0% found this document useful (0 votes)
89 views6 pages

Partial Differential Equations Example Sheet 1: Books

This document provides references for further reading on partial differential equations (PDEs). It recommends several books that cover PDE topics beyond the course and are very good resources. These include introductions to PDEs, Fourier analysis, distributions, and more advanced treatments. It also lists previous Cambridge lecture notes that are still useful for the PDE topics in the course.

Uploaded by

Nasih Ahmad
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
89 views6 pages

Partial Differential Equations Example Sheet 1: Books

This document provides references for further reading on partial differential equations (PDEs). It recommends several books that cover PDE topics beyond the course and are very good resources. These include introductions to PDEs, Fourier analysis, distributions, and more advanced treatments. It also lists previous Cambridge lecture notes that are still useful for the PDE topics in the course.

Uploaded by

Nasih Ahmad
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Partial Differential Equations Example sheet 1

David Stuart
[email protected]

Books
In addition to the sets of lecture notes written by previous lecturers ([1, 2]) which
are still useful, the books [4, 3, 5] are very good for the PDE topics in the course,
and go well beyond the course also. If you want to read more on distributions [6]
is most relevant. Also [7, 8] are useful; the books [9] are more advanced, but the
first volume may be helpful.

References
[1] T.W. Körner, Cambridge Lecture notes on PDE, available at
https://www.dpmms.cam.ac.uk/ twk
[2] M. Joshi and A. Wassermann, Cambridge Lecture notes on PDE,available at
http://www.damtp.cam.ac.uk/user/dmas2
[3] G.B. Folland, Introduction to Partial Differential Equations, Princeton 1995,
QA 374 F6
[4] L.C. Evans, Partial Differential Equations,AMS Graduate Studies in Mathe-
matics Vol 19, QA377.E93 1990
[5] E DiBenedetto, Partial Differential Equations, Birkhauser, Boston 2010
QA377 D534 2010
[6] F.G. Friedlander, Introduction to the Theory of Distributions, CUP 1982,
QA324
[7] F. John, Partial Differential Equations, Springer-Verlag 1982, QA1.A647
[8] Rafael Jos Iorio and Valria de Magalhes Iorio, Fourier analysis and partial
differential equations CUP 2001, QA403.5 .I57 2001
[9] M.E. Taylor, Partial Differential Equations,Vols I-IIISpringer 96, QA1.A647
1 Introduction
1.1 Notation

We write partial derivatives as ∂t = ∂t , ∂j = ∂x∂ j etc and also use suffix on a function
to indicate partial differentiation: ut = ∂t u etc. A general k th order linear partial
differential operator (pdo) acting on functions u = u(x1 , . . . xn ) is written:
X
P = aα ∂ α u . (1.1)
|α≤k

Here α = (α1 , . . . αn ) ∈ Zn+ is a multi-index of order |α| = αj and


P

Y α Y α
∂α = ∂j j , xα = xj j . (1.2)
Q
For a multi-index we define the factorial α! = αj !. For (real or complex) con-
stants aα the formula (1.1) defines a constant coefficient linear pdo of order k. (Of
course assume always that at least one of the aα with |α| = k is non-zero so that it
is genuinely of order k.) If the coefficients depend on x it is a variable coefficient
linear pdo. The word linear means that
P (c1 u1 + c2 u2 ) = c1 P u1 + c2 P u2 (1.3)
holds for P applied to C k functions u1 , u2 and arbitrary constants c1 , c2 .

1.2 Basic definitions


If the coefficients depend on the partial derivatives of a function of order strictly
less than k the operator
X
u 7→ P u = aα (x, {∂ β u}|β|<k ) ∂ α u (1.4)
|α≤k

is called quasi-linear and (1.3) no longer holds. The corresponding equation P u = f


for f = f (x) is a quasi-linear partial differential equation (pde). In such equations
the partial derivatives of highest order - which are often most important - occur
linearly. If the coefficients of the partial derivatives of highest order in a quasi-
linear operator P depend only on x (not on u or its derivatives) the equation is
called semi-linear. If the partial derivatives of highest order appear nonlinearly the
equation is called fully nonlinear; such a general pde of order k may be written
F (x, {∂ α u}|α|≤k ) = 0 . (1.5)

Definition 1.2.1 A classical solution of the pde (1.5) on an open set Ω ⊂ Rn is a


function u ∈ C k (Ω) which is such that F (x, {∂ α u(x)}|α|≤k ) = 0 for all x ∈ Ω .
Classical solutions do not always exist and we will define generalized solutions
later in the course. The most general existence theorem for classical solutions is the
Cauchy-Kovalevskaya theorem, to state which we need the following definitions:

2
Definition 1.2.2 Given an operator (1.1) we define
• Pprincipal = |α=k aα ∂ α u , (principal part)
P

• p = |α≤k aα (iξ)α , ξ ∈ Rn , (total symbol)


P

• σ = |α=k aα (iξ)α , ξ ∈ Rn , (principal symbol)


P

• Charx (P ) = {ξ ∈ Rn : σ(x, ξ) = 0} , (the set of characteristic vectors at x)


• Char(P ) = {(x, ξ) : σ(x, ξ) = 0} = ∪x Charx (P ) , (characteristic variety) .
Clearly σ, p depend on (x, ξ) ∈ R2n for variable coefficient linear operators, but are
independent of x in the constant coefficient case. For quasi-linear operators we
make these definitions by substituting in u(x) into the coefficients, so that p, σ and
(also the definition of characteristic vector) depend on this u(x).

Definition 1.2.3 The operator (1.1) is elliptic at x (resp. everywhere) if the


principal symbol is non-zero for non-zero ξ at x (resp. everywhere). (Again the
definition of ellipticity in the quasi-linear case depends upon the function u(x) in
the coefficients.)

1.3 The Cauchy-Kovalevskaya theorem


The Cauchy problemis the problem of showing that for a given pde and given
data on a hypersurface S ⊂ Rn there is a unique solution of the pde which agrees
with the data on S. This is a generalization of the initial value problem for
ordinary differential equations, and by analogy the appropriate data to be given
on S consists of u and its normal derivatives up to order k − 1. A crucial condition
is the following:
Definition 1.3.1 A hypersurface S is non-characteristic at a point x if its normal
vector n(x) is non-characteristic, i.e. σ(x, n(x)) 6= 0. We say that S is non-
characteristic if it is non-characteristic for all x ∈ S.)
Again for quasi-linear operators it is necessary to substitute u(x) to make sense of
this definition, so that whether or not a hypersurface is non-characteristic depends
on u(x), which amounts to saying it depends on the data which are given on S.

Theorem 1.3.2 (Cauchy-Kovalevskaya theorem) In the real analytic case there


is a local solution to the Cauchy problem for a quasi-linear pde in a neighbourhood
of a point as long as the hypersurface is non-characteristic at that point.
This becomes clearer with a suitable choice of coordinates which emphasizes
the analogy with ordinary differential equations: let the hypersurface be the level
set xn = t = 0 and let x = (x1 , . . . xn−1 ) be the remaining n − 1 coordinates. Then
a quasi-linear P takes the form
X
P u = a0k ∂tk + ajα ∂tj ∂ α u (1.6)
|α|+j≤k,j<k

3
with the coefficients depending on derivatives of order < k, as well as on (x, t).
Since the normal vector to t = 0 is n = (0, 0, . . . 0, 1) ∈ Rn the non-characteristic
condition is just a0k 6= 0, and ensures that the quasi-linear equation P u = f can
be solved for ∂tk u in terms of {∂tj ∂ α u}|α|+j≤k,j<k to yield an equation of the form:

∂tk u = G(x, t, {∂tj ∂ α u}|α|+j≤k,j<k ) (1.7)


to be solved with data
u(x, 0) = φ0 (x), ∂t u(x, 0) = φ1 (x) . . . ∂tk−1 u(x, 0) = φk−1 (x) . (1.8)
Theorem 1.3.3 Assume that G, φ0 , . . . φk−1 are all real analytic functions in some
neighbourhood of the origin. Then there exists a unique real analytic function which
satisfies (1.8)-(1.7) in some neighbourhood of the origin.
Notice that the non-characteristic condition ensures that the k th normal derivative
∂tk u(x, 0) is determined by the data. Differentiation of (1.7) gives further relations
which can be shown to determine all derivatives of the solution at t = 0, and
the theorem can be proved by showing that the resulting Taylor series defines a
real-analytic solution of the equation. Read section 1C of the book of Folland for
the full proof.
In the case of first order equations with real coefficients the method of character-
istics gives an alternative method of attack which does not require real analyticity.
In this case we consider a pde of the form
n
X
aj (x, u)∂j u = b(x, u) (1.9)
j=1

with data
u(x) = φ(x) , x∈S (1.10)
where S ⊂ Rn is a hypersurface, given in paramteric form as xj = gj (s) , s =
(s1 , . . . sn−1 ) ∈ Rn−1 . (Think of S = {xn = 0} parametrized by g(s1 , . . . sn−1 ) =
(s1 , . . . sn−1 , 0).)
Theorem 1.3.4 Let S be a C 1 hypersurface properly paramterized by a C 1 func-
tion g, and assume that the aj , b, φ are all C 1 functions. Then there exists a unique
C 1 solution of (1.9)-(1.10) defined on a neighbourhood of S as long as the non-
characteristic condition Σnj=1 aj (x, φ(x))nj (x) 6= 0 holds.
This is proved by considering the characteristic curves which are obtained by
integrating the system of n + 1 characteristic ordinary differential equations (ode):
dxj dz
= aj (x, z) , = b(x, z) (1.11)
dt dt
with data xj (s, 0) = gj (s), z(s, 0) = φ(g(s)); let (X(s, t), Z(s, t)) ∈ Rn × R be
this solution. The non-characteristic condition implies (via the inverse function
theorem) that the “restricted flow map” which takes (s, t) 7→ X(s, t) = x is locally
invertible, with inverse sj = σj (x), t = τ (x) and this allows one to recover the
solution as u(x) = Z(σ(x), τ (x)).

4
2 Example sheet 1
1. Write out the multinomial expansion for (x1 +. . . xn )N and the n-dimensional
Taylor expansion using multi-index notation.
2. Consider the problem of solving the heat equation ut = ∆u with data
u(x, 0) = f (x). Is the non-characteristic condition satisfied? How about
for the wave equation utt = ∆u with data u(x, 0) = f (x) and ut (x, 0) =
g(x)? For which of these problems, and for which data, does the Cauchy-
Kovalevskaya theorem ensure the existence of a local solution? How about
the Cauchy problem for the Schrödinger equation?
3. Find the characteristic vectorsPfor the operator P = ∂1 ∂2 (n = 2). Is it
elliptic? Do the same for P = m 2
Pn 2

j=1 j − j=m+1 ∂j (1 < m < n).
4. Solve the linear PDE x1 ux2 − x2 ux1 = u with boundary condition u(x1 , 0) =
f (x1 ) for f a C 1 function. Where is your solution valid? Classify the f
for which a global C 1 solution exists. (Global solution here means a solution
which is C 1 on all of R2 .Solve the linear PDE x1 ux2 −x2 ux1 = u with boundary
condition u(x1 , 0) = f (x1 ) for f a C 1 function. Where is your solution valid?
Classify the f for which a global C 1 solution exists. (Global solution here
means a solution which is C 1 on all of R2 .
5. Solve Cauchy problem for the semi-linear PDE ux1 + ux2 = u4 , u(x1 , 0) =
f (x1 ) for f a C 1 function. Where is your solution C 1 ?
6. For the quasi-linear Cauchy problem ux2 = x1 uux1 , u(x1 , 0) = x1
(a) Verify that the Cauchy-Kovalevskaya theorem implies existence of an
analytic solution in a neighbourhood of all points of the initial hypersurface
x2 = 0 in R2 ,
(b) Solve the characteristic ode and discuss invertibility of the restricted flow
map X(s, t) (this may not be possible explicitly),
(c) give the solution to the Cauchy problem (implicitly).
7. For the quasi-linear Cauchy problem ux2 = A(ux1 +1)/(B−x1 −u), u(x1 , 0) =
0:
(a) Find all points on the intial hypersurface where the Cauchy-Kovalevskaya
theorem can be applied to obtain a local solution defined in a neighbourhood
of the point.
(b) Solve the characteristic ode and invert (where possible) the restricted
flow map, relating your answer to (a).
(c) Give the solution to the Cauchy problem, paying attention to any sign
ambiguities that arise.
(In this problem take A, B to be positive real numbers).
8. For the Cauchy problem
ux1 + 4ux2 = αu u(x1 , 0) = f (x1 ), (2.12)
with C 1 initial data f , obtain the solution u(x1 , x2 ) = eαx2 /4 f (x1 − x2 /4)
by the method of characteristics. For fixed x2 write u(x2 ) for the function
x1 7→ u(x1 , x2 ) i.e. the solution restricted to “time” x2 . Derive the following
well-posedness properties for solutions u(x1 , x2 ) and v(x1 , x2 ) corresponding
to data u(x1 , 0) and v(x1 , 0) respectively:
(a) for α = 0 there is global well-posedness in the supremum (or L∞ ) norm
uniformly in time in the sense that if for fixed x2 the distance between u and

5
v is taken to be
ku(x2 ) − v(x2 )kL∞ ≡ sup |u(x1 , x2 ) − v(x1 , x2 )|
x1

then
ku(x2 ) − v(x2 )kL∞ ≤ |u(0) − v(0)|L∞ for all x2 .
Is the inequality ever strict?
(b) for all α there is well-posedness in supremum norm on any finite time
interval in the sense that for any time interval |x2 | ≤ T there exists a number
c = c(T ) such that

ku(x2 ) − v(x2 )kL∞ ≤ c(T )ku(0) − v(0)kL∞ .

and find c(T ). Also, for different α, when can c be assumed independent of
time for positive (respectively negative) times x2 ?
(c) Try to do the same for the L2 norm, i.e. the norm defined by
Z
ku(x2 ) − v(x2 )kL2 (dx1 ) = |u(x1 , x2 ) − v(x1 , x2 )|2 dx1 .
2

9. For which real numbers a can you solve the Cauchy problem

u x1 + u x2 = 0 u(x1 , ax1 ) = f (x1 )

for any C 1 function f . Explain both in terms of the non-characteristic condi-


tion and by explicitly trying to invert the (restricted) flow map, interpreting
your answer in relation to the line x2 = ax1 on which the initial data are
given.
10. (a). Consider the equation

ux1 + nux2 = f

where n is an integer and f is a smooth function which is 2π- periodic in


both variables:
f (x1 + 2π, x2 ) = f (x1 , x2 + 2π) = f (x1 , x2 ).

Apply the method of characteristics to find out for which f there is a solution
which is also 2π- periodic in both variables:

u(x1 + 2π, x2 ) = u(x1 , x2 + 2π) = u(x1 , x2 ).

(b) Consider the problem in part (a) using fourier series representations of f
and u (both 2π- periodic in both variables) and compare your results.

You might also like