484 Real Analysis
Now we turn to the geometric mean. Applying the Cesàro–Stolz theorem to the sequences
un=ln
P( 1 )
1 m
+ln
P( 2 )
2 m
+···+ln
P (n)
nm
andvn=n,n≥1, we obtain
lim
n→∞
un
vn
=lim
n→∞
ln
Gn
(n!)m/n
=lim
n→∞
ln
P (n)
nm
=lnam.
We therefore have
nlim→∞
An
Gn
·
(√n
n!
n
)m
=
1
m+ 1
.
Now we can simply invoke Stirling’s formula
n!≈nne−n
√
2 πn,
or we can argue as follows. If we letun=nnn!, then the Cesàro–Stolz theorem applied to
lnunandvn=nshows that ifunu+n^1 converges, then so doesn
√
un, and to the same limit.
Because
nlim→∞
un+ 1
un
=nlim→∞
(
n
n+ 1
)n
=
1
e
,
we have
lim
n→∞
√nn!
n
=
1
e
.
Therefore,
nlim→∞
An
Gn
=
em
m+ 1
.
(Gazeta Matematica ̆(Mathematics Gazette, Bucharest), 1937, proposed by T. Popo-
viciu)
347.Clearly,(an)n≥ 0 is an increasing sequence. Assume thatanis bounded. Then it must
have a limitL. Taking the limit of both sides of the equation, we have
lim
n→∞
an+ 1 = lim
n→∞
an+lim
n→∞
1
√kan,
orL=L+√k^1 L, contradiction. Thus limn→∞an=+∞and dividing the equation by
an, we get limn→∞ana+n^1 =1.