226 CHAPTER 9. INTEGRATION

55. The problem in Apostol’s book mentioned in Problem 54 does not require nan tobe decreasing and is as follows. Let {ak}∞

k=1 be a decreasing sequence of non-negative numbers which satisfies limn→∞ nan = 0. Then ∑

∞k=1 ak sin(kx) converges

uniformly on R. You can find this problem worked out completely in Jones [19].Fill in the details. This is a very remarkable observation. It says for example that∑

∞k=1

1k1+ln(k) sin(kx) converges uniformly.

Always let p be so large that (p−1)ap−1 < ε . Also, note that |sinx| ≤ |x| for allx and for x ∈ (0,π/2) ,sinx ≥ x

2π. (You could just graph sinx− x

2πto verify this.)

Also, we can assume all ak are positive since there is nothing to show otherwise.Define b(k) ≡ sup

{ja j : j ≥ k

}. Thus k → b(k) is decreasing and b(k)→ 0 and

b(k)/k ≥ ak.

Suppose x < 1/q so each sin(kx)> 0. Then∣∣∣∣∣ q

∑k=p

ak sin(kx)

∣∣∣∣∣≤ q

∑k=p

b(k)k

sin(kx)≤q

∑k=p

b(k)k

kx≤ b(p)(q− p)1q≤ b(p) (9.14)

Next recall that

n

∑k=1

sin(kx) =cos( x

2

)− cos

((n+ 1

2

)x)

2sin( x

2

) ≡ mn (x) ,

|mn (x)| ≤ n, |mn (x)| ≤1

sin(x/2)if x ∈ (0,π) .

This is from the process for finding the Dirichlet kernel. Then use the process ofsummation by parts to obtain in every case that∣∣∣∣∣ q

∑k=p

ak sin(kx)

∣∣∣∣∣≤ ∣∣aqmq (x)−ap−1mp−1 (x)∣∣+ ∣∣∣∣∣q−1

∑k=p

mk (x)(ak−ak+1)

∣∣∣∣∣≤ 2ε +

q−1

∑k=p|mk (x)|(ak−ak+1)≤ 2ε +

1sin(x/2)

(ap−aq) (9.15)

We will only consider x ∈ (0,π) for the next part. Then for such x, It remains toconsider x∈ (0,π) with x≥ 1/q. In this case, choose m such that q > 1

x ≥m≥ 1x −1.

Thus x < 1m ,

1m+1 < x. Then from 9.15, 9.14,

∣∣∣∣∣ q

∑k=p

ak sin(kx)

∣∣∣∣∣≤≤b(p)︷ ︸︸ ︷∣∣∣∣∣ m

∑k=p

ak sin(kx)

∣∣∣∣∣+∣∣∣∣∣ q

∑k=m+1

ak sin(kx)

∣∣∣∣∣≤ b(p)+2ε +

1sin(x/2)

(am+1−aq)≤ b(p)+2ε +4π

xam+1

≤ b(p)+2ε +4π

xb(m+1)

m+1≤ b(p)+2ε +

xb(m+1)x

≤ b(p)+2ε +4πb(p)≤ 2ε +17b(p) , limp→∞

b(p) = 0.

226CHAPTER 9. INTEGRATION55. The problem in Apostol’s book mentioned in Problem 54 does not require na, tobe decreasing and is as follows. Let {a,};_, be a decreasing sequence of non-negative numbers which satisfies lim,_,..md, = 0. Then )7_, a, sin (kx) convergesuniformly on R. You can find this problem worked out completely in Jones [19].Fill in the details. This is a very remarkable observation. It says for example thatLie=1 gv Sin (kx) converges uniformly.Always let p be so large that (p—1)ap_1 < €. Also, note that |sinx| < |x| for allx and for x € (0,2/2),sinx > 35. (You could just graph sinx — 3° to verify this.)Also, we can assume all a, are positive since there is nothing to show otherwise.Define b(k) = sup { ja; : j >k}. Thus k — b(k) is decreasing and b(k) + 0 andb(k) /k > ag.Suppose x < 1/q so each sin (kx) > 0. Then4 1 b(k 1 b(k 1y ay sin (kx)| < y 10) n(x) < y "ee < b(p)(q—p)- <b(p) (.14)k=p k=p k=p qNext recall thatne cos (¥) —cos ((n+ 4) x) _py sin(kx) = sin (3 =m, (x),1 .lmn(x)| <a, rm IS Sapa fre 07).This is from the process for finding the Dirichlet kernel. Then use the process ofsummation by parts to obtain in every case thatq q-ly ay sin (kx)} < Jaging (x) — ap—1mp—1 (x)| + y my, (x) (ag — p41)k=p k=pq-1S 26+ Yelena (2)I (ak — aes) S2€+ Say a) (9.15)We will only consider x € (0,2) for the next part. Then for such x, It remains toconsider x € (0,2) with x > 1/q. In this case, choose m such that q > + >m> + 1.Thus x < 4, 1) <x. Then from 9.15, 9.14,<b(p)q m qY- ag sin (kx) < Y. ag sin (kx) + y ag sin (kx)k=p k=p k=m+1< d( )+2e+—1 _( — aq) <b( )42e4%— P sin (x/2) m1 ~ Gg) = O\P yom4n b(m+1) An<b 26+ ———_ <b 2e+—b 1< b(p)+2e+ a mal = (p) +2e€+ x (m+1)x< b(p)+2e+4nb(p) <2e+17b(p), lim b(p) =0.Dp ‘CO