626 Differential Equations with Applications and Historical Notes

Problems
1. Find the exact solution of the initial value problem

y′ = y2, y(0) = 1.

Starting with y0(x) = 1, apply Picard’s method to calculate y1(x), y2(x),

y3(x), and compare these results with the exact solution.
2. Find the exact solution of the initial value problem

y′ = 2x(l + y), y(0) = 0.

Starting with y0(x) = 0, calculate y1(x), y2(x), y3(x), y4(x), and compare these
results with the exact solution.
3. It is instructive to see how Picard’s method works with a choice of the
initial approximation other than the constant function y0(x) = y0. Apply
the method to the initial value problem (4) with
(a) y0(x) = ex;
(b) y0(x) = 1 + x;
(c) y0(x) = cos x.

70 Picard’s Theorem
As we pointed out at the end of the last section, the principal value of Picard’s
method of successive approximations lies in the contribution it makes to the
theory of differential equations. This contribution is most clearly illustrated
in the proof of the following basic theorem.

Theorem A. (Picard’s theorem.) Let f (x, y) and ∂f/∂y be continuous functions of

x and y on a closed rectangle R with sides parallel to the axes (Figure 105). If (x0, y0)
is any interior point of R, then there exists a number h > 0 with the property that the
initial value problem

y′ = f (x, y), y(x0) = y0 (l)

has one and only one solution y = y(x) on the interval |x − x0| ≤ h.
Proof. The argument is fairly long and intricate, and is best absorbed in easy
stages.
The Existence and Uniqueness of Solutions 627

y
R

(x0 , y0)
y = y(x)
R΄

x0 – h x0 + h x

FIGURE 105

First, we know that every solution of (1) is also a continuous solution of the
integral equation

y( x ) = y 0 +
ò f éët, y(t)ùû dt,
x0
(2)

and conversely. This enables us to conclude that (1) has a unique solution on
an interval |x − x0| ≤ h if and only if (2) has a unique continuous solution on
the same interval. In Section 69 we presented some evidence suggesting that
the sequence of functions yn(x) defined by

y0 ( x) = y0 ,
x

ò
y1( x) = y0 + f [t , y0 (t)] dt ,
x0
x

ò
y 2 ( x) = y0 + f [t , y1(t)] dt ,
x0
(3)

x

ò
y n ( x) = y0 + f [t , y n -1(t)] dt ,
x0

converges to a solution of (2). We next observe that yn(x) is the nth partial sum
of the series of functions
628 Differential Equations with Applications and Historical Notes

y0 ( x) + å[y (x) - y
n =1
n n -1 ( x)] = y0 ( x) + [ y1( x) - y0 ( x)]

+ [ y 2 ( x) - y1( x)] +  + [ y n ( x) - y n -1( x)] + , (4)

so the convergence of the sequence (3) is equivalent to the convergence of this

series. In order to complete the proof, we produce a number h > 0 that defines
the interval |x − x0| ≤ h and then we show that on this interval the following
statements are true: (i) the series (4) converges to a function y(x); (ii) y(x) is a
continuous solution of (2); (iii) y(x) is the only continuous solution of (2).
The hypotheses of the theorem are used to produce the positive number h,
as follows. We have assumed that f (x, y) and ∂f/∂y are continuous functions
on the rectangle R. But R is closed (in the sense that it includes its boundary)
and bounded, so each of these functions is necessarily bounded on R. This
means that there exist constants M and K such that

|f (x, y)| ≤ M (5)

and

¶
f (x, y) £ K (6)
¶y

for all points (x, y) in R. We next observe that if (x, y1) and (x, y2) are distinct
points in R with the same x coordinate, then the mean value theorem guar-
antees that

¶
f ( x , y1 ) - f ( x , y 2 ) = f ( x , y*) y1 - y 2 (7)
¶y

for some number y* between y1 and y2. It is clear from (6) and (7) that

|f (x, y1) − f (x, y2)| ≤ K|y1 − y2| (8)

for any points (x, y1) and (x, y2) in R (distinct or not) that lie on the same verti-
cal line. We now choose h to be any positive number such that

Kh < 1 (9)

and the rectangle R′ defined by the inequalities |x − x0| ≤ h and |y − y0| ≤ Mh

is contained in R. Since (x0, y0) is an interior point of R, there is no difficulty in
seeing that such an h exists. The reasons for these apparently bizarre require-
ments will of course emerge as the proof continues.
The Existence and Uniqueness of Solutions 629

From this point on, we confine our attention to the interval |x − x0| ≤ h. In
order to prove (i), it suffices to show that the series

y 0 ( x ) + y1 ( x ) - y 0 ( x ) + y 2 ( x ) - y1 ( x ) +  + y n ( x ) - y n - 1 ( x ) +  (10)

converges; and to accomplish this, we estimate the terms |yn(x) − yn−1(x)|. It is

first necessary to observe that each of the functions yn(x) has a graph that lies
in R′ and hence in R. This is obvious for y0(x) = y0, so the points [t, y0(t)] are in
R′, (5) yields |f [t, y0(t)]| ≤ M, and

y1 ( x ) - y 0 =
ò f [t, y (t)] dt £ Mh,
x0
0

which proves the statement for y1(x). It follows in turn from this inequality
that the points [t, y, (t)] are in R′, so |f [t, y1, (t)]| ≤ M and

y 2 ( x) - y0 =
ò f [t, y (t)] dt £ Mh .
x0
1

Similarly,

y 3 ( x) - y0 =
ò f [t, y (t) dt] £ Mh,
x0
2

and so on. Now for the estimates mentioned above. Since a continuous func-
tion on a closed interval has a maximum, and y1(x) is continuous, we can
define a constant a by a = max|y1(x) − y0| and write

|y1(x) − y0(x)| ≤ a.

Next, the points [t, y1(t)] and [t, y0(t)] lie in R′, so (8) yields

|f [t, y1(t)] − f [t, y0(t)]| ≤ K|y1(t) − y0(t)| ≤ Ka

and we have

y 2 ( x ) - y1 ( x ) =
ò( f [t, y (t)] - f [t, y (t)]) dt
x0
1 0

£ Kah = a(Kh).
630 Differential Equations with Applications and Historical Notes

Similarly,

|f [t, y2(t)] − f [t, y1(t)]| ≤ K|y2(t) − y1(t)| ≤ K 2 ah,

so
x

y 3 ( x) - y 2 ( x) =
ò( f [t, y (t)] - f [t, y (t)]) dt
x0
2 1

£ (K 2 ah)h = a(kh)2 .

By continuing in this manner, we find that

|yn(x) − yn−1(x)| ≤ a(Kh)n−1

for every n = 1, 2,... Each term of the series (10) is therefore less than or equal
to the corresponding term of the series of constants

y0 + a + a (Kh) + a (Kh)2 +  + a (Kh)n -1 + .

But (9) guarantees that this series converges, so (10) converges by the com-
parison test, (4) converges to a sum which we denote by y(x), and yn(x) → y(x).
Since the graph of each yn(x) lies in R′, it is evident that the graph of y(x) also
has this property.
Now for the proof of (ii). The above argument shows not only that yn(x)
converges to y(x) in the interval, but also that this convergence is uniform.
This means that by choosing n to be sufficiently large, we can make yn(x)
as close as we please to y(x) for all x in the interval; or more precisely, if ∈ > 0
is given, then there exists a positive integer n0 such that if n ≥ n0 we have
|y(x) − yn(x)| < ∈ for all x in the interval. Since each yn(x) is clearly continu-
ous, this uniformity of the convergence implies that the limit function y(x)
is also continuous.2 To prove that y(x) is actually a solution of (2), we must
show that

ò
y( x) - y0 - f [t , y(t)] dt = 0.
x0
(11)

2 We will not discuss this in detail, but the reasoning is quite simple and rests on the inequality

y( x) - y( x ) = [ y( x) - y n ( x)] + [ y n ( x) - y n ( x )] + [ y n ( x ) - y( x )]
£ y( x ) - y n ( x ) + y n ( x ) - y n ( x ) + y n ( x ) - y( x ) .
The Existence and Uniqueness of Solutions 631

But we know that

ò
y n ( x) - y0 - f [t , y n -1(t)] dt = 0,
x0
(12)

so subtracting the left side of (12) from the left side of (11) gives

x x

ò ò
y( x) - y0 - f [t , y(t)] dt = y( x) - y n ( x) + ( f [t , y n -1(t)] - f [t , y(t)]) dt,
x0 x0

and we obtain

ò
y( x) - y0 - f [t , y(t)] dt
x0

ò
£ y( x) - y n ( x) + ( f [t , y n -1(t)] - f [t , y(t)]) dt .
x0

Since the graph of y(x) lies in R′ and hence in R, (8) yields

ò
y( x) - y0 - f [t , y(t)] dt
x0

£| y( x) - y n ( x)|+ Kh max|y n -1( x) - y( x)|. (13)

The uniformity of the convergence of yn(x) to y(x) now implies that the right
side of (13) can be made as small as we please by taking n large enough. The
left side of (13) must therefore equal zero, and the proof of (11) is complete.
In order to prove (iii), we assume that y ( x) is also a continuous solution
of (2) on the interval |x − x0| ≤ h, and we show that y ( x) = y( x) for every x
in the interval. For the argument we give, it is necessary to know that the
graph of y ( x) lies in R′ and hence in R, so our first step is to establish this
fact. Let us suppose that the graph of y ( x) leaves R′ (Figure 106). Then the
properties of this function [continuity and the fact that y ( x0 ) = y0] imply that
there exists an x1 such that |x1 − x0| < h,| y ( x1 ) - y0 |= Mh and | y ( x) - y0 |< Mh
if |x − x0| < |x1 − x0|. It follows that

| y ( x1 ) - y0 | Mh Mh
= > = M.
|x1 - x0 | |x1 - x0 | h
632 Differential Equations with Applications and Historical Notes

y
–y(x)
R΄

Mh

y0

x0 – h x0 x1 x 0 + h x

FIGURE 106

However, by the mean value theorem there exists a number x* between x0

and x1 such that

| y ( x1 ) - y0 |
=| y ¢ ( x*)|=| f [x*, y ( x*)]|£ M ,
|x1 - x0 |

since the point [x*, y ( x*)] lies in R′. This contradiction shows that no point
with the properties of x1 can exist, so the graph of y ( x) lies in R′. To complete
the proof of (iii), we use the fact that y ( x) and y(x) are both solutions of (2) to
write
x

| y ( x) - y( x)|=
ò{ f [t, y(t)] - f [t, y(t)]}dt .
x0

Since the graphs of y ( x) and y(x) both lie in R′, (8) yields

| y ( x) - y( x)|£ Kh max|y ( x) - y( x)|,

so
max|y ( x) - y( x)|£ Kh max| y ( x) - y( x)|.

This implies that max y ( x) - y( x) = 0, for otherwise we would have 1 ≤ Kh

in contradiction to (9). It follows that y ( x) = y( x) for every x in the inter-
val |x − x0| ≤ h, and Picard’s theorem is fully proved.

Remark 1. This theorem can be strengthened in various ways by weakening

its hypotheses. For instance, our assumption that ∂f/∂y is continuous on R is
stronger than the proof requires, and is used only to obtain the inequality (8).
The Existence and Uniqueness of Solutions 633

We can therefore introduce this inequality into the theorem as an assump-

tion that replaces the one about ∂f/∂y. In this way we arrive at a stronger form
of the theorem since there are many functions that lack a continuous partial
derivative but nevertheless satisfy (8) for some constant K. This inequality,
which says that the difference quotient

f ( x , y1 ) - f ( x , y 2 )
y1 - y 2

is bounded on R, is called a Lipschitz3 condition in the variable y.

Remark 2. If we drop the Lipschitz condition, and assume only that f (x,y) is
continuous on R, then it is still possible to prove that the initial value problem
(1) has a solution. This result is known as Peano’s theorem.4 The only known
proofs depend on more sophisticated arguments than those we have used
above.5 Furthermore, the solution whose existence this theorem guarantees
is not necessarily unique. As an example, consider the problem
y′ = 3y2/3, y(0) = 0, (14)
and let R be the rectangle |x| ≤ 1, |y| ≤ 1. Here f(x,y) = 3y2/3 is plainly continu-
ous on R. Also, y1(x) = x3 and y2(x) = 0 are two different solutions valid for all
x, so (14) certainly has a solution that is not unique. The explanation for this
nonuniqueness lies in the fact that f (x,y) does not satisfy a Lipschitz condi-
tion on the rectangle R, since the difference quotient

f (0, y ) - f (0, 0) 3 y 2/3 3

= = 1/3
y=0 y y

is unbounded in every neighborhood of the origin.

3 Rudolf Lipschitz (1832–1903) was a professor at Bonn for most of his life. He is remembered
chiefly for his role in simplifying and clarifying Cauchy’s original theory of the existence and
uniqueness of solutions of differential equations. However, he also extended Dirichlet’s theo-
rem on the representability of a function by its Fourier series, obtained the formula for the
number of ways a positive integer can be expressed as a sum of four squares as a consequence
of his own theory of the factorization of integral quaternions, and made useful contributions
to theoretical mechanics, the calculus of variations, Bessel functions, quadratic differential
forms, and the theory of viscous fluids.
4 Guiseppe Peano (1858–1932), Italian logician and mathematician, strongly influenced
Hilbert’s axiomatic treatment of plane geometry and the work of Whitehead and Russell on
mathematical logic. His postulates for the positive integers have led generations of students
to wonder whether all of modern algebra is some kind of conspiracy to render the obvious
obscure (it is not!). In 1890 he astounded the mathematical world with his remarkable con-
struction of a continuous curve in the plane that completely fills the square 0 ≤ x ≤ l, 0 ≤ y ≤ 1.
Unfortunately for a man who valued logic so highly, his 1886 proof of the above existence
theorem for solutions of y′ = f(x,y) was inadequate, and a satisfactory proof was not found
until many years later.
5 See, for example, A. N. Kolmogorov and S. V. Fomin, Elements of the Theory of Functions and
Functional Analysis, vol, 1, p. 56, Graylock, Baltimore, 1957.
634 Differential Equations with Applications and Historical Notes

Remark 3. Theorem A is called a local existence and uniqueness theorem

because it guarantees the existence of a unique solution only on some inter-
val |x − x0| ≤ h where h may be very small. There are several important cases
in which this restriction can be removed. Let us consider, for example, the
first order linear equation

y′ + P(x)y = Q(x),

where P(x) and Q(x) are defined and continuous on an interval a ≤ x ≤ b. Here
we have

f(x,y) = −P(x)y + Q(x);

and if K = max|P(x)| for a ≤ x ≤ b, it is clear that

|f(x,y1) − f(x,y2)| = |−P(x)(y1 − y2)| ≤ K |y1 −y2|.

The function f(x,y) is therefore continuous and satisfies a Lipschitz condition

on the infinite vertical strip defined by a ≤ x ≤ b and −∞ < y < ∞. Under these
circumstances, the initial value problem

y′ + P(x)y = Q(x), y(x0) = y0

has a unique solution on the entire interval a ≤ x ≤ b. Furthermore, the point

(x0,y0) can be any point of the strip, interior or not. This statement is a special
case of the next theorem.

Theorem B. Let f (x,y) be a continuous function that satisfies a Lipschitz condition

|f(x,y1) − f(x,y2)| ≤ K|y1 − y2|

on a strip defined by a ≤ x ≤ b and −∞ < y < ∞. If (x0,y0) is any point of the strip, then
the initial value problem

y′ = f (x,y), y(x0) = y0 (15)

has one and only one solution y = y(x) on the interval a ≤ x ≤ b.

Proof. The argument is similar to that given for Theorem A, with certain
simplifications permitted by the fact that the region under discussion is not
bounded above or below. In particular, we start the proof in the same way
and show that the series (4)— and therefore the sequence (3)—is uniformly
convergent on the whole interval a ≤ x ≤ b. We accomplish this by using a
somewhat different method of estimating the terms of the series (10).
The Existence and Uniqueness of Solutions 635

First, we define M0, M1, and M by

M0 = |y0|, M1 = max|y1(x)|, M = M0 + M1,

and we notice that |y0(x)| ≤ M and |y1(x) − y0(x)| ≤ M. Next, if x0 ≤ x ≤ b, it fol-

lows that

| y 2 ( x ) - y1 ( x ) =
ò{ f [t, y (t)] - f [t, y (t)]} dt
x0
1 0

ò
£ | f [t , y1(t)] – f [t , y0 (t)]|dt
x0

ò
£ K | y1(t) – y0 (t)|dt
x0

£ KM( x – x0 ),
x

| y 3 ( x) – y 2 ( x)| =
ò{ f [t, y (t) – f [t, y (t)]} dt
x0
2 1

ò
£ K | y 2 (t) – y1(t)|dt
x0

x
( x – x 0 )2
ò
£ K 2 M (t – x0 ) dt = K 2 M
x0
2
,

and in general

( x - x 0 )n - 1
y n ( x ) - y n -1 ( x ) £ K n -1 M .
(n - 1)!

The same argument is also valid for a ≤ x ≤ x0, provided only that x − x0 is
replaced by |x − x0|, so we have

|x - x0 |n -1
y n ( x ) - y n -1 ( x ) £ K n -1 M
(n - 1)!

( b - a )n - 1
£ K n -1 M
(n - 1)!
636 Differential Equations with Applications and Historical Notes

for every x in the interval and n = 1, 2,.... We conclude that each term of the
series (10) is less than or equal to the corresponding term of the convergent
series of constants

( b - a )2 ( b - a )3
M + M + KM(b - a) + K 2 M + K 3M + ,
2! 3!

so (3) converges uniformly on the interval a ≤ x ≤ b to a limit function y(x).

Just as before, the uniformity of the convergence implies that y(x) is a solu-
tion of (15) on the whole interval, and all that remains is to show that it is the
only such solution. We assume that y ( x) is also a solution of (15) on the inter-
val. Our strategy is to show that y n ( x) ® y ( x) for each x as n→∞; and since we
also have yn(x) → y(x), it will follow that y ( x) = y( x). We begin by observing
that y ( x) is continuous and satisfies the equation

y( x) = y0 +
ò f [t, y(t)] dt.
x0

If A = max|y ( x) - y0 |, then for x0 ≤ x ≤ b we see that

| y ( x ) - y1( x)|=
ò{ f [t, y(t)] - f [t, y (t)]} dt
x0
0

ò
£ | f [t , y (t)] - f [t , y0 (t)]|dt
x0

ò
£ K | y (t) - y0 |dt
x0

£ KA( x - x0 ),
x

| y ( x) - y 2 ( x)|=
ò{ f [t, y(t)] - f [t, y (t)]} dt
x0
1

ò
£ K | y (t) - y1(t)|dt
x0

x
( x - x 0 )2
ò
£ K A (t - x0 ) dt = K 2 A
2
,
2
x0
The Existence and Uniqueness of Solutions 637

and in general

( x - x 0 )n
y( x) - y n ( x) £ K n A ,
n!

A similar result holds for a ≤ x ≤ x0 so for any x in the interval we have

n
x - x0 ( b - a )n
y( x) - y n ( x) £ K n A £ K nA .
n! n!

Since the right side of this approaches zero as n → ∞, we conclude that

y ( x) = y( x) for every x in the interval, and the proof is complete.

Problems
1. Let (x0,y0) be an arbitrary point in the plane and consider the initial
value problem

y′ = y2, y(x0) = y0.

Explain why Theorem A guarantees that this problem has a unique

solution on some interval |x − x0| ≤ h. Since f(x,y) = y2 and ∂f/∂y = 2y are
continuous on the entire plane, it is tempting to conclude that this solu-
tion is valid for all x. By considering the solutions through the points
(0,0) and (0,1), show that this conclusion is sometimes true and some-
times false, and that therefore the inference is not legitimate.
2. Show that f(x,y) = y1/2
(a) does not satisfy a Lipschitz condition on the rectangle |x| ≤ 1 and
0 ≤ y ≤ 1;
(b) does satisfy a Lipschitz condition on the rectangle |x| ≤ 1 and
c ≤ y ≤ d, where 0 < c < d.
3. Show that f(x,y) = x2|y| satisfies a Lipschitz condition on the rectangle
|x| ≤ 1 and |y| ≤ 1 but that ∂f/∂y fails to exist at many points of this
rectangle.
4. Show that f(x,y) = xy2
(a) satisfies a Lipschitz condition on any rectangle a ≤ x ≤ b and c ≤ y ≤ d;
(b) does not satisfy a Lipschitz condition on any strip a ≤ x ≤ b> and
–∞ < y < ∞.