ABSTRACT
An equation containing two variables at most and derivatives of the rst
or higher order of one of the variables with respect to the other is called a
dierential equation The order n of such an ordinary dierential equation
is the order of the highest derivative that appears To solve a dierential
equation of nth order n integrations are necessary Each integration delivers
an integration constant These arbitrary constants can be used to adapt the
general solution to the particular solution of the problem in question
Many dierential equations represent mathematical models describing a
physical or technical problem The free fall of a parachutist is described by
the equation of motion
m
d
x t
dt
mg a
dx
dt
Herem is the mass of the parachutist mg is the gravitational attraction and
the last term on the righthand side describes aerodynamic drag By dening
the velocity v of fall
v t
dx t
dt
x
the equation which is of second order can be reduced to a dierential
equation of rst order
dv t
dt
g
a
m
v
t
This equation is of the form
P tdtQ vdv
and is called dierential equation with variables separable separable equation
since can be written in the form
dt
dv
t
p
m
p
ag
lnC
avm
p
agm
avm
p
agm
This integration has been executed using the Mathematica command
Integrate[1/(a*vˆ2/m-g),v] and observe ln x x arctanh x
If we now assume the initial condition v v
then we obtain the
integration constant C from Thus our solution of for
v t reads
avm
p
agm
avm
p
agm
exp
p
ag
p
m
t
For t this gives the nal falling speed
v
p
gma v
Inserting into yields a dierential equation for x t and a second
integration constant after integration Instead of solving for
x t we may derive a dierential equation for x v Substituting dt from
into we have
dx v
dv
g av
m
and after integration using the substitution u v
one has
x C
m
a
ln
v
v
In order to determine the second integration constant C let us consider that
the parachutist leaves the airplane at time t with zero velocity v
at a
certain height x
above the earths surface x Thus we have
x
C
m
a
ln v
as the second initial condition and the solution of reads
x x
m
a
ln
v
v
In this example of a dierential equation of second order we used
two initial conditions ie two conditions imposed on the solution at the
same location x x
We could however impose two conditions at dierent
Then we can assume that the parachutist should contact the earths surface
x with a velocity v of fall equivalent to one tenth of v
We then have
the nal condition boundary condition that for x v v
or from
C
m
a
ln
v
Then the general solution becomes the particular solution
x
m
a
ln
v
v
If we solve using the command
x t C
m
a
ln
cosh
p
g atmC
p
am
Here C
and C
are integration constants Using the two initial conditions
v and x x
one obtains the particular solution
x t x
m
a
ln
cosh
t
p
ga
p
m
For m kg a x
m and g ms
this function
is shown in Figure This gure has been produced with the help of the
commands
m=70; a=0.08; x0=5000; g=9.81; x[t_]=x0-m*Log[Cosh[t*Sqrt[g*a]/Sqrt[m]]]/a; Plot[x[t],{t,0,60}] In this expression we have used a semicolon This allows more than one
command to be written in one line The command to plot a function needs
a function that can be evaluated at any arbitrary value of the independent
variable t To obtain such a variable we replaced t by t_
We now treat an example to nd the force producing a given trajectory Let
us consider the relativistic onedimensional equation of motion of an astronaut
traveling into space along a straight line and returning to earth Neglecting
gravitation such a motion is described by three equations
relativistic equation of motion
d
dt
m
v
p
v
c
dm
dt
v
p
v
c
Numbers in brackets designate bibliographical references see the list at the end of the
Figure
Trajectory of a parachutist
conservation of energy
d
dt
m
c
p
v
c
dm
dt
c
p
v
c
addition theorem of velocities
v
w v
v
wc
In these equations we used the following designations m
and m
are the
rest masses of the space rocket and the exhausted gases respectively v
and v
are the respective velocities and w is the exhaust velocity relative to the space
ship The repulsive force which we shall later designate by F is represented
by the rhs term of The quantities m
m
v
and v
are functions
of time t c is the constant velocity of light
We now turn to the integration of the equation of motion Dier
entiation delivers
A
dm
v
p
v
c
m
dv
p
v
c
m
v
d
p
v
c
dm
v
p
v
c
whereas the energy theorem results in
A
dm
c
p
v
c
m
c
d
p
v
c
dm
c
p
v
c
Here A
and A
are abbreviations for the lhs terms Using now the addition
theorem we build the expressions
B
p
v
c
v
wc
p
w
c
p
v
c
and
B
v
p
v
c
w v
p
w
c
p
v
c
These abbreviations allow the rewriting of the equations and
in the simple forms
A
dm
B
and A
dm
c
B
This allows the elimination of dm
The resulting equation is A
B
c
B
A
The four abbreviations only contain the timedependent variables
v
t and m
t After some straightforward algebra the resulting equation
takes the form
dm
w
p
v
c
c
v
m
dv
p
v
c
c
v
m
w c
v
d
p
v
c
or simplied
m
t
dv
t
dt
dm
t
dt
w t
v
t
c
F
When treating the motion of the parachutist we assumed a given force and we
derived his motion Now we assume a given trajectory x
t of the space ship
and we are determining from the repulsive force F producing this tra
jectory During space travel the space ship leaving earth will rst accelerate
then slow down to zero velocity make a degree turn and accelerate again
to return to earth Slowing down is again necessary prior to landing Since
always dm
dt the slowing down force F must be realized by a change
of sign of the exhaust speed w Therefore w will change its sign and must
x
t Let be the total duration of the trip into space then we can assume
x
t x
sin
t
v
t x
cos
t
v
cos
t
Here x
is the largest distance from earth x
The trajectory satises
the following conditions x
surface of the earth x
x
maximum distance travelled v
v
initial speed and v
v
landing speed Inserting into we obtain for v
fc f
m
t
dm
t
dt
w t
cf
sin
t
f
cos
t
G t
There seem to be several possibilities for the choice of m
t and w t
Whereas w t is quite arbitrary the function m
t has to satisfy the two
boundary conditions m
M
s
m
M
f
where M
s
and M
f
are the
start rest mass of the space rocket andM
f
the landing rest mass respectively
Equations involving more than one independent variable and partial deriva
tives with respect to these variables are called partial dierential equations
The order of a partial dierential equation is the order of the partial deriva
tive of highest order that occurs in the equation The problem of nding a
solution to a given partial dierential equation that will meet certain specied
requirements for a given set of values of the independent variables boundary
conditions constitutes now a boundary value problem The given set of val
ues is then given on a twodimensional curve or on a threedimensional
surface
If the values given by the boundary condition are all zero the boundary
condition is called homogeneous If the values do not vanish the boundary
condition is called inhomogeneous
For instance the vibrations of a membrane are described in cartesian coor
dinates by the socalled twodimensional wave equation
u x y t
c
u
t
where the Laplacian is given by
u
u
x
u
y
u
xx
u
yy
and c is a constant speed describing the material properties of the membrane
Trying to separate the independent variable t we use the setup
v
v
c
T
d
T
dt
c
Since both sides of this equation depend on dierent independent variables
the lefthand side and the righthand side must be equal to the same con
stant that we called
c
We thus have to solve the ordinary dierential
equation
d
T
dt
T
together with some initial or two point condition and the partial dierential
equation
v
xx
v
yy
k
v
where we used k
c
The dimension of k is given by m
Equation
is usually called the Helmholtz equation
One may look for the general solution of or one may be interested
in the vibrations of a rectangular membrane clamped along its boundary
Considering a rectangle as shown in Figure the homogeneous boundary
conditions for a membrane clamped at the boundaries read
x
a
a
b b
b b
y
Figure
Rectangular boundary
va y for b y b
va y for b y b
vx b for a x a
vxb for a x a
To nd a solution to we make the ansatz
X
X
Y
Y
k
Since both second derivatives divided by X and Y respectively must be
constants like
or
we now have the following choice
X
X
Y
Y
k
either a
Y
Y
X
k
X
or b
Y
Y
X
k
X
which we will discuss later in detail The constants are usually called
separation constants They are dependent on the boundary conditions
From Figure one can see that a solution satisfying the boundary condi
tions will be symmetric in both independent variables x and y We
hence take the solutions
Xx A cos x
Y y B cos y
where A and B are integration constants partial amplitudes
If the solutions are used the boundary conditions
determine the separation constants This is the usual method From
and we have
vx a y A cos a B cos y
vx y b A cos x B cos b
These boundary conditions are satised by
a
m b
n m n
Since there appear several even an innite number of solutions for the sep
aration constants we should write
m
n
membrane From we then have
k
mn
m
n
Apparently our boundary value problem with has solutions
only for special discrete values of k If a dierential equation like
or has solutions only for special values of a parameter the prob
lem is called an eigenvalue problem k is the eigenvalue and the solutions
X
m
x A
m
cos
m
x are called eigenfunctions In contrast to this solu
tions of will be called modes or particular solutions
Let us now consider One possible solution is
T t cost
Due to and the denition used for k in the constant angular
frequency has the dimension sec
and assumes the discrete values
mn
c
m
a
n
b
The
mn
are called eigenfrequencies
A partial dierential equation is called linear if it is of the rst degree in
the dependent variable ux y ux y t and their partial derivatives u
x
u
xx
ie if each term either consists of the product of a known function of the inde
pendent variables and the dependent variable or one of its partial derivatives
The equation may also contain a known function of the independent variables
only inhomogeneous equation If this term does not appear the equation is
called homogeneous It is a general property of all linear dierential equations
that two particular solutions can be superposed which means that the sum of
two solutions is again a new solution superposition principle So the gen
eral solution is a superposition of all particular solutions Hence the general
symmetric solution of may be written as
ux y t
X
mn
A
mn
cos
x
a
m
cos
y
b
n
cos
mn
t
where the
mn
are given by and the partial amplitudes A
mn
may be
determined from an initial condition like
ux y fx y
where fx y is a given function ie the deection of the membrane at the
time t
Using now the solutions we have
vx y A cos
p
k
x
cos y
the general solution reads
vx y
X
n
A
n
cos
p
k
n
x
cos
n
y
In order to satisfy boundary conditions the unknowns A
n
n
and k have to
be determined We will discuss this problem later is the starting
point for a boundary point collocation method
Problems
InMathematica a function fx is represented by f1[x] We now dene f1[x]=1/(1-a*xˆ2) and integrate it by the command
F1[x]=Integrate[f1[x],x] which results in ArcTanh p
a x
p
a
Now we verify by dierentiation D[F1[x],x] which gives f1[x]. Be careful the letter l looks very much like the number
If we want to dene an expression like
R[x]=a+b*x+c*xˆ2; f2[x]=x/Sqrt[R[x]]; we may end the ex pression or the line with a semicolon Then Mathematica allows to put
several commands on the same line If there is a last semicolon in the
line as above no outprint will be given
Integration F2[x]=Integrate[f2[x],x] results in
p
a bx cx
c
b Log
b ax
p
c
p
a bx cx
c
To verify again one may give the command D[F2[x],x] yielding
b cx
c
p
a bx cx
b
p
c
b cx
p
a bx cx
c
b cx
p
c
p
a bx cx
Apparently this can be simplied We give the command Simplify[%]
to obtain
x
p
a bx cx
Here the symbol means use the last result generated and means
the nexttolast result and ktimes indicates the kth previous
Mathematica knows how to solve dierential equations
DSolve[x[t]+aˆ2*x[t]==0, x[t],t] This solves x t a
xt with respect to the independent variable
t and gives the solution xt
xt C Cosa t C Sina t
As usual Mathematica drops the sign ! for the multiplication Do not
use quotation marks for the second derivative use two apostrophes
Specifying the integration constants we can plot the result in the
interval x by the command Plot[2 Sin[x],{x,0,2 Pi}]
ffyx
e
x
e
x
C e
x
Cgg
ffyx
e
x
e
x
C e
x
Cgg
x
x
x
x
x
x
Integrate[(x+2)/(x+1),x] x Log x
p
x x
ArcSin
x
a
b
Tan x
a
a
b
Tanx
D[Cosh[xˆ2-3*x+1],x] x Sinh x x
ffyx e
x
C e
x
e
x
x
p
Erfx
gg
ffxt
g m t
e
a t
C
Cgg
Classication of partial dierential equations
Ordinary dierential equations containing only one independent variable and
two point boundary conditions are of minor importance for problems of physics
and engineering Hence we mainly concentrate on partial dierential equa
tions
Partial dierential equations possess a large manifold of solutions Instead
of integration constants arbitrary functions appear in the solution As an
example we consider in a specialization of two independent variables
c
u
xx
x t u
tt
x t
By inserting into it is easy to prove that
ux t fx ct gx ct
is a solution of Here f and g are arbitrary functions Such a solu
tion of a partial dierential equation of order n with n arbitrary functions
is called a general solution If a partial dierential equation contains p inde
pendent variables x y one can nd a complete solution which contains p
integration constants If a function satises the partial dierential equation
and the accompanying boundary conditions and has no arbitrary function or
constants it is called a particular solution If a solution is not obtainable by
assigning particular values to the parameters in the general complete solu
tion it is called a singular solution It describes an envelope of the family of
curves represented by the general solution But the expression is also used for
a solution containing a singular point a singularity We will later consider
such solutions containing singular points where the solution tends to innity
There are other essential properties characterizing various types of partial
dierential equations We give some examples below
u
xx
x y u
yy
x y
This Laplace equation is linear homogeneous and has constant coe"cients
The equation
u
t
x t x
u
x
x t
is of the rst order linear homogeneous and has variable coe"cients For a
rstorder partial dierential equation a boundary problem cannot be formu
lated An example of a nonlinear homogeneous partial dierential equation
is given by
u
x
x y u
yy
x y
and the equation
is given by the linear Poisson equation
u
xx
x y u
yy
x y x y
where is a given function The most general linear partial dierential equa
tion of two independent variables has the form
ax yu
xx
x y bx yu
xy
x y cx yu
yy
x y dx yu
x
x y
ex yu
y
x y gx yux y hx y
Now equations satisfying
b
x y ax ycx y
in a certain domain in the x y plane are called hyperbolic equations
b
x y ax ycx y
characterizes elliptic equations and if
b
x y ax ycx y
the equation is called parabolic Thus an equation is hyperbolic elliptic or
parabolic within a certain domain in the x y plane This is an important
distinctive mark determining the solvability of a boundary problem
Boundary curves or surfaces may be open or closed A closed boundary
surface is one that surrounds the domain everywhere conning it to a nite
surface or volume A simple closed smooth curve is called Jordan curve An
open surface is one that does not completely enclose the domain but lets it
extend to innity in at least one direction Then the Dirichlet boundary
conditions rst boundary value problem x the value ux y on the bound
ary Neumann boundary conditions second boundary value problem x the
value of the normal derivative u n on the boundary and a Cauchy condi
tion xes both value and normal derivative at the same place The Cauchy
condition actually represents an initial condition The normal derivative is
the directional derivative of a function ux y in the direction of the normal
at the point of the boundary where the derivative is taken A generalized
Neumann boundary condition third boundary value problem xes
kx y
ux y
n
lx yux y mx y
on the boundary This condition is of importance in heat ow and uid
mechanics It is possible to prove the solvability of boundary problems
see Table
Table Solvability of boundary problem
Boundary Equation
condition hyperbolic elliptic parabolic
Cauchy
open boundary solvable indeterminate overdeterminate
one closed boundary overdeterminate overdeterminate overdeterminate
Dirichlet
open boundary indeterminate indeterminate solvable
one closed boundary indeterminate solvable overdeterminate
Neumann
open boundary indeterminate indeterminate solvable
one closed boundary indeterminate solvable overdeterminate
In this connection the term solvable means solvable by an analytic solution If
an elliptic boundary problem has two closed boundaries an analytic solution
is no longer possible and singularities have to be accepted
If the value ux y or its derivatives or mx y in vanish on the
boundary the boundary condition is said to be homogeneous If the values on
the boundary do not vanish the boundary condition is called inhomogeneous
A boundary problem is called homogeneous if the dierential equation and
the boundary condition are both homogeneous If the dierential equation
or the boundary condition or both are inhomogeneous then the boundary
problem is said to be inhomogeneous Inhomogeneous boundary conditions
of a homogeneous equation can be transformed into homogeneous conditions
of an inhomogeneous equation This is made possible by the following fact
The general solution ux y of a linear inhomogeneous dierential equation
consists of the superposition of the general solution wx y of the matching
homogeneous equation and a particular solution vx y of the inhomogeneous
equation We consider an example Let
ux y x y
be an inhomogeneous equation with the homogeneous condition uboundary
If we insert the ansatz
ux y wx y vx y
into we obtain
wx y vx y x y
Putting
vx y x y
we obtain an inhomogeneous equation for v and a homogeneous equation for
w which reads
vboundary wboundary
We now see that the inhomogeneous condition that belongs to the
homogeneous equation has been converted into a homogeneous con
dition uboundary and an inhomogeneous equation We thus
have to construct a function vx y satisfying the inhomogeneous condition
for w and producing the term x y by application of on v as in
We now give an example of the conversion of an inhomogeneous boundary
condition for a homogeneous equation into a homogeneous condition matching
an inhomogeneous equation As the homogeneous equation we choose the
Laplacian and write it now in the form
w
xx
x y w
yy
x y
We consider again the rectangle of Figure but instead of the homogeneous
boundary conditions we now use inhomogeneous conditions
wa y for b y b
wa y for b y b
wx b fx for a x a
wxb fx for a x a
Then the corresponding inhomogeneous equation is given by and its
homogeneous boundary conditions are written for ux y To nd a
solution of we have thus to solve
u
xx
x y u
yy
x y x y
together with the homogeneous boundary conditions
ua y for b y b
uxb for a x a
The solution of the inhomogeneous equation is now a superposition
of a general solution u
x y of the homogeneous equation and a particular
solution vx y u
x y A setup u
x y Xx Y y together with the
symmetry expressed by and delivers
u
x y
X
m
A
m
cos
m
x cosh
m
y
To make the problem a little easier we assume x y p const This
tive must give the constant p we use the setup
u
x y c
x
c
xy c
y
c
x c
y c
which seems to be a general possibility where the c
i
are constants We obtain
from
u
c
c
p
Since the particular solution u
has to satisfy separately the boundary condi
tions we have from
c
a
c
ay c
y
c
a c
y c
for b y b hence also for y this gives c
c
c
and
c
p from From we get
u
xb
p
x
c
x c
or with
u
xb
p
x
a
If
m
is given by the homogeneous solution satises too
From we have
p
x
a
X
m
A
m
cos
m
a
x
cosh
m
a
b
This indicates that we have to expand the lefthand side for a x a
into a cosFourier series to satisfy the boundary condition This
procedure gives the A
m
So the solution ux y u
x y u
x y reads
ux y
X
m
A
m
cos
m
a
x
cosh
m
a
y
p
x
a
It satises the inhomogeneous equation for x y p and also the
associated boundary conditions and
The particular solution u
vx y has now to satisfy or
v
xx
x y v
yy
x y p
For vx y we have
u
x y vx y
p
x
a
Then gives an identity Due to the solution satises
the boundary conditions and fx
p
x
a
The inhomogeneous boundary conditions have been homogenized and
according to the solution of reads
wx y ux y vx y
X
m
A
m
cos
m
a
x
cosh
m
a
y
which satises due to the relation It is clear that the method
described can be used for x y const too
Problems
Determine if the following partial dierential equations are hyperbolic
h elliptic e parabolic p or of a mixed type m which means the
type depends on the domain in the x y plane
u
xx
u
yy
e u
xx
u
yy
h
u
x
u
yy
p u
xx
x
u
yy
y
m
u
xx
xu
yy
m u
xx
u
y
p
u
xx
u
xy
u
yy
x p x
u
xx
u
yy
u u
x
h
Using Mathematica calculate
Z
x
x
dx x
x
x
Z
x
x
dx x logx
Z
x
p
x x
dx
p
x x
arcsin
x
d
dx
ab
arctan
a
b
tanx
sec
x
b
a
tan
x
d
dx
coshx
x sinhx
x x
What happens if reads
u
x c
x
c
y
c
x
xx
yy
results in c
x c
c
y ax by c so that c
a c
b c
c
f1[x]=(xˆ2+2)ˆ2*3*xˆ2 F1[x]=Integrate[f1[x],x] D[F1[x],x]; Simplify[%] Explanations of these commands will be given later
Types of boundary conditions and the collocation
method
If a boundary curve or surface can be described by coordinate lines or sur
faces and if the partial dierential equation in question is separable into ordi
nary dierential equations in this coordinate system by a setup like
the boundary problem can be solved quite easily compare the calculations
through To express boundary conditions in a simple way
one must have coordinate surfaces that t the physical boundary of the prob
lem Very often however the situation is more complicated even for partial
dierential equations that are separable in only a few coordinate systems
if the boundary cannot be described in the corresponding coordinate system
On the other hand there are problems that belong to partial dierential equa
tions that cannot be separated at all into ordinary dierential equations As
an example we mention equation
u
xx
x y fx yu
yy
x y
If the coe"cient function fx y cannot be represented by a product fx y
gx hy then a separation of into ordinary dierential equations is
rarely possible However in a problem of plasma physics or
u
zz
r z u
rr
r z
r
u
r
r z
r
ur z
ur z a br cr
cz
u
is an example demonstrating the opposite but has to be solved In this
case the ansatz ur z Rr Zz leads to a separation into two ordinary
dierential equations
R
r
R
r
R
a br cr
k
R
arability in general terms On the other hand it is well known that the
Helmholtz equation u k
u is separable in coordinate systems
only These coordinate systems are
rectangular coordinates x y z circular cylinder coord r z
elliptic cylinder coord z parabolic cylinder coord z
spherical coordinates r prolate spheroidal coord
oblate spheroidal coord parabolic coordinates
conical coordinates r ellipsoidal coordinates
paraboloid coordinates
These coordinate systems are formed from rst and seconddegree sur
faces There are also systems built from fourthdegree surfaces that may
have practical applications
However the theory of separability of partial dierential equations is
of no great interest to us since we will be discussing methods to solve non
separable problems in this book In principle each boundary problem has a
solution The problem is though how to nd the solution In this book
we will discuss a method to solve boundary problems of various kinds see
Table In this table the term boundary tted means that the boundary
can be described by coordinate lines of the coordinate system in which the
partial dierential equation is separable
Table Various boundary problems
di equ boundary example solution
separable tted
yes yes rectangular membrane classical
yes yes circular ring membrane singularity
yes no circ membrane cartes coordinates possible
no no Cassini curve membrane numerical
yes yes boundaries from coordsystems sing nonuniform
no no toroidal problems singularity
no corners nonJordan curve special solution
We will discuss some examples of boundary problems that are mentioned in
Table A membrane described by the Helmholtz equation and bound by
a Cassini curve is such an example Membranes with holes and exhibiting
two closed boundaries and toroidal problems or boundaries with corners will
be treated If there are two boundaries it may happen that they belong
to two dierent coordinate systems This type of problem can be called a
nonuniform boundary problem It can be solved by special methods
In principle each reasonable boundary problem can be solved using the
fact that the general solution of a partial dierential equation contains one or
reasonable function can be expanded into an innite series of partial solutions
Thus an innite set of constants like partial amplitudes A
m
is equivalent to
an arbitrary function
In the case of dierent boundary problems one has two possibilities one
can choose an expression that satises either the dierential equation or the
boundary conditions exactly Coe"cients contained in the expression can
then be used to satisfy the boundary condition or the dierential equation
respectively We give an example We consider the problem
u
xx
u
yy
with the boundary condition on the square jxj jyj
u
x
u x y
u
y
u y x
Instead of using the method that we used to solve we rst write
down another expression satisfying the dierential equation For the
general solution of we require a particular solution of it together with a
solution of the homogeneous equation A particular solution u
is apparently
given by
u
x
y
compare
As the solution u
of the homogeneous equation satisfying the given symme
try conditions we could use or the real parts of xiy
n
n
which deliver the socalled harmonic polynomials Multiplying them by coef
cients we have
u u
u
x
y
a
a
x
x
y
y
a
x
x
y
x
y
x
y
y
This is a solution of for We now must determine the coe"
cients in such a way that the solution satises the boundary conditions
To do this we use a collocation method This means that we choose a set of n
socalled collocation points x
i
y
i
i n on the boundary curve on which
the boundary conditions must be satised The coe"cients a
n
will then be
calculated from the boundary conditions Due to the double
symmetry we need only consider the part of the boundary along x where
u
u
y
a
a
y
y
a
y
y
y
y
In order to determine the three unknown coe"cients a
a
a
we need three
equations This means that we have to choose three collocation points y
y
y
along the boundary line x y We choose y
y
y
and obtain from the three equations i
y
i
a
a
y
i
y
i
a
y
i
y
i
y
i
y
i
The method to determine the unknown coe"cients a
a
and a
from the
boundary conditions is called boundary collocation
On the other hand we can use another ansatz satisfying the boundary
conditions from the start Then the coe"cients have to be determined in
such a way that the dierential equation is satised in the whole domain
interior collocation Since we now need more collocation points to cover the
whole square this method is more expensive
In order rst to satisfy the boundary conditions we make the ansatz
u a
a
a
x
y
a
x
y
a
a
a
x
y
a
x
y
a
x
y
x
y
Each term has to satisfy the boundary conditions separately To obtain
this we calculate the a
a
and so on from whereas the coe"cients
a
a
have to be determined in such a way that the dierential equation is
satised Inserting into we have for the rst term
a
a
y
a
a
a
y
a
y
which is solved by a
a
a
The second term yields
a
a
a
y
a
y
a
a
a
y
a
a
y
a
y
a
y
which delivers a
a
a
a
We now have the
ansatz
u a
x
y
x
y
a
x
y
satises the dierential equation Inserting we obtain
a
x
y
a
x
y
x
y
x
y
For this expression with two unknowns we have to choose two collocation
points within the square domain We can dene x
y
and
x
y
Inserting this into we can calculate a
and a
to
obtain an approximate solution
Since collocation will be discussed later on in detail we postpone problems
Dierential equations as models for nature
In the last sections we have discussed boundary conditions but where are
the dierential equations coming from There are apparently two methods
to derive dierential equations as models for phenomena in nature and en
gineering intuition and derivation from fundamental laws of nature like the
energy theorem etc We will give two examples the rst for intuition Let
us assume we want to study the spread of an epidemic disease By St we
designate the number of healthy persons It will be the number of persons
contracting the disease and let Rt be the number of persons being immune
against the disease Apparently one then has a theorem for the conservation
of the number of people if nobody dies or is born during the short time period
considered This number balance reads
St It Rt const
Now intuition and experience come into play Apparently the number of
persons newly infected will be proportional to the number St of healthy
persons and to the number It of sick persons
dSt
dt
St It
The parameter can be called rate of infection On the other hand the
number of persons becoming immune after having recovered from the disease
would be
dRt
dt
It
where can be called rate of immunization We thus have three equations
dR
dt
dS
dt
dI
dt
I
Calculation of I and dIdt from and inserting into gives a
nonlinear dierential equation of second order
d
S
dt
S
d lnS
dt
dS
dt
S
We now assume the initial condition St S
Thus the number of
healthy persons at t t
is given by S
Neglecting d lnSdt we obtain
near t
d
S
dt
dS
dt
S
Thus the curvature S
t
depends on the conditions S
and S
respectively This represents exactly the empirical basic theorem of epidemi
ology An epidemic starts if the number S
of healthy but predisposed
persons exceeds a specic threshold
Another method to derive dierential equations is given by a derivation from
fundamental laws Whereas boundary conditions describe actual situations
and are used to specify an actual particular solution by determining integra
tion components dierential equations of order n deliver the general solution
containing n integration constants For example the dierential equation de
scribing transverse vibrations of a thin uniform plate can be derived from the
empirical law of Hookean deformation plus the energy theorem A more
elegant way of deriving dierential equations is variational calculus Let us
assume that the eigenfrequencies of transversal vibrations of plates of vary
ing thickness are suddenly of practical interest eg for the investigation of
ssures in an airplane wing
The fundamental problem of the calculus of variation is to determine the
minimum of the integral
Jux y t
ZZZ
F x y t ux y t u
x
x y t u
y
x y t u
t
x y t
u
xx
x y t u
yy
x y t u
xy
x y tdxdydt
for a given functional F The minimum of the integral J delivers this function
ux y t which actually makes J a minimum This function is determined
by the Euler equations which will be derived later on in section
F
u
x
F
u
x
y
F
u
y
t
F
u
t
F
u
xx
F
u
xy
F
u
yy
x
F x y t u u
x
u
y
with respect to its variables u u
x
etc
For a physical or engineering problem the functional F is given by the La
grange functional dened by the dierence kinetic energy T minus potential
energy # If we assume that ux y t is the local transversal deection of a
plate the kinetic energy of a plate with modestly varying thickness hx y is
given by
T
o
Z
G
Z
hx y
u
t
dxdy
Here
is the constant surface mass density per unit of thickness so that
hx y is the local surface mass density The surface integral T is taken
over the area G of the plate
In order to nd the minimum of we have to vary the integral
t
Z
t
Z
G
Z
hx y
u
t
dxdy #
A
dt
is the variational symbol and # is the total elastic energy ie the local
elastic potential integrated over the domain G
Using the designations E for Youngs modulus and for Poissons ratio
the local elastic free energy per unit volume of the plate is given by
fx y z u
xx
u
yy
u
xy
z
E
u
xx
u
yy
u
xy
u
xx
u
yy
The total elastic energy is then given by integration over the volume of the
plate Integration rst over z
dz alone from hx y to hx y delivers
h
x y Thus the total elastic energy # is then given by
#x y u
xx
u
yy
u
xy
E
ZZ
n
h
x yu
xx
u
yy
h
x yu
xy
u
xx
u
yy
o
dxdy
If the plate has to carry a load px y then we have to add the term
Z
G
Z
px y ux ydxdy
This term describes the work done by the external forces when the points on
the plate are displaced by the displacement u Now the total functional is
F
hu
t
pu
Eh
u
xx
u
yy
u
xx
u
yy
u
xy
u
xx
u
yy
Since then F
u
x
F
u
y
takes the form
F
u
t
F
u
t
x
F
u
xx
xy
F
u
xy
y
F
u
yy
From the rst two terms of we thus obtain
F
u
p F
u
t
hu
t
t
F
u
t
hu
tt
Furthermore we then get the plate equation for varying thickness hx y in
the form
Eh
u
xxxx
u
xxyy
u
yyyy
hu
tt
Eh
fh h
x
u
xxx
u
xyy
h
y
u
yyy
u
xxy
u
xx
h
x
hh
xx
h
y
hh
yy
u
yy
h
x
hh
xx
h
y
hh
yy
u
xy
h
x
h
y
hh
xy
g px y
This is the plate equation for weakly varying thickness hx y
This derivation does however not answer the question where the energy
theorem like or a consequence of Hookes law comes from
There are people believing that these fundamental laws are preexistent in
nature or have been originated by a creator Modern natural philosophy
tends to another view So mathematicians know that dierential equations
are invariant under special transformations of coordinates If for instance the
equation of motion is submitted to a simple translation along the x coordinate
axis then the momentum mv
x
remains constant it will be conserved Emmy
Noether has shown that the invariance of a dierential equation against a
transformation has the consequence of the existence of a conservation theorem
for a related physical quantity Thus the energy theorem is a consequence of
the invariance of the equation of motion under a translation along the time
axis Human beings assume that the laws of nature are independent against
a time translation But intelligent lizards as coldblooded animals would
ton But the results of the lizard physics would be the same as in human
physics Apparently human assumptions on coordinate transformations cre
ate the laws we nd in nature But how could we nd out if these laws are
correct and true In his Discours de la M$ethode Poincar
e has shown that
there are always several true models or theories describing natural phenom
ena As an example we can mention that Dives theory of elliptic waves
and the special relativity theory give exactly the same results up to the or
der vc
Why have we chosen special relativity to describe nature When
we have to decide between two fully equivalent theories we should take into
account
Aesthetic points of view
Machs principle of economic thinking
The extensibility of a theory to broader elds of applications like the
extension of special to general relativity respectively
Problems
Derive the equation for transverse vibrations ux y t for a plate with
constant thickness h const see section
Try to solve using Mathematica Not possible
Type the command
gives St
e
t S
C
S
C
and plot the result But there is now a problem the solution is not given
by S[t]= . So it is necessary to dene a new function ut which gives a value for any arbitrary t This is done by replacing t by t_ We rst
select the integration constants C[2]=0, C[1]1=S0*(S0* ) to obtain St S and write u[t_]=Exp[t*(S0 )]*S0 This ut may be plotted for given arbitrary values of S
Learn partial derivatives Dene
u[x,y]=xˆ2+a*xˆ3*yˆ4+yˆ3 D[u[x,y],x] x a x y D[u[x,y],{x,2}] a x y D[u[x,y],y] a x y y