polynomial is
m
2
= 0. Observe that this has a double root at m = 0. So we cannot come up with
two independent series arising from the two solutions for m. Instead we will develop the
series for just one of them, using m = 0, and make some comments on the other. The
right-most column of the table shows the recursion to be
a
n1
+a
n+1
(m+n + 1)
2
= 0 consequently a
n+2
=
a
n
(m+n + 2)
2
Because m = 0 and the recursion goes in steps of two, we get a series with only even
powers of x. Allowing a
0
= 1, we have
J
0
(x) = 1
x
2
(2)
2
+
x
4
(2 4)
2

x
6
(2 4 6)
2
+ =

n=0
(1)
n
x
2n
2
2n
n!
2
.
Given that (1) = 1, this is the same as the series given in on page 358 with n = 0.
Using the same type of argument as in the previous example, we see that this series
converges for all x.
We return to the canonical form of the original equation. Recall that the canonical form
is d
2
y/dx
2
+ P (x) dy/dx + Q(x) y = 0. In the case of J
0
, we have P (x) = 1/x and
Q(x) = 1. The series development we just did results in a Taylor series taken around
x = 0, which is precisely the point at which P (x) is discontinuous. Compare that with
example 4 on page 357, which is already in canonical form with P (x) = 0 and Q(x) = x.
In that example, both coecient functions are continuous and dierentiable everywhere.
The distinction has a bearing on how we use the roots of the indicial equation. This
topic is quite involved, and we can only touch on it here. Ordinary Dierential Equations
and Their Solutions by George M. Murphy (van Nostrand, 1960) goes into full detail
in part I, unit B1. Other advanced books on dierential equation also deal with these
issues. When P and Q are continuous and dierentiable at the point around which you
are taking the series, then you can use both roots of the indicial equation as we did in
example 4 on page 357, provided the roots are distinct. If, however, there is a pole in
either P or Q at the center-point of the series, as we have in the equation for J
0
, the
situation becomes more complicated. In that case, if the two roots dier by an integer,
then the two series developed from those two roots will not be independent of each
other. This happens when you develop the series for J
n
, where n is any integer other
than zero. The roots of the indicial equation are m = n. The series from the root,
m = n, has negative powers of x. The denominators of those terms contain a factor
of (k), where k is a nonnegative integer. But the function of nonnegative integers is
innite, so those terms of the series are all zero. The terms that remain are all a scalar
multiple of those of the other series developed from the root, m = n. Hence the two
series are dependent. In addition when you have a double root in the indicial equation,
as we do for J
0
, we also have no way of developing two independent series.
We can use the formula on page 307 to establish a second independent solution when
we already know a particular solution. In the of the zeroth order Bessel function, we
know that J
0
(x) solves equation (d). Using P (x) from the canonical form, we have
Y
0
(x) = J
0
(x)

1
J
0
(x)
2
e

P(x)dx
dx,
S-59
where Y
0
(x) is the second independent solution to (d). Since P (x) = 1/x, the integral
becomes
Y
0
(x) = J
0
(x)

dx
xJ
0
(x)
2
.
Observe that J
0
(x) is nonzero at x = 0. Based upon the series solution for J
0
(x), it
is clear that 1/J
0
(x)
2
and all its derivatives exist at x = 0. This means that 1/J
0
(x)
2
can be expanded as a Taylor series around x = 0, and that series will have a radius of
convergence equal to the lowest magnitude x that solves J
0
(x) = 0. Carrying out that
expansion is beyond the scope of this tutorial. It is clear, however, that the constant
term of that expansion will be equal to 1. That term divided by the x that appears
inside the integral is responsible for J
0
(x) log x term that appears in the expansion for
Y
0
(x). The rst few terms of this expansion is shown on page 358, although the book
uses the old symbol, K
0
, for this function rather than the modern symbol, Y
0
.
Developing a series solution around points other than x = 0: That is, if we seek
a series around x = b, for example, we are looking to establish the a
k
s in
y (x) = a
0
+a
1
(x b) +a
2
(x b)
2
+a
3
(x b)
3
+. . . =

n=0
a
n
(x b)
n
.
The method here is a straightforward variation on the method of Frobenius already
introduced on page S-57. Simply make the substitution, z = x b, into the original
dierential equation. As an example, the zeroth order Bessel equation, (d), becomes
d
2
y
dz
2
+
1
z +b
dy
dz
+y = 0 or equivalently b
d
2
y
dz
2
+
d
dz

z
dy
dz

+ (z +b) y = 0.
Develop the series solution for z around z = 0 using the methods already presented,
then back-substitute at the end. In the case of the zeroth order Bessel function, the
series is more complicated than any we have attempted. We will only go as far as to ll
out the table for this example.
z
m
z
m+1
z
m+n
zy a
0
a
n1
by a
0
b a
1
b a
n
b
d
dz

z
dy
dz

a
1
(m+ 1)
2
a
2
(m + 2)
2
a
n+1
(m+n + 1)
2
b
d
2
y
dz
2
a
2
b (m+2) (m+1) a
3
b (m+3) (m+2) a
n+2
b (m+n+2) (m+n+1)
To form the indicial equation, sum up the left-most column. Observe that we have
mixed indices of a. Let k be the highest of those indices in this case k = 2. Replace
m with mk. The resulting equation is
a
2
b +a
1
(m1)
2
+a
0
b m(m1) = 0.
By assumption, a
0
is the lowest index coecient that is nonzero. It follows that a
2
and
a
1
are both zero. So the indicial equation is m(m1) = 0. Summing the right-most
column, though, we see that the recursion is nasty and involves four dierent indices
of a. In this example there is no clear path to a closed form for a
n
. Were we able to
establish a
n
, however, we could not expect that either the m = 0 or the m = 1 series to
be equal to either J
0
or Y
0
. Instead each would be some linear combination of the two.
And because Y
0
has a pole at z = b, we would expect that the radius of convergence
of either series would be b.
S-60
Resonant systems revisited: On page S-26 is a paragraph introducing the canonical
form of the equation for resonant systems. Resonant systems occur again and again in
engineering. Here are three physical examples.
i. An RLC-circuit. The di-
agram shows such a circuit. A
voltage, v
in
, which varies sinu-
soidally in time, is applied on the
left and results in an output volt-
age, v
out
, on the right. Here v
in
is the forcing function and v
out
is the particular integral deriving
from the dierential equation implicit in the circuit acting upon the forcing function.
We let v
in
= cos
f
t, where we shall call
f
, the forcing frequency. When
f
= 0, the
impedance of the capacitor, C, is innite and the impedance of the inductor, L, is zero.
So we expect then that v
out
= v
in
. As
f
goes to innity, the impedance of the inductor
goes to innity as well while the impedence of the capacitor goes to zero. Under those
conditions we should expect that v
out
should approach zero. Analysis of this circuit
network yields
v
out
=
1
LC
D
2
+
R
L
D +
1
LC
v
in
.
The analysis is based upon the impedance of the inductor being LD, the impedence of
the capacitor being 1/ (CD), and the impedence of the resistor being R.
Letting =

1
LC
and =
1
2
R

C
L
, we nd that
v
out
=

2
D
2
+ 2 D +
2
v
in
,
which is similar to the resonant equation on page S-26, where is the resonant frequency
and is the damping factor. Substituting v
in
= cos
f
t, this becomes
v
out
=

2
D
2
+ 2 D +
2
cos
f
t. (a)
The goal is to see how v
out
varies with the forcing frequency,
f
. We can solve for the
particular integral, v
out
, using the method detailed in case 5 on page 313. Replacing D
2
with
2
f
,
v
out
=

2
2 D

2
f

2
cos
f
t.
Now multiply top and bottom by 2 +

2
f

2

to get
v
out
=

2D +

2
f

2

2
4
2

2
D
2

2
f

2

2
cos
f
t.
Again replacing D
2
with
2
f
,
v
out
=

2D +

2
f

2

2
4
2

2
f
+

2
f

2

2
cos
f
t.
To simplify, divide top and bottom by
4
and replace
f
/ with the frequency ratio, .
v
out
=
2

D +

2
1

4
2

2
+ (
2
1)
2
cos
f
t =

1
2

cos
f
t + 2 sin
f
t
(1
2
)
2
+ 4
2

2
. (b)
It is possible to resolve this function into two components. One is the amplitude of the
S-61
resulting sinusoid. To nd that we take the square root of the sum of the squares of the
coecients of cos
f
and sin
f
.
|v
out
|
|v
in
|
=
1

(
2
1)
2
+ 4
2

2
. (c)
A plot of this function
is shown to the right for
various values of . This
type of plot showing the
magnitude of output of
a system as a function
of sinusoidal frequency
of input is known as
a Bode plot, after Hen-
drik Wade Bode (1905-
1982), who invented it.
Note that the axes of
the plot are scaled by
the log base 10 of fre-
quency and the log base
10 of amplitude.
Based upon this
plot we can see
that, as expected, for

f
< (that is for
0 < < 1), we have
|v
out
|/|v
in
| 1, which
is consistent with our
early observation that v
out
= v
in
when
f
= 0. When
f
, (that is when 1),
we see that on the log/log scale, the plot decreases with a slope of 2. This is equivalent
to |v
out
| varying inversely as
2
, which is consistent with our early observation that as

f
goes to innity, we expect v
out
to go to zero. Of course the region of greatest interest
is near the resonant frequency that is the region around = 1, or equivalently the
region around
f
= . When < 1, we see that the plot reaches a maximum just to the
left of = 1. As gets closer and closer to zero, the maximum becomes sharper and
sharper, as well as becoming higher. The position of the maximum approaches = 1.
Observe that at = 0.1, we have the amplitude of v
out
being ve times the input voltage
at the point of resonance. Indeed, when = 1, equation (c) predicts that
|v
out
| =
|v
in
|
2
.
So as approaches zero, at resonance the ratio of the output amplitude to the input
amplitude goes to innity. Circuits such as this one designed to have a sharp resonance
(that is a very small value for ) are useful in radio receivers, among other applications,
where it is necessary to select a narrow band of frequencies out of a broader spectrum
input.
By contrast, look at what happens when is much greater than 1. The trace for = 5
shows that for
f
up to about 0.1, we continue to have |v
out
| |v
in
|. From about
0.1 to about 10, the trace decreases with a slope of 1. Only for
f
> 10 do we see
the decrease rate become a slope of 2. What is happening here is that when > 1, the
denominator of equation (a) will have two real roots, one of them with [D[ < , the other
with [D[ > . In the case of = 5, we see D =

5 +

24

and D =

5 +

24

,
or D 0.1 and D 10. When
f
falls between those two roots, the factor
S-62
(D + 0.1) is growing proportionally larger, whereas the factor (D + 10) is changing
very little. So between the two roots, the amplitude varies in inverse proportion to
the forcing frequency. But when the forcing frequency is larger than the magnitude of
both the roots, the amplitude varies in inverse proportion to the square of the forcing
frequency. Also observe that as becomes larger and larger, the two roots will spread
farther and farther apart (although both roots will always be negative).
Recalling that the method of nding the particular integral given on page 313 replaces
D
2
with
2
f
, this means that it also replaces D with
f
, where =

1. If the two
roots are at 0.1 and 10, then the contribution to the denominator of (a) of the
two roots will be
f
+0.1 and
f
+10 respectively. The use of complex quantities
in the analysis of linear circuits is commonplace in the world of electrical engineering.
Indeed if all you had done to equation (a) was replace D with
f
and evaluated the
resulting complex function, the real part of the result would give you the correct cosine
coecient and the imaginary part the correct sine coecient. From those coecients
the right-hand version of equation (b) would follow immediately. And just so you know,
the customary symbol in electrical engineering for

1 is j,which distinguishes it from

i, which electrical engineers use as a variable name for electrical current. Electrical
engineers also use the symbol, s, instead of D.
The discussion of equa-
tion (b) already stated
that it has a second
component besides its
magnitude. That sec-
ond component is its
phase angle. When you
add the sine and co-
sine terms of equation
(b), you get a sinu-
soid that lags in phase
from v
in
. The plot here
shows that phase dier-
ence as a function of
=
f
/ for various
values of . We can
see that when the forc-
ing frequency is substantially less than , there is very little phase lag between the
output and the input. At the resonant frequency, the output always lags the input by
exactly 90

. And as the forcing frequency becomes much greater than the resonant fre-
quency, the phase lag approaches 180

, which means that the v

out
is almost completely
opposite to v
in
. The other trend to notice is that as approaches zero, the phase lag
transition from nearly 0

to nearly 180

occurs more and more sharply right at the

resonant frequency.
The function used to construct the above plot is:
v
out
[v
in
= tan
1
2
1
2
,
then summing in 180

where the arctangent was negative to account for quadrant.

If you had processed equation (a) by replacing D with
f
, then you would have had a
point on the complex plane for each value of
f
. Drawing a ray from the origin through
that point would make an angle with the real axis that is exactly equal to the phase lag
of v
out
to v
in
corresponding to that
f
. In addition, the distance of that point from the
origin would be exactly equal to the corresponding magnitude, |v
out
|/|v
in
|.
S-63
As promised earlier, there are other examples of physical systems that follow the reso-
nance equation, (a).
ii. An automobiles suspension: The axle of an automobile is connected to the
chassis by a spring and a dashpot (which mechanics call a shock-absorber) in parallel.
Let the chassis have mass, m, and let x
in
be the vertical position of the axle and x
out
be the position of the chassis. The spring applies a force, k (x
in
x
out
) to the chassis.
The dashpot applies a force, c

dxin
dt

dxout
dt

to the chassis. Be aware that this is a very

simplied description of an automobile suspension. Other factors in real suspensions
are couplings between the four wheels, that the suspension is responsive not only in
the vertical axis but the lateral and longitudinal axes as well, and a number of other
engineering details. But the single-wheel, single-axis system described here will show
how resonant systems occur in mechanics.
By Newtons laws of motion, we have
f
total
= k (x
in
x
out
) +c

dx
in
dt

dx
out
dt

= m
d
2
x
out
dt
2
, (d)
or equivalently
k (x
in
x
out
) +c (x
in
x
out
) D = mx
out
D
2
.
Solving for x
out
in terms of x
in
,
x
out
=
k
m
+
c
m
D
D
2
+
c
m
D +
k
m
x
in
.
If we let =

k
m
and =
1
2
c

km
, the equation becomes
x
out
=

2
+ 2D
D
2
+ 2D +
2
x
in
.
Finally we let x
in
= cos
f
t:
x
out
=

2
+ 2D
D
2
+ 2D +
2
cos
f
t, (e)
which is similar to equation (a), in that it has the same denominator, but is dierent in
that it has an extra term in the numerator. We will solve this for x
out
by substituting
D with
f
, as suggested on the previous page.
x
out
=

2
+ 2
f

2
f
(
f

2
)
cos
f
t,
Multiply top and bottom by 2
f
+

f

2

, then divide top and bottom by

4
and replace
f
/ with the frequency ratio, , take the cancellation, rearrange a bit, and
you have
x
out
=
4
2

2
1

2
3
4
2

2
+ (
2
1)
2
cos
f
t. (f)
S-64
A plot of the magnitude of
|x
out
|/|x
in
| for various val-
ues of is shown to the
right. Observe that like
Bode plot of equation (c),
for
f
< , the output
magnitude is nearly equal
to the input magnitude.
The region around the reso-
nance point is nearly identi-
cal to that of (c). But for

f
> , the behavior is
dierent. For low values
of the trace slopes down
steeply after resonance. For
high values of , the trace
doesnt even start to slope
down appreciably until well
to the right of resonance.
For all values of , the slope
far to the right of resonance
eventually approaches 1.
This of course begs the ques-
tion of what value of
should be chosen for an au-
tomobile suspension. The goal is that the suspension will isolate the chassis from irreg-
ularities in the road. So for high values of
f
, which correspond to rapid changes in
the road surface, we would like to see them well below the 1.0 line in the plot. For low
values of
f
we would like the chassis to follow the axle. A low value for does that
the best except for the region right around resonance. In that region a low amplies
those frequencies, and the eect will be that the car bounces at the resonant rate each
time it goes over a bump. So the actual value of chosen by automotive designers is a
compromise. Typically the suspension is designed for 2 radians/second and with
between 0.5 and 1.0.
iii. A servo-system: A potentiometer is arranged in a circuit so that its wiper-pin
produces a voltage that is in proportion to how far the potentiometer shaft is turned.
That voltage goes to an amplier that delivers a current to a servomotor in proportion
to the input voltage. The servomotor delivers a torque to a ywheel in proportion to
its input current. The ywheel also has a mechanical damper that provides a resistive
torque in proportion to how fast the ywheel spins. Finally, the shaft of the motor is
coupled to the shaft of a second potentiometer arranged in the same way as the rst.
The wiper-pin voltage of that potentiometer is subtracted from the wiper-pin voltage
of the rst, and it is actually the dierence of the two voltages that is the input of the
amplier.
The desired eect is that if an operator turns the shaft of the rst potentiometer by
some angle, the servo-system will turn the ywheel by the same angle, after which the
ywheel will come to rest. Or more broadly, we desire the ywheels angular position
to follow that of the rst potentiometer shaft.
Let k be conversion factor that relates servomotor torque to potentiometer position
dierence. Let c be the resistive factor that relates resistive torque on the ywheel to
the angular velocity of the ywheel. Let m be the moment of inertia of the ywheel.
Then the equation descibing the whole servosystem is identical to (d) in the previous
S-65
example, where x
in
is the angular position of the rst potentiometer shaft, x
out
is the
angular position of the ywheel, and f
total
is replaced by total torque. Observe that
for low values of , the ywheel will overshoot, then oscillate around the control point
before settling, each time the rst potentiometer is moved. For high values of , the
ywheel response will be sluggish, undershooting the rst potentiometers change of
position and settling into the control point only over time. Again the ideal choice for
is a compromise value between 0.5 and 1.0.
The the galvanometer on page 323 is yet another example forcing an input on a resonant
system.
Dierential Equation Solution to Keplerian Orbits: Here we solve the problem
of two bodies attracted to each other by gravity. In particular we solve the problem
where one of the two bodies outweighs the other by a large factor (for example, the
suns mass is over 300,000 times the mass of the earth). This way we can consider the
larger mass as xed and solve only for the orbit of the smaller mass, m.
We set the problem up in polar coordinates. In the radial axis we have
m
d
2
r
dt
2
mr

d
dt

2
= f (r) . (a)
Here f (r) is the attractive gravitational force felt by the smaller object as a function of
how far it is from the larger object. The rst term on the left is simply Newtons law
of motion. The second term is the so-called centrifugal force. That is it is the apparent
force a rotating object feels away from the center of rotation.
The gravitational force acts only in the radial direction. Hence in the tangential axis we
have conservation of angular momentum. Angular momentum, L, is given by
L = mr
2
d
dt
. (b)
By saying that this quantity is conserved, we are saying that it is constant and therefore
its time derivative is zero.
m
d
dt

r
2
d
dt

= 0, or equivalently r
d
2

dt
2
+ 2
dr
dt
d
dt
= 0.
Observe that geometrically, angular momentum, L, is the mass, m, times twice the rate
of radial area swept by the smaller object as it orbits the larger object. Johanas Kepler
observed that planets orbiting the sun sweep equal area in equal time (Keplers second
law). So we see that conservation of angular momentum is equivalent to Keplers second
law.
Solving (b) for d/dt and substituting the result into (a), we have
m
d
2
r
dt
2
mr

L
mr
2

2
= f (r) , or equivalently m
d
2
r
dt
2
=
L
2
mr
3
f (r) .
Newtons universal law of gravitation postulates that
f (r) =
GMm
r
2
,
where M and m are the masses of the two bodies, G is a universal physical constant,
and r is the distance between the two objects. So the equation we must solve is
m
d
2
r
dt
2
=
L
2
mr
3

GMm
r
2
. (c)
The trick to solving this is to make the substitution, r = 1/u. Then dierentiating r
once yields
m
dr
dt
= m
1
u
2
du
dt
= m
1
u
2
du
d
d
dt
=
&&
m
1
&
&
u
2
du
d
L&
&
u
2
&&
m
.
S-66
Dierentiating a second time yields
m
d
2
r
dt
2
= L
d
dt
du
d
= L
d
2
u
d
2
d
dt
=
L
2
u
2
m
d
2
u
d
2
.
Substituting this result into (c) we get

L
2
u
2
m
d
2
u
d
2
=
L
2
u
3
m
GMmu
2
, or equivalently
d
2
u
d
2
+u =
GMm
2
L
2
. (d)
The general solution to (d) (see case 3 on page 309) is
u = u
0
cos ( +) +
GMm
2
L
2
; back-substuting, r =
L
2
u
0
L
2
cos ( +) + GMm
2
, (e)
where u
0
and are arbitrary constants. Equation (e) does not tell us anything about the
time-behavior of the orbit, but it does give the shape of the orbit. Setting = 0 aligns
the symmetry of the orbit with the x-axis. By substituting r
2
= x
2
+y
2
and x = r cos ,
we nd that the resulting Cartesian formula is the equation of a conic section, of which
an ellipse with one of the foci at the origin is one of the possibilities tting the equation.
The latter is mathematical conrmation of Keplers rst law. Replacing with nonzero
values simply rotates the major axis of the ellipse away from the x-axis.
The time-behavior of the orbit can only be determined by numeric methods. Such
methods, as well as more details on the derivation above, can be found in An Introduction
to Celestial Mechanics by Forest Ray Moulton (1914 MacMillan. Available since 1970
from Dover Publications).
Introduction to Fouriers Heat Equation and its Solution: This topic is covered
here only as the barest of outlines. Imagine a long thin strand of heat conducting
material, such as a length of wire. We also imagine that it is completely insulated
from the outside world except maybe at its endpoints. So heat conduction is restricted
to heat traveling along the length of the wire. A ow of heat along the wire results
in a temperature gradient according to the constant, . That is, for example, if =
1 calorie/C

cm second, then if the temperature gradient is 1 C

per cm, heat ow past

any point of that gradient will be 1 calorie per second. We also let the thermal density of
the wire be represented by . Then, for example, if = 1 calorie/cm C

and a point on
the wire is at temperature 100 C

, then that point on the wire contains 100 calories/cm

of heat.
If we take the case where one end of the wire of length, L, is held at temperature, T
1
and
the other at T
2
, and the temperature gradient is constant throughout the length at a
rate of (T
2
T
1
) /L, then heat ow will be (T
2
T
1
) /L. If the initial heat gradient is
constant at this value, then as long as the ends are held at T
1
and T
2
, the heat gradient
along the entire length of the wire will remain unchanged over time. That is, a constant
heat gradient of (T
2
T
1
) /L is the steady state condition for the boundary conditions
of T
1
and T
2
held at the respective ends of the wire.
Having constant heat gradient implies that
2
T/x
2
= 0, where x represents position
along the wire, and whenever this condition is true, the wire is in steady state, which
is to say that the temperature prole will remain unchanged through time. But when

2
T/x
2
= 0, we nd that the temperature prole changes over time even if the bound-
ary conditions at either end of the wire are held constant. Finding the time function of
the temperature prole under this condition is the problem Jean Baptist Joseph Fourier
(1768-1830) sought to solve in 1822 (see God Created the Integers, a collection of math-
ematical breakthrough papers edited and commented by Stephen Hawking, Running
Press, 2005 for more on Fourier and his work).
S-67
We can see intuitively that wherever there is a hump in temperature, that is where

2
T/x
2
< 0, there is a net outward ow of heat from that region to dissipate the hump,
and wherever there is a dip in temperature, that is where
2
T/x
2
> 0, there is a net
inward ow of heat to that region to ll up the dip. Fouriers heat equation,

2
T
x
2
=
1
k
T
t
, (a)
where k = / quanties this. This equation is similar to example 1 on page 351, and
page S-52 oers a solution to this equation in terms of decaying sinusoids,
T (x, t) =

e
k
2
t
sin (x +

) ,
or equivalently
T (x, t) =

e
k
2
t
A

cos (x) +B

sin(x) , (b)
where A

and B

are sets of constants, one A and one B for each value of used in
the solution.
What Fourier recognized and is remembered for is the observation that you can always
convert any initial bounded function, T (x, 0), into a weighted sum of sines and cosines,
and likewise, you can always convert any weighted sum of sines and cosines into a
function of x. So the procedure is to rst convert the initial T (x, 0) into a set of
A

s and B

s for some set of values of . Second, to nd what has happened after

some elapsed time, t, we apply the decay function, e
k
2
t
, for each value of to the
corresponding sines and cosines. Third and nally, we sum up the decayed sines and
cosines each weighted by the A

s and B

s that we found in the rst step.

If we restrict our consideration of a nite length, L, of the wire, then the set of s we
need to consider is
k
= 2k/L, for each nonnegative integer, k. To demonstrate how
we break an arbitrary function T (x, 0), dened over 0 x L, into sines and cosines,
we rst demonstrate the principle of orthogonality of sines and cosines. Observe that

L
0
cos (
j
x) sin (
k
x) dx = 0 for any j and k
and

L
0
cos (
j
x) cos (
k
x) dx = 0 and

L
0
sin (
j
x) sin(
k
x) dx = 0 whenever j = k
In words, we have a collection of functions, the product of any two of them when
integrated from 0 to L yields zero, provided the two functions are distinct. Such functions
are said to be orthogonal to one another under this operation. The operation itself is
fundamental to this problem and we give it the name, scalar product. We dene the
scalar product of two function, f (x) and g (x) as
'f, g` =

L
0
f (x) g (x) dx,
and we say that nonzero f and g are orthogonal whenever 'f, g` = 0.
If we assume that the function, T (x, 0), is the sum,
T (x, 0) =

k
A
k
cos (
k
x) +B
k
sin (
k
x) ,
for some set of A
k
and B
k
, then taking the scalar product of T with any any cos (
j
x)
yields
'T (x, 0) , cos (
j
x)` = 'A
j
cos (
j
x) , cos (
j
x)` .
Observe that by orthogonality, all the terms in the original sum except the one above
to the right of the equal are zero. Only for the cosine term where k = j do we get a
S-68
nonzero term. Likewise with
'T (x, 0) , sin (
j
x)` = 'B
j
sin (
j
x) , sin(
j
x)` .
It is easily veried using our denition of scalar product that
'Af (x) , g (x)` = A'f (x) , g (x)` and 'f (x) , Bg (x)` = B'f (x) , g (x)` ,
for any functions, f and g, and for any constants, A, and B. This makes it easy to solve
for the value of any A
k
or B
k
. Simply take the product of T (x, 0) with corresponding
cos (
k
x) or sin (
k
x) and integrate the product from 0 to L, then divide by L/2. The
dividing by L/2 is because
'cos (
k
x) , cos (
k
x)` = 'sin (
k
x) , sin(
k
x)` =
L
2
for all k.
Proving this last fact is left as an exercise for the reader.
To prove that the desired set of A
k
s and B
k
s always exists, we sample the interval,
[0, L] at 2n + 1 equally spaced points.
x
i
=
iL
2n
for i from 0 to 2n.
We also isolate 2n + 1 sinusoid functions, cos (
k
x) for k from 0 to n and sin (
k
x) for
k from 1 to n. Note that we omit consideration of sin (
0
x) because it is zero for all x.
We allow rows k = 0 through n of a matrix, M, to be cos (
k
x
i
), where i goes from 0 to
2n. We allow rows k = n +1 through 2n to be sin (
kn
x
i
). This gives us an invertible
square matrix. Then taking a vector of n + 1 A
k
s and n B
k
s times the matrix, M,
gives a weighted sum of the sine and cosine functions over the various sample points,
x
i
. It follows that
M
1
T (x
i
, 0)
must give back a vector of the weights, A
k
and B
k
. As we let n go to innity, we see that
we can reproduce the original function, T (x, 0), on a denser and denser set of sample
points.
There is much more can be said about solving Fouriers heat equation. Indeed there
are entire textbooks devoted to this one equation. The point here is to introduce you
to the notion of decomposing a function into weighted sinusoids. This procedure is
known as Fourier analysis, and is the topic of chapter VIII of J.W. Mellors book. It
is also applicable to a wide variety of problems other than the solving Fouriers heat
equation. In an earlier section, for example, where we made Bode plots of the responses
of resonant systems to sinusoidal inputs, we were laying the groundwork for applying
Fourier analysis to this problem as well. In real life the inputs to such systems are not
sinusoids, but could be any function. But the fact that any legitimate input function can
be decomposed into sinusoids allows us to solve the action of the system on each sinusoid,
then sum the output sinusoids to determine the systems response to the original input.
Even the action of the human ear is connected with Fouriers method. The human ear is
an instrument that resolves an input function, namely sound acting upon the eardrum,
into a set of sinusoids of various frequencies and transmits the magnitudes of the resolved
frequencies to the brain.
Also be aware that Fourier analysis is just a single example of a more general method-
ology of decomposing an arbitrary function into a weighted sum of some set of basis
functions. In the more general setting, the basis set need not be sinusoids. Depending
upon the type of problem, resolution into other basis functions is often convenient. As
mentioned on page 358 of the text, Bessel functions provide a useful basis for systems
that have cylindrical symmetry. And there are others for other symmetries.
S-69
Selected referenced sections of original text begin
20. Leibnitz Theorem. To nd the nth dierential coecient of the product
of two funnctions of x in terms of the dierential coecients of each function.
On page 26 the dierential coecient of the product of two variables was shown to
be
dy
dx
=
d (uv)
dx
= v
du
dx
+u
dv
dx
,
where u and v are fuuctions of x. By successive dierentiation and analogy with the
binomial theorem it may be shown that
The co-
ecient,
n, is a
binomial
coe-
cient,
as are
the re-
spective
unshown
coe-
cients
in this
equation.
KH
d
n
(uv)
dx
n
= v
d
n
u
dx
n
+n
dv
dx
d
n1
u
dx
n1
+ +u
d
n
v
dx
n
. (1)
This formula, due to Leibnitz, will be found very convenient in Chapter VII How to
Solve Dierential Equations. The reader must himself prove the formula, as an exercise
on (' 19, by comparing the values of d
2
(uv)/dx
2
, d
3
(uv)/dz
3
,..., with the developments
of (x +h)
2
, (x +h)
3
,..., of page 22.
49
22. Eulers Theorem on Homogeneous Functions. The following
discussion is convenient for reference:
To show that if u is an homogeneous function

of the nth degree, say u = ax

,where
+ = n, then
x
u
x
+y
u
y
= nu. (14)
By dierentiation of the homogeneous function,
u = ax

+bx

1
y

1
+ =

ax

,
where + =
1
+
1
= = n, we obtain
u
x
=

ax
1
y

and
u
y
=

ax

y
1
.
Hence
x
u
x
+y
u
y
=

a ( +) x

= n

ax

.
The theorem may be extended to include any number of variables (see footnote, page
340).
EXAMPLES. (1) If u = x
2
y +y
2
x +3xyz, then x
u
x
+y
u
y
+z
u
z
= 3u. Prove this
result by actual dierentiation. It of course follows directly from Eulers theorem, since
the equation. is homogeneous and of the third degree.
(2) If u =
x
3
+x
2
y+y
3
x
2
+xy+y
2
, x
u
x
+ y
u
y
= u, since the equation is of the rst degree and
homogeneous.
(3) Put Eulers theorem into words. Ansr. In any homogeneous function, the terms
of the products of each variable with the partial dierential coecients of the original
function with respect to that variable is equal to the product of the original function with
its degree.

An homogeneous function is one in which all the terms containing the variables have
the same degree. Examples: x
2
+bxy +z
2
; x
4
+xyz
2
+x
3
y +x
2
+z
2
are homogeneous
functions of the second and fourth degrees respectively.
The sign is to be read the sum of all terms of the same type as. . . or here
the sum of all terms containing x, y and a constant. The symbol is sometimes
used in the same way for the product of all terms of the type.
56
23. Successive Partial Dierentiation. We can get the higher par-
tial derivatives by successive dierntiation, using processes analogous to
those used on page 47. Thus when
u = x
2
+y
2
+x
2
y
2
,
u
x
= 2x + 2y
3
x;
u
y
= 2y + 3x
2
y
2
; (1)
repeating the dierentiation,

2
u
x
2
= 2

1 +y
3

;

2
u
y
2
= 2

1 + 3x
2
y

; (2)
If we had dierentiated u/x with respect to y, and u/y with respect
to x, we should have obtained two identical results, viz.:

2
u
yx
= 6y
2
x, and

2
u
xy
= 6y
2
x. (3)
This rule is general.
The higher partial derivatives are independent of the order of dier-
entiation. By dierentiation of u/x with respect to y, assuming x to
be constant, we get

u
x

y, which is written

2
u
yx
; on the other
hand, by the dierentiation of u/y with respect to x, assuming y to be
constant, we obtain

2
u
xy
. That is to say

2
u
yx
=

2
u
xy
. (4)
This was only proved in (3) for a special case. As soon as the reader. has
got familiar with the idea of dierentiation, he will no doubt be able to
deduce the general proof for himself, although it is given in the regular
text books. The result stated in (4) is of great importance.
24. Exact Dierentials. To nd the condition that u may be a func-
tion of x and y in the equation
du = M dx +N dy, (5)
where M and N are functions of x and y.
We have just seen that if u is a function of x and y
du =
u
x
dx +
u
y
dy, (6)
that is to say, by comparing (5) and (6)
M =
u
x
; N =
u
y
.
57
Dierentiating the rst with respect to y, and the second with respect
to x, we have, from (4)
M
y
=
N
x
. (7)
In the chapter on dierential equations this condition is shown to be nec-
essary and sucient in order that certain equations may be solved, or
integrated as it is called. Equation (7) is therefore called the criterion
of integrability. An equation that satises this condition is said to be a
complete or an exact dierential. For examples, see page 290.
25. Integrating Factors. The equation
M dx +N dy = 0 (8)
can always be made exact by multiplying through with some function of x,
called an integrating factor. (M and N are functions of x and y.)
Since M and N are functions of x and y, (8) may be written
dy
dx
=
M
N.
(9)
or the variation of y with respect to x is as M is to N; that is to say, x
is some function of y, say
f (x, y) = a,
then from (5), page 52,
f (x, y)
x
dx +
f (x, y)
y
dy = . (10)
By a transformation of (10), and a comparison of the result with (9), we
nd that
dy
dx
=
f (x, y)
x
dx

f (x, y)
y
dy =
M
N
. (11)
Hence
f (x, y)
x
= M; and
f (x, y)
y
= N, (12)
where , is either a function of x and y, or else a constant. Multiplying the
original equation by the integrating factor , and substituting the values
of M, , N so obtained in (12), we obtain
f (x, y)
x
dx +
f (x, y)
y
dy = 0,
which fulls the condition of exactness.
EXAMPLE. Show that the equation y xdy = 0 becomes exact when multiplied by
1/y
2
.
M
y
=
1
y
2
;
N
x
=
1
y
2
.
Hence M/y = N/x, the condition required by (7). In the same way show that
1/xy and 1/x
2
are also integrating factors.
58
67. Envelopes.
The equation
m
a
+ax,
represents a family of curves, since for each value of a, we get a distinct
curve. If a varies continuously it will determine a succession of curves, each
of which is a member of the family denoted by the above equation. a is
said to be the variable parameter of the family, since the dierent members
of the family are obtained by assigning arbitrary values for a. Let the
equations
y
1
=
m
a
+ax (1)
y
2
=
m
a +a
+ (a +a) x (2)
y
3
=
m
a + 2a
+ (a + 2a) x (3)
142
be three successive members of the family. As a general rule two distinct
curves in the same family will have a point of intersection.
Fig. 78. Envelope
Let P (Fig. 78) be the point of intersection
of curves (1) and (2); P
1
the point of inter-
section of curves (2) and (3), then, since P
1
,
and P
2
, are both situated on the curve (2),
PP
1
, is part of the locus of a curve whose
are PP
1
, coincides with an equal part of the
curve (2). It can be proved, in fact, that
the curve P P,. . . touches the whole family
of curves represented by the original equa-
tion. Such a curve is said to be an envelope
of the family.
To nd the equation to the envelope,
bring all the terms of the original equation
to one side,
y =
m
a
ax = 0.
Then dierentiate with respect to the variable parameter, and put
m
a
2
x = 0.
Eliminate a, between these equations,
y

m x x

m
x
= 0, or, y 2

m x = 0.
y
2
= 4mx.
EXAMPLES. Find the envelope of the family of circles
(x a)
2
+ y
2
= r
2
,
where a is the variable parameter. Dierentiate with respect a and x a = 0;
Fig. 79. Double Envelope
eliminating a, we get y = a, which
is the required envelope. The enve-
lope y = a represents two straight
lines parallel to the x-axis and at a
distance +a, and a from it. Shown
Fig. 79.
(2) Show that the envelope of the
family of curves (x ma)
2
+y
2
=
4ma, is a parabola y
2
= 4mx. See
'' 126 and 138.
143
72. How to nd a Value for the Integration Constant. It is
perhaps unnecessary to remind the reader that integration constants must
not be confused with the constants belonging to the original equation. For
instance, in the law of descent of a falling body
dv/dt = g;

dv = g

dt, or, v = gt +C. (1)

Here g is a constant representing the increase of velocity due to the earths
attraction, C is the constant of integration. The student will nd some
instructive remarks in ' 118.
There are two methods in general use for the evaluation of the integra-
tion constant.
First Method. Returning to the falling body and to its equation of
motion,
v = gt +C.
On attempting to apply this equation to an actual experiment, we
should nd that, at the moment we began to calculate the velocity, the
body might be moving upwards or downwards, or starting from a position
of rest. All these possibilities are included in the integration constant C.
Let v
0
, denote the initial velocity of the body. The computation begins
when t = 0, hence
v
0
= g 0 +C, or, C = v
0
.
If the body starts to fall from a position of rest, v
0
= C = 0, and

dv = gt, or, v = gt.

This suggests a method for evaluating the constant whenever the nature
of the problem permits us to deduce the value of the function for particular
values of the variable.
If possible, therefore, substitute particular values of the variables in
the equation containing the integration constant and solve She resulting
expression for C.
EXAMPLE. Find the value of C in the equation
t =
1
k
log
1
a x
+C, (2)
which is a standard velocity equation of physical chemistry. t represents the time
required for the formation of an amount of substance x. When the reaction is just
beginning, x = 0 and t = 0. Substitute these values of x and t in (2).
1
k
log
1
a
+C = 0, or, C =
1
k
log
1
a
.
Substitute this value of C in the given equation and we get
t =
1
k

log
1
a x
log
1
a

=
1
k
log
a
a x
.
Second Method. Another way is to nd the values of x corresponding to
two dierent values of t. Substitute the two
162
sets of results in the given equation. The constant can then be made to
disappear by subtraction.
EXAMPLE. In the above equation, (2), assume that when t = t
1
, x = x
1
and when
t = t
2
, x = x
2
; where x
1
, x
2
, t
1
, and t
2
, are numerical measurements. Substitute these
results in (2).
t
1
=
1
k
log
1
a x
1
+C; t
2
=
1
k
log
1
a x
2
+C.
By subtraction and rearrangement of terms
t
2
t
1
=
1
k
log
a x
1
a x
2
.
The result of this method is to eliminate, not evaluate the constant.
Numerous examples of both methods will occur in the course of this
work. Some have already been given in the discussion on the Compound
Interest Law in Nature.
163