Dierential Equations (part 3): Systems of First-Order Dierential Equations (by Evan Dummit, 2016, v. 2.

00)

Contents
6 Systems of First-Order Linear Dierential Equations 1
6.1 General Theory of (First-Order) Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

6.2 Eigenvalue Method (Nondefective Coecient Matrices) . . . . . . . . . . . . . . . . . . . . . . . . . . 3

6 Systems of First-Order Linear Dierential Equations

In many (perhaps most) applications of dierential equations, we have not one but several quantities which change
over time and interact with one another. Examples include the populations of the various species in an ecosystem,
the concentrations of molecules involved in a chemical reaction, the motion of objects in a physical system, and the
availability and production of items (goods, labor, materials) in economic processes.

In this chapter, we will outline the basic theory of systems of dierential equations. As with the other dierential
equations we have studied, we cannot solve arbitrary systems in full generality: in fact it is very dicult even
to solve individual nonlinear dierential equations, let alone a system of nonlinear equations. We will therefore
restrict our attention to systems of linear dierential equations: as with our study of higher-order linear dierential
equations, there is an underlying vector-space structure to the solutions which we will explain. We will discuss how
to solve many examples of homogeneous systems having constant coecients.

6.1 General Theory of (First-Order) Linear Systems

• Before we start our discussion of systems of linear dierential equations, we rst observe that we can reduce
any system of linear dierential equations to a system of rst-order linear dierential equations (in more
variables): if we dene new variables equal to the higher-order derivatives of our old variables, then we can
rewrite the old system as a system of rst-order equations.

• Example: Convert the single 3rd-order equation y 000 + y 0 = 0 to a system of rst-order equations.

◦ If we dene new variables z = y0 and w = y 00 = z 0 , then the original equation tells us that y 000 = −y 0 , so
0 000 0
w =y = −y = −z .
◦ Thus, this single 3rd-order equation is equivalent to the rst-order system y 0 = z , z 0 = w, w0 = −z .
• Example: Convert the system y100 + y1 − y2 = 0 and y200 + y10 + y20 sin(x) = ex to a systen of rst-order equations.

◦ If we dene new variables z1 = y10 and z2 = y20 , then z10 = y100 = −y1 +y2 and z20 = y200 = ex −y10 −y20 sin(x) =
x
e − z1 − z2 sin(x).
◦ So this system is equivalent to the rst-order system y10 = z1 , y20 = z2 , z10 = −y1 + y2 , z20 = ex − z1 −
z2 sin(x).
• Thus, whatever we can show about solutions of systems of rst-order linear equations will carry over to
arbitrary systems of linear dierential equations. So we will talk only about systems of rst-order linear
dierential equations from now on.

• Denition: The standard form of a system of rst-order linear dierential equations with unknown functions
y1 , y2 , . . . , yn is

y10 = a1,1 (x) · y1 + a1,2 (x) · y2 + · · · + a1,n (x) · yn + q1 (x)

y20 = a2,1 (x) · y1 + a2,2 (x) · y2 + · · · + a2,n (x) · yn + q2 (x)
. .
. .
. .

yn0 = an,1 (x) · y1 + an,2 (x) · y2 + · · · + an,n (x) · yn + qn (x)

for some functions ai,j (x) and qi (x) for 1 ≤ i, j ≤ n.

1
  
a1,1 (x) ··· a1,n (x)
. .. .
◦ We can write this system more compactly using matrices: if A =  . . , q =
 
. . .
an,1 (x) · · · an,n (x)
y10 (x)
     
q1 (x) y1 (x)
. . .
. , and y= . so that y0 =  . , we can write the system more compactly as
     
 . .  .
0
qn (x) yn (x) yn (x)
y0 = Ay + q.
◦ We say that the system is homogeneous if q = 0, and it is nonhomogeneous otherwise.

◦ Most of the time we will be dealing with systems with constant coecients, in which the entries of A are
constant functions.

◦ An initial condition for this system consists of n pieces of information: y1 (x0 ) = b1 , y2 (x0 ) = b2 , . . . ,
yn (x0 ) = bn , where x0 is the starting value for x and the bi are constants. Equivalently, it is a condition
of the form y(x0 ) = b for some vector b.

• We also have a version of the Wronskian in this setting for checking whether function vectors are linearly
independent:
   
y1,1 (x) yn,1 (x)
. .
• Denition: Given n vectors v1 =  . , · · · , vn =  . of length n with functions as entries,
   
. . 
y1,n (x) yn,n (x)
y1,1 y1,2 · · · y1,n
y2,1 y2,2 · · · y2,n
their Wronskian is dened as the determinant W = . . .. . .
. . . .
. . .
yn,1 yn,2 ··· yn,n

◦ By our results on row operations and determinants, we immediately see that n function vectors v1 , . . . , vn
of length n are linearly independent if their Wronskian is not the zero function.

• Many of the theorems about general systems of rst-order linear equations are very similar to the theorems
about nth order linear equations.

• Theorem (Existence-Uniqueness): For a system of rst-order linear dierential equations, if the coecient
functions ai,j (x) and nonhomogeneous terms pj (x) are each continuous in an interval around x0 for all 1≤
i, j ≤ n, then the system

y10 = a1,1 (x) · y1 + a1,2 (x) · y2 + · · · + a1,n (x) · yn + p1 (x)

y20 = a2,1 (x) · y1 + a2,2 (x) · y2 + · · · + a2,n (x) · yn + p2 (x)
. .
. .
. .

yn0 = an,1 (x) · y1 + an,2 (x) · y2 + · · · + an,n (x) · yn + pn (x)

with initial conditions y1 (x0 ) = b1 , . . . , yn (x0 ) = bn has a unique solution (y1 , y2 , · · · , yn ) on that interval.

◦ This theorem is not trivial to prove and we will omit the proof.

◦ Example: The system y 0 = ex · y + sin(x) · z , z 0 = 3x2 · y has a unique solution for every initial condition

y(x0 ) = b1 , z(x0 ) = b2 .

• Proposition: Suppose ypar is one solution to the matrix system y0 = Ay + q. Then the general solution ygen
to this equation may be written as ygen = ypar + yhom , where yhom is a solution to the homogeneous system
y0 = Ay.

◦ y1 and y2 are solutions to the general equation. Then (y2 − y1 )0 = y20 − y10 =
Proof: Suppose that
(Ay1 +q)−(Ay2 +q) = A(y1 −y2 ), meaning that their dierence y2 −y1 is a solution to the homogeneous
equation.

2
 • Theorem (Homogeneous Systems): If the coecient functions ai,j (x) are continuous on an interval I for each
1 ≤ i, j ≤ n, then the set of solutions y to the homogeneous system y0 = Ay on I is an n-dimensional vector
space.

◦ Proof: First, the solution space is a subspace, since it satises the subspace criterion:

∗ [S1]: The zero function is a solution.

∗ [S2]: If y1 and y2 are solutions, then (y1 + y2 )0 = y10 + y20 = A(y1 + y2 ) so y1 + y2 is also a solution.
0 0
∗ [S3]: If α is a scalar and y is a solution, then (αy) = αy = α(Ay) = A(αy) so αy is also a solution.

◦ Now we need to show that the solution space is n-dimensional. We will do this by nding a basis.

∗ Choose any x0 in I. 1≤i≤n

By the existence part of the existence-uniqueness theorem, for each
there exists a function zi such that zi (x0 ) is the ith unit coordinate vector of Rn , with zi,i (x0 ) = 1
and xi,j (x0 ) for all j 6= i.

∗ The functions z1 , z2 , . . . , zn are linearly independent because their Wronskian matrix evaluated at
x = x0 is the identity matrix. (In particular, the Wronskian is not the zero function.)
 
c1
∗ Now suppose y is any solution to the homogeneous equation, with y(x0 ) =  ... .
 

cn
 
c1
∗ Then the function z = c1 z1 + c2 z2 + · · · + cn zn also has z(x0 ) =  ...  and is a solution to the
 

cn
homogeneous equation.

∗ But by the uniqueness part of the existence-uniqueness theorem, there is only one such function, so
we must have y(x) = z(x) for all x: therefore y = c1 z1 + c2 z2 + · · · + cn zn , meaning that y is in the
span of z1 , z2 , . . . , zn .
∗ This is true for any solution function y, so z1 , z2 , . . . , zn span the solution space. Since they are also
linearly independent, they form a basis of the solution space, and because there are n of them, we
see that the solution space is n-dimensional.

• If we combine the above results, we can write down a fairly nice form for the solutions of a general system of
rst-order dierential equations:

• Corollary: The general solution to the nonhomogeneous equation y0 = Ay + q has the form y = ypar + C1 z1 +
C2 z2 + · · · + Cn zn , where ypar is any one particular solution of the nonhomogeneous equation, z1 , . . . , zn are
a basis for the solutions to the homogeneous equation, and C1 , . . . , Cn are arbitrary constants.

◦ This corollary says that, in order to nd the general solution, we only need to nd one function which
satises the nonhomogeneous equation, and then solve the homogeneous equation.

6.2 Eigenvalue Method (Nondefective Coecient Matrices)

• We now restrict our discussion to homogeneous rst-order systems with constant coecients: those of the
form

y10 = a1,1 y1 + a1,2 y2 + · · · + a1,n yn

y20 = a2,1 y1 + a2,2 y2 + · · · + a2,n yn
. .
. .
. .

yn0 =an,1 y1 + an,2 y2 + · · · + an,n yn

   
y1 a1,1 a1,2 ··· a1,n
 y2   a2,1 a2,2 ··· a2,n 
which we will write in matrix form as y0 = Ay with y =  .  and A =  . .
   
. .. .
 ..   .. .
. . .
. 
yn an,1 an,2 ··· an,n

3
 
c1
 c2 
• Our starting point for solving such systems is to observe that if v =  .  is an eigenvector of A with
 
 .. 
cn
 
c1
 c2 
eigenvalue λ, then y =  .  eλx is a solution to y0 = Ay.
 
 .. 
cn

◦ This follows simply by dierentiating y = eλx v with respect to x: we see y0 = λeλx v = λy = Ay.
◦ In the event that A has n linearly independent eigenvectors, we will therefore obtain n solutions to the
dierential equation.

◦ If these solutions are linearly independent, then since we know the solution space is n-dimensional, we
would be able to conclude that our solutions are a basis for the solution space.

• Theorem (Eigenvalue Method): If A has n linearly independent eigenvectors v1 , v2 , . . . , vn with associated

eigenvalues λ1 , λ2 , . . . , λn , then the general solution to the matrix dierential system y0 = Ay is given by
y = C1 eλ1 x v1 + C2 eλ2 x v2 + · · · + Cn eλn x v2 , where C1 , · · · , Cn are arbitrary constants.

◦ Recall that if λ is a root of the characteristic equation k times, we say that λ has multiplicity k . If the
eigenspace for λ has dimension less than k , we say that λ is defective. The theorem allows us to solve
the matrix dierential system for any nondefective matrix.

◦ Proof: By the observation above, each of eλ1 x v1 , eλ2 x v2 , · · · , eλn x vn is a solution to y0 = Ay. We claim
that they are a basis for the solution space.

◦ To show this, we know by our earlier results that the solution space of the system y0 = Ay is n-
dimensional: thus, if we can show that these solutions are linearly independent, we would be able to
conclude that our solutions are a basis for the solution space.

◦ We can compute the Wronskian of these solutions: after factoring out the exponentials from each column,
 
| | |
we obtain W = e(λ1 +···+λn )x det(M ), where M =  v1 ··· vn .
| | |
◦ The exponential is always nonzero and the vectors v1 , v2 , . . . , vn are (by hypothesis) linearly independent,
meaning that det(M ) is nonzero. Thus, W is nonzero, so eλ1 x v1 , eλ2 x v2 , · · · , eλn x vn are linearly
independent.

◦ Since these solutions are therefore a basis for the solution space, we immediately conclude that the general
solution to y0 = Ay has the form y = C1 eλ1 x v1 + C2 eλ2 x v2 + · · · + Cn eλn x v2 , for arbitrary constants
C1 , · · · , Cn .

• The theorem allows us to solve all homogeneous systems of linear dierential equations whose coecient
matrix A has n linearly independent eigenvectors. (Such matrices are called nondefective matrices.)

y10 = y1 − 3y2
• Example: Find all functions y1 and y2 such that .
y20 = y1 + 5y2
 
1 −3 t−1 3
◦ The coecient matrix is A= , whose characteristic polynomial is det(tI−A) = =
1 5 −1 t−5
(t − 1)(t − 5) + 3 = t2 − 6t + 8 = (t − 2)(t − 4).
◦ Thus, the eigenvalues of A are λ = 2, 4.
   
2−1 3 1 3
◦ For λ = 2, we want to nd the nullspace of = . By row-reducing we nd
  −1 2 − 5 −1 −3  
1 3 −3
the row-echelon form is , so the 2-eigenspace is 1-dimensional and is spanned by .
0 0 1

4
    
4−1 3 3 3
◦ For λ = 4, we want to nd the nullspace of = . By row-reducing we nd
  −1 4 − 5 −1 −1  
1 1 −1
the row-echelon form is , so the 4-eigenspace is 1-dimensional and is spanned by .
0 0 1
     
y1 −3 −1
◦ Thus, the general solution to the system is = C1 e2x + C2 e4x .
y2 1 1

y10 = y1 − 3y2 + 7y3

• Example: Find all functions y1 , y2 , y3 such that y20 = −y1 − y2 + y3 .
y30 = −y1 + y2 − 3y3
 
1 −3 7
◦ The coecient matrix is A =  −1 −1 1 , whose characteristic polynomial is det(tI − A) =
−1 1 −3
t−1 3 −7
1 t+1 −1 = t3 + 3t2 + 2t = t(t + 1)(t + 2).
1 −1 t+3
◦ Thus, the eigenvalues of A are λ = 0, −1, −2.
 
−1 3 −7
◦ For λ = 0, we want to nd the nullspace of  1 1 −1 . By row-reducing we nd the row-echelon
1 −1 3
   
1 0 1 −1
form is  0 1 −2 , so the 0-eigenspace is 1-dimensional and is spanned by  2  .
0 0 0 1
 
−2 3 −7
◦ For λ = −1, we want to nd the nullspace of  1 0 −1 . By row-reducing we nd the row-echelon
  1 −1 2  
1 0 −1 1
form is  0 1 −3 , so the (−1)-eigenspace is 1-dimensional and is spanned by  3  .
0 0 0 1
 
−3 3 −7
◦ For λ = −2, we want to nd the nullspace of  1 −1 −1 . By row-reducing we nd the row-echelon
  1 −1 1  
1 −1 0 1
form is  0 0 1 , so the (−2)-eigenspace is 1-dimensional and is spanned by  1  .
0 0 0 0
       
y1 −1 1 1
−x
◦ Thus, the general solution to the system is  y2  = C1  2  + C2  3  e + C3  1  e−2x .
y3 1 1 0

• In the event that the coecient matrix has complex-conjugate eigenvalues, we generally want to rewrite the
resulting solutions as real-valued functions.

◦ SupposeA has a complex eigenvalue λ = a+bi with associated eigenvector v = w1 +iw2 . Then λ̄ = a−bi
v = w1 − iw2 (the conjugate of v), so we obtain the two solutions eλx v and eλ̄x v̄ to
has an eigenvector
0
the system y = Ay.
1 1
◦ Now we observe that (eλx v+eλ̄x v̄) = eax (w1 cos(bx)−w2 sin(bx)), and (eλx v−eλ̄x v̄) = eax (w1 sin(bx)+
2 2i
w2 cos(bx)), and the latter solutions are real-valued.
◦ Thus, to obtain real-valued solutions, we can replace the two complex-valued solutions eλx v and eλ̄x v̄
ax ax
with the two real-valued solutions e (w1 cos(bx) − w2 sin(bx)) and e (w1 sin(bx) + w2 cos(bx)), which
are simply the real part and imaginary part of eλx v respectively.

y10 = y2
• Example: Find all real-valued functions y1 and y2 such that .
y20 = −y1

5
  
0 1 t −1
◦ The coecient matrix is A= , whose characteristic polynomial is det(tI − A) = =
−1 0 1 t
t2 + 1 .
◦ Thus, the eigenvalues of A are λ = ±i.
 
i −1
◦ For λ = i, we want to nd the nullspace of . By row-reducing we nd the row-echelon form
  1 i  
i −1 1
is , so the i-eigenspace is 1-dimensional and spanned by .
0 0 i
 
1
◦ For λ = −i we can take the complex conjugate of the eigenvector for λ=i to see that is an
−i
eigenvector.
     
y1 1 1
◦ The general solution, as a complex-valued function, is + C2 e−ix . = C1 e ix
y2 i −i
   
1 1
◦ We want real-valued solutions, so we must replace the complex-valued solutions ix
e and e−ix
i −i
with real-valued ones.
      
1 1 0 0
◦ We have λ=i and v= +
w1 = and w2 = i so that
.
0 0 1 1
       
1 0 cos(x) 1
◦ Thus, the equivalent real-valued solutions are cos(x)− sin(x) = and sin(x)+
0 1 − sin(x) 0
   
0 sin(x)
cos(x) = .
1 cos(x)
     
y1 cos(x) sin(x)
◦ The system's solution is then = C1 + C2 .
y2 − sin(x) cos(x)

y10 = 3y1 − 2y2

• Example: Find all real-valued functions y1 and y2 such that .
y20 = y1 + y2
 
3 −2 t−3 2
◦ The coecient matrix is A= , whose characteristic polynomial is det(tI−A) = =
1 1 −1 t−1
t2 − 4t + 5.
◦ Thus, the eigenvalues of A are λ=2±i by the quadratic formula.
 
−1 + i 2
◦ For λ = 2 + i, we want to nd the nullspace of . By row-reducing we nd the row-
  −1 1+i  
1 −1 − i 1+i
echelon form is , so the i-eigenspace is 1-dimensional and spanned by .
0 0 1
 
1−i
◦ For λ = 2−i we can take the complex conjugate of the eigenvector for λ = 2+i to see that is
1
an eigenvector.
     
y1 1+i (2+i)x 1−i
◦ The general solution, as a complex-valued function, is = C1 e +C2 e(2−i)x .
y2 1 1
 
1+i
◦ We want real-valued solutions, so we must replace the complex-valued solutions e(2+i)x and
1
 
1−i
e(2−i)x with real-valued ones.
1
       
1 1 1 1
◦ We have λ=2+i and v= w1 = + and w2 = i so that
.
1 1 0 0
   
     
1 1 1 1
◦ Thus, the real-valued solutions are e2x cos(x) − sin(x) and e2x sin(x) + cos(x) .
1 0 1 0
     
y1 cos(x) − sin(x) sin(x) + cos(x)
◦ The system's solution is then = C1 e2x + C2 e2x .
y2 cos(x) sin(x)

6
 y10 = 3y1 − 7y2 − 3y3
• Example: Find all real-valued functions y1 , y2 , y3 such that y20 = y1 − 4y2 − 2y3 .
y30 = y1 + 2y2 + 2y3
 
3 −7 −3
◦ The coecient matrix is A =  1 −4 −2 , whose characteristic polynomial is det(tI − A) =
1 2 2
t−3 7 3
−1 t+4 2 = (t + 1)(t2 − 2t + 2).
−1 −2 t−2
◦ The eigenvalues of A are λ = −1, 1 ± i by the quadratic formula.
 
−4 7 3
◦ For λ = −1, we want to nd the nullspace of  −1 3 2 . By row-reducing we nd the row-echelon
  −1 −2 −3  
1 0 1 −1
form is  0 1 1 , so the (−1)-eigenspace is 1-dimensional and spanned by  −1  .
0 0 0 1
 
−2 + i 7 3
◦ For λ = 1 + i, we want to nd the nullspace of  −1 5+i 2 . By row-reducing we nd the
  −1 −2 −1 + i  
5 0 1 − 3i −1 + 3i
row-echelon form is  0 5 2 − i , so the (1 + i)-eigenspace is spanned by  −2 + i  .
0 0 0 5
 
−1 − 3i
◦ For λ = 1 − i we can take the complex conjugate of the eigenvector for λ = 1 + i to see that  −2 − i 
5
is an eigenvector.
    
y1 −1 −1 + 3i
−x
◦ The general solution, as a complex-valued function, is  y2  = C1  −1  e +C2  −2 + i  e−(1+i) +
  y3 1 5
−1 − 3i
C3  −2 − i  e−(1−i) , but we need to replace the complex-valued solutions with real-valued ones.
5
       
−1 3 −1 3
◦ We have λ = 1 + i and v =  −2  +  1  i so that w1 =  −2  and w2 =  1 .
5 0 5 0
         
−1 3 −1 3
x 2x 
◦ Thus, the real-valued solutions are e  −2  cos(x) −  1  sin(x) and e −2  sin(x) +  1  cos(x).
5 0 5 0
       
y1 −1 − cos(x) − 3 sin(x) − sin(x) + 3 cos(x)
−x x x
◦ Then  y2  = C1  −1  e + C2 e  −2 cos(x) − sin(x)  + C3 e  −2 sin(x) + cos(x)  .
y3 1 5 cos(x) 5 sin(x)

Well, you're at the end of my handout. Hope it was helpful.

Copyright notice: This material is copyright Evan Dummit, 2012-2016. You may not reproduce or distribute this
material without my express permission.