Ordinary differential equations

Pierre Gosselet — pierre.gosselet@univ-lille.fr

LaMcube, Univ-Lille/CNRS/Centrale Lille

Master Strains – 2021-2022

Generalities

First order
Separated and separable variables
Homogeneous equations
Exact differential
Linear equations
Special nonlinear cases: Bernoulli, Riccati, Clairaut

Second order
Linear equations
Sturm-Liouville theory

More Exercices

Systems of ODEs

Note: this document is partly inspired from the course “outils mathématiques pour
la physique” by Guy Royer of faculté des sciences et techniques de Nantes.
 Ordinary differential equations

An (implicit) ODE of order n is an equation of the form

F (x , y , y ′ , . . . , y (n) ) = 0

The equation is . . .
▶ explicit if F (x , y , y ′ , . . . , y (n−1) ) = y (n)
▶ autonomous if F does not depend on x
Pn−1
▶ linear if y (n) = ai (x )y (i) + r (x ). r is the source term
i=0
▶ Homogeneous if r = 0, the solution is a complementary function yc (can be 0).
▶ Nonhomogeneous (inhomogeneous) if r ̸= 0, the solution is a particular integral yp
plus a complementary function y = yc + yp .
It can be complemented with n boundary or initial conditions.
 Very general theorems

Peano existence theorem

Let D ⊂ R × R be an open domain, f : D → R continuous, we consider the explicit
first-order differential equation with initial value (x0 , y0 ) ∈ D:

y ′ (x ) = f (x , y (x )) with y (x0 ) = y0

there exists a solution defined in a neighborhood of x0 (no uniqueness).

The Carathéodory’s existence theorem enables to relax the continuity condition.

Picard-Lindelöf (Cauchy-Lipschitz) existence and unicity theorem

with the same notations as above, if f is uniformly Lipschitz continuous in y and
continuous in x , then for some ε > 0, there exists a unique solution on the interval
[x0 − ε, x0 + ε].
Extension : Global uniqueness and maximum domain of definition. For each initial
condition, there exists a unique maximum (possibly infinite) open interval
Imax = (x− , x+ ) ∋ x0 , where any solution is a restriction of the solution on Imax .

Note that if x± ̸= ±∞ then either there is an “explosion in finite time”

(lim supx →x± |y (x )| = ∞) or an exit from the domain of definition
(lim |y (x )| →x →x± ∈ ∂D).
 Examples
Example
Verify that functions of the form:

a(t − α) 3


 , t⩽α

 3
y (t) = t = 0, α⩽t⩽β
  3
 a(t − β) ,

 t⩾β
3

where −∞ ⩽ α ⩽ 0 ⩽ β ⩽ +∞, are all solutions to y ′ = ay 2/3 with y (0) = 0.

Explain the non-uniqueness by studying f (y ) = y 2/3 .

Example
Prove that y ′ = y 2 satisfies the hypothesis of Picard-Lindelöf theorem.
Verify that the solution is of the form y (x ) = (x −xy0)y +1
0 0
Verify that the maximum domain of existence depends on y0 :
if y0 = 0,
R


 1

−∞, x0 + if y0 > 0,
Imax = y0
  

 x0 + 1
, +∞ if y0 < 0.
y0
and that we observe an explosion in finite time.
 Grönwall’s inequality (lemma)

Let, a<b in R, I = [a, b] or [a, b)] with b potentially +∞.

Let β and u be continuous function defined in I.
◦
If u is differentiable in I and satisfies:
◦
u ′ (t) ⩽ β(t)u(t), t ∈ I,

then Z t 
u(t) ⩽ u(a) exp β(s)ds , ∀t ∈ I.
a

Exercise
Prove the lemma. Rt

Hint: introduce v (t) = exp β(s)ds , see which ODE is satisfied by v and study
a
the monotony of u/v .
 First order equations

Separated variables

f (x ) + g(y )y ′ = 0
R R
Use direct integration1 : ( f )(x ) + ( g)(y ) = c ∈ R.

Exercise: solve 3x 2 + x sin(x ) + y 4 y ′ = 0.

Separable variables

f1 (x )g1 (y ) + f2 (x )g2 (y )y ′ = 0
Study the cases g1 (y ) = 0 and eliminate the domain where f2 (x ) = 0. Then
separate the variables.
Exercise: solve 2x (1 − y 2 ) − y ′ = 0.

1
R
f is the notation for a primitive of f .
 First order equations

Homogeneous equations

y
y′ = f ( )
x
Use change of variable z = y /x so that y ′ = z ′ x + z, obtain a separable equation
in (z, x ).
2x 2 +xy −2y 2
Exercise: solve y ′ = x2
.

Exact differential

∂f ∂g
f (x , y ) + g(x , y )y ′ = 0 with =
∂y ∂x
then dF = f (x , y )dx + g(x , y )dy is an exact differential which can be obtained by
integration:
∂F ∂F
=f and = g.
∂x ∂y

Exercise: Solve (4x 3 y − 20x 3 ) + (x 4 + 3)y ′ = 0.

 Linear first order equations

y ′ + a(x )y = b(x )

Homogeneous case
If b = 0, then it is an equation with separated variables:
 Z 
yc (x ) = K exp − a

General solution (superposition principle)

The general solution y is the sum of a particular solution yp and a complement
function yc solution to the homogeneous equation.

Particular solution when a is a constant

▶ If b is a polynomial of degree n, search yp as a polynomial of degree n.
▶ If b(x ) = (c1 cos αx + c2 sin αx ) e βx , search yp = (λ1 cos αx + λ2 sin αx ) e βx .
▶ If b(x ) = pn (x )e βx with pn a polynomial of degree n,
▶ If β = −a, search yp = xqn (x )e βx with qn a polynomial of degree n,
▶ If β ̸= −a, search yp = qn (x )e βx with qn a polynomial of degree n.
 Linear first order equations

Solution by Lagrange’s method aka variation of the constant

Search the solution under the form
 Z 
y (x ) = K (x ) exp − a

y
Exercise: solve y ′ − x
= 1.
 Special nonlinear cases
Bernoulli’s equations

y ′ + a(x )y = b(x )y n
Set z = y 1−n and reduce the problem to a linear equation.
3
y
Exercise: Solve y ′ + 2x
= − xy2 .

Riccati’s equations

y ′ = a(x )y 2 + b(x )y n + c(x )

1
If y1 is a known particular equation, set y = y1 + z
and z is the solution to a linear
equation.

Clairaut’s equations

y = xy ′ + f (y ′ ), with f continuously derivable.

Search two-times differentiable y by deriving the equation. You obtain two family
of solutions: the general solutions yg′′ = 0 and the singular solution x + f ′ (ys′ ) = 0
which forms the envelope of the family of the general solutions.
Exercise: consider the case f (z) = z 2 . Once ys found, seek extra solutions of the
form: y (x ) = m(x ) + ys (x ).
 Linear second order equations

y ′′ + a(x )y ′ + b(x )y = c(x )

Generality
▶ Superposition: solution to homogeneous equation + particular solution,
▶ If y1 and y2 are independent solutions to the homogeneous equation, the
general solution is y = λ1 y1 + λ2 y2 , λi ∈ R.
▶ If y1 is one solution to the homogeneous equation, one can set y2 = y1 z and
z ′ is the solution to a first order equation.
▶ To find the particular solution we can use the variation of the constants :
y = λ1 (x )y1 + λ2 (x )y2 .
▶ The examination of the RHS (c) can suggest a particular solution.
′
Exercise: search solutions to y ′′ + 4 yx + 2 xy2 = 0 under the form y = x m .
 Linear second order equations with constant coefficients

y ′′ + ay ′ + by = c(x )

Homogeneous equation c = 0
Searching for solutions as e rx leads to the characteristic equation r 2 + ar + b = 0
▶ If a2 − 4b > 0, two real roots and y = Ae r1 x + Be r2 x .
▶ If a2 − 4b = 0, one real root and y = (Ax + B)e r1 x .
▶ If a2 − 4b < 0, two conjugate complex roots r = α ± iβ and
y = e α x (A cos(βx ) + B sin(βx )).
√ √
Remark: in the case a = 0 and b < 0, y = A cosh( −bx ) + B sinh( −bx ).

Exercises: solve y ′′ − 2y ′ − 15y = 0, y ′′ − 6y ′ + 9y = 0, y ′′ + 4y ′ + 7y = 0,

y ′′ − 4y = 0, y ′′ + 9y = 0.

Particular solution
▶ If c(x ) = pn (x )e µx , then yp = x d qn (x )e µx where d = 0 if µ is not a root of
the characteristic equation, else d is the multiplicity of the root µ.
▶ If c(x ) = (α cos(ωx ) + β sin(ωx ))e γx , then y = x d (λ cos(ωx ) + µ sin(ωx ))e γx ,
where d = 0 if γ + iω is not a root of the characteristic equation, else d = 1.
 Special nonlinear second order equation

Equation of the form F (x , y ′ , y ′′ ) = 0 (no y )

Set z = y ′ , it is the solution to a first order problem.
Exercise: solve xy ′′ − y ′ = 5x .
 Sturm-Liouville theory

Sturm-Liouville operator and eigenvalue problem

I interval of R, p ∈ C 1 (I), q ∈ C 0 (I), s ∈ C 0 (I) with s > 0.

1 d dX
   
SL(X ) = p + qX
s dx dx

Find X (let’s say C 2 (I)) and λ ∈ R such that:

SL(X ) + λX = 0

Let L2s (I) be the Hilbert space of s-weighted square-integrable functions with inner
product: Z
(f , g)L2 = fgs dx
s
I
Exercise: compute (SL(f ), g)L2 − (f , SL(g))L2 , show that only boundary terms
s s
remain.
In the following we make hypothesis on the boundary conditions to get rid of those
terms, so that SL is self-adjoint.
 Regular Sturm-Liouville problem
Let I = [a, b], we assume p > 0 and use zero Dirichlet or Neumann or Robin bcs:
a0 X (a) + a1 X ′ (a) = 0
 
|a0 | + |a1 | ̸= 0
with (a0 , a1 , b0 , b1 ) ∈ R4
b0 X (b) + b1 X ′ (b) = 0 |b0 | + |b1 | ̸= 0

Theorem
i. There exists a countable unbounded family of real simple eigenvalues :

|λ0 | < |λ1 | < . . . < |λn | with lim |λn | = +∞

n→+∞

ii. For any n ∈ N, the eigenfonction Xn associated with λn satisfies :

d dXn
 
p + (q + sλn )Xn = 0
dx dx
iii. For any n ∈ N, the eigenfonction Xn as eaxctly n zeros in ]a, b[; more, the
zeros of Xn−1 are inserted between the zeros of Xn .
!
X
iv. If f ∈ L2s ([a, b]), then the series of functions (f , Xn )L2 ([a,b]) Xn
s
n
converge to f in L2s ([a, b]) (quadratic convergence).
If f is piecewise C 1 , the convergence is simple to the regularized function:
f (x + )+f (x − )
x 7→ 2
If more, f is continuous, the convergence is uniform in [a, b].
 Periodic Sturm-Liouville problem

Let I = [a, b], we assume p > 0 and periodic bcs:


p(a) = p(b) 
X (a) = X (b)
q(a) = q(b) and
 X ′ (a) = X ′ (b)
s(a) = s(b)

Theorem
Almost the same result as the regular case except that
λ0 is a simple eigenvalue, the others can be multiple eigenvalues.

Application: Fourier series for periodic signal analysis.

Exercise: solve on [0, 1] :  ′′

 f + λf = 0
f (0) = f (1)
f ′ (0) = f ′ (1)

 Singular Sturm-Liouville

The theory remains valid for many cases, and we can recover many famous
functions:
▶ Legendre polynomials
▶ Hermite polynomials
▶ Laguerre polynomials
▶ Tchebychev polynomials
▶ Bessel functions
all these functions form Hilbert basis of well-chosen spaces.
 Exercices

Solve:
▶ y ′ − 16y = 0
▶ xy ′ − 2y = x 5
▶ y ′ + y = e 2x + e x + 3 sin(x )
▶ x 2y ′ − ey = 0
▶ (x 2 + 1)y ′ + 3xy = x
▶ x 3 y ′ + (2 − 3x 2 )y = x 3
▶ y ′ − (2x + 3x 2 )y − (2x + 3x 2 ) = 0
▶ x (y − 3)y ′ − 4y = 0
▶ y ′′ − 81y = 0
▶ y ′′ + 256y = 0
▶ y ′′ + 4y = 4x 3
▶ y ′′ − 5y ′ + 4y = e x
▶ y ′′ − 2y ′ + y = (x 2 + 1)e x
▶ y ′′ + 2y ′ + 2y = e −x cos(x )
 Systems of ODEs

Definition and general properties

y(x ) ∈ Rm , F takes values in Rm : F(x , y, y′ , . . . , y(n) )

∂F(x ,u0 ,...,un )
If at least locally the Jacobian matrix ∂un
is non-singular, there exists an
explicit form y(n)
=e F(x , y, y′ , . . . , y(n−1) ).
Picard’s existence and uniqueness theorem still applies.

Transforming high order ODE in first order system of ODEs

We consider y and f with values in R. The system f (x , y , y ′ , . . . , y (n) ) is
equivalent to:
′
0 = f (x , y , y1 , . . . , yn−1 , yn−1 )
y1 = y ′
..
.
′
yn−1 = yn−2

The evolution of such system is observed in the phase space.

 Linear systems of first order ODEs

y′ (t) = A(t)y(t) + b(t)


In Rn ,
y(t0 ) = y0

Constant coefficients A
P∞ Ai
We introduce the matrix exponential: e A = i=0 i!
.
Exercise: use the change of unknown function z(t) = e tA y(t), what is the equation
verified by z, give y in case of an homogeneous equation (b(t) = 0).
Exercise: why this method doesn’t work with non-constant matrix A ? Introduce a
primitive of A(t) and see where things stop working.

Resolvent matrix
Let R2 ∋ (t1 , t2 ) 7→ R(t1 , t2 ) ∈ Rn×n be the solution to:

∂
∀(t1 , t2 ), R(t1 , t1 ) = I and R(t1 , t2 ) = A(t1 )R(t1 , t2 )
∂t1

Then y(t) = R(t, t0 )y0 solves the homogeneous equation.

Properties: R(t1 , t2 )R(t2 , t3 ) = R(t1 , t3 ) and R(t1 , t2 )−1 = R(t2 , t1 ).
Exercise: show that the solution to the non-homogeneous equation can be
obtained by the variation of the constant technique.
 Linear system of first order homogeneous ODEs

In Rn , y′ (t) = A(t)y(t)

Solution space and fundamental system of solutions

The solution space S is a vector space of dimension n,
If Y(t) = (y1 , . . . , yn ) is a basis of S, it is called a fundamental system of solutions.

Wronskian
The Wronskian is the determinant of the system:

W (t) = det(y1 , . . . , yn )

If we expand det(Y′ ) = det(A) det(Y), we obtain the Wronskian’s ODE:

W ′ (t) = tr(A)W (t)

Z t 
(Liouville’s formula) W (t) = W (t0 ) exp tr(A(s)) ds
t0

Note that since Y(t) = R(t, t0 )Y0 , we have:

Z t 
det(R(t, t0 )) = exp tr(A(s)) ds
t0
 Exercises

▶ Solve:
y1′ = a(y2 − y1 )


y2′ = a(y1 − y2 )
with given initial conditions.
▶ Solve the system corresponding to two masses connected by a spring:

m1 y1′′ = k(y2 − y1 )


m2 y2′ = k(y1 − y2 )

with given initial conditions.

▶ Let (a, b) ∈ R2 , let y be a real-valued function. We consider the following
second order homogeneous scalar equation:

y ′′ = ay ′ + by

Consider two independent solutions y 1 and y2 ,

y1 (t) y2 (t)
their Wronskian is det .
y1′ (t) y2′ (t)
Show the equation satisfied by the Wronskian and solve it.
Show that the above developments correspond to putting the equation under
the form of a first order system.
 References I

Grigorian, Alexander (2008). Ordinary Differential Equation. https://www.math.uni-bielefeld.de/ grigor/odelec2008.pdf.

Nagy, Gabriel (2021). Ordinary Differential Equations. https://users.math.msu.edu/users/gnagy/teaching/ode.pdf.