Gabriel

c Nagy

The Stolz-Cesaro Theorem

P
The Theorem. If (bn ) n=1 is a sequence of positive real numbers, such that n=1 bn = ,

then for any sequence (an )n=1 R one has the inequalities:
a1 + a2 + + an an
lim sup lim sup ; (1)
n b1 + b2 + + bn n bn
a1 + a2 + + an an
lim inf lim inf . (2)
n b1 + b2 + + bn n bn

In particular, if the sequence (an /bn )

n=1 has a limit, then

a1 + a2 + + an an
lim = lim .
n b1 + b2 + + bn n bn

Proof. . It is quite clear that we only need to prove (1), since the other inequality follows
by replacing an with an .
The inequality (1) is trivial, if the right-hand side is +. Assume then that the quantity
L = lim supn (an /bn ) is either finite or , and let us fix for the moment some number
` > L. By the definition of lim sup, there exists some index k N, such that
an
`, n > k. (3)
bn
Using (3) we get the inequalities

a1 + a2 + + an a1 + + ak + `(bk+1 + bk+2 + . . . bn ), n > k. (4)

If we denote for simplicity the sums a1 + + an by An and b1 + + bn by Bn , the above

inequality reads:
An Ak + `(Bn Bk ), n > k,
so dividing by Bn we get
An Ak `Bk
`+ . (5)
Bn Bn
Since Bn , by fixing k and taking lim sup in (5), we get lim supn (An /Bn ) `. In
other words, we obtained the inequality
a1 + + an
lim sup `, ` L,
n b1 + + bn
which in turn forces
a1 + + an
lim sup L.
n b1 + + bn

1
 Remark. An equivalent formulation of the above Theorem is as follows: If (yn ) n=1
is a strictly increasing sequence with limn yn = , then for any sequence (xn )
n=1 the
,
following inequalities hold:
xn xn xn1
lim sup lim sup ; (6)
n yn n yn yn1
xn xn xn1
lim inf lim inf . (7)
n yn n yn yn1

 
xn xn1
In particular, if the sequence has a limit, then
yn yn1 n=1
xn xn xn1
lim = lim .
n yn n yn yn1

Indeed (assuming all the yn s are positive, which happens anyway for n large enough), if we
consider the sequences (an )
n=1 and (bn )n=1 , defined by a1 = x1 , b1 = y1 , and an = xn xn1 ,
bn = yn yn1 , n 2, then everything is clear, since xn = a1 + + an and yn = b1 + + bn .

The Stolz-Cesaro Theorem has numerous applications in Calculus. Below are three of
the most significant ones.
Additive Cesaros Theorem. For any sequence (an ) n=1 R one has the inequali-
ties:
a1 + a2 + + an
lim sup lim sup an ; (8)
n n n
a1 + a2 + + an
lim inf lim inf an . (9)
n n n

In particular, if the sequence (an )

n=1 has a limit, then

a1 + a2 + + an
lim = lim an .
n n n

Proof. Particular case of Stolz-Cesaro Theorem with bn = 1.

Remark. An equivalent formulation of the above Theorem (proven using the alternative
version of Stolz-Cesaro Theorem) is as follows: For any sequence (xn )
n=1 , the following
inequalities hold:
xn
lim sup lim sup(xn xn1 ); (10)
n n n
xn
lim inf lim inf (xn xn1 ). (11)
n n n

In particular, if the sequence (xn xn1 )

n=1 has a limit, then

xn
lim = lim (xn xn1 ).
n n n

2
 Multiplicative Cesaros Theorem. For any sequence of positive numbers (an )
n=1
one has the inequalities:

lim sup n a1 a2 an lim sup an ; (12)
n n

lim inf n a1 a2 an lim inf an . (13)
n n

In particular, if the sequence (an )

n=1 has a limit, then

lim n a1 a2 an = lim an .
n n

 
b1 + + bn
Proof. Let bn = ln an , so that n a1 a2 . . . an = exp . Everything then follows
n
from the Additive Cesaro Theorem.
Remark. An equivalent formulation of the above Theorem (proven using the alternative
version of Additive Cesaro Theorem) is as follows: For any sequence of positive numbers
(xn )
n=1 , the following inequalities hold:
xn
lim sup n xn lim sup ; (14)
n n xn1
xn
lim inf n xn lim inf . (15)
n n xn1

In particular, if the sequence (xn /xn1 )

n=1 has a limit, then
xn
lim n xn = lim .
n n xn1

LHopitals Rule. Suppose f and g are differentiable on some interval that has A as
f 0 (x)
an accumualtion point, limxA g(x) = , and limxA 0 = L. Then:
g (x)
f (x)
lim = L.
xA g(x)

(Comment: The notation limxA can include any kind of limit: honest, one-sided, or
A = . It is also implicitly assumed that g 0 (x) 6= 0 for x near A. Nothing is assumed
about limxA f (x), not even its existence!)
Proof. By a simple change-of-variable argument, it suffices to consider only the case A = .
Since g 0 (x) 6= 0 on some interval (, ) and limx g(x) = , it follows that g is strictly
increasing on (, ). By a standard argument, it suffices to show that
(?) Whenever (tn )
n=1 is a strictly increasing sequence in (, ), with limn tn = , it
follows that:
f (tn )
lim = L. (16)
n g(tn )

3
 To prove (?) fix a sequence (tn ) n=1 as above, and let us consider the sequences xn = f (tn )
and yn = g(tn ). Using the Lagranges Mean Value Theorem, we know that for every n 2,
there exists some sn in (tn1 , tn ), such that

f (tn ) f (tn1 ) f 0 (sn )

= 0 . (17)
g(tn ) g(tn1 ) g (sn )

f 0 (sn )
Since (sn )
n=1 is obviously increasing, with limn sn = , it follows that limn =
g 0 (sn )
L. Going back to (17), we now have
xn xn1
lim = L,
n yn yn1

with (yn )
n=1 strictly increasing (because g is strictly increasing) and limn yn = . By
xn
the (alternative version of) Stolz-Cesaro Theorem, it follows that limn = L, which is
yn
precisely the desired conclusion (16).