Peano's Existence Theorem Revisited: February, 2012
Peano's Existence Theorem Revisited: February, 2012
Peano's Existence Theorem Revisited: February, 2012
February, 2012
Introduction
y(t0 ) = y0 .
(2.2)
The proof. As usual, we start by fixing some constants a > 0 and b > 0
such that the function f is defined and continuous in the (n+1)dimensional
interval [t0 , t0 + a] {y Rn : ky y0 k b}, which is a compact subset of
Rn+1 . Hence there exists L > 0 such that
kf (t, y)k L whenever 0 t t0 a and ky y0 k b.
We now define the real interval I = [t0 , t0 + c] with length
c = min{a, b/L},
and we consider the set A of all functions : I Rn such that (t0 ) = y0
and which satisfy a Lipschitz condition with constant L, i.e.,
k(t) (s)k L|t s| for all s, t I.
The previous choice of the constant c guarantees that every function
A satisfies k(t) y0 k b for all t I. Hence, for every A
the composition t I 7 f (t, (t)) is welldefined, bounded (by L) and
continuous in I. It is therefore possible to construct a mapping
F : A [0, +)
4
which is a sort of measure of how far the function is from being a solution.
In fact, the Fundamental Theorem of Calculus ensures that if F () = 0 for
some A then is a solution of the initial value problem (1.1) in the
interval I (the converse is also true, but we do not need it for this proof.)
It is easy to check that the mapping F is continuous in A (equipped
with the topology of the uniform convergence in I). In turn, by the Arzel`a
Ascoli Theorem, the domain A is compact. Hence F attains a minimum at
some A.
To show that F () = 0 (which implies that is a solution) it suffices to
prove that F assumes arbitrarily small positive values in A. To do so, we
follow Tonelli, and for k N, k 2, we consider the approximate problem1
y0
for all t [t0 , t0 + c/k],
k (t) =
Hence
Lc
.
f (s, k (s)) ds
k
tc/k
t
Lc
,
k
thus proving that F () = 0 because k can be chosen as big as we wish.
0 F () F (k )
The differential equations in the approximate problems belong to the class of differential equations with delay. See, for instance, [17].
(t t1 )3 , if t < t1 ,
(t) =
0,
if t1 t t2 ,
3
(t t2 ) , if t > t2 ,
the greatest solutions to (1.1). This result goes back precisely to Peano
[14], where the greatest solution defined on the right of t0 was obtained
as the infimum of all strict upper solutions. A strict upper solution to the
problem (1.1) on some interval I is, roughly speaking, some function =
(t) satisfying (t) > f (t, (t)) for all t I, and (t0 ) y0 . Peano also
showed in [14] that the least solution is the supremum of all strict lower
solutions, which we define by reversing all the inequalities in the definition
of strict upper solution.
Notice, for instance, that Peanos example (2.3) has infinitely many solutions, the least one being
(t) = t3 for t < 0,
(t) = 0 for t 0,
Here we present a very easy proof of the existence of the least and the
greatest solutions which does not lean on lower/upper solutions (as in [4, 7,
14, 16]) or on special sequences of approximate solutions (as in [4, 18, 20]).
Basically, we obtain the greatest solution as the solution having the greatest
integral. This idea works in other settings, see [5].
Theorem 3.1 (Second version of Peanos Theorem) Consider problem (1.1)
in dimension n = 1, and assume that there exist constants a, b, L (0, +)
such that the function
f : [t0 , t0 + a] [y0 b, y0 + b] R
is continuous and |f (t, y)| L for all (t, y) [t0 , t0 + a] [y0 b, y0 + b].
Then there exist solutions of (1.1) , : I = [t0 , t0 + c] R, where
c = min{a, b/L}, such that every solution of (1.1) : I R satisfies
(t) (t) (t)
for all t I.
Let S denote the set of solutions of (1.1) defined on I. Our first version of
Peanos Theorem ensures that S is not an empty set. Moreover standard
arguments show that S A and that S is a compact subset of C(I). Hence
the continuous mapping
Z t0 +c
I : S 7 I() =
(s) ds
t0
(t), if t [t2 , t3 ],
1 : t I 7 1 (t) =
(t), otherwise.
Elementary arguments with side derivatives show that 1 S. Moreover
1 in I, with strict inequality in a subinterval, hence
I( ) < I(1 ),
but this is a contradiction with the choice of .
Similarly, one can prove that I attains a minimum at certain S,
and that is the least element in S.
Can Theorem 3.1 be adapted to systems? Yes, it can, but more than
continuity must be required for the function f , as we will specify below.
The need of some extra conditions is easily justified with examples of the
following type.
Example 3.1 Consider the system
2/3
y1 = 3y1 , y1 (0) = 0,
y2 = y1 ,
y2 (0) = 0.
The first problem can be solved independently, and it has infinitely many
solutions (this is Peanos example again). Notice that the greater the solution
8
we choose for y1 is then the smaller the corresponding y2 becomes on the right
of t0 = 0. Therefore this system does not have a solution which is greater
than the other ones in both components.
Notice however that the system in Example 3.1 has a solution whose first
component is greater than the first component of any other solution, and
the same is true replacing greater by smaller or first component by
second component. This observation leads us naturally to the following
question: in the conditions of Peanos Existence Theorem we fix a component i {1, 2, . . . , n}, can we ensure the existence of a solution with the
greatest i-th component? The following example answers this question on
the negative.
Example 3.2 Let : R R be a continuously differentiable function
such that (0) = 0 and assumes both negative and positive values in every
neighborhood of 0 (hence (0) = 0)2 .
The idea is to construct a system whose solutions have a component
which is a translation of , and then those specific components cannot be
compared. To do so it suffices to consider the two dimensional system
2/3
y1 (0) = 0,
y1 = 3y1 ,
y = y 1/3 , y (0) = 0.
2
2
1
Let > 0 be fixed; we are going to prove that there is not a solution of
the system whose second component is greater than the second component of
any other solution on the whole interval [0, ].
First, note that we can compute all the solutions. For each a [0, ] we
have a solution (t) = (1 (t), 2 (t)) given by
1 = 0 on [0, a]
and
Now let us consider two arbitrary solutions of the system. They are
given by the above formulas for some corresponding values a = b and a = b ,
with 0 b < b , and then their respective second components cannot be
compared in the subinterval (b, b ).
The previous example still has a solution with the greatest first component, but, in the authors opinion, this is just a consequence of the fact
that the first equation in the system is uncoupled and we can solve it independently (in particular, Theorem 3.1 applies). However this remark raises
the open problem of finding a two dimensional system which has neither a
solution with the greatest first component nor a solution with the greatest
second component.
A multidimensional version of Theorem 3.1 is valid if the nonlinear part
f (t, y) = (f1 (t, y), f2 (t, y), . . . , fn (t, y))
is quasimonotone nondecreasing, i.e., if for each component i {1, 2, . . . , n}
the relations yj yj , j 6= i, imply
fi (t, (y1 , . . . , yi1 , y, yi , . . . , yn )) fi (t, (
y1 , . . . , yi1 , y, yi , . . . , yn )).
The reader is referred to [1, 2, 8, 9, 19, 20] and references therein for more
information on quasimonotone systems.
The power of lower and upper solutions: Existence for nonlocal problems
The real power of lower and upper solutions reveals when we want to guarantee the existence of solution to (1.1) on a given interval, and not merely
in an unknown (possibly very small) neighborhood of t0 .
Let a > 0 be fixed, let f : [t0 , t0 + a] R R be continuous, and
consider the nonlocal problem
y = f (t, y) for all t I = [t0 , t0 + a],
y(t0 ) = y0 .
(4.4)
Note that we can find in the literature some other denominations for upper (lower)
solutions, such as upper (lower) functions, or superfunctions (subfunctions).
10
for all t I.
t0
Lemma 4.2 For all t1 , t2 I such that t1 < t2 , and all y1 [(t1 ), (t1 )],
there exists A such that
Z t2
f (s, (s)) ds.
(t1 ) = y1 and (t2 ) = y1 +
t1
existence of functions
and in A1 such that G(
) 0 G().
Next we show how to construct one such
from (the construction of
from is analogous and we omit it). If (t1 ) = y1 we simply take
= .
If, on the other hand, (t1 ) < y1 then we define
in three (or two) pieces:
first, we define
on [t0 , t1 ] as an adequate convex linear combination of
and to have
(t1 ) = y1 ; second, we define
on the right of t1 as the
function whose graph is the line with slope L starting at the point (t1 , y1 )
and on the interval [t1 , t3 ], where t3 is the first point in the interval (t1 , t0 +a)
such that the line intersects with the graph of , and finally we continue
Now we carry on with the final part of the proof of Theorem 4.1. Let
> 0 be fixed and consider a partition of the interval [t0 , t0 + a], say
t0 , t1 , . . . , tk = t0 + a (k N), such that
0 < tj tj1 <
4
2L
This is not true in dimension n > 1, so this approach does not work in that case.
12
tj
f (s, (s)) ds
tj1
(s))
ds
f
(s,
(s))
ds
t0
tj1
Z t
| (t) (tj1 )| +
|f (s, (s))| ds
tj1
y(t0 ) = y0 ,
13
Example 4.1 The function (t) = 0 for all t [0, ] is a lower solution to
the initial value problem
y = 1 + y2,
y(0) = 0,
which has no solution defined on [0, ] (Its unique solution is (t) = tan t
for all t (/2, /2)).
Theorem 4.1 does not guarantee that every solution of (4.4) is located
between the lower and the upper solutions. To see it, simply note that
solutions are lower and upper solutions at the same time, so the zero function
is both lower and upper solution for Peanos example (2.3) which has many
solutions above the zero function.
However it is true that, in the conditions of Theorem 4.1, we have a
greatest and a least solution between and , and we have their respective
Goodmans characterizations in terms of lower and upper solutions, see [7].
Corollary 4.3 Suppose that problem (4.4) has a lower solution and an
upper solution such that (t) (t) for all t I and let
[, ] = { C(I) : on I}.
Then problem (4.4) has solutions , [, ] such that every solution
of (1.1) [, ] satisfies
(t) (t) (t)
for all t I.
(t I),
(4.5)
(t I). (4.6)
Remark 4.2 Corollary 4.3 can be proven directly via Perrons method, see
[4, 7, 12, 16]. This means starting at (4.6) as a definition and then showing
that is a solution in the conditions of the statement. Perrons method
involves careful work with sets of onesided differentiable functions, which
we avoid. Corollary 4.3 can also be proven easily from Theorem 3.1.
Finally we deduce from Corollary 4.3 the Peanos characterizations of the
least and the greatest solutions in terms of strict lower and upper solutions.
In doing so we are finally proving the real Peanos Theorem, because our
next result is the closest to the one proven in [14].
We say that : I R is a strict lower solution of (4.4) if it is continuously differentiable in I, (t0 ) y0 , and (t) < f (t, (t)) for all t I.
A strict upper solution is defined analogously by reversing the relevant inequalities. Notice that if is a strict lower solution and is a strict upper
solution, then < on (t0 , t0 +a], for otherwise we could find t1 (t0 , t0 +a]
such that < in (t0 , t1 ) and (t1 ) = (t1 ), but then we would have
(t1 ) > f (t1 , (t1 )) = f (t1 , (t1 )) > (t1 ) (t1 ),
a contradiction.
(4.7)
(t I), (4.8)
and
(t) = inf{(t) : strict upper solution of (4.4)}
(t I). (4.9)
Proof. Corollary 4.3 guarantees that (4.4) has the least and the greatest
solutions in [, ] = { C(I) : }, which we denote, respectively,
by and .
Let us prove that solutions of (4.4) satisfy (4.7) which, in particular,
ensures that all of them belong to [, ]. Reasoning by contradiction, assume
that a certain solution : [t0 , t0 +] R ( (0, a)) satisfies (t1 ) (t1 )
for some t1 (t0 , t0 + ). The initial conditions ensure that we can find some
t2 (t0 , t1 ) such that < in (t0 , t2 ) and (t2 ) = (t2 ), but then we have
(t2 ) > f (t2 , (t2 )) = f (t2 , (t2 )) = (t2 ) (t2 ),
a contradiction. This proves that every solution is smaller than on its
domain, and a similar argument shows that every solution is greater than .
Therefore Theorem 4.1 ensures that every solution of (4.4) can be continued
to the whole interval I = [t0 , t0 + a] as a solution of (4.4) and between
and . Hence (4.7) obtains and, moreover, and are, respectively, the
least and the greatest among all the solutions of (4.4).
Next we show that (4.9) is satisfied. The proof of (4.8) is similar and we
omit it.
The previous arguments still work if we replace by any other strict
upper solution : I R (necessarily greater than on I). Hence
(t) inf{(t) : strict upper solution of (4.4)}
(t I).
for all t I.
y(t0 ) = y0 ,
(4.10)
and, in turn, is a lower solution. Hence, for all sufficiently large values
of k there exists k , a solution of (4.10), between and .
Obviously, k is a strict upper solution to (4.10) with k replaced by k+1,
hence k k+1 on I. Thus we can define a limit function
(t) = lim k (t) (t) (t I).
k
(4.11)
Concluding remarks
1. The assumption on I in Theorem 4.1 can be omitted. Marcelli
and Rubbioni proved in [13] that we have solutions between the minimum of and and the maximum of them. Furthermore, we do not
even need that lower or upper solutions be continuous, see [11].
2. Theorem 4.1 and Corollary 4.3 can be extended to quasimonotone
systems, see [1, 2, 9]. Theorem 4.1 for systems even works when the
lower and upper solutions are not ordered, see [13], and f may be
discontinuous or singular, see [1].
3. The lower and upper solutions method is today a most acknowledged
effective tool in the analysis of differential equations and, specially,
boundary value problems. A detailed account on how far the method
has evolved (just for secondorder ODEs!) is given in the monograph
by De Coster and Habets [3]. As far as the author is aware, the first
use of lower and upper solutions is due Peano in [14].
17
References
[1] D. C. Biles and R. L
opez Pouso, Firstorder singular and discontinuous
differential equations, Bound. Value Problems, 2009 (2009), Article ID
507671, 25 pages.
[2] D. C. Biles and E. Schechter, Solvability of a finite or infinite system of
discontinuous quasimonotone differential equations, Proc. Amer. Math.
Soc. 128 (2000), 33493360.
[3] C. De Coster and P. Habets, Twopoint boundary value problems: Lower
and Upper solutions, Mathematics in Science and Engineering, vol. 205,
Elsevier, 2006.
[4] M. A. Dow and R. V
yborn
y, Elementary proofs of Peanos Existence
Theorem, J. Aust. Math. Soc. 15 (1973), 366372.
[5] R. Figueroa and R. L
opez Pouso, Minimal and maximal solutions to
second-order boundary value problems with state-dependent deviating arguments, Bull. Lond. Math. Soc. 43 (2011), 164174.
[6] C. Gardner, Elementary proof of Peanos Existence Theorem, Amer.
Math. Monthly 83 (1976), 556560.
[7] G. S. Goodman, Subfunctions and the initialvalue problem for differential equations satisfying Caratheodorys hypotheses, J. Differential Equations 7 (1970), 232242.
[8] E. R. Hassan and W. Rzymowski, Extremal solutions of a discontinuous
scalar differential equation, Nonlinear Anal. 37 (1999), 9971017.
[9] S. Heikkila and V. Lakshmikantham, Monotone iterative techniques for
discontinuous nonlinear differential equations, Marcel Dekker, New York,
1994.
[10] H. C. Kennedy, Is there an elementary proof of Peanos Existence Theorem for first order differential equations?, Amer. Math. Monthly 76
(1969), 10431045.
[11] R. L
opez Pouso, Nonordered discontinuous upper and lower solutions
for first-order ordinary differential equations, Nonlinear Anal. 45 (2001),
391406.
18
[12] R. L
opez Pouso, Necessary conditions for solving initial value problems
with infima of superfunctions, Math. Inequal. Appl. 8 (2005), 633641.
[13] C. Marcelli and P. Rubbioni, A new extension of classical M
ullers
theorem, Nonlinear Anal. 28 (1997), 17591767.
[14] G. Peano, Sullintegrabilit`
a delle equazioni differenziali di primo ordine,
Atti. Accad. Sci. Torino 21 (1886), 677685.
[15] G. Peano, Demonstration de lintegrabilite des equations differentielles
ordinaires, Math. Ann. 37 2 (1890), 182228.
[16] O. Perron, Ein neuer Existenzbeweis f
ur die Integrale der Differentialgleichung y = f (x, y), Math. Ann. 76 (1915), 471484.
[17] H. Smith, An introduction to delay differential equations with applications to the life sciences, Springer, New York, 2011.
[18] J. Walter, On elementary proofs of Peanos Existence Theorem, Amer.
Math. Monthly 80 (1973), 282286.
[19] W. Walter, Differential and Integral Inequalities, Springer, Berlin,
1970.
[20] W. Walter, There is an elementary proof of Peanos Existence Theorem,
Amer. Math. Monthly 78 (1971), 170173.
19