Professional Documents
Culture Documents
Partial Differential Equations - LibreTexts
Partial Differential Equations - LibreTexts
EQUATIONS
Erich Miersemann
Universität Leipzig
Universität Leipzig
Partial Differential Equations
Erich Miersemann
This text is disseminated via the Open Education Resource (OER) LibreTexts Project (https://LibreTexts.org) and like the hundreds
of other texts available within this powerful platform, it is freely available for reading, printing and "consuming." Most, but not all,
pages in the library have licenses that may allow individuals to make changes, save, and print this book. Carefully
consult the applicable license(s) before pursuing such effects.
Instructors can adopt existing LibreTexts texts or Remix them to quickly build course-specific resources to meet the needs of their
students. Unlike traditional textbooks, LibreTexts’ web based origins allow powerful integration of advanced features and new
technologies to support learning.
The LibreTexts mission is to unite students, faculty and scholars in a cooperative effort to develop an easy-to-use online platform
for the construction, customization, and dissemination of OER content to reduce the burdens of unreasonable textbook costs to our
students and society. The LibreTexts project is a multi-institutional collaborative venture to develop the next generation of open-
access texts to improve postsecondary education at all levels of higher learning by developing an Open Access Resource
environment. The project currently consists of 14 independently operating and interconnected libraries that are constantly being
optimized by students, faculty, and outside experts to supplant conventional paper-based books. These free textbook alternatives are
organized within a central environment that is both vertically (from advance to basic level) and horizontally (across different fields)
integrated.
The LibreTexts libraries are Powered by MindTouch® and are supported by the Department of Education Open Textbook Pilot
Project, the UC Davis Office of the Provost, the UC Davis Library, the California State University Affordable Learning Solutions
Program, and Merlot. This material is based upon work supported by the National Science Foundation under Grant No. 1246120,
1525057, and 1413739. Unless otherwise noted, LibreTexts content is licensed by CC BY-NC-SA 3.0.
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not
necessarily reflect the views of the National Science Foundation nor the US Department of Education.
Have questions or comments? For information about adoptions or adaptions contact info@LibreTexts.org. More information on our
activities can be found via Facebook (https://facebook.com/Libretexts), Twitter (https://twitter.com/libretexts), or our blog
(http://Blog.Libretexts.org).
1: INTRODUCTION
1.0: INTRODUCTION
1.1: EXAMPLES
1.2: ORDINARY DIFFERENTIAL EQUATIONS
1.3: PARTIAL DIFFERENTIAL EQUATIONS
1.E: INTRODUCTION (EXERCISES)
3: CLASSIFICATION
3.1: LINEAR EQUATIONS OF SECOND ORDER
3.1.1: NORMAL FORM IN TWO VARIABLES
3.2: QUASILINEAR EQUATIONS OF SECOND ORDER
3.2.1: QUASILINEAR ELLIPTIC EQUATIONS
3.3: SYSTEMS OF FIRST ORDER
3.3.1: EXAMPLES
3.4: SYSTEMS OF SECOND ORDER
3.4.1: EXAMPLES
3.5: THEOREM OF CAUCHY-KOVALEVSKAYA
3.5.1 APPENDIX: REAL ANALYTIC FUNCTIONS
3.E: CLASSIFICATION (EXERCISES)
BACK MATTER
INDEX
4: HYPERBOLIC EQUATIONS
Here we consider hyperbolic equations of second order, mainly wave equations.
1 3/30/2022
INDEX
5: FOURIER TRANSFORM
5.1: DEFINITION AND PROPERTIES
5.1.1: PSEUDODIFFERENTIAL OPERATORS
5.E: FOURIER TRANSFORM (EXERCISES)
6: PARABOLIC EQUATIONS
Here we consider linear parabolic equations of second order.
BIBLIOGRAPHY
BACK MATTER
INDEX
GLOSSARY
2 3/30/2022
CHAPTER OVERVIEW
1: INTRODUCTION
Thumbnail: Visualization of heat transfer in a pump casing, created by solving the heat equation. Heat is being generated internally in the
casing and being cooled at the boundary, providing asteady state temperature distribution. (CC-BY-SA-3.0; Wikipedia A1).
1.0: INTRODUCTION
1.1: EXAMPLES
1.2: ORDINARY DIFFERENTIAL EQUATIONS
1.3: PARTIAL DIFFERENTIAL EQUATIONS
1.E: INTRODUCTION (EXERCISES)
1 3/30/2022
1.0: Introduction
Preface
These lecture notes are intented as a straightforward introduction to partial differential equations which can serve as a textbook for
undergraduate and beginning graduate students.
For additional reading we recommend following books: W. I. Smirnov [21], I. G. Petrowski [17], P. R. Garabedian [8], W. A.
Strauss [23], F. John [10], L. C. Evans [5] and R. Courant and D. Hilbert[4] and D. Gilbarg and N. S. Trudinger [9]. Some material
of these lecture notes was taken from some of these books.
Introduction
Ordinary and partial differential equations occur in many applications. An ordinary differential equation is a special case of a
partial differential equation but the behavior of solutions is quite different in general caused by the fact that the functions for which
we are looking at are functions of more than one independent variable.
Equation
$$F(x,y(x),y'(x),\ldots,y^{(n)})=0\]
is an ordinary differential equation of n-th order for the unknown function y(x), where F is given.
An important problem for ordinary differential equations is the initial value problem
′
y (x) = f (x, y(x))
y(x0 ) = y0 ,
where f is a given real function of two variables x,y and x ,, y are given real numbers.
0 0
(ii) There is a constant K such that |f (x, y)| ≤ K for all (x, y) ∈ Q.
(ii) Lipschitz condition: There is a constant L such that
|f (x, y2 ) − f (x, y1 )| ≤ L| y2 − y1 | (1.0.2)
′′ ′
y (x) = f (x, y(x), y (x))
y(x0 ) = y0 , y(x1 ) = y1 .
Example 1.1.2:
ux = uy , where u = u(x, y) . A change of coordinates transforms this equation into an equation of the first example. Set
ξ = x + y, η = x − y , then
ξ +η ξ −η
u(x, y) = u ( , ) =: v(ξ, η). (1.1.1)
2 2
Assume u ∈ C , then
1
1
vη = (ux − uy ). (1.1.2)
2
If u = u , then v = 0 and vice versa, thus v = w(ξ) are solutions for arbitrary C -functions w(ξ). Consequently, we have a
x y η
1
large class of solutions of the original partial differential equation: u = w(x + y) with an arbitrary C -function w.
1
Example 1.1.3:
A necessary and sufficient condition such that for given C -functions M , N the integral
1
P1
is independent of the curve which connects the points P0 with P1 in a simply connected domain Ω ⊂R
2
is that the partial
differential equation (condition of integrability)
My = Nx (1.1.4)
in Ω.
C -function. It follows from Gauss theorem that these are all C -solutions of the above differential equation.
2 1
for given C -functions M , N . Then we seek a C -function μ(x, y) such that μM dx + μN dy is a total differential, i. e., that
1 1
(μM ) = (μN )
y xis satisfied. This is a linear partial differential equation of first order for μ :
$$
M\mu_y-N\mu_x=\mu (N_x-M_y).
\]
Example 1.1.5:
Two C -functions u(x, y) and v(x, y) are said to be functionally dependent if
1
ux uy
det ( ) = 0, (1.1.6)
vx vy
which is a linear partial differential equation of first order for u if v is a given C -function. A large class of solutions is given
1
by
Set f (z) = u(x, y) + iv(x, y) , where z = x + iy and u, v are given C (Ω)-functions. Here Ω is a domain in R . If the
1 2
function f (z) is differentiable with respect to the complex variable z then u, v satisfy the Cauchy-Riemann equations
ux = vy , uy = −vx . (1.1.8)
It is known from the theory of functions of one complex variable that the real part u and the imaginary part v of a differentiable
function f (z) are solutions of the Laplace equation
△u = 0, △v = 0, (1.1.9)
where △u = u xx + uyy .
Let u(x, t) be the temperature of a point x ∈ Ω at time t , where Ω ⊂ R is a domain. Then u(x, t) satisfies in Ω × [0, ∞) the
3
heat equation
ut = k△u, (1.1.11)
where u 0 (x) is given, is an initial condition associated to the above heat equation. The condition
u(x, t) = h(x, t), x ∈ ∂Ω, t ≥ 0, (1.1.13)
v = g(x) on ∂Ω.
where u = u(x, t), c is a positive constant, describes oscillations of membranes or of three dimensional domains, for example.
In the one-dimensional case
2
utt = c uxx (1.1.15)
where u 0, u1 are given functions. Thus the initial position and the initial velocity are prescribed.
If the string is finite one describes additionally boundary conditions, for example
$$
u(0,t)=0,\ \ u(l,t)=0\ \ \mbox{for all}\ t\ge 0.
\]
Euler equation
Let u ∈ V be a solution of (P), then
$$\frac{d}{dx}f_{u'}(x,u(x),u'(x))=f_u(x,u(x),u'(x))\]
in (a, b).
For fixed [a, b] with ϕ(a) = ϕ(b) = 0 and real ϵ , |ϵ| < ϵ , set g(ϵ) = E(u + ϵϕ) . Since g(0) ≤ g(ϵ)
ϕ ∈ C
2
0 it follows
′
g (0) = 0 . Integration by parts in the formula for g (0) and the following basic lemma in the calculus of variations imply
′
Euler's equation.
∫ h(x)ϕ(x) dx = 0 (1.2.1)
a
for all ϕ ∈ C 1
0
(a, b) . Then h(x) ≡ 0 on (a, b).
Proof. Assume h(x 0) >0 for an x0 ∈ (a, b) , then there is a δ >0 such that (x0 − δ, x0 + δ) ⊂ (a, b) and h(x) ≥ h(x0 )/2 on
(x − δ, x + δ) .
0 0
Set
$$
\phi(x)
=\left\{
2
2 2
(δ − |x − x0 | ) x ∈ (x0 − δ, x0 + δ)
(1.2.2)
0 x ∈ (a, b) ∖ [ x0 − δ, x0 + δ]
\right. .
\]
Thus ϕ ∈ C 1
0
(a, b) and
i=1
in Ω.
Proof. Exercise. Hint: Extend the above fundamental lemma of the calculus of variations to the case of multiple integrals. The
interval (x − δ, x + δ) in the definition of ϕ must be replaced by a ball with center at x and radius δ .
0 0 0
⎛ ⎞ ⎛ ⎞
∂ ux ∂ ux
1 2
In fact, the additional assumption u ∈ C (Ω) is superfluous since it follows from regularity considerations for quasilinear
2
elliptic equations of second order, see for example Gilbarg and Trudinger [9].
Let Ω = R . Each linear function is a solution of the minimal surface equation (1.3.2). It was shown by Bernstein [2] that
2
there are no other solutions of the minimal surface quation. This is true also for higher dimensions n ≤ 7 , see Simons [19].
If n ≥ 8 , then there exists also other solutions which define cones, see Bombieri, De Giorgi and Giusti [3].
The linearized minimal surface equation over u ≡ 0 is the Laplace equation △u = 0 . In R linear functions are solutions but
2
also many other functions in contrast to the minimal surface equation. This striking difference is caused by the strong
nonlinearity of the minimal surface equation.
More general minimal surfaces are described by using parametric representations. An example is shown in Figure 1.2.2.21. See
[18], pp. 62, for example, for rotationally symmetric minimal surfaces.
¯
¯¯¯
for all ϕ ∈ C 1
(Ω) . Assume additionally u ∈ C 2
(Ω) , then u is a solution of the Neumann type boundary value problem
n
∂
∑ Fu − Fu = 0 in Ω
x
i
∂xi
i=1
∑ Fux νi − gu = 0 on ∂Ω,
i
i=1
∂u
= h on ∂Ω.
∂ν
ν ⋅ T u = cos γ on ∂Ω,
The above problem describes the ascent of a liquid, water for example, in a vertical cylinder with cross section Ω. Assume the
gravity is directed downwards in the direction of the negative x -axis. Figure 1.2.2.3 shows that liquid can rise along a vertical
3
wedge which is a consequence of the strong non-linearity of the underlying equations, see Finn [7]. This photo was taken from
[15].
Q1.1
Find nontrivial solutions u of
$$
u_xy-u_yx=0 \ .
\]
Q1.2
Prove: In the linear space C 2 2
(R ) there are infinitely many linearly independent solutions of △u = 0 in R . 2
Hint: Real and imaginary part of holomorphic functions are solutions of the Laplace equation.
Q1.3
Find all radially symmetric functions which satisfy the Laplace equation in
R ∖ {0} for n ≥ 2 . A function u is said to be radially symmetric if u(x) = f (r), where r = (∑ .
n n 2 1/2
x )
i i
′
Hint: Show that a radially symmetric u satisfies △u = r 1−n
(r
n−1 ′
f ) by using ∇u(x) = f ′
(r)
x
r
.
Q1.4
Prove the basic lemma in the calculus of variations:
Let Ω ⊂ R be a domain and f ∈ C (Ω) such that
n
∫ f (x)h(x) dx = 0 (1.E.1)
Ω
for all h ∈ C 2
0
(Ω) . Then f ≡0 in Ω.
Q1.5
Write the minimal surface equation (1.2.2.1) as a quasilinear equation of second order.
Q1.6
Prove that a sufficiently regular minimizer in
¯
¯¯¯
C (Ω) of
1
∑ Fu νi − gu = 0 on ∂Ω,
x
i
i=1
∂Ω .
Hint: The angle between two surfaces is by definition the angle between the two associated normals at the intersection of the
surfaces.
Q1.8
¯
¯¯¯
Let Ω be bounded and assume u ∈ C 2
(Ω) is a solution of
div T u = C in Ω
∇u
ν ⋅ −−−−−−−− = cos γ on ∂Ω,
2
√ 1 + |∇u|
where C is a constant.
Prove that
$$
C={|\partial\Omega|\over|\Omega|}\cos\gamma\ .
\]
Hint: Integrate the differential equation over Ω.
Q1.9
Assume Ω = B R (0) is a disc with radius R and the center at the origin.
−−−−−−
Show that radially symmetric solutions u(x) = w(r), r = √x 2
1
+x
2
2
, of the capillary boundary value problem are solutions of
′ ′
rw
( − −−−−−) = κrw in 0 < r < R
√ 1 + w′2
′
w
− −−−−− = cos γ if r = R.
√ 1 + w′2
Remark. It follows from a maximum principle of Concus and Finn [7] that a solution of the capillary equation over a disc must be
radially symmetric.
Q1.10
Find all radially symmetric solutions of
′ ′
rw
( − −−−−−) = C r in 0 < r < R
√ 1 + w′2
′
w
− −−−−− = cos γ if r = R.
√ 1 + w′2
1 3/30/2022
2.0: Prelude to First Order Equations
For a given sufficiently regular function F the general equation of first order for the unknown function ) is
$$F(x,u,\nabla u)=0\]
in n . The main tool for studying related problems is the theory of ordinary differential equations. This is quite different for systems
of partial differential of first order. The general linear partial differential equation of first order can be written as
$$\sum_{i=1}^na_i(x)u_{x_i}+c(x)u=f(x)\]
for given functions ai, c and f . The general quasilinear partial differential equation of first order is
$$\sum_{i=1}^na_i(x,u)u_{x_i}+c(x,u)=0.\]
Assume there is a C -solution z = u(x, y). This function defines a surface S which has at P
1
= (x, y, u(x, y)) the normal
1
N = −−−− −−−− (−ux , −uy , 1) (2.1.2)
2
√ 1 + |∇u|
Set p = u (x, y), q = u (x, y) and z = u(x, y). The tuple (x, y, z, p, q) is called surface element and the tuple (x, y, z) support of
x y
the surface element. The tangential plane is defined by the surface element. On the other hand, differential equation (2.1.1)
defines at each support (x, y, z) a bundle of planes if we consider all (p, q) satisfying this equation. For fixed (x, y), this family of
planes Π(λ) = Π(λ; x, y) is defined by a one parameter family of ascents p(λ) = p(λ; x, y), q(λ) = q(λ; x, y).
The envelope of these planes is a line since
′ 1
L : l(σ) = x0 + σ x (τ0 ), σ ∈ R , (2.1.6)
Tx
0
= Π(λ0 ) and consider two planes
Π(λ0 ) : z − z0 = (x − x0 )p(λ0 ) + (y − y0 )q(λ0 )
it follows
′ ′
a1 p (λ0 ) + a2 q (λ0 ) = 0. (2.1.10)
Consequently
′
x (τ )
′ ′
(x (τ ), y (τ )) = (a1 (x(τ ), y(τ )), a2 (x(τ ), y(τ )), (2.1.11)
a1 (x(τ , y(τ ))
It follows x (t) = a
′
1 (x,
′
y), y (t) = a2 (x, y) . We denote x(τ (t)) by x(t) again.
Now we consider the initial value problem
′ ′
x (t) = a1 (x, y), y (t) = a2 (x, y), x(0) = x0 , y(0) = y0 . (2.1.13)
From the theory of ordinary differential equations it follows (Theorem of Picard-Lindelöf) that there is a unique solution in a
neighbourhood of t = 0 provided the functions a , a are in C . From this definition of the curves (x(t), y(t)) is follows that
1 2
1
the field of directions (a (x , y ), a (x , y )) defines the slope of these curves at (x(0), y(0)).
1 0 0 2 0 0
Definition. The differential equations in (2.1.13) are called characteristic equations or characteristic system and solutions of the
associated initial value problem are called characteristic curves.
Definition. A function ϕ(x, y) is said to be an integral of the characteristic system if ϕ(x(t), y(t)) = const. for each characteristic
curve. The constant depends on the characteristic curve considered.
Proposition 2.1. Assume ϕ ∈ C is an integral, then u = ϕ(x, y) is a solution of (2.1.1).
1
Proof. Consider for given (x 0, y0 ) the above initial value problem (2.1.13). Since ϕ(x(t), y(t)) = const. it follows
′ ′
ϕx x + ϕy y =0 (2.1.14)
Thus
$$
\phi_x(x_0,y_0)a_1(x_0,y_0)+\phi_y(x_0,y_0)a_2(x_0,y_0)=0.
\]
□
Remark. If ϕ(x, y) is a solution of equation (2.1.1) then also H (ϕ(x, y)), where H (s) is a given C -function. 1
Consider
a1 ux + a2 uy = 0, (2.1.15)
is constant along each characteristic curve. Consequently, see Proposition 2.1, u = a2 x − a1 y is a solution of the differential
equation. From an exercise it follows that
u = H (a2 x − a1 y), (2.1.19)
where H (s) is an arbitrary C -function, is also a solution. Since u is constant when a x − a y is constant, equation (2.1.19)
1
2 1
defines cylinder surfaces which are generated by parallel straight lines which are parallel to the (x, y)-plane, see Figure 2.1.2.
Example 2.1.2:
Consider the differential equation
x ux + y uy = 0. (2.1.20)
where A, B are arbitrary constants. Then an integral is y/x, x ≠ 0 , and for a given C -function the function u = H (x/y) is a
1
solution of the differential equation. If y/x = const. , then u is constant. Suppose that H (s) > 0 , for example, then u defines
′
Example 2.1.3:
Consider the differential equation
y ux − x uy = 0. (2.1.23)
If follows
′ ′
x x + yy = 0, (2.1.25)
or, equivalently,
d 2 2
(x + y ) = 0, (2.1.26)
dt
which implies that x + y = const. along each characteristic. Thus, rotationally symmetric surfaces defined by
2 2
Example 2.1.4:
The associated characteristic equations to
ay ux + bx uy = 0, (2.1.27)
It follows bx x′
− ay y
′
=0 , or equivalently,
d 2 2
(b x − ay ) = 0. (2.1.29)
dt
Solutions of the differential equation are u = H (bx − ay ) , which define surfaces which have a hyperbola as the intersection
2 2
with planes parallel to the (x, y)-plane. Here H (s) is an arbitrary C -function, H (s) ≠ 0 . 1 ′
homogeneous linear equations in two variables. The additional equation 3 follows from
′ ′ ′
z (τ ) = p(λ)x (τ ) + q(λ)y (τ )
= p a1 + q a2
= a3 ,
see also Section 2.3, where the general case of nonlinear equations in two variables is considered.
ψx + ψu ux = 0, ψy + ψu uy = 0. (2.2.1.1)
Assume ψ u ≠0 , then
ψx ψy
ux = − , uy = − . (2.2.1.2)
ψu ψu
where z := u .
We consider the associated system of characteristic equations
′
x (t) = a1 (x, y, z)
′
y (t) = a2 (x, y, z)
′
z (t) = a3 (x, y, z).
One arrives at this system by the same arguments as in the two-dimensional case above.
Proposition 2.2. (i) Assume w ∈ C , w = w(x, y, z), is an integral, i. e., it is constant along each fixed solution of (2.2.1.1), then
1
(ii) The function z = u(x, y), implicitly defined through ψ(x, u, z) = const. , is a solution of (2.2.1), provided that ψz ≠0 .
(iii) Let z = u(x, y) be a solution of (2.2.1) and let (x(t), y(t)) be a solution
of
′ ′
x (t) = a1 (x, y, u(x, y)), y (t) = a2 (x, y, u(x, y)), (2.2.1.3)
then z(t) := u(x(t), y(t)) satisfies the third of the above characteristic equations.
Proof. Exercise.
be a regular curve in R and denote by C the orthogonal projection of Γ onto the (x, y)-plane, i. e.,
3
$$
\mathcal{C}:\ \ x=x_0(s),\ \ y=y_0(s).
\]
Initial value problem of Cauchy: Find a C -solution u = u(x, y) of
1
(⋆) such that u(x0 (s), y0 (s)) = z0 (s) , i. e., we seek a
surface S defined by z = u(x, y) which contains the curve Γ.
x(s, 0) = x0 (s)
y(s, 0) = y0 (s)
z(s, 0) = z0 (s).
Let x = x(s, t) , y = y(s, t) , z = z(s, t) be the solution, s ≤ s ≤ s , |t| < η for an η > 0 . We will show that this set of curves,
1 2
see Figure 2.2.2.1, defines a surface. To show this, we consider the inverse functions s = s(x, y) , t = t(x, y) of x = x(s, t) ,
y = y(s, t) and show that z(s(x, y), t(x, y)) is a solution of the initial problem of Cauchy. The inverse functions s and t exist in a
neighborhood of t = 0 since
∂(x, y) ∣ xs xt ∣
∣ ′ ′
det =∣ ∣ = x (s)a2 − y (s)a1 ≠ 0, (2.2.2.2)
∣ 0 0
∂(s, t) t=0 ∣ ys yt ∣
t=0
The following calculation shows that u is also a solution of the differential equation (⋆).
a1 ux + a2 uy = a1 (zs sx + zt tx ) + a2 (zs sy + zt ty )
= zs (a1 sx + a2 sy ) + zt (a1 tx + a2 ty )
= zs (sx xt + sy yt ) + zt (tx xt + ty yt )
= a3
since 0 = s t = sx xt + sy yt and 1 = t
t = tx xt + ty yt .
(ii) Uniqueness. Suppose that v(x, y) is a second solution. Consider a point (x , y ) in a neighborhood of the curve (x
′ ′
0 (s), y(s)) ,
s − ϵ ≤ s ≤ s + ϵ , ϵ > 0 small. The inverse parameters are s = s(x , y ) , t = t(x , y ) , see Figure 2.2.2.2.
′ ′ ′ ′ ′ ′
1 2
be the solution of the above initial value problem for the characteristic differential equations with the initial data
′ ′ ′ ′ ′ ′
x(s , 0) = x0 (s ), y(s , 0) = y0 (s ), z(s , 0) = z0 (s ). (2.2.2.6)
According to its construction this curve is on the surface S defined by u = u(x, y) and u(x , y ′ ′ ′ ′
) = z(s , t ) . Set
then
′ ′ ′ ′
ψ (t) = vx x + vy y − z
= xx a1 + vy a2 − a3 = 0
and
′ ′ ′
ψ(0) = v(x(s , 0), y(s , 0)) − z(s , 0) = 0 (2.2.2.8)
since v is a solution of the differential equation and satisfies the initial condition by assumption. Thus, ψ(t) ≡ 0 , i. e.,
Set t = t , then
′
′ ′ ′ ′
v(x , y ) − z(s , t ) = 0, (2.2.2.10)
Remark. In general, there is no uniqueness if the initial curve Γ is a characteristic curve, see an exercise and Figure 2.2.2.3, which
illustrates this case.
Examples
Example 2.2.2.1:
is given by
i. e.,
x = t + s, y = t + 1, z = z0 (s). (2.2.2.16)
It follows s = x − y + 1, t = y − 1 and that u = z 0 (x − y + 1) is the solution of the Cauchy initial value problem.
Here the constants k are positive, these constants define the velocity of the reactions in consideration, and the function u (y)
j 0
is given. The variable x is the time and y is the height of a tube, for example, in which the chemical reaction takes place, and u
is the concentration of the chemical substance.
In contrast to our previous assumptions, the initial data are not in C
1
. The projection C1 ∪ C2 of the initial curve onto the
(x, y)-plane has a corner at the origin, see Figure 2.2.2.4.
It follows x = t + c , y = t + c with constants c . Thus the projection of the characteristic curves on the (x, y)-plane are
1 2 j
straight lines parallel to y = x . We will solve the initial value problems in the domains Ω and Ω , see Figure 2.2.2.4, 1 2
separately.
(i) The initial value problem in Ω . The initial data are
1
It follows
x = x(s, t) = t + s, y = y(s, t) = t. (2.2.2.21)
Thus
′ −k1 (t+s)
z (t) = (k0 e + k2 )(1 − z), z(0) = 0. (2.2.2.22)
Consequently
k0
−k1 x −k1 (x−y)
u1 (x, y) = 1 − exp( e − k2 y − k0 k1 e ) (2.2.2.24)
k1
$$
\lim_{x\to\infty} u_1(x,y)=1-e^{-k_2y}.
\]
(ii) The initial value problem in Ω . The initial data are here
2
It follows
Thus
′ −k1 t
z (t) = (k0 e + k2 )(1 − z), z(0) = 0. (2.2.2.27)
Consequently
k0 k0
−k1 x
u2 (x, y) = 1 − (1 − u0 (y − x)) exp( e − k2 x − ) (2.2.2.29)
k1 k1
is the solution in Ω .
2
If x = y , then
k0 k0
−k1 x
u1 (x, y) = 1 − exp( e − k2 x − )
k1 k1
k0 k0
−k1 x
u2 (x, y) = 1 − (1 − u0 (0)) exp( e − k2 x − ).
k1 k1
is known.
Let
is satisfied. Set
g(s) = ϕ(x0 (s), y0 (s)) (2.2.2.33)
Example 2.2.2.3:
Consider equation
ux + uy = 0 (2.2.2.34)
and
u0 (ϕ + 1) = u0 (x − y + 1) (2.2.2.37)
Definition. Equations (2.3.2)--(2.3.6) are said to be characteristic equations of equation (2.3.1) and a solution
$$(x(t),y(t),z(t),p(t),q(t))\]
of the characteristic equations is called characteristic strip or Monge curve.
and for which v = p , v = q at (x , y ). In the case of quasilinear equations these set of planes is a bundle of planes which all
x y 0 0
contain a fixed straight line, see Section 2.1. In the general case of this section the situation is more complicated.
Consider the example
2 2
p +q = f (x, y, z), (2.3.7)
where f is a given positive function. Let E be a plane defined by z = v(x, y) and which contains (x 0, y0 , z0 ) . Then the normal on
the plane E directed downward is
$${\bf N}=\frac{1}{\sqrt{1+|\nabla v|^2}}(p,q,-1),\]
where p = v (x , y ) , q = v (x , y ) . It follows from (2.3.7) that the normal N makes a constant angle with the z -axis, and the
x 0 0 y 0 0
$$p(\lambda)=p(\lambda;x,y,z),\ q(\lambda)=q(\lambda;x,y,z)\]
of solutions of (2.3.1). These (p(λ), q(λ)) define a family Π(λ) of planes.
Let
$${\bf x}(\tau)=(x(\tau),y(\tau),z(\tau))\]
be a curve on the surface S which touches at each point its Monge cone, see Figure 2.3.4. Thus we assume that at each point of the
surface S the associated tangent plane coincides with a plane from the family Π(λ) at this point.
The above straight line l is the limit of the intersection line of two neighboring planes which envelopes the Monge cone:
z − z0 = (x − x0 )p(λ0 ) + (y − y0 )q(λ0 )
Assume x (τ
′
0) ≠0 and Fp ≠0 , then we obtain from (2.3.10), (2.3.11)
′
y (τ0 ) Fq
= ,
′
x (τ0 ) Fp
′ ′ ′ ′
x (τ ) = (x (τ ), y (τ ), z (τ ))
Fq Fq
′ ′ ′
= ( x (τ ), x (τ ) , x (τ ) (p + q ))
Fp Fp
′
x (τ )
= (Fp , Fq , p Fp + q Fq ),
Fp
where F = F (x(τ ), y(τ ), z(τ ), p(λ(τ )), q(λ(τ ))) . Introducing the new parameter t by the inverse of τ = τ (t) , where
τ
Fx + Fz p + Fp px + Fq py = 0, (qx = py )
′ ′
Fx + Fz p + x (t)px + y (t)py = 0
′
Fx + Fz p + p (t) = 0
since p = p(x, y) = p(x(t), y(t)) on the curve x(t) . Thus equation (2.3.5) of the characteristic system is shown. Differentiating
the differential equation (2.3.1) with respect to y , we get finally equation (2.3.6).
Remark. In the previous quasilinear case
F (x, y, z, p, q) = a1 (x, y, z)p + a2 (x, y, z)q − a3 (x, y, z) (2.3.14)
The point is that the right hand sides are independent on p or q. It follows from Theorem 2.1 that there exists a solution of the
Cauchy initial value problem provided the initial data are non-characteristic. That is, we do not need the other remaining two
characteristic equations.
The other two equations (2.3.5) and (2.3.6) are satisfied in this quasilinear case automatically if there is a solution of the equation,
see the above derivation of these equations.
The geometric meaning of the first three characteristic differential equations (2.3.2)--(2.3.6) is the following one. Each point of the
curve
A : (x(t), y(t), z(t)) (2.3.16)
corresponds a tangential plane with the normal direction (−p, −q, 1) such that
′ ′ ′
z (t) = p(t)x (t) + q(t)y (t). (2.3.17)
This equation is called strip condition. On the other hand, let z = u(x, y) defines a surface, then z(t) := u(x(t), y(t)) satisfies the
strip condition, where p = u and q = u , that is, the ''scales'' defined by the normals fit together.
x y
Proposition 2.3. F (x, y, z, p, q) is an integral, i. e., it is constant along each characteristic curve.
Proof.
d
′ ′ ′ ′ ′
F (x(t), y(t), z(t), p(t), q(t)) = Fx x + Fy y + Fz z + Fp p + Fq q
dt
= Fx Fp + Fy Fq + p Fz Fp + q Fz Fq
−Fp fx − Fp Fz p − Fq Fy − Fq Fz q
= 0.
Corollary. Assume F (x 0, y0 , z0 , p0 , q0 ) = 0 , then F =0 along characteristic curves with the initial data (x 0, y0 , z0 , p0 , q0 ) .
Proposition 2.4.
Let z = u(x, y), u ∈ C , be a solution of the nonlinear equation (2.3.1). Set
2
where (x(t), y(t), z(t), p(t), q(t)) is the solution of the characteristic system (2.3.2)--(2.3.6) with initial data (x 0, y0 , z0 , p0 , q0 )
is a solution of the characteristic system. We recall that the solution exists and is uniquely determined.
Set z(t) = u(x(t), y(t)) , then (x(t), y(t), z(t)) ⊂ S , and
′ ′ ′
z (t) = ux x (t) + uy y (t) = ux Fp + uy Fq . (2.3.20)
Finally, from the differential equation F (x, y, u(x, y), u x (x, it follows
y), uy (x, y)) = 0
′
p (t) = −Fx − Fu p
′
q (t) = −Fy − Fu q.
is satisfied. Moreover, we assume that the initial strip satisfies the nonlinear equation, that is,
Similar to the quasilinear case we will show that the set of strips defined by the characteristic system which are fixed onto the
initial strip, see Figure 2.3.1.1, fit together and define the surface for which we are looking at.
condition (2.3.1.2) and the differential equation (2.3.1.3) there exists exactly one solution z = u(x, y) of the Cauchy initial value
problem in a neighborhood of the initial curve (x (s), y (s), z (s)) , i. e., z = u(x, y) is the solution of the differential equation
0 0 0
(2.3.1) and
u(x0 (s), y0 (s)) = z0 (s)
We will show that the surface defined by x = x(s, t), y(s, t) is the surface defined by z = u(x, y), where u is the solution of the
Cauchy initial value problem. It turns out that u(x, y) = z(s(x, y), t(x, y)), where s = s(x, y) , t = t(x, y) is the inverse of
x = x(s, t) , y = y(s, t) in a neighborhood of t = 0 . This inverse exists since the initial strip is non-characteristic by assumption:
$$\det\frac{\partial(x,y)}{\partial(s,t)}\Big|_{t=0}=x_0F_q -y_0F_q\not=0.\]
Set
$$P(x,y)=p(s(x,y),t(x,y)),\ \ Q(x,y)=q(s(x,y),t(x,y)).\]
From Proposition 2.3 and Proposition 2.4 it follows F (x, y, u, P , Q) = 0 . We will show that P (x, y) = ux (x, y) and
Q(x, y) = u (x, y). To see this, we consider the function
y
$$h(s,t)=z_s-px_s-qy_s.\]
One has
∂
= (zt − p xt − q yt ) + ps xt + qs yt − qt ys − pt xs
∂s
= (p xs + q ys )Fz + Fx xs + Fy zs + Fp ps + Fq qs .
Since F (x(s, t), y(s, t), z(s, t), p(s, t), q(s, t)) = 0 , it follows after differentiation of this equation with respect to s the
differential equation
$$h_t=-F_zh.\]
Hence h(s, t) ≡ 0 , since h(s, 0) = 0 .
Thus we have
zs = p xs + q ys
zt = p xt + q yt
zs = ux xs + uy ys
zt = ux yt + uy yt .
The first equation was shown above, the second is a characteristic equation and the last two follow from
z(s, t) = u(x(s, t), y(s, t)) . This system implies
(P − ux )xs + (Q − uy )ys = 0
(P − ux )xt + (Q − uy )yt = 0.
It follows P = ux and Q = u .
y
uy (x(s, t), y(s, t)) = q(s(x, y), t(x, y)) = q(s, t).
Example 2.3.1.1:
$$u_x^2+u_y^2+u_z^2=f(x,y,z).\]
Thus we have to extend the previous theory from R to R , n ≥ 3 .
2 n
where
n 1
x = (x1 , … , xn ), z = u(x) : Ω ⊂ R ↦ R , p = ∇u. (2.4.2)
′
p (t) = −∇x F − Fz p.
Let
x0 (s) = (x01 (s), … , x0n (s)), s = (s1 , … , sn−1 ), (2.4.3)
∂ x0 (s)
rank = n − 1. (2.4.4)
∂s
Assume
1
z0 (s) : D ↦ R , p0 (s) = (p01 (s), … , p0n (s)) (2.4.5)
l = 1, … , n − 1 , strip condition.
The initial strip manifold is said to be non-characteristic if
Fp Fp ⋯ Fp
⎛ 1 2 n ⎞
∂x ∂x02 ∂x0n
⎜ 01
⋯ ⎟
⎜ ∂s ∂s1 ∂s1
⎟
det ⎜ ⎟ ≠ 0,
1
(2.4.8)
⎜ ⎟
⎜... ... ... ... ⎟
⎜ ⎟
∂x01 ∂x02 ∂x0n
⎝ ⋯ ⎠
∂sn−1 ∂sn−1 ∂sn−1
Initial value problem of Cauchy. Seek a solution z = u(x) of the differential equation (2.4.1) such that the initial manifold is a
subset of {(x, u(x), ∇u(x)) : x ∈ Ω}.
As in the two dimensional case we have under additional regularity assumptions
Theorem 2.3. Suppose the initial strip manifold is not characteristic and satisfies differential equation (2.4.1), that is,
F (x (s), z (s), p (s)) = 0 . Then there is a neighborhood of the initial manifold (x (s), z (s)) such that there exists a unique
0 0 0 0 0
be the inverse of x = x(s, t) which exists in a neighborhood of t = 0 . Then, it turns out that
z = u(x) := z(s1 (x1 , … , xn ), … , sn−1 (x1 , … , xn ), t(x1 , … , xn )) (2.4.11)
i. e., the equation is linear in p n+1 and does not depend on z explicitly.
Remark. Formally, one can write equation (2.4.1)
$$F(x_1,\ldots,x_n,u,u_{x_1},\ldots,u_{x_n})=0\]
as an equation of type (2.5.1). Set x n+1 =u and seek u implicitly from
$$\phi(x_1,\ldots,x_n,x_{n+1})=const.,\]
where ϕ is a function which is defined by a differential equation.
Assume ϕ xn+1 ≠0 , then
0 = F (x1 , … , xn , u, ux1 , … , uxn )
ϕx1 ϕxn
= F (x1 , … , xn , xn+1 , − ,…,− )
ϕxn+1 ϕxn+1
= : G(x1 , … , xn+1 , ϕ1 , … , ϕx ).
n+1
Suppose that G ϕx
n+1
≠0 , then
$$\phi_{x_{n+1}}=H(x_1,\ldots,x_n,x_{n+1},\phi_{x_1},\ldots,\phi_{x_{n+1}}).\]
The associated characteristic equations to (2.5.1) are
′
x (τ ) = Fp =1
n+1 n+1
′
x (τ ) = Fp = Hp , k = 1, … , n
k k k
n+1 n
′
z (τ ) = ∑ pl Fp = ∑ pl Hp + pn+1
l l
l=1 l=1
= ∑ pl Hp − H
l
l=1
′
p (τ ) = −Fx − Fz pn+1
n+1 n+1
= −Fxn+1
′
p (τ ) = −Fx − Fz pk
k k
= −Fx , k = 1, … , n.
k
Here is
$$x=(x_1,\ldots,x_n),\ p=(p_1,\ldots,p_n).\]
′
z (t) = p ⋅ ∇p H − H .
Definition. The function H (x, t, p) is called Hamilton function, equation (2.5.1) Hamilton-Jacobi equation and the system (2.5.3),
(2.5.4) canonical system to H.
There is an interesting interplay between the Hamilton-Jacobi equation and the canonical system. According to the previous theory
we can construct a solution of the Hamilton-Jacobi equation by using solutions of the canonical system. On the other hand, one
obtains from solutions of the Hamilton-Jacobi equation also solutions of the canonical system of ordinary differential equations.
Definition.
A solution ϕ(a; x, t) of the Hamilton-Jacobi equation, where a = (a1 , … , an ) is an n -tuple of real parameters, is called a
complete integral of the Hamilton-Jacobi equation if
$$
\det (\phi_{x_ia_l})_{i,l=1}^n\not=0.
\]
Remark. If u is a solution of the Hamilton-Jacobi equation, then also u + const.
Theorem 2.4 (Jacobi). Assume
2
u = ϕ(a; x, t) + c, c = const. , ϕ ∈ C in its arguments, (2.5.5)
with respect to x l = xl (a, b, t) , where b i i = 1, … , n are given real constants, and then by setting
n
∂xk
ϕtai + ∑ ϕxk ai = 0
∂t
k=1
ϕtai + ∑ ϕxk ai H pk = 0.
k=1
′ ′
p (t) = ϕxi t + ∑ ϕxi xk x (t)
i k
k=1
0 = ϕx t + ∑ ϕx xk Hp + Hx
i i k i
k=1
′
0 = ϕx t + ∑ ϕx xk x (t) + Hx
i i k i
k=1
has to replace U by −U . See Landau and Lifschitz [12], Vol 1, for instance, concerning the related physics.
Then
r
dρ
t − t0 = − ∫ −−−−−−−−−−−. (2.5.14)
2
2
r0 2k β
√ 2α + −
ρ ρ
2
The inverse function r = r(t) , r(0) = r , is the r-coordinate depending on time t , and
0
r
dρ
θ − θ0 = β ∫ −−−−−−−−−−−. (2.5.15)
2
2
r0 2k β
2
ρ √ 2α + −
ρ ρ
2
Substitution τ −1
=ρ yields
1/r
dτ
θ − θ0 = −β ∫ − −−−−−−−−−−−− −
1/r0 √ 2α + 2 k2 τ − β 2 τ 2
2 2
β 1 β 1
−1 −1
k
2 r k
2 r0
= − arcsin ( −−−−−−− ) + arcsin ( −−−−−−− ).
2 2
2αβ 2αβ
√1 + √1 +
4 4
k k
Set
2
β 1
2
−1
k r0
θ1 = θ0 + arcsin ( −−−−−−−) (2.5.16)
2
2αβ
√1 +
4
k
and
−−−−−−−−
2 2
β 2
2αβ
p = , ϵ = √1 + , (2.5.17)
2 4
k k
It follows
p
r = r(θ) = , (2.5.19)
2
1 −ϵ sin(θ − θ1 )
which is the polar equation of conic sections. It defines an ellipse if 0 ≤ ϵ < 1 , a parabola if ϵ = 1 and a hyperbola if ϵ > 1 ,
see Figure 2.5.2 for the case of an ellipse, where the origin of the coordinate system is one of the focal points of the ellipse.
1
Hamilton, William Rowan, 1805--1865
2
Jacobi, Carl Gustav, 1805--1851
Q2.1
Suppose u : R 2
↦ R
1
is a solution of
Q2.2
Find a solution u ≢ const. of
ux + uy = 0 (2.E.2)
such that
3 2
graph(u) := {(x, y, z) ∈ R : z = u(x, y), (x, y) ∈ R } (2.E.3)
Q2.3
Let ϕ(x, y) be a solution of
Prove that level curves SC := {(x, y) : ϕ(x, y) = C = const. } are characteristic curves, provided that ∇ϕ ≠ 0 and
(a , a ) ≠ (0, 0) .
1 2
Q2.4
Prove Proposition 2.2.
Q2.5
Find two different solutions of the initial value problem
ux + uy = 1, (2.E.5)
where the initial data are x 0 (s) =s ,y0 (s) =s ,z0 (s) =s .
Hint: (x 0, y0 ) is a characteristic curve.
Q2.6
Solve the initial value problem
x ux + y uy = u (2.E.6)
Q2.8
Solve the initial value problem
u ux + uy = 1, (2.E.8)
Q2.9
Solve the initial value problem
u ux + u uy = 2, (2.E.9)
Q2.10
Solve the initial value problem 2 2
ux + uy = 1 + x with given initial data
x0 (s) = 0, y0 (s) = s, u0 (s) = 1, p0 (s) = 1, q0 (s) = 0 , −∞ < s < ∞ .
Q2.11
Find the solution Φ(x, y) of
(x − y)ux + 2y uy = 3x (2.E.10)
Q2.12
Solve the following initial problem of chemical kinetics.
−k1 x 2
ux + uy = (k0 e + k2 ) (1 − u ) , x > 0, y > 0 (2.E.11)
Q2.13
Solve the Riemann problem
ux1 + ux2 = 0
u(x1 , 0) = g(x1 )
in Ω = {(x
1 1, x2 ) ∈ R
2
: x1 > x2 } and in Ω2 = {(x1 , x2 ) ∈ R
2
: x1 < x2 } ,
where
with constants u l ≠ ur .
Q2.14
Determine the opening angle of the Monge cone, that is, the angle between the axis and the apothem (in German: Mantellinie) of
the cone, for equation
2 2
ux + uy = f (x, y, u), (2.E.13)
where f >0 .
Q2.15
Solve the initial value problem
2 2
ux + uy = 1, (2.E.14)
where x 0 (θ) = a cos θ, y0 (θ) = a sin θ, z0 (θ) = 1, p0 (θ) = cos θ ,q
0 (θ) = sin θ if 0 ≤ θ < 2π ,
a = const. > 0 .
Q2.16
Show that the integral ϕ(α, β; θ, r, t), see the Kepler problem, is a complete integral.
Q2.17
−−−−−−
−−−−−
a) Show that S = √−
−
α x + √1 − α y + β , α, β ∈ R
1
, 0 < α < 1 , is a complete integral of S x − √1 − Sy
2
=0 .
b) Find the envelope of this family of solutions.
Q2.18
Determine the length of the half axis of the ellipse
$$
r=\frac{p}{1-\varepsilon^2\sin(\theta-\theta_0)},\ 0\le\varepsilon<1.
\]
Q2.19
Find the Hamilton function H (x, p) of the Hamilton-Jacobi-Bellman differential equation if h = 0 and f = Ax + Bα , where
A, B are constant and real matrices, A : R , B is an orthogonal real n × n -Matrix and p ∈ R is given. The set of
m n n
↦ R
f (x, α) and h(x, α) are given. See for example, Evans [5], Chapter 10.
The classification of differential equations follows from one single question: can we calculate formally the solution if sufficiently many
initial data are given? Consider the initial problem for an ordinary differential equation y (x) = f(x, y(x)), y(x ) = y . Then one can
′
0 0
determine formally the solution, provided the function f(x, y) is sufficiently regular. The solution of the initial value problem is formally
given by a power series. This formal solution is a solution of the problem if f(x, y) is real analytic according to a theorem of Cauchy. In the
case of partial differential equations the related theorem is the Theorem of Cauchy-Kowalevskaya. Even in the case of ordinary differential
equations the situation is more complicated if y is implicitly defined, i. e., the differential equation is F (x, y(x), y (x)) = 0 for a given
′ ′
function F .
Thumbnail: Photograph of the Russian Mathematician Sofja Wassiljewna Kowalewskaja, the first major Russian female mathematician and
responsible for important original contributions to analysis, partial differential equations and mechanics.
INDEX
1 3/30/2022
CHAPTER OVERVIEW
FRONT MATTER
TITLEPAGE
INFOPAGE
1 3/30/2022
3: Classification
This text is disseminated via the Open Education Resource (OER) LibreTexts Project (https://LibreTexts.org) and like the hundreds
of other texts available within this powerful platform, it is freely available for reading, printing and "consuming." Most, but not all,
pages in the library have licenses that may allow individuals to make changes, save, and print this book. Carefully
consult the applicable license(s) before pursuing such effects.
Instructors can adopt existing LibreTexts texts or Remix them to quickly build course-specific resources to meet the needs of their
students. Unlike traditional textbooks, LibreTexts’ web based origins allow powerful integration of advanced features and new
technologies to support learning.
The LibreTexts mission is to unite students, faculty and scholars in a cooperative effort to develop an easy-to-use online platform
for the construction, customization, and dissemination of OER content to reduce the burdens of unreasonable textbook costs to our
students and society. The LibreTexts project is a multi-institutional collaborative venture to develop the next generation of open-
access texts to improve postsecondary education at all levels of higher learning by developing an Open Access Resource
environment. The project currently consists of 14 independently operating and interconnected libraries that are constantly being
optimized by students, faculty, and outside experts to supplant conventional paper-based books. These free textbook alternatives are
organized within a central environment that is both vertically (from advance to basic level) and horizontally (across different fields)
integrated.
The LibreTexts libraries are Powered by MindTouch® and are supported by the Department of Education Open Textbook Pilot
Project, the UC Davis Office of the Provost, the UC Davis Library, the California State University Affordable Learning Solutions
Program, and Merlot. This material is based upon work supported by the National Science Foundation under Grant No. 1246120,
1525057, and 1413739. Unless otherwise noted, LibreTexts content is licensed by CC BY-NC-SA 3.0.
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not
necessarily reflect the views of the National Science Foundation nor the US Department of Education.
Have questions or comments? For information about adoptions or adaptions contact info@LibreTexts.org. More information on our
activities can be found via Facebook (https://facebook.com/Libretexts), Twitter (https://twitter.com/libretexts), or our blog
(http://Blog.Libretexts.org).
where n , u : Ω ⊂ R ↦ R , Du ≡ ∇u and
1
u stands for all second derivatives. The function F is given and sufficiently regular
with respect to its 2n + 1 + n arguments.
2
ik
∑ a (x)uxi xk + f (x, u, ∇u) = 0. (3.1.2)
i,k=1
i
f = ∑ b (x)uxi + c(x)u + d(x). (3.1.3)
i=1
ik
∑ a (x)uxi xk (3.1.4)
i,k=1
plays the essential role. Suppose u ∈ C , then we can assume, without restriction of generality, that a
2 ik
=a
ki
, since
n n
ik ik ⋆
∑ a uxi xk = ∑ (a ) uxi xk , (3.1.5)
i,k=1 i,k=1
where
ik ⋆
1 ik ki
(a ) = (a +a ). (3.1.6)
2
λi = λi (x1 , … , xn ), i = 1, … , n − 1,
∂(λ1 , … , λn )
det ≠0 (3.1.7)
∂(x1 , … , xn )
in n . It is assumed that i and i are sufficiently regular. Such a mapping λ = λ(x) exists, see an exercise.
The above transform maps S onto a subset of the hyperplane defined by λ n =0 , see Figure 3.1.2.
2
∂λi ∂λl ∂ λi
uxj xk = vλi λl + vλi .
∂xj ∂xk ∂ xj ∂ xk
Since v (λ , … , λ
λk 1 n−1 , , , are known, see (3.1.9), it follows that v
0) n λk λl , l = 1, … , n − 1 , are known on S . Thus we know all
0
is the limit of
vλ (λ1 , … , λl + h, λl+1 , … , λn−1 , 0) − vλ (λ1 , … , λl , λl+1 , … , λn−1 , 0)
k k
(3.1.12)
h
as h → 0 .
Then the differential equation can be written as
jk
∂λn ∂λn
∑ a (x) vλn λn = terms known on S0 . (3.1.13)
∂xj ∂xk
j,k=1
ij
∑ a (x)χx χx ≠0 (3.1.14)
i j
i,j=1
on S . This is a condition for the given equation and for the given surface S .
Definition. The differential equation
n
ij
∑ a (x)χxi χxj = 0 (3.1.15)
i,j=1
is called it characteristic differential equation associated to the given differential equation (3.1.2).
If χ, ∇χ ≠ 0 , is a solution of the characteristic differential equation, then the surface defined by χ =0 is called characteristic
surface.
Remark. The condition (3.1.14) is satisfied for each χ with ∇χ ≠ 0 if the quadratic matrix (a (x)) is positive or negative ij
definite for each x ∈ Ω, which is equivalent to the property that all eigenvalues are different from zero and have the same sign.
This follows since there is a λ(x) > 0 such that, in the case that the matrix (a ) is poitive definite, ij
ij 2
∑ a (x)ζi ζj ≥ λ(x)|ζ | (3.1.16)
i,j=1
for all ζ ∈ R . Here and in the following we assume that the matrix (a ij
) is real and symmetric.
The characterization of differential equation (3.1.2) follows from the signs of the eigenvalues of (a ij
(x)) .
Definition. The differential equation (3.1.2) is said to be of type (α, β, γ) at x ∈ Ω if α eigenvalues of ij
(a )(x) are positive, β
Remarks:
1. According to this definition there are other types aside from elliptic, parabolic or hyperbolic equations.
2. The classification depends in general on x ∈ Ω. An example is the Tricomi equation, which appears in the theory of transsonic
flows,
y uxx + uyy = 0. (3.1.17)
Examples:
This equation is elliptic since for every manifold S given by {(x, y, z) : χ(x, y, z) = 0} , where χ is an arbitrary sufficiently
regular function such that ∇χ ≠ 0 , all derivatives of u are known on S , provided u and ∇u are known on S .
Example 3.1.2:
The wave equation u = u + u tt xx yy + uzz , where u = u(x, y, z, t) , is hyperbolic. Such a type describes oscillations of
mechanical structures, for example.
Example 3.1.3:
Example 3.1.4:
Consider the case that the (real) coefficients a in equation (3.1.2) are {\it constant}. We recall that the matrix A = (a ) is
ij ij
symmetric, that is, A = A . In this case, the transform to principle axis leads to a normal form from which the classification of
T
T ⎜0 λ2 ⋯ 0 ⎟
U AU = ⎜ ⎟. (3.1.19)
⎜... ... ... ...⎟
⎝ ⎠
0 0 ⋯ λn
Set y = U T
x and v(y) = u(U y) , then
n n
ij
∑ a uxi xj = ∑ λi vy yj . (3.1.20)
i
i,j=1 i=1
2 2
aχx + 2b χx χy + c χy = 0. (3.1.1.2)
We show that an appropriate coordinate transform will simplify equation (3.1.1.1) sometimes in such a way that we can solve the
transformed equation explicitly.
Let z = ϕ(x, y) be a solution of (3.1.1.2). Consider the level sets {(x, y) : ϕ(x, y) = const. } and assume ϕ ≠ 0 at a point y
(x0 , y0 ) of the level set. Then there is a function y(x) defined in a neighborhood of x such that ϕ(x, y(x)) = const. It follows
0
$$y'(x)=-\dfrac{\phi_x}{\phi_y},\]
which implies, see the characteristic equation (3.1.1.2),
′2 ′
ay − 2b y + c = 0. (3.1.1.3)
1 − − −−−−
2
μ1,2 = (b ± √ b − ac ) . (3.1.1.4)
a
\right),\]
see an exercise.
ψx
′
y ≡ μ2 = − ,
2
ψy
where μ and μ are given by (3.1.1.4). Thus ϕ and ψ are solutions of the linear homogeneous equations of first order
1 2
Assume ϕ(x, y), ψ(x, y) are solutions such that ∇ϕ ≠ 0 and ∇ψ ≠ 0 , see an exercise for the existence of such solutions.
Consider two families of level sets defined by ϕ(x, y) = α and ψ(x, y) = β, see Figure 3.1.1.1.
alt
(ii):
$$
\mu_2-\mu_1=\dfrac{\phi_x}{\phi_y}-\dfrac{\psi_x}{\psi_y}.
\]
□
Proposition 3.1. The mapping ξ = ϕ(x, y), η = ψ(x, y) transforms equation (3.1.1.1) into
vξη = lower order terms, (3.1.1.7)
ux = vξ ϕx + vη ψx
uy = vξ ϕy + vη ψy
2 2
uxx = vξξ ϕx + 2 vξη ϕx ψx + vηη ψx + lower order terms
2 2
uyy = vξξ ϕy + 2 vξη ϕy ψy + vηη ψy + lower order terms.
Thus
auxx + 2b uxy + c uyy = α vξξ + 2β vξη + γ vηη + l. o. t. , (3.1.1.3)
where
2 2
α : = aϕx + 2b ϕx ϕy + c ϕy
β : = aϕx ψx + b(ϕx ψy + ϕy ψx ) + c ϕy ψy
2 2
γ : = aψx + 2b ψx ψy + c ψy .
The coefficients α and γ are zero since ϕ and ψ are solutions of the characteristic equation. Since
2 2 2
αγ − β = (ac − b )(ϕx ψy − ϕy ψx ) , (3.1.1.4)
it follows from the above lemma that the coefficient β is different from zero.
□
Example 3.1.1.1:
Consider the differential equation
uxx − uyy = 0. (3.1.1.5)
ψx − ψy = 0.
ξ = x − y, η = x + y (3.1.1.8)
where g is an arbitrary C function. Thus each C -solution of the differential equation can be written as
2 2
where f , g ∈ C . 2
ij
∑ a (x, u, ∇u)uxi xj + b(x, u, ∇u) = 0 (3.2.1)
i,j=1
i,j=1
In contrast to linear equations, solutions of the characteristic equation depend on the solution considered.
i,j=1
for all ζ ∈ R .
Definition. Equation (3.2.1) is called uniformly elliptic if Λ/λ is uniformly bounded in U .
An important class of elliptic equations which are not uniformly elliptic (non-uniformly elliptic) is
n ⎛ ⎞
∂ uxi
∑ ⎜ −−−−−−−− ⎟ + lower order terms = 0. (3.2.1.1)
∂xi 2
i=1 ⎝ √ 1 + |∇u| ⎠
The main part is the minimal surface operator (left hand side of the minimal surface equation). The coefficients a are ij
−1/2 pi pj
ij 2
a (x, z, p) = (1 + |p | ) ( δij − ), (3.2.1.2)
2
1 + |p|
k
∑ A (x, u)uuk + b(x, u) = 0, (3.3.1)
k=1
where A are m × m -matrices, sufficiently regular with respect to their arguments, and
k
$$
u=\left(
u1
(3.3.2)
⋮
um
\right),\ \
u_{x_k}=\left(
u1,x
k
(3.3.3)
⋮
um,x
k
\right),\ \
b=\left(
b1
(3.3.4)
⋮
bm
\right).
\]
We ask the same question as above: can we calculate all derivatives of u in a neighborhood of a given hypersurface S in R defined
by χ(x) = 0 , ∇χ ≠ 0 , provided u(x) is given on S ?
For an answer we map S onto a flat surface S by using the mapping
0 λ = λ(x) of Section 3.1 and write equation (3.3.1) in new
coordinates. Set v(λ) = u(x(λ)) , then
$$\sum_{k=1}^nA^k(x,u)\chi_{x_k}v_{\lambda_n}=\mbox{terms known on}\ \mathcal{S}_0.\]
We can solve this system with respect to v , provided that
λn
$$\det\left(\sum_{k=1}^nA^k(x,u)\chi_{x_k}\right)\not=0\]
on S .
Definition. Equation
$$\det\left(\sum_{k=1}^nA^k(x,u)\chi_{x_k}\right)=0\]
is called characteristic equation associated to equation (3.3.1) and a surface S : χ(x) = 0 , defined by a solution χ, ∇χ ≠ 0 , of this
characteristic equation is said to be characteristic surface.
Set
$$C(x,u,\zeta)=\det\left(\sum_{k=1}^nA^k(x,u)\zeta_k\right)\]
for ζ k ∈ R .
Definition.
for all (η
1, , where
… , ηn−1 )
ii. System (3.3.1) is parabolic if there exists a regular linear mapping ζ = Qη such that D is independent of κ , that is, D depends
on less than n parameters.
iii. System (3.3.1) is elliptic if C (x, u, ζ) = 0 only if ζ = 0 .
Remark. In the elliptic case all derivatives of the solution can be calculated from the given data and the given equation.
W uy + avx + b vy = 0, (3.3.1.2)
where W , a, b, c are given functions depending of (x, y), W ≠0 and the matrix
$$
\left(
a b
(3.3.1.1)
b c
\right)
\]
is positive definite.
The Beltrami system is a generalization of Cauchy-Riemann equations. The function f (z) = u(x, y) + iv(x, y) , where
z = x + iy , is called a quasiconform mapping, see for example [9], Chapter 12, for an application to partial differential
equations.
Set
$$
A^1=\left(
W −b
(3.3.1.2)
0 a
\right),\ \
A^2=\left(
0 −c
(3.3.1.3)
W b
\right).
\]
Then the system (3.3.1.1), (3.3.1.2) can be written as
$$
A^1\left(
ux
(3.3.1.4)
vx
\right)+
A^2\left(
uy
(3.3.1.5)
vy
\right)=\left(
0
(3.3.1.6)
0
\right).
\]
Thus,
which is different from zero if ζ ≠ 0 according to the above assumptions. Thus the Beltrami system is elliptic.
where
E = (e1 , e2 , e3 )
T
electric field strength, e = e (x, t) , x = (x
i i 1, x2 , x3 ) ,
H = (h1 , h2 , h3 )
T
magnetic field strength, h = h (x, t) ,
i i
c speed of light,
λ specific conductivity,
ϵ dielectricity constant,
μ magnetic permeability.
0 ϵp0 /c 0 −p3 0 p1
0 0 ϵp0 /c p2 −p1 0
(3.3.1.7)
0 −p3 p2 μp0 /c 0 0
p3 0 −p1 0 μp0 /c 0
−p2 p1 0 0 0 μp0 /c
\right|=0.
\]
The following manipulations simplifies this equation:
i. multiply the first three columns with μp /c, 0
ii. multiply the 5th column with −p and the the 6th column with p and add the sum to the 1st column,
3 2
iii. multiply the 4th column with p and the 6th column with −p and add the sum to the 2th column,
3 1
iv. multiply the 4th column with −p and the 5th column with p and add the sum to the 3th column,
2 1
v. expand the resulting determinant with respect to the elements of the 6th, 5th and 4th row.
We obtain
$$
\left|
2
q +p p1 p2 p1 p3
1
2
p1 p2 q +p p2 p3 (3.3.1.8)
2
2
p1 p3 p2 p3 q +p
3
\right|=0,
\]
where
$$
q:=\frac{\epsilon\mu}{c^2}p_0^2-g^2
solutions of
$$
\frac{\epsilon\mu}{c^2}\chi_t^2=|\nabla_x\chi|^2.
\]
Functions defined by χ = f (n ⋅ x − V t) are solutions of this equation.
Here is f (s) an arbitrary function with f (s) ≠ 0 , n is a unit vector and V
′ −−
= c/ √ϵμ .
The associated characteristic surfaces S(t) are defined by
$$
\chi(x,t)\equiv f(n\cdot x-Vt)=0,
\]
here we assume that 0 is in he range of f : R ↦ R . Thus, S(t) is defined by n ⋅ x − V t = c , where c is a fixed constant. It
1 1
follows that the planes S(t) with normal n move with speed V in direction of n , see Figure 3.3.1.1.
Figure 3.3.1.1: d ′
(t) is the speed of plane waves
V is called speed of the plane wave S(t) .
Remark. According to the previous discussions, singularities of a solution of Maxwell equations are located at most on
characteristic surfaces.
A special case of Maxwell equations are the telegraph equations, which follow from Maxwell equations if \div E = 0 and
div H = 0$ i. e., E and H are fields free of sources. In fact, it is sufficient to assume that this assumption is satisfied at a fixed
Since
$$
\text{rot}_x\ \text{rot}_x\ A=\mbox{grad}_x\ \text{div}_x\ A-\triangle_xA
\]
for each C -vector field A , it follows from Maxwell equations the uncoupled system
2
ϵμ λμ
△x E = Ett + Et
2
c c2
ϵμ λμ
△x H = Htt + Ht .
2 2
c c
ρ density, ρ = ρ(x, t) ,
(v ⋅ ∇x )v ≡ (v ⋅ ∇x v1 , v ⋅ ∇x v2 , v ⋅ ∇x v3 ))
T
.
The second equation is
$$
\rho_t+v\cdot\nabla_x\rho+\rho\ \text{div}_x\ v=0\ \ \ \mbox{(conservation of mass)}.
\]
Assume the gas is compressible and that there is a function (state equation)
$$
p=p(\rho),
\]
where p (ρ) > 0 if ρ > 0 . Then the above system of four equations is
′
1 ′
vt + (v ⋅ ∇)v + p (ρ)∇ρ = f (3.3.1.5)
ρ
ρt + ρ div v + v ⋅ ∇ρ = 0, (3.3.1.6)
where ∇ ≡ ∇ and div ≡ div , i. e., these operators apply on the spatial variables only.
x x
dχ 1 ′
0 0 p χx2
dt ρ
(3.3.1.9)
dχ 1 ′
0 0 p χx3
dt ρ
dχ
ρχx1 ρχx2 ρχx3
dt
\right|=0,
\]
where
$$\dfrac{d\chi}{dt}:=\chi_t+(\nabla_x\chi)\cdot v. \]
Evaluating the determinant, we get the characteristic differential equation
2 2
dχ dχ 2
′
( ) (( ) − p (ρ)| ∇x χ| ) = 0. (3.3.1.7)
dt dt
This equation implies consequences for the speed of the characteristic surfaces as the following consideration shows.
Consider a family S(t) of surfaces in R defined by χ(x, t) = c , where
3
One of the two normals on S(t) at a point of the surface S(t) is given by, see an exercise,
∇x χ
n = . (3.3.1.8)
| ∇x χ|
and t0 < t1 ,t
1 −t 0 small, see Figure 3.3.1.2.
Proof. The proof follows from χ(Q 0, t0 ) = 0 and χ(Q 0 + dn, t0 + △t) = 0 , where d = |Q 1 − Q0 | and △t = t1 − t0 .
□
χt v ⋅ ∇x χ 1 dχ
V = P − vn = − − =− . (3.3.1.12)
| ∇x χ| | ∇x χ| | ∇x χ| dt
kl
∑ A (x, u, ∇u)uxk xl + lower order terms = 0, (3.4.1)
k,l=1
kl
∑ A χxk χxl vλn λn = terms known on S, (3.4.2)
k,l=1
k,l=1
If there is a solution χ with ∇χ ≠ 0 , then it is possible that second derivatives are not continuous in a neighborhood of S .
Definition. The system is called elliptic if
n
kl
det ( ∑ A ζk ζl ) ≠ 0 (3.4.4)
k,l=1
for all ζ ∈ R , ζ ≠ 0 .
divx v = 0,
where
ρ is the (constant and positive) density of liquid,
p = p(x, t) pressure.
ρ (constant) density,
The characteristic equation is det C =0 where the entries of the matrix C are given by
2 2
cij = (λ + μ)χxi χxj + δij (μ| ∇x χ| − ρχ ) . (3.4.1.2)
t
−−−−−−
−
−
λ + 2μ μ
P1 = √ , and P2 = √ . (3.4.1.4)
ρ ρ
i
ux = ∑ a (x, u)ux + b(x, u) (3.5.1)
n i
i=1
Here is u = (u 1, … , um )
T
, b = (b
1, … , bn )
T
and a are (m × m) -matrices.
i
We assume that a , b and f are in C with respect to their arguments. From (3.5.1) and (3.5.2) it follows that we can calculate
i ∞
i=0 i i
ϕ (x) does not converge, in contrast to the power series method of this section. See [15] for some asymptotic formulas in
∞
∑ i
i=0
capillarity.
Theorem 3.1. (Cauchy-Kovalevskaya). There is a neighborhood of 0 ∈ R such there is a real analytic solution of the initial value
problem (3.5.1), (3.5.2). This solution is unique in the class of real analytic functions.
Proof. The proof is taken from F. John \cite{John}. We introduce u − f as the new solution for which we are looking at and we add
a new coordinate u to the solution vector by setting u (x) = x . Then
⋆ ⋆
n
u1,xn
⋮
(3.5.3)
um,xn
⋆
ux
n
\right)=
\sum_{i=1}^{n-1}\left(
i
a 0
(3.5.4)
0 0
\right)
\left(
u1,x
i
⋮
(3.5.5)
um,xi
⋆
ux
i
⋮
(3.5.6)
bm
\right),
\]
where the associated initial condition is u(x 1, … , xn−1 , 0) = 0 .
The new u is u = (u1 , … , um )
T
, the new a
i
are i
a (x1 , … , xn−1 , u1 , … , um , u )
⋆
and the new b is
b = (x1 , … , xn−1 , u1 , … , um , u )
⋆ T
.
Thus we are led to an initial value problem of the type
n−1 N
i
uj,xn = ∑ ∑ a (z)uk,xi + bj (z), j = 1, … , N (3.5.7)
jk
i=1 k=1
functions are real analytic again, see Proposition A7 of the appendix to this section. The resulting power series on the left and on
the right have the same coefficients caused by the calculation of the derivatives D u (0) from (3.5.7). It follows that u (x), α
j j
j = 1, … , n , defined by its formal power series are solutions of the initial value problem (3.5.7), (3.5.8).
Set
$$d=\left(\frac{\partial}{\partial z_1},\ldots,\frac{\partial}{\partial z_{N+n-1}}\right)\]
Lemma A. Assume u ∈ C ∞
in a neighborhood of 0 ∈ R . Then
α β i γ
D uj (0) = Pα (d a (0), d bj (0)) , (3.5.9)
jk
where |β|, |γ| ≤ |α| and P are polynomials in the indicated arguments with non-negative integers as coefficients which are
α
independent of a and of b .
i
Moreover, it follows from (3.5.7) that the polynomials P have integers as coefficients. The initial condition (3.5.8) implies
α
where , that is, α = 0 . Then, the proof is by induction with respect to α . The induction starts with
α = (α1 , … , αn−1 , 0) n n
αn = 0 , then we replace $D^\delta u_k(0)$ in the right hand side of (3.5.10) by (3.5.11), that is by zero. Then it follows from (
3.5.10) that
i
Uj,xn = ∑ ∑ A (z)Uk,xi + Bj (z) (3.5.13)
jk
i=1 k=1
j = 1, … , N ,A
, B real analytic, is called majorizing problem to (3.5.7), (3.5.8) if
i
jk j
$$
a_{jk}^i<<A_{jk}^i\ \ \mbox{and}\ b_j<<B_j.
\]
Lemma B. The formal power series
1 α α
∑ D uj (0)x , (3.5.15)
α!
α
where D u (0) are defined in Lemma A, is convergent in a neighborhood of 0 ∈ R if there exists a majorizing problem which has
α
j
Proof. It follows from Lemma A and from the assumption of Lemma B that
α β i γ
|D uj (0)| ≤ Pα (| d a (0)|, | d bj (0)|)
jk
β i γ α
≤ Pα (| d A (0)|, | d Bj (0)|) ≡ D Uj (0).
jk
is convergent since
1 α α
1 α α
∑ |D uj (0)x | ≤∑ D Uj (0)| x |. (3.5.18)
α! α!
α α
section that there are positive constants M and r such that all these functions are majorized by
Mr
. (3.5.19)
r − z1 − … − zN +n−1
Uj (x) = 0 if xn = 0,
j = 1, … , N .
The solution of this problem is
Uj (x1 , … , xn−1 , xn ) = V (x1 + … + xn−1 , xn ), j = 1, … , N , (3.5.20)
where V (s, t), s = x 1 + … + xn−1 , t = x , is the solution of the Cauchy initial value problem
n
Mr
Vt = (1 + N (n − 1)Vs ) ,
r −s−NV
V (s, 0) = 0.
This function is real analytic in (s, t) at (0, 0). It follows that Uj (x) are also real analytic functions. Thus the Cauchy-
Kovalevskaya theorem is shown.
□
y(x0 ) = y0 ,
where x ∈ R and y ∈ R are given. Assume f (x, y) is real analytic in a neighborhood of (x , y ) ∈ R × R . Then it
0
1
0 0 0
1
follows from the above theorem that there exists an analytic solution y(x) of the initial value problem in a neighborhood of x . 0
This solution is unique in the class of analytic functions according to the theorem of Cauchy-Kovalevskaya. From the Picard-
Lindel\"of theorem it follows that this analytic solution is unique even in the class of C -functions. 1
u(x, 0) = ϕ(x)
uy (x, 0) = ψ(x).
We assume that ϕ, ψ are analytic in a neighborhood of x = 0 and that f is real analytic in a neighbourhood of
′ ′
(0, 0, ϕ(0), ϕ (0), ψ(0), ψ (0)). (3.5.22)
There exists a real analytic solution in a neighborhood of 0 ∈ R of the above initial value problem.
2
In particular, there is a real analytic solution in a neighborhood of 0 ∈ R of the initial value problem 2
△u = 1
u(x, 0) = 0
uy (x, 0) = 0.
Set U = (u, p, q, r, s, t)
T
, b = (q, 0, t, 0, 0, f
y + fu q + fq t)
T
and
0 0 0 0 0 0
⎛ ⎞
⎜0 0 1 0 0 0 ⎟
⎜ ⎟
⎜0 0 0 0 0 0 ⎟
A =⎜ ⎟. (3.5.24)
⎜ ⎟
⎜0 0 0 0 1 0 ⎟
⎜ ⎟
⎜0 0 0 0 0 1 ⎟
⎝ ⎠
0 0 fp 0 fr fs
′ ′′ ′
U (x, 0) = (ϕ(x), ϕ (x), ψ(x), ϕ (x), ψ (x), f0 (x)) , (3.5.25)
where f 0 (x)
′ ′′
= f (x, 0, ϕ(x), ϕ (x), ψ(x), ϕ (x), ψ (x))
′
.
α! = α1 ! α2 ! ⋅ … ⋅ αn !
α α1 α2 αn
x = x x ⋅ … ⋅ xn (for a monom)
1 2
∂
Dk =
∂xk
D = (D1 , … , Dn )
Du = (D1 u, … , Dn u) ≡ ∇u ≡ grad u
|α|
α α1 α2 αn
∂
D = D D ⋅ … ⋅ Dn ≡ .
1 2 α1 α2 αn
∂x ∂x … ∂ xn
1 2
where
0, 1 ∈ R.
β+γ=α
here is Dα
f (0) := (D
α
f (x)) |
x=0
.
3. Let x = (x 1, … , xn ) and m ≥ 0 an integer, then
$$
β α
D x = 0 otherwise.
7. Directional derivative:
m
d |α|!
α α
f (x + ty) = ∑ (D f (x + ty)) y , (3.5.1 Appendix.8)
m
dt α!
|α|=m
where x, y ∈ R and t ∈ R . 1
where
1 α α
Rm = ∑ (D u(y + δ(x − y))) x , 0 < δ < 1, (3.5.1 Appendix.10)
α!
|α|=m+1
δ = δ(u, m, x, y) ,
or
1
1 m (m+1)
Rm = ∫ (1 − t) Φ (t) dt, (3.5.1 Appendix.11)
m! 0
Power series
Here we collect some definitions and results for power series in R.
Definition. Let c α
1
∈ R (or ∈ R
m
) . The series
∞
⎛ ⎞
∑ cα ≡ ∑ ∑ cα (3.5.1 Appendix.12)
⎝ ⎠
α m=0 |α|=m
is said to be convergent if
∞
⎛ ⎞
∑ | cα | ≡ ∑ ∑ | cα | (3.5.1 Appendix.13)
⎝ ⎠
α m=0 |α|=m
is convergent.
Remark. According to the above definition, a convergent series is absolutely convergent. It follows that we can rearrange the order
of summation.
Using the above multi-index notation and keeping in mind that we can rearrange convergent series, we have
10. Let x ∈ R, then
n ∞
α αi
∑x = ∏ (∑ x )
i
α i=1 αi =0
1
=
(1 − x1 )(1 − x2 ) ⋅ … ⋅ (1 − xn )
1
= ,
1
(1 − x)
j
= ∑(x1 + … + xn )
j=0
1
= .
1 − (x1 + … + xn )
β!
=
1+β
(1 − x)
|β|!
= .
1+|β|
(1 − x1 − … − xn )
and the series (3.34) is uniformly convergent for all x ∈ Q(z), where
$$
Q(z):\ \ |x_i|\le|z_i|\ \ \mbox{for all}\ \ i.
\]
For given x in a fixed compact subset D of Q(z) there is a q, 0 < q < 1 , such that
| xi | ≤ q| zi | for all i. (3.5.1 Appendix.15)
Set
$$
f(x)=\sum_\alpha c_\alpha x^\alpha.
\]
Proposition A1. (i) In every compact subset D of Q(z) one has f ∈ C (D) and ∞
the formal differentiate series, that is ∑ D c x , is uniformly convergent on the closure of D and is equal to D
α
β
α
α β
f .}
(ii)
β −|β|
| D f (x)| ≤ M |β|! r in D, (3.5.1 Appendix.16)
where
μ
M = , r = (1 − q) min | zi |. (3.5.1 Appendix.17)
n
(1 − q) i
Proof. See F. John [10], p. 64. Or an exercise. Hint: Use formula 12. where x is replaced by (q, … , q).
for all x ∈ N (y), and the series converges (absolutely) for each x ∈ N (y).
A function f is called {\it real analytic in Ω} if it is real analytic for each y ∈ Ω .
We will write f ∈ C (Ω) in the case that f is real analytic in the domain Ω.
ω
A vector valued function f (x) = (f (x), … , f ) is called real analytic if each coordinate is real analytic.
1 m
for all x ∈ N (y), and the series converges (absolutely) for each x ∈ N (y), and
β −|β|
| D f (x)| ≤ M |β|! r . (3.5.1 Appendix.20)
α
Ω1 = {x ∈ Ω : D f (x) = 0 for all α},
α
Ω2 = {x ∈ Ω : D f (x) ≠ 0 for at least one α}.
Since z ∈ Ω , i. e., Ω
1 1 ≠∅ , it follows Ω 2 =∅ .
□
It was shown in Proposition A2 that derivatives of a real analytic function satisfy estimates.
On the other hand it follows, see the next proposition, that a function f ∈ C is real analytic if these estimates are satisfied.
∞
for all β.
Proposition A4. f ∈ C
ω
(Ω) if and only if f ∈ C
∞
(Ω) and for every compact subset S ⊂Ω there are positive constants M, r
such that
Proof. See F. John [10], pp. 65-66. We will prove the local version of the proposition, that is, we show it for each fixed y ∈ Ω . The
general version follows from Heine-Borel theorem. Because of Proposition A3 it remains to show that the Taylor series
1 α α
∑ D f (y)(x − y ) (3.5.1 Appendix.24)
α!
α
where d is a sufficiently small positive constant. Set Φ(t) = f (y + t(x − y)) . The one-dimensional Taylor theorem says
j−1
1 (k)
f (x) = Φ(1) = ∑ Φ (0) + rj , (3.5.1 Appendix.26)
k!
k=0
where
1
1 j−1 (j)
rj = ∫ (1 − t) Φ (t) dt. (3.5.1 Appendix.27)
(j − 1)! 0
−j j
= Mr (| x1 − y1 | + … + | xn − yn |)
j
d
≤ M( ) .
r
Choose d > 0 such that d < r , then the Taylor series converges (absolutely) in N (y) and it is equal to d f (x) since the remainder
satisfies, see the above estimate,
$$
|r_j|=\left|\frac{1}{(j-1)!}\int_0^1\ (1-t)^{j-1}\Phi^j(t)\ dt\right|\le M\left(\frac{d}{r}\right)^j.
\]
We recall that the notation f << F (f is majorized by F ) was defined in the previous section.
Proposition A5. (i) f = (f1 , … , fm ) ∈ CM,r (0) if and only if f << (Φ, … , Φ) , where
Mr
Φ(x) = . (3.5.1 Appendix.29)
r − x1 − … − xn
}
(ii) \(f\in C_{M,r}(0)\) and f (0) = 0 if and only if
f << (Φ − M , … , Φ − M ), (3.5.1 Appendix.30)
where
M (x1 + … + xn )
Φ(x) = . (3.5.1 Appendix.31)
r − x1 − … − xn
Proof.
$$
D^\alpha\Phi(0)=M|\alpha|!r^{-|\alpha|}.
\]
□
α β γ
D hk (0) = Pα (δ gl (0), D fj (0)), (3.5.1 Appendix.33)
α β γ
|D hk (0)| ≤ Pα (| δ gl (0)|, | D fj (0)|)
β γ
≤ Pα (δ Gl (0), D Fj (0))
α
= D Hk (0).
Using this result and Proposition A4, which characterizes real analytic functions, it follows that compositions of real analytic
functions are real analytic functions again.
Proposition A7. Assume f (x) and g(u) are real analytic, then g(f (x)) is real analytic if f (x) is in the domain of definition of g .
Proof. See F. John [10], p. 68. Assume that f maps a neighborhood of y ∈ R in R
m
and g maps a neighborhood of v = f (y) in
${\mathbb R}^m$. Then f ∈ C (y) and g ∈ C (v) implies
M,r μ,ρ
In the above formulas v, y are considered as fixed parameters. From Proposition~A5 it follows
∗
f (x) << (Φ − M , … , Φ − M ) =: F
∗
g (u) << (Ψ, … , Ψ) =: G,
where
Mr
Φ(x) =
r − x1 − x2 − … − xn
μρ
Ψ(u) = .
ρ − x1 − x2 − … − xn
where
μρ
χ(x) =
ρ − m(Φ(x) − M )
μρ(r − x1 − … − xn )
=
ρr − (ρ + mM )(x1 + … + xn )
μρr
<<
ρr − (ρ + mM )(x1 + … + xn )
μρr/(ρ + mM )
= .
ρr/(ρ + mM ) − (x1 + … xn )
Contributors:
Prof. Dr. Erich Miersemann (Universität Leipzig)
Integrated by Justin Marshall.
Q3.1
Let χ : n
R
1
→ R in C
1
, ∇χ ≠ 0. Show that for given x0 ∈ R
n
there is in a neighborhood of x0 a local diffeomorphism
λ = Φ(x) , Φ : (x 1, … , xn ) ↦ (λ , … , λ ) , such that λ
1 n n = χ(x) .
Q3.2
Show that the differential equation
a(x, y)uxx + 2b(x, y)uxy + c(x, y)uyy + lower order terms = 0 (3.E.1)
is elliptic if ac − b 2
>0 , parabolic if ac − b 2
=0 and hyperbolic if ac − b 2
<0 .
Q3.3
Show that in the hyperbolic case there exists a solution of ϕ x + μ1 ϕy = 0 , see equation (3.9), such that ∇ϕ ≠ 0 .
Hint: Consider an appropriate Cauchy initial value problem.
Q3.4
Show equation (3.4).
Q3.5
Find the type of
and transform this equation into an equation with vanishing mixed derivatives by using the orthogonal mapping (transform to
principal axis) x = U y, U orthogonal.
Q3.6
Determine the type of the following equation at (x, y) = (1, 1/2).
Lu := x uxx + 2y uxy + 2xy uyy = 0. (3.E.3)
Q3.7
Find all C -solutions of
2
Hint: Transform to principal axis and stretching of axis lead to the wave equation.
Q3.8
Oscillations of a beam are described by
1
wx − σt = 0
E
σx − ρwt = 0,
Q3.9
Find nontrivial solutions (∇χ ≠ 0 ) of the characteristic equation to
2
x uxx − uyy = f (x, y, u, ∇u), (3.E.5)
where f is given.
Q3.10
Determine the type of
uxx − x uyx + uyy + 3 ux = 2x, (3.E.6)
Q3.11
Transform equation
2
uxx + (1 − y )uxy = 0, (3.E.7)
Q3.12
Transform the Tricomi-equation
Q3.13
Transform equation
2 2
x uxx − y uyy = 0, (3.E.9)
Q3.14
Show that
1 1
λ = , Λ = . (3.E.10)
3/2 1/2
2 2
(1 + |p | ) (1 + |p | )
−1/2 pi pj
ij 2
a = (1 + |p | ) ( δij − ). (3.E.11)
2
1 + |p|
Q3.15
Show that Maxwell equations are a hyperbolic system.
Q3.16
Consider Maxwell equations and prove that div E = 0 and div H = 0 for all t if these equations are satisfied for a fixed time t . 0
Q3.18
Prove formula (3.22) for the normal on a surface.
Q3.19
Prove formula (3.23) for the speed of the surface S(t) .
Q3.20
Write the Navier-Stokes system as a system of type (3.4.1).
Q3.21
Show that the following system (linear elasticity, stationary case of (3.4.1.1) in the two-dimensional case) is elliptic
Q3.22
Discuss the type of the following system in stationary gas dynamics (isentrop flow) in R . 2
2
ρu ux + ρvuy + a ρx = 0
2
ρu vx + ρvvy + a ρy = 0
ρ(ux + vy ) + u ρx + vρy = 0.
Q3.23
Show formula 7. (directional derivative).
Hint: Induction with respect to m.
Q3.24
Let y = y(x) be the solution of:
′
y (x) = f (x, y(x))
y(x0 ) = y0 ,
as x → x . 0
△u = 1
u(x, 0) = uy (x, 0) = 0.
Q3.26
Solve the Cauchy initial value problem
Mr
Vt = (1 + N (n − 1)Vs )
r −s−NV
V (s, 0) = 0.
Q3.27
Write △ 2
u = −u as a system of first order.
Hint: △ 2
u ≡ △(△u) .
Q3.28
Write the minimal surface equation
⎛ ⎞ ⎛ ⎞
∂ ux ∂ uy
⎜ −−−−−−−−−⎟+ ⎜ −−−−−−−−−⎟ =0 (3.E.15)
∂x 2 2 ∂y 2 2
⎝ √ 1 + ux + uy ⎠ ⎝ √ 1 + ux + uy ⎠
Q3.29
Let f : R
1 m
×R → R
m
be real analytic in (x 0, y0 ) . Show that a real analytic solution in a neighborhood of x of the problem
0
′
y (x) = f (x, y)
y(x0 ) = y0
Index
C G P
CAPILLARY EQUATION Green's Function Parabolic Equations
1.3: Partial Differential Equations 7.4: Green's Function for \(\Delta\) 6: Parabolic Equations
Poisson's formula
D I 6.1: Poisson's Formula
Diffusion in a tube Inhomogeneous Equations
6.4.1: Fourier's Method 4.3: Inhomogeneous Equations
DIRICHLET INTEGRAL
1.3: Partial Differential Equations
CHAPTER OVERVIEW
4: HYPERBOLIC EQUATIONS
Here we consider hyperbolic equations of second order, mainly wave equations.
1 3/30/2022
CHAPTER OVERVIEW
FRONT MATTER
TITLEPAGE
INFOPAGE
1 3/30/2022
4: Hyperbolic Equations
This text is disseminated via the Open Education Resource (OER) LibreTexts Project (https://LibreTexts.org) and like the hundreds
of other texts available within this powerful platform, it is freely available for reading, printing and "consuming." Most, but not all,
pages in the library have licenses that may allow individuals to make changes, save, and print this book. Carefully
consult the applicable license(s) before pursuing such effects.
Instructors can adopt existing LibreTexts texts or Remix them to quickly build course-specific resources to meet the needs of their
students. Unlike traditional textbooks, LibreTexts’ web based origins allow powerful integration of advanced features and new
technologies to support learning.
The LibreTexts mission is to unite students, faculty and scholars in a cooperative effort to develop an easy-to-use online platform
for the construction, customization, and dissemination of OER content to reduce the burdens of unreasonable textbook costs to our
students and society. The LibreTexts project is a multi-institutional collaborative venture to develop the next generation of open-
access texts to improve postsecondary education at all levels of higher learning by developing an Open Access Resource
environment. The project currently consists of 14 independently operating and interconnected libraries that are constantly being
optimized by students, faculty, and outside experts to supplant conventional paper-based books. These free textbook alternatives are
organized within a central environment that is both vertically (from advance to basic level) and horizontally (across different fields)
integrated.
The LibreTexts libraries are Powered by MindTouch® and are supported by the Department of Education Open Textbook Pilot
Project, the UC Davis Office of the Provost, the UC Davis Library, the California State University Affordable Learning Solutions
Program, and Merlot. This material is based upon work supported by the National Science Foundation under Grant No. 1246120,
1525057, and 1413739. Unless otherwise noted, LibreTexts content is licensed by CC BY-NC-SA 3.0.
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not
necessarily reflect the views of the National Science Foundation nor the US Department of Education.
Have questions or comments? For information about adoptions or adaptions contact info@LibreTexts.org. More information on our
activities can be found via Facebook (https://facebook.com/Libretexts), Twitter (https://twitter.com/libretexts), or our blog
(http://Blog.Libretexts.org).
where u = u(x, t) is a scalar function of two variables and c is a positive constant. According to previous considerations, all C
2
-
solutions of the wave equation are
The Cauchy initial value problem for the wave equation is to find a C -solution of 2
1
utt − uxx = 0
2
c
u(x, 0) = α(x)
ut (x, 0) = β(x),
where α, β ∈ C 2
(−∞, ∞) are given.
Theorem 4.1. There exists a unique C
2 1
(R ×R )
1
-solution of the Cauchy initial value problem, and this solution is given by
d'Alembert's1 formula
x+ct
α(x + ct) + α(x − ct) 1
u(x, t) = + ∫ β(s) ds. (4.1.3)
2 2c x−ct
Proof. Assume there is a solution u(x, t) of the Cauchy initial value problem, then it follows from (4.1.2) that
u(x, 0) = f (x) + g(x) = α(x) (4.1.4)
′ ′
ut (x, 0) = c f (x) − c g (x) = β(x). (4.1.5)
Then
x
α(x) 1
f (x) = + ∫ β(s) ds + C1
2 2c 0
x
α(x) 1
g(x) = − ∫ β(s) ds + C2 .
2 2c 0
see (4.1.4). Thus each C -solution of the Cauchy initial value problem is given by d'Alembert's formula. On the other hand, the
2
function u(x, t) defined by the right hand side of (4.1.3) is a solution of the initial value problem.
□
dependence}.
1
M (r, t) = ∫ u(y, t) dSy , (4.2.4)
ωn rn−1 ∂ Br (x)
where
$$
\omega_n=(2\pi)^{n/2}/\Gamma(n/2)
\]
is the area of the n-dimensional sphere, ω nr
n−1
is the area of a sphere with radius r.
From the mean value theorem of the integral calculus we obtain the function u(x, t) for which we are looking at by
u(x, t) = lim M (r, t). (4.2.5)
r→0
1
Mt (r, 0) = ∫ g(y) dSy =: G(r), (4.2.7)
n−1
ωn r ∂ Br (x)
It follows
n
1
= ∫ ∑ uy (y, t)ξi dSy .
i
n−1
ωn r ∂ Br (x)
i=1
since $\xi\equiv (y-x)/r$ is the exterior normal at ∂ B r (x) . Assume u is a solution of the wave equation, then
n−1
1
r Mr = ∫ utt (y, t) dy
2
c ωn Br (x)
r
1
= ∫ ∫ utt (y, t) dSy dc.
2
c ωn 0 ∂ Bc (x)
n−1 2
r ∂ 1
= ( ∫ u(y, t) dSy )
2 2
c ∂t ωn rn−1 ∂ Br (x)
n−1
r
= Mtt .
2
c
The right hand side of the previous formula is well defined if the domain of dependence [x − ct, x + ct] is a subset of (0, ∞). We
can extend F and G to F and G which are defined on (−∞, ∞) such that rF and rG are C (R ) -functions as follows.
0 0 0 0
2 1
Set
$$
F_0(r)=\left\{
F (r) r >0
f (x) r =0 (4.2.1.2)
F (−r) r <0
\right.\
\]
The function G 0 (r) is given by the same definition where F and f are replaced by G and g , respectively.
Lemma. rF 0 (r), rG0 (r) ∈ C
2
(R )
2
.
Proof. From definition of F (r) and G(r) , r > 0 , it follows from the mean value theorem
$$\lim_{r\to+0} F(r)=f(x),\ \ \ \lim_{r\to+0} G(r)=g(x).\]
Thus rF (r) and rG (r) are C (R )-functions. These functions are also in C
0 0
1 1
(R )
1
. This follows since F and G are in
0 0 C
1 1
(R ) .
We have, for example,
n
1
′
F (r) = ∫ ∑ fy (x + rξ)ξj dSξ
j
ωn ∂ B1 (0)
j=1
′
1
F (+0) = ∫ ∑ fy (x)ξj dSξ
j
ωn ∂ B1 (0) j=1
n
1
= ∑ fy (x) ∫ nj dSξ
j
ωn ∂ B1 (0)
j=1
= 0.
d
= (tF (ct)) + tG(ct).
dt
Theorem 4.2. Assume f ∈ C (R ) and g ∈ C (R ) are given. Then there exists a unique solution
3 3 2 3
u ∈ C
2 3
(R × [0, ∞)) of the
initial value problem (4.2.2)-(4.2.3), where n = 3 , and the solution is given by the Poisson's formula
1 ∂ 1
u(x, t) = ( ∫ f (y) dSy ) (4.2.1.3)
2
4πc ∂t t ∂ Bct (x)
1
+ ∫ g(y) dSy . (4.2.1.4)
2
4π c t ∂ Bct (x)
Proof. Above we have shown that a C -solution is given by Poisson's formula. Under the additional assumption f
2
∈ C
3
it follows
from Poisson's formula that this formula defines a solution which is in C , see F. John [10], p. 129.
2
Corollary. From Poisson's formula we see that the domain of dependence for u(x, t0 ) is the intersection of the cone defined by
|y − x| = c|t − t | with the hyperplane defined by t = 0 , see Figure 4.2.1.2.
0
where f ∈ C
3
, g ∈ C
2
.
Using the formula for the solution of the three-dimensional initial value problem we will derive a formula for the two-dimensional
case. The following consideration is called Hadamard's method of decent.
Let v(x, y, t) be a solution of (4.2.2.1)-(4.2.2.3), then
$$u(x,y,z,t):=v(x,y,t)\]
is a solution of the three-dimensional initial value problem with initial data f (x, y), g(x, y), independent of z , since u satisfies (
4.2.2.1)-(4.2.2.3). Hence, since u(x, y, z, t) = u(x, y, 0, t) + u (x, y, δz, t)z, 0 < δ < 1 , and u = 0 , we have
z z
$$v(x,y,t)=u(x,y,0,t).\]
Poisson's formula in the three-dimensional case implies
1 ∂ 1
v(x, y, t) = ( ∫ f (ξ, η) dS)
2
4πc ∂t t ∂ Bct (x,y,0)
1
+ ∫ g(ξ, η) dS. (4.2.2.4)
2
4π c t ∂ Bct (x,y,0)
exterior normal n at S is n = ∇χ/|∇χ| and the surface element is given by dS = (1/|n 3 |)dξdη , where the third coordinate of n
is
$$n_3=\pm\frac{\sqrt{c^2 t^2-(\xi-x)^2-(\eta-y)^2}}{ct}.\]
The positive sign applies on S
+
, where ζ >0 and the sign is negative on S
−
where ζ <0 , see Figure 4.2.2.1. We have
¯
¯¯¯¯¯
¯
S =S
+
∪S .−
−−−−−−−−−−−−−− −
Set ρ = √(ξ − x ) 2
+ (η − y )
2
. Then it follows from (4.2.2.4)
Theorem 4.3. The solution of the Cauchy initial value problem (4.2.2.1)-(4.2.2.3) is given by
1 ∂ f (ξ, η)
v(x, y, t) = ∫ − −−−−− − dξdη
2πc ∂t Bct (x,y) √ c2 t2 − ρ2
1 g(ξ, η)
+ ∫ dξdη.
−−−−−− −
2πc 2 2 2
Bct (x,y) √c t − ρ
only. Therefore, see formula of Theorem 4.3, if f , g have supports in a compact domain D ⊂ R , then these functions have
2
influence on the value v(x, y, t) for all time t > T , T sufficiently large.
where □u := u tt
2
− c △u . We assume f ∈ C
3
, g ∈ C and w ∈ C , which are given.
2 1
Set u = u + u , where u is a solution of problem (4.3.1)-(4.3.3) with w := 0 and u is the solution where f
1 2 1 2 =0 and g = 0 in (
4.3.1)-(4.3.3). Since we have explicit solutions u in the cases n = 1 , n = 2 and n = 3 , it remains to solve
1
n 1
□u = w(x, t) on x ∈ R , t ∈ R (4.3.4)
u(x, 0) = 0 (4.3.5)
ut (x, 0) = 0. (4.3.6)
The following method is called Duhamel's principle which can be considered as a generalization of the method of variations of
constants in the theory of ordinary differential equations.
To solve this problem, we make the ansatz
t
and
v(x, s, s) = 0. (4.3.9)
= w(x, t).
Thus necessarily v t (x, , see (4.3.8). We have seen that the ansatz provides a solution of (4.3.4)-(4.3.6) if for all s
t, t) = w(x, t)
Let v ∗
(x, t, s) be a solution of
then
$$v(x,t,s):=v^*(x,t-s,s)\]
is a solution of (4.3.11).
In the case n = 3 , where v is given by, see Theorem 4.2,
∗
1
= ∫ w(ξ, s) dSξ .
2
4π c (t − s) ∂ Bc( t−s) (x)
t
1 w(ξ, s)
= ∫ ∫ dSξ ds.
2
4πc 0 ∂ Bc( t−s) (x)
t −s
1 w(ξ, t − r/c)
= ∫ dξ,
2
4πc Bct (x)
r
where r = |x − ξ| .
Formulas for the cases n = 1 and n =2 follow from formulas for the associated homogeneous equation with inhomogeneous
initial values for these cases.
Theorem 4.4. The solution of
□u = w(x, t), u(x, 0) = 0, ut (x, 0) = 0, (4.3.13)
Case n = 3 :
1 w(ξ, t − r/c)
u(x, t) = ∫ dξ, (4.3.14)
2
4πc Bct (x)
r
where r = |x − ξ| , x = (x 1, x2 , x3 ) , ξ = (ξ 1, ξ2 , ξ3 ) .
Case n = 2 :
t
1 w(ξ, τ )
u(x, t) = ∫ (∫ − −− − −−− − −−− − dξ) dτ , (4.3.15)
4πc 0 Bc( t−τ ) (x) √ c2 (t − τ )2 − r2
x = (x1 , x2 ) , ξ = (ξ
1, ξ2 ) .
Case n = 1 :
t x+c(t−τ)
1
u(x, t) = ∫ (∫ w(ξ, τ ) dξ) dτ . (4.3.16)
2c 0 x−c(t−τ)
Remark. The integrand on the right in formula for n = 3 is called retarded potential. The integrand is taken not at t , it is taken at
an earlier time t − r/c .
ii. S is not a characteristic curve. Moreover assume that the characteristic curves, which are lines here and are defined by
x = const. and y = const. , have at most one point of intersection with S , and such a point is not a touching point, i. e.,
characteristics associated to these first partial differential of first order is defined by x (t) = 1, y (t) = 0 , see Chapter 2.
′ ′
Set
P = −(ux v − xx u + 2buv)
Q = uy v − vy u + 2auv.
= ∮ P dx + Qdy, (4.4.6)
where integration in the line integral is anticlockwise. The previous equation follows from Gauss theorem or after integration by
parts:
$$\int_\Omega\ (-P_y+Q_x)\ dxdy=\int_{\partial\Omega}\ (-Pn_2+Qn_1)\ ds,\]
where n = (dy/ds, −dx/ds) , s arc length, (x(s), y(s)) represents ∂Ω .
Assume u is a solution of the initial value problem (4.4.1)-(4.4.4) and suppose that v satisfies
$$Mv=0\ \ \mbox{in}\ \Omega.\]
The line integral from B to A is known from initial data, see the definition of P and Q.
Since
$$u_xv-v_xu+2buv=(uv)_x+2u(bv-v_x),\]
it follows
Let v be a solution of the initial value problem, see Figure 4.2.2 for the definition of domain D(P ),
M v = 0 in D(P ) (4.4.9)
bv − vx = 0 on C1 (4.4.10)
av − vy = 0 on C2 (4.4.11)
v(P ) = 1. (4.4.12)
Example 4.4.1:
Example 4.4.2:
bounded at 0. This bounded solution is the Bessel function J (x) of first kind and of order n , see [1], for example.
n
solutions u(x, t) defined in a bounded domain Ω ⊂ R and for all t ∈ R and which satisfy additional boundary conditions on
n 1
∂Ω.
Assume the initial data f , g are sufficiently regular. This implies compatibility conditions f (0) = f (l) = 0 and g(0) = g(l) .
Fourier's method
To find solutions of differential equation (4.5.1.1) we make the separation of variables ansatz
u(x, t) = v(x)w(t). (4.5.1.1)
or, if v(x)w(t) ≠ 0 ,
′′ ′′
w (t) v (x)
= . (4.5.1.3)
w(t) v(x)
and
′′
v (x)
= −λ (4.5.1.5)
v(x)
are
−− −−
sin(√λn t), cos(√λn t). (4.5.1.9)
Set
−− −−
wn (t) = αn cos(√λn t) + βn sin(√λn t), (4.5.1.10)
where α , β ∈ R .
n n
1
n=1
which satisfies the differential equation (4.5.1.1) and the boundary conditions (4.5.1.4). Consider the formal solution of (4.5.1.1), (
4.5.1.4)
∞
−− −− −−
u(x, t) = ∑ (αn cos(√λn t) + βn sin(√λn t)) sin(√λn x). (4.5.1.7)
n=1
''Formal'' means that we know here neither that the right hand side converges nor that it is a solution of the initial-boundary value
problem. Formally, the unknown coefficients can be calculated from initial conditions (4.5.1.2), (4.5.1.3) as follows. We have
∞
−−
u(x, 0) = ∑ αn sin(√λn x) = f (x). (4.5.1.12)
n=1
−−
Multiplying this equation by sin(√λ k x) and integrate over (0, l), we get
l l
2
−− −−
αn ∫ sin (√λk x) dx = ∫ f (x) sin(√λk x) dx. (4.5.1.13)
0 0
We recall that
l
−− −− l
∫ sin(√λn x) sin(√λk x) dx = δnk . (4.5.1.14)
0
2
Then
l
2 πk
αk = ∫ f (x) sin( x) dx. (4.5.1.8)
l 0
l
n=1
Under additional assumptions f ∈ C (0, l), g ∈ C (0, l) it follows that the right hand side of (4.5.1.7), where α , β are given by
4
0 0
3
n n
(4.5.1.8) and (4.5.1.9), respectively, defines a classical solution of (4.5.1.1)-(4.5.1.4) since under these assumptions the series for
u and the formal differentiate series for u , u , u , u
t tt x converges uniformly on 0 ≤ x ≤ l , 0 ≤ t ≤ T , 0 < T < ∞ fixed, see an
xx
exercise.
1
utt (x, t) = △x u in Ω × R , (4.5.2.1)
¯
¯¯¯
u(x, 0) = f (x), x ∈ Ω, (4.5.2.2)
¯
¯¯¯
ut (x, 0) = g(x), x ∈ Ω, (4.5.2.3)
1
u(x, t) = 0 on ∂Ω × R . (4.5.2.4)
As in the previous subsection for the string, we make the ansatz (separation of variables)
$$
u(x,t)=w(t)v(x)
\]
which leads to the eigenvalue problem
−△v = λv in Ω, (4.5.2.5)
v = 0 on ∂Ω. (4.5.2.6)
Let λ are the eigenvalues of (4.5.2.5), (4.5.2.6) and v a complete associated orthonormal system of eigenfunctions. We assume
n n
Ω is sufficiently regular such that the eigenvalues are countable, which is satisfied in the following examples. Then the formal
αn = ∫ f (x)vn (x) dx
Ω
1
βn = ∫ g(x)vn (x) dx.
−−
√λn Ω
Note
In general, eigenvalues of (4.5.2.5), (4.5.2.5) are not known explicitly. There are numerical methods to calculate these values.
In some special cases, see next examples, these values are known.
Examples
Example 4.5.2.1: Rectangle membrane
Let
Ω = (0, a) × (0, b). (4.5.2.1)
Using the method of separation of variables, we find all eigenvalues of (4.5.2.5), (4.5.2.6) which are given by
−−−−−−−
2 2
k l
λkl =√ + , k, l = 1, 2, … (4.5.2.2)
2 2
a b
u(R, θ) = 0, (4.5.2.7)
here is u = u(r, θ) := v(r cos θ, r sin θ) . We will find eigenvalues and eigenfunctions by separation of variables
where v(R) = 0 and q(θ) is periodic with period 2π since u(r, θ) is single valued.
This leads to
1 1
′ ′ ′′
− ((rv ) q + vq ) = λvq. (4.5.2.5)
r r
which implies
′′
q (θ)
= const. =: −μ. (4.5.2.6)
q(θ)
It follows that eigenvalues μ are real and nonnegative. All solutions of the differential equation are given by
− −
q(θ) = A sin(√μ θ) + B cos(√μ θ), (4.5.2.7)
it follows
x +y x −y
sin x − sin y = 2 cos sin
2 2
x +y x −y
cos x − cos y = −2 sin sin
2 2
which implies, since A , B are not zero simultaneously, because we are looking for q not identically zero,
− −
sin(√μ π) sin(√μ θ + δ) = 0 (4.5.2.10)
Inserting q ′′
(θ)/q(θ) = −n
2
into (4.5.2.8), we obtain the boundary value problem
2 ′′ ′ 2 2
r v (r) + rv (r) + (λ r − n )v = 0 on (0, R) (4.5.2.9)
v(R) = 0 (4.5.2.10)
− −
Set z = √λr and v(r) = v(z/√λ) =: y(z) , then, see (4.5.2.9),
2 ′′ ′ 2 2
z y (z) + zy (z) + (z − n )y(z) = 0, (4.5.2.12)
where z > 0 . Solutions of this differential equations which are bounded at zero are Bessel functions of first kind and n -th order
−
J (z) . The eigenvalues follows from boundary condition (4.5.2.10), i. e., from J (√λ R) = 0 . Denote by τ
n the zeros of
n nk
2
τnk
λnk = ( ) (4.5.2.13)
R
as x → ∞ . It follows from this formula that there are infinitely many zeros of J n (x) .
1
utt = Lu + f (x, t) in Ω × R (4.5.3.1)
¯
¯¯¯
u(x, 0) = ϕ(x) x ∈ Ω (4.5.3.2)
¯
¯¯¯
ut (x, 0) = ψ(x) x ∈ Ω (4.5.3.3)
1
u(x, t) = 0 for x ∈ ∂Ω and t ∈ R , (4.5.3.4)
2. L = △ , oscillating membrane.
x
3. n
∂
ij
Lu = ∑ (a (x)ux ) , (4.5.3.1)
i
∂xj
i,j=1
¯
¯¯¯
where a = a are given sufficiently regular functions defined on Ω. We assume L is uniformly elliptic, that is, there is a constant
ij ji
$$\sum_{i,j=1}^na^{ij}\zeta_i\zeta_j\ge\nu|\zeta|^2\]
for all x ∈ Ω and ζ ∈ R . n
4. Let u = (u 1, … , um ) and
$$Lu=\sum_{i,j=1}^n\frac{\partial}{\partial x_j}\left(A^{ij}(x)u_{x_i}\right),\]
¯
¯¯¯
where A = A are given sufficiently regular (m × m) -matrices on Ω. We assume that L defines an elliptic system. An example
ij ji
v = 0 on ∂Ω. (4.5.3.6)
k=1
with functions wk (t) which will be determined later. It is assumed that all series are convergent and that following calculations
make sense.
Let
∞
k=1
which follows from (4.5.3.8) after multiplying with v (x) and integrating over Ω.
l
Set
k=1
k=1
t
1 −−
+ − − ∫ ck (τ ) sin(√λk (t − τ )) dτ ,
√λk 0
where
$$ a_k=\langle\phi,v_k\rangle,\ \ \ b_k=\frac{1}{\sqrt{\lambda_k}}\langle\psi,v_k\rangle.\]
Summarizing, we have
Proposition 4.2. The (formal) solution of the initial-boundary value problem (4.5.3.1)-(4.5.3.4) is given by
∞
k=1
where v is a complete orthonormal system of eigenfunctions of (4.5.3.5), (4.5.3.6) and the functions w are defined by (4.5.3.10).
k k
1 ω −−
= Avn (x) ( −− sin(√λk t) − sin(ωt)) ,
2
ω − λn √λn
−−
provided ω ≠ √λ . It follows
n
$$u(x,t)\to\frac{A}{2\sqrt{\lambda_n}}v_n(x)\left(\frac{\sin(\sqrt{\lambda_n}t)}{\sqrt{\lambda_n}}-t\cos(\sqrt{\lambda_n}
t)\right)\]
−− −−
Consequently |u| can be arbitrarily large at some points x and at some times t if ω = √λn . The frequencies √λn are called
critical frequencies at which resonance occurs.
A Uniqueness Result
¯
¯¯¯
The solution of the initial-boundary value problem (4.5.3.1)-(4.5.3.4) is unique in the class C 2
(Ω × R )
1
.
Proof. Let u , u are two solutions, then u = u
1 2 2 − u1 satisfies
1
utt = Lu in Ω × R
¯
¯¯¯
u(x, 0) = 0 x ∈ Ω
¯
¯¯¯
ut (x, 0) = 0 x ∈ Ω
n
u(x, t) = 0 for x ∈ ∂Ω and t ∈ R .
′ ij
E (t) = 2 ∫ ( ∑ a (x)ux ux t + ut utt ) dx
i j
Ω
i,j=1
n
ij
= 2∫ ( ∑ a (x)uxi ut nj ) dS
∂Ω i,j=1
= 0.
It follows E(t) = const. From u (x, 0) = 0 and u(x, 0) = 0 we get E(0) = 0 . Consequently E(t) = 0 for all t , which implies,
t
¯
¯¯¯
since L is elliptic, that u(x, t) = const. on Ω × R . Finally, the homogeneous initial and boundary value conditions lead to
1
¯
¯¯¯
u(x, t) = 0 on Ω × R .1
2
utt = c uxx (4.E.1)
holds for all parallelograms ABC D in the (x, t)-plane, which are bounded by characteristic lines, see Figure 4.E.1.
Prove that v k (x)wk (t) is a solution of the partial differential equation (wave equation) u tt = uxx .
Q4.3
Solve for given f (x) and μ ∈ R the initial value problem
1
1 1
ut + ux + μuxxx = 0 in R × R+
u(x, 0) = f (x) .
Q4.4
Let S := {(x, t); t = γx} be space-like, i.e., |γ| < 1/c ) in (x, t)-space, x = (x , x
2
1 2, x3 ) . Show that the Cauchy initial value
problem □u = 0 with data for u on S can be transformed using the Lorentz-transform
2
x1 − γ c t
x1 = (4.E.4)
− −−− − −−
√ 1 − γ 2 c2
′ ′
x = x2 , x = x3 (4.E.5)
2 3
t − γx1
′
t = (4.E.6)
−−−− −−−
2 2
√1 − γ c
Prove |α n| ≤ c/ n
4
, provided f ∈ C
0
4
(0, l) .
Q4.6
Let Ω be the rectangle (0, a) × (0, b) . Find all eigenvalues and associated eigenfunctions of −△u = λu in Ω, u = 0 on ∂Ω . Hint:
Separation of variables.
Q4.7
2
ℏ
n 1
iℏψt = − △x ψ + V (x)ψ in R ×R , (4.E.9)
2m
n
2
∫ |ψ(x, t)| dx = 1 , (4.E.10)
R
2m
n
△u + (E − V (x))u = 0 in R (4.E.11)
2
ℏ
n 2
under the side condition ∫ R
|u | dx = 1 , u : R n
↦ C .
Here is
n 1
ψ : R ×R ↦ C (4.E.12)
Q4.8
Find nonzero solutions by using separation of variables of u = △ tt xu in Ω × (0, ∞) , u(x, t) = 0 on ∂Ω , where Ω is the circular
cylinder Ω = {(x , x , x ) ∈ R : x + x < R , 0 < x < h} .
1 2 3
n 2
1
2
2
2
3
Q4.9
Solve the initial value problem
3 utt − 4 uxx = 0
u(x, 0) = sin x
ut (x, 0) = 1 .
Q4.10
Solve the initial value problem
2 2 1
utt − c uxx = x , t > 0, x ∈ R
u(x, 0) = x
ut (x, 0) = 0 .
Hint: Find a solution of the differential equation independent on t , and transform the above problem into an initial value problem
with homogeneous differential equation by using this solution.
Q4.12
2 2
utt − c uxx = λ u, λ = const. (4.E.14)
2 2 2 2 2 2
u(x, t) = f (x − c t ) = f (s), s := x −c t (4.E.15)
2 ′′ ′ 2 2
z f (z) + zf (z) + (z − n )f = 0, z > 0 (4.E.16)
with n = 0 .
Remark. The above differential equation for u is the transformed telegraph equation (see Section 4.4).
Q4.13
Find the formula for the solution of the following Cauchy initial value problem uxy = f (x, y), where S : y = ax + b , a >0 , and
the initial conditions on S are given by
u = αx + βy + γ,
ux = α,
uy = β,
Q4.14
Find all eigenvalues μ of
′′
−q (θ) = μq(θ)
Index
C G P
CAPILLARY EQUATION Green's Function Parabolic Equations
1.3: Partial Differential Equations 7.4: Green's Function for \(\Delta\) 6: Parabolic Equations
Poisson's formula
D I 6.1: Poisson's Formula
Diffusion in a tube Inhomogeneous Equations
6.4.1: Fourier's Method 4.3: Inhomogeneous Equations
DIRICHLET INTEGRAL
1.3: Partial Differential Equations
CHAPTER OVERVIEW
5: FOURIER TRANSFORM
Fourier's transform is an integral transform which can simplify investigations for linear differential or integral equations since it transforms a
differential operator into an algebraic equation.
Thumbnail: The real and imaginary parts of the Fourier transform of a delayed pulse. The Fourier transform decomposes a function into
eigenfunctions for the group of translations. (CC-BY-SA-4.0; Sławomir_Biały).
1 3/30/2022
5.1: Definition and Properties
Definition. Let f ∈ C
0
s n
(R ) , s = 0, 1, … . The function f^ defined by
given by
−n/2 iξ⋅x
g̃ (x) = (2π ) ∫ e g(ξ) dξ (5.1.2)
n
R
is called inverse Fourier transform, provided the integrals on the right hand side
exist.
From (5.1.1) it follows by integration by parts that differentiation of a function is changed to multiplication of its Fourier
transforms, or an analytical operation is converted into an algebraic operation. More precisely, we have
Proposition 5.1.
α̂ |α| α ˆ
D f (ξ) = i ξ f (ξ), (5.1.3)
where |α| ≤ s .
The following proposition shows that the Fourier transform of f decreases rapidly for |ξ| → ∞ , provided f ∈ C
0
s n
(R ) . In
particular, the right hand side of (5.1.2) exists for g := f^ if f ∈ C n+1
0
(R ) .
n
1/2
n
2 −
|ξ| = ( ∑ ξ ) ≤ √n | ξj | (5.1.4)
k
k=1
which implies
s
s
s k
(1 + |ξ|) = ∑( )|ξ |
k
k=0
s
s k/2 k
≤ 2 ∑n | ξj |
k=0
s s/2 α
≤ 2 n ∑ |ξ |.
|α|≤s
|α|≤s
s s/2 α
≤ 2 n ∑ ∫ | D g(x)| dx =: M .
n
R
|α|≤s
−n/2 iξ⋅x ˆ
(2π ) ∫ e f (ξ) dξ = f (x) (5.1.5)
n
R
here. The proof of the other relation is left as an exercise. All integrals appearing in the following exist, see Proposition 5.2 and the
special choice of g .
(i) Formula
ˆ ix⋅ξ
∫ g(ξ)f (ξ)e dξ = ∫ ĝ (y)f (x + y) dy (5.1.6)
n n
R R
−n/2 −iξ⋅(y−x)
= (2π ) ∫ (∫ g(ξ)e dξ) f (y) dy
n n
R R
(ii) Formula
−n/2 −iy⋅ξ −n
(2π ) ∫ e g(εξ) dξ = ε ĝ (y/ε) (5.1.7)
n
R
for each ε > 0 follows after substitution z = εξ in the left hand side of (5.1.1).
(iii) Equation
ˆ ix⋅ξ
∫ g(εξ)f (ξ)e dξ = ∫ ĝ (y)f (x + εy) dy (5.1.8)
n n
R R
follows from (5.1.6) and (5.1.7). Set G(ξ) := g(εξ) , then (5.1.6) implies
ˆ ix⋅ξ ˆ
∫ G(ξ)f (ξ)e dξ = ∫ G(y)f (x + y) dy. (5.1.9)
n n
R R
ˆ −n/2 −iy⋅ξ
G(y) = (2π ) ∫ e g(εξ) dξ
n
R
−n
= ε ĝ (y/ε),
ˆ −n
∫ g(εξ)f (ξ) = ∫ ε ĝ (y/ε)f (x + y) dy
n n
R R
= ∫ ĝ (z)f (x + εz) dz.
n
R
Letting ε → 0 , we get
ˆ ix⋅ξ
g(0) ∫ f (ξ)e dξ = f (x) ∫ ĝ (y) dy. (5.1.10)
n n
R R
Set
2
−|x| /2
g(x) := e , (5.1.11)
then
n/2
∫ ĝ (y) dy = (2π ) . (5.1.12)
n
R
Since g(0) = 1 , the first assertion of Theorem 5.1 follows from (5.1.10) and (5.1.12). It remains to show (5.1.12).
(iv) Proof of (5.1.12). We will show
2
−n/2 −|x| /2 −ix⋅x
ĝ (y) : = (2π ) ∫ e e dx
n
R
2
−|y | /2
= e .
The proof of
2
−|y | /2 n/2
∫ e dy = (2π ) (5.1.13)
n
R
it follows
2 2 2
−|x| /2 −ix⋅y −η −|y | /2
∫ e e dx = ∫ e e dx
n n
R R
2 2
−|y | /2 −η
= e ∫ e dx
n
R
2 2
n/2 −|y | /2 −η
= 2 e ∫ e dη
n
R
where
x y
η := – +i –. (5.1.15)
√2 √2
where the integration is along the curve C which is the union of four curves as indicated in Figure ??? .
It follows
−η
2
−
lim ∫ e dη = √π (5.1.18)
R→∞
C3
since
2
−η
lim ∫ e dη = 0, k = 2, 4. (5.1.19)
R→∞
Ck
The case n > 1 can be reduced to the one-dimensional case as follows. Set
x y
η = – +i – = (η1 , … , ηn ), (5.1.20)
√2 √2
where
xl yl
ηl = – +i –. (5.1.21)
√2 √2
l=1
it follows
n
2 2
−eta −η
∫ e dη = ∏ ∫ e l
dηl , (5.1.23)
n
R Γl
l=1
There is a useful class of functions for which the integrals in the definition of fˆ and f˜ exist.
For u ∈ C (R ) we set
∞ n
$$
q_{j,k}(u):=\max_{\alpha:\ |\alpha|\le k}\left(\sup_{\mathbb{R}^n}\left((1+|x|^2)^{j/2}|D^\alpha u(x)|\right)\right).
\]
Definition. The Schwartz class of rapidly decreasing functions is
$$
{\mathcal{S}}(\mathbb{R}^n)=\left\{u\in C^\infty(\mathbb{R}^n): \ q_{j,k}(u)<\infty\ \mbox{for any}\ j,k\in{\mathbb N}\cup\
{0\}\right\}.
\]
This space is a Frechét space.
Proposition 5.3. Assume u ∈ S(R ), then û and ũ ∈ S(R
n n
) .
Proof. See [24], Chapter 1.2, for example, or an exercise.
According to Theorem 5.1 and Proposition 5.3, we have, at least for u ∈ S(R ), n
−n/2 ix⋅ξ
p(x, D)u(x) = (2π ) ∫ e p(x, ξ)û (ξ) dξ, (5.1.1.1)
n
R
where
$$p(x,\xi)=\sum_{|\alpha|\le m}a_\alpha(x)\xi^\alpha.\]
The right hand side of (5.1.1.1) makes sense also for more general functions p(x, ξ), not only for polynomials.
Definition. The function p(x, ξ) is called symbol and
$$(Pu)(x):=(2\pi)^{-n/2}\int_{\mathbb{R}^n}\ e^{ix\cdot\xi}p(x,\xi)\widehat{u}(\xi)\ d\xi\]
is said to be pseudodifferential operator.
An important class of symbols for which the right hand side in this definition of a pseudodifferential operator is defined is S
m
−n/2 ix⋅ξ α
(P u)(x) = (2π ) ∫ K(x, y) ∫ e ξ û (ξ) dξ
n n
R R
Q5.1
Show
2
−|y | /2 n/2
∫ e dy = (2π ) . (5.E.1)
n
R
Q5.2
Show that u ∈ S(Rn implies u
)
^, ũ ∈ S(R
n
) .
Q5.3
Give examples for functions p(x, ξ) which satisfy p(x, ξ) ∈ S . m
Q5.4
Find a formal solution of Cauchy's initial value problem for the wave equation by using Fourier's transform.
Contributors:
Prof. Dr. Erich Miersemann (Universität Leipzig)
1 3/30/2022
6.1: Poisson's Formula
Assume u is a solution of (6.2), then, since Fourier transform is a linear mapping,
ˆ
ut − △u = ^
0. (6.1.1)
provided the transforms exist. Thus we arrive at the ordinary differential equation for the Fourier transform of u
dû 2
+ |ξ | û = 0, (6.1.3)
dt
since û(ξ, 0) = ϕ
ˆ
(ξ) . From Theorem 5.1 it follows
2
−n/2 −|ξ| t iξ⋅x
u(x, t) = (2π ) ∫ ϕ̂ (ξ)e e dξ
n
R
2
−n iξ⋅(x−y)−|ξ| t
= (2π ) ∫ ϕ(y) ( ∫ e dξ) dy.
n n
R R
Set
$$K(x,y,t)=(2\pi)^{-n}\int_{\mathbb{R}^n}e^{i\xi\cdot (x-y)-|\xi|^2t}\ d\xi.\]
By the same calculations as in the proof of Theorem 5.1, step (vi), we find
2
−n/2 −|x−y | /4t
K(x, y, t) = (4πt) e . (6.1.5)
Thus we have
1 2
−|x−z| /4t
u(x, t) = ∫ ϕ(z)e dz. (6.1.6)
−
− n
(2 √πt)
n
R
Definition. Formula (6.1.6) is called Poisson's formula} and the function K defined by (6.1.5) is called heat kernel or
fundamental solution of the heat equation.
Proposition 6.1 The kernel K has following properties:
1. (i) K(x, y, t) ∈ C (R × R × R ) ,
∞ n n 1
+
t>0
uniformly for x ∈ R.
Proof. (i) and (iii) are obviously, and (ii) follows from the definition of K . Equations (iv) and (v) hold since
2
−n/2 −|x−y | /4t
∫ K(x, y, t) dy = ∫ (4πt) e dy
n n
R ∖ Bδ (x) R ∖ Bδ (x)
2
−n/2 −|η|
= π ∫ e dη
n
R ∖ Bδ/√4t (0)
Theorem 6.1. Assume ϕ ∈ C (R ) and sup |ϕ(x)| < ∞ . Then u(x, t) given by Poisson's formula (6.1.6) is in
n
R
n C
∞ n
(R ×R
1
+
) ,
continuous on R × [0, ∞) and a solution of the initial value problem (6.2), (6.3).
n
}u(x,t)=\phi(\xi).
\]
< 2ε
2. u(x, t) defined by Poisson's formula depends on all values ϕ(y), y ∈ R . That means, a perturbation of ϕ , even far from a fixed
n
x, has influence to the value u(x, t). This means that heat travels with infinite speed, in contrast to the experience.
u(x, 0) = ϕ(x),
ˆ
û (ξ, 0) = ϕ (ξ).
t
2
−|ξ| (t−τ) ˆ(ξ, τ ) dτ ) dξ.
+ ∫ e f
0
From the above calculation for the homogeneous problem and calculation as in the proof of Theorem 5.1, step (vi), we obtain the
formula
1 2
−|y−x| /(4t)
u(x, t) = −
− ∫ ϕ(y)e dy
n
(2 √πt) R
n
t
1 2
−|y−x| /(4(t−τ))
+∫ ∫ f (y, τ ) e dy dτ .
−−− −−− − n
n
0 R (2 √ π(t − τ ) )
This function u(x, t) is a solution of the above inhomogeneous initial value problem provided
$$\phi\in C(\mathbb{R}^n),\ \ \sup_{\mathbb{R}^n}|\phi(x)|<\infty\]
and if
$$f\in C(\mathbb{R}^n\times[0,\infty)),\ \ M(\tau):=\sup_{\mathbb{R}^n}|f(y,\tau)|<\infty,\ 0\le\tau<\infty.\]
Theorem 6.2
Assume u ∈ C (D ¯
¯¯¯¯¯
¯
T ) , that u , u t xi xk exist and are continuous in D , and T
Proof
¯¯¯¯¯¯¯¯¯¯
¯ ¯¯¯¯¯¯¯¯¯¯
¯
Assume initially u t − △u < 0 in D . Let ε > 0 be small and
T 0 <ε <T . Since u ∈ C (DT −ε ) , there is an (x0 , t0 ) ∈ DT −ε
such that
$$u(x_0,t_0)=\max_{\overline{D_{T-\varepsilon}}} u(x,t).\]
Case (i)
Let (x 0, t0 ) ∈ DT −ε . Hence, since D T −ε is open,
ut (x0 , t0 ) = 0 ,u xl (x0 , t0 ) = 0 , l = 1, … , n and
$$\sum_{l,k=1}^n u_{x_lx_k}(x_0,t_0)\zeta_l\zeta_k\le0\ \ \mbox{for all}\ \zeta\in\mathbb{R}^n.\]
The previous inequality implies that u xk xk (x0 , t0 ) ≤ 0 for each k . Thus we arrived at a contradiction to u t − △u < 0 in D . T
Case (ii)
Assume (x , t ) ∈ Ω × {T − ε} . Then it follows as above △u ≤ 0 in (x , t ), and from
0 0 0 0 u(x0 , t0 ) ≥ u(x0 , t) , t ≤ t0 , one
concludes that u (x , t ) ≥ 0 . We arrived at a contradiction to u − △u < 0 in D again.
t 0 0 t T
$$u(\overline{x},\overline{t})=\max_{\overline{D_T}} u(x,t)\]
since ST −ε ⊂ ST and S T is compact. Thus, theorem is shown under the assumption ut − △u < 0 in DT . Now assume
ut − △u ≤ 0 in D . Set T
≤ max v(x, t) + kT
¯
¯¯¯
¯¯¯
¯
DT
= max v(x, t) + kT
ST
≤ max u(x, t) + kT ,
ST
Letting k → 0 , we obtain
$$\max_{\overline{D_T}} u(x,t)\le\max_{S_T} u(x,t).\]
¯
¯¯¯¯¯
¯
Since S T ⊂ DT , the theorem is shown.
□
If we replace in the above theorem the bounded domain Ω by R , then the result remains true provided we assume an {\it
n
additional} growth assumption for u. More precisely, we have the following result which is a corollary of the previous theorem. Set
for a fixed T , 0 < T < ∞ ,
$$D_T=\{ (x,t):\ x\in\mathbb{R}^n,\ 0<t<T\}.\]
Proposition 6.2
Assume u ∈ C (D ¯
¯¯¯¯¯
T
¯
) , that u , u
t xi xk exist and are continuous in D , T
Corollary
The initial value problem u − △u = 0 in D , u(x, 0) = f (x), x ∈ R , has a unique solution in the class defined by
t T
n
$$u(x,t)\le\sup_{y\in\mathbb{R}^n}u(y,k\tau)\le\sup_{y\in\mathbb{R}^n}u(y,0)\]
for kτ ≤ t ≤ (k + 1)τ , k = 0, … , N − 1 , where N = T /τ .
There is an ϵ > 0 such that 4a(T + ϵ) < 1 . Consider the comparison function
for fixed y ∈ R and for a constant μ > 0 . Since the heat kernel K(ix, iy, t) satisfies K
n
t = △Kx , we obtain
$$\frac{\partial}{\partial t}v_\mu-\triangle v_\mu=u_t-\triangle u\le0.\]
Set for a constant ρ > 0
$$D_{T,\rho}=\{(x,t):\ |x-y|<\rho,\ 0<t<T\}.\]
Then we obtain from Theorem 6.2 that
$$v_\mu(y,t)\le\max_{S_{T,\rho}}v_\mu,\]
where S T ,ρ ≡ ST of Theorem 6.2 with Ω = B ρ (y) , see Figure 6.3.1.
On the bottom of S T ,ρ we have, since μK > 0 ,
$$v_\mu(x,0)\le u(x,0)\le\sup_{z\in\mathbb{R}^n}f(z).\]
On the cylinder part |x − y| = ρ , 0 ≤ t ≤ T , of S T ,ρ it is
2
2
a|x| −n/2 ρ /(4(T +ϵ−t))
vμ (x, t) ≤ M e − μ (4π(T + ϵ − t)) e
2 2
a(|y|+ρ) −n/2 ρ /(4(T +ϵ))
≤ Me − μ (4π(T + ϵ)) e
≤ sup f (z)
n
z∈R
for all ρ > ρ (μ) , ρ sufficiently large. We recall that 4a(T + ϵ) < 1 .
0 0
Summarizing, we have
$$\max_{S_{T,\rho}}v_\mu(x,t)\le\sup_{z\in\mathbb{R}^n}f(z)\]
if ρ > ρ 0 (μ) . Thus
$$v_\mu(y,t)\le\max_{S_{T,\rho}}v_\mu(x,t)\le\sup_{z\in\mathbb{R}^n}f(z)\]
if ρ > ρ 0 (μ) .
Since
$$v_\mu(y,t)=u(y,t)-\mu\left(4\pi(T+\epsilon-t)\right)^{-n/2}\]
it follows
$$u(y,t)-\mu\left(4\pi(T+\epsilon-t)\right)^{-n/2}\le\sup_{z\in\mathbb{R}^n}f(z).\]
Letting μ → 0 , we obtain the assertion of the proposition.
□
The above maximum principle of Theorem 6.2 holds for a large class of parabolic differential operators, even for degenerate
equations.
Set
$$Lu=\sum_{i,j=1}^na^{ij}(x,t)u_{x_ix_j},\]
where a ij
∈ C (DT ) are real, a ij
=a
ji
, and the matrix (a ij
) is non-negative, that is,
$$\sum_{i,j=1}^na^{ij}(x,t)\zeta_i\zeta_j\ge 0\ \ \mbox{for all}\ \zeta\in\mathbb{R}^n,\]
and (x, t) ∈ D . T
Proof
(i) One proof is a consequence of the following lemma: let A , B real, symmetric and non-negative matrices. Non-negative
n
means that all eigenvalues are non-negative. Then trace~(AB) ≡ ∑ a b ≥ 0 , see an exercise.
i,j=1
ij
ij
(ii) Another proof is more directly: let U = (z , … , z ) , where z is an orthonormal system of eigenvectors to the eigenvalues
1 n l
n n
ij 2
0 ≤ ∑ a (x0 , t0 )ζi ζj = ∑ λi η
i
i,j=1 i=1
n n
2
0 ≥ ∑ uxi xj ζi ζj = ∑ vy yi η .
i i
i,j=1 i=1
It follows λ
i ≥0 and v yi yi ≤0 for all i.
Consequently
$$\sum_{i,j=1}^na^{ij}(x_0,t_0)u_{x_ix_j}(x_0,t_0)=\sum_{i=1}^n\lambda_iv_{y_iy_i}\le0.\]
□
¯
¯¯¯
c(x, 0) = c0 (x) x ∈ Ω (6.4.2)
∂c
= 0 on ∂Ω × (0, ∞). (6.4.3)
∂n
Here is Ω ⊂ R , n the exterior unit normal at the smooth parts of ∂Ω , D a positive constant and c
n
0 (x) a given function.
Remark. In application to diffusion problems, c(x, t) is the concentration of a substance in a solution, c0 (x) its initial
concentration and D the coefficient of diffusion.
The first Fick's rule says that
w = D∂c/∂n, (6.4.4)
where w is the flow of the substance through the boundary ∂Ω . Thus according to the Neumann boundary condition (6.4.3), we
assume that there is no flow through the boundary.
∂v
= 0 on ∂Ω, (6.4.1.2)
∂n
Assume Ω is bounded and ∂Ω sufficiently regular, then the eigenvalues of (6.4.1.1), (6.4.1.2) are countable and
$$0=\lambda_0<\lambda_1\le\lambda_2\le\ldots\to\infty.\]
Let v (x) be a complete system of orthonormal (in L
j
2
(Ω) ) eigenfunctions.
Solutions of (6.4.1.3) are
$$w_j(t)=C_je^{-D\lambda_jt},\]
where C are arbitrary constants.
j
Diffusion in a tube
Consider a solution in a tube, see Figure 6.4.1.1.
uniqueness result below that the solution of the initial-boundary value problem c(x , x , x , t) is independent of x and x .
1 2 3 1 2
ct = Dczz
c(z, 0) = c0 (z)
cz = 0, z = 0, z = l.
l
2 π
Cn = ∫ c0 (z) cos( nz) dz, n ≥ 1.
l 0
l
c(x, 0) = 0
∂c
= 0 on ∂Ω × (0, ∞).
∂n
Call option
Here is V (S, t) := C (S, t) , where C (S, t) is the value of the (European) call option. In this case we have following side
conditions to (6.5.1):
C (S, T ) = max{S − E, 0} (6.5.2)
C (0, t) = 0 (6.5.3)
where E and T are positive constants, E is the exercise price and T the expiry.
Side condition (6.5.2) means that the value of the option has no value at time T if S(T ) ≤ E , condition (6.5.3) says that it makes
no sense to buy assets if the value of the asset is zero, condition (6.5.4) means that we buy assets if its value becomes large, see
Figure 6.5.1, where the side conditions are indicated.
where
x
1 −y
2
/2
N (x) = −− ∫ e dy,
√2π −∞
2
ln(S/E) + (r + σ /2)(T − t)
d1 = −−− − ,
σ√ T − t
2
ln(S/E) + (r − σ /2)(T − t)
d2 = .
−−− −
σ√ T − t
where
$$k=\frac{r}{\sigma^2/2}.\]
Initial condition (6.5.6) implies
x
v(x, 0) = max{ e − 1, 0}. (6.5.7)
u(x, 0) = u0 (x),
with
$$u_0(x)=\max\left\{e^{(k+1)x/2}-e^{(k-1)x/2},0\right\}.\]
A solution of this initial value problem is given by Poisson's formula
$$u(x,\tau)=\frac{1}{2\sqrt{\pi \tau}}\int_{-\infty}^{+\infty}\ u_0(s)e^{-(x-s)^2/(4\tau)}\ ds.\]
−−
Changing variable by q = (s − x)/(√2τ ) , we get
+∞
1 −− 2
−q /2
u(x, τ ) = −− ∫ u0 (q √2τ + x)e dq
√2π −∞
= I1 − I2 ,
where
∞
1 2
(k+1)(x+q√2τ ) −q /2
I1 = −− ∫ e e dq
√2π −x/( √2τ)
∞
1 2
(k−1)(x+q√2τ ) −q /2
I2 = −− ∫ e e dq.
√2π −x/( √2τ)
where
x 1 −−
d2 = −− + (k − 1)√2τ
√2τ 2
di
1 2
−s /2
N (di ) = −− ∫ e ds, i = 1, 2.
√2π −∞
Combining the formula for u(x, τ ), definition (6.5.8) of v(x, τ ) and the previous settings x = ln(S/E) , 2
τ = σ (T − t)/2 and
C = Ev(x, τ ) , we get finally the formula of Theorem 6.4.
In general, the solution u of the initial value problem for the heat equation is not uniquely defined, see for example [10], pp. 206.
Uniqueness. The uniqueness follows from the growth assumption (6.5.4). Assume there are two solutions of (6.5.1), (6.5.2)-(
6.5.4), then the difference W (S, t) satisfies the differential equation (6.5.1) and the side conditions
where S = Ee , t = T − 2τ /(σ
x 2
) . Thus we have the growth property
a|x| 1
|u(x, τ )| ≤ M e , x ∈ R , (6.5.10)
0 ≤ τ ≤ σ T /2 , with the initial condition u(x, 0) = 0 is uniquely defined in the class of functions satisfying the growth condition
2
Put option
Here is V (S, t) := P (S, t) , where P (S, t) is the value of the (European) put option. In this case we have following side conditions
to (6.5.1):
−r(T −t)
P (0, t) = Ee (6.5.12)
Q6.1
Show that the solution u(x, t) given by Poisson's formula satisfies
\[ \inf_{z\in \mathbb{R}^n}\varphi(z)\le u(x,t)\le\sup_{z\in\mathbb{R}^n}\varphi(z)\ , $$ provided φ(x) is continuous and bounded on
R .
n
Q6.2
Solve for given f (x) and μ ∈ R the initial value problem
1
1 1
ut + ux + μuxxx = 0 in R × R+
u(x, 0) = f (x) .
Q6.3
Show by using Poisson's formula: (i) Each function f ∈ C ([a, b]) can be approximated uniformly by a sequence f n ∈ C
∞
[a, b] . (ii) In (i)
we can choose polynomials f (Weierstrass's approximation theorem).
n
2 2
|y−x| |y−x|
Hint: Concerning (ii), replace the kernel K = exp(− 4t
) by a sequence of Taylor polynomials in the variable z = − 4t
.
Q6.4
Let u(x, t) be a positive solution of
′
ut = μuxx , t > 0, $$where\(μ\)isaconstant. Showthat\(θ := −2μux /u\)isasolutionof Burger sequation$$ θt (6.E.1)
Q6.5
Assume u1 (s, t), . . . , un (s, t) are solutions of ut = uss . Show that ∏
n
k=1
uk (xk , t) is a solution of the heat equation ut − △u = 0 in
R
n
× (0, ∞) .
Q6.6
Let A , B are real, symmetric and non-negative matrices. Non-negative means that all eigenvalues are non-negative. Prove that trace
n
≥0 .
ij
(AB) ≡ ∑ a b ij
i,j=1
Hint: (i) Let U = (z1 , … , zn ) , where z is an orthonormal system of eigenvectors to the eigenvalues λ of the matrix B . Then
l l
−−
√λ1 0 ⋯ 0
⎛ ⎞
−−
⎜0 √λ2 ⋯ 0 ⎟
T
X =U ⎜ ⎟U (6.E.2)
⎜ ⎟
⎜⋯ ⋯ ⋯ ⋯ ⎟
−−
⎝ √λn ⎠
0 0 ⋯
T ⎜0 λ2 ⋯ 0⎟
U BU = ⎜ ⎟. (6.E.3)
⎜⋯ ⋯ ⋯ ⋯⎟
⎝ ⎠
0 0 ⋯ λn
(ii) trace (QR) = trace (RQ). (iii) Let μ (C ), … μ (C ) are the eigenvalues of a real symmetric
1 n n×n -matrix. Then trace
C =∑
n
l=1
μl (C ) , which follows from the fundamental lemma of algebra:
n n−1
det (λI − C ) = λ − (c11 + … + cnn )λ +…
≡ (λ − μ1 ) ⋅ … ⋅ (λ − μn )
n n+1
= λ − (μ1 + … + μn )λ +…
Q6.8
Prove the following comparison principle: Assume Ω is bounded and u, v satisfy the regularity assumptions of the maximum principle.
Then
ut − △u ≤ vt − △v in DT
u ≤ v on ST
imply that u ≤ v in D . T
Q6.9
Show that the comparison principle implies the maximum principle.
Q6.10
Consider the boundary-initial value problem
ut − △u = f (x, t) in DT
¯
¯¯¯¯¯
¯
where f , ϕ are given.\\ Prove uniqueness in the class u, u t, ux
i
xk ∈ C (DT ) .
Q6.11
¯
¯¯¯¯¯
¯
Assume u, v 1, v2 ∈ C
2
(DT ) ∩ C (DT ) , and u is a solution of the previous boundary-initial value problem and v , v satisfy 1 2
v1 ≤ ϕ ≤ v2 on ST .
Q6.12
Show by using the comparison principle: let u be a sufficiently regular solution of
ut − △u = 1 in DT
u = 0 on ST ,
then 0 ≤ u(x, t) ≤ t in DT .
Q6.13
Discuss the result of Theorem 6.3 for the case
n n
i,j=1 i
Q6.14
Show that
∞
2
−n t
u(x, t) = ∑ cn e sin(nx), (6.E.6)
n=1
where
π
2
cn = ∫ f (x) sin(nx) dx, (6.E.7)
π 0
u(x, 0) = f (x),
u(0, t) = 0,
u(π, t) = 0,
if f ∈ C
4
(R) is odd with respect to 0 and 2π-periodic.
Q6.15
(i) Find the solution of the diffusion problem c = Dc in 0 ≤ z ≤ l$, $0 ≤ t < ∞ ,
t zz D = const. > 0 , under the boundary conditions
c (z, t) = 0 if z = 0 and z = l and with the given initial concentration
z
c0 = const. 0 ≤z ≤h
c(z, 0) = c0 (z) := { (6.E.8)
0 h < z ≤ l.
Q6.16
Solve the initial-boundary value problem (rotationally symmetric solution in a ball): find c(r, t) on (0, R) × (0, ∞) of
∂c 1 ∂ 2
∂c
= (Dr ) − kc (6.E.9)
2
∂t r ∂r ∂r
⎧ 0 0 < r < R0
h(r) = ⎨ r−R0 , (6.E.13)
⎩ c0 R0 < r < R
R−R0
Q6.17
Prove the Black-Scholes formula for an European put option.
Hint: Put-call parity.
Q6.18
Prove the put-call parity for European options
−r(T −t)
C (S, t) − P (S, t) = S − Ee (6.E.14)
by using the following uniqueness result: Assume W is a solution of (6.5.1) under the side conditions W (S, T ) = 0 , W (0, t) = 0 and
W (S, t) = O(S) as S → ∞ , uniformly on 0 ≤ t ≤ T . Then W (S, t) ≡ 0 .
Q6.19
Prove that a solution V (S, t) of the initial-boundary value problem (6.5.1) in Ω under the side conditions (i) V (S, T ) = 0 , S ≥0 , (ii)
¯
¯¯¯
V (0, t) = 0 , 0 ≤ t ≤ T , (iii) lim S→∞ V (S, t) = 0 uniformly in 0 ≤ t ≤ T , is uniquely determined in the class C 2
(Ω) ∩ C (Ω) .
Q6.20
Prove that a solution V (S, t) of the initial-boundary value problem (6.5.1) in Ω, under the side conditions (i) V (S, T ) = 0 , S ≥ 0 , (ii)
V (0, t) = 0 , 0 ≤ t ≤ T , (iii) V (S, t) = S + o(S) as S → ∞ , uniformly on 0 ≤ t ≤ T , satisfies |V (S, t)| ≤ cS for all S ≥ 0 and
0 ≤t ≤T .
$$ \triangle u=0. \]
Solutions of the Laplace equation are called potential functions or harmonic functions. The Laplace equation is called also potential equation.
The general elliptic equation for a scalar function u(x), x ∈ Ω ⊂ R , is n
$$Lu:=\sum_{i,j=1}^na^{ij}(x)u_{x_ix_j}+\sum_{j=1}^n b^j(x)u_{x_j}+c(x)u=f(x),\]
$$\sum_{i,j=1}^na^{ij}u_{x_ix_j}=\triangle v,\]
where v(y) ,
:= u(T y) T stands for the above composition of mappings.
1 3/30/2022
7.1: Fundamental Solution
Here we consider particular solutions of the Laplace equation in R of the type n
$$u(x)=f(|x-y|),\]
where y ∈ R is fixed and f is a function which we will determine such that u defines a solution if the Laplace equation.
n
Set r = |x − y| , then
xi − yi
′
ux = f (r)
i
r
2 2
(xi − yi ) 1 (xi − yi )
′′ ′
ux xi = f (r) + f (r) ( − )
i
2 3
r r r
′′
n−1 ′
△u = f (r) + f (r).
r
c1 ln r + c2 n =2
(7.1.1)
2−n
c1 r + c2 n ≥3
\right.\]
with constants c , c .
1 2
\right.
\]
is called singularity function associated to the Laplace equation. Here is ωn the area of the n-dimensional unit sphere which is
given byω = 2π /Γ(n/2), where
n
n/2
Proof. Let B ρ (y) ⊂Ω be a ball. Set Ω ρ (y) = Ω ∖ Bρ (y) . See Figure 7.2.2 for notations.
∂v ∂v ∂v
∫ u dS = ∫ u dS + ∫ u dS.
∂ Ωρ (y)
∂n ∂Ω
∂n ∂ Bρ (y)
∂n
n−1
≤ M (∫ s(ρ) dS + C ωn ρ ),
∂ Bρ (y)
∂u O(ρ| ln ρ|) n =2
∫ v dS = { (7.2.4)
∂ Bρ (y) ∂n O(ρ) n ≥ 3.
1
n−1
= ∫ u dS + O(ρ )
n−1
ωn ρ ∂ Bρ (y)
1
n−1
= u(x0 ) ∫ dS + O(ρ ),
n−1
ωn ρ ∂ Bρ (y)
n−1
= u(x0 ) + O(ρ ).
for an x0 ∈ ∂ Bρ (y) .
Combining this estimate and (7.2.4), we obtain the representation formula of the theorem.
□
△u = 0 in Ω .
D u(y) exist and are continuous for all l since one can change differentiation with integration in right hand sides of the
l
representation formulas.
□
Remark. In fact, a function which is harmonic in Ω is even real analytic in Ω, see an exercise.
Proposition 7.2 (Mean value formula for harmonic functions). Assume u is harmonic in Ω. Then for each B ρ (x) ⊂⊂ Ω
1
u(x) = ∫ u(y) dSy . (7.2.1.1)
n−1
ωn ρ ∂ Bρ (x)
Proof. Consider the case n ≥ 3 . The assertion follows from (7.2.6) where Ω := B ρ (x) since r = ρ and
1 ∂u 1 ∂u
∫ dSy = ∫ dSy
n−2 n−2
∂ Bρ (x) r ∂ny ρ ∂ Bρ (x)
∂ny
1
= ∫ △u dy
ρn−2 Bρ (x)
= 0.
We recall that a domain Ω ∈ R is called connected if Ω is not the union of two nonempty open subsets
n
Ω1 , Ω2 such that
Ω ∩ Ω = ∅ . A domain in R is connected if and only if its path connected.
n
1 2
Proposition 7.3 (Maximum principle). Assume u is harmonic in a connected domain and achieves its supremum or infimum in Ω.
Then u ≡ const. in Ω.
Proof. Consider the case of the supremum. Let x 0 ∈ Ω such that
u(x0 ) = sup u(x) =: M . (7.2.1.2)
Ω
Set
Ω1 := {x ∈ Ω : u(x) = M } and Ω 2 := {x ∈ Ω : u(x) < M } . The set Ω is not empty since x
1 0 ∈ Ω1 . The set Ω is open since
2
¯¯¯¯¯¯¯¯¯¯¯¯¯
¯
u ∈ C
2
. Consequently, Ω is empty if we can show that Ω is open. Let x ∈ Ω , then there is a ρ > 0 such that
(Ω) 2 1
¯¯
¯
1 0 Bρ (x)
0
¯¯
¯
⊂Ω
and u(x) = M for all x ∈ B (x̄) . If not, then there exists ρ > 0 and x̂ such that
ρ
0
¯
¯
¯
¯¯¯
Corollary. Assume Ω is connected and bounded, and u ∈ C 2
(Ω) ∩ C (Ω) is harmonic in Ω. Then u achieves its minimum and its
maximum on the boundary ∂Ω .
Remark. The previous corollary fails if Ω is not bounded as simple counterexamples show.
u = Φ on ∂Ω, (7.3.1.2)
$$
\Omega=\{x\in B_1(0):\ x_2>0\},
\]
u = 0 on ∂Ω ∖ {0}.
∂u
= Φ on ∂Ω, (7.3.2.2)
∂n
∂u
+ hu = Φ on ∂Ω, (7.3.3.2)
∂n
where Φ and h are given and continuous on ∂Ω .e Φ and h are given and continuous on ∂Ω .
Proposition 7.6. Assume Ω is bounded and sufficiently regular, then a solution to the mixed problem is uniquely determined in the
¯
¯¯¯
class u ∈ C 2
(Ω) provided h(x) ≥ 0 on ∂Ω and h(x) > 0 for at least one point x ∈ ∂Ω .
Proof. Exercise. Hint: Multiply the differential equation △w = 0 by w and integrate the result over Ω.
where γ(y, x) is a fundamental solution. In general, u does not satisfies the boundary condition in the above boundary value
problems. Since γ = s + ϕ , see Section 7.2, where ϕ is an arbitrary harmonic function for each fixed x, we try to find a ϕ such
that u satisfies also the boundary condition.
Consider the Dirichlet problem, then we look for a ϕ such that
γ(y, x) = 0, y ∈ ∂Ω, x ∈ Ω. (7.4.2)
Then
∂γ(y, x)
u(x) = − ∫ u(y) dSy , x ∈ Ω. (7.4.3)
∂Ω
∂ny
Suppose that u achieves its boundary values Φ of the Dirichlet problem, then
∂γ(y, x)
u(x) = − ∫ Φ(y) dSy , (7.4.4)
∂Ω
∂ny
We claim that this function solves the Dirichlet problem (7.3.1.1), (7.3.1.2).
A function γ(y, x) which satisfies (7.4.2), and some additional assumptions, is called Green's function. More precisely, we define a
Green function as follows.
¯
¯¯¯
Definition. A function G(y, x), y, x ∈ Ω , x ≠ y , is called Green function associated to Ω and to the Dirichlet problem (7.3.1.1),
(7.3.1.2) if for fixed x ∈ Ω, that is we consider G(y, x) as a function of y , the following properties hold:
¯
¯¯¯
(i) G(y, x) ∈ C 2
(Ω ∖ {x}) ∩ C (Ω ∖ {x}) ,△ y G(y, x) = 0, x ≠ y .
¯
¯¯¯
(ii) G(y, x) − s(|x − y|) ∈ C 2
(Ω) ∩ C (Ω) .
(iii) G(y, x) = 0 if y ∈ ∂Ω , x ≠ y .
Remark. We will see in the next section that a Green function exists at least for some domains of simple geometry. Concerning the
existence of a Green function for more general domains see [13].
It is an interesting fact that we get from (i)-(iii) of the above definition two further important properties, provided Ω is bounded,
sufficiently regular and connected.
Proposition 7.7. A Green function has the following properties. In the case n = 2 we assume {\rm diam} Ω < 1 .
(A) G(x, y) = G(y, x)\ \ (symmetry).
(B) 0 < G(x, y) < s(|x − y|), x, y ∈ Ω, x ≠ y .
¯
¯¯¯
¯¯
Proof. (A) Let x
(1) (2)
, x ∈ Ω . Set Bi = Bρ (x
(i)
) , i = 1, 2 . We assume Bi ⊂ Ω and B1 ∩ B2 = ∅ . Since G(y, x
(1)
) and
¯
¯¯¯¯
¯ ¯
¯¯¯¯
¯
G(y, x
(2)
) are harmonic in Ω ∖ (B 1 ∪ B2 ) we obtain from Green's identity, see Figure 7.4.1 for notations,
(1)
∂ (2)
0 = ∫ (G(y, x ) G(y, x )
¯
¯¯¯
¯¯¯
∂(Ω∖( B1 ∪B2 ))
¯
¯¯¯
¯¯¯
∂ny
∂
(2) (1)
−G(y, x ) G(y, x ))dSy
∂ny
∂ ∂
(1) (2) (2) (1)
= ∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x )) dSy
∂Ω ∂ny ∂ny
(1)
∂ (2) (2)
∂ (1)
+ ∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x )) dSy
∂B1
∂ny ∂ny
∂ ∂
(1) (2) (2) (1)
+ ∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x )) dSy .
∂B2 ∂ny ∂ny
The integral over ∂Ω is zero because of property (iii) of a Green function, and
(1)
∂ (2) (2)
∂ (1)
∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x ))dSy
∂B1 ∂ny ∂ny
(1) (2)
→ G(x ,x ),
∂ ∂
(1) (2) (2) (1)
∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x )) dSy
∂B2
∂ny ∂ny
(2) (1)
→ −G(x ,x )
as ρ → 0 .
This follows by considerations as in the proof of Theorem 7.1.
(B) Since
From the definition of s(|x − y|) it follows that ϕ(y, x) < 0 if y ∈ ∂Ω . Thus, since △y ϕ = 0 in Ω , the maximum-minimum
principle implies that ϕ(y, x) < 0
for all y, x ∈ Ω. Consequently
G(y, x) < s(|x − y|), x, y ∈ Ω, x ≠ y. (7.4.7)
Fix x ∈ Ω and let B ρ (x) be a ball such that B ρ (x) ⊂Ω for all 0 < ρ < ρ . There is a sufficiently small ρ
0 0 >0 such that for each
ρ, 0 < ρ < ρ ,
0
G(y, x) = 0 if y ∈ ∂Ω
the sphere ∂ B (0) , that is, in particular |y||y | = R , see Figure 7.4.1.1 for notations.
R
′ 2
Set
$$G(x,y)=s(r)-s\left(\frac{\rho}{R}r_1\right),\]
where s is the singularity function of Section 7.1, r = |x − y| and
$$\rho^2=\sum_{i=1}^ny_i^2,\ \ \ r_1=\sum_{i=1}^n\left(x_i-\frac{R^2}{\rho^2}y_i\right)^2.\]
This function G(x, y) satisfies (i)-(iii) of the definition of a Green function. We claim that
$$u(x)=-\int_{\partial B_R(0)} \frac{\partial}{\partial n_y}G(x,y)\Phi\ dS_y\]
is a solution of the Dirichlet problem (7.3.1.1), (7.3.1.2). This formula is also true for a large class of domains Ω ⊂ R , see [13]. n
Lemma.
$$-\frac{\partial}{\partial n_y}G(x,y)\bigg|_{|y|=R}=\frac{1}{R\omega_n}\frac{R^2-|x|^2}{|y-x|^n}.\]
Proof. Exercise.
Set
2 2
1 R − |x |
H (x, y) = n
, (7.4.1.1)
Rωn |y − x|
¯
¯¯¯
is the solution of the first boundary value problem (7.3.1.1), (7.3.1.2) in the class C 2
(Ω) ∩ C (Ω) .
Proof. The proof follows from following properties of H (x, y):
i. H (x, y) ∈ C , |y| = R, |x| < R, x ≠ y ,
∞
= I1 + I2 ,
where
$$H(x,y)<\frac{\epsilon}{2M\omega_nR^{n-1}}\]
if x and y satisfy |x − ζ| < δ ′
, |y − ζ| > δ , see Figure 7.4.1.2 for notations.
Remark. Define δ ∈ [0, π] through cos δ = x ⋅ y/(|x||y|), then we write Poisson's formula of Theorem 7.2 as
$$u(x)=\frac{R^2-|x|^2}{\omega_nR}\int_{\partial B_R(0)}\Phi(y)\frac{1}{\left(|x|^2+R^2-2|x|R\cos\delta\right)^{n/2}}\ dS_y.\]
In the case n = 2 we can expand this integral in a power series with respect to ρ := |x|/R if |x| < R , since
2 2 2
R − |x | 1 −ρ
=
2 2
|x| + R − 2|x|R cos δ ρ − 2ρ cos δ + 1
∞
n
= 1 +2 ∑ρ cos(nδ),
n=1
see [16], pp. 18 for an easy proof of this formula, or [4], Vol. II, p. 246.
where z = x 1 + i x2 ,z
0 = y1 + i y2 .
u = 0 on ∂Ω, (7.5.2)
where f is given.
We need the following lemma concerning volume potentials. We assume that Ω is bounded and sufficiently regular such that all the
following integrals exist. See [6] for generalizations concerning these assumptions.
Let for x ∈ R , n ≥ 3 ,
n
1
V (x) = ∫ f (y) dy (7.5.3)
n−2
Ω |x − y|
∂ 1
Vxi (x) = ∫ f (y) ( ) dy, if n ≥ 3,
n−2
Ω
∂xi |x − y|
∂ 1
Vxi (x) = ∫ f (y) (ln( )) dy if n = 2.
Ω
∂xi |x − y|
(ii) If f ∈ C
1
(Ω) , then V ∈ C
2
(Ω) and
△V = −(n − 2)ωn f (x), x ∈ Ω, n ≥ 3
△V = −2πf (x), x ∈ Ω, n = 2.
we obtain
∂ 1 ∂ 1
= ∫ f (y) ( ) dy + ∫ f (y) ( ) dy
Ω∖Bρ
∂xi |x − y| Bρ
∂xi |x − y|
∂ 1
= ∫ f (y) ( ) dy
Ω∖Bρ ∂xi |x − y|
∂ 1 1
+∫ (− (f (y) ) + fy (y) ) dy
i
Bρ ∂yi |x − y| |x − y|
∂ 1
= ∫ f (y) ( ) dy
Ω∖Bρ ∂xi |x − y|
1 1
+∫ fy (y) dy − ∫ f (y) ni dSy ,
i
Bρ |x − y| ∂Bρ |x − y|
where n is the exterior unit normal at ∂B . It follows that the first and second integral is in C
ρ
1
. The second integral is also in
(Ω)
−1
Because of △ x (|x − y| ) = 0, x ≠ y , it follows
n
∂ 1
△V = ∫ ∑ fy (y) ( ) dy
i
Bρ
∂xi |x − y|
i=1
n
∂ 1
−∫ f (y) ∑ ( ) ni dSy .
∂Bρ
∂xi |x − y|
i=1
△V = I1 + I2 , (7.5.7)
where
n
yi − xi
I1 = ∫ ∑ fy (y) dy
i
3
Bρ (x) i=1 |x − y|
1
I2 = − ∫ f (y) dSy .
2
∂ Bρ (x) ρ
We recall that n ⋅ (y − x) = ρ if y ∈ ∂ B ρ (x) . It is I 1 = O(ρ) as ρ → 0 and for I we obtain from the mean value theorem of the
2
1
¯¯
I2 = − f (ȳ )∫ dSy
2
ρ ∂ Bρ (x)
¯¯
= −ωn f (ȳ ),
In the following we assume that Green's function exists for the domain Ω, which is the case if Ω is a ball.
¯
¯¯¯
Theorem 7.3. Assume f ∈ C
1
(Ω) ∩ C (Ω) . Then
where ϕ is a potential function with respect to x or y , we obtain from the above lemma that
1 1
△u = △∫ f (y) dy + ∫ △x ϕ(x, y)f (y) dy
4π Ω |x − y| Ω
= −f (x),
where x ∈ Ω. It remains to show that u achieves its boundary values. That is, for fixed x 0 ∈ ∂Ω we will prove that
lim u(x) = 0. (7.5.11)
x→x0 , x∈Ω
Set
u(x) = I1 + I2 , (7.5.12)
where
we obtain, if x ∈ B ρ (x0 ) ∩Ω ,
M dy
2
| I2 | ≤ ∫ + O(ρ )
4π Ω∩Bρ ( x0 ) |x − y|
M dy
2
≤ ∫ + O(ρ )
4π B2ρ( x) |x − y|
2
= O(ρ )
since G(x 0, y) = 0 if y ∈ Ω ∖ B
ρ (x0 ) and G(x, y) is uniformly continuous in x ∈ B ρ/2
(x0 ) ∩ Ω and y ∈ Ω ∖ B ρ (x0 ) , see Figure
7.5.1.
Remark. For the proof of (ii) in the above lemma it is sufficient to assume that f is H\"older continuous. More precisely, let
f ∈ C (Ω) , 0 < λ < 1 , then V ∈ C (Ω) , see for instance [9].
λ 2,λ
Q7.1
Let γ(x, y) be a fundamental solution to △, y ∈ Ω . Show that
$$
-\int_\Omega\gamma(x,y)\ \triangle\Phi(x)\ dx=\Phi(y)\quad\hbox{for all}\ \
\Phi\in C_0^2(\Omega)\ .
\]
Hint: See the proof of the representation formula.
Q7.2
Show that |x | sin(k|x|) is a solution of the Helmholtz equation
−1
$$
\triangle u+k^2u=0 \ \mbox{in}\ \mathbb{R}^n\setminus\{0\}.
\]
Q7.3
¯
¯¯¯
Assume u ∈ C 2
(Ω) , Ω bounded and sufficiently regular, is a solution of
3
△u = u in Ω
u = 0 on ∂Ω.
Show that u = 0 in Ω.
Q7.4
Let Ω α = {x ∈ R
2
: x1 > 0, 0 < x2 < x1 tan α} , 0 < α ≤ π . Show that
π
k
π
u(x) = r α
sin( kθ) (7.E.1)
α
is a harmonic function in Ω satisfying u = 0 on ∂Ω , provided k is an integer. Here (r, θ) are polar coordinates with the center at
α α
(0, 0).
Q7.5
¯
¯¯¯
Let u ∈ C
2
(Ω) be a solution of △u = 0 on the quadrangle Ω = (0, 1) × (0, 1) satisfying the boundary conditions
¯
¯¯¯
u(0, y) = u(1, y) = 0 for all y ∈ [0, 1] and u (x, 0) = u (x, 1) = 0 for all x ∈ [0, 1]. Prove that u ≡ 0 in Ω.
y y
Q7.6
Let u ∈ C 2
(R )
n
be a solution of △u = 0 in R satisfying u ∈ L
n 2
(R )
n
, i. e., ∫ R
n
2
u (x) dx < ∞. Show that u ≡ 0 in R . n
Hint: Prove
2
const. 2
∫ |∇u | dx ≤ ∫ |u | dx, (7.E.2)
2
BR (0) R B2R (0)
ϵ
2
Q7.7
Show that a bounded harmonic function defined on R must be a constant (a theorem of Liouville).
n
Q7.8
¯¯¯¯¯¯¯¯¯¯¯
¯
Assume u ∈ C 2
(B1 (0)) ∩ C (B1 (0) ∖ {(1, 0)}) is a solution of
△u = 0 in B1 (0)
Q7.9
Assume Ω ⊂R
n
is bounded and u, v ∈ C
2
(Ω) ∩ C (Ω)
¯
¯¯¯
satisfy △u = △v and max∂Ω |u − v| ≤ ϵ for given ϵ>0 . Show that
max¯¯¯¯ |u − v| ≤ ϵ .
Ω
Q7.10
¯¯¯¯¯¯¯¯¯¯¯
¯ ¯
¯¯¯
Set Ω = R ∖ B (0) and let u ∈ C (Ω) be a harmonic function in Ω satisfying lim
n
1
2
|x|→∞ u(x) = 0 . Prove that
$$
\max_{\overline{\Omega}}|u|=\max_{\partial\Omega}|u|\ .
\]
Hint: Apply the maximum principle to Ω ∩ B R (0) , R large.
Q7.11
Let Ωα = {x ∈ R
2
: x1 > 0, 0 < x2 < x1 tan α} , 0 <α ≤π , Ωα,R = Ωα ∩ BR (0) , and assume f is given and bounded on
¯
¯¯¯¯¯¯¯¯
¯
Ωα,R .
¯
¯¯¯¯¯¯¯¯
¯
Show that for each solution u ∈ C (Ω ) ∩ C (Ω 1
α,R
2
α,R ) of △u = f in
Ω satisfying u = 0 on ∂ Ω ∩ B (0) , holds:
α,R α,R R
Q7.13
¯¯¯¯¯¯¯¯¯¯¯
¯
Let u ∈ C (B (0)) ∩ C (B (0)) satisfying u ≥ 0, △u = 0 in B
2
a a a (0) . Prove (Harnack's inequality):
$$
{a^{n-2}(a-|\zeta|)\over (a+|\zeta|)^{n-1}}u(0)\le u(\zeta)\le
{a^{n-2}(a+|\zeta|)\over (a-|\zeta|)^{n-1}}u(0)\ .
\]
Hint: Use the formula (see Theorem 7.2)
2 2
a − |y | u(x)
u(y) = ∫ dSx (7.E.4)
n
aωn |x|=a |x − y|
for y = ζ and y = 0 .
Q7.14
Let ϕ(θ) be a 2π-periodic C -function with the Fourier series
4
n=0
Show that
∞
n
u = ∑ (an cos(nθ) + bn sin(nθ)) r (7.E.6)
n=0
Q7.15
Assume u ∈ C 2
(Ω) satisfies △u = 0 in Ω. Let B a (ζ) be a ball such that its closure is in Ω.
Show that
|α|
|α|γn
α
|D u(ζ)| ≤ M ( ) , (7.E.7)
a
Q7.16
Use the result of the previous exercise to show that u ∈ C 2
(Ω) satisfying △u = 0 in Ω is real analytic in Ω.
Hint: Use Stirling's formula
n −n −−− 1
n! = n e ( √2πn + O ( − )) (7.E.8)
√n
n ≤ c e n! which follows from Stirling's formula. See Section 3.5 for the definition of a real analytic function.
n n
Q7.17
Assume Ω is connected and u ∈ C (Ω) is a solution of △u = 0 in Ω. Prove that u ≡ 0 in Ω if D
2 α
u(ζ) = 0 for all α , for a point
ζ ∈ Ω . In particular, u ≡ 0 in Ω if u ≡ 0 in an open subset of Ω .
Q7.18
Let Ω = {(x 1, x2 , x3 ) ∈ R
3
: x3 > 0} , which is a half-space of R . Show that 3
1 1
G(x, y) = − , (7.E.9)
¯
¯¯
4π|x − y| 4π|x − y |
where ȳ = (y
¯
¯
1, y2 , −y3 ) , is the Green function to Ω.
Q7.19
Let Ω = {(x 1, x2 , x3 ) ∈ R
3
: x
2
1
+x
2
2
+x
2
3
2
< R , x3 > 0} , which is half of a ball in R . Show that 3
1 R
G(x, y) = −
4π|x − y| 4π|y||x − y ⋆ |
1 R
− + ,
¯
¯¯ ⋆
¯
¯¯
4π|x − y | 4π|y||x − y |
⋆
where ȳ = (y , y , −y ) , y
¯
¯
1 2 3
⋆ 2
= R y/(|y | )
2
and ȳ
¯
¯ ¯¯
= R ȳ
2
/(|y | )
2
,
is the Green function to Ω.
Q7.20
Let Ω = {(x 1, x2 , x3 ) ∈ R
3
: x2 > 0, x3 > 0} , which is a wedge in R . Show that 3
1 1
G(x, y) = −
¯¯
4π|x − y| 4π|x − ȳ |
1 1
− + ,
′ ′
4π|x − y | ¯¯
4π|x − ȳ |
′
where y = (y , y , −y ) , y
¯
¯¯
1 2 3
′
= (y1 , −y2 , y3 ) and y ¯
¯¯
= (y1 , −y2 , −y3 ) ,
is the Green function to Ω.
Q7.21
Find Green's function for the exterior of a disk, i. e., of the domain Ω = {x ∈ R 2
: |x| > R} .
Q7.22
Find Green's function for the angle domain Ω = {z ∈ C : 0 < arg z < απ} , 0 < α < π .
Q7.23
Find Green's function for the slit domain Ω = {z ∈ C : 0 < arg z < 2π} .
Q7.24
Let for a sufficiently regular domain Ω ∈ R , a ball or a quadrangle for example,
n
¯
¯¯¯ ¯
¯¯¯
where K(x, y) is continuous in Ω × Ω where x ≠ y , and which satisfies
c
|K(x, y)| ≤ (7.E.11)
α
|x − y|
Q7.25
Prove (i) of the lemma of Section 7.5.
Hint: Consider the case n ≥ 3 . Fix a function η ∈ C 1
(R )
1
satisfying 0 ≤η ≤1 , 0 ≤η
′
≤2 , η(t) = 0 for t ≤1 , η(t) = 1 for
t ≥ 2 and consider for ϵ > 0 the regularized integral
dy
Vϵ (x) := ∫ f (y)ηϵ , (7.E.12)
n−2
Ω |x − y|
∂ 1
∫ f (y) ( ) dy (7.E.13)
n−2
Ω ∂xi |x − y|
as ϵ → 0 .
C G P
CAPILLARY EQUATION Green's Function Parabolic Equations
1.3: Partial Differential Equations 7.4: Green's Function for \(\Delta\) 6: Parabolic Equations
Poisson's formula
D I 6.1: Poisson's Formula
Diffusion in a tube Inhomogeneous Equations
6.4.1: Fourier's Method 4.3: Inhomogeneous Equations
DIRICHLET INTEGRAL
1.3: Partial Differential Equations
Glossary
Sample Word 1 | Sample Definition 1