Chern-Ching Chao
Institute of Statistical Science, Academia Sinica, Taipei 115, Taiwan,
Republic of China
1. INTRODUCTION
Martingale limit theory is well known for its variety of applications, from mathe-
matics to statistics, where classical independence-based methods are not available.
It has recently been applied with success to combinatorial problems, such as
random graphs see, e.g., w1, 12x. and random trees see, e.g., w13]15x..
The key idea when applying martingale limit theory is to reduce a sequence of
random variables to a martingale difference sequence or array to which one can
apply the martingale limit theorem. Mahmoud, Smythe, and Szymanski have
successfully employed this approach to obtain striking results for recursive trees
and plane-oriented recursive trees w13]15x. The particular sequences X n4 which
they considered share the common property that Ew X n y X ny1 < Fny1 x is a linear
323
324 CHAO
2. MAIN RESULT
and
kn
p
Zn2, k 1 2.
ks1
d
imply kks 1 Z n, k N 0, 1 , the standard normal distribution.
n .
Our main result is presented as follows.
Suppose that X n ; n G 14 is a sequence of random variables on the probability
space V, F , P . and Fn ; n G 04 is an increasing sequence of sub-s-fields of F , such
that X n if Fn-measurable and Ew X n2 x F ` for each n. Denote D X n s X n y X ny1
D X 1 s X 1 . and Dn s D X n y Ew D X n N Fny1 x. Notation h n 7 k n means h n s O k n .
and k n s O h n .. Restricting to the case that Ew D X n N Fny1 x is a linear function of
X ny 1 , we have the following theorem.
E w D X n N Fny1 x s a n q bn X ny1 , 3.
APPLICATIONS OF MARTINGALE TO RANDOM PERMUTATIONS 325
where
n
bn , n s 1, bn , k s 1 q bj . , k s 1, . . . , n y 1,
jskq1
n
mn ' E w Xn x s bn , k a k ,
ks1
n
sn2 ' Var w X n x s bn2, k E D k2 ,
ks1
2
E Dn2 N Fny1 s c n q Yny1 y e n X ny1 y m ny1 . ,
lim inf c n ) 0, 4.
1
en s o / , 5.
n
n
bn2, k c k 7 ncnq1 , 6.
ks1
and
Xn y mn d
N 0, 1 . . 9.
sn
326 CHAO
Proof.
a. Since
X n s X ny1 q E w D X n N Fny1 x q Dn
s a n q Dn q 1 q bn . X ny1
n
s bn , k a k q D k .
ks1
Hence,
n n
sn2 s bn2, k c k y e k s ky1
2
.s bn2, k c k q o c k . . .
ks1 ks1
For any given ) 0, fix n 0 such that c k ) 0 and < o c k .< F 2 c k for all k ) n 0 . Then,
for n ) n 0 ,
n
bn2, k o c k .
ks1
n0
n n0
F bn2, k o c k . q bn2, k c k y bn2, k c k
ks1 2
ks1 ks1
/
o ck . n0
n
F M 2 max q < ck < q bn2, k c k .
kFn 0 ck 2 / ks1 2
ks1
Observe that nks 1 bn,2 k c k ` due to 4. and 6.. Thus, find n1 ) n 0 such that
n 2 o ck . n0
bn2, k c k ) q < ck <
ks1
M 2 max
kFn 0 ck 2 /
ks1
and
n p
WnU ' sny2 bn2, k E D k2 N Fky1 y E D k2 . 0. 11 .
ks1
Since
E Dk2 y E D k2 N Fky1 . Dh2 y E D h2 N Fhy1 . s0
for k / h, and noting that assumption III. and 8. imply ess sup D k2 s o ny1 r2sn2 .,
we have
n
2
E Wn2 s sny4 bn4, k E Dk2 y E D k2 N Fky1 . F sny4 n o ny1sn4 . . 0,
ks1
Since sny2 nks1 bn,2 k Ew < Yky1 <x 0, by 7., and sny2 nks1 bn,2 k e k s ky1
2
0, 11. is
proved.
3. APPLICATIONS
P n v . s pn 1 . , . . . , pn n . . if v g A p n ,
328 CHAO
where I ?4 is the indicator function. Inserting n into the given P ny 1 , one has
1
P D X n s k N Fny1 4 s for k s 0, . . . , n y 1.
n
Hence,
E w D X n N Fny1 x s 12 n y 1 . ,
E Dn2 N Fny1 s 121 n2 y 1 . ,
bn , k s 1, k s 1, . . . , n,
m n s 14 n n y 1 . ,
and
sn2 s 721 2 n3 q 3n2 y 5n . .
The same result by different derivations can be found in w2, 9x and w8, pp.
256]257x.
where P n, nq1 s P n, 1. For n G 2 and given P ny1 , inserting n between P ny1, i and
P ny 1, iq1 , for some i, one gets D X n s 0 if P ny1, i - P ny1, iq1 , and D X n s 1
otherwise. Therefore,
X ny 1
P D X n s 0 N Fny1 4 s 1 y P D X n s 1 N Fny1 4 s , n G 2.
ny1
APPLICATIONS OF MARTINGALE TO RANDOM PERMUTATIONS 329
Hence, for n G 2,
1
E w D X n N Fny1 x s 1 y X ny1 ,
ny1
ky1 n
bn , k s , k s 1, . . . , n, mn s ,
ny1 2
1 1 2
E Dn2 N Fny1 s y 2 X ny1 y m ny1 . ,
4 n y 1.
and
n
sn2 s q O 1. .
12
Related results, mostly under the name of Eulerian number, can be found in
w2, 4, 7, 10, 17x and w6, pp. 150]154x.
2
P D X n s 0 N Fny1 4 s 1 y P D X n s 1 N Fny1 4 s X ny1 ,
ny1
2
E w D X n N Fny1 x s 1 y X ny1 ,
ny1
k y 1. k y 2. n
bn , k s , k s 1, . . . , n, mn s ,
n y 1. n y 2. 3
and
2 2 4 2
E Dn2 N Fny1 s y X ny1 y m ny1 . y X ny1 y m ny1 . .
9 3 n y 1. n y 1.
2
Hence,
2
sn2 s n q O 1. ,
45
d
and X n y m n .rsn N 0, 1. since ny1 Ew < X ny1 y m ny1 <x s o1..
Related results are given in w3x without proof, and w18x.
330 CHAO
1 ny1
E D X n s . N Fny1 s 2 n s y P ny1,
s < s <
i y P ny1, iq1 y P ny1, i y P ny1, iq1 4
s s
ny1 is1
2s 1
s ns q o ns . y X ny 1 s . .
sq1 ny1
Therefore,
ky1
bn , k s , k s 3, . . . , n,
ny1
2s
mn s . s n sq 1 q o n sq1 . .
s q 1. s q 2.
Also,
4 s2
E Dn2 N Fny1 s 2
n2 s q o n2 s .
s q 1. s q 2.
8
y n sy 1 q o n sy1 . / X ny 1 s . y m ny1 s . .
sq2
2
q X ny 1 2 s . y m ny1 2 s . .
ny1
1 2
y 2 X ny 1 s . y m ny1 s . . .
n y 1.
Hence,
4 s2
sn2 s. s 2
n2 sq1 q o n2 sq1 . ,
s q 1. s q 2. 2 s q 3.
d
and X n s . y m n s ..rsn s . N 0, 1. since
and
sqtq2 1
s n sqt q o n sqt . y X ny 1 s, t . .
s q 1. t q 1. ny1
Therefore,
ky1
bn , k s , k s 2, . . . , n,
ny1
1
m n s, t . s n sqtq1 q o n sqtq1 . .
s q 1. t q 1.
Also,
E Dn2 s, t . N Fny1
s 2 t 2 2 s q 2 t q 3.
s 2 2
n2 sq2 t q o n2 sq2 t .
s q 1. t q 1. 2 s q 1. 2 t q 1.
1
q X ny 1 2 s, 2 t . y m ny1 2 s, 2 t . .
n y 1.
2 st q 2 s q 2 t .
q n sqty1 q o n sqty1 . X ny 1 s, t . y m ny1 s, t . .
s q 1. t q 1. /
y 2 n sy 1 q o n sy1 . . X ny 1 s, 2 t . y m ny1 s, 2 t . .
y 2 n ty1 q o n ty1 . . X ny 1 2 s, t . y m ny1 2 s, t . .
1 2
y 2 X ny 1 s, t . y m ny1 s, t . . .
n y 1.
Hence,
s2 t 2
sn2 s, t . s 2 2
n2 sq2 tq1 q o n2 sq2 tq1 . ,
s q 1. t q 1. 2 s q 1. 2 t q 1.
d
and X n s, t . y m n s, t ..rsn s, t . N 0, 1. since
and
n sy 1 E < X ny 1 s, 2 t . y m ny1 s, 2 t . < s o n2 sq2 t . .
REFERENCES