Differential Equations

Assignment #1: answers.

Note. A mathematical text is written in human language; every formula should be part of a
proper sentence. Needless (?) to say, a sentence begins with a capital letter and ends with a
punctuation symbol. Otherwise the text is not meant to be read.
This elementary principle is often disregarded; however writing lines and lines of formulas

ns
is not acceptable. The idea is that your assignments should show your reasoning, not your
computation; that someone with less knowledge than you should be able to see the key ideas,
and the names of the methods.
By reading the solutions below, you will be impressed by the little amount of symbols. But
high-level mathematics proceeds in sentences, not in formulas. Do not focus on the computa-
tions, focus on the logic.
tio
Exercise 1. Consider the following equation:
1
x0 (t) = 2tx(t) + exp(t2 )
t
defined on R>0 , with initial condition x(1) = 0.
1. Write the so-called homogeneous equation (without any initial condition). Give the general
solution (of the homogeneous equation).
lu
2. Give one special solution of the original equation. Then give the general solution (of the
original equation).
3. Now give the only solution satisfying the initial condition.
Solution.
so

1. The associated homogeneous equation is:

x0 (t) = 2tx(t)
2
which as we know solves into x(t) = exp( 2tdt) = λet for λ ∈ R.
R

Notes.
2
(a) Many of you wrote “λ(t) · et ”. This is simply wrong — λ must be a constant.
(b) Some of you have been using the so-called “separable variables method”, as follows:
0 2
Write xx = 2t, which integrates as log(x) = et + C. . .
This is mathematically not rigorous for two reasons, and naive for one.
i. Division by x: but x could vanish. Indeed, the map x(t) = 0 is a solution.
ii. Use of the logarithm: there is no well-defined log function on R.
2
The slightly more correct log |x| = et does not solve this issue: again, what
about x = 0?
iii. How do you expect to generalise this division to the vector case?
In short, nothing beats quick use of the exponential. For desperate (non-linear)
scalar situations, separating variables gives some intuition; its rigorous treatment
is however much heavier than anything we discussed so far. Refrain from using
something you cannot justify!

1
 2. We use the method called “variation of the parameter”, and look for one special solution
2
of the original equation in the form x(t) = λ(t) · et .
For this to be a solution, we need to have:
2 1 2
λ0 (t) · et = et
t
and therefore λ0 (t) = 1t .
2
This suggests to let x(t) = log(t) · et , which is well-defined on I = R>0 . A quick compu-
tation shows that this is a solution of the original equation. Hence the general solution
has the form:
2
(λ + log(t)) · et

ns
for some λ ∈ R.

Notes.

(a) Do not forget to check that the x you define is a solution indeed. (A necessary
condition is not always sufficient, so getting the condition on λ is a priori no serious
guarantee.)
tio
(b) It is mathematically and grammatically incorrect to write “the one special solu-
tion”.

3. The linear case of the Cauchy-Lipschitz theorem predicts that there will be exactly one
maximal solution to the equation with initial condition, and that it will be defined on R>0 .
2
For an arbitrary solution x(t) = (λ + log(t)) · et to satisfy x(1) = 0, one needs λ = 0.
2
One quickly checks that x(t) = log(t) · et is a solution of the equation, and satisfies the
lu
initial condition; it is even unique as such by the Cauchy-Lipschitz theorem.
Exercise 2. We consider the following differential equation:

(EH ) : x(4) = x(3) + 7x(2) − 13x0 + 6x.

1. Provide a basis of the space solution SH .

so

2. Give the explicit solution with the initial conditions

x(0) = 2; x0 (0) = 0; x00 (0) = −2 and x(3) (0) = −4

Solution.
1. We consider the characteristic polynomial associated to the problem:

P (λ) = λ4 − λ3 − 7λ2 + 13λ − 6.

The value 1 is an obvious root; dividing (there is no need for Euclidean division here, you
can do it mentally) one finds P (λ) = (λ − 1)(λ3 − 7λ + 6). Here again 1 is a root of the
second factor, yielding P (λ) = (λ − 1)2 (λ2 − λ − 6). The latter term has discriminant 25,
and solutions 1±5
2 , namely −2 and 3.
Hence the characteristic polynomials has roots (counted with multiplicity): −2, 1, 1, 3. So
the functions et , tet , e−2t , e3t are solutions of the differential equation (EH ).
Because it is linear homogeneous of order 4, the Cauchy-Lipschitz theorem predicts that
the space SH of solutions has dimension 4; since the solutions we gave are clearly linearly
independent, they form a basis of SH .
As a conclusion, a basis of SH is {et , tet , e−2t , e3t }.

2
 Notes.

(a) One should not need to write down a polynomial division here; the factorisation
was too simple.
(b) Never introduce the discriminant ∆ without saying the word “discriminant” (this
is a typical example of unclear mathematical exposition; as a rule, never use an
unexplained notation, however universal you think it is).
(c) A vector space almost never has a unique basis. It is therefore mathematically
incorrect to write “the basis is”.

2. We now take the initial condition into account. The general solution has the form:

ns
x(t) = aet + btet + ce−2t + de3t .

We can then compute and evaluate the successive derivatives:

x(0) = a + c + d;
x0 (0) = a + b − 2c + 3d;
tio x00 (0) = a + 2b + 4c + 9d;
x000 (0) = a + 3b − 8c + 27d.

Hence the initial condition gives rise to the following system in matrix form, where we let
Λ be the vector of coordinates in the given basis:
   
1 0 1 1 2
1 1 −2 3 
 0 
 
 · Λ =  .

lu
1 2 4 9 −2


1 3 −8 27 −4
This system is quickly handled by introducing an augmented matrix and using Gauß
elimination, as follows:
   
1 0 1 1 2 1 0 1 1 2
so

 1 1 −2 3 0   0 1 −3 2 −2 
 −−−−−−−→
   
1 2 4 9 −2 0 2 3 8 −4 
  
  L2 ←L2 −L1 
L3 ←L3 −L1
1 3 −8 27 −4 L4 ←L4 −L1 0 3 −9 26 −6
 
1 0 1 1 2
 0 1 −3 2 −2 
−−−−−−−−→  .
 
L3 ←L3 −2L2  0 0 9 4 0 
L4 ←L4 −3L2
0 0 0 20 0
At this stage one clearly has c = d = 0; finally a = 2 and b = −2. So the only solution
with initial condition is x(t) = 2et − 2tet = 2(1 − t)et .

Note. More than ever in this question, what matters is not the mathematical formulas,
but the sentences linking them.

Exercise 3. Consider the following matrix:

!
0 −1
A=
1 0

Prove that: !
cos(1) − sin(1)
exp(A) =
sin(1) cos(1)

3
Solution. We treat A as a complex-valued matrix. Its characteristic polynomial is λ2 +1, which
has roots ±i. The 2 × 2 matrix A has two distinct eigenvalues, so it is diagonalisable.
Let us compute the eigenspaces:
! !
−i −1  i 
ker(A − iI) = ker = ker 1 −i = C
1 −i 1
and likewise, ! !
i −1  1 
ker(A + iI) = ker = ker 1 i = C .
1 i i
! !!
i 1
Hence , is an eigenbasis, which suggests to indtroduce the following coordinate
1 i

ns
change matrix: !
i 1
P = .
1 i
Its inverse is easily computed: !
1 i −1
P −1 =− ;
2 −1 i
tio
by construction (no computation is required here),

−1 i
!
P AP = .
−i

Now exp(P AP −1 ) is easily computed, which yields:

! ! ! !
−i 1
ei −1 i 1 ei 2 2
exp A = P P = · · −i
e−i 1 i e−i 1
lu
2 2
! ! !
−i
iei e−i 2
1
2
1 ei + e−i i(ei − e−i )
= · −i =
ei ie−i 1
2 2 2 −i(e − e ) ei + e−i
i −i
!
cos(1) − sin(1)
= .
sin(1) cos(1)
so

Notes.
√ √
1. The function · is defined on R≥0 . Writing −1 is simply meaningless.

2. An eigenvector, the eigenspace. It is not correct to write “the eigenvector” (since any
multiple will be another such).
By the way, the notation Eλ (M ) refers to the eigenspace; it is therefore incorrect to
write Eλ (M ) = v, one should write Eλ (v) = Cv.

3. When you first introduce P , you must write “Let P =”. Otherwise the reader may
believe (s)he has missed an earlier definition.

4. Last but not least, a technical remark. Because the matrix here was not too hard, it
was possible to “brute-force” the exponential by actually computing the series. Now be
honest and ask yourself:

(a) if you would have found the correct answer had it not been given to you;
(b) what the odds of success of this method might be in dimension 4.

The general method proceeds by conjugating the matrix to one with simpler exponential;
in practice, and although it is the definition, computing the series never works.