MATH 7770-01
LIMIT THEOREMS
FALL 2010
Lecture 35. Improvements in limit theorems with power βtailsβ, continued.
The integrals in the sums over 1 < πΌπ < 2 have been already evaluated in Lecture 14;
the summands in them are equal to
Ξ(2 β πΌπ ) [ +
πΌπ π
πΌπ π ] πΌπ
β
β
(ππΌπ + πβ
β π β
(π+
β£π‘β£
πΌπ ) cos
πΌπ β ππΌπ ) sign(π§) β
sin
πΌπ β 1
2
2
(35.1)
(see formula (14.12)); and it can be rewritten in exactly the form (34.11) if we understand
the function Ξ as being continued analytically to all complex numbers except non-positive
integers (see the text after formula (14.12)). So we can extend the ο¬rst sum in (34.20) to
all πΌπ between 0 and 2.
β«
β
β«
(πππ‘π₯ β 1 β ππ‘π₯) ππ΅ + (π₯) and
If 1 < π½ < 2, the last two integrals in (34.20) are
0
0
(πππ‘π₯ β 1 β ππ‘π₯) ππ΅ β (π₯), and we handle them by integrating by parts; say, for the
ββ
second one:
[ ππ‘π₯
]0
(π β 1 β ππ‘π₯) β
π΅β (π₯) ββ β ππ‘
β«
0
(πππ‘π₯ β 1) β
π΅β (π₯) ππ₯.
(35.2)
ββ
The term outside the integral is equal to 0 because limπ₯β0β β£(πππ‘π₯ β 1 β ππ‘π₯) β
π΅β (π₯)β£ β€
limπ₯β0β (π‘2 π₯2 /2)β
π(β£π₯β£βπ½ ) = 0, limπ₯βββ β£(πππ‘π₯ β1βππ‘π₯)β
π΅β (π₯)β£ β€ limπ₯βββ (2+β£π‘β£β£π₯β£)×
× π(β£π₯β£βπ½ ) = 0; and the absolute value of the integral does not exceed
[β«
β«
β£π‘β£β1
constβ
2β
β£π₯β£
ββ
βπ½
0
ππ₯+
β£π‘β£β£π₯β£β
β£π₯β£
βπ½
]
ππ₯ = constβ
β£π‘β£β1
(
2
1 ) π½β1
+
β
β£π‘β£
. (35.3)
π½β1 2βπ½
So we have:
π (π‘) β 1
=β
β
π: 0<πΌπ
[
πΌπ π
πΌπ π ] πΌπ
β
β
Ξ(1 β πΌπ ) β
(π+
β π β
(π+
β£π‘β£
πΌπ + ππΌπ ) cos
πΌπ β ππΌπ ) sign(π‘) β
sin
2
2
<2
+ ππ1 π‘ + π(β£π‘β£π½ )
(π‘ β 0)
(35.4)
(the term ππ1 π‘ is inο¬nitely small compared to all terms with πΌπ < 1, and inο¬nitely large
compare to those with πΌπ > 1, so perhaps we shouldnβt put all terms with 0 < πΌπ < 2
together, but write the terms with 0 < πΌπ < 1 ο¬rst, then ππ1 π‘, then the terms with
1 < πΌπ < 2, and ο¬nally π(β£π‘β£π½ ). But of course (35.4) looks shorter as it is written).
Problem 72 . Find the ο¬rst pseudo-moment for the distribution of Problem 70
with accuracy to 0.1.
So we can write an asymptotic expansion for π (π‘) for the distribution of Problem 70
(this way I am telling you that we can take π½ > 1 in this problem).
1
If π½ > 2, we subtract from the integrand in the last two integral in (34.20) and
subtract from it β π‘2 π₯2 /2; the second pseudo-moment π2 is deο¬ned as
β«
β
π2 =
[
π₯ π πΉ (π₯)β1+
2
0
β
βπΌπ
π+
πΌπ π₯
]
β«
0
+
ββ
π: 0<πΌπ <2
β
[
π₯2 π πΉ (π₯)β
βπΌπ
πβ
πΌπ β£π₯β£
]
(35.5)
π: 0<πΌπ <2
(the second pseudo-moment, in contrast with the second moment, neednβt be nonnegative).
We add to and subtract from πππ‘π₯ β1βππ‘π₯ in (34.20) the next Maclaurin term βπ‘2 π₯2/2:
π (π‘) β 1
=β
β
π: 0<πΌπ
[
πΌπ π
πΌπ π ] πΌπ
β
β
Ξ(1 β πΌπ ) β
(π+
β π β
(π+
β£π‘β£
πΌπ + ππΌπ ) cos
πΌπ β ππΌπ ) sign(π‘) β
sin
2
2
<2
+ ππ1 π‘ β π2 π‘2/2
β«
β
+
β
[
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2/2) π πΉ (π₯) β 1 +
0
βπΌπ
π+
πΌπ π₯
]
π: 0<πΌπ <2
β«
0
[
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2/2) π πΉ (π₯) β
+
β
]
βπΌπ
πβ
.
πΌπ β£π₯β£
(35.6)
π: 0<πΌπ <2
Now
we
add
to
and
subtract
from
the
expressions
in
the
brackets
the
sums
β
β
β
βπΌπ
+ βπΌπ
(the only thing we seem to be doing is adding
, π: 2<πΌπ <3 ππΌπ β£π₯β£
π: 2<πΌπ <3 ππΌπ π₯
and subtracting the same things again and again). We get instead of the last two integrals:
ββ
[
β
β«
π+
πΌπ
π: 2<πΌπ <3
β
0
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2/2) π(βπ₯βπΌπ )
β« 0
]
β
+ ππΌπ
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2/2) πβ£π₯β£βπΌπ
ββ
β«
β
+
β
[
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2/2) π πΉ (π₯) β 1 +
0
β«
βπΌπ
π+
πΌπ π₯
]
(35.7)
π: 0<πΌπ <3
0
+
[
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2/2) π πΉ (π₯) β
ββ
β
]
βπΌπ
πβ
.
πΌπ β£π₯β£
π: 0<πΌπ <3
Letβs evaluate the ο¬rst tow integrals. For 2 < πΌπ < 3, integrating by parts, we get:
β«
β
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2/2) π(βπ₯βπΌπ )
0
[
]β
= (πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2/2) β
(βπ₯βπΌπ ) 0 β ππ‘
β«
β
(πππ‘π₯ β 1 β ππ‘π₯) β
(βπ₯βπΌπ ) ππ₯.
0
(35.8)
1
π(β π₯1βπΌπ ),
The term outside the integral is equal to 0. Rewriting (βπ₯βπΌπ ) ππ₯ as
πΌπ β 1
we integrate by parts another time (again with the term outside the integral being equal
2
to 0), and get:
β« β
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2/2) π(βπ₯βπΌπ ) = β
0
2
π‘
=β
(1 β πΌπ )(2 β πΌπ )
β«
π‘2
πΌπ β 1
β
(π
ππ‘π₯
β«
β
(πππ‘π₯ β 1) β
π₯1βπΌπ ππ₯
0
(35.9)
β 1) π(βπ₯
2βπΌπ
).
0
Since 0 < πΌπ β 2 < 1, the integral here is of the kind that we have considered in Lecture
14, and it is equal
(
)
βΞ 1 β (πΌπ β 2) β
πβπ(πΌπ β2)β
sign(π‘)β
π/2 β£π‘β£πΌπ β2 .
(35.10)
Multiplying by the factor in front of the integral in (35.9) and taking into account that
Ξ(3 β πΌπ )
β1 = πβπβ
2β
sign(π‘)β
π/2 and
= Ξ(1 β πΌπ ) (with the analytically-continued
(2 β πΌπ )(1 β πΌπ )
Ξ-function), we get that the π-th summand in the ο¬rst sum in (35.7) is equal to
[
πΌπ π
πΌπ π ] πΌπ
β
β
β π β
(π+
β£π‘β£ ,
βΞ(1 β πΌπ ) β
(π+
πΌπ + ππΌπ ) cos
πΌπ β ππΌπ ) sign(π‘) β
sin
2
2
(35.11)
that is, it is expressed by exactly the same formula as for 0 < πΌπ < 1 or 1 < πΌπ < 2.
So we get:
π (π‘) β 1
=β
β
π: 0<πΌπ
[
πΌπ π ] πΌπ
πΌπ π
β
β
β π β
(π+
β£π‘β£
Ξ(1 β πΌπ ) β
(π+
πΌπ β ππΌπ ) sign(π‘) β
sin
πΌπ + ππΌπ ) cos
2
2
<3
+ ππ1 π‘ β π2 π‘2 /2
β«
β
+
β
[
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2 /2) π πΉ (π₯) β 1 +
0
β«
βπΌπ
π+
πΌπ π₯
]
π: 0<πΌπ <3
0
+
β
[
(πππ‘π₯ β 1 β ππ‘π₯ + π‘2 π₯2 /2) π πΉ (π₯) β
]
βπΌπ
πβ
β£π₯β£
.
πΌπ
(35.12)
Let us formulate what we are getting this way (adding and subtracting the same
things, expressing integrals through the Ξ-function, and from time to time saying βif π½
is so and soβ and estimating the remainder term) as a lemma (not a theorem, because it
is not a result about the limiting behavior of the distribution of the normalized sum with
improvements).
Lemma 35.1. Let the distribution function satisfy the conditions (34.3), (34.4) with
non-integer πΌπ , π½; let conditions (34.9) be satisο¬ed if π½ < 1. Then
ββ
π: 0<πΌπ <3
π (π‘) β 1
β
[
πΌπ π
πΌπ π ] πΌπ
β
β
=β
Ξ(1 β πΌπ ) β
(π+
β π β
(π+
β£π‘β£
πΌπ + ππΌπ ) cos
πΌπ β ππΌπ ) sign(π‘) β
sin
2
2
(35.13)
π
+
βπ½β
β
ππ ππ π‘π /π ! + π(β£π‘β£π½ )
π =1
3
(the ο¬rst sum is taken over all π).
If the conditions (34.1), (34.2) are satisο¬ed, and condition (34.8) for π½ < 1, then the
asymptotic expansion (35.13) holds with π(β£π‘β£π½ ) replaced with π(β£π‘β£π½ ) as π‘ β 0.
The proof has in fact been carried out for the remainder term π(β£π‘β£π½ ); the only thing
that remained was to say βetc.β The statement with π(β£π‘β£π½ ) is proved in the same way,
only ±β£π‘β£β1 as( the limits
of integration are replaced with some functions ± π’(π‘) = π(β£π‘β£β1 ),
)
such that π΅± ± π’(π‘) (or the corresponding total variations on the interval from π’(π‘) to β
or from β β to β π’(π‘)) are π(β£π‘β£βπ½ ) as π‘ β 0.
We know what we are doing after this: we write the asymptotic expansion for ln π (π‘)
as π‘ β 0 using the Maclaurin expansion ln π€ = (π€ β 1) β (π€ β 1)2 /2 + (π€ β 1)3 /3 β ... ;
( π§ )
after this, the expansion in (negative) powers of π of π ln π
, and after that, one for
( π§ )π
{
( π§ )}
π΅π
πππ (π§) = π
= exp π ln π
.
π΅π
π΅π
4
© Copyright 2026 Paperzz