Example 1 ([24]). Put xn=pn, the nth prime and denote Xn=
2 pn, 3
pn, . . . ,pn−1 pn ,pn
pn
.
The sequence of blocks Xn is u.d. and therefore the ratio sequence pm/pn, m= 1,2, . . . , n,n= 1,2, . . . is u.d. in [0,1].This generalizes a result of A. S c h i n -z e l (cf. W. S i e r p i ´n s k i (1964, p. 155)). Note that from u.d. ofXn applying for theL2 discrepancy ofXn we get the following interesting limit
n→∞lim 1 n2pn
n i,j=1
|pi−pj|= 1 3.
Example 2 ([24, Ex. 11.1]). Let γ, δ, and a be given real numbers satisfying 1≤γ < δ≤a. Letxnbe an increasing sequence of all integer points lying in the intervals
(γ, δ),(γa, δa), . . . ,(γak, δak), . . . ThenG(Xn) =
gt(x);t∈[0,1]
, wheregt(x) has constant values gt(x) = 1
ai(1 +t(a−1)) for x∈ (δ, aγ)
ai+1(tδ+ (1−t)γ), i= 0,1,2, . . . and on the component intervals it has a constant derivative
gt(x) = tδ+ (1−t)γ
(δ−γ)(a−11 +t) for x∈ (γ, δ)
ai+1(tδ+ (1−t)γ), i= 0,1,2, . . . and x∈
γ
tδ+ (1−t)γ,1
, where
F(Xnk, x)→gt(x) fornk for which xnk =
akγ+tak(δ−γ)
. (53) Here we write (xz, yz) = (x, y)z and (x/z, y/z) = (x, y)/z. Then the set G(Xn) has the following properties:
10. Everyg ∈G(Xn) is continuous.
20. Every g ∈ G(Xn) has infinitely many intervals with constant values, i.e., with g(x) = 0, and in the infinitely many complement intervals it has a constant derivative g(x) = c, where 1
d ≤ c ≤ d1 and for lower d and upper dasymptotic density ofxn we have
d= (δ−γ)
γ(a−1), d= (δ−γ)a δ(a−1). 30. The graph of everyg∈G(Xn) lies in the intervals
1 a,1
× 1
a,1
∪ 1
a2,1 a
× 1
a2,1 a
∪. . . Moreover, the graph g in 1
ak,ak−11
×1
ak,ak−11
is similar to the graph of g in 1
ak+1,a1k
× 1
ak+1,a1k
with coefficient 1a. Using the parametric expression, it can be written for all x∈ 1
ai+1,a1i
that gt(x) = gt(aaiix), i= 0,1,2, . . .
40. G(Xn) is connected and the upper distribution function g(x) = g0(x) ∈ G(Xn) and the lower distribution function g(x) ∈/ G(Xn). The graph ofg(x) on1
a,1
×1
a,1
coincides with the graph of y(x) =
1 +1
d 1
x −1 −1
onγ
δ,1
, further, on1
a,γδ
we haveg(x) =a1. 50. G(Xn) = gg0(xβ)
0(β);β∈1
a,aγδ ! . For the proofs of 10.−50. we only note:
Assume that xn ∈ ak(γ, δ), i, i+ 1, i+ 2,· · · ∈aj(γ, δ) for somej < k, and let F(Xn, x) →g(x) for some sequence of n. Then g(x) has a constant derivative in the intervals containing xin,i+1xn,i+2xn, . . ., since
1 n i+1
xn −xin = xn n , and thus xnn must be convergent tog(x), so 1
d ≤g(x)≤ 1d. For xn=
takδ+ (1−t)akγ we can find
g(x) = lim
n→∞
xn n = lim
k→∞
ak(tδ+ (1−t)γ)
#k−1
j=0aj(δ−γ) +ak(tδ+ (1−t)γ)−akγ
= tδ+ (1−t)γ (δ−γ) 1
a−1+t.
Using Theorem 18 and [3, Ex. 3] we shall add the following properties more-over:
60. By definition (5) of the local asymptotic densitydg and by (53) forg(x) = gt(x) we have
dgt = lim
k→∞
nk
xnk = lim
k→∞
#k−1
i=0 ai(δ−γ) +tak(δ−γ) akγ+tak(δ−γ)
= (δ−γ)(1 +t(a−1))
(a−1)(γ+t(δ−γ)) (54)
and for t= 0 we havedg0 =dand for t= 1 we havedg1 =dand we see gt(x) = 1
dgt
(55) forxwith the constant derivative ofgt(x).
70. For the function h1,g(x) defined in (26), putting g(x) =gt(x), we have:
d
dgt = γ+t(δ−γ)
γ(1 +t(a−1)),1−dgt
1−d = γ
γ+t(δ−γ), d
dgt
1−dgt
1−d = 1
1 +t(a−1).
Then
h1,gt(x) =
xγ(1+t(a−1))γ+t(δ−γ) forx∈
0,γ+t(δ−γ)γ , xd1
gt + 1− d1gt, forx∈ γ
γ+t(δ−γ),1
, (56)
see the following figure.
,, ,, ,, ,, ,, ,, ,, ,
gt(x)
h1,gt(x)
(1,1)
(1a,1a)
(a12,a12)
Figure:gt(x) andh1,gt(x).
80. In the proof of the upper bound (29) we have proved that 1−1
0 h1,g(x) dxis maximal fordg = min(√
d, d). Lett0∈[0,1] be such thatdgt0 = min(√ d, d) andt0 can be computed by inverse formula to (54)
t= dgt(a−1)γ−(δ−γ)
(δ−γ)(a−1)(1−dgt). (57) 90. LetP(t) be the area in1
a,1
×1
a,1
bounded by the graph ofgt(x). Then 1
0
gt(x) dx=P(t) 1 1−a12
+ 1
a+ 1
= 1 2+ 1
2. 1
(a+ 1). (γa−δ)
(1 +t(a−1))(γ+t(δ−γ)) + 1
2. t(δ−γa)
(1 +t(a−1))(γ+t(δ−γ)) (58) and sinceg0(x) =g(x) we have that the maxt∈[0,1]1
0 gt(x) dx is attained att= 0. Using derivative ofP(t) it can be see that the mint∈[0,1]1
0 gt(x) dx
is attained at t= 1. It also follows from the fact that for xn+1 = xn+ 1 we have
1 n+ 1
n+1
i=1
xi
xn+1 − 1 n
n i=1
xi
xn
= 1
n+ 1− 1
xn+ 1+ 1 n+ 1. 1
1 +x1n 1
n n i=1
xi xn
>0 becausec1(x)∈/G(Xn) and thus lim supn→∞n1#n
i=1 xi
xn <1. Now, denot-ing the indexnk forxnk = [akδ], the lim sup of 1n#n
i=1 xi
xn is attained over n = nk, k = 0,1,2, . . . and for such nk we have F(Xnk, x) → g1(x) for x∈[0,1].
100. Thus we have lim inf
n→∞
1 n
n i=1
xi xn = 1−
1 0
g0(x) dx= 1 2− 1
2. 1 (a+ 1)
γa−δ γ
, (59)
lim sup
n→∞
1 n
n i=1
xi xn = 1−
1 0
g1(x) dx= 1 2+ 1
2. 1 (a+ 1)
γa−δ δ
. (60)
The upper bound (29) coincides with the maximal value of 1−1
0 h1,g(x) dx attained for dg = min(√
d, d). Since 1−1
0 g1(x) dx is maximal for all 1−1
0 gt(x) dx, t∈[0,1] and 1−1
0 g1(x) dx≤1−1
0 h1,g1(x) dxthen the upper bound (60) satisfies (29).
110. Using explicit formulas
d= (δ−γ)
γ(a−1), d= (δ−γ)a
δ(a−1) (61)
for asymptotic densities we see again that (59) and (60) satisfy (28) and (29), respectively, in Theorem 19.
Example 3 ( [9, Ex. 2]). Let xn and yn, n = 1,2, . . ., be two strictly in-creasing sequences of positive integers such that for the related block sequences Xn=x1
xn, . . . ,xxn
n
andYn=y1
yn, . . . ,yyn
n
, we have singleton for bothG(Xn) = g1(x)
and G(Yn) = g2(x)
. Furthermore, let nk, k = 1,2, . . ., be an in-creasing sequence of positive integers such thatNk=#k
i=1ni satisfies Nnk
k →1.
Denote by zn the following increasing sequence of positive integers composed by blocks (here we use the notationa(b, c, d, . . .) = (ab, ac, ad, . . .))
(x1, . . . , xn1), xn1(y1, . . . , yn2), xn1yn2(x1, . . . , xn3), xn1yn2xn3(y1, . . . , yn4), . . .
Then the sequence of blocks Zn=z1
zn, . . . ,zznn
has the set of d.f.s G(Zn) =
g1(x), g2(x), c0(x)
∪
g1(xyn);n= 1,2, . . .
∪
g2(xxn);n= 1,2, . . .
∪
$ 1
1 +αc0(x) + α
1 +αg1(x);α∈[0,∞)
%
∪
$ 1
1 +αc0(x) + α
1 +αg2(x);α∈[0,∞)
% , whereg1(xyn) = 1 if xyn≥1, similarly forg2(xxn).
P r o o f. For everyn= 1,2, . . . there exists an integerk such that Nk−1 < n≤Nk
(hereN0= 0). Put n=n−Nk−1. For everyn we have zn=
xn1yn2. . . xnk−1yn if kis even, xn1yn2. . . ynk−1xn if kis odd.
Firstly we assume thatk is even. ThenZnhas the form Zn=
. . . ,xn1yn2. . . ynk−2(x1, . . . , xnk−1)
xn1yn2. . . xnk−1yn ,xn1yn2. . . xnk−1(y1, . . . , yn) xn1yn2. . . xnk−1yn
=
. . . , 1 xnk−1yn
y1
ynk−2, . . . ,ynk−2 ynk−2
, 1
yn
x1
xnk−1, . . . ,xnk−1 xnk−1
,
y1
yn, . . . ,yn
yn
and thus forx > x 1
nk−1 we have
F(Zn, x) =Nk−2+nk−1F(Xnk−1, xyn) +nF(Yn, x) Nk−1+n
= Nk−2 Nk−1+n +
nk−1
Nk−1
1 +Nn
k−1
F(Xnk−1, xyn) + 1
1 +Nk−1n F(Yn, x).
Ifn→ ∞, then the first term tends to zero. IfF(Zn, x)→g(x) for some sequence ofn, we can select a subsequence ofn’s such that Nn
k−1 →αfor someα∈[0,∞), or Nn
k−1 → ∞. For suchn we distinguish the following cases:
(a) If n= constant, then
nk−1
Nk−1
1 +Nn
k−1
F(Xnk−1, xyn)→g1(xyn)( hereg1(xyn) = 1 forxyn >1) 1
1 +Nk−1n F(Yn, x)→0
and thusF(Zn, x)→g1(xyn).
(b) If n→ ∞, then F(Xnk−1, xyn)→1; precisely F(Xnk−1, xyn)→c0(x).
(b1) If Nn
k−1 →0, thenF(Zn, x)→c0(x).
(b2) If Nn
k−1 →α∈(0,∞), then F(Zn, x)→ 1+α1 c0(x) +1+αα g2(x).
(b3) If Nn
k−1 → ∞, then F(Zn, x)→0 +g2(x).
Fork-odd we use a similar computation.
Now, identify xn = yn and select xn such that g1(x) = x (e.g., xn = n or xn =pn, thenth prime) and putnk = 2k2 fork = 1,2, . . .. Then the set of all d.f.s
G(Zn) =
g1(x), c0(x)
∪
g1(xxn);n= 1,2, . . .
∪
$ 1
1 +αc0(x) + α
1 +αg1(x);α∈[0,∞)
%
is disconnected, as it can be seen in the figure on the page 174.
Example 4. Letxn,n= 1,2, . . ., be an increasing sequence of positive integers for which there exists a sequencenk, k= 1,2, . . ., of positive integers such that (ask→ ∞)
(i) nnk−1
k →0, (ii) xnk
nk →0, (iii) xnk−1x
nk →0, and
(iv) xnk−i =xnk−i fori= 0,1, . . . , nk−nk−1−1.
Then the sequence of blocks Xn=
x1 xn, x2
xn, . . . ,xn
xn
has
G(Xn) =
hα(x);α∈[0,1]
.
P r o o f. For givenθ∈[0,1] andn=nk−
θ(nk−nk−1)
and by (iv) we have xn=xnk−
θ(nk−nk−1) .
Fori≤nwe distinguish two cases:xi ∈(xnk−1, xn] andxi ≤xnk−1. (I) Forxi ∈(xnk−1, xn] we have
xi
xn ∈
&
xnk−(nk−nk−1) + 1 xnk−[θ(nk−nk−1)] ,1
'
→[1,1]
asn→ ∞and for anyθ∈[0,1]. The number of suchxi’s is (nk−nk−1)−
θ(nk−nk−1)
= (1−θ)(nk−nk−1) +O(1).
(II) For xi≤xnk−1 we have xi
xn ∈
&
0, xnk−1
xnk−[θ(nk−nk−1)]
'
→[0,0].
We thus get, for anyx∈(0,1) and any sufficiently largen, F(Xn, x) = nk−1
n = nk−1
nk−1+ (1−θ)(nk−nk−1) +O(1). This gives:
(a) If θ≤ε0<1, for some fixedε0, then F(Xn, x)→c1(x).
(b) If θ= 1, then
F(Xn, x)→c0(x).
(c) For anyα∈(0,1) there exists a sequence θk→1, ask→ ∞, such that nk−1
nk−1+ (1−θk)(nk−nk−1) →α, and in this case
F(Xn, x)→hα(x).
Note that the sequences nk= 2k2 and xnk = 2(k+1)2 satisfy the assumptions (i), (ii), (iii) and (iv). We also see thatG(Xn) is connected but
F(Xnk+1, x)→c0(x), and F(Xnk, x)→c1(x),
a.e. on [0,1] and thusρ(tnk+1, tnk)→1. Using the permutationπ:N→N 1,2, . . . , n1, n2, n2−1, n2−2, . . . , n1+ 1, n2+ 1, n2+ 2, . . . n3, n4, n4−1, n4−2, . . . , n3+ 1, n4+ 1, n4+ 2, . . . , n5, n6, n6−1, n6−2, . . . , n5+ 1, . . . we haveρ(tπ(n+1), tπ(n))→0 asn→ ∞, because the “neighbouring” d.f. oftπ(n) satisfies the scheme
c1(x), c1(x), . . . , c0(x), c0(x), . . . , c1(x), c1(x), . . . , c0(x), c0(x), . . . , c1(x), c1(x), . . . , c0(x), c0(x), . . .
Example 5. In [8] is proved that xxn
n+1 → 1 does not imply that G(Xn) is a singleton. This is a negative answer to the Problem 1.9.2 in [20].
Let ak, nk, k = 1,2, . . ., and xn, n = 1,2, . . . be three increasing integer sequences andh1< h2 be two positive integers. Assume that
(i) nnk
k+1 →0 fork→ ∞; (ii) nak
k+1 →0 fork→ ∞; (iii) for oddk we have
ahk2 ≤xnk = (ak−1+nk−nk−1)h1 ≤(ak+ 1)h2 and xi= (ak+i−nk)h2 fornk< i≤nk+1;
(iv) for even k we have
ahk1 ≤xnk = (ak−1+nk−nk−1)h2 ≤(ak+ 1)h1 and xi= (ak+i−nk)h1 fornk< i≤nk+1.
Then xxn
n+1 →1 and the setG(Xn) of all distribution functions of the sequence of blocks XnisG(Xn) =G1∪G2∪G3∪G4, where
G1 =
xh12.t;t∈[0,1]
, G2 =
xh12(1−t) +t;t∈[0,1]
, G3 =
max(0, xh11 −(1−xh11)u);u∈[0,∞) and G4 =
min(1, xh11.v);v∈[1,∞) .
In [24, Th. 5.2, p. 762 ] = Theorem 15, it is proved that the condition xxn
n+1 →1 implies the connectivity ofG(Xn)
P r o o f. 1.Firstly we prove that for anyh1 < h2 the sequencesak, nk, xn satis-fying (i)–(iv) exist:
Fori= 1, . . . , n1 we putxi=ih1 and then we find a1 such that ah12 ≤xn1 ≤ (a1+ 1)h2. If we have selected, for an odd step k, allai, i= 1,2, . . . , k−1, xi, i = 1,2, . . . , nk, then we find ak such that ahk2 ≤ xnk < (ak+ 1)h2, and then we putxi= (ak+i−nk)h2 fornk < i≤nk+1, where we choosenk+1sufficiently large to satisfy the limits (i) and (ii). For an even step k we proceed similarly replacingh2 byh1.
2.In contrary to the independence of ak and nk+1 we have ak
n
h1 h2
k
→1 for oddk → ∞, ak n
h2 h1
k
→1 for evenk→ ∞. (62) This follows from (iii) and (iv), directly, e.g., from (iii) we have
ahk2 nhk1 <
ak−1 nk
+ 1−nk−1 nk
h1
< (ak+ 1)h2 nhk1 . As an application of (62) we have
ak
nk →0 for odd k→ ∞, ak
nk → ∞ for evenk→ ∞. (63) 3. Now we prove xxi
i+1 →1 as i → ∞. Let i ∈ (nk, nk+1) and let, e.g., k be odd. Then by (iii)
xi xi+1 =
1− 1
ak+i+ 1−nk h2
>
1− 1
ak h2
and for i=nk again xnk
xnk+1 > ahk2 (ak+ 1)h2 >
1− 1
ak
h2
which implies the limit 1 as oddk→ ∞. Similarly for evenk.
4.LetN ∈[nk, nk+1] be an integer sequence (we shall omit the index inNk) fork→ ∞. Forx∈(0,1) we have
F(XN, x) = #{1≤i≤nk−1;xxNi < x} N
+ #{nk−1< i≤nk;xxi
N < x}
N + #{nk< i≤N;xxi
N < x} N
=o(1) + A N + B
N. (64)
To compute NA for oddk we use xi
xN = (ak−1+i−nk−1)h1
(ak+N−nk)h2 < x⇐⇒i−nk−1 < xh11(ak+N−nk)
h2
h1 −ak−1
and we have A
N =min(nk−nk−1,max(0,[xh11(ak+N−nk)
h2
h1 −ak−1]))
N . (65)
Similarly, for even k A
N =min(nk−nk−1,max(0,[xh12(ak+N−nk)
h1
h2 −ak−1]))
N . (66)
For NB and odd k we use xi
xN
=
ak+i−nk
ak+N−nk
h2
< x⇐⇒i−nk< xh12(ak+N−nk)−ak which gives
B
N = min(N−nk,max(0,[xh12(ak+N−nk)−ak]))
N . (67)
Similarly, for even k we have B
N = min(N−nk,max(0,[xh11(ak+N−nk)−ak]))
N . (68)
In the following we will distinguish three cases nk
N →t >0, nk
N →0 and N
nk+1 →0, and N
nk+1 →t >0.
5.Now, let nNk →t >0 as k→ ∞.
a) Assume that k is odd and compute the limit of NA by (65). We have
nk−nk−1
N →tand ift <1 we see xh11
ak
N
h1 h2
+ N
N
h1 h2
1− nk
N hh21
−ak−1
N → ∞
since N
N
h1 h2
forh1< h2 is unbounded and by (62) ak
N
h1 h2
= ak n
h1 h2
k
nk N
hh1
2 →t
h1 h2
is bounded. Thus, for 0< t <1, we have A
N →t for odd k→ ∞. (69)
a1) Let for the momentt= 1. We have ak
n
h1 h2 k
→1 and
xh11
ak N
h1 h2
+ N−nk N
h1 h2
hh2
1
− ak−1
N →xh11(1 +u)
h2 h1
assuming the limit N−nk
N
h1 h2
→u, whereu∈[0,∞) can be arbitrary. Putv= (1+u)
h2 h1. Thus fort= 1 and corresponding v∈[1,∞) we have
A
N →min(1, xh11v) for odd k→ ∞. (70) If N−nk
N
h1 h2
→ ∞, then
A
N →1 for odd k→ ∞. (71)
b) Now, again 0< t≤1. For even k in (66) we have xh12
ak N
h2 h1
+ N
N
h2 h1
1− nk
N hh12
− ak−1
N →xh12.t since by (62)
ak N
h2 h1
= ak n
h2 h1
k
nk N
hh2
1→t
h2 h1. Thus
A
N →xh12.t for even k→ ∞. (72)
c) For the limit BN as odd k → ∞we compute (67) by using N−nN k →1−t and
xh12 ak
N + 1− nk N
−ak
N →xh12(1−t) since by (63) we have aNk = nakknNk →0. Thus
B
N →xh12(1−t) for odd k→ ∞. (73) d) Again by (63), for evenkwe have aNk = ankknNk → ∞, then (assumingx <1)
xh11 ak
N + 1−nk N
− ak
N → −∞. Thus
B
N →0 for even k→ ∞. (74)
e) Summing up (69), (72), (73) and (74) we find, for everyx∈(0,1), F(XN, x)→
xh12(1−t) +t for oddk→ ∞,
xh12.t for evenk→ ∞ (75)
for nNk →t, 0< t <1. For nNk →t= 1, N−nk
N
h1 h2
→uand v= (1 +u)
h2
h1 we have applying (70)
F(XN, x)→min(1, xh11.v) for odd k→ ∞, (76) and for N−nk
N
h1 h2
→ ∞we have
F(XN, x)→c0(x) for odd k→ ∞, (77) wherec0(x) = 1 for x∈(0,1).
6. In the case nNk →0 and nN
k+1 →0 we have NA =o(1) and then it suffices to compute the limit NB by (67) or (68).
a) Assume that odd k → ∞. Since N−Nnk → 1 and by (63) we have aNk =
ak
nk
nk
N →0 and thus xh12
ak
N + 1−nk N
−ak
N →xh12. (78)
b) Assume that evenk→ ∞. In this case (by (62) and (ii)) we have ak
N = ak
n
h2 h1
k
n
h2 h1
k
N , ak
n
h2 h1
k
→1, ak
nk+1 →0, then n
h2 h1
k
nk+1 →0.
Thus, for anyu∈[0,∞) we can find a subsequence ofN such that n
h2 h1
k
N →u. (79)
Then
xh11 ak
N + 1−nk N
− ak
N →xh11 −(1−xh11)u. (80) c) Summing up (78) and (80) we find for everyx∈(0,1)
F(XN, x)→
xh12 for oddk → ∞,
max(0, xh11 −(1−xh11)u) for evenk→ ∞ (81) for nNk →0, nN
k+1 →0 and foru∈(0,∞) satisfying (79) ifkis even. If n
h2 h1 k
N → ∞
then
F(XN, x)→c1(x) for even k→ ∞, (82) wherec1(x) = 0 for x∈(0,1).
7.Finally, let nN
k+1 →t >0. Then aNk →0, because (ii) nak
k+1 →0. Computing the limit BN by (67) or (68) we find
F(NN, x)→
xh12 for oddk→ ∞,
xh11 for evenk→ ∞. (83) 8.Now, assume that F(XN, x)→g(x) for some sequence of N ∈[nk, nk+1], i.e., g(x)∈G(Xn). Then we can find subsequence of N (denoting again asN) such that nNk, N−nk
N
h1 h2
, nN
k+1, and n
h2 h1 k
N converge. Consequently g(x) is contained in the collection of (75), (76), (77), (81), (82) and (83).
Thus the proof is finished.
L. M i ˇs´ık (2004, personal communication) found the following sequence xn for which c1(x)∈ G(Xn) andc0(x) ∈/ G(Xn) and consequently the implication Q.7 in [9] does not hold.
Example 6. Letxn,n= 1,2, . . ., be an increasing sequence of positive integers which satisfies the following conditions
(i) ifnk= (k+ 1)(k−1)!2k(k−1)2 fork= 1,2, . . ., thenxnk = (k+ 1)nk, (ii) ifnk=k(k−2)!2k(k−1)2 then xnk =k2nk,
(iii) ifn= 2ink−1+j, 0≤j <2ink−1 and 0≤i < k−1 fork= 1,2, . . ., then xn=xnk−1(i+ 1)2i+ (i+ 3)kj (i.e., n∈[nk−1, nk]),
(iv) ifn∈[nk, nk] fork= 1,2, . . ., then xn=xnk+n−nk. Then for the sequence of blocks
Xn= x1
xn, x2
xn, . . . ,xn xn
we havec1(x)∈G(Xn) but c0(x)∈/G(Xn).5 P r o o f. We start with the following figure:
Here for n running through [2ink−1, 2i+1nk−1], the xn is equi-distributed in [x2ink−1, x2i+1nk−1] with difference Δi, wherei= 0,1, . . . , k−2.
5This and the Theorem 13 imply thatG(Xn)⊂ {cα(x);α∈[0,1]}.
10.Using the definition of xn we can see that xxnk
nk →1 and nnk
k →0 and thus we havec1(x)∈G(Xn).
20. On the contrary, assume that there exists increasing sequence ml < ml, l= 1,2, . . ., such thatml∈[nk−1, nk],k=k(l), (i) xxml
ml →0 and (ii) mml
l →1 as l→ ∞.
a) If [2jnk−1,2j+1nk−1]⊂[ml, ml] for some 0≤j ≤k−2, then ml
ml ≤ 2jnk−1 2j+1nk−1 = 1
2 which contradicts (ii).
b) If [ml, ml]⊂[2jnk−1,2j+2nk−1], then xml
xml ≥ x2jnk−1 x2j+2nk−1
= (j+ 1)2j (j+ 3)2j+2 =
1− 2
j+ 3 1
4 which contradicts (i).
c) If [nk, nk]⊂[ml, ml], then ml ml ≤ nk
nk →0 which contradicts (ii).
d) Ifml ∈[2k−2nk−1, nk] and ml ∈[nk, nk], i.e.,ml=nk+i, then (because nk = 2k−1mk−1 andxml =xnk+i)
xml
xml ≥ x2k−2nk−1
xml = x2k−2nk−1
x2k−1nk−1
· xnk
xml =
k−1 k
· 1 2· 1
1 +xi
n k
, ml
ml ≤ nk
ml = 1 1 +ni
k
. Furthermore, (i) implies ni
k → 0 and (ii) implies xi
n k
= k2in
k → ∞ which is impossible.
e) If [2nk,22nk]⊂[ml, ml] then ml
ml ≤ 2nk
22nk = 1 2 which contradicts (ii).
f) Finally, assume thatml ∈[nk, nk] andml ∈[nk,2nk]. Sincex2nk = 4xnk, we have
xml
xml ≥ xnk
x2nk = xnk 4xnk → 1
4
which contradicts (i).