Solutions to Problems in Amer. Math. Monthly, Math.

Magazine, College Math. J., Elemente der Math.,Crux Math.,

Raymond Mortini

arXiv (arXiv: 2501.05096v1)

Generated on May 13, 2025

Abstract
In this arxiv-post I present my solutions (published or not) to Problems that appeared in Amer. Math.
Monthly, Math. Magazine, Elemente der Mathematik and CRUX, that were mostly done in collaboration
with Rudolf Rupp. Some of them (including a few own proposals which were published) were also done
in cooperation with Rainer Br\"uck, Bikash Chakraborty, Pamela Gorkin, Gerd Herzog, J\'er\^ome
No\"el, Peter Pflug and Amol Sasane.

Solutions to Problems in Amer. Math. Monthly Math. Magazine College Math. J. Elemente der Math
Crux Math. EMS Newsletter Math. Gazette Edited by Raymond Mortini 10.1.2025 1arXiv:2501.05096v1
[math.HO] 9 Jan 2025
2 In this arxiv-post I present my solutions to Problems that appeared in Amer. Math. Monthly, Math.
Magazine, Elemente der Mathematik and CRUX, that were mostly done in collaboration with Rudolf
Rupp. Some of them (including a few own pro- posals which were published) were also done in
cooperation with Rainer Br¨ uck, Bikash Chakraborty, Pamela Gorkin, Gerd Herzog, J´ erˆ ome No¨ el,
Peter Pflug and Amol Sasane. A few of these contributions to “Recreational Mathematics” actually were
the base for interesting generalizations that led to some of my publications (partially co-authored) in
research journals ([1, 2, 3, 4, 5, 6, 7, 8]). The content will surely be attractive to all
undergraduate/graduate students in Mathematics who want to solve challenging problems in Analysis
by calculating explicit values of funny looking integrals, sums and products, by deriving astonishing
inequal- ities and by solving functional equations. It is also a valuable source for teachers in
mathematics in preparing exercise sheets for their students. Moreover, I think that it is worth to see in
most cases quite different solutions than those already published in the above listed journals. My main
reason to post this collection of (mainly unpublished solutions), is to keep also for future generations an
archive for historians in Mathematics, interested in the work of one of the very few mathematicians with
Luxembourgish Nationality. With- out this digital archiving, these contributions to education and science
would for ever disappear in a few years. Some technical remarks: The first items for each journals are
still blacked, as the submission deadline has not yet occurred. Regular updates are planned. Own
proposals are presented with a yellow background. We obtained permission from the MAA, the EMS,
Math. Gazette and CRUX to post a scan of the original statements and to reproduce our published (and
non-published) solutions. 10.1.2025 Raymond Mortini, Pontpierre (Luxembourg). Prof. emeritus, Univ.
Lorraine, Metz, France Visiting researcher Univ. Luxembourg, Esch-sur-Alzette
3 published proposals See [9] for the whole bibliographic list.
4 .1.American Math. Monthly ©Mathematical Association of America, 2025. Solution to problem 12501
in Amer. Math. Monthly 131 (2024), p. 906 Raymond Mortini and Rudolf Rupp - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Substituting x/(1 +x) =:t, equivalently x=t/(1−t) with
t∈]0,1[ yields I:=ˆ∞ 0■ logx 1 +x■4 log■ x3(1 +x)17■ dx =ˆ1 0log4t■ 3 log■ t 1−t■ + 17 log■ 1 1−t■■
1 1−tdt (1−t)2 =ˆ1 0log4t■ 3 logt−20 log(1 −t)■ 1−tdt = 3ˆ1 0log5t 1−tdt−20ˆ1 0log4tlog(1−t) 1−tdt =: 3
I1−20I2. Now, as the integrands have constant sign,´P=P´ , and so, using 5-times integration by parts,
I1=ˆ1 0∞X n=0tnlog5t dt=∞X n=0ˆ1 0tnlog5t dt =−∞X n=05! (n+ 1)6=−120ζ(6). LetHk= 1 +1 2+1 3+···+1
kdenote the k-th harmonic number. Using that for 0 ≤x <1, log(1−x) 1−x=−∞X k=1Hkxk
(Cauchy-product), we obtain I2=−ˆ1 0log4t∞X k=1Hktkdt=−∞X k=1ˆ1 0tklog4t dt =−4!∞X k=1Hk (k+ 1)5.
Due to Euler’s formula (see [10, p. 416], [11]), 2∞X k=1Hk (k+ 1)n=nζ(n+ 1)−n−2X k=1ζ(n−k)ζ(k+
1),(n≥2), we conclude that ∞X k=1Hk (k+ 1)5=1 2■ 5ζ(6)−ζ(4)ζ(2)−ζ(3)2−ζ(2)ζ(4)■ =5 2ζ(6)−ζ(4)ζ(2)−1
2ζ(3)2. Since ζ(6) =π6 945andζ(4)ζ(2) =π4 90π2 6=π6 540=945 540ζ(6) =7 4ζ(6), we deduce that
5 I= 3I1−20I2=−360ζ(6) + 20 ·24■5 2ζ(6)−ζ(4)ζ(2)−1 2ζ(3)2■ =ζ(6)■ −360 + 1200 −840■ −240ζ(3)2
=−240ζ(3)2.
6 Solution to problem 12494 in Amer. Math. Monthly 131 (2024), p. 815 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let S:=S(r) :=∞X n=1∞X
m=11 m2n+mn2+rmn. We additionally show that for r∈N={0,1,2,3, . . .} S(r) =ˆ1 0xr−1log2(1−x)dx= 
2ζ(3) if r= 0 r−1X j=0■r−1 j■ (−1)j2 (j+ 1)3ifr■= 0. If, more generally, r≥0 is a real number, then this
formula also holds in the form S(r) =∞X j=0■r−1 j■ (−1)j2 (j+ 1)3. 1.1.The calculations. Letm, n∈N∗:=N\
{0}andr∈R, r≥0. Then 1 m2n+mn2+rmn=1 mn1 m+n+r=1 m(m+r)■1 n−1 n+ (m+r)■ =1 m(m+r)ˆ1
0(xn−1−xn+m+r−1)dx =1 m(m+r)ˆ1 0xn−1(1−xm+r)dx. Since all terms considered are positive,´P=P´ .
Hence, by partial integration, ∞X n=11 m2n+mn2+rmn=1 m(m+r)ˆ1 0(1−xm+r)∞X n=1xn−1dx (1) =1
m(m+r)ˆ1 01−xm+r 1−xdx =−1 mˆ1 0xm+r−1log(1−x)dx Consequently, if r≥0,r∈R, S(r) = −∞X m=11 mˆ1
0xm+r−1log(1−x)dx =−ˆ1 0xr−1 ∞X m=1xm m! log(1−x)dx=ˆ1 0xr−1log2(1−x)dx (2) =ˆ1 0(1−x)r−1log2x
dx. (3)
7 Hence, if r∈N∗, S(r) =r−1X j=0■r−1 j■ (−1)jˆ1 0xjlog2x dx =r−1X j=0■r−1 j■ (−1)j2 (j+ 1)3. where the
latter identity is shown using twice partial integration. Ifr= 0, we have the convergent integral (two
singularities), S=ˆ1 0(1−x)−1log2x dx =ˆ1 0∞X j=0xjlog2x dx =∞X j=0ˆ1 0xjlog2x dx =∞X j=02 (j+ 1)3=
2ζ(3). Ifr∈Randr >0, S(r) =∞X j=0■r−1 j■ (−1)jˆ1 0xjlog2x dx =∞X j=0■r−1 j■ (−1)j2 (j+ 1)3. Next, if
r∈N∗, we derive the desired equality from 2. We do this using the following trick: consider the function
f(a) := (1 −x)a−1, a > 0. Then d2 da2(1−x)a−1=d da(1−x)a−1log(1−x) = (1 −x)a−1log2(1−x). In particular,
log2(1−x) =d2 da2(1−x)a−1 a=1. Fora >0, let G(a) :=ˆ1 0xr−1d2 da2(1−x)a−1dx. The continuity of Gon
]0,∞[ and ( x, a)7→xr−1d2 da2(1−x)a−1on ]0,1[×]0,∞[ yields that lim aˆ =ˆ lim a. Hence G(1) = lim
a→1ˆ1 0xr−1d2 da2(1−x)a−1dx=ˆ1 0xr−1lim a→1d2 da2(1−x)a−1dx =ˆ1 0xr−1log2(1−x)dx. Now, in view
of the definition of the Eulerian beta-function, B(r, a), and the fact that d2 da2´ =´d2 da2
8 G(a) =d2 da2ˆ1 0xr−1(1−x)a−1dx =d2 da2B(r, a) =d2 da2Γ(r)Γ(a) Γ(a+r) =d2 da2■(r−1)! a(a+
1)···(a+r−1)■ . Fora >0, put L(a) := log1Qr−1 j=0(a+j)=−r−1X j=0log(a+j). Then F(a) :=B(r, a)
(r−1)!=eL(a). Hence G(1) = ( r−1)!F′′(1) = ( r−1)!F(1)■ (L′(1))2+L′′(1)■ =(r−1)! r!■ (Hr)2+H(2) r■ .
Remark . Using that for r∈N∗ 1 m(m+r)ˆ1 01−xm+r 1−xdx=1 m(m+r)ˆ1 0m+r−1X k=0xkdx =Hm+r
m(m+r), we also have that, in view of (1), S(r) =∞X m=1Hm+r m(m+r).
9 Solution to problem 12490 in Amer. Math. Monthly 131 (2024), p. 814 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - a1) We prove thatP´ is
well defined and that ∞X n=1ˆ1 0sin(2πnx) nf(x)dx≤0. Asf′≥0, partial integration yields In:=ˆ1 0sin(2πnx)
nf(x)dx=−cos(2 πnx) 2πn2f(x) 1 0+ˆ1 0cos(2 πnx) 2πn2f′(x)dx ≤f(0)−f(1) 2πn2+1 2πn2ˆ1 0f′(x)dx= 0.
Since|In| ≤c n2the series converges S:=PInand so S≤0. a2) We prove that´Pis well defined and that ˆ1 0
∞X n=1sin(2πnx) nf(x)! dx≤0. A standard exercise in Fourier analysis tells us S(x) :=∞X n=1sin(2πnx)
n=π−2πx 2for 0 < x < 1. Since f′≥0, ˆ1 0S(x)f(x)dx=π 2■ˆ1 0f(x)dx−2ˆ1 0xf(x)dx■ =π 2■ xf(x) 1 0−ˆ1
0xf′(x)dx−x2f(x) 1 0+ˆ1 0x2f′(x)dx■ =π 2■ (x−x2)f(x) 1 0+ˆ1 0(x2−x)f′(x)dx■ =π 2ˆ1 0(x2−x)f′(x)dx≤0.
10 b) Since g′′≥0, Jn=ˆ1 0cos(2 πnx)g(x)dx=−1 2πnˆ1 0sin(2πnx)g′(x)dx =1 4π2n2cos(2 πnx)g′(x) 1 0−1
4π2n2ˆ1 0cos(2 πnx)|{z} ≤1g′′(x)dx ≥g′(1)−g′(0) 4π2n2−g′(1)−g′(0) 4π2n2= 0. Since|Jn| ≤c n2, the series
S∗:=PJnconverges and so S∗≥0. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Second proof of b)
Extend g1-periodically to R. The Fourier series for gis given by g(x) =a0 2+∞X n=1(ancos(2 πnx)
+bnsin(2πnx)) with an= 2´1 0g(x) cos(2 πnx)dxandbn= 2´1 0g(x) sin(2 πnx)dx. Since gis smooth,
Dirichlet’s rule yields (4)g(0) + g(1) 2=a0 2+∞X n=1an=ˆ1 0g(x)dx+ 2∞X n=1ˆ1 0cos(2 πnx)g(x)dx.
Nowg(0)+g(1) 2≥´1 0g(x)dxsince the convexity of gimplies g(x)≤g(0) + x(g(1)−g(0)) and soˆ1
0g(x)dx≤g(0)·1 +1 2(g(1)−g(0)) =g(0) + g(1) 2. Hence, in view of (4),∞X n=1ˆ1 0cos(2 πnx)g(x)dx≥0.
Second proof of a) This follows from b): Given fin a), put g(x) :=´x 0f(t)dt. Then g′′≥0; hence gis convex.
Now, by partial integration, ˆ1 0g(x) cos(2 πnx)dx=g(x)sin(2πnx) 2πn 1 0−ˆ1 0sin(2πnx) 2πng′(x)dx =−ˆ1
0sin(2πnx) 2πnf(x)dx. Thus b) implies a).
11 Solution to problem 12481 in Amer. Math. Monthly 131 (2024), p. 630 Raymond Mortini, Peter Pflug
and Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The answer
is yes, and this is an easy classical exercise in a complex analysis course; see [14, p. 168], [15, p. 353],
[13, p. 161], [17, p. 164, Ex. 300/302], [19, p. 41], [20]. Since the modulus of a holomorphic function is
subharmonic, it satisfies the maximum principle. Thus, if the maximum value of f∈H(U) is taken inside
the domain U, then fis constant. This establishes the problem for n= 1. Now if n≥2, we use that the finite
sum uof subharmonic functions is subharmonic again, a fact best seen by using the definition via the
mean-value inequality u(x0)≤1 2πˆ2π 0u(x0+reiθ)dθ for all closed disks D(x0, r)⊆U. Thus g=Pn j=1|fj|is
subharmonic, and so is constant c under the assumption of the problem. Next, fix j0∈ {1, . . . , n }. Then
|fj0|=c−nX j=1 j■=j0|fj| |{z} :=v. The subharmonicity of vimplies that −vis superharmonic. Hence, by
adding the constant c, c−vis superharmonic, too. Consequently, |fj0|is superharmonic, as well as
subharmonic. In other words, |fj0|is harmonic. Any holomorphic function fin a domain Uwhose modulus
is harmonic, is constant though. In fact, let Dbe a closed disk in Usuch that fhas no zeros in D. Then
fadmits a holomorphic square-root qinD; that is q2=f([12, p. 816]). Now |q|2is harmonic by assumption.
Since ∆ |q|2= 4|q′|2(where ∆ is the Laplacian [12, p. 222]), we get that q′≡0 inD, and so qis constant in
D. Consequently, f=q2is constant in D. By the uniqueness theorem for holomorphic functions [15, p.
347], fis constant in the connected open setU. Remark 1. A more direct, but not so elegant way to prove
that the harmonicity of |f|implies constancy, is purely computational and uses the ∂-calculus, that is the
Wirtinger derivatives ∂u=uzand∂u=uz(see [12, sect. 4]): Lets(z) :=|f(z)|,f■≡0. Then s2=ffand∂s2=
2s∂s=f′f. Hence, outside the discrete zero set of f,∂s=f′f 2s, and so ∆s= 4∂∂s=4 2f′f′s−f′ff′f 2s s2!= 0⇐⇒
| f′|2■ 2s2− |f|2■ = 0⇐⇒ | f′|2|f|2= 0⇐⇒ f′= 0. This could of course also be established via the
”real”-method by calculating sxx+syywith s(z)∼s(x, y),z=x+iyands=√ u2+v2, where f=u+iv, and applying
the Cauchy- Riemann equations in their ‘real’ form, instead of the shorter form fz= 0 above. Remark 2.
An analysis of the proof shows that one obtains the same conclusion replacingPn j=1|fj|byPn j=1|fj|αj,
where αj>0. Just note the following two points: 1) if f■≡0 is holomorphic, then u:=αlog|f|is subharmonic,
and so its left composition euwith the exponential function yields the subharmonicity of |f|α(see [16,
Chap.1, §6] or [18, p.44]). 2) The square root qoffabove is replaced by an α/2-root of f, so that q2=fα.
12 Solution to problem 12479 in Amer. Math. Monthly 131 (2024), p. 630 Raymond Mortini, Peter Pflug
and Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - For
technical reasons, we use the new enumeration r0, r1, r2, . . .to denote the real roots of the equation 2
cos(√ 3x) =−e−3x. We show that ■ ■ S:=∞X n=0r−6 n=8 5. Recall the Bourbaki notation N:={0,1,2, . .
.}andN∗:=N\ {0}. Let g(z) := 2 cos(√ 3z) +e−3z, and, by multiplication with e3z, f(z) := 2 e3zcos(√ 3z) +
1. Note that falso writes as f(z) =e(3+i√ 3)z+e(3−i√ 3)z+ 1. Depending on which is best adapted to the
computations, we shall use in the sequel all these variants. Note that the functions f,ghave the same
zero sets. Property 1 Let α:=ei2π/3=−1 2+i√ 3 2. Then f(αz) =e(−3+i√ 3)zf(z). In fact, by using that −1
2+i√ 3 2!■ 3 +i√ 3■ =−3 +√ 3i and −1 2+i√ 3 2!■ 3−i√ 3■ = 2i√ 3, we obtain f(αz) = e(−3+√ 3i)z+e2i√
3z+ 1 =e√ 3iz■ e−3z+ei√ 3z■ + 1 =e√ 3iz■ e−3z+ei√ 3z+e−i√ 3z■ −1 + 1 =e√ 3iz■ e−3z+ 2 cos(√ 3z)■
=e√ 3ize−3zf(z). As an immediate consequence we have the following useful property: Property 2 f(z) =
0 if and only f(αz) = 0 .
13 Property 3 The function fhas infinitely many positive zeros and no negative one. More precisely, for
every n∈Neach interval In:=■2nπ√ 3,2(n+ 1)π√ 3■ contains exactly two positive zeros of f, one
contained in the left part of the intervalhπ 2+2nπ√ 3,3π 2+2nπ√ 3i , and one in the right part. Proof. •Let
us first show that g(hence f) has no negative zeros. In fact if −π 2√ 3≤x≤0, then cos(√ 3x)≥0, and so g(x)
=e−3x+ 2 cos(√ 3x)>0. If x <−π 2√ 3, then e−3x> e√ 3π 2∼ 15.1909···>2, and again g(x)>0. •Now we
deal with the positive solutions to g(x) = 0; or equivalently f(x) = 2 e3xcos(√ 3x)+ 1 = 0. Consider the
points an=2nπ√ 3, b n=2nπ+π 2√ 3, c n=2nπ+π√ 3, d n=2nπ+3π 2√ 3, e n=2nπ+ 2π√ 3. To achieve our
goal, we study for x≥0 the variation of f(x) = 2 e3xcos(√ 3x)+1. Note that f′(x) = 2 e3x■ 3 cos(√ 3x)−√ 3
sin(√ 3x)■ . In particular, f′(x) = 0 ⇐⇒√ 3 = tan√ 3x⇐⇒ x=xn:=arctan√ 3+nπ√ 3=π 3+nπ√ 3, n∈N. Thus
fis increasing on [2nπ√ 3,π 3+2nπ√ 3], decreasing on [π 3+2nπ√ 3,4π 3+2nπ√ 3] and increasing on [4π
3+2nπ√ 3,2π+2nπ√ 3]. Now f(an) = 2 e3an+ 1, f(bn) = 1 , f(cn) =−2e3cn+ 1<0, f(dn) = 1 , f(en) = 2 e3en+
1. Since an<π 3+2nπ√ 3< bn<π+2nπ√ 3=cn<4π 3+2nπ√ 3<3π 2+2nπ√ 3=dn, we see that there are
exactly two zeros on [ an, en], namely one between bnandcnand one between cnanddn1. Let us denote
the zeros by rn, in increasing order ( n∈N). Note that rnis eventually close to the zero
■ sn:=π 2+nπ√ 3 of 2 cos(√ 3x), since e−3xtends rapidly to 0 as x→ ∞ , and that for n∈N, (5) s2n< r2n<
r2n+1< s2n+1< s2n+2< r2n+2 (because the cosinus values at rnmust be negative). ■ Here are some
numerical examples: r0∼0.924906 . . ., s0=π 2√ 3∼0.9068996 . . .,|s0−r0| ∼0.0180064 <0.125 = 1 /8,
r1∼2.720616677 . . .,s1=3π 2√ 3∼2.7206990 . . .,|r1−s1| ∼0.00008223 ···, r2∼4.53449876 . . ., s2=5π 2√
3∼4.53449841 . . .,|r2−s2| ∼0.00000035 ···. |r1−r0| ∼1.7957107 ···,|r2−r1| ∼1.81388208 ··· The items
(1)–(4) below are nice additional properties of the real zeros of f. We do not need these, though. 1Using
only the information f(an)>0, f(cn)<0 and f(en)>0 (so without studying the variation off), allows us to
conclude that fhas at least two zeros in In= [an, en]. Using Rouch´ es theorem and the additional
information that in the right half-plane the complex zeros of cos(√ 3z), all simple, are exactly the real
zerosπ 2+2nπ√ 3and3π 2+2nπ√ 3, gives us the possibility to conclude that fhas no other zeros in the
strips {z∈C: Rez∈In}(see Property 6).
14 Property 4 (1)|rn−sn|<1 16√ 3<1 8for all n∈N. (2)|rn−sn| →0. (3)|rn−rn−1|>1 for all n∈N∗.
(4)|rn−rn−1| →π√ 3∼1.813799 ···. (5)r3 n∼cn3, for some constant c. Noticing that {r2n, r2n+1} ⊆In, one
can easily show that |r2n+1−r2n|increases toπ√ 3and |r2n−r2n−1|decreases toπ√ 3. Proof. Recall that
f(z) = 2 e3zcos(√ 3z) + 1. (1) For n= 2meven, it is sufficient to prove that sn< rn< sn+1 16√ 3=:pn,
where sn, rn∈ Im. Since f(sn) = 1 and f(rn) = 0, and fis decreasing on [π 3+2mπ√ 3,4π 3+2mπ√ 3]⊇[sn,
pn], this proof is done by showing that f(pn)<0. Now, noticing that the cosinus term is negative, f(pn) = 2
e√ 3(π 2+nπ+1 16)cos■π 2+1 16■ + 1 ≤2e√ 3(π 2+1 16)cos■π 2+1 16■ + 1∼ −1.114587 . . . . Forn=
2m+ 1 odd, it is proved in the same way that qn:=sn−1 16√ 3< rn< sn, where sn, rn∈Im. In fact, fis
increasing on [4π 3+2mπ√ 3,2π+2mπ√ 3]⊇[qn, sn]. So, by noticing that the cosinus term is negative,
f(qn) = 2 e√ 3(3π 2+(n−1)π−1 16)cos■3π 2−1 16■ + 1 ≤2e√ 3(3π 2−1 16)cos■3π 2−1 16■ + 1∼
−391.97708 ··· (2) This works in the same way as above; just replace 1 /16 be an number ε >0 arbitrary
close to 0. Then the cosinus term, still negative, may be very small. But the power nin the exponential
factor can be made sufficiently big (depending on ε), so that f(pn), resp. f(qn) is strictly negative. (3) Let
nbe even. Then |rn−rn−1|=rn−rn−1≥sn−sn−1=π√ 3∼1.813799 ···>1. Ifnis odd, then by (1)
|rn−rn−1|=rn−rn−1≥(sn−1 16√ 3)−(sn−1+1 16√ 3) =π√ 3−1 8√ 3∼1.7416305 ···>1. (4) Due to (2)
rn−rn−1= (rn−sn) + (sn−sn−1) + (sn−1−rn−1)→0 +π√ 3+ 0. (5) Using (5), we obtain that (6)π 2+ 2nπ√ 3<
r2n< r2n+1<3π 2+ 2nπ√ 3. Hencern n→π√ 3and so r3 n∼cn3. ■ LetZ:={rn, αrn, α2rn}. Combining the
properties (1) and (2), we see that Zis a symmet- ric zero-set for f, resp. g.
15 Property 5 The elements of Zare simple zeros for f(hence for g)2. Proof. Suppose, contrariwise, that
for some real z, (7) f(z) = 2 e3zcos(√ 3z) + 1 = 0 , and (8) f′(z) = 3·2e3zcos√ 3z−2·e3z√ 3 sin√ 3z= 0.
Plugging equality (7) into (8) yields (9) 0 = f′(z) =−3−2√ 3e3zsin√ 3z. Hence 1 = sin2(√ 3z) + cos2(√
3z)(9)= (7)3 4e−6z+1 4e−6z=e−6z. Therefore z= 0. But f(0) = 3, a contradiction. We conclude that, due
to Property 2, all elements in Zare simple zeros for f. ■ Property 6 The exact zero-set of fcoincides with
Z. Proof. Numerically it can be shown that for k∈N∗ 1 2πiˆ |z|=2kπ/√ 3f′ fdz= 6k, which yields the
assertion, as we already know that within the annuli ■ z∈C:2nπ√ 3≤ |z| ≤2(n+ 1)π√ 3■ there are 2 real
zeros and their rotations by αandα2. For our genuine proof, we first restrict the calculations to the case
where Re z≥0. The other case will be deduced at the very end of the proof. •We know that |cosz|2=
(cos x)2+ (sinh y)2, where z=x+iy. In fact, 4|cosz|2=|ei(x+iy)+e−i(x+iy)|2=|e−yeix+eye−ix|2 =e−2y+e2y+
2Re( e2ix) = 2 cosh(2 y) + 2 cos(2 x) = 2(cosh2y+ sinh2y) + 2(cos2x−sin2x) = 2(1 + 2 sinh2y) +
2(cos2x−(1−cos2x)) = 4(sinh2y+ cos2x). •Recall that g(z) =e−3z+ 2 cos(√ 3z). We now show that g■= 0
on [0 ,∞[×[1,∞[. In fact, |g(z)| ≥2|cos(√ 3z)| − |e−3z|= 2q cos2(√ 3x) + sinh2(√ 3y)−e−3x≥2 sinh√ 3−1>0.
The same proof also shows that g■= 0 on [0 ,∞[×]− ∞,−1]. Next we use Rouch´ e’s theorem for the
rectangles Rk:=h 2kπ√ 3,2(k+1)π√ 3i ×[−1,1],k∈N. If x=2kπ√ 3orx=2(k+1)π√ 3, then |g(z)−2 cos(√
3z)|=e−3x≤1<2|cos√ 3iy|= 2(q 1 + sinh2(√ 3y) = 2|cos(√ 3z)| ≤ |g(z)|+ 2|cos(√ 3z)|. 2Also note here, that
Rouch´ e’s theorem (applied as in the proof of Property 6) automatically yields this fact, too. We
preferred though a straightforward elementary proof.
16 If|y|= 1, then |g(z)−2 cos(√ 3z)|=e−3x≤1<2 sinh√ 3≤2q cos2(√ 3x) + sinh2(√ 3)≤2|cos(√ 3z)| ≤ |g(z)|+
2|cos(√ 3z)|. By Rouch´ e’s Theorem [12, p. 852], gand cos(√ 3z) have the same number of zeros on
Rk, namely 2. As we already know that ghas two real zeros in these intervalsh 2kπ√ 3,2(k+1)π√ 3i , we
are done. •Having only the real zeros rkoff(equivalently g) in the right-half plane, the symmetry f(z) = 0
⇐⇒ f(ei2π/3z) = 0 now implies that no other zeros are in the left half-plane excepted the rotations
ei2π/3rkandei4π/3rkas any ζ:=reiθwith π/2≤θ≤3π/2 would yield that at least one of the points
ei4π/3ζorei2π/3ζbelongs to the right-half plane. This finishes the proof of Property 6. ■ We are finally
ready to derive the value for the desired sum S. The main idea is to apply the residue theorem for the
function F(z) :=1 z6f′(z) f(z), and the formula Res■ hf′ f, p■ =h(p), where pis a simple pole of fandhis
holomorphic in a neighborhood of p. Observe that3 f′(z) f(z)= 2−4z2−24 5z5−212 35z8−14736
1925z11−1694832 175175z14− ··· Thus, Res■1 z6f′ f, z= 0■ =−24 5. Note that the zeros zn=αrn, resp.
zn=α2rnhave the property that z6 n=rn. Recall from (6) that for n∈Nthe numbers r2n, r2n+1belong to
■2nπ+π 2√ 3,2nπ+3π 2√ 3■ . ForN∈N, letCNbe the circle centered at the origin and with radius
RN:=2π(N+1)√ 3. Then the associated disk contains r0, r1, . . . , r 2N+1and (1)RN> N, (2)R3 N−r3
2N+1≥1. Note that (2) is a consequence to RN−r2N+1≥π/2. Due to the residue theorem L:= lim N→∞1
2πiˆ CN1 z6f′(z) f(z)dz= Res■1 z6f′ f,0■ +X ξ∈ZRes■1 z6f′ f, ξ■ =−24 5+ 3X n1 r6n. We claim that L=
0, from which we deduce that S=X n1 r6n=8 5. To prove this claim, we use that by Property (4),P1
r3nconverges. Hence the infinite product p(z) :=∞Y n=0 1−■z rn■3! 3Obtained e.g. with wolframalpha.
17 converges locally uniformly on C. According to Weierstrass’s factorization theorem, f=pq, where qis
a zero-free entire function. Note that f′ f=p′(z) p(z)+q′(z) q(z) and that ˆ CN1 z6f′(z) f(z)dz=ˆ CN1 z6p′(z)
p(z)dz+ 0. Now p′(z) p(z)=−3∞X n=0z2 r3n−z3, which is a locally uniformly convergent series on
C\Z(note that r3 n∼c·n3). Hence´P=P´ and so, by Cauchy’s integral theorem, 2πiJN:=ˆ CN1 z6p′(z)
p(z)dz=−32N+1X n=0ˆ CNz−4 r3n−z3dz. Now z−4 r3n−z3 ≤R−4 N R3 N−r3n≤R−4 N. Thus
0≤2πJN≤3(2N+ 2)R−4 N2πRN≤C·N−2. We conclude that L= lim N→∞JN= 0. ■ Related problems are
given in [23, p. 279] and [22] (cos xcoshx+1 = 0) and [21] (tan x=kx) 4. Second, but more complicated
proof of Property 6 Note that g(z) =e−3z+ 2 cos(√ 3z) = 0 implies that 4|cos2(√ 3z)|=e−6x, and so 4
sinh2(√ 3y) =e−6x−4 cos2(√ 3x). In particular (10) 4 cos2(√ 3x)≤e−6x. Ifk∈N, equality in (10) holds if
x=rkand if x=π 2+kπ√ 3=sk, then trivially 0 = 4 cos2(√ 3x)≤e−6x. In the figure below, we display one
”branch” of the curves yj, given by (11) y2=1√ 3arsinhr 1 4e−6x−cos2(√ 3x) and y1= 2 cos(√ 3x) +e−3x.
They are very tiny, as the coordinates show. To recapitulate, if zis a zero of gwith non-negative real
part, then it belongs to these tiny arcs. •To show that in the right half-plane ghas only the zeros rk, it
remains to study the behaviour of g(z) = 2 cos(√ 3z) +e−3zon the disks D(rk,1/8). A major difficulty will
be to show that the union of these disks contains the graph of the curve y2. A tool will be the following
result from complex analysis: Zero-set criterium 1. [12, p. 365] LetΦbe bounded by 1and holomorphic in
Dand suppose thatΦ(0) = 0 as well as |Φ′(0)| ≥δ >0. Then Φhas no zeros on {0<|z|< δ}. 4Communicated
to us by A. Sasane.
18 Figure 1. The curves y2=1√ 3arsinhq 1 4e−6x−cos2(√ 3x) and y1= 2 cos(√ 3x) +e−3xands0=π 2√ 3
We first note that for z=x+iywith x≥0 and |y| ≤1,gis bounded by 8 : |g(z)| ≤ 2|cos(√ 3z)|+e−3x= 2(q
cos2(√ 3x) + sinh2(√ 3y)) + 1 ≤2(q 1 + sinh2√ 3) + 1 = 2 cosh(√ 3) + 1 ≤e√ 3+ 2≤8. Moreover |g′(rk)| ≥1,
since g′(z) =−2√ 3 sin(√ 3z)−3e−3z, and so |g′(rk)|=| −2√ 3 sin(√ 3rk) + 6 cos(√ 3rk)| →2√ 3∼3.4641···.
respectively |g′(rk)|=|(−1)k√ 3p 4−e−6rk−3e−3rk| Figure 2. The curves y=| ±√ 3√ 4−e−6x−3e−3x| The
value of the lower branch at x= 1/6 is∼1.481371 ···and the branch is increasing. So |g′(rk)| ≥1 for reven.
The upper branch is always bigger than 3. So, |g′(rk)| ≥1, too if ris odd.
19 We conclude from the zero set criterium 1 that g■= 0 on D(rk,1/8), excepted at rk. •The final part
which remains is now the proof that the graph of the (different branches) of the curve y2is contained in
the union of the disks D(rk,1/8). We first determine the positive zeros of y2. These are given by e−6x−4
cos2(√ 3x) = 0. Of course the rk, the positive zeros of 2 cos(√ 3x)−e−3xare zeros of y2, too. The other
ones are given by the zeros of the function h∗(x) = 2 e3xcos(√ 3x)−1 =h(x)−2. This works as in the proof
oh Property 3; note that ( h∗)′=h′. h∗(an) = 2 e3an−1>0, h(bn) =−1, h(cn) =−2e3cn−1<0, h(dn) =−1,
h(en) = 2 e3en−1>0. Since an<π 3+2nπ√ 3< bn<π+2nπ√ 3=cn<4π 3+2nπ√ 3<3π 2+2nπ√ 3=dn, and h∗is
increasing on [2nπ√ 3,π 3+2nπ√ 3], decreasing on [π 3+2nπ√ 3,4π 3+2nπ√ 3] and increasing on [4π
3+2nπ√ 3,2π+2nπ√ 3], we see that there are exactly two zeros on [ an, en], namely one betweenπ
3+2nπ√ 3andbn=π 2+2nπ√ 3and one between dn=3π 2+2nπ√ 3anden=2π+2nπ√ 3. We enumerate these
in an increasing order, say r∗ n, and r∗ 2n< s2n< r2nas well as r2n+1< s2n+1< r∗ 2n+1. As in the proof
of Property 5 (1) we see that |r∗ n−sn|<1 16√ 3. In fact, Forn= 2meven, it is sufficient to prove that p∗
n:=sn−1 16√ 3< r∗ n< sn, where sn, r∗ n∈Im=h 2πm√ 3,2πm+2π√ 3i . Since h(sn) =−1 and h(r∗ n) = 0,
and his decreasing on [π 3+2mπ√ 3,4π 3+2mπ√ 3]⊇ [p∗ n, sn], this is done by showing that h∗(pn)>0.
Now, noticing that the cosinus term is positive, h∗(p∗ n) = 2 e√ 3(π 2+nπ−1 16)cos■π 2−1 16■ −1 ≥2e√
3(π 2−1 16)cos■π 2−1 16■ −1∼0.7029349 ···>0. Forn= 2m+ 1 odd, it is proved in the same way that
sn< r∗ n< q∗ n:=sn+1 16√ 3, where sn, r∗ n∈Im. In fact, h∗is increasing on [4π 3+2mπ√ 3,2π+2mπ√
3]⊇[sn, q∗ n]. So, by noticing that the cosinus term is positive, h∗(q∗ n) = 2 e√ 3(3π 2+(n−1)π+1
16)cos■3π 2+1 16■ −1 ≥2e√ 3(3π 2+1 16)cos■3π 2+1 16■ −1∼486.971814 ··· Next we estimate the
local maxima of y2. Since r∗ n≥r∗ 0> π/ 3, y2≤1√ 3arsinhr 1 4e−6π/3=1√ 3arsinhe−π 2∼0.01247381
···<0.02 := ρ Now the rectangle R:=■ r0−1 8√ 3, r0+1 8√ 3■ ×[0, ρ]⊆D(r0,1 8), because the vertex (1 8√
3, ρ) satisfies ■1 8√ 3■2 +ρ2∼0.00568 ···<0.01562 =1 64.
20 It is easily shown that {(x, y2(x)) :r∗ 2n≤x≤r2n}and{(x, y2(x)) :r2n+1≤x≤r∗ 2n+1}are contained in the
rectangle R. ■ •Having only the real zeros rkoffin the right-half plane, the symmetry f(z) = 0 ⇐⇒
f(ei2π/3z) = 0 now implies that no other zeros are in the left half-plane excepted the rotations
ei2π/3rkandei4π/3rkas any ζ:=reiθwith π/2≤θ≤3π/2 would yield that at least one of the points
ei4π/3ζorei2π/3ζbelongs to the right-half plane. This finishes the proof of Property 6. ■ A detailed
analysis of the zeros of the functions eaz+e−az+ 1 appears in [1]. That paper is based on the preceding
methods and text.
21 Solution to problem 12470 in Amer. Math. Monthly 131 (2024), p. 536 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that■ ■
S:=∞X n=11 2nlog■tanh 2n tanh 2n−1■ = log( e2+ 1)−2∼0.1269280110 ···. One has to transform this
into a telescoping series. tanh 2n tanh 2n−1=sinh 2n sinh 2n−1cosh 2n−1 cosh 2n=2 sinh 2n−1cosh
2n−1 sinh 2n−1cosh 2n−1 cosh 2n = 2(cosh 2n−1)2 cosh 2n. Hence 1 2nlog■tanh 2n tanh 2n−1■ =1
2n■ log 2 + 2 log(cosh 2n−1)−log cosh 2n■ =log 2 2n+1 2n−1log(cosh 2n−1)−1 2nlog(cosh 2n). Note
that εn:=1 2nlog(cosh 2n)→1 since (by using l’Hospital’s rule). lim x→∞log(ex+e−x) x= 1. Consequently
the series below converges and S=∞X n=1log 2 2n+∞X n=1■1 2n−1log(cosh 2n−1)−1 2nlog(cosh 2n)■
= 1·log 2 + log cosh 1 −lim nεn= log 2 + log■e1+e−1 2■ −1 = log( e2+ 1)−2.
22 Solution to problem 12460 in Amer. Math. Monthly 131 (2024), 354 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - (2) =⇒(1): If f(x)
=mx+candg(x) =m, then trivially S(n) :=f(an) +bng(bn) =man+c+bnm=m(an+bn) +c and so S(n)
converges whenever an+bnconverges. (1)=⇒(2): Let N:={0,1,2, . . .}. Fix a, q∈Q. Consider the
sequences (an)n∈N= (a+q, a, a +q, a, a +q, a, . . . ) and ( bn)n∈N= (−q,0,−q,0, . . .). Then an+bn=afor
all n. Moreover sn:=f(an) +bng(bn) =( f(a+q)−qg(−q) if nis even f(a) if nis odd .. As by assumption ( sn)
converges, we deduce that (12) f(a+q)−qg(−q) =f(a) Next consider the sequences ( an)n∈N= (q,0, q,0, .
. .) and ( bn)n∈N= (−q,0,−q,0, . . .). Then an+bn= 0 and rn:=f(an) +bng(bn) =( f(q)−qg(−q) if nis even f(0)
if nis odd . Since by assumption also ( rn) converges, we have (13) f(q)−qg(−q) =f(0). Now (12)–(13)
yields that for every q∈Qanda∈Q f(a+q)−f(q) =f(a)−f(0). Since fis assumed to be continuous, f(a+x)−f(a)
=f(x)−f(0) for every x∈R. This implies that fis an affine function. In fact, let h(x) =f(x)−f(0). Then h(a+x)
=h(x) +h(a), that is, his a continuous additive function. By a classical result due to Cauchy, his linear;
that is h(x) =mxfor some m∈R. Consequently f(x) =mx+f(0). Let c:=f(0). Then the condition on fandghas
the form Sn=f(an) +bng(bn) =man+c+bng(bn) =c+m(an+bn) +bn(g(bn)−m). Letp, q∈Qand consider the
sequences ( bn)n∈N= (p, q, p, q, . . . ) and ( an)n∈N= (−p,−q,−p,−q, . . . ). Then an+bn= 0 for all nand,
by assumption, Sn=( c+p(g(p)−m) if nis even c+q(g(q)−m) if nis odd converges. Hence the function
x(g(x)−m) must be constant on Q, hence on R(due to conti- nuity). Consequently g(x) =mfor every x∈R.
23 Solution to problem 12459 in Amer. Math. Monthly 131 (2024), 354 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We prove that for a >1
■ ■ I(a) :=ˆ∞ 0Li2(−xa) + Li 2(−x−a) 1 +xadx=π3 3a■sin2(π/a)−3 sin3(π/a)■ . To start with, we use the
known formula [24] Li2(z) + Li 2(1 z) =−π2 6−1 2log2(−z), z∈C\[0,∞[, forz=−xa. So the integral to be
computed is I(a) =−π2 6ˆ∞ 01 1 +xadx−1 2ˆ∞ 0log2(xa) 1 +xadx. The change of variable xa7→e−tnow
yields I(a) =−π2 6aˆ∞ −∞e−t a 1 +e−tdt−1 2aˆ∞ −∞t2e−t a 1 +e−tdt. We solve this with the help of the
residue theorem. So, for m= 0 or m= 2, let fm(z) :=zme−z a 1 +e−z. In order the obtained series
converge and the path-integrals tend to 0 when ”blowing up” the contours, we consider for 0 < r < 1 the
auxiliary functions ur(z) :=r−iz/2π:=e−ilogr 2πz (which converge locally uniformly to 1 as r→1) and
Fm,r(z) :=fm(z)ur(z), and calculate the integral (14) Jr(a) :=−π2 6aˆ∞ −∞e−t ar−it/2π 1 +e−tdt−1 2aˆ∞
−∞t2e−t ar−it/2π 1 +e−tdt. As|ur(t)| ≤1, we deduce from Lebesgue’s dominated convergence theorem
that lim r→1Jr(a) =I(a). Note that Fm,ris meromorphic in Cwith simple poles at zn=iπ(1 + 2 n) for n∈Z.
We integrate Fm,rover the positively oriented boundary Γ N=γ1+γ2+γ3of the rectangles
[−2Nπ,2Nπ]×[0,2Nπ], where N∈N∗. Let sN:= 2πN. Then γ1(t) = sN(1 +it),0≤t≤1, γ[−1] 2(t) =
sN(t+i),−1≤t≤1, γ[−1] 3(t) = sN(−1 +it),0≤t≤1.
24 By the residue theorem ˆ ΓNFm,r(z)dz= 2πiN−1X n=0Res ( Fm,r, zn). Now Fm,r=g/hand so Res (
Fm,r, zn) =g(zn) h′(zn). Moreover,ˆ ΓNFm,r(z)dz→0 asN→ ∞ . To see this we have to consider three
cases: Since on γjforj= 1,3, we have |ur(γj(t))| ≤etNlogr≤1, we see that ˆ γ1Fm,r(z)dz ≤ˆ1 0|sN(1
+it)|m|e−sN(1+it)/a||ur(sN(1 +it))| |1 +e−sN(1+it)|sNdt ≤CNm+1e−sN/a 1−e−sN→0 asN→ ∞ . ˆ
γ3Fm,r(z)dz ≤ˆ1 0|sN(−1 +it)|m|e−sN(−1+it)/a||ur(sN(−1 +it))| |1 +e−sN(−1+it)|sNdt ≤CNm+1esN/a
esN−1·e−sN e−sN =CNm+1e−sN(1−1 a) 1−e−sN→0 asN→ ∞ . Next we observe that on γ2we have
|u(γ2(t))|= e−iNlogr(t+i) =eNlogr, and that Nm+1eNlogr=e(m+1) log N+Nlogr→0 asN→ ∞ . Hence ˆ
γ2Fm,r(z)dz ≤ˆ1 −1|sN(i+t)|m|e−sN(t+i)/a||ur(sN(i+t))| |1 +e−sN(t+i)|sNdt ≤CNm+1eNlogrˆ1 −1e−sNt/a 1
+e−tsNdt ≤CNm+1eNlogr ˆ1 0e−sNt/a |{z} ≤1dt+ˆ0 −1esNt(1−1 a)|{z} ≤1dt  →0·0 asN→ ∞ , where
the property lim´ =´ lim is used (Lebesgue’s dominated convergence theorem: the integrands are
bounded (in moduli) by 1 and converge to 0 on the associated open intervals). By letting N→ ∞ , we
conclude that Jr(a) = 2 πi −π2 6a∞X n=0Res ( F0,r, zn)−1 2a∞X n=0Res ( F2,r, zn)! . That the series
converge will be clear in a moment. To this end, we need to calculate the residua. Note that ur(zn)
=e−ilogr 2πiπ(1+2n)= (√r)1+2n. Hence Res ( Fm,r, zn) = zm ne−zn/a −e−znur(zn) =−(iπ(1 + 2
n))me−iπ(1+2n)/a e−iπ(1+2n)ur(zn) = (√r)1+2n(iπ(1 + 2 n))me−iπ(1+2n)/a.
25 Letζ:=e−iπ/a. Then Jr(a) = 2 πi −π2 6a∞X n=0(√rζ)2n+1−1 2a∞X n=0(iπ(1 + 2 n))2(√rζ)2n+1! = 2 πi
−π2 6a∞X n=0(√rζ)2n+1+π2 2a∞X n=0(2n+ 1)2(√rζ)2n+1! . It is straightforward to check the following
result: (15)∞X n=0(2n+ 1)2z2n+1=zz4+ 6z2+ 1 (1−z2)3=:S(z). Hence Jr(a) = 2 πi■ −π2 6a√rζ 1−rζ2+π2
2aS(√rζ)■ , and so I= lim r→1Jr(a) = 2 πi■ −π2 6aζ 1−ζ2+π2 2aS(ζ)■ . Note that ζζ= 1, but ζ■= 1. A
short calculation yields S(ζ) =ζ2+ 6 + ζ2 (ζ−ζ)3. Consequently, by using that ζ2+ 6 + ζ2= (ζ−ζ)2+ 8, I= 2
πi −π2 6aζ 1−ζ2+π2 2aζ2+ 6 + ζ2 (ζ−ζ)3! = 2 πi■ −π2 6a1 ζ−ζ+π2 2a(ζ−ζ)2+ 8 (ζ−ζ)3■ = 2 πi■ −π2
6a(ζ−ζ)2 (ζ−ζ)3+π2 2a(ζ−ζ)2+ 8 (ζ−ζ)3■ =−π3 3a■ i(−4) sin2(π/a) (2i)3sin3(π/a)−i(−12) sin2(π/a) + 24
(2i)3sin3(π/a)■ =−π3 3a■3−sin2(π/a) sin3(π/a)■ . Remark A more classical way is to compute I(a) =−π2
6ˆ∞ 01 1 +xadx−1 2ˆ∞ 0log2(xa) 1 +xadx ”directly” without the change of variable by applying the
residue theorem to the functions 1 1+za,logz 1+zaandlog2z 1+zafor the standard branches of the
power and logarithm and the boundary of the sectors {z∈C:|z|< R, 0≤argz≤2π a}, which contains one
simple pole.
26 Solution to problem 12451 in Amer. Math. Monthly 131 (2024) Raymond Mortini and Rudolf Rupp - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - As usual, M0=Iswhere Mis a
square s×smatrix. Via induction ■A B 0 0■k =■ AkAk−1B 0 0■ . Hence exp■A B 0 0■ =∞X k=01 k!■A
B 0 0■k =In+m+∞X k=11 k!■ AkAk−1B 0 0■ =■ expAP∞ k=11 k!Ak−1B 0 Im■ . Butˆ1 0exp(At)dt=∞X
j=0Ajˆ1 0tj j!=∞X j=01 (j+ 1)!Aj. Hence exp■A B 0 0■ = expA■´1 0exp(At)dt■ B 0 Im! .
27 Solution to problem 12436 in Amer. Math. Monthly 130 (2023) Raymond Mortini and Rudolf Rupp - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that■ ■ P(x) :=nY
k=1■ x+ sin2■kπ 2n■■ = 2−2n+2(x+ 1)Un−1(2x+ 1) , where U0= 1 and Un(x) =■n 2■X k=0(−1)k■n−k
k■ (2x)n−2kn≥1= 2nnY k=1■ x−cos■kπ n+ 1■■ is the Chebyshev polynomial of the second kind. This
is very easy, though. P(x) =nY k=1■ x+ sin2■kπ 2n■■ =nY k=1■ x+1 2■ 1−cos■kπ n■■■ = 2−nnY
k=1■ 2x+ 1−cos■kπ n■■ = 2−nn−1Y k=1■ 2x+ 1−cos■kπ n■■ (2x+ 1−cos(π)) = 2−2n+2(x+
1)Un−1(2x+ 1).
28 Solution to problem 12433 in Amer. Math. Monthly 130 (2023), ? Raymond Mortini and Rudolf Rupp
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetG:={z∈C: Imz <0.5}be the
shifted lower half-plane, and let log z= log|z|+iargz with−π <argz < π be the standard holomorphic
branch of the logarithm. Since for z∈Gwe have Re (0 .5 +iz)>0, the function (0.5 +iz)x=exlog(0.5+iz) is
well defined and holomorphic in G. Consequently, the function f(z) :=tanh( πz) (0.5 +iz)x, is
meromorphic in Gwith simple poles at zk:=−i(0.5 +k)∈G, where k∈N={0,1,2, . . .}. We apply now the
residue theorem to f. To this end, we integrate for N≥1 the function falong the positively oriented
boundary Γ Nof the rectangles RN:= [−N, N ]×[0,−N] and conclude thatˆ ΓNf(z)dz= 2πi∞X k=0n(ΓN,
zk)Res( f, zk), where n(Γ, z) denotes the number of times the point zis surrounded by Γ. Observe that at
most a finite number of terms in this sum are not equal to 0 as n(ΓN, zk) =( 1 if k= 0,1, . . . , N 0 if k > N.
Let us calculate the residue now. We use the formula Res (g h, a) =g(a) h′(a), whenever ais a simple
zero of h. That is, when we choose g(z) =sinh(πz) (0.5 +iz)xandh(z) = cosh( πz), Res(f, zk) =sinh(πz) (1
+k)x1 πsinh(πz) z=−i(0.5+k)=1 π1 (1 +k)x. It remains to show that the integral along the three parts Γj
Nof Γ Nthat are contained in the lower half plane Im z <0 tends to zero. First note that tanh( πz)
=e2πz−1 e2πz+ 1. i) Let z(t) =−N−it, where 0 ≤t≤N. Then for n≥N0, |tanhz(t)|= e−2πNe−2iπt−1
e−2πNe−2iπt−1 ≤1 +e−2πN 1−e−2πN≤2. Moreover, |0.5 +iz(t)|x=|0.5−iN+t|x≥Nx. ii) Let z(t) =N−it, where
0 ≤t≤N. Then for n≥N0, |tanhz(t)|= e2πNe−2iπt−1 e2πNe−2iπt+ 1 ≤e2πN+ 1 e2πN−1≤2. Moreover
29 |0.5 +iz(t)|x=|0.5 +iN+t|x≥Nx. iii) Let z(t) =t−iNwhere −N≤t≤N. Then |tanhz(t)|= e2πte−2iπN−1
e2πte−2iπN+ 1 =e2πt−1 e2πt+ 1≤1. Moreover |0.5 +iz(t)|x=|0.5 +it+N|x≥(N+ 0.5)x≥Nx. Since x >1, we
conclude that for N≥N0 3X j=1ˆ Γj Nf(z)dz ≤2X j=1ˆN 0|f(zj(t))|dt+ˆN −N|f(z3(t))|dt ≤2·2N Nx+ 2N1 Nx=6
Nx−1→0 asN→ ∞ . We conclude that i 2ˆ∞ −∞tanh( πt) (0.5 +it)xdt=−i 22πi∞X k=01 π1 (1 +k)x=ζ(x).
Note that the minus sign comes from the fact that the upper boundary of the rectangle RNis run through
from the right to the left.
30 Solution to problem 12422 in Amer. Math. Monthly 130 (2023), 862 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We solve (a). Put
R:=∞X n=11 an2+bn+c. Letr1andr2be the zeros of the polynomial p(x) =ax2+bx+c. Suppose that a■= 0
and b2−4ac=k2a2for some k∈ {1,2,3, . . .}. Then r1=−b−ka 2aandr2=−b+ka 2a andr1−r2=−k. As an
example we mention a= 1,b= 5 and c= 4,k= 3,r1=−4,r2=−1. Now the partial fraction decomposition of 1
/p(n) reads as 1 an2+bn+c=1 a(n−r1)(n−r2)=1 a(r1−r2)■1 n−r1−1 n−r2■ . Hence, for n≥n0>1 and
n0chosen so that n−rj−1>0, S:=∞X n=n01 an2+bn+c=1 a(r2−r1)∞X n=n0ˆ1 0(xn−r2−1−xn−r1−1)dx =1
ak∞X n=n0ˆ1 0xn−1(x−r2−x−r1)dx(1)=1 akˆ1 0(x−r2−x−r1)∞X n=n0xn−1dx =1 akˆ1 0xn0−1x−r2−x−r1
1−xdx=1 akˆ1 0xn0−1x−r21−xr2−r1 1−xdx =1 akˆ1 0xn0−r2−11−xk 1−xdx=1 akˆ1 0k−1X
j=0xn0−1−r2+jdx =1 akk−1X j=01 n0−r2+j. Hence Sis rational, and therefore Ris rational, too. Note that
in (1) the interchanging´P=P´ is possible, since xn−1(x−r2−x−r1) has constant sign.
31 Solution to problem 12415 in Amer. Math. Monthly 130 (2023), 765 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Letn∈N={0,1,2, . . .}and
let Sn:=2nX j=0jX k=■j/2■■2n+ 2 2k+ 1■ ■n+ 1 2k−j■ . We show that ■ ■■ ■Sn= 23n+1. First we
interchange the two summations. Figure 3. k≤j≤2k+ 1,k= 0,1,2,3,4, or■j/2■ ≤k≤jforj= 0,1,2,3,4 Sn=2nX
k=02k+1X j=k■2n+ 2 2k+ 1■ ■n+ 1 2k−j■ =2nX k=0 ■2n+ 2 2k+ 1■2kX j=k■n+ 1 2k−j■  =nX
k=0"■2n+ 2 2k+ 1■kX m=0■n+ 1 m■# It is well known thatnX k=0■2n+ 2 2k+ 1■ = 22n+1. In fact
22n+2=2n+2X k=0■2n+ 2 k■ and 0 = (1 + ( −1))2n+2=2n+2X k=0(−1)k■2n+ 2 k■ . Substraction yields
that 22n+2= 22n+2X k=0 kodd■2n+ 2 k■ = 2nX m=0■2n+ 2 2m+ 1■ .
32 -101234567812345k j Figure 4. k≤j≤2k,k= 0,1,2,3,4, or■j/2■ ≤k≤jforj= 0,1,2,3,4 -101234567812345k
j Also, 2n+1= (1 + 1)n+1=n+1X j=0■n+ 1 j■ =kX j=0■n+ 1 j■ +n+1X j=k+1■n+ 1 j■ =kX j=0■n+ 1 j■
+n+1X j=k+1■n+ 1 n+ 1−j■ = i=n+1−jkX j=0■n+ 1 j■ +n−kX i=0■n+ 1 i■ . Hence Sn=nX k=0■2n+ 2
2k+ 1■kX m=0■n+ 1 m■ = k=n−jnX j=0■2n+ 2 2n−2j+ 1■n−jX i=0■n+ 1 i■ =nX j=0■2n+ 2 (2n+
2)−(2n−2j+ 1)■n−jX i=0■n+ 1 i■ =nX j=0■2n+ 2 1 + 2 j■n−jX i=0■n+ 1 i■ = j→knX k=0■2n+ 2 2k+
1■n−kX i=0■n+ 1 i■ .
33 Addition yields 2Sn=nX k=0■2n+ 2 2k+ 1■kX j=0■n+ 1 j■ +nX k=0■2n+ 2 2k+ 1■n−kX i=0■n+ 1
i■ =nX k=0■2n+ 2 2k+ 1■ kX j=0■n+ 1 j■ +n−kX i=0■n+ 1 i■  = 22n+1·2n+1= 23n+2. Hence Sn=
23n+1. Remarks (1)Note that Sn= j=k−mnX k=0 ■2n+ 2 2k+ 1■kX j=0■n+ 1 k−j■  This has the
form∞X k=0akkX j=0bk−j, which is a little bit different from the Cauchy product ∞X k=0ak! ∞X k=0bk!
=∞X k=0kX j=0ajbk−j. (2) Replacing ■j/2■by■j/2■yields the same result Rn:=2nX j=0jX k=■j/2■■2n+
2 2k+ 1■ ■n+ 1 2k−j■ = 23n+1 (see below), alhough the associated index-grid is different (see figure 3
and 4). Just note that Rn=2nX k=02kX j=k■2n+ 2 2k+ 1■ ■n+ 1 2k−j■ =2nX k=02k+1X j=k■2n+ 2 2k+
1■ ■n+ 1 2k−j■ =Sn
34 Solution to problem 12407 in Amer. Math. Monthly 130 (2023), ** Raymond Mortini and Rudolf Rupp
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Given r >0, let I(r) :=ˆ∞ 0xr−1
(1 +x2)(1 + x2r)dx. We show that
■ I(r) =π 4r. First it is clear that the integral converges since at ∞we have that the integrand fr(x) is
similar to 1/xr+3and at 0 fr(x) is similar to xr−1, where r−1>−1. We make the change of the variable
x→1/y. Then I(r) =ˆ∞ 0y1−r (1 +y−2)(1 + y−2r)dy y2=ˆ∞ 0y1+r (1 +y2)(1 + y2r)dy =ˆ∞ 0(y2+ 1−1)yr−1 (1
+y2)(1 + y2r)dy=ˆ∞ 0yr−1 1 +y2rdy−I(r). Hence 2I(r) =1 rarctan( yr) ∞ 0=π 2r, from which we deduce
that I(r) =π/(4r).
35 Solution to problem 12406 in Amer. Math. Monthly 130 (2023), 679 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Letp∈R. Consider the
functional equation (16) f(x2) + 2pf(x) = (x+p)2. We claim that all solutions of (16) on [0 ,1] and
continuous at {0,1}are actually continuous on [0,1] and are given by f(x) =x+p2 1 + 2 p whenever
p■=−1/2. •Ifp=−1/2, then f(x2)−f(x) = (x−1 2)2has no solution on [0 ,1] (independently of being
continuous or not) since for x= 1, we would get 0 = f(1)−f(1) = 1 /4. •Ifp= 0, then f(x2) =x2implies that on
[0 ,1] one has f(x) =x. •Letp■=−1/2. We first determine the polynomial solutions. So let qbe a
polynomial solving (16). Then the degree of qis at most 1. Say q(x) =ax+b. Pulling into the functional
equation yields ax2+b+ 2p(ax+b) =x2+ 2px+p2 or equivalently (a−1)x2+ 2p(a−1)x+b(1 + 2 p)−p2= 0.
Hence a= 1 and b=p2 1+2p. It is straighforward to check that q(x) =x+p2 1+2pis indeed a solution to
(16). We conclude that all polynomial solutions are given by the linear function qabove. Next we
determine the general solution (16). So let fbe a solution on [0 ,1] continuous at 0,1. Now put h(x) :=
f(x)−q(x). Then hsatisfies on [0 ,1] the functional equation (of Schroeder type) (17) h(x2) =−2ph(x). Of
course this implies that h(0) = 0. i) Let p <−1/2 orp≥1/2. Via induction h(x2n) = (−2p)nh(x). Since his
continuous at 0, and x2n→0 for 0 < x < 1,h(0) = 0, and |2p|n→ ∞ respectively (−2p)n= (−1)nifp= 1/2, we
deduce that h(x) = 0 for 0 < x < 1, too. ii) If 0 <|p|<1/2, we rewrite (17) as (18) h(√x) =−1 2ph(x). Via
induction h(x1/2n) =■ −1 2p■n h(x). Since his assumed to be continuous at 1 and h(1) = 0 by (17),■ 1
2p■n → ∞ implies that 0 =h(1) =∞ ·h(x),
36 and so h(x) = 0 for x >0. We conclude that for p■=−1/2, the general solution to (16) on [0 ,1], and
continuous at 0 ,1, is given by our poynomial q(x) =x+p2 1 + 2 p. Thus the solution is completely
established. The Schr¨ oder type functional equations f(xm) =rf(x) are analyzed in detail in [2].
37 Solution to problem 12398 in Amer. Math. Monthly 130 (2023), 587 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We suppose that this
agglomeration csch of letters is nothing but 1 /sinh. So let S:=∞X n=01 sinh 2n. We prove that■ ■■
■S=2 e−1. This is very simple though. Since 2 sinh x=ex−e−xand (e2n+ 1)( e2n−1) = e2n+1−1, we
obtain S= 2∞X n=01 e2n−e−2n= 2∞X n=0e2n+1−1 e2n+1−1 = 2∞X n=0■1 e2n−1−1 e2n+1−1■ =2 e−1.
Another possibility would be to use the formula 1 sinhx= coth( x/2)−cothx. Then S=∞X n=0■
coth(2n−1)−coth 2n■ = coth(1 /2)−1 =2e−1/2 e1/2−e1/2=2 e−1.
38 Solution to problem 12389 in Amer. Math. Monthly 130 (2023), 386 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Our tool will be the fact
that for Hn:=nX j=11 jwe have Hn−logn■γ, where γis the Euler-Mascheroni constant. First note that ∞X
j=1N (N+j)2≤∞X j=1N (N+j)(N+j−1) =N∞X j=1■1 N+j−1−1 N+j■ = 1 . Fix 0 < x < 1 and let N:=N(x) :=■1
x■ . Let ε∈]0,1/2]. Since |siny| ≤yfory≥0, we obtain for ∞X n=1|sin(nx)| n2≤NX n=1nx n2+∞X n=N+11
n2≤xNX n=11 n+1 N ≤x■ HN−logN−γ■ +xγ+xlogN+1 N. Hence, for xsmall enough, Nis big, and so H(x)
:=P∞ n=1|sin(nx)| n2 xlog(1/x)≤1 log(1/x)+γ log(1/x)+log■1 x■ log(1/x)+1 xlog(1/x)■1 x■. We conclude
that 0≤lim sup x→0H(x)≤0 + 0 + 1 + 0 = 1 . Now we estimate lim inf x→0H(x). Let ε∈]0,1/2]. Since
x7→(sinx)/xis decreasing on [0, π/2], we see that for 0 < u≤ε sinu u≥sinε ε. For 0 < x < ε putN:=N(x) :=jε
xk . Then, N > 0 and for n≤Nwe have nx≤Nx=jε xk x≤ε xx=ε, and so sin(nx) nx≥sinε ε. Hence ∞X
n=1|sin(nx)| n2≥sinε εNX n=1nx n2=xsinε εNX n=11 n ≥xsinε εlogN.
39 We deduce that for 0 < x < ε H(x)≥sinε εlog■ε x■ log(1/x)≥sinε εlog(ε x−1) log(1/x) =sinε
εlog(ε−x)−logx −logx. Since lim x→0log(ε−x) −logx= log ε·0 we conclude that lim inf x→0H(x)≥sinε ε.
Now ε→0 yields that lim inf x→0H(x)≥1. Consequently lim x→0H(x) = 1.
40 Solution to problem 12388 in Amer. Math. Monthly 130 (2023), 385 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Fora∈[0,2π], let I(a)
:=ˆ∞ 0(logx)2arctan x 1−2xcosa+x2dx. We prove that I(a) =  πa sina(2π−a)(π−a)
12if 0< a < 2π,a■=π π3 6ifa= 0 or a= 2π π3 12ifa=π. Ifais arbitrary, we replace abya−2kπ, where k∈Zis
chosen so that 2 kπ≤a <2(k+ 1)π. First we let ”disappear” the arctangent: the substitution u=
1/x,dx=−1/u2and the formula arctan(1 /x) + arctan x=π/2 for x >0 yield I(a) =ˆ∞ 0(logu)2■ π 2−arctan u■
1−21 ucosa+1 u2du u2=−I+π 2ˆ∞ 0(logx)2 1−2xcosa+x2dx, and so I(a) =π 4ˆ∞ 0(logx)2
1−2xcosa+x2dx. Using again the transformation u= 1/x, we obtain that ˆ1 0(logx)2 1−2xcosa+x2dx=ˆ∞
1(logx)2 1−2xcosa+x2dx, and so■ ■ I(a) =π 2ˆ1 0(logx)2 1−2xcosa+x2dx. Next we use that for a /∈
{kπ:k∈Z} 1 1−2xcosa+x2=A x−eia−A x−e−ia, where A=−i 2 sina. Hence, in that case, I(a) = 2Re■π 2Aˆ1
0(logx)2 x−eiadx■ =π 2 sinaImˆ1 0(logx)2 x−eiadx =π 2 sinaIm (−eia)ˆ1 0(logx)2 1−xe−iadx
41 SinceP∞ n=0xn(logx)2is an L1(0,1)-majorant, we have´P=P´ . Thus I(a) =−π 2 sinaIm■∞X
n=0e−ia(n+1)ˆ1 0xn(logx)2dx■ . By twice partial integration, ˆ1 0xn(logx)2dx=2 (n+ 1)3. We conclude
that I(a) =−π 2 sinaIm■∞X n=0e−ia(n+1) 2 (n+ 1)3■ =π sina∞X n=0sin(n+ 1)a (n+ 1)3. Let h(a) :=∞X
n=0sin(n+ 1)a (n+ 1)3. Then h′(a) =∞X n=0cos(n+ 1)a (n+ 1)2. Since1 3π2+ 4P∞ n=1cosnx n2is the
Fourier series of the function ( x−π)2, 0≤x <2π, extended 2π-periodically, we see that for 0 < a < 2π,
h′(a) =(a−π)2 4−π2 12. Ash(0) = 0, we deduce that for 0 < a < 2π, h(a) =(a−π)3 12−π2 12a+π3
12=a3−3πa2+ 2π2a 12. Consequently, for 0 < a < 2π,a■=π, I(a) =πa sina(2π−a)(π−a) 12. Now let an■0
and fn(a) :=(logx)2 1−2xcosan+x2. As fnis positive and increases to(logx)2 (1−x)2, we deduce from
Beppo-Levi’s monotone convergence theorem that I(an)→I(0). Hence I(0) = π3/6. Moreover, I(bn)→π3
12asbn■π. This is also the value of I(π) =π 2ˆ1 0(logx)2 (1 +x)2dx. Just write (logx)2 (1 +x)2=∞X
n=1(−1)n−1nxn−1(logx)2, and use again that´P=P´ . Finally, I(2π) =I(0) = π3/6.
42 Solution to problem 12380 in Amer. Math. Monthly 130 (2023), 285 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Lemma Letf,g,hbe
positive increasing functions on [0 ,∞[ satisfying for every xj≥0 and 0≤tj≤1 withPn j=1tj= 1 the concavity
inequality (19) f■nX j=1tjxj■ ≥nX j=1tjf(xj) and similarily for g, h. Let M2, Pj∈[0,∞[×[0,∞[. Then the
function Ggiven by G(M2) :=G(x, y) :=f(x+g(y)) also satisfies G■nX j=1tjPj■ ≥nX j=1tjG(Pj), Similarily, if
M3, Qj∈[0,∞[×[0,∞[×[0,∞[, then the function Hgiven by H(M3) :=H(x, y, z ) :=f(x+g(y+h(z))) satisfies
H■nX j=1tjQj■ ≥nX j=1tjH(Qj). Proof G■nX j=1tjxj,nX j=1tjyj■ =f■nX j=1tjxj+g■nX j=1tjyj■■ ≥f■nX
j=1tjxj+nX j=1tjg(yj)■ =f■nX j=1tj(xj+g(yj))■ ≥nX j=1tjf(xj+g(yj)) =nX j=1tjG(Pj).
43 Now applying this, we get H(nX j=1tjQj) = f■nX j=1tjxj+g■nX j=1tjyj+h(nX j=1tjzj)■■ ≥f■nX
j=1tjxj+nX j=1tjg(yj+h(zj))■ =f■nX j=1tj(xj+g(yj+h(zj))■ ≥nX j=1tjf(xj+g(yj+h(zj))) =nX j=1tjH(Qj). Now we
are ready to give the solution to the problem. Let S(a, b, c ) :=mr a+nq b+p√c+mr b+nq c+p√a+mr c+nq
a+p√ b. ForM:= (x, y, z )∈R3, x, y, z ≥0, let f(M) :=f(x, y, z ) :=mr x+nq y+p√z. By Lemma, for P:= (a, b, c
),Q= (b, c, a ) and R= (c, a, b ) we have (20)1 3■ f(P) +f(Q) +f(R)■ ≤f■P+Q+R 3■ . Since a+b+c= 3, we
deduce that f(P) +f(Q) +f(R)≤3·f(1,1,1) = 3mq 1 +n√ 2. In case mnp > 1, at least one functionr√xforr∈
{m, n, p }is strictly concave and we have strict inequality in (19) whenever not all the xjare the same and
0 < tj<1. The proof of the Lemma in particular then yields that equality holds in (20) only if P=Q=R, and
so a=b=c. Thus, due to a+b+c= 3, we deduce that a=b=c= 1. Hence S(a, b, c ) = 3mq 1 +n√ 2 if and
only if ( a, b, c ) = (1 ,1,1). Ifm=n=p= 1, then f(M) is linear in R3, and so for all a, b, c with a+b+c= 3 we
have equality: S(a, b, c ) = 3( a+b+c) = 9 = 3mq 1 +n√ 2.
44 Solution to problem 12372 in Amer. Math. Monthly 130 (2023), 187 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Fora > 0, let I:=ˆ1
0log|xa−(1−x)a| xdx, which is a double improper integral (with singularities at 0 ,1/2). We show that
■ I=−a2+ 2 12aπ2. To this end, we first note that xa≤(1−x)aif and only if 0 ≤x≤1/2. Hence, by
substituting x→1−xin the second integral I=ˆ1/2 0log■ (1−x)a−xa■ xdx+ˆ1/2 0log■ xa−(1−x)a■ xdx
=ˆ1/2 0log■ (1−x)a−xa■ xdx+ˆ1/2 0log■ (1−x)a−xa■ 1−xdx =ˆ1/2 0log■ (1−x)a−xa■ x(1−x)dx =ˆ1/2
0log■ 1−■ x 1−x■a■ +alog(1−x) x(1−x)dx. Next we substitute x/(1−x) =y. Equivalently, x=y/(1+y). Note
that 0 →0 and 1 /2→1, dx=1 (1+y)2dyand 1 −x=1 1+y. Hence I=ˆ1 0log(1−ya)−alog(1 + y) y (1+y)21 (1
+y)2dy. Consequently, I=ˆ1 01 xlog■1−xa (1 +x)a■ dx. Using partial integration for´1−η εwith u′:=
1/xandv= log■1−xa (1+x)a■ , and passing to the limits ε, η→0, we obtain I= 0 + aˆ1 0■xa−1 1−xa+1 1
+x■ logx dx =aˆ1 0 ∞X n=0xa−1xna+∞X n=0(−1)nxn! logx dx.
45 Note that Ihas the form I=´P. Now let us calculate J:=P´ . To do this, we apply for β >−1 the formula
ˆ1 0xβlogx dx =−1 (β+ 1)2, (which can easily be obtained by partial integration u= log x, v′=xβ). Hence
J/a =−1 a2∞X n=01 (n+ 1)2+∞X n=0(−1)n+1 1 (n+ 1)2(21) =−1 a2π2 6−π2 12(22) =−a2+ 2 12a2π2. (23)
To finish the proof, we need to show that´P=P´ . As the summands in the first sumP∞
n=0xa−1xnalogxdo not change sign, we may use Beppo-Levi’s theorem. In the second sum,P∞
n=0(−1)nxnlogx, we have absolute convergence, in particular any rearrangement converges (to the
same function), and so we apply Beppo-Levi to the sum over the odd integers and the sum over the
even integers. Thus´P even=P even´ and´P odd=P odd´ . Similarily to (21), it can be shown that the
values ofP odd´ andP even´ are finite. Hence ˆX =ˆX even−ˆX odd=X evenˆ −X oddˆ =Xˆ .
46 Solution to problem 12375 in Amer. Math. Monthly 130 (2023), ?? Raymond Mortini - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - A change of the variable x→1/xyields that
J:=ˆ∞ 0■ 1−x2sin2■1 x■■n dx=ˆ∞ 0(x2−sin2x)n x2n+2dx. Now we ”linearize” the trigonometric
powers: using sin2x= (1/2)(1−cos 2x), we obtain J=1 2ˆ∞ −∞■ x2−1 2+1 2cos(2 x)■n x2n+2dx =1 2ˆ∞
−∞1 x2n+2nX j=0■n j■1 2j■ x2−1 2■n−j cosj(2x)dx. Noticing that cosj(2x) =1 2jjX k=0■j k■ cos(2(
j−2k)x), we finally obtain that with I:= 2J I=ˆ∞ −∞nX ■=01 x2n+2p■(x) cos(2 ■x)dx where p■is a
polynomial of degree at most 2 nand with rational coefficients. Next we consider the functions f(z) :=nX
■=01 z2n+2p■(z)e2i■z and F(z) :=f(z)−p(z) z2n+2, wherep(z) z2n+2=q(z) z2n+2+r zis the principal
part of the meromorphic function f. Note that degp≤2n+ 1, deg q≤2n, and r∈Q+iQ. In particular, Fhas a
holomorphic extension to the origin, hence is an entire function. Therefore´ ΓF(z)dz= 0, where Γ is the
boundary of the half-disk |z| ≤R, Imz≥0, consisting of the half circle Γ Rand the interval [ −R, R]. Hence,
by letting R→ ∞ and taking real parts, 0 = Re lim R→∞ˆ ΓRF(z)dz+I. By Jordan’s Lemma, lim supR→∞
´ ΓReinzdz <∞. Hence, by noticing that the differences of the degrees of the polynomials in the
denominator and numerator if fis bigger than 2, lim R→∞ˆ ΓRF(z)dz= 0 + 0 + lim R→∞ˆ ΓRr zdz=i r π.
We conclude that the value of the original integral Jis rational.
47 . Solution to problem 12362 in Amer. Math. Monthly 129 (2022), 986 Raymond Mortini - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We reduce the present problem to Problem
12340, telling us that for each f: [0,1]→R continuous, (24) lim n→∞n 2nˆ1 0f(x) xn+ (1−x)ndx=π 4f■1 2■
. First we note that one may replace of course nbyt,t→ ∞ . Later we shall take t=n/2. As a result we
obtain (25)■ ■■ ■lim n→∞ˆπ/2 0n■√ 2 cosx■n +■√ 2 sinx■ndx=π 2 To see this, let u:= sin x. Then,
dx= (1−u2)−1/2duand so with In:=ˆπ/2 0n■√ 2 cosx■n +■√ 2 sinx■ndx, we obtain In=nˆ1 0(1−u2)−1/2
2n/2(1−u2)n/2+ 2n/2(u2)n/2dx Now let y:=u2. Then du=1 2√ydy, and so In=n 2·2n/2ˆ1 0y−1/2(1−y)−1/2
(1−y)n/2+yn/2dy Letgε(y) = (y+ε)−1/2(1−y+ε)−1/2andg:=g0. Then (26)n 2·2n/2ˆ1 0gε(y)
(1−y)n/2+yn/2dy≤In. Next we estimate from above. Let x∈[0,1] satisfy |x−1/2| ≥δ, where δ >0 is small.
Then, fort≥1, xt+ (1−x)t≥(1/2 +δ)t+ (1/2−δ)t. Hencet 2t1 xt+ (1−x)t≤t (1 + 2 δ)t+ (1−2δ)t=:mt→0 ast→ ∞
Now for ε >0, choose δso small that |g(x)−g(1/2)|< εfor|x−1/2| ≤δ. Then t 2tˆ1 0g(x) xt+ (1−x)tdx≤t 2tˆ
|x−1/2|≤δg(1/2) +ε xt+ (1−x)tdx+mtˆ |x−1/2|>δg(x)dx ≤t 2tˆ1 0g(1/2) +ε xt+ (1−x)tdx+mt||g||1−→ t→∞π
4(g(1/2) +ε). Together with (26), we obtain that π 4gε(1/2)≤lim inf nIn≤lim sup nIn≤π 4(g(1/2) +ε). Hence,
by letting ε→0, lim nIn=π 4g(1/2) = π/2.
48 Solution to problem 12347 in Amer. Math. Monthly 129 (2022), 786 Raymond Mortini - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that on Rthere are exactly 4
continuous solutions to the functional equation (27) f(f(x))−(a+b)f(x) +abx= 0, whenever f(0) = 0 and 0 <
a < 1< b. Namely F1(x) =ax,F2(x) =bx,F3(x) =( axifx≤0 bxifx >0andF3(x) =( bxifx≤0 axifx >0. It is easy to
check that Fjare solutions. Now suppose that fis a solution. i)fis injective: let f(x) =f(y). Then abx=−f(f(x))
+ ( a+b)f(x) =−f(f(y)) + ( a+b)f(y) =aby and so x=y. ii)fis strictly increasing: monotonicity implies that
M±:= lim x→±∞ f(x) exists in [−∞,∞]. Now M±cannot be finite, since (1) and continuity would imply that
f(M±)− (a+b)M±+±∞= 0, which is impossible. But M+■=−∞, either, since otherwise f(f(x)) +abx=
(a+b)f(x)→ −∞ asx→ ∞ , and so lim x→∞f(f(x)) =−∞. Hence, with y:=f(x)→ −∞ , we deduce that lim
y→−∞f(y) =−∞= lim x→∞f(x), contradicting the monotonicity of f. We conclude that fis strictly increasing,
f(x)≥0 for x≥0,f(x)≤0 for x≤0, and lim x→−∞ f(x) =−∞, lim x→∞f(x) =∞. iii)The inverse
h:=f−1:R→Rsatisfies the functional equation (28) h(h(y))−■1 a+1 b■ h(y) +1 aby= 0. Just take
x:=h(h(y)) in (27) and note that h■f=f■h= id. Then y−(a+b)h(y) +ab h(h(y)) = 0 . Now divide by ab. We
also deduce the following identity: (29)■ f(y)−ay■ +ab■ f−1(y)−(1/a)y■ = 0. In particular f−1is
increasing, too. iv)The only fixed point of fis 0: let f(s) =s. Ifs■= 0, then, by (27) s−(a+b)s+abs= 0. Thus
1 + ab=a+b, or equivalently, b(a−1) = a−1. That is, b= 1 (since a < 1). A contradiction. We conclude that
for x >0 either f(x)< xorf(x)> xfor every x >0. v)Letfn:=f■ ··· ■ f|{z} n-timesbe the n-th iterate of f5. 5We
never use the exponent nto designate the n-th iterate when working with functions, as the risk to mix it
up with the n-th power is too big.
49 •Suppose that there is x0>0 such that f(x0)< x 0. Then fn(x)→0 for every x≥0. Indeed, by iv), 0 <
f(x)< xforx >0. Hence fn+1(x) =f(fn(x))< fn(x) and so M(x) := lim n→∞fn(x) exists for every x > 0.
Plugging fn(x) into the functional equation (27), yields M(x)−(a+b)M(x) +abM(x) = 0 . Consequently,
M(x)(1+ab−(a+b)) = 0. But 1+ ab−(a+b) = (1−a)+b(a−1) = (1 −a)(1−b)■= 0. Hence M(x) = 0. •Suppose
that there is x0>0 such that f(x0)> x0. Then, by iv) f(x)> xfor every x >0 and the sequence ( fn(x)) of
iterates is increasing for each x >0. As its limit M(x) can’t be finite, in particular not 0, we see that lim
n→∞fn(x) =∞for every x >0. vi)For each x∈Rwe obtain the following three terms difference equations:
fn+2(x)−(a+b)fn+1(x) +abfn(x) = 0 , with initial condition f0(x) :=xandf1(x) :=f(x). The associated
characteristic polynomial is p(z) =z2−(a+b)z+ab, which has as roots a andb. Hence, there exist real
coefficients AxandBxdepending on the initial value xsuch that (30) fn(x) =Axan+Bxbn. Iff(x)< xfor every
x >0, then lim nfn(x) = 0 implies that Bx= 0, because b >1 and 0< a < 1. Hence f(x) =Axa. As the initial
value f0(x) equals x, we deduce from (30) that Ax=x. Thus, for x >0,f(x) =axwhenever there exists x0>0
with f(x0)< x0. Iff(x)> x for every x > 0, then x > f−1(x) (note that by iii) h:=f−1is increasing). Hence, the
difference equations, (31) hn+2(x)−■1 a+1 b■ hn+1(x) +1 abhn(x) = 0 with initial values h0(x) =xhave
for x >0 the solutions (32) hn(x) =Cx1 an+Dx1 bn for real coefficients CxandDx. Using (31), we see as
above that lim n→∞hn(x) = 0. Hence Cx= 0. Thus h(x) =h1(x) =Dx1 b. As h0(x) =x, we deduce from (32)
that Dx=x. To some up, f−1(x) =h(x) =x/band so f(x) =bxfor every x >0 whenever there exists x0>0 with
f(x0)> x0. vii) The case for negative arguments follows from the observation that if fis a solution to (27),
then the function ggiven by g(x) =−f(−x) is a solution, too: g(g(x))−(a+b)g(x) +abx =−f(−f(−x)) + (
a+b)f(−x) +abx =−■ f(f(−x))−(a+b)f(−x) +ab(−x)■ = 0.
50 Solution to problem 12340 in Amer. Math. Monthly 129 (2022), 686 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Asg: [0,1]→Ris
continuous, ||g||∞= max {|g(x)|: 0≤x≤1}<∞. Let In:=n 2nˆ1 0g(x) xn+ (1−x)ndx. We claim that lim n→∞In=
(π/4)g(1/2). To see this, we split the integral into two parts and use two different change of variables:
In=n 2nˆ1/2 0g(x) xn+ (1−x)ndx | {z } x=:1 2−s 2n+n 2nˆ1 1/2g(x) xn+ (1−x)ndx | {z } x=:1 2+s 2n =n 2nˆn
0g(1 2−s 2n) (1 2−s 2n)n+ (1 2+s 2n)n1 2nds+n 2nˆn 0g(1 2+s 2n) (1 2+s 2n)n+ (1 2−s 2n)n1 2nds =1
2ˆn 0g(1 2−s 2n) +g(1 2+s 2n) (1−s n)n+ (1 +s n)nds. Note that n7→(1 +s n)nis increasing; so the
integrand is dominated for s≥1 by ||g||∞ (1 +s 2)2≤ ||g||∞4s−2. Hence, as n→ ∞ , lim n→∞In=1
22g(1/2)ˆ∞ 0ds e−s+esds =g(1/2)ˆ∞ 0es 1 + (es)2ds =g(1/2)■ arctan es■∞ 0=g(1/2)■π 2−π 4■ =π
4g(1/2). Generalizations appear in [3].
51 Solution to problem 12338 in Amer. Math. Monthly 129 (2022), 686 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Letf(x) :=cosx−1
x(ex−1)andI:=´∞ 0f(x)dx. Note that lim x→0f(x) = 1 /2, that fis bounded, and that the integral converges
(absolutely). Using the Laplace transform of f, we are going to show that I=1 2log■π sinhπ■ . So let
F(s) :=ˆ∞ 0e−sxf(x)dx. Also this integral converges absolutely and uniformly in s≥0, as the integrand is
dominated on [1,∞[ by 4 e−x. Moreover, Fis continuous on [0 ,∞[ with F(0) = I. Now, by a similar reason,
G(s) :=−ˆ∞ 0xe−sxf(x)dx is absolutely convergent, as the integrand is dominated on [1 ,∞[ by 4 xe−x.
Hence F′(s) =G(s). Moreover, by considering for x >0 the geometric series for (1 −e−x)−1, G(s) = −ˆ∞
0e−(s+1)x 1−e−x(cosx−1)dx=ˆ∞ 0∞X k=0e−(s+1+k)x(1−cosx)dx. As all the summands are positive,
Beppo Levi’s monotone convergence theorem for Lebesgue integrals implies that´P=P´ . Hence, by
using that for a >0 ˆ∞ 0e−axcosxdx=a a2+ 1, we obtain G(s) =∞X k=0ˆ∞ 0e−(s+1+k)x(1−cosx)dx=∞X
k=0■1 s+ 1 + k−s+ 1 + k (s+k+ 1)2+ 1■ n=k+1=∞X n=11 (s+n)3+ (s+n). The convergence being
absolut and uniform on [0 ,∞[ (a majorant is given byP∞ n=1n−3), we can integrate termwise to
re-obtain F. Note that a primitive PofGon [0,∞[ is given by P(s) =∞X n=1ˆ1 (s+n)3+ (s+n)ds=∞X
n=1logs+np (s+n)2+ 1 =−1 2∞X n=1log(s+n)2+ 1 (s+n)2=−1 2∞X n=1log■ 1 +1 (s+n)2■ . Now
F=P+cfor some constant c. Since Pis uniformly convergent, it easily follows that lims→∞P(s) = 0 (just
take a tail uniformly small, and use that the limit of the remaining finitely many summands is 0). But also
lim s→∞F(s) = 0, because |F(s)| ≤ ||f||∞´∞ 0e−st=||f||∞/s. Hence c= 0 and so F(0) =−1 2∞X n=1log■ 1
+1 n2■ . Next we use that sinh(πz) =πz∞Y n=1■ 1 +z2 n2■ .
52 So, log sinh( π) = log π+∞X n=1log■ 1 +1 n2■ . Hence F(0) = I=−1 2log sinh( π) +1 2logπ=1 2logπ
sinhπ.
53 A different solution to problem 12338 in Amer. Math. Monthly 129 (2022), 686 Raymond Mortini and
Rudolf Rupp LetI:=ˆ∞ 0cosx−1 x(ex−1)dx. We show that I=1 2log■π sinhπ■ . Recall that N={0,1,2, . . .}.
First we develop for x > 0 the integrand into a double, absolutely convergent series (so this is
independent of the arrangement): g(x) :=cosx−1 x(ex−1)=cosx−1 xe−x 1−e−x=∞X n=1(−1)n
(2n)!x2n−1∞X k=1e−kx=∞X k=1∞X n=1(−1)n (2n)!x2n−1e−kx | {z } :=akn =∞X k=1∞X neven1
(2n)!x2n−1e−kx−∞X k=1∞X nodd1 (2n)!x2n−1e−kx. Note that lim x→0g(x) = 1 /2, but that both
absolutely convergent double series at the right vanish at 0. Beppo Levi’s monotone convergence
theorem for Lebesgue integrals applied twice, gives (33)ˆX k≥2X neven=X k≥2X nevenˆ andˆX k≥2X
nodd=X k≥2X noddˆ . As the calculations below show, the sumsP nodd´ |akn|andP neven´
|akn|converge for k≥2, but diverge for k= 1, thoughP n´ a1nconverges. Moreover,P k≥2P nodd´ andP
k≥2P neven´ are finite; hence (by (33)) , (34)X k≥2X nˆ =ˆX k≥2X n. So, at the end, by adding in (34) the
termP n´ a1,n, respectively´P na1,n(which coincide, too; see addendum) we see that ˆX kX n=X nX kˆ .
To complete the calculations, we use that for m∈Nandk∈N\ {0}, ˆ∞ 0xme−kxdx=m!/km+1. Hence ∞X
k=1∞X n=1(−1)n (2n)!ˆ∞ 0x2n−1e−kxdx=∞X k=1∞X n=1(−1)n (2n)!(2n−1)! k2n= ∞X k=1 ∞X n=1(−1)n
2n1 k2n! =−1 2∞X k=1log■ 1 +1 k2■ , where the last identity comes from the fact that for 0 ≤y≤1 h(y)
:=∞X n=1(−1)n 2ny2n=−1 2log(1 + y2)
54 (note that for y= 1 there is no absolute convergence). Next we use that sinh(πz) =πz∞Y n=1■ 1 +z2
n2■ . So, log sinh( π) = log π+∞X n=1log■ 1 +1 n2■ . Hence I=−1 2log sinh( π) +1 2logπ=1 2logπ sinhπ
Addendum (35) J:=ˆ∞ 0e−xcosx−1 xdx=∞X n=1(−1)n 2n=−1 2log 2. First we note that, as above,P n´
a1,n=P∞ n=1(−1)n 2n=−1 2log 2. To show that J= −1 2log 2,we interprete this as the Laplace transform
L(q)(s) of the function q(x) = (cos x−1)/x evaluated at s= 1. By a well-known formula, if L(F(t))(s) =f(s),
then L(q)(s) =L(F(t) t)(s) =ˆ∞ sf(u)du, where f(s) =ˆ∞ 0e−st(cost−1)dt=1 s3+s. Hence L(q)(s) =−1 2log(1
+ s−2) and so J=L(q)(1)) = −1 2log 2. Another way to calculate the Laplace transform J(s) :=L(q)(s) ofqis
to take derivatives: J′(s) =−ˆ∞ 0e−st(cost−1)dt=f(s). Note thatd ds´ =´d ds, since both integrands are
locally (in s) dominated by L1[0,∞[ functions. Remark This integral Jappears also on the web, see [25].
55 Solution to problem 12312, AMM 129 (3) (2022), p. 286 Gerd Herzog, Raymond Mortini - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that the constant function 1 is
the only solution Lety=y(x) :=´x 0f(t)dtand suppose that the continuous function f: [0,∞[→Rsatisfies on
]0,∞[ f(x)■ f(x)−1 xˆx 0f(t)dt■ ≥(f(x)−1)2. Then (36) y′■ 2−y x■ ≥1 for x >0 and y(0) = 0 . Note that this
implies that y′(0) = 1, because, by letting x→0, y′(0)(2−y′(0))≥1⇐⇒ (y′(0)−1)2≤0 Let the function w:
[0,∞[→Rbe given by w(x) :=(y(x) xifx >0 y′(0) if x= 0. Then w∈C([0,∞[)∩C1( ]0,∞[). We claim that (37)
w(x) = 1 for every x≥0, from which we conclude that y(x) =xand so f(x) =y′(x) = 1 for x≥0. To see this,
note that by (36), w(x)■= 2. Since wis continuous on [0 ,∞[,w(0) = 1, and w does not take the value 2,
we have that w(x)<2 for each x >0. Hence, for x >0, w′(x) =xy′(x)−y(x) x2≥1 x■1 2−w(x)−w(x)■ =1
x·(1−w(x))2 2−w(x)(38) Thus we may deduce from (38) that w′≥0; that is wis increasing6. Now suppose
that (37) is not true. Case 1 There is x0>0 with w(x0)<1. This is not possible, though, as wis increasing,
butw(0) = 1. Case 2 There is x0>0 with w(x0)>1. As wis increasing, w > 1 for x≥x0. Note that we already
know that w <2. Since the map t7→(1−t)2 2−tis increasing on [1 ,2[, we deduce from (38) that for x≥x0
w′(x)≥1 x·(1−w(x0))2 2−w(x0)=:c1 x. Hence, by integration, for x≥x0, w(x)≥w(x0) +clog(x/x0)→ ∞ (x→ ∞
). An obvious contradiction. ■ 6in the weak sense; or funnily called nondecreasing, a very ambiguous
word.
56 Solution to problem 12308, AMM 129 (3) (2022), p. 285 , by Raymond Mortini - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that the minimal value is given by 105 /2
and is obtained by the polynomial f(x) =−105/16x4+ 105 /8x2−33/16 Letpbe any polynomial. Then, by
Cauchy-Schwarz, ■ˆ1 0f′p dx■2 ≤■ˆ1 0f′2dx■ ■ˆ1 0p2dx■ . A primitives of f′pis given by fp−´ fp′dx.
Now choose pso that p(0) = p(1) = 0 and p′(x) =αx2+β. To this end, put p(x) =ax(x2−1). Then I:=ˆ1 0f′p
dx =fp 1 0−ˆ1 0f(3ax2−a)dx=−3a+a=−2a Moreover,ˆ1 0p2dx=a2ˆ1 0(x6+x2−2x4)dx=a2■1 7+1 3−2 5■ .
Hence ˆ f′2dx≥4a2 a2■1 7+1 3−2 5■=105 2. Equality in the Cauchy-Schwarz inequality is given
whenever f′=p. Thus f(x) =a 4x4−a 2x2+c. Now aandchave to be chosen so that´ f=´ x2f= 1. This yields
the linear system −7a+ 60c= 60 −27a+ 140 c= 420 whose solution is a=−105/4 and c=−33/16.
Consequently f(x) =−105/16x4+ 105 /8x2−33/16. Note that f′(x)2=■ −105 4x(x2−1)■2.
57 Solution to problem 12326, AMM 129 (5) (2022), p. 487 Raymond Mortini, Peter Pflug, Amol Sasane
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - By considering the symmetric
function p(x, y) := f(x+y)−f(x)−f(y) we get from the assumption that as well p(·, y) and p(x,·) are
polynomials in their variables separately. Hence, by [26], p(x, y) is a polynomial. Case 1 f∈C1(R). Write
p(x, y) =Pai,jxiyjwith symmetrical coefficients and a0,0= −f(0) (the sum being finite of course) If we take
y= 0, then for all x −f(0) = f(x+ 0)−f(x)−f(0) = a0,0+X ai,0xi. Hence ai,0= 0 for all i≥1. Due to symmetry,
we also have a0,j= 0 for all j≥1. Thus we have only coefficients ai,jfori, j≥1. Consequenlty
f(x+y)−f(x)−(f(y)−f(0))) y=X i,j≥1ai,jxiyj−1. Asfis assumed to be differentiable, we may take y→0 and get
f′(x)−f′(0) =X i≥1ai,1xi. Integration yields f(x)−f(0)−xf′(0) =X i≥1ai,1xi+1 i+ 1. Thus fis a polynomial. Case
2 f∈C(R). Let F(x) :=´x 0f(t)dtbe a primitive of f. Then with G(x, y) :=F(x+y)−F(x)−F(y) G(x, y) =ˆx+y
0f(t)dt−ˆx 0f(t)dt−ˆy 0f(t)dt = t=y+sˆx −yf(y+s)ds−ˆx 0f(t)dt−ˆy 0f(t)dt =ˆ0 −yf(y+s)ds+ˆx 0■ f(y+s)−f(s)■
ds−ˆy 0f(t)dt = t=y+sˆy 0f(t)dt+ˆx 0■ f(y+s)−f(s)■ ds−ˆy 0f(t)dt =ˆx 0p(y, s)ds+f(y)x which is a polynomial
in x. Again, by symmetry, and the Carroll argument, Gis a polynomial. Hence, by Case 1, Fis a
polynomial and so does f=F′. A detailed analysis of the functional equation f(x+y)−f(x)−f(y) =p(x, y) (and
based on these methods) appears in [4].
58 Solution to problem 12290 in Amer. Math. Monthly 128 (2021), 946 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that all
solutions are given by az,bsin(kz) and csinh(kz) where a, b, c ∈Cand k∈R. First we note that any
solution fnecessarily satisfies f(0) = 0. Now let h(z) :=|f(z)|2= (ff)(z). Since fx=f′andfy=ifx=if′, we see that
fxy= (f′)y=i(f′)x=if′′. Moreover (f)x=fx. Hence hxy= (fxf+ffx)y=fxyf+fxfy+fyfx+ffxy = 2Re( fxyf) + 0 = 2Re(
if′′f) =−2Im(f′′f). Now|f(z)|2=|f(x)|2+|f(iy)|2implies that the mixed derivative of the right hand side is 0. We
conclude that Im( f′′f) = 0 in C. Let U=C\Z(f), where Z(f) ={z∈C:f(z) = 0}. Then on U, this is equivalent to
0 = Im■f′′ f|f|2■ = Im■f′′ f■ . Thus, a necessary condition for f■≡0 being a solution is that f′′/fis a real
constant λ. The differential equation f′′=λfinChas the solutions az+difλ= 0, or αe√ λ z+βe−√ λ zif λ >0,
and αei√ |λ|z+βe−i√ |λ|zifλ <0. Since f(0) = 0, we have d= 0 and β=−α. So, with k:=p |λ|, f(z)
=az,csinhkzifλ >0 and csinkzifλ <0. It is now easy to check that these are solutions indeed (wlog for k=
1): sin(x+iy) = cos( iy) sinx+ cos xsin(iy) =e−y+ey 2sinx−icosxe−y−ey 2 = cosh ysinx+icosxsinhy |sinz|2=
sin2xcosh2y+ cos2xsinh2y = sin2xcosh2y+ (1−sin2x) sinh2y = sin2x(cosh2y−sinh2y) + sinh2y = sin2x+
sinh2y = sin2x+|sin2(iy)|. as sin( iy) =isinhy
59 Solution to problem 12288 in Amer. Math. Monthly 128 (2021), 946 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - A change of the
variable x→1/xyields that J:=ˆ∞ 0■ 1−x2sin2■1 x■■2 dx=ˆ∞ 0(x2−sin2x)2 x6dx. Note that
(x2−sin2x)2=x4−2x2sin2x+ sin4x. Now we ”linearize” the trigonometric powers: sin2x= (1/2)(1−cos 2x)
and sin4x= (3/8)−(1/2) cos 2 x+ (1/8) cos 4 x. Thus J=I/2, where I:=ˆ R3 8+x4−x2+ (x2−1 2) cos(2 x) +1
8cos(4 x) x6dx. Next we consider the meromorphic function f(z) :=3 8+z4−z2+ (z2−1 2)e2iz+1 8e4iz z6.
Then we add in the numerator the polynomial p(z) :=i■1 2z−4 3z3+2 5z5■ , that is we consider the
function F(z) :=f(z) +p(z) z6. Note that this polynomial is chosen so that Fhas a removable singularity at
z= 0 (in other words, −p(z) z6is the principal part in the Laurent expansion of faround the origin).
Hence´ ΓF(z)dz= 0, where Γ is the boundary of the half-disk |z| ≤R, Imz≥0, consisting of the half circle Γ
Rand the interval [ −R, R]. Hence, by letting R→ ∞ and taking real parts, 0 = Re lim R→∞ˆ ΓRF(z)dz+I.
By Jordan’s Lemma, lim supR→∞´ ΓR|einz||dz|<∞. Hence, lim R→∞ˆ ΓRF(z)dz= 0 + 0 + ilim R→∞ˆ
ΓR2 5z5 z6dz=−2π 5. We conclude that the value of the original integral Jisπ/5.
60 Solution to problem 12256 in Amer. Math. Monthly 128 (2021), 478 Raymond Mortini and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Using that 4 ab=
(a+b)2−(a−b)2, we obtain 4ˆ1 0log(1 + x) log(1 −x) xdx=ˆ1 0log2(1−x2) xdx−ˆ1 0log21+x 1−x xdx=:I1−I2.
ForI1, we make the substitution 1 −x2=t2. Hence, due to −xdx=tdt, I1=ˆ1 0log2t2 1−t2t dt Using
that´P=P´ (Lebesgue), and twice integration by parts, I1= 4∞X n=0ˆ1 0t2n+1log2t dt= 8∞X n=01 (2n+
2)3=ξ(3). For the second one, I2, we make the substitution t=1+x 1−x. Then x=t−1 t+1anddx=2 (t+1)2dt.
Hence I2= 2ˆ∞ 1log2t 1−t2dtt=1/s= 2ˆ1 0log2s 1−s2ds= 2∞X n=0ˆ1 0s2nlog2s ds= 4∞X n=01 (2n+ 1)3=
47 8ξ(3). Consequently, 4 I= (1−7 2)ξ(3) =−5 2ξ(3) and so ˆ1 0log(1 + x) log(1 −x) xdx=−5 8ξ(3).
61 original statement Given a, b, α, β ∈Cwith|a|<1,|b|<1 and |α|=|β|= 1, let φa(z) = (a−z)/(1−az) and ρ(a,
b) =|a−b|/|1−ab|the pseudohyperbolic distance between aandb. i) Show that whenever a, b∈]−1,1[,
M−:= max |z|≤1|φa(z)−φb(z)|= 2ρ(a, b) and M+:= max |z|≤1|φa(z) +φb(z)|= 2. ii) Determine M:= max
|z|=1|αφa(z)−βφb(z)| and m:= min |z|=1|αφa(z)−βφb(z)|. Solution to problem 11684 AMM 120 (2013), 76
Raymond Mortini, Rudolf Rupp i) That M+= 2 is easy: just take z= 1 and evaluate: |φa(1) + φb(1)|=|
−1−1|= 2. Since M+≤2, we are done. ii) We first observe that ϕbis its own inverse. Let c= (b−a)/(1−ab)
and λ=−(1−ab)/(1− ab). Since ϕbis a bijection of the unit circle onto itself, max |z|=1|αφa(z)−βφb(z)|=
max |z|=1|αβφa(φb(z))−z|= max |z|=1|αβλφ c(z)−z|. The same identities hold when replacing the
maximum with the minimum. Putγ:=αβλand let −π <argγ≤π. For|z|= 1 we obtain H(z) :=|γϕc(z)−z|=
γz(cz−1) 1−cz−z = γ1−cz 1−cz+ 1 = γw w+ 1 , where w= 1−cz= 1−c1 z.
62 Ifzmoves on the unit circle, then wmoves on the circle |w−1|=|c|. Let w=|w|eiθ. Then (see figure 5)
the domain of variation of θis the interval [ −θm, θm] with |θm|< π/ 2 and sinθm=|c|=ρ(a, b). Now H(z)
=|γe2iθ+ 1|= 2|cos(argγ 2+θ)|. Hence, M= max |z|=1H(z) = 2 max {|cos(argγ 2+θ)|:|θ| ≤arcsin( ρ(a, b))}
and m= min |z|=1H(z) = 2 min {|cos(argγ 2+θ)|:|θ| ≤arcsin( ρ(a, b))}. In particular, if a, b∈]−1,1[ and α=β=
1, then γ=−1, and so (using the maximum principle at ∗) M−∗= max |z|=1H(z) = 2 max {|sinθ|:|θ| ≤arcsin(
ρ(a, b))}= 2ρ(a, b). Ifa, b∈]−1,1[ and α= 1, β=−1, then γ= 1, and so M+∗= max |z|=1H(z) = 2 max
{|cosθ|:|θ| ≤arcsin( ρ(a, b))}= 2. We note that m= 0, that is H(z0) = 0 for some z0with|z0|= 1, if and only if
γϕchas a fixed point on the unit circle (namely z0). This is equivalent to the condition |cos(argγ 2)| ≤ |c|.
Moreover, M= 2 if and only if |sin(argγ 2)| ≤ |c|. ! Figure 5. The domain of variation of arg w
63 ©Mathematical Association of America, 2025.
64 Solution to problem 11584 AMM 118 (2011), 558 Raymond Mortini, J´ erˆ ome No¨ el - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - By the Schwarz-Pick inequality,(1−|z|2)|B′(z)|
1−|B(z)|2≤1 for any holomorphic self-map of the unit disk. Then, if we let Bbe the Blaschke product B(z)
=∞Y n=1|an| anan−z 1−anz associated with the zeros ( an), we get: |B′(0)| 1− |B(0)|2≤1. But B′(z)
B(z)=−∞X n=11− |aj|2 (1−ajz)(aj−z). Hence ∞X j=11− |aj|2 aj =|B′(0)| |B(0)|≤1− |B(0)|2 |B(0)|=1−Q∞
j=1|aj|2 Q∞ j=1|aj|. Motivation for posing this as a problem to AMM: We are interested in a direct
elementary proof.
65 Solution to problem 11578 in Amer. Math. Monthly 118 (2011), 464 Raymond Mortini - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Lemma 2. Let0<||x||<1ands∈S. Let s′be the
(second) uniquely determined intersection point of the half-line starting at sand passing through xwith
S. Then the map Q:S→ [0,∞[, s7→ ||x−s||/||x−s′||is a nonconstant continuous map. Proof. Qobviously is
continuous. If we suppose that Qis constant κ, then this constant is necessarily 1 (just interchange
swith s′). Now x= (1−t)s+ts′. Thus x−s=t(s′−s) and x−s′= (1−t)(s−s′) and so Q(s) =t/(1−t). Hence 1 = κ=t
1−t. Sot= 1/2. Now x/||x|| and−x/||x||belong to Sand with t= (1− ||x||)/2 we have x= (1−t)x ||x||+t−x ||x||.
Sot= 1/2 implies that x= 0. ■ Lemma 3. The unit sphere Sis connected whenever dimE≥2. Proof. Letx,
y∈S,x■=y. Ifxis linear independent of y, then the segment {tx+ (1−t)y: 0≤t≤1}does not pass through the
origin; hence t7→tx+ (1−t)y ||tx+ (1−t)y|| is a path joining ywith xonS. Ify=λxfor some λ∈R, then we use
the hypothesis that dim E≥2 to guarantee the existence of a vector ulinear independent of x. Thus
v:=u/||u|| ∈S. By the first case, we may join xwith vand then vwith yby a path in S. ■ The first step is to
show that f(0) = 0. (1) Let x= 0,y=−f(0). Then f(f(−f(0))) = f(0)−f(0) = 0; (2) Let x=y= 0. Then f(f(0)) = f(0);
(3) Let x=−f(y). Then f(0) = f(−f(y)) +y. With y= 0 this gives f(0) = f(−f(0)). (4) Applying fyields f(f(0)) =
f(f(−f(0))) = (1)0. Thus, by (2), f(0) = 0. (5) Let x= 0. Then f(f(y)) =f(0) + y=y. Hence fis an involution.
(6)fis additive since f(x+y) = (5)f(x+f(f(y))) = f(x) +f(y). (7) Next we show that fisQ-homogeneous by
induction. Indeed, by (5), f((n+ 1)x) =f(nx+x) =f(nx|{z} X+f(f(x)|{z} Y)) =f(nx) +f(x). Thus f(mx) =mf(x) for
every m∈N. Now 0 =f(0) = f(−x+x) = (5)f(−x+f(f(x))) = f(−x) +f(x). Thus f(−x) =−f(x). Hence, for p∈Z, we
have f(px) =pf(x). Next, if n∈N, then nf(x n) =f(x n) + (n−1)f(x n) =f(x n) +f(n−1 nx) =f■x n|{z} X+f■
f■n−1 nx■ |{z} Y■■ = (5)f■x n+n−1 nx■ =f(x)
66 Hence f■x n■ =1 nf(x). Therefore f■p n■ =p nf(x) for p∈Zandn∈N. (8) By hypothesis, ||f(s)|| ≤Cfor
every s∈S. Let 0 <||x||<1. Consider, as in Lemma 2, the map H:S→[0,∞[, s7→ ||x−s||/||x−s′||.His
continuous and non-constant. Since dimE≥2,Sis connected by Lemma 3. Hence H(S) is an interval. In
particular, there is s∈Ssuch that r:=||x−s||/||x−s′||is rational. Thus, with t=r/(1 +r), x= (1−t)s+ts′ is a
rational convex-combination of two elements in the sphere. Since fisQ-linear, we conclude that ||f(x)||
≤(1−t)||f(s)||+t||f(s′)|| ≤(1−t)C+tC=C. Now let x∈Ebe arbitrary. Choose a null-sequence ■nof positive
numbers so that qn:= ||x||+■nis rational. Then, ||x/qn|| ≤1. Since fisQ-linear, we obtain
||f(x)||=qn||f(x/qn)|| ≤qnC. Letting ntend to infinity, we get ||f(x)|| ≤C||x||. Thus fis continuous at the origin.
Since fis additive, we deduce that fis continuous everywhere; just use f(x0+x) =f(x0) +f(x)→f(x0) ifx→0.
(9) It easily follows now that fis homogeneous: if α∈R, choose a sequence ( rn) of rational numbers
converging to α. Then, due to continuity, f(αx) = lim nrnf(x) =αf(x). To sum up, we have shown that fis a
continuous linear involution. Remarks Ifn= 1, then the unit sphere Sis just a two point set, and so every
function is automatically bounded on S. There exist, though, non-continuous linear involutions in R. To
this end, let B be a Hamel basis of the Q-vector space R, endowed with the usual Euclidean norm. We
may assume that Bis dense in R. Fix two elements b0andb1∈ B. Let fbe defined by f(b0) =b1, f(b1)
=b0andf(b) =bifb∈ B\{ b0, b1}. Linearly extend f(in a unique way). Then, obviously, fis a linear
involution. But fis not continuous at b0. In fact, let ( bk)n≥2∈ BNconverge to b0. Then f(bk)
=bk→b0=f(b1)■=f(b0).
67 Solution to problem 11548 in Amer. Math. Monthly 118 (2011), 85 Raymond Mortini and J´ erˆ ome
No¨ el - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Letf∈C2([−1,1]),f(0) =
0. Then ■ˆ1 −1f(x)dx■2 ≤1 10ˆ1 −1(f′′(x))2dx. Moreover, the constant 1 /10 is best possible. Solution
We consider the auxiliary integral I=1 2■ˆ1 0(t−1)2f′′(t)dt+ˆ0 −1(1 +t)2f′′(t)■ . We first show that I=´1
−1f(t)dt. In fact, twice integration by parts yields: ˆ1 0(t−1)2f′′(t)dt=−f′(0)−2ˆ1 0(t−1)f′(t)dt=−f′(0) + 2ˆ1
0f(t)dt, as well asˆ0 −1(t+ 1)2f′′(t)dt=f′(0)−2ˆ0 −1(t+ 1)f′(t)dt=f′(0) + 2ˆ0 −1f(t)dt. This proves the first
claim. Now we use the Cauchy-Schwarz inequality to estimate I: ■ˆ1 0(t−1)2f′′(t)dt■2 ≤ˆ1 0(t−1)4dtˆ1
0(f′′(t))2dt=1 5ˆ1 0(f′′(t))2dt, and similarily for the second integral. Hence, by using that (
A+B)2≤2(A2+B2), we obtain I2≤21 4■1 5ˆ1 0(f′′(t))2dt+1 5ˆ0 −1(f′′(t))2dt■ =1 10ˆ1 −1(f′′(t))2dt. The
constant 1 /10 is obtained for the function f(t) =( 1 12t4+1 3t3+1 2t2if−1≤t≤0 1 12t4−1 3t3+1 2t2if 0≤t≤1.
Indeed, this follows from the fact that in the Cauchy-Schwarz inequality we actually have equality if the
functions are colinear: p′′(t) = (1 + t)2if−1≤t≤0 and p′′(t) = (1 −t)2if 0≤t≤1. A computation then shows that
■ˆ1 −1p(x)dx■2 =1 10ˆ1 −1(p′′(x))2dx=1 25. Remark Iff∈C2([−1,1]) satisfies f(1) = f(−1) = f′(1) = f′(−1) =
0, then the inequality above holds, too. In fact. ˆ1 −1f(x)dx=ˆ1 −11·f(x)dx=xf(x)|1 −1−ˆ1 −1xf′(x)dx= =−1
2x2f′(x)|1 −1+1 2ˆ1 −1x2f′′(x)dx=1 2ˆ1 −1x2f′′(x)dx
68 Thus, by Cauchy-Schwarz, ■ˆ1 −1f(x)dx■2 ≤1 4ˆ1 −1x4dxˆ1 −1(f′′(x))2dx=1 4·2 5ˆ1 −1(f′′(x))2dx.
69 Solution to problem 11456 AMM 116 (2009), 747 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - am:= 1−1 m+5 41 m2=1 +■2m−1 2■2 m2 nY m=1am=Qn
m=1■ 1 +■2m−1 2■2■ Qn m=1m2Q2n m=1m2 Qn m=1(2m−1)2Qn m=1(2m)2 =Qn m=1■ 1
(2m−1)2+1 4■ (2n)!2 4n(n!)4=Qn m=1■ 4 (2m−1)2+ 1■ (2n)!2 16n(n!)4. Now, by Stirlings formula,
(2n)! 4nn!2∼(2n)2ne−2n√ 4πn (nne−n√ 2πn)222n=1√πn. Since cos( πz) =Q∞ n=1■ 1−4z2 (2n−1)2■ , we
have lim nnnY m=1am=cos(πi) π=coshπ π. We note thatvuutnY m=1am=1 n!nY m=1 i−2m−1 2 = (n+
1)2√ 5|f(n+1)(0)| (n+ 1)!, where f(z) = (1 −z)i+1 2, an interesting function in the Wiener algebra (its
Taylor coefficients behave like n−3/2by the above calculations).
70 Solution to problem 11402, AMM 115 (10), (2008), p. 949 Raymond Mortini - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The problem obviously is equivalent to show the
existence of two points 0 < a < b < 1 with f(a) =f(b) =b−a, or in other words, find 0 < a < b < 1 with b−f(b)
=aandf(b) =f(a). To this end, consider the function h(x) :=f(x−f(x))−f(x), where we have continuously
extended fby the value 0 for x <1 and x >1. Then his continuous. We have to show that h admits a zero
bin ]0,1[ with f(b)< b. Then a:=b−f(b)∈]0,1[ and b−a=f(b) =f(a). To do this, we prove that htakes positive
and negative values on [0 ,1]. Since h(0) = h(1) = 0, the continuity of himplies that hhas a zero bin ]0,1[.
Our construction will guarantee that f(b)< b. Letξ0be the largest fixed point of f(note hat 0 ≤ξ0<1). For
later purposes, we note thatf(x)≤xwhenever ξ0≤x≤1. If ξ0= 0, we let x0be the be the smallest point for
which f(x0) =M:= max x∈[0,1]f(x). Note that x0∈]0,1[. Finally, let x1∈[ξ0,1[ be the largest point with f(x1)
=M1:= max x∈[ξ0,1]f(x). Then 0 < x0≤x1<1. Since the function x−f(x) is 0 at ξ0and 1 at 1, the
intermediate value theorem for continuous functions implies that there exists y1∈]ξ0,1[ such that
y1−f(y1) =x1. Since f >0,y1> x1. Thus h(y1) =f(y1−f(y1))−f(y1) =f(x1)−f(y1) =M1−f(y1)>0. On the other
hand, h(ξ0) =f(ξ0−f(ξ0))−f(ξ0) = 0−f(ξ0)<0 ifξ0>0, and if ξ0= 0, then, h(x0) =f(x0−f(x0))−f(x0)<0 (since
x0−f(x0) is left from the smallest maximal point x0off.) In both cases, there exists bsuch that h(b) = 0.
Since ξ0< b < y 1ifξ0>0 and 0 < x0< b < y 1ifξ0= 0, we see that f(b)< b.
71 Solution to problem 11333, AMM 114 (10), (2007), p. 926 Raymond Mortini - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let PN=NY n=2 ■n2−1 n2■2(n2−1)■n+ 1 n−1■n! . a)
We have the following equalities: NY n=2■n2−1 n2■n2−1 =(N+ 1)N2−1 NN(N+2)(N!)2, b) NY n=2■n+
1 n−1■n =(N+ 1)NNN+1 2(N!)2. Hence p PN=(N+ 1)N2−1 NN(N+2)(N!)2(N+ 1)N/2N(N+1)/2 √ 2N!=
■N+ 1 N■N2−1NN2−1 NN(N+2)N!(N+ 1)N/2N(N+1)/2 √ 2= ■N+ 1 N■N2−1 N!(N+ 1)N/2 NN/2√
2N(N+1)/2NN/2 N2N+1= ■N+ 1 N■N2−1 N!■ 1 +1 N■N/2 √ 2√ N NN+1. We are now using Stirling’s
formula telling us that n!∼e−nnn√ 2πn. Hence p PN∼√e√ 2NNe−N√ 2πN■N+ 1 N■N2−1√ N NN+1=
√e√πe−N■N+ 1 N■N2−1 . ButaN:=e−N■N+ 1 N■N2−1 →1√easN→ ∞ ; in fact, by taking logarithms we
obtain logan= (N2−1) log(1 +1 N)−N∼N2log(1 +1 N)−N=N2(1 N−1 2N± ··· )−N∼ −1 2.
Hence√PN→√πand so PN→π.
72 Solution to problem 11226, AMM 113 (5), (2006), p. 460 Raymond Mortini - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Leth(x) =Qn j=1(1−xaj). Then h′(x)/h(x) =−Pn j=1ajxaj−1
1−xajand hence h′(x) =−nX j=1ajxaj−1Y k■=j(1−xak). Clearly h′(0) = h′(1) = 0. Let f(x) = (1 −x)−1nY
j=1(1−xaj) if 0≤x < 1. Note that f(0) = 1 and lim x→1f(x) =−h′(1) = 0. Thus, if we show that f′(0)>0 and
that the derivative of fhas a unique zero in the open interval ]0 ,1[, we are done (that is we can then
conclude by the intermediate value theorem that there is a unique x0with 0< x0<1 so that f(x0) = 1, and
hence h(x0) = 1−x0.) Now, f′(x)/f(x) =1 1−x+h′(x)/h(x). In particular, f′(0) = 1. Thus we have to look for
x∈]0,1[ so that g(x) :=Pn j=1ajxaj−11−x 1−xaj= 1. But g(0) = 0, and, by de l’Hopital’s rule, limx→1g(x)
=n. The intermediate value theorem yields the existence of x. The uniqueness of such an xfollows from
the fact that gis strictly increasing. This is due to the fact that the functionxa−1−xa 1−xais strictly
increasing on ]0 ,1[ whenever a >1. The latter follows from the fact that d dxxa−1−xa 1−xa=xa−2■ (a−1)
+xa−ax■ (1−xa)2 and that k(x) :=a−1 +xa−ax≥0 for 0 ≤x≤1, because k(0) = a−1>0,k(1) = 0 and k′(x)
=a(xa−1−1)≤0.
73 Solution to problem 11210, AMM 113 (3), (2006), p. 267 Raymond Mortini - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We note that pn:=(2n+ 1)4−(2/π)4 (2n+ 1)4= 1−■2
π(2n+ 1)■4! = ■ 1−4 π2(2n+ 1)2■ ■ 1 +4 π2(2n+ 1)2■ . Multiplying in the numerator and denominator
(which is 1) with the ”missing” factors ■ 1−4 π2(2n)2■ ■ 1 +4 π2(2n)2■ we obtain P:=∞Y n=0pn=∞Y
k=1■ 1−4 π2k2■■ 1 +4 π2k2■ ■ 1−1 π2k2■■ 1 +1 π2k2■. Using the standard infinite product
representation of the sinus sinz z=∞Y k=1■ 1−z2 π2k2■ , we obtain P=sin 2 2sin(2i) 2i sin 1 1sini i=
cos 1 cosh 1 = (cos 1)e2+ 1 2.
74 Solution to problem 11202, AMM 113 (2), (2006), p. 179 Raymond Mortini - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The assertion is an immediate consequence of H¨
older’s inequality: Wlog let 0 ≤aj≤1 and letq∈]0,1[ be such that p+q= 1 (note that p∈]0,1[.) np−1nX
j=1ap j=np−1 NX j=1ap j+nX j=N+1ap j·1 ≤ N n1−p+ nX j=N+1(ap j)1/p p nX j=N+111/q q 1
n1−p≤ N n1−p+ ∞X j=N+1aj p nq n1−p=N n1−p+ ∞X j=N+1aj p ≤■ ifNandn > N is sufficiently big.
75 Solution to problem 11185 AMM 112 (2005), 840 Rainer Br¨ uck, Raymond Mortini - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that I(α, p) converges if and only if (
α, p)∈]1,∞[×Nor (α, p)∈■1 2,1■ ×(2N+ 1). First we discuss the behaviour of the integrand at the origin.
For α >0 we have log■ 1 +sinpx xα■ ≤ log (1 + x−α). Substituting1 xbyt, we obtain ˆ1 0log■ 1 +x−α■
dx=ˆ∞ 1log (1 + tα) t2dt , and this integral is convergent. Hence, our integral I(α, p) converges at 0 for
every α >0 and p∈N. Now we discuss the behaviour at infinity. Since lim t→0log (1+ t) t= 1, we see that
at infinity A(x) := log■ 1 +sinpx xα■ ∼sinpx xα=:B(x). Hence´∞ 1A(x)dxconverges absolutely if and only
if´∞ 1B(x)dxdoes. Note that by Riemann’s convergence test´∞ 1|B(x)|dx≤´∞ 1dx xα<∞whenever α >1.
Hence,´∞ 1A(x)dxis absolutely convergent for α >1. Now suppose that 0 < α≤1. On the intervals Jk:=■π
6+ 2kπ,π 2+ 2kπ■ ,k≥1, we have |sinx| ≥1 2andx≥1. Hencesinpx xα≥2−p x≥2−p 2π(k+1). Therefore, ˆ
Jk|B(x)|dx≥1 3·2−p−1 k+ 1. Since´∞ 1|B(x)|dx≥P∞ k=1´ Jk|B(x)|dx, we see that´∞ 1|B(x)|dxand hence´∞
1|A(x)|dx diverges (absolutely) for 0 < α≤1. In particular,´∞ 1A(x)dxdiverges whenever pis even, since in
that case |A(x)|=A(x). To continue, we may thus assume that p= 2n+ 1 is odd. We use that for every α
>0 and n∈Nthe integral´∞ 1sin2n+1x xαdxconverges. Indeed, let Im(x) :=´x 1sinmt tαdtand let Fmbe a
primitive of sinmtwith Fm(1) = 0. For modd, Fmis periodic, hence bounded. By partial integration we
obtain I2n+1(x) =F2n+1(x) xα+αˆx 1F2n+1(t) tα+1dt , and we conclude that I2n+1(x) converges as x→
∞ . Now we use the Taylor development log (1 + u) =m−1X k=1(−1)k−1 kuk+(−1)m−1 mum(1 +ε(u)),
where εis a continuous function of uandε(0) = 0. In particular, |ε(u)|<1 whenever |u| ≤δ with δ >0
sufficiently small. Now, we set u=u(x) =sin2n+1x xα, where x >0 is so large that
76 |u| ≤δ. Then for sufficiently large real numbers M > N , we have I:=ˆM Nlog■ 1 +sin2n+1x xα■
dx=m−1X k=1(−1)k−1 kˆM N■sin2n+1x xα■k dx +(−1)m−1 mˆM N■sin2n+1x xα■m (1
+ε(u(x)))dx=:m−1X k=1Ik+eIm. Choosing m∈Nsuch that mα > 1 and ( m−1)α≤1, the boundedness of
ε(u) yields the absolute convergence of the last integral eIm. If1 2< α≤1, then m= 2 and hence I=I1+eI2.
ButI1andeI2converge, and hence Iconverges. If 0 < α≤1 2, then m≥3 and at least a third integral
I2above appears. That integral is divergent, since the exponent of the sin is an even one (note that by
the choice of m, the exponent of xis still at most 1). Since all those divergent integrals I2qcome up with
the same sign, we finally get the divergence of I1+I2+···+Im−1, and thus Idiverges. Finally, we note that
the example p= 1 and α=1 2yields examples of functions fandgsuch that at infinity, f∼g, but for which´∞
0f(x)dxdiverges and´∞ 0g(x)dxconverges, namely f(x) = log■ 1 +sinx√x■ andg(x) =sinx√x.
77 Solution to problem 11147 AMM 112 (2005), 366 Pamela Gorkin, Raymond Mortini - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetS(z) = exp■ −1+z 1−z■ be the atomic inner
function. Put f=1/e−S 1−(1/e)S. Then fis an inner function (that is it has radial limts of modulus one
almost everywhere). Since fdoes not have radial limit zero, it must be a pure Blaschke product (see
Garnett, p.76), that is f(z) =eiθzY n∈Z\{0}|an| anan−z 1−anz. Its zeros are exactly the numbers
anforn∈Z\ {0}, including the the origin. Since the derivative of SisS′(z) =−S(z)2 (1−z)2, it follows that the
derivative of fdoes not vanish either. But S′(z) S(z)=1 z+X n∈Z\{0}■1 z−an−1 z−a∗n■ .
78 Solution to problem 11136 AMM 112 (2005), 181 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetDn=D(λn, rn) be a sequence of pairwisw disjoint, closed
disks contained in the open unit disk Dsuch that the area measure of D\SDnis zero. Noticing that by the
mean-value area theorem for harmonic functionsˆ ˆ D(λ,r)u(z)dA(z) =πr2u(λ), we obtain the assertion 0
=u(0) =ˆ ˆ Du(z)dA(z) =X nˆ ˆ Dnu(z)dA(z) =πX nr2 nu(λ). Remark The problem was motivated by the
question, circulating in England, and communi- cated to me by Joel F. Feinstein, whether the set of
exponentials {eiλz:λ∈C}is countably linear independent! The method for the proof above presumably
appeared for the first time in a paper of J. Wolff [Comptes Rendus Acad. Sci. Paris 173 (1921),
1056-1058].
79 Solution to problem 11070, AMM 111 (2004), p. 258 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetN={1,2,···} andf, g∈Cn(Ω). Then the result follows from
the following formula: (f■g)(n)(z) =nX j=1f(j)(g(z))■X k∈Nj |k|=nCn kg(k)(z)■ , (Mo n) where k= (k1, k2,
. . . , k j)∈Njis an ordered multi-index with k1≤k2≤ ··· ≤ kj, |k|=Pj i=1ki, g(k)=g(k1)g(k2). . . g(kj)andCn
k=1Q i[Ak(i)!]■n k■ . Here Ak(i) denotes the cardinal of how often iappears within the ordered index
kand■n k■ =n! k1!k2!...kj!. This formula has many advantages vis-■ a-vis the Faa di Bruno formula
(f■g)(n)=X■n p■ (f(p)■g)nY j=1■g(j) j!■pj , where pj∈
{0,1,2,···},p=p1+p2+···+pnandp1+2p2+···+npn=n, since one immediately can write down all the factors
that occur without solving the above equations for pj. Case 1 : Let f(z0) =g(z0) =z0,A:=f′(z0) =g′(z0)■=
0,Ap■= 1∀p∈Nandf■g=g■f. In order to show that f≡git is enough to prove that f(n)(z0) =g(n)(z0) for all
n. The proof is done inductively: n= 2: Since ( f■g)′′= (f′′■g)g′2+(f′■g)g′′andf■g=g■fwe get:
f′′(z0)A2+Ag′′(z0) = g′′(z0)A2+Af′′(z0). Hence f′′(z0)(A−1) = g′′(z0)(A−1). Since A■= 1 we obtain that
f′′(z0) =g′′(z0). n→n+ 1: (f■g)(n+1)= (f′■g)g(n+1)+nX j=2(f(j)■g)X k∈Nj |k|=n+1Cn+1 kg(k)+
(f(n+1)■g)(g′)n+1 Evaluating at z0and noticing that, by induction hypotheses, all derivatives appearing
in the middle term coincide at z0with those when fis replaced by g, we get that Ag(n+1)(z0)
+f(n+1)(z0)An+1=Af(n+1)(z0) +g(n+1)(z0)An+1. Hence f(n+1)(z0)(An−1) = g(n+1)(z0)(An−1), from
which we conclude that f(n+1)(z0) = f(n+1)(z0), because An■= 1. Case 2 :f(z0) =g(z0) =z0,f(j)(z0)
=g(j)(z0) = 0 for 1 ≤j < n 0, but f(n0)(z0) = g(n0)(z0)■= 0 and f■g=g■f. Suppose that f(j)(z0) =g(j)(z0)
has been shown to be true for j < n , where n=pn0+q, with 0 ≤q < n 0andp≥1. We show that this holds
then for j=n. LetN=n2 0+ (p−1)n0+qand consider ( f■g)(N)(x0). All the terms in (Mo) Nwith j < n 0
disappear, since f(j)(g(z0)) =f(j)(z0) = 0. Moreover, as we are going to show, all other terms, excepted
the term for j=n0and the index k= (n0,···, n0, pn0+q)∈Nn0, coincide for f andg; hence can be thrown off
when regarding the equality ( f■g)(N)= (g■f)(N). Thus that equality is equivalent to
f(n0)(g(z0))(g(n0))n0−1(z0)g(pn0+q)(z0) =g(n0)(f(z0))(f(n0))n0−1(z0)f(pn0+q)(z0)
80 But this implies of course that f(pn0+q)(z0) =g(pn0+q)(z0), which is what we were after. That one
can restrict to this single index k= (n0,···, n0, pn0+q)∈Nn0is seen as follows: Letk′∈Nn0, be an ordered
index with |k′|=|k|= (n0−1)n0+pn0+q=N. Suppose that the last coordinate of k′(which is the maximum) is
strictly bigger than the last coordinate of k. Then at least one of the previous coordinates of k′must be
strictly smaller than n0. But the associated derivatives of g(resp f) vanish at z0. Thus this term does not
appear in the formula for ( f■g)(N)(z0). On the other hand, if the last coordinate of k′is strictly less than
pn0+q(hence all of the coordinates of k′), then by induction all the associated derivatives of g(in
(f■g)(N)) coincide with those for f(in (g■f)(N)) atz0. Thus these terms can be thrown away. Now let
k′∈Njwith n0< j≤Nand|k′|=N. Then the maximum of the coordinates of k′ is strictly less than pn0+q,
since otherwise |k′| ≥(j−1)n0+pn0+q≥n2 0+pn0+q > N , a contradiction. Thus, as above, also these
terms can be thrown away.
81 No own Solution to problem 10991, AMM 110 (2003), p. 155
82 10890. Proposed by Raymond Mortini, Universit´ e de Metz, France. Letd1andd2 be two metrics on
a nonempty set Xwith the property that every ball in ( X, d 1) contains a ball in ( X, d 2) and vice versa.
Must d1andd2generate the same topology? Solution to problem 10890, AMM 108 (2001), p. 668
Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Letddenote
the Euclidean metric on Rand let fbe an injective real-valued function on R. It is easy to see that the
function ρ(x, y) =|f(x)−f(y)|defines a second metric on R, i.e. satisfies the axioms (D1) ρ(x, y)≥0, ρ(x, y) =
0⇐⇒x=y, (D2) ρ(x, y) =ρ(y, x) (D3) ρ(x, y)≤ρ(x, z) +ρ(z, y) for all x, y, z ∈R. LetBd(x0, ■) resp. Bρ(x0,
■) denote the open balls of radius ■and center x0with respect to the distances dandρ. Let us now
additionally assume that fis increasing, one-sided continuous but not continuous, and has only a finite
number of discontinuities. This guarantees that I:=f(R) is a union of non-degenerated intervals, with
pairwise disjoint closures. The inverse function f−1:I→R then is continuous on I. Fix x0. Hence for every
■ >0 there exists δ >0 such that Bρ(x0, δ)⊆ Bd(x0, ■). Letx0be a point at which fis, say, left-continuous.
Then for every ■ >0 there exists δ >0 such that for all x < x 0,d(x, x0)< δimplies ρ(x, x0) =|f(x)−f(x0)|< ■.
Let x1=x0−1 2δ. Then Bd(x1, δ/2)⊆Bρ(x0, ■). Thus each ball in the d-metric contains a ball in the
ρ-metric, and vice-versa. It is clear that the identity map id: ( R, ρ)→(R, d), although being continuous,
has no continuous inverse. Note that id : ( R, d)→(R, ρ) is continuous at x0if and only if fis continuous at
x0. Thus the two topologies are distinct. Remark If we additionally assume that ( X, d j) are topological
vector spaces, then the answer is yes. This is due to the fact that these topologies can be generated by
translation invariant metrics d′ 1andd′ 2. In fcat, ∀ε >0∃δ >0 :Bd′ 1(x0, δ)⊆Bd′ 2(0, ε/2). In particular,
x0and−x0 are in Bd′ 2(0, ε/2). Hence Bd′ 1(0, δ) =−x0+Bd′ 1(x0, δ)⊆Bd′ 2(0, ε/2) +Bd′ 2(0, ε/2)⊆Bd′ 2(0,
ε). The problem was also solved by Matthias Bueger and Dietmar Voigt (Germany).
83 Solution to problem 10857 (a), AMM 108 (2001), p. 172 Raymond Mortini - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetC2n=nX j=0x2j (2j)!andS2n+1=nX j=0x2j+1 (2j+ 1)!.
We show that, for every x >0, the sequence (S2n+1 C2n) is strictly decreasing, whereas (S2n−1 C2n) is
stricly increasing. Since both sequences con- verge to tanh xwe get thatS2n−1 C2n<tanh <S2n+1 C2n.
i) We have the following equivalences: (S2n+1 C2n)■⇐⇒S2n+1 S2n−1<C2n C2n−2⇐⇒S2n−1+x2n+1
(2n+1)! S2n−1<C2n−2+x2n (2n)! C2n−2⇐⇒ ⇐⇒1 +x2n+1 (2n+1)! S2n−1<1 +x2n (2n)!
C2n−2⇐⇒xC2n−2<(2n+ 1)S2n−1⇐⇒ n−1X j=0x2j+1 (2j)!<(2n+ 1)n−1X j=0x2j+1 (2j+ 1)!(1) But1
(2j)!<(2n+ 1)1 (2j+1)!⇐⇒2j+ 1<2n+ 1, which is true. Since x >0 we get (1). ii) That (S2n−1 C2n) is
stricly increasing, is shown in exactly the same way. To sum up, we get C2n+2 C2n<S2n+1
S2n−1<C2n C2n−2.
84 . Solution to problem 10854 AMM 108 (2001), p. 171 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Suppose that f:R→Ris a function, continuous at the origin,
and satisfying f(x+ 2f(y)) =f(x) +f(y) +y (1) for all x, y∈R. First, we shall show that fis continuous
everywhere. In fact, f■ x+ 2f(x+ 2f(y))■ =f(x+ 2[f(x) +f(y) +y]) =f■ [x+ 2y+ 2f(y)] + 2 f(x)■ = =f■ (x+ 2y) +
2f(y)■ +f(x) +x=f(x+ 2y) +f(y) +y+f(x) +x. (2) On the other hand: f■ x+ 2f(x+ 2f(y))■ =f(x) +f(x+ 2f(y)) +x+
2f(y) = =f(x) + [f(x) +f(y) +y] +x+ 2f(y) = 2 f(x) + 3f(y) +y+x. (3) By (2) and (3) we get that f(x+ 2y) =f(x) +
2f(y)∀(x, y)∈R2. In particular, by setting x=y= 0, we see that f(0) = 0. It easily follows that fis continuous
at every point x∈R. So, in order to continue, we may assume that fis a continuous solution of (1).
Letx=y. Then f(y+ 2f(y)) =y+ 2f(y). (4) First we shall determine all continuous solutions of (4). Let g(y)
=y+ 2f(y). Since gis continuous, g(R) is either a singleton or a nondegenerate interval I. Ifgis constant,
say g≡c, then f(y) =c−y 2and so c=f(y+ 2f(y)) = f(c), from which we conclude that c= 0. Hence f(y) =−y 2.
Ifgis not constant, take z∈I; that is y+ 2f(y) =g(y) =zfor some y. Then f(z) =z. Hence fis the identity on I.
It follows that 3 z=z+ 2f(z) =f(z+ 2f(z)) =g(z). Therefore 3 z∈Iand so I= m,∞[ for some m∈R∪ {−∞} .
Thus f(x) =xfor every x > m . Since g≥m, we have that f≥m−y 2on ]− ∞, m]. To prove the converse,
choose m∈R. Let f∗be any continuous function on ] − ∞, m] such thatf∗(y)≥m−y 2fory≤mand so that
f∗(m) =m. Then ef(y) =( f∗(y) if y≤m y ify≥m(5) is a continuous solution of (4). We deduce that any
continuous solution of (1) necessarily has the form (5) or equals −1 2y. We shall now show that only for
f∗=id, we really get a solution of (1). So let fbe a continuous solution of (1). Then f=effor some f∗. Fix x <
m . Take y > m so that x+ 2y > m . Then f(x+ 2f(y)) =f(x+ 2y) =x+ 2yandf(x) +y+f(y) =f∗(x) + 2y. Hence
(1) implies that f∗(x) =x.
85 We conclude that fis a continuous solution of (1) if and only if f(x) =xorf(x) =−x 2on R.
86 Solution to problem 10768 AMM 106 (1999), 963 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - a) Let f(x) =p |x|sin1 xforx■= 0 and f(0) = 0. Then fis
continuous on R. Let g be a differentiable function on R. Then, in every neigborhood of 0, h:=f+g−g(0)
takes negative and positive values. In fact, suppose that h≥0 on [0 , ε]. Thenh(x) x≥0 on [0 , ε]. But lim
inf x→0+h(x) x=g′(0) + lim inf x→0+1√xsin1 x=−∞, a contradiction. Thus f+gis not monotone on any
interval centered at 0. b) Let f(x) =x2sin1 x2forx■= 0 and f(0) = 0. Then fis differentiable on R,f′(0) = 0,
butf′(x) = 2 xsin1 x2−2 xcos1 x2takes arbitrarily large negative and positive values in any neighborhood
Uof 0. Let gbe any C1(R) function. In particular, g′is bounded on every compact interval centered at 0.
Hence f′+g′takes arbitrary large negative and positive values inU. Thus f+gis not monotone on any
interval centered at 0. c) We show that for every function f∈C1(R) there exists an entire function g(that
is a function holomorphic on the whole plane), real-valued on R, such that f+gis increasing on R. In fact,
f′+ 2|f′|+ 2ε≥2ε >0 onR. Let q= 2|f′|+ 2ε. Then qis continuous on R. By Carleman’s theorem (see [C] and
[G], p. 125), there exists an entire function Qsuch that q−Q∞≤ε, where ·∞denotes the supremum norm
on R. Let G(x) = Re Q(x). Then q−G≤ε. Moreover, the function H(z) =1 2■ Q(z) +Q(z)■ is analytic in C,
and Hcoincides onRwith G. Now it is easy to check that f′+G≥ε >0. Let gbe a primitive of G. Then gis
the trace of an entire function and f+gis (strictly) increasing, since its derivative is strictly positive.
References [G] Dieter Gaier, Approximation im Komplexen , Birkhæuser-Verlag, Basel, 1980. [C]
Carleman, T,: sur un th´ eor` eme de Weierstrass , Ark. Mat. Astronom. Fys., 20B (1927), 1-5.
87 Solution to problem 10747 AMM 106 (1999), p. 685 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that all differentiable solutions foff′(f(t)) = 2 f(t), t∈R,
f(1) = 1, and having only one real root, have the form f(t) =t2fort≥0 and f(t) =g(t) for t <0, where g is an
arbitrary differentiable function, defined on ] − ∞,0] satisfying g(t)>0 for t <0 and g(0) = g′(0) = 0. The
assumption, that fshould be twice differentiable in a neighboorhood of 0, is not important. Proof Letfbe a
solution of the problem. Put h=f■f−f2. Then h′= (f′■f)f′− 2f′f=f′(f′■f−2f)≡0. Hence his a constant, say C.
Because h(1) = 0, we see that C= 0 and so f■f=f2. Let y∈f(R). Then f(x) = yfor some x∈R. Therefore
f(y) =f(f(x)) = f2(x) =y2. By hypothesis, {0,1} ⊆f(R). By continuity we conclude that [0,1]⊆f(R). Since the
left derivative at x= 1 is 2, the differentiability of fnow implies that there exists points x0greater than 1 for
which f(x0)> f(1) = 1. Since fn+1=f2n, we obtain that fn+1(x0) = [f(x0)]2n→ ∞ . Hence fis unbounded. By
the intermediate value theorem, we then get that [0 ,∞]⊆f(R). Hence f(x) =x2forx≥0. To determine the
behaviour of ffor negative values, we use the hypothesis that fshould have only one zero. Since f(0) =
0, by continuity, we conclude that either f(x)<0 for all x <0 or f(x)>0 for all x <0. But f(x0)<0 for some (all)
x0<0 implies that f(f(x0)) =f2(x0)>0, a contradiction. Thus f(x)>0 for x >0. It is easy to check that every
function of the form f(x) =x2forx≥0 and f(x) =g(x) for x <0, where g >0 is differentiable and satisfies g(0)
= g′(0) = 0, is a solution of f■f=f2. Hence, by differentiating, f′(f(x))f′(x) = 2 f′(x)f(x). If f′(x)■= 0, then we
are done. If f′(x0) =g′(x0) = 0 for some x0<0, then we use the fact that y:=f(x0)>0 and that for these
positive values f(y) =y2. Hence, f′(f(x0)) = 2 f(x0). So we obtain a solution of our functional equation.
88 Solution to problem 10739 AMM 106 (1999), p. 586 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetH(x) =f(x)−f(0) x−0=f(x) x. Since f′′(x)>0, the function fis
strictly convex and both its derivative and the quotient Hare strictly increasing (see e.g. W. Walter,
Analysis 1, Springer-Verlag, p. 303). Moreover, His continuous on ]0 ,1]. Note that H(1) = f(1) and that
H(0) := lim x→0f′(x) exists in [ −∞, f(1)]. Hence, by the intermediate value theorem, there exists for
every value wwith H(0)< w < H (1) a point b∈]0,1[ with H(b) =w. Now choose a∈]0,1[ such that w:=f′(a)
satisfies H(0)< w < H (1) (such a choice obviously is possible). Thus there exists b∈]0,1[ so thatf(b)
b=H(b) =f′(a). Choose xa∈]0, a[ so that H(a) =f′(xa). Due to the monotonicity of f′we obtain: H(a)
=f′(xa)< f′(a) =H(b). Since His monotone, bis unique and satisfies a < b < 1.
89 Solution to problem 10697 AMM 105 (1998), p. 955 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - This is nothing but a Lagrange interpolatory argument: In fact
let w1,···, wn∈C. Then p(z) =nX k=1wkQn j=1,j■=k(z−zj)Qn j=1,j■=k(zk−zj) is the unique polynomial of
degree at most n−1 satisfying p(zk) =wk, k= 1,···, n. Now choose wk= 1 for every k. Since q(z)≡1
satisfies the interpolation q(zk) =wk, we obtain from uniqueness that q=p. Let z= 0. Then 1 =q(0) =nX
k=1nY j=1 j■=k(−zj) zk−zj= (−1)n−1nX k=1Qn j=1,j■=kzjQn j=1,j■=k(zk−zj). Dividing byQn j=1zj, yields
the assertion nX k=11 zknY j=1 j■=k1 zk−zj=(−1)n−1 Qn j=1zj.
90 Solution to problem 10651 AMM 105 (1998), p. 271 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We prove a stronger version than in the formulated problem
(see [5], which was based on this). Proposition 1 Letuandvbe two non constant harmonic functions on a
domain D⊆C. Suppose that uvis harmonic. Then uhas an harmonic conjugate ˜uonDand there are
constants a, b∈Rsuch that v=a˜u+b. (1) Remarks . (1) If uis a constant, then (1) is not true (because
vmay be chosen to be any harmonic function). (2) If vis a constant then (1) is true for a= 0, provided a
harmonic conjugate exists. A well known sufficient condition for the existence of a harmonic conjugate
being that Dis simply connected. (3) Of course, if vis any harmonic function satisfying (1), then uvis
harmonic. Solution Let ∆ be the Laplace operator. Because ∆ u= ∆v= 0 we obtain: 0 = ∆( uv) =■ uxxv+
2uxvx+vxx■ +■ uyyv+ 2uyvy+vyy■ = 2■ uxvx+uyvy■ . Letf=ux−iuyandg=vx−ivy. The harmonicity of
uandvimply that fandgsatisfy the Cauchy-Riemann differential equations; hence fandgare holomorphic.
It is easy to see that Refg=uxvx+uyvy. Thus Re fg≡0 on D. LetZ(g) ={z∈D:g(z) = 0}denote the zero set
of g. It is a discrete subset of Dprovided thatg■≡0. Since vis assumed not to be a constant, we see that
g■≡0. Then on D\Z(g) we have Ref g= Refg |g|2. Thus Ref g≡0 onD\Z(g). This implies, in view of the
analyticity, that f gis a pure imaginary constant, sayf g≡iλonD\Z(g). Hence f=iλgonD. The definitions
offandgnow yield that ux=λvyanduy=−λvx. Consequently, by the Cauchy-Riemann equations, the
function u+iλvis holomorphic on D. In particular, uhas an harmonic conjugate onD. (Note that we do not
have assumed that Dis simply connected.) Thus, for any other harmonic conjugate ˜ uofu, we have λv=
˜u+cfor some constant c∈R. Note that unot constant implies that λ■= 0. Thus vhas the desired form (1).
A natural question now is the following. Let uandvbe two harmonic functions on a domain D⊆C. Then (
u+iv)2=u2−v2+ 2iuv. Assume that u2−v2is harmonic. What can be said forv? We have the following
result: Proposition 2 Assume that u,vandu2−v2are harmonic in a simply connected domain D⊆C. Then
there exists a∈Randθ∈[0,2π[such that v= cos θ u−sinθ˜u+a. (2) Conversely, every function vsatisfying
(2) for a harmonic function uhas the property that u2−v2is harmonic. Proof Because ∆ u= ∆v= 0 we
obtain: 0 = ∆( u2−v2) = 2■ u2 x+u2 y−(v2 x+v2 y)■ .
91 Hence u2 x+u2 y=v2 x+v2 y. Again, let f=ux−iuyandg=vx−ivy. As above, fandgare holomorphic on D.
Moreover |f|2=|g|2. Thus gis a rotation of f, say g=eiθf. Letz0∈D. Since Dis simply connected,
uandvhave harmonic conjugates ˜ uand ˜v respectively, satisfying ˜ u(z0) = ˜v(z0) = 0. Let
F=u+i˜uandG=v+i˜v. Then, by Cauchy- Rieman, F′=ux+i˜ux=ux−iuy=f. Similiarly G′=g. Thus
G=eiθF+cfor some constant c∈C. Taking real parts yields v= cos θ u−sinθ˜u+a for some real constant
a. The converse is easy to check. The above results are related to the following more general result:
Proposition 3. Lethbe an entire function and let u:D→Randv:D→Rbe two nonconstant harmonic
functions in a simply connected domain D. Let ˜ube a harmonic conjugate of uinD. Then h(u+iv)
:D→Cis harmonic if and only if v=±˜u+afor a constant a∈R. Proof Since his holomorphic, we have, by
Cauchy-Riemann, hy=ihxandhx=h′. Hence hxx=h′′, hxy=hyx=ih′′andhyy=−h′′. As above, let
f=ux−iuyandg=vx−ivy.Then ∆[h■(u+iv)] =h′′■(u+iv)·[(|f|2− |g|2) + 2iRefg]. Obviously h′′■q■≡0 for any
nonconstant continuous function q. Hence h(u+iv) is harmonic if and only if |f|=|g|and Re fg= 0. By the
paragraphs above we conclude that f=iλg for some λ∈R. Hence |λ|= 1. Thus ux=±vyand−uy=±vx. So
vor−vis a harmonic conjugate of uinD. Therefore v=±˜u+a. To prove the converse, we have simply to
note that the composition of a holomorphic function with a holomorphic or anti-holomorphic function is
harmonic.
92 Solution to problem 10638 AMM 105 (1998), p. 69 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - In the following we present a solution to problem number
10638. We shall not only compute the functions S0,···, S3, but we will give an explicit value for all m∈N.
To this end we need the following Lemma. Lemma Letf(z) =zez. Then fis invertible in a neighborhood of
the origin in Cand the inverse function has the Taylor representation f−1(w) =∞X n=1nn−1 n!(−1)n−1wn,
which converges for |w|<1 e. Proof By the residue theorem it is easy to see that whenever fis
holomorphic and injective in a disque D⊆C(or even a simply connected domain), then f−1(w)n(Γ,
f−1(w)) =1 2πiˆ Γzf′(z) f(z)−wdz, where Γ is an arbitrary cycle (=finite union of closed, piecewise
C1-curves) in D. Applying this formula for f(z) =zezand the disk |z|<2δ,δsmall enough, we obtain : dn
(dw)nf−1(w) =n! 2πiˆ |z|=δz(z+ 1)ez (zez−w)n+1dz. Thus, for the power series f−1(w) =P∞ n=0anwnwe
have a0= 0 and for n≥1: an=1 2πiˆ |z|=δz+ 1 zne−nzdz=1 2πi∞X k=0(−1)kˆ |z|=δz(nz)k+ (nz)k znk!dz=
(−1)n−1nn−1 n!. By d’Alembert’s rule it is easy to check that the radius of convergence is 1 /e. ■
Proposition For0< λ < 1andm∈Z, letgm(λ) =λmSm(λ), where Sm(λ) =∞X n=1e−λn(λn)n−m/n!. (1) Then,
for m∈ {1,2,···},gmis a polynomial of degree mvanishing at the origin, say gm(λ) =−Pm n=1bn,m(−λ)n,
and the coefficients bn,mare given by the recurrence relation bn,m=1 n(bn,m−1+bn−1,m−1), b 1,1= 1.
(3) Solving these difference equations yields bn,m=nX j=11 n!■n j■ (−1)j−1■1 j■m−n . (4) Proof We
note that, by Stirling’s formula, the series gm(λ) converges locally uniformly in 0≤λ <1, but does not
converge whenever λ= 1 and m= 0. Note that gm(0) = 0. Due to
93 local uniform convergence, it is easy to see that, in order to obtain g′ m(λ), one can differentiate the
series for gmterm by term. This yields that for m∈Z g′ m(λ) =1−λ λgm−1. (5) Later we shall show that
g1(λ) =λ. Hence, by induction on (5), it is clear that for m= 1,2,···the function gmis a polynomial
vanishing at the origin, say gm(λ) =−Pm n=1bn,m(−λ)n. If we let x=−λ, then we obtainPm n=1nbn,mxn=
(1 + x)Pm n=1bn,m−1xn.Comparing coeffi- cients, finally yields (3). This difference equation can be
solved by the usual methods. May be Maple or Mathematica gives the solution. In any case, by the
uniqueness of the solution, it suffices to show that (4) verifies the difference equation. Note also, that
for n > m , the bn,min (4) are 0. This follows from the fact that the p-th difference operator Dp(an) =Pp
j=0■n j■ (−1)jan−jvanishes identically whenever anis a polynomial (in n) of degree strictly less than p.
For the readers convenience, here are the coefficients for m= 1,···,5: 1 11 2 13 41 6 17 811 361 24 115
1685 21625 2881 120 The case m=1 In that case we have g1(λ) =λ∞X n=1e−λn(λn)n−1/n! =∞X
n=1nn−1 n!(λe−λ)n. Letw=−λe−λ. Now, for w∈C,|w|<1 e, the function h(w) =P∞ n=1nn−1
n!(−1)n−1wnis, by Lemma 1, nothing but the inverse function of the holomorphic function f(z) =zez,|z|<
δfor sufficiently small δ >0. Thus g1(λ) =λ. The case m=0 By (5) we see that 1 = g′ 1(λ) =1−λ λg0(λ).
Hence, g0(λ) =λ 1−λ. Using (5) it is also easy to derive, inductively, the values of gmfor negative
integers m. For example we get: g−1(λ) =λ (1−λ)3, g −2(λ) =λ (1−λ)5(1 + 2 λ), g −3(λ) =λ (1−λ)7(1 + 8
λ+ 6λ2). In general, one can convince oneself that for m∈Z, m < 0,gm(λ) has the form gm(λ) = λ
(1−λ)−2m+1Qm(λ), where Qis a polynomial of degree −m−1 with value 1 at the origin and satisfying the
differential equations Qm−1(λ) =λ(1−λ)Q′ m(λ) + (1 −2mλ)Qm(λ). Due to lack of time we were not able
to solve this explicitely. May be Maple and Mathematica will be helpfull.
94 Solution to problem 10624 AMM 104 (1997), p. 871 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - By Leibniz’s criteria, we know that Snactually converges and
that Sn≥0 for every n∈N. Since Sn=an−Sn+1, we see that Sn≤anand soPanSn≤Pa2 n. Thus the
convergence ofPa2 nimplies the convergence ofPanSn. Now assume thatPanSn=P(Sn+Sn+1)Sn(1)
converges. Since all the terms are positive, we deduce the convergence of the sumsPS2
nandPSn+1Sn. A shift of the variable yields thatPS2 n+1converges. HencePSn+1(Sn+1+Sn) (2)
converges. Summing (1) and (2) yields thatPa2 n=P(Sn+1+Sn)2=P(Sn+Sn+1)Sn+PSn+1(Sn+1+Sn) is
convergent.
95 Solution to problem 10605 (b) AMM 104 (1997), p. 567 Raymond Mortini - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Writen2s−m2s n2s+m2s=1−(m/n)2s 1 + (m/n)2s. Let
y=■m n■2 . Then 1−ys 1 +ys=Qs−1 j=0■ 1−yexp(−i2πj s)■ Qs−1 j=0■ 1−yexp(−iπ+2πj s)■. Since
ε∈Cis an s-root of 1 [resp. (-1)] if and only if εis an s-root, we obtain: 1−ys 1 +ys=(1−y)(1 + y)Qp−1 j=1
1−yexp(−i2πj s) 2 Qp−1 j=1 1−yexp(−iπ(2j+1) s) 2 ifs= 2pand 1−ys 1 +ys=(1−y)Qp j=1 1−yexp(−i2πj s) 2
(1 +y)Qp−1 j=1 1−yexp(−iπ(2j+1) s) 2 ifs= 2p+ 1. This can be written by a single formula: 1−ys 1 +ys=
(1−y)(1 + y)(−1)ss−1Y k=1 1−yexp(−iπk s) 2(−1)k . (1) In particular s−1Y k=1 1−exp(−iπk s) 2(−1)k = lim
y→11−ys 1 +ys (1−y)(1 + y)(−1)s=s 2·2(−1)s. (2) It is easy to check that P:=s−1Y k=1(2π2m2)(−1)k=( 1,
ifsis odd 1 2π2m2,ifsis even.(3)
96 Now use the infinite product representation of the function sin πz. This gives: sinπz πz=∞Y n=1■
1−z2 n2■ , and ∞Y n=1■ 1 +z2 n2■ =siniπz iπz=sinhπz πz. Moreover we have by de l’Hˆ opital’s rule
that ∞Y n■=m■ 1−m2 n2■ = lim z→msinπz πz 1−■z m■2 =(−1)m+1 2. Finally we need that |sinz|2=1
2(cosh 2 y−cos 2x) for z=x+iy. Put all this together to get from (1) 2(−1)sY n■=mn2s−m2s
n2s+m2s=(−1)m+1 2■sinhπm πm■(−1)s ·s−1Y k=1 Y n■=m 1−■m nexp(−iπk 2s)■2! 2(−1)k =
=(−1)m+1 2■sinhπm πm■(−1)ss−1Y k=1 sin■ πmexp(−iπk 2s)■ πm■ 1−exp(−iπk s)■ 2(−1)k =
=(−1)m+1 2■sinhπm πm■(−1)sQs−1 k=1■ 1 2■ cosh■ 2πmsinπk 2s■ −cos■ 2πmcosπk 2s■■■ (−1)k
Qs−1 k=1 1−exp(−iπk s) 2(−1)k ·Qs−1 k=1(πm)2(−1)k= =(−1)m+1 2■sinhπm πm■(−1)sQs−1 k=1■
cosh■ 2πmsinπk 2s■ −cos■ 2πmcosπk 2s■■(−1)k Qs−1 k=1 1−exp(−iπk s) 2(−1)k ·Qs−1
k=1(2π2m2)(−1)k= =(−1)m+1(sinh πm)(−1)sQs−1 k=1■ cosh■ 2πmsinπk 2s■ −cos■ 2πmcosπk
2s■■(−1)k (πm)(−1)ss 2(−1)s·P Clearly 1 (πm)(−1)ss·P= 2(1+(−1)s)/2■πm s■ . Putting ε= (1 + (
−1)s)/2, we get the final equality: P2s=Y n■=mn2s−m2s n2s+m2s= = (−1)m+12εmπ s(sinh
πm)(−1)ss−1Y k=1■ cosh■ 2πmsinπk 2s■ −cos■ 2πmcosπk 2s■■(−1)k . Ifs= 1 we interpret the empty
product as 1. This gives P2(m) = (−1)m+1πm/sinh(πm).
97 Solution to problem 10588/10595 AMM 104 (1997), p. 456 Raymond Mortini - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that P=∞Y n=1e−1 n■ 1 +1 n+1 2n2■
=eπ/2+e−π/2 πeγ. Let Γ(z) =" eγzz∞Y n=1■ 1 +z n■ ez/n#−1 be the Gamma function and let ε=1 2(1 +i).
Then ε= 1−ε. Hence, as is well known, Γ(ε)Γ(ε) = Γ( ε)Γ(1−ε) =π sinπε. Therefore sinπε π=eγεε∞Y n=1■
1 +ε n■ e−ε/n×eγεε∞Y n=1■ 1 +ε n■ e−ε/n= =eγ1 2∞Y n=1■ 1 +ε n■■ 1 +ε n■ e−1/n= =eγ1 2∞Y n=1■
1 +1 n+1 2n2■ e−1/n. Hence P=2 sinπε πeγ=2 cosh π/2 πeγ, which is the assertion.
98 Solution to problem 6654 AMM 98 (1991), p. 273 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - - -
------------------------------
99 Solution to problem 6648 AMM 98 (1991), p. 63 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - - - - -
-----------------------------
100 Solution to problem E3329 AMM 96 (1989), p. 445 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - -
--------------------------------
101 Solution to problem E3325 AMM 96 (1989), p. 445 Raymond Mortini - - - - - - - - - - - - - - - - - - - - - -
--------------------------------
102 2.Mathematics Magazine ©Mathematical Association of America, 2025. Solution to problem 2202
Math. Mag. 97 (2) 2024, p. 434 Raymond Mortini, Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - Using complex analysis, we prove that ■ ■■ ■I1:=´2π 0cos(cos t)
cosh(sin t)dt= 2π and ■ ■■ ■I2=´2π 0sin(cos t) cosh(sin t)dt= 0. First note that cosh z= cos( iz). Hence,
by using that cosx+ cos y= 2 cos■x+y 2■ cos■x−y 2■ , and sinx+ sin y= 2 sin■x+y 2■ cos■x−y 2■ ,
we obtain cos(cos t) cosh(sin t) = cos(cos t) cos( isint) = cos■eit+e−it 2■ cos■eit−e−it 2■ =1 2■
cos(eit) + cos( e−it)■ , and sin(cos t) cosh(sin t) = sin(cos t) cos( isint) = sin■eit+e−it 2■ cos■eit−e−it
2■ =1 2■ sin(eit) + sin( e−it)■ . Now, by Cauchy’s residue theorem in complex analysis, I1=1 2ˆ2π 0■
cos(eit) + cos( e−it)■ dtz=eit =1 2i‰cosz+ cos(1 /z) zdz =1 2i2πi■ Reshcosz z,0i + Res■cos(1 /z)
z,0■■ =π(1 + 1) = 2 π, and
103 I2=1 2ˆ2π 0■ sin(eit) + sin( e−it)■ dtz=eit =1 2i‰sinz+ sin(1 /z) zdz =1 2i2πi■ Res■sinz z,0■ +
Res■sin(1/z) z,0■■ =π(0 + 0) = 0 .
104 Solution to problem 2191 Math. Mag. 97 (2) 2024, p. 223 Raymond Mortini, Rudolf Rupp - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We prove that this area Antends to■ ■■
■8/π∼2.5464790 . . .. To this end, we use the Fourier series for f(x) :=|sinx| (39) |sinx|=2 π−4 π∞X
k=1cos(2 kx) 4k2−1, x∈R, which can easily be obtained by noticing that that fis even and π-periodic,
and so f(x)∼a0 2+∞X n=1akcos(2 kx) with ak=2 πˆπ 0f(x) cos(2 kx)dx=2 πˆπ 0sinxcos(2 kx)dx=1 πˆπ 0■
sin(1+2 k)x+sin(1 −2k)x■ dx. Note that the desired area Anis given by´π 0|cosx−cos(nx)|dx. By the
addition theorem |cosx−cos(nx)|= 2|sin ((n−1)(x/2)) sin (( n+ 1)( x/2))|. Hence q(x) := 2|sin(n−1)(x/2) sin(
n+ 1)( x/2)|= 2 2 π−4 π∞X k=1cosk(n−1)x 4k2−1! 2 π−4 π∞X k=1cosk(n+ 1)x 4k2−1! . Since for m,
n∈N={0,1,2, . . .},m■=n, ˆπ 0cos(mx) cos( nx)dx=1 2ˆπ 0■ cos(m−n)x+ cos( m+n)x■ dx= 0, we deduce
from the fact that the Fourier series (39) is absolutely and uniformly convergent, that for n≥2 An=ˆπ
0q(x)dx=8 π2π+32 π2∞X j,k=1ˆπ 0cosj(n−1)xcosk(n+ 1)x (4k2−1)(4j2−1)dx =8 π+16π π2∞X j,k=1
j(n−1)=k(n+1)1 (4k2−1)(4j2−1). Now j(n−1) = k(n+ 1) if j=r(n+ 1)mandk=r(n−1)mfor some m∈Nand r=( 1
if nis even 1 2ifnis odd In fact, if nis even, then the odd numbers n−1 and n+ 1 are relatively prime,
since otherwise a joint divisor must also divide 2 = ( n+ 1)−(n−1). Hence k=m(n−1) and j=m(n+ 1) for
some m∈N. And if nis odd, then n−1 and n+ 1 are even and 2 then is the gcd of n−1 and
105 n+1 since ( n+1)−(n−1) = 2. Now let n≥3. Due to 1 ≤2(1/4)m2(n±1)2≤2r2m2(n+1)2, we finally
conclude that 8 π≤An=8 π+16π π2∞X m=11 (4r2m2(n−1)2−1)(4r2m2(n+ 1)2−1) ≤8 π+16π π2∞X m=11
2r2m2(n+ 1)2=8 π+1 (n+ 1)216π π21 2r2∞X m=11 m2 →8 π. Remark. Using that for a /∈Z,∞X m=11
m2−a2=1−aπcot(aπ) 2a2, we obtain An=8 π+16 π1 16nr2∞X m=1 1 m2−1 4r2(n−1)2−1 m2−1 4r2(n+1)2!
=8 π+16 π1 16nr2 1−π 2r(n−1)cot(π 2r(n−1)) 2 4r2(n−1)2−1−π 2r(n+1)cot(π 2r(n+1)) 2 4r2(n+1)2! =8 π+2
πnr2■ r2(n−1)2−r(n−1)π 2cotπ 2r(n−1)−r2(n+ 1)2+r(n+ 1)π 2cotπ 2r(n+ 1)■ =−n−1 nrcotπ 2r(n−1)+n+ 1
nrcotπ 2r(n+ 1). For example, A5=4 5(3√ 3−2)∼2.5569219 . . . A4=5 4q 5 + 2√ 5−3√ 3 4∼2.54806631 . . .
A3=8 3∼2.6666666 . . . A2=3√ 3 2∼2.59807621 . . .. It is now very easy to determine the limit directly
(one may replace rabove even by any x■= 0): Let a:=π 2x(n+1)undb:=π 2x(n−1). Then xAn= (cot
a−cotb) +1 n(cota+ cot b), and so xAn=sin(b−a) sinasinb+1 nsin(a+b) sinasinb. Note that b−a=π x1
n2−1anda+b=π xn n2−1. Since lim x→0sinx x= 1,xAnhas the same asymptotic as π x1 n2−1 π 2x(n+1)π
2x(n−1)+1 nπ xn n2−1 π 2x(n+1)π 2x(n−1)=4x π+4x π=8x π.
106 Solution to problem 2193 Math. Mag. 97 (2) 2024, p. 223 Raymond Mortini, Rudolf Rupp - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetN={0,1,2, . . .}. We assume that the
terms of the seriesP∞ j=1ajare regrouped as follows (this regrouping does not change the partial
sums): 1√ 1−■1√ 2+1√ 3■ +■1√ 4+1√ 5+1√ 6■ −■1√ 7+1√ 8+1√ 9+1√ 10■ +−··· =:∞X k=1(−1)k−1Pk,
where for k∈N Pk:=Ak+1−1X j=Ak1√j, and where the number Ak=k(k−1) 2+ 1 is the j-index of the first
summand in the k-th group. The sum Pkhasksummands. We now estimate each Pk: (40) Pk≤Ak+1−1X
j=Ak1√j≤k√Ak=k√ 2p k(k−1) + 2≤2√ 2. Pk≥kp Ak+1−1=√ 2kp k(k+ 1)≥√ 2k k+ 1≥√ 2 2. By Cauchy’s
criterion, the series diverges as the blocks Pkdo not go to zero. Next we estimate the partial sums NX
k=1(−1)k−1Pk. To this end, we use the useful inequality 2(√ n+ 1−√n)≤1√n≤2(√n−√ n−1). This yields (via
telescoping property) Pk≤2(p Ak+1−1−p Ak−1) = 2√ 2√ k√ k+ 1 +√ k−1=√ 2■ 1 +1 81 k2■ +O(1 k2)
Pk≥2(p Ak+1−p Ak)≥2√ 2kp k(k+ 1) + 2 +p k(k−1) + 2=√ 2■ 1−7 81 k2■ +O(1 k2). Numerical
computations let us guess that the Pkareincreasing . Anyway, the partial sums formed with full blocks
write as LN:=NX k=1(−1)k−1Pk=( (P1−P2) + (P3−P4) +···+ (PN−1−PN) if Neven (P1−P2) + (P3−P4)
+···+ (PN−2−PN−1) +PNifNodd, and the general one is given by
107 (41) Sn:=nX j=1aj=NX k=1(−1)k−1Pk+ (−1)NnX j=AN+11√j, where Nis the unique number for which
AN+1=N(N+1) 2+1≤n <(N+2)(N+1) 2=AN+2−1. It remains to estimate the differences |Pk−Pk+1|.
Pk−Pk+1≤ √ 2 +√ 2 81 k2+O(1 k2)! − √ 2−7√ 2 81 (k+ 1)2+O(1 (k+ 1)2)! ≤√ 2 k2+O(1 k2) Pk−Pk+1≥ √
2−7√ 2 81 k2+O(1 k2)! − √ 2 +√ 2 81 (k+ 1)2+O(1 (k+ 1)2)! ≥ −√ 2 k2+O(1 k2). Hence, for all k≥k0,
|Pk−Pk+1| ≤√ 2 k2+O(1 k2)≤√ 2 + 1 k2 and so, for all k, (42) |Pk−Pk+1| ≤Ck2. We deduce from (40) and
(41) that for every nandNchosen as above |Sn| ≤ | LN|+nX j=AN+11√j ≤■N/2■X j=1|P2j−1−P2j|+ PN|{z}
comes from the case Nodd+nX j=AN+11√j ≤∞X k=1|Pk−Pk+1|+PN+PN+1 (40) ≤ (42)C∞X k=11 k2+ 4√
2 =: ˜C. Remark We also deduce that the associated parenthesized series converges:■1√ 1−1√ 2−1√
3■ +■1√ 4+1√ 5+1√ 6−1√ 7−1√ 8−1√ 9−1√ 10■ +■1√ 11+1√ 12+1√ 13+1√ 14+1√ 15−1√ 16−1√ 17−1√
18−1√ 19−1√ 20−1√ 21■ +···
108 Solution to problem 2187 Math. Mag. 97 (1) 2024, p. 81 Raymond Mortini, Rudolf Rupp - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that for r > s≥0, ■ ■ P:=∞Y
n=0■ 1 +cosh(2ns) cosh(2nr)■ =sinhr coshr−coshs. To this end we first show via induction that (43)kY
n=0cosh(2nx) =sinh(2k+1x) 2k+1sinhx. In fact, let k= 0. Thensinh 2 x 2 sinh x= cosh x. If (43) is correct
for some k, then k+1Y n=0cosh(2nx) = kY n=0cosh(2nx)! ·cosh(2k+1x) =sinh(2k+1x)
2k+1sinhx·cosh(2k+1x) =sinh(2k+2x) 2k+2sinhx. A way to come up with such a formula, is to use the
well-known funny formula kY n=0(1 +w2n) =1−w2k+1 1−w forw=e−xand by writing cosh x=ex(1
+e−2x)/2. Now 1 +coshu coshv=coshu+ cosh v coshv= 2cosh(u+v 2) cosh(u−v 2) coshv. Hence, with
u= 2nsandv= 2nr, Pk:=kY n=0■ 1 +cosh 2ns cosh 2nr■ = 2k+1kY n=0cosh(2n−1(r+s)) cosh(2n−1(r−s))
cosh 2nr = 2k+1cosh(r+s 2) cosh(r−s 2)Qk−1 j=0cosh 2j(r+s)Qk−1 j=0cosh 2j(r−s) Qk n=0cosh 2nr =
2k+1cosh■r+s 2■ cosh■r−s 2■sinh 2k(r+s) 2ksinh( r+s)sinh 2k(r−s) 2ksinh( r−s)2k+1sinhr sinh 2k+1r
= 4 cosh■r+s 2■ cosh■r−s 2■sinhr sinh( r+s) sinh( r−s)sinh 2k(r+s) sinh 2k(r−s) sinh 2k+1r Next we
claim that for r > s , lim k→∞sinh 2k(r+s) sinh 2k(r−s) sinh 2k+1r=1 2.
109 In fact, using that sinh x=ex 2(1−e−2x) we obain sinh 2k(r+s) sinh 2k(r−s) sinh 2k+1r=1
4e2k(r+s)e2k(r−s)(1−e−2k+1(r+s))(1−e−2k+1(r−s)) e2k+1r 2(1−e2k+2r) =1
2(1−e−2k+1(r+s))(1−e−2k+1(r−s)) (1−e2k+2r)→1 2. Now note that cosh2r−sinh2s= sinh( r+s) sinh( r−s),
and so 4 cosh■r+s 2■ cosh■r−s 2■sinhr sinh(r+s) sinh( r−s)= 2coshr+ cosh s sinh(r+s) sinh( r−s)sinhr
=2 sinh r coshr−coshs. Thus lim k→∞Pk=sinhr coshr−coshs.
110 Solution to problem 2186 Math. Mag. 97 (1) 2024, p. 81 Raymond Mortini, Rudolf Rupp - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - A function “arctanh x” does not exist in
the terminology we have learned (see R. Burckel, Classical Analysis in the plane, 2021, p. 135). It is
either arctan xor artanh x. Not knowing whether the letter cor the letter hin your statement is
superfluous, we consider both cases. So we will prove the following: (1)ˆ1 0artanh( x√ 2−x2) xdx=3
16π2∼1.850550825204 ···, (2)ˆ1 0arctan( x√ 2−x2) xdx=1 2C+π 4log(√ 2 + 1) ∼1.15021199360 ···.
where Cis the Catalan constant. We need the following well-known integral: Lemma 4. LetIn:=´π/2
0(sinx)ndx. Then I0=π/2andI1= 1. For n∈N∗:={1,2,3···} we have I2n=1 2·3 4·5 6···2n−1 2nπ 2=(2n)!
4n(n!)2·π 2=■2n n■ 4n·π 2. Proof. I2n=2n−1 2nI2n−2forn∈N∗andI0=π 2, because
2nI2n−(2n−1)I2n−2=ˆπ/2 0(sinx)2n−2■ 2nsin2x−(2n−1)■ dx =−ˆπ/2 0(sinx)2n−2■ (2n−1) cos2x−sin2x■
dx =−■ (sinx)2n−1cosx■π/2 0= 0. ■ Moreover, we will use that∞X n=01 (2n+ 1)2=π2 8as well as∞X
n=0(−1)n (2n+ 1)2=C. Finally we need that for |x| ≤1 arcsin x=∞X n=0■2n n■ 4nx2n+1 2n+ 1and arsinh
x=∞X n=0(−1)n■2n n■ 4nx2n+1 2n+ 1. Note that in view of Stirling’s formula(2n n) 4n∼1√π√n, so the
series converge absolutely for x=±1.
111 (1) We make the substitution x=√ 2 sint,dx=√ 2 cost dt. Then, by using´P=P´ (all terms are
positive), ˆ1 0artanh( x√ 2−x2) xdx =ˆπ/4 0artanh(sin(2 t)) sintcost dt =ˆπ/4 0∞X n=01 2n+ 1(sin(2
t))2n+1 sintcost dt sin(2t)=2 sin tcost=ˆπ/4 0∞X n=01 2n+ 1(sin(2 t))2n2 cos2t|{z} =1+cos(2 t)dt =∞X
n=01 2n+ 1 ˆπ/4 0(sin(2 t))2ndt+ˆπ/4 0(sin(2 t))2ncos(2 t)dt! 2t=u=1 2∞X n=01 2n+ 1 ˆπ/2
0(sinu)2ndu+ˆπ/2 0(sinu)2ncosudu! =1 2∞X n=01 2n+ 1 ■2n n■ 4n·π 2+1 2n+ 1! =π 4arcsin 1 +π2 16=3
16π2. (2) In this case case, the factor1 2n+1is replaced by(−1)n 2n+1. Moreover,´P=P´ , since NX
n=0(−1)n 2n+ 1(sinu)2n(1 + cos u) ≤∞X n=01 2n+ 1(sinu)2n(1 + cos u), which is an integrable majorant
by (1). Hence ˆ1 0arctan( x√ 2−x2) xdx=π 4arsinh 1 +1 2C. Since arsinh x= log( x+√ 1 +x2), we are
done.
112 Solution to problem 2184 Math. Mag. 96 (5) 2023, p. 567 Raymond Mortini, Peter Pflug and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We give two proofs
(one geometric/intuitive informal one and one analytic one) of the fol- lowing result: Proposition Letf(x)
=x 1+x2. Then the set of all those (a, b)∈R2for which the line y=ax+bcuts the graph G:={(x, f(x))
:x∈R}offin exactly one point is given by the ”exterior” E7of the closed Jordan curve (displayed in red
below) (44) Γ( t) =■a(t) b(t)■ =■1−t2 (1+t2)2 2t3 (1+t2)2■ ,fort∈R,and Γ( ±∞) =■0 0■ , together with
{(0,0)} ∪ {(1,0)} ∪ {(0,±1 2)} ∪ {(−1 8,±3√ 3 8)}and deleted by the half lines {0}×]1/2,∞[and{0}×]−
∞,−1/2[. Figure 6. The red curve 7This is the unbounded component of the complement of the curve.
113 Figure 7. Graph of fand one tangent Proof. We first discuss the geometry of the graph of f. •Note
that f′(x) =1−x2 (1 +x2)2. Hence the red curve Γ, excepted the point (0 ,0), is the set of (a, b) = (a(x),
b(x)) such that s7→as+bis a tangent to the graph of fat the point ( x, f(x)) (since a(x)x+b(x) =f(x) and
a(x) =f′(x)). Next, f′′(x) =2x(x2−3) (1 +x2)3. Since f′′(0) = f′′(±√ 3) = 0, (45) max f′=f(0) = 1 and min
f′=f′(±√ 3) =−1 8. Moreover f(±√ 3) =±√ 3 4∼ ±0.433013 . . .and 0 and ±√ 3 are inflection points for fand
maxf=f(1) =1 2, respectively min f=f(−1) =−1 2. If in (44) t=±√ 3, then a(t) =−1 8and b(t) =±3√ 3 8∼
±0.64951 . . .. Figure 8. fandf′ •Observe that if s7→as+bcuts the graph of fin at least two different
points, then a∈[−1/8,1]. In fact, by the mean value theorem, if xjare two intersection points, then
a=(ax1+b)−(ax2+b) x1−x2=f(x1)−f(x2) x1−x2=f′(η). Now (45) yields the assertion. Consequently, if a
/∈[−1/8,1], then the line s7→as+bis either disjoint from the graph of for cuts it in a single point. •The
only lines s7→as+bwhich do not intersect the graph of fare those that are parallel to the real axis (that
is a= 0) and for which |b|>1/2 (obviously clear by having a glimpse at
114 the figure 8 of the graph Goff). In fact, any ”oblique” line L(and any vertical line) has points in both
domains determined by Gand so the connectedness of the line implies that L∩G■=∅. Moreover, if a=
0, then b= 0·x+b=x 1+x2is equivalent to bx2−x+b= 0. So no solution exists if and only if the discriminant
1 −4b2is negative; that is if |b|>1/2. •We will see below that the only tangents meeting the graph of fat a
single point are the lines y=±1/2 and y=xandy=−1 8x±3√ 3 8(those tangents associated with the
extrema and the inflection points of f). All other tangents have another point of intersection: this is seen
”geometrically” by looking at the graph and by considering the three cases (and of course the
associated opposites) : 0 ≤x0≤1, 1< x 0≤√ 3 and x0>√ 3, and by noticing that on the interior Ijof these
three intervals the tangents are on one side of the graph {(x, f(x) :x∈Ij}, as we have no change of
curvature ( fis either convex or concave on Ij.) See figure 9. •The behavior of the lines of the form
s7→as+bwith a∈[−1/8,1] and b > b (x) orb < b (x) can be intuitively guessed by looking at the graph of
f(for a precise analytic proof, see next section). ■ Figure 9. Thea(x) and b(x)’s for x∈Ij 2.1.An analytic
proof. The intersection condition is equivalent to solving, for a■= 0, the cubic polynomial equation
ax+b=x 1 +x2⇐⇒ ax3+bx2+ (a−1)x+b= 0 and for a= 0 the quadratic equation b=x 1 +x2⇐⇒ bx2−x+b=
0. Put p(z) :=pa,b(z) :=az3+bz2+ (a−1)z+b. Then several cases occur when discussing the equation
pa,b(z) = 0, a■= 0: i) one real solution and two complex ones (which are conjugated), ii) three distinct
real solutions, iii) one double real solution and a second real solution, iv) a triple real solution. A way to
deal with this, is to use the discriminant. For■ ■■ ■a■= 0 , let D:=a4(z1−z2)2(z2−z3)2(z1−z3)2 be the
discriminant of this cubic equation. Here z1, z2, z3are the zeros. Then, D=−4a4−8a2b2−4b4+
12a3−20ab2−12a2+b2+ 4a.
115 A lenghthier calculation (a posteriori verified by Maple and wolframalpha) gives D=D(a, b) =−4"■
b2+a2+5 2a−1 8■2 −8■ a+1 8■3# . It is well-known that the cubic equation has a multiple zero if and
only if the discriminant is zero. In other words, if and only if ■ b2+a2+5 2a−1 8■2 = 8■ a+1 8■3 . Also,
D > 0 if and only if the cubic equation (with real coefficients) has three distinct real zeros, and D < 0 if
and only if there is a unique real zero. In our situation here, D < 0 if and only if −4"■ b2+a2+5 2a−1 8■2
−8■ a+1 8■3# <0, equivalently■ b2+a2+5 2a−1 8■2 >8■ a+1 8■3 . Now we have the following result:
Lemma 5. Let(a, b)∈R2. The following assertions are equivalent: (1)D(a, b) = 0 ifa■= 0or(a, b) = (0
,±1/2)ifa= 0. (2)pa,b(z) =az3+bz2+ (a−1)z+bhas a multiple zero. (3)The line L:s7→as+bis tangent to the
graph Goffat the point (x, f(x))for some 8x∈R, and a=a(x) =1−x2 (1 +x2)2andb=b(x) =2x3 (1 +x2)2.
Proof. (1)⇐⇒ (2): Discussed above for the case a■= 0. The case a= 0 follows since the discriminant of
the quadratic bz2−z+bis 1−4b2. (2) =⇒(3): Suppose that x∈Ris a multiple zero of p. Recall that p′(z) = 3
az2+2bz+(a−1). Then p(x) =p′(x) = 0 imply that ax+b=x 1+x2and 3ax2+ 2x■x 1 +x2−ax■ + (a−1) = 0 .
Thus a=1−x2 (1 +x2)2. (In case a= 0, x=±1). Consequently, s7→as+bis a tangent to the graph of
fatx(since ax+b=f(x) and a=a(x) =f′(x)). Moreover, b=x 1 +x2−1−x2 (1 +x2)2x=2x3 (1 +x2)2. (In case a=
0,b=±1/2). (3) =⇒(1): Suppose that s7→as+bis a tangent at ( x, f(x)) and that aandbhave the form given
in the assumption (3). If a=a(x)/∈ {0,1}, then xis (at least !) a double zero of pa,b, since pa,b(x) = 0
(equivalently ax+b=f(x)), and p′ a,b(x) = 0 because 31−x2 (1 +x2)2x2+ 22x3 (1 +x2)2x+1−x2 (1
+x2)2−1≡0. Moreover, if a=a(x) = 1, then xis a triple zero of pa,bandb=b(x) = 0. Hence, as (2) = ⇒
(1),D(a(x), b(x)) = 0. If a=a(x) = 0, then x=±1 and b=±1/2 . Thus (1) holds. ■ 8Later we shall see that xis
uniquely determined; so a line Lcan be tangent to Gat at most one point.
116 Conclusion: The set ( a, b)∈R2of points where pa,bhas a multiple zero is in a one to one
correspondance with those lines s7→as+bwhich are tangent to the graph of f. It coincides with {(a,
b)∈R2\■ {0} ×R■ :D(a, b) = 0} ∪n (0,−1 2),(0,1 2)o , and is the Jordan arc parametrized by Γ(t) =■a(t)
b(t)■ =■1−t2 (1+t2)2 2t3 (1+t2)2■ , t∈R. To see that Γ is injective, suppose that there exists ( a,
b)∈R2such that a=a(t) =a(t′) andb=b(t) =b(t′) for t■=t′. By Lemma 5 (and its proof), the line s7→as+bis
tangent to the graph of fat the points ( t, f(t)) and ( t′, f(t′)), and so tandt′are (at least ) double zeros
ofpa,b. This would imply that the degree of pa,bis bigger than 4. A contradiction. The two components
determined by the closure J:= Γ(R)∪{(0,0)}of this Jordan arc (which is a closed Jordan curve) coincide
with9 ˜D(a, b)−1( ]0,∞[ ) and ˜D(a, b)−1( ]− ∞,0[ ), respectively, where10 ˜D(a, b) =( D(a, b) if a■= 0
b2(1−4b2) if a= 0. The following observations now will show that the exterior of this Jordan domain is
the set where ˜D(a, b)<0. Always have in mind figure 9. But attention: this is not yet the final set the
problem is asking for. Consider a tangent at xwith 0 < a(x)≤1. Since a(x) =f′(x) we deduce that |x|<1<√
3. This implies that b(x)2+a(x)2+5 2a(x)−1 8=1 8(3−x2)3 (x2+ 1)3>0. Hence, if b > b (x), we get 8■ a(x)
+1 8■3 =■ b(x)2+a(x)2+5 2a(x)−1 8■2 <■ b2+a(x)2+5 2a(x)−1 8■2 , and so D(a(x), b)<0. This implies
that there is a unique real zero of pand so the line s7→a(x)s+bcuts the graph of fat a single point. Next,
if b= 0 and if a→ −∞ , then D(a, b)→ −∞ . So again ˜D < 0 in that part of the exterior of Jthat is contained
in the left-hand plane. Finally, if a= 0, the discriminant 1 −4b2ofbz2−z+bis negative if and only if
p0,bhas no real zeros; so no intersection points of ax+bexist whenever a= 0 and |b|>1/2, but two if
0<|b|<1/2 and one if b= 0. Consequently, the exterior of the Jordan curve is the set where ˜D < 0. To
achieve the solution to the problem, a last case has to be investigated: for which ( a, b) the polynomial
pa,bhas triple zero (as this yields tangents which cut the graph Goffat a single point). So let pa,b(x) =p′
a,b(x) =p′′ a,b(x) = 0. By Lemma 5, s7→as+bis tangent to the graph G off. Hence a=a(x) =1−x2
(1+x2)2andb=b(x) =2x (1+x2)2. Now p′′ a,b(z) = 6 az+ 2b. Hence x=−b 3a=2x (1 +x2)2 31−x2 (1
+x2)2=−2 3x3 1−x2. 9Take e.g. two points in the exterior complemented component of J, denoted by Ω.
Join those with an arc inside Ω. Then ˜Dmust have the same sign at both points; otherwise this arc
would meet the set where ˜Dis zero. As this set coincides with the boundary of Ω, that is the Jordan
curve J, we get a contradiction. 10In order to have continuity of ˜D, we need to add the factor b2.
117 Consequently, either x= 0 or x=±√ 3. This yields the values ( a, b) = (1 ,0) and ( a, b) = (−1 8,±3√ 3
8). •We are now able to answer the question, which lines s7→as+bintersect the graph Gof f(x) =x/(1
+x2) in a single point: i) All points ( a, b)∈R2for which a■= 0 and D(a, b)<0. ii) The 6 points ( a, b)∈
{(0,0),(1,0),(0,±1 2),(−1 8,±3√ 3 8)}, which induce via the map s7→as+btangents to Gwhenever ( a,
b)■= (0,0).
118 Quicky 1140 Math. Mag. 97 (2024) by Raymond Mortini and Rudolf Rupp - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Submitted statement: (a) Let m, n∈N={0,1,2, . . .}.
Determine the value of B(n, m) :=nX k=0(−1)k■m+k k■ ■m+n+ 1 n−k■ . (b) Let z, w∈C\ {−1,−2,−3. . .}.
Suppose that z−w /∈ {− 1−2,−3. . .}. Using that for these parameters■z w■ :=Γ(z+1) Γ(w+1)Γ( z−w+1)is
well defined, show that for a, b∈Cwith Re a >0, Reb >1, and b /∈Z, the series S(a, b) :=∞X
k=0(−1)k■a+k−1 k■ ■a+b−1 b−k−1■ . converges absolutely and that S(a, b) = 1. Solution (a) Note that
■m+k k■ ■m+n+ 1 n−k■ =(m+k)! m!k!(m+n+ 1)! (m+k+ 1)!( n−k)! =(m+n+ 1)! m!1 k!(m+k+ 1)(
n−k)!=(m+n+ 1)! m!n!■n k■1 m+k+ 1. Hence nX k=0(−1)k■m+k k■ ■m+n+ 1 n−k■ =(m+n+ 1)! m!n!nX
k=0(−1)k■n k■1 m+k+ 1. Put f(x) :=nX k=0(−1)k■n k■1 m+k+ 1xm+k+1. Then f′(x) =nX k=0(−1)k■n k■
xm+k=xm(1−x)n. Consequently, as´1 0f′(x)dx=f(1)−f(0) and f(0) = 0, B(n, m) =(m+n+ 1)! m!n!ˆ1
0xm(1−x)ndx. The value of the integral is given by Euler’s βfunction β(m+ 1, n+ 1) =Γ(m+ 1) Γ( n+ 1)
Γ(m+n+ 2)=m!n! (m+n+ 1)!.
119 Hence B(n, m) =(m+n+ 1)! m!n!m!n! (m+n+ 1)!= 1. (b) First we note that under the conditions on
aandb, the complex binomial coefficients are well defined. Since Γ( z+ 1) = zΓ(z) for z∈C\(−N), and
since the Γ-function has no zeros, we have ■a+k−1 k■ ■a+b−1 b−k−1■ =Γ(a+k) Γ(a) Γ(k+ 1)Γ(a+b)
Γ(a+k+ 1)Γ( b−k) =Γ(a+b) Γ(a)Γ(b)Γ(b) (a+k)Γ(k+ 1)Γ( b−k) =Γ(a+b) Γ(a)Γ(b)■b−1 k■1 a+k. Hence S(a,
b) =∞X k=0(−1)k■a+k−1 k■ ■a+b−1 b−k−1■ =Γ(a+b) Γ(a)Γ(b)∞X k=0(−1)k■b−1 k■1 a+k. It is known
that the binomial series∞X k=0■b−1 k■ converges absolutely for Re b > 1 (see [27, p. 140]). Hence
S(a, b) converges. Now consider for c∈Cand 0 < x≤1 the functions xc:= exp( clogx), and f(x) := 
xa∞X k=0(−1)k■b−1 k■1 a+kxkif 0< x≤1 0 if x= 0, which is continuous11on [0,1]. Using for 0 < x < 1
the Newton-Abel formula for the binomial series with complex powers (see [27, p. 158]), we obtain f′(x)
=∞X k=0(−1)k■b−1 k■ xa+k−1=xa−1∞X k=0■b−1 k■ (−x)k=xa−1(1−x)b−1. Consequently, as´1
0f′(x)dx=f(1)−f(0) and f(0) = 0, S(a, b) =Γ(a+b) Γ(a)Γ(b)ˆ1 0xa−1(1−x)b−1dx. This integral is the
β-function. Note that Re a >0 and Re b >0. Hence this integral is well defined and β(a, b) =Γ(a)Γ(b)
Γ(a+b)(see [28, p. 67 ff]. Consequently, S(a, b) = 1. For this proposal, we were motivated by Problem
4862 Crux Math. 49 (7) 2023, 375. We hope that this sum has not been considered earlier. 11Note that
Re a > 0 and so 0a:= 0 is the correct value if one wants continuity: |xa| ≤ exp(Re alogx)→exp(−∞) = 0
as x→0+. Also, as usual in the realm of power series, 00:= 1.
120 Solution to problem 2185 Math. Mag. 96 (5) 2023, p. 567 Raymond Mortini and Rudolf Rupp - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that the value of the
integral In:=´1 −1Pn(x)dxis
■ (−1)nin n+ 2■ 1 + (−1)n■ 2n+2 2cos■ nπ 4■ . Another representation is In=εn2n+4 2 n+ 2, where
εn=  1 if n≡0 mod 8 0 if n≡1 mod 8 0 if n≡2 mod 8 0 if n≡3 mod 8 −1 if
n≡4 mod 8 0 if n≡5 mod 8 0 if n≡6 mod 8 0 if n≡7 mod 8. A very strange result! In fact, dn dxn■1 1 +ix■
=in(−1)nn! (1 +ix)n+1anddn dxn■1 1−ix■ =inn! (1−ix)n+1. Hence dn dxn■1 1 +x2■ =1 2dn dxn■1 1
+ix+1 1−ix■ =in 2n!(1 +ix)n+1+ (−1)n(1−ix)n+1 (1 +x2)n+1. From this we get that Pnis a polynomial of
degree nwith n+ 1 as leading coefficient. We are now ready to calculate the integral:
121 In=ˆ1 −1Pn(x)dx=(−1)n 2inˆ1 −1■ (1 +ix)n+1+ (−1)n(1−ix)n+1■ dx =(−1)n 2in■(1 +i)n+2−(1−i)n+2
i(n+ 2)+ (−1)n(1−i)n+2−(1 +i)n+2 (−i)(n+ 2)■ =(−1)nin−1 2(n+ 2)(1 + (−1)n)■ (1 +i)n+2−(1−i)n+2■ .
Since (1 +i)n+2−(1−i)n+2=√ 2n+1 ■1 +i√ 2■n+2 −■1−i√ 2■n+2! = 2n+1 2■ e(n+2)iπ/4−e−(n+2)iπ/4■ =
2 i2n+1 2sin■ (n+ 2)π 4■ , we conclude that In=(−1)nin n+ 2■ 1 + (−1)n■ 2n+2 2sin■ (n+ 2)π 4■
=(−1)nin n+ 2■ 1 + (−1)n■ 2n+2 2cos■ nπ 4■ .
122 Solution to problem 2181 Math. Mag. 96 (5) 2023, p. 566 Raymond Mortini, Peter Pflug and Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - a) The double series
S(x) :=∞X k=0∞X m=01 k!1 [2(m+ 1)]!1 2m+ 2 + kx2m+2+k converges since for every j∈Nthe partial
sums can be estimated as follows: jX n=0jX m=01 k!1 [2(m+ 1)]!1 2m+ 2 + kx2m+2+k≤jX n=0jX m=01
k!1 [2(m+ 1)]!xkx2m+2 = jX n=01 k!xk! jX m=01 [2(m+ 1)]!x2m+2! Hence the series Pconverges
absolutely (and so does any re-arrangement) locally uniformly to some finite value P(x). b) By the same
reason the formal derivated series H(x) :=∞X k=0∞X m=0(−1)k k!(−1)m+1 [2(m+ 1)]!x2m+1+k
converges absolutely and locally uniformly for x≥0. Hence P′=H. Thus H(x) = ∞X k=0(−1)k k!xk! ∞X
m=0(−1)m+1 [2(m+ 1)]!x2m+1! =e−xcosx−1 x. Consequently Pis a primitive of e−xcosx−1 xwhich
vanishes at 0. Hence J:= lim x→∞P(x) =ˆ∞ 0e−xcosx−1 xdx. Next we show that J=−1 2log 2 by
interpreting this integral as the Laplace transform L(q)(s) of the function q(x) = (cos x−1)/xevaluated at
s= 1. By a well-known formula, if L(F(t))(s) = f(s), then L(q)(s) =L(F(t) t)(s) =ˆ∞ sf(u)du, where f(s) =ˆ∞
0e−st(cost−1)dt=1 s3+s. Hence L(q)(s) =−1 2log(1 + s−2) and so J=L(q)(1) = −1 2log 2. Remark A
formal (but probably unjustifiable) way to calculate the value of Jwould be the following: J:=ˆ∞
0e−xcosx−1 xdx=ˆ∞ 0e−x∞X n=1(−1)n (2n)!x2n−1dx
123 !=∞X n=1(−1)n (2n)!ˆ∞ 0x2n−1e−xdx Since for m∈Nandk∈N\ {0}, ˆ∞ 0xme−kxdx=m!/km+1,
wewould obtain J=∞X n=1(−1)n (2n)!(2n−1)! =∞X n=1(−1)n 2n=−1 2log 2. Note also that the softwares
Wolframalpha/mathematica give the exact value of the integral, too. The problem itself come to our
mind when solving Problem number 12338 in Amer. Math. Soc..
124 Solution to problem 2176 Math. Mag. 96 (3) 2023, p. 468 Raymond Mortini and Rudolf Rupp - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetI:=ˆ1 0log(1 + x+x2) 1 +x2dx. We
make the substitution x= tan u,dx/du = 1 + tan2u= 1 cos2u,x= 0→u= 0 and x= 1→u=π/4. Then I =ˆπ/4
0log(tan2u+ tan u+ 1) 1 + tan2u(1 + tan2u)du=ˆπ/4 0log■1 + cos usinu cos2u■ du =ˆπ/4 0log■ 1 +1
2sin(2u)■ du−2ˆπ/4 0log(cos u)du = Lem.61 2ˆπ/2 0log■ 1 +1 2sin(v)■ dv−2■C 2−π 4log 2■ =1 2ˆπ/2
0∞X k=1(−1)k+1 k1 2k(sinx)kdx+π 2log 2−C Lem.4= unif.abs.conv.´P=P´1 2∞X n=01 2n+ 11 22n+14n
(2n+ 1)■2n n■−1 2∞X n=11 2n1 22n■2n n■ 4nπ 2+π 2log 2−C =1 4∞X n=01 (2n+ 1)2■2n n■−π 8∞X
n=11 n■2n n■ 16−n+π 2log 2−C Lemm .8= Lemm .91 4■ C−1 8πlog(2 +√ 3)■8 3−π 82 log 1−p
1−4(1/16) 2(1/16)! +π 2log 2−C = −1 3C+π 6log(2 +√ 3). 2.2.Appendix. Here we present for
completeness the proofs of all those known results used above to derive the value of the integral.
Lemma 6. [29, formula (8)] C= 2ˆπ/4 0log(2 cos x)dx. Proof. Since on ]0 ,1] the integrable function
|logx|dominates the modulus of the partial sumsPN n=0(−1)nx2nlogx, we have ˆ1 0logx 1 +x2dx=∞X
n=0ˆ1 0(−1)nx2nlogx dx =∞X n=0(−1)n−1 (n+ 1)2=−C. Hence, with x= tan uanddx= (1 + tan2u)du, (46)
C=−ˆπ/4 0log(tan u)du=ˆπ/4 0log(cos u)du−ˆπ/4 0log(sin u)du=:Lc−Ls.
125 Now, using the standard result that´π/2 0log sin x dx=−π 2log 2, we obtain (47) Lc+Ls+π 4log 4
=ˆπ/4 0log(2 sin(2 u))du=ˆπ/2 0log(2 sin x)dx= 0. Adding (46) and (47), yields C−π 4log 4 = 2 Lc. In other
words, 2ˆπ/4 0log(2 cos x)dx=C. ■ Lemma 7. LetIn:=´π/2 0(sinx)ndx. Then I0=π/2,I1= 1and for n∈N∗,
(1)I2n=1 2·3 4·5 6···2n−1 2nπ 2=(2n)! 4n(n!)2·π 2=(2n n) 4n·π 2. (2)I2n+1=2 3·4 5·6 7···2n 2n+1=4n(n!)2
(2n+1)!=4n (2n+1)(2n n). Proof. (1)I2n=2n−1 2nI2n−2forn∈N∗andI0=π 2, because
2nI2n−(2n−1)I2n−2=ˆπ/2 0(sinx)2n−2■ 2nsin2x−(2n−1)■ dx =−ˆπ/2 0(sinx)2n−2■ (2n−1) cos2x−sin2x■
dx =−■ (sinx)2n−1cosx■π/2 0= 0. (2)I2n+1=2n 2n+1I2n−1forn∈N∗andI1= 1, because (2n+
1)I2n+1−2nI2n−1=ˆπ/2 0(sinx)2n−1■ (2n+ 1) sin2x−2n■ dx =−ˆπ/2 0(sinx)2n−1■ 2ncos2x−sin2x■ dx
=−■ (sinx)2ncosx■π/2 0= 0. ■
126 Lemma 8. [30] (1)∞X n=0■2n n■ xn=1√1−4xfor|x|<1/4. (2)∞X n=11 n■2n n■ xn= 2 log■1−√1−4x
2x■ for|x|<1/4. Proof. (1) Note that ■−1/2 n■ =(−1 2)·(−3 2)· ···(−1 2−n+ 1) n!= (−1)n1·3·5···(2n−1) 2nn!
= (−1)n(2n)! (2nn!)2= (−1)n■2n n■ 4n. Hence, by Newton’s binomial theorem ∞X n=0■2n n■ xn=∞X
n=0(−1)n■−1/2 n■ (4x)n= (1−4x)−1/2. (2) Let f(x) :=∞X n=11 n■2n n■ xn. Then, for 0 < x < 1/4, f′(x) =1
x∞X n=1■2n n■ xn=1 x√1−4x−1 x. To calculate the primitive, we make the transformation u:=√1−4x, or
equivalently x=1−u2 4. Sinceˆ4 1−u21 u■ −u 2■ du= log■1−u 1 +u■ , we deduce that f(x) =
log■1−√1−4x 1 +√1−4x■ −logx= log■(1−√1−4x)2 4x■ −logx = log■(1−√1−4x)2 (2x)2■ = 2
log■1−√1−4x 2x■ . The following formula is due to Ramanujan. Lemma 9. [29, formula (62)] C=1
8πlog(2 +√ 3) +3 8∞X n=01 (2n+ 1)2■2n n■, equivalently ∞X n=01 (2n+ 1)2■2n n■=8 3C+1 3πlog(2−√
3). Proof. First we note that (48) artanh z=1 2log1 +z 1−z=∞X n=01 2n+ 1z2n+1,|z|<1. Hence, by
Lemma 7 J:=ˆπ/2 0log■1 +1 2sinx 1−1 2sinx■ dx=ˆπ/2 02∞X n=01 22n+1(sinx)2n+1 2n+
1dxunif.abs.conv.=´P=P´∞X n=01 22n4n (2n+ 1)2■2n n■ =∞X n=01 (2n+ 1)2■2n n■.
127 To calculate this integral we combine calculations done in [31] and [33], where it is also shown that
J=ˆ1 0artanhp u(1−u)p u(1−u)du (just put u= sin2x). See [34], too. Let us introduce the parametric
integral I(a) :=ˆπ/2 0log■1 + sin asinx 1−sinasinx■ dx. Nowd da´ =´d da(as all functions considered
here are continuously differentiable). Hence I′(a) =ˆπ/2 0■cosasinx 1 + sin asinx+cosasinx 1−sinasinx■
dx=ˆπ/2 02 cosasinx 1−sin2asin2xdx =2 cosa sin2aˆπ/2 0sinx 1 sin2a+ cos2x−1dx=2 cosa sin2aˆπ/2
0sinx cot2a+ cos2xdx =−2 sinaarctan(cos xtana) π/2 0=2a sina. Thus, using partial integration, and the
fact that tan( x/2) =sinx 1+cos x, J=I(π/6) = I(π/6)−I(0) =ˆπ/6 0I′(a)da=ˆπ/6 02a sinada = 2ˆπ/6 0x■ log■
tanx 2■■′ dx= 2xlog■ tanx 2■ π/6 0−2ˆπ/6 0log■ tanx 2■ dx =x 2=tπ 3log(2−√ 3)−4ˆπ/12 0log(tan t)dt.
Now we follow [33]12. Recall that on ]0 , π[ the Fourier series for −log tan( t/2) is 2∞X n=01 2n+ 1cos(2
n+ 1)t. Since the Fourier series converges in the L2-norm, hence L1-norm on ]0 , π[, we haveP´ =´P.
Hence −ˆπ/12 0log(tan x)dx = x=t/2ˆπ/6 0∞X n=01 2n+ 1cos(2 n+ 1)t dt=∞X n=01 2n+ 1ˆπ/6 0cos(2 n+
1)t dt =∞X n=01 (2n+ 1)2sin■π 6(2n+ 1)■ | {z } :=S!=2 3∞X n=01 (2n+ 1)2(−1)n=2 3C. To show the
penultimate identity, we follow [32]. To this end we first note that sin■π 6(2k+ 1)■ = 
1 2ifk≡0 mod 6 1 if k≡1 mod 6 1 2ifk≡2 mod 6 −1 2ifk≡3 mod 6 −1 if k≡4 mod 6 −1 2ifk≡5 mod
6. 12We thank Roberto Tauraso for providing us this link.
128 Hence S=1 2∞X n=01 (12n+ 1)2+∞X n=01 (12n+ 3)2+1 2∞X n=01 (12n+ 5)2 −1 2∞X n=01 (12n+
7)2−∞X n=01 (12n+ 9)2−1 2∞X n=01 (12n+ 11)2 =1 9∞X n=0■1 (4n+ 1)2−1 (4n+ 3)2■ +1 2■1 12+1
52−1 72−1 112+···■ =1 9C+1 2■1 12−1 32+1 52−1 72+1 92−1 112+···■ +1 2■1 32−1 92+1 152− ···■
=1 9C+1 2C+1 2·1 32■1 12−1 32+1 52− ···■ =1 9C+1 2C+1 18C=12 18C=2 3C. We conclude that J=π
3log(2−√ 3) +8 3C=8 3C−π 3log(2 +√ 3). ■ Here is a second proof to calculate the value of´π/12
0log(tan t)dt. Proof. We follow [32]. Consider for a >0 the integral Q(a) =−ˆπ/12 0artanh■2 cos 2 x
a+a−1■ dx=−ˆπ/12 0artanh■2acos 2x a2+ 1■ dx. (Note that a+ 1/a≥2, so this is well defined). Againd
da´ =´d da. Using that (artanh z)′= 1 1−z2, and that d da■1 a+a−1■ =1−a2 (a2+ 1)2, Q′(a) = −ˆπ/12
01−a2 (1+a2)22 cos 2 x 1−4 cos2(2x) (a+a−1)2dx=ˆπ/12 0(a2−1) 2 cos 2 x (a2+ 1)2−4a2cos22xdx =−1
2aarctan■2asin 2x 1−a2■ π/12 0=arctana a2−1 2a. Hence, by using that Q(0) = 0, and that arctan u+
arctan v= arctan(u+v 1−uv), log(tan x) =−1 2log■1 + cos 2 x 1−cos 2x■ =−artanh(cos 2 x).
Consequently, as C=−´1 0arctan x xdx(use the power series for arctan x) ˆπ/12 0log(tan x)dx = Q(1) =ˆ1
0Q′(a)da=ˆ1 0arctana a2−1 2ada = −ˆ1 0■arctan a 2a+arctan a3 2a■ da = = a3→b−■1 2+1 6■ˆ1
0arctan a ada=−2 3C. We conclude that J=π 3log(2−√ 3) +8 3C=8 3C−π 3log(2 +√ 3). ■
129 2.3.Remarks. (1) The integral L:=´∞ 0log(1+ x+x2) 1+x2dxis mentioned on wikipedia [36] (with- out
a source) under the form C=3 4L−π 4arcosh 2 . We notice that L= 2I+ 2C. In fact, I = u=1/xˆ∞
1log(u2+u+ 1)−log(u2) 1 +u2du =ˆ∞ 1log(u2+u+ 1) u2+ 1−2C. Hence 2I=I+I=ˆ∞ 0log(u2+u+ 1) u2+
1−2C. Using the assertion of the problem dealt with here, L= 2■π 6log(√ 3 + 2) −C 3■ + 2C, and so
C=3 4L−π 4log(√ 3 + 2) . Now just note that arcosh 2 = log(√ 3 + 2). ■ This integral Lis also calculated
in [35].
130 Solution to problem 2171 Math. Mag. 96 (3) 2023, p. 359 Raymond Mortini and Rudolf Rupp - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - For logical reasons, we think that the
exponent of ( −1) in the two statements above has to ben+ 1, since one starts with −(−1)n, where n= 0.
Since the double series is absolutely convergent, we may arrange as we wish. (a) Let C(x) :=∞X n=0■
cosx−1 +x2 2!−x4 4!+···+ (−1)n+1x2n (2n)!■ and let Tnbe the 2 n-th Taylor polynomial for cos x, which
is given by Tn(x) =nX j=0(−1)jx2j (2j)!. Then C(x) =∞X n=0(cosx−Tn(x)). Hence C(x) =∞X n=0∞X
k=n+1(−1)kx2k (2k)!=∞X k=1k−1X n=0(−1)kx2k (2k)! =∞X k=1k(−1)k+1x2k (2k)!=1 2∞X k=12k(−1)kx2k
(2k)! =x 2∞X k=1(−1)x2k−1 (2k−1)! =−1 2xsinx. For (b) we give two solutions.
131 Similarily to (a) , let S(x) :=∞X n=0■ sinx−x+x3 3!−x5 5!+···+ (−1)n+1x2n+1 (2n+ 1)!■ . Then S(x)
=∞X n=0∞X k=n+1(−1)kx2k+1 (2k+ 1)!=∞X k=1k−1X n=0(−1)kx2k+1 (2k+ 1)! =∞X k=1k(−1)kx2k+1 (2k+
1)!=1 2∞X k=12k(−1)kx2k+1 (2k+ 1)! =1 2∞X k=1(2k+ 1)(−1)kx2k+1 (2k+ 1)!−1 2∞X k=1(−1)kx2k+1
(2k+ 1)! =x 2∞X k=1(−1)kx2k (2k)!−1 2(sinx−x) =x 2(cosx−1)−1 2(sinx−x) =xcosx−sinx 2 The second
method is to integrate termwise and then to interchange the sum with the integral (uniform convergence
on compacta). S(x) =∞X n=0ˆx 0■ cost−1 +t2 2!−t4 4!+···+ (−1)n+1t2n (2n)!■ dt =ˆx 0∞X n=0■ cost−1
+t2 2!−t4 4!+···+ (−1)n+1t2n (2n)!■ dt = (a)−1 2ˆx 0(tsint)dt =xcosx−sinx 2.
132 Solution to problem 2167 Math. Mag. 96 (2) 2023, p. 190 Raymond Mortini and Rudolf Rupp - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let L:=en/2nY j=2ej2nY j=2■ 1−1
j2■j4 . Then, by using that for |x|<1,−log(1−x) =∞X k=11 kxk, logL =n 2+nX j=2j2+nX j=2j4log■ 1−1 j2■
=n 2+nX j=2j2−nX j=2∞X k=1j4 k1 j2k =n 2+nX j=2j2−nX j=2j2−1 2nX j=21−∞X k=31 knX j=21 (j2)k−2
m:=k−2−→ n→∞1 2−∞X m=1ζ(2m)−1 m+ 2. So we need to show that (49)■ ■ ∞X m=1ζ(2m)−1 m+ 2=7
4−logπ−3 π2ζ(3)∼0.239888629 ···, from which we conclude that L=πexp■3 π2ζ(3)−5 4■
∼1.2970745345 ··· To achieve our goal we use the partial fraction decomposition of πzcot(πz) = 1 +∞X
n=12z2 z2−n2, z∈C\Z, a formula which implies (see [37, p. 182]) that ∞X n=1(ζ(2n)−1)x2n=1 2■
1−πxcot(πx)■ −x2 1−x2=−∞X n=2x2 x2−n2. Since z= 0 and z=±1 are removable singularities for1 2■
1−πzcot(πz)■ −z2 1−z2, the holomorphy in |z|<2 implies that we have uniform convergence ofP∞
n=1(ζ(2n)−1)x2non [0,1]. Also, for x∈]0,1[, ∞X n=1(ζ(2n)−1)xn+1=x 2−π 2x3/2cot(π√x)−x2 1−x.
133 A primitive on ]0 ,1[ is then given by ∞X n=1ζ(2n)−1 n+ 2xn+2=x2 4−ˆ■π 2x3/2cot(π√x) +x2 1−x■
dx. Using again uniform convergence on [0 ,1], the substitution s=√xand integration between 0 and 1
yields ∞X n=1ζ(2n)−1 n+ 2=1 4−ˆ1 0■π 2s3cot(πs) +s4 1−s2■ 2sds =1 4−ˆ1 0■ πs4cot(πs) +2s5 1−s2■
ds Let ■ ■ I:=ˆ1 0■ π t4cot(πt) +2t5 1−t2■ dt. We claim that (50) I=3 π2ζ(3)−3 2+ log π∼0.0101113705
··· To determine the value of I, we first calculate a primitive of f(x) :=π x4cot(πx) on ]0,1[. This is done by
using partial integration and the 1-periodic Fourier series ∞X k=1cos(2 kπx) k=−log(2 sin( πx)) =−log
2−log(sin( πx)), where the convergence is considered in the L2-norm on ]0 ,1[, which also guarantees
that´P=P´ below. ˆ f(x)dx=x4log(sin( πx))−4ˆ x3log(sin( πx))dx =x4log(sin( πx)) + 4ˆ∞X k=1x3cos(2 kπx)
kdx+ 4ˆ x3log 2 dx =x4log(sin( πx)) + 4∞X k=1ˆ x3cos(2 kπx) kdx+ 4ˆ x3log 2 dx Before evaluating at the
boundary points, we need to add 2x5 1−x2=−2x3−2x−1 1 +x+1 1−x, since the integral´1 0f(x)dxis
divergent (at 1). Defining the symbol [ h(x)]1 0below as [h(x)]1 0:= lim x→1−h(x)−lim x→0h(x), we
obtain I=ˆ1 0■ π x4cot(πx) +2x5 1−x2■ dx=" x4log(sin( πx))−log(1−x)−x4 2−x2−log(1 + x)#1 0 +4∞X
k=1ˆ1 0x3cos(2 kπx) kdx+ log 2 . Note that x4log(sin( πx))−log(1−x) =x4logsin(πx) 1−x+ (x4−1) log(1
−x)→logπasx→1.
134 Also, three times partial integration yields ˆ1 0x3cos(2 kπx) kdx=3 4k3π2. Hence I= log π−3 2−log 2
+ 4∞X k=13 4k3π2+ log 2 = log π−3 2+3 π2ζ(3), yielding (50). We conclude that (49) is satisfied, that is
∞X n=1ζ(2n)−1 n+ 2=1 4−I=7 4−logπ−3 π2ζ(3)∼0.23988862 ··· Remarks (1) The value for Iis also given
directly by Maple (2) Using Wolframalpha’s representation below of a primitive of π t4cot(πt) +2t5
1−t2and evaluating at the boundary points, we also obtain the value of I. Just note that ζ(2) = π2/6
andζ(4) = π4/90: I= 2 iζ(2) π+ 3ζ(3) π2−3iζ(4) π3−3ζ(5) 2π4+iπ 5−1 2−1 + lim s→1■
s4log(1−e−2πis)−log(1−s2)■ +3ζ(5) 2π4−lim s→0■ s4log(1−e−2πis)−log(1−s2)■ = 3ζ(3) π2+i■1 3−1
30■ π+iπ 5−3 2+ (−iπ 2+ log π)−(0) = 3ζ(3) π2−3 2+ log π. (3) Generalizations of formula 49 are given in
[6].
135 Solution to problem 2147 Math. Mag. 95 (2) 2022, p. 242 Raymond Mortini and Rudolf Rupp - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that, in accordance with
WolframAlpha, P:=∞Y n=2n4+ 4 n4−1=2 sinh π 5π. Due to sin( iz) =isinhz, we have P(z) :=∞Y n=1■
1−z4 n4■ =∞Y n=1■ 1−z2 n2■∞Y n=1■ 1 +z2 n2■ =sinπzsinhπz π2z2 and so Q(z) :=∞Y n=2■ 1−z4
n4■ =P(z) 1−z4 Note that P=∞Y n=21 +4 n4 1−1 n4. We put either z= 1 or z= 1 + i. Note that (1 +
i)4=−4 and that lim z→1Q(z) = lim z→1sinπz 1−zlim z→11 (1 +z)(1 + z2)sinhπ π2=1 4sinhπ π. Hence
P=1 1−(1 +i)4sin(π(1 +i)) sinh( π(1 +i)) π2(1 +i)2 1 4sinhπ π=2 5sinhπ π
136 Solution to problem 2141 Math. Mag. 95 (2) 2022, p. 157 Raymond Mortini and Rudolf Rupp - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The value of the integral Iis
2πcos(φ/2). First we note that ( u+eiφ)(u+e−iφ) = 1 + 2 ucosφ+u2. Case 1 cosφ■= 0 (or equivantly φ /∈
{π+ 2kπ:k∈Z}). Let log z= log |z|+iargzbe the main branch of the complex logarithm (that is −π < argz < π
). Put H:=C\]− ∞,0]. Note that for z∈S:=C\ {it:|t|>1}we have (51) arctan z=:1 2ilog1 +iz 1−iz is a primitive
of 1 /(1 +z2). Now for x∈R, 1/x2+e±iφ∈Hand so f(x) := log(1 /x2+e±iφ) is well defined. A primitive is
given by F(x) =xlog(1/x2+e±iφ)−ˆ xd dxlog(1/x2+e±iφ) =xlog(1/x2+e±iφ) +ˆ2 1 +x2e±iφdx. Since
φ■=±π,z:=xe±iφ/2∈Sand so, by using (51) and the fact that1+iz 1−izmaps the right-half plane onto the
upper-half-plane, ˆ2 1 +x2eiφdx= 2e−iφ/2arctan( eiφ/2x)−→ x→∞2e−iφ/2π/2 Now arg z+ arg z= 0 and so
log z+ log z= log|z|2,z∈H. Hence, with z= 1/x2+eiφ∈H, xlog(1/x2+eiφ) +xlog(1/x2+e−iφ) =xlog(1 + 2
x−2cosφ+x−4)x→∞−→ x→00 Hence lim x→∞I(x) = 0 + π(e−iφ/2+eiφ/2) = 2 πcosφ/2. Case 2 cosφ= 0. In
other words I=´∞ 0log((1 −1/x2)2)dx, which is improper at 0 and 1. In this case I= 0. In fact, for x >0 and
x■= 1, h1(x) := xlog■1 x2−1■2 + log■1 +x 1−x■2 = ( x−1) log( x−1)2+ 2(x+ 1) log(1 + x)−4xlogx is a
primitive of log((1 −1/x2)2). Hence ˆ1 0log((1 −1/x2)2)dx= log 16 andˆ∞ 1log((1 −1/x2)2)dx= log(1 /16).
137 A related method (for cos φ■= 0) is to apply partial integration directly to B(x) := log(1 +
2x−2cosφ+x−4) and which givesˆ B(x)dx=xB(x)−ˆ xB′(x)dx with xB′(x) = 41 + cos φ x2 x4+ 2 cos φ x2+
1=: 4R(x). This rational function writes as R(x) =A x2+eiφ+B x2+e−iφ with A=eiφcosφ−1 2isinφ, B=A. By
using (51), we obtain ˆ∞ 0dx x2+b2= lim x→∞1 barctan( x/b) =( π 2bif Re b >0 −π 2bif Re b <0. Since
lim x→0 x→∞xB(x) = 0, we deduce (with b:=eiφ/2) that´∞ 0B(x)dx= 2πcos(φ/2).
138 Solution to problem 2128 Math. Mag. 94 (2021), p. 308 Raymond Mortini - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We first show that lim m→∞(1−bm)1 mam= 0 and lim
m→∞(1−am)1 mam= 1. In fact, taking logartithms, this is equivalent to show that L:= lim
m→∞log(1−bm) mam=−∞ andR:= lim m→∞log(1−am) mam= 0. SinceP mmamconverges for
|a|<1,mam→0. Hence we have an indeterminate form 0 /0 and may use l’Hospital’s rule. Using that for
x >1, we have lim xm/m=∞, we obtain L= lim m→∞1 1−bm(−bm) logb am(1 +mloga)=−logblim m→∞■b
a■m1 1 +mloga =−logb logalim m→∞■b a■m1 m=−∞. Moreover R= lim m→∞1 1−am(−am) loga am(1
+mloga)=−logalim m→∞1 mloga= 0. Next we use that for sm:=1 mamandxm:= log(1 −bm)→0 the
inequalities (sm−1)xm<■sm■xm≤smxm imply that the limits RandLdo not change if we replace
smby■sm■. Consequently, lim m→∞(1−bm)■1 mam■= 0 and lim m→∞(1−am)■1 mam■= 1 Hence,
for each ε∈]0,1[ we obtain m, n∈Nsuch that (1−bm)n< εand (1 −am)n>1−ε.
139 Solution to problem 2118 Math. Mag. 94 (2021), p. 150 Raymond Mortini - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The series converges. This is an immediate
consequence to Abel’s theorem telling us that if (an) is a sequence of positive numbers with an■0, then
the trigonometric series S(t) :=P∞ n=0aneintconverges for all t■∈ {2kπ:k∈Z}(see i.e. Appendix 4 in my
encyclopedic monograph: R. Mortini, R. Rupp, Extension Problems and Stable Ranks, A Space
Odyssey, Birkh¨ auser 2021, ca 2150 pages): Just take an=e−■logn■,t= 1, and the imaginary part of
S(t). The proof is based on the Abel-Dirichlet rule, telling us that with bn=eint, and |b0+b1+···+bm|=|1
+eit+···+eimt|= = 1−e(m+1)it 1−eit ifeit■= 1. we obtain for t /∈2πZthat (52) |b0+b1+···+bm| ≤2 |1−eit|=:M.
Hence the seriesP∞ n=0anbnis convergent.
140 Solution to problem 2117 in Math. Mag. 94 (2021), p. 150 Raymond Mortini, Rudolf Rupp and Amol
Sasane - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - There are only the
three solutions ( n, m)∈ {(1,1),(1,2),(2,4)}. It is easy to check that these are solutions. Now suppose that
n≥m≥2. Then ( n, m) cannot be a solution since (m+ 1)n≥(m+ 1)m> mm> m!, so ( m+ 1)n> m! + 1. Now,
if 2 = n < m , then (m+ 1)2=m! + 1⇐⇒ m+ 2 = ( m−1)! which is obviously only satisfied for m= 4. Next
let 2 < n < m . Then we see that if ( n, m) is a solution to ( m+ 1)n=m! + 1, then m must be even.
(Actually, by Wilson’s theorem, m+ 1 divides m! + 1 if and only m+ 1 is prime; but we do not need this
result). In particular, m≥4. Note that the equation ( m+1)n−1 =m! under discussion is equivalent to
(53)n−1X k=0(m+ 1)k= (m−1)!. 1■m= 4. Then, due to (53), 6 = 3! = 1 + 5 + ···implying that n= 2. A
contradiction to the assumption 2 < n < m .. 2■m≥6. Then 2 < m/ 2< m−1. Hence the integer m/2
divides ( m−1)!. Since m/2>2, additionally the number 2 divides ( m−1)!. Thus m= 2·(m/2) divides (
m−1)!. Now, (53) yields n≡0, mod m. That is, mdivides nand so m≤n. This is again a contradiction to the
assumption 2 < n < m .
141 Solution to problem 2116 Math. Mag. 94 (2021), 150 Raymond Mortini, Rudolf Rupp - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that
142
143 Solution to Quicky 1075 in Math. Mag. 90 (2017), 384 Raymond Mortini - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Figure 10. Intersecting curves Yes. We have to show that
for every d∈R, there is x0∈Rsuch that f(x0) =f(x0−d). 1) Non-elementary geometric approach. Putg(x)
:=f(x−d) and choose a, b∈Rsuch that m=f(a),M=f(b) and m < f (x)< M forx∈]a, b[. We may assume that
a < b . Of course, m≤g(x)≤M. Let a′< aandb > b′. Then x7→(x, g(x)), a′≤x≤b′is a curve in the rectangle
R:= [a′, b′]×[m, M ] starting at the left of the graph F:={(x, f(x)) :a≤x≤b}offand ending at the right (here we
need that the Jordan arc Fis a cross-cut of R). Thus this curve meets the graph: that is there is a′≤x0≤b′
such that ( x0, g(x0))∈F. Hence, there is a≤x1≤bsuch that ( x0, g(x0)) = ( x1, f(x1)). Consequently,
x0=x1and so f(x0) =f(x0−d). 2) Analytic approach. LetH:=f−g. Then H(a) =m−g(a)≤0 and H(b)
=M−g(b)≥0. Ifg(a) =morg(b) =M, then we are done. So we may assume that H(a)<0 and H(b)>0. Hence,
by the intermediate value theorem, there is x0∈]a, b[ such that H(x0) = 0. We conclude thatf(x0) =g(x0)
=f(x0−d). Let us point out that the assertion does not hold whenever merely inf Rfand supRfexist: just
look at f(x) = arctan x. Motivation for the problem came from the paper: Peter Horak, Partitioning Rninto
connected components. Am. Math. Mon. 122, No. 3, 280-283 (2015), where periodic functions were
considered.
144 Solution to problem 1947 Math. Mag. 87 (2014), 230 Raymond Mortini, J´ erˆ ome No¨ el - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - nX k=0|cosk| ≥1 +nX k=1(cosk)2= 1 +nX
k=1cos(2 k) + 1 2= 1 +n 2+1 2Re nX k=1e2ik! . NownX k=1e2ik=e2i1−e2in 1−e2i=e2iein eisinn sin
1=ei(n+1)sin(n) sin 1. HencenX k=0|cosk| ≥1 +n 2+cos(n+ 1) sin n 2 sin 1≥1 +n 2−1 2 sin 1 =n 2+■ 1−1
2 sin 1■ |{z } >0≥n 2, because 2 sin 1 >1 (note that π/4<1< π/ 3 implies 1 <√ 2<2 sin 1 <√ 3). Let us
remark that in the very first step it was important to begin the sum at k= 1 in order to have the summand
1. Otherwise we would have obtained nX k=0|cosk| ≥n+ 1 2+cosnsin(n+ 1) 2 sin 1≥n+ 1 2−1 2 sin 1 =n
2+1 2■ 1−1 sin 1■ , an estimate that is less than n/2.
145 Solution to problem 1871 Math. Mag. 84 (2011), p. 229 Raymond Mortini - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The solution is a based on the following Lemma:
Lemma 10. LetFbe a continuous, real-valued function on R×R. Suppose that Fis not zero outside the
diagonal Dand not constant 0onD. Then either F≥0orF≤0everywhere. Proof. LetP+={(x, y)∈R2;x <
y}andP−={(x, y)∈R2;x > y}. Case 1: if F(x0, y0)<0 and F(x1, y1)>0 for some points P0= (x0, y0) and
P1= (x1, y1) in P+, then Fmust have a zero on the segment Sjoining P0andP1inP+(since the image of
S under Fis an interval). Case 2: if F(P0)<0 and F(P1)>0 for some P0∈P+andP1∈P−, then we may
choose an arc A(piecewise parallel to the axis) such that F■= 0 on A∩D, which is a singleton. By the
the intermediate value theorem, there is a zero of Fon the arc A, but outside D. Case 3: if F(Q0)<0 and
F(Q1)>0 for some Q0, Q1∈D, then there are P0∈P+and P1∈P−such that F(P0)<0 and F(P1)>0. Hence
we are in the second case. Thus, all cases yield a contradiction to the assumption. Hence, in the image
space, 0 is a global extremum. ■ Solution to the problem Without loss of generality, we may assume
that g >0. Let H(a, b) =  ´b af(x)dx ´b ag(x)dxifa■=b f(a) g(a)ifa=b. We claim that His
continuous on R×R. In fact, it suffices to prove continuity at the diagonal. So let ( a0, a0)∈D. Then, for (
a, b)∈R2\D, there is ξ∈]a, b[ such that´b af(x)dx/(b−a) = f(ξ)→f(a0) if (a, b)→(a0, a0). Thus lim H(a, b)
=H(a0). By assumption, H(a, b)■=f(c)/g(c) whenever ( a, b) is outside the diagonal in R2. Case 1:
H≡f(c)/g(c) on the diagonal D. Then the function x7→f(x)/g(x) has derivative 0 everywhere, and so
satisfies the assertion of the problem. Case 2: Hnot constant f(c)/g(c) onD. Then, by Lemma 10 applied
to F=H−f(c)/g(c), we see that H≥f(c)/g(c) onR×RorH≤f(c)/g(c) onR×R. In particular, cis an extrema of the
function x7→f(x)/g(x) and so the differentiability of f/gimplies that ( f/g)′(c) = 0.
146 Solution to problem 1867 Math. Mag. 84 (2011), p. 150 Raymond Mortini - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Iff≡1, then the assertions are trivially true (just take any
npoints in ]0 ,1[ ). If f■≡1, then there exist points at which fis strictly less than 1 and points where fis
strictly bigger than one (note that this is the only occasion where we have used the hypothesis that´1
0f(t)dt= 1). Hence, due to intermediate value theorem, there is at least one point at which ftakes the
value 1. In particular, if h=forh= 1/f, and noticing that the image of [0 ,1] under fis an interval containing
the point 1 in its interior, there exist b∈[0,1] with M:=h(b)>1 and a sequence (ai) with h(ai)<1 and lim
h(ai) = 1. By compactness, we may assume that ( ai) is converging to some a∈[0,1]. Hence h(a) = 1
and m(δ) := min {h(x) :x∈[a−δ, a+δ]∩[0,1]} →1 ifδ→0. For later purposes, we note that m(δ)<1. Choose
δso small that (n−1)(1−m(δ))≤M−1. Then n−M= (n−1)−(M−1)≤(n−1)m(δ). Now choose n−1 distinct points
x1, . . . , x n−1in [a−δ, a+δ]∩]0,1[ such that m(δ)< h(xj)<1. Then A:=Pn−1 j=1h(xj) satisfies (n−1)m(δ)≤A
< n −1. Thus n−M≤Aand so 1 < n−A≤M. Again, by the intermediate value theorem, there is xn∈]0,1[
such that h(xn) =n−A. Hence nX j=1h(xj) =n. Note that xn/∈ {x1, . . . , x n−1}. Alternate proof concerning
the existence of the cj LetF(x) =´x 0f(t)dtbe the primitive of fvanishing at the origin. Let xj=j/n, j = 0,1, . . .
, n . Then, by the mean-value theorem of differential calculus, there exist cj∈]xj−1, xj[⊆ ]0,1[ such that 1
=F(1)−F(0) =nX j=1(F(xj)−F(xj−1)) =nX j=1F′(cj)(xj−xj−1) =1 nnX j=1f(cj).
147 Solution to problem 1863, Math. Mag. 84 (2011), 64. Raymond Mortini - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We use Carath´ eodory’s definition of differentiability: A
function f:I→Ris differentiable at a point x0∈I,I⊆Ran interval, if there exists a function
g=gx0:I→Rcontinuous at x0 such that f(x) =f(x0) + (x−x0)g(x); just define gx0(x) =(f(x)−f(x0)
x−x0ifx■=x0 f′(x0) if x■=x0. Now if f∈C1[a, b], then gx0is continuous and, by Rolle’s theorem, gx0(x)
=f′(ξ) for some ξ∈]a, b[,ξdepending on x0andx. Hence sup a≤s≤b|gx0(s)| ≤max a≤t≤b|f′(t)|=:M. Letc=
(a+b)/2. Then, using the hypotheses that´b af(x)dx= 0 and the fact that´b a(x−c)dx= 0 we obtain the
following equalities: J:=ˆb axf(x)dx=ˆb a(x−c)f(x)dx= ˆb a(x−c)(f(x)−f(c))dx=ˆb a(x−c)2gc(x)dx. Thus |J|
≤ˆb a(x−c)2Mdx =1 3■ (x−c)3■b aM= 2 3■b−a 2■3 M=1 12(b−a)3M. Iff(x) =xanda=−1, b= 1 then´1
−1f(x)dx= 0 and´1 −1xf(x)dx= 1/3 = ( b−a)3/12.
148 Solution to problem 1860, Math. Mag. 83 (2010), 392 Raymond Mortini - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We use the Schwarz-Pick Lemma telling us that
holomorphic selfmaps of the unit disk are contractions with respect to the (pseudo)-hyperbolic metric
ρand that ρ(f(z), f(w)) =ρ(z, w) for some pair ( z, w)∈D2,z■=wimplies that fis a conformal selfmap of
D(hence of the form eiθb−z 1−bz) and so a (pseudo)-hyperbolic isometry. Note that zn+αzn−1+az+1 =
0 for some z∈Dif and only if zn−1=−az+1 a+z. Now suppose that there are two solutions z, winD. Let
f(z) =−az+1 a+z. Then ρ(z, w) =ρ(f(z), f(w)) =ρ(zn−1, wn−1). But this would imply that zn−1is a bijection
of Donto itself; a contradiction since n≥3. Thus the equation zn+αzn−1+αz+ 1 = 0 has at most one
solution in D. Since zis a solution if and only if1 zis a solution, we see that this polynomial of degree
nmust have at least n−2 solutions (multiplicities counting) on the unit circle. Next we note that u∈Ts a
solution of modulus one of zn+αzn−1+αz+ 1 = 0 if and only ifuis a fixed point on Tof the selfmap φ(z)
=f−1(zn−1) ofD. Since the derivative of φdoes not vanish at boundary fixed points, we conclude that
there are at least n−2distinct solutions of unit modulus.
149 3.College Math. J. ©Mathematical Association of America, 2025. Solution to problem 1283 College
Math. J. 55 (2024), 353 Raymond Mortini, Peter Pflug - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - We shall use the mean value theorem (MVT) for integrals. Let h(x) :=xˆx
0tf(t)dt−ˆx 0t2f(t)dt. Then this c∈]0,1[ with h(c) = 0 exists once we can show that I:=´1 0h(x)dx= 0. This
is true, though, in view of the assumption ˆ1 0(1−t2)tf(t) = 0 . In fact, using Fubini’s theorem, and
noticing that 0 ≤t≤x≤1, ˆ1 0h(x)dx=ˆ1 0ˆx 0■ xtf(t)−t2f(t)■ dtdx =ˆ1 0ˆ1 t■ xtf(t)−t2f(t)■ dxdt =ˆ1 0ˆ1
t(x−t)tf(t)dxdt =ˆ1 0■1−t2 2−(1−t)t■ tf(t)dt =1 2ˆ1 0(1−t)2tf(t)dt= 0.
150 @@ 4.Elemente der Mathematik Solution to problem 1453 Elem. Math. 79 (2024), 176 Raymond
Mortini, Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - This is
entirely trivial. First we have that ■ 1−1 x■n ≤1−n x+n−1=x−1 x+n−1=x1−1 x x+n−1⇐⇒■ ■ ■ 1−1
x■n−1 ≤x x+n−1. Now the right inequality is shown by induction ( x >1): n= 0:■ 1−1 x■−1=x x−1. n→n+
1: ■ 1−1 x■nind.hyp ≤x x+n−1■ 1−1 x■ =x x+n−1x−1 x=x+n−1−n x+n−1 = 1 −n x+n−1≤1−n x+n=x x+n.
151 Solution to problem 1449 Elem. Math. 79 (2024), 131 Raymond Mortini - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Es sei f(x) =(x+ 1)2n+1 x. Dann gilt f′′(x) =(x+ 1)2n−1■
2n(2n−1)x2−(4n−2)x+ 2■ x3 sowie f′′(−2) = 2 n2−1 4 Nun berechnen wir die Laurentreihe und deren
zweite Ableitung: f(x) =2n+1X k=0■2n+ 1 k■ xk−1. f′′(x) =2n+1X k=0■2n+ 1 k■ (k−1)(k−2)xk−3 =
22n+1X k=0■2n+ 1 k■■k−1 2■ xk−3 Auswertung an x=−2 ergibt: f′′(−2) = 22n+1X k=0(−1)k−1■2n+ 1
k■■k−1 2■ 2k−3 = 2( −1)1■2n+ 1 0■■−1 2■ 2−3+ 2A =−1 4+ 2A. Folglich gilt A=n2. This was a very
special case of problems treated in [38, p.8].
152 Solution to problem 1447 Elem. Math. 79 (2024), 84 Raymond Mortini, Rudolf Rupp - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Zu untersuchen ist f¨ ur welche n, m die
Gleichung (20 + 24√ 2)n= (24 + 20√ 2)mgilt. Wir zeigen, dass nur die 1 als gemeinsame Zahl vorkommt
(dies f¨ ur n=m= 0). In der Tat, bezeichnet N(a+b√ 2) := a2−2b2die sogenannte Norm des Elements
a+b√ 2 im quadratischen Zahlk¨ orper Q[√ 2], so erhalten wir wegen deren Multiplikativit¨ at N(xy)
=N(x)N(y) folgende Bedingung: N(4n(5 + 6√ 2)n) =N(4m(6 + 5√ 2)m). ¨Aquivalent: 42n(25−72)n=
42m(36−50)m also 16n(−47)n= 16m(−14)m. Da 47 eine Primzahl ist, die rechte Seite aber nicht durch
diese Zahl teilbar ist falls n, m≥1, bleibt nur n=m= 0 ¨ ubrig.
153 Solution to problem 1443 Elem. Math. 79 (2024), 38 Raymond Mortini - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We use that (see [39, p. 453]) ˆπ/4 0log √ 2 cosx+ 1√ 2
cosx−1! dx= 2C. Now put v:=√ 2 cosx, 1≤v≤√ 2, dv=−√ 2 sinx dx =−√ 2p 1−cos2x dx =−√ 2(p
1−v2/2dx=−p 2−v2dx. Hence 2C=ˆ√ 2 1log■ v+1 v−1■ √ 2−v2dv.
154 Solution to problem 1441 Elem. Math. 78 (2023), 180 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Diese Aufgabe ist unseren Ermessens
nach nicht korrekt formuliert, da per Definition eine Kurve das stetige Bild eines Intervals in einen
topologischen Raum ist. Deshalb liegen hier zwei Kurven vor und wir werden folgendes zeigen13:
4.1.Nichtnegative Parameter. Proposition 11. Es seien a, b∈R,a, b≥0undΓ±die Kurven, welche
gegeben sind durch die Parameterdarstellung Γ+:z(t) = (x(t), y(t)) =■ t−a t, t2+b t■ , t >0,
beziehungsweise Γ−:z(t) = (x(t), y(t)) =■ t−a t, t2+b t■ , t <0. Dann gilt: (1) Γ±besitzen keine Selbst¨
uberschneidungen; sind also Jordanb¨ ogen. (2)F¨ ura >0schneidet Γ+die Kurve Γ−in genau einem
Punkt. (3)F¨ ura= 0 liegt der Graph einer Funktion auf R\ {0}vor, und folglich sind keine Selbst¨
uberschneidungen vorhanden. (4) Γ+schneidet Γ−f¨ ura, b∈N={0,1,2. . .}mit senkrechten Schnittwinkel
genau dann wenn (a, b) = (2 r2+ 1, r(2r2+ 1)) f¨ ur ein r∈N. L¨ osung (1) (2) Zu betrachten ist das
folgende System von Gleichungen f¨ ur s, t∈R\ {0}:  t−a/t =s−a/s t2+b/t =s2+b/s.⇐⇒  t−s=a■1
t−1 s■ =a(s−t)1 st t2−s2=b■1 s−1 t■ =b(t−s)1 st. F¨ urs■=tunda■= 0 ist dies ¨ aquivalent zu■st =−a
st(t+s) = b⇐⇒■st =−a t+s=−b a. Dies f¨ uhrt auf die L¨ osung der quadratischen Gleichung 0
=x2−(s+t)x+st=x2+b ax−a. Die L¨ osungen hierzu sind (54) s=−b+√ b2+ 4a3 2aundt=−b−√ b2+ 4a3 2a.
Es liegen also ein negativer und ein positiver Wert der Kurven-Parameter sundtvor. Folglich schneidet
Γ−die Kurve Γ+in genau einem Punkt und es liegen keine Selbst¨ uberschneidungen vor. 13Die
Schreibweise mit −AundB, anstatt AundB, l¨ asst uns vermuten, dass bloss nichtnegative Zahlen
gemeint waren mit der Bezeichnung ”ganzzahlig”. Im zweiten Abschnitt betrachten wir auch den Fall
wo a, bbeliebig sind.
155 (3) ist klar. (4) Es sei M:= (z(s), z(t)) dieser eindeutige Schnittpunkt mit s < 0 und t >0. Diese
Kurven Γ−und Γ+schneiden sich nun senkrecht in Mgenau dann wenn gilt 0 = ■■z(s),■z(t)■= ■x(s)
■x(t) + ■y(s) ■y(t) ⇐⇒■ 1 +a s2■ ■ 1 +a t2■ =−■ 2s−b s2■ ■ 2t−b t2■ . Einsetzen von
st=−aundst(s+t) =bergibt −■ 1−t s■■ 1−s t■ =■ 2s−t(t+s) s■ ■ 2t−s(t+s) t■ . Multiplikation mit stliefert
−(s−t)(t−s) =■ 2s2−t(t+s)■ ■ 2t2−s(t+s)■ = (s−t)(t+ 2s) (t−s)(s+ 2t) = (s−t)(t−s)(t+ 2s)(s+ 2t). Das■=t,
erh¨ alt man schliesslich (55) ( t+ 2s)(s+ 2t) =−1. Umformen ergibt −1 =ts+ 2s2+ 2t2+ 4st= 5st+ 2■
(s+t)2−2st■ =st+ 2(s+t)2. Durch Einsetzen von st=−aunds+t=−b/aergibt das (56) −1 =−a+ 2■b a■2
⇐⇒ − a2=−a3+ 2b2⇐⇒ 2b2+a2−a3= 0. Wir m¨ ussen nun alle L¨ osungspaare ( a, b)∈N2dieser
diophantischen Gleichung bestimmen. Ein Umschreiben ergibt (57) 2b2 a2−a=−1. Es sei r:=b a. Gem¨
ass der Voraussetzung ist r∈Q. Also hat adie Form a= 2r2+ 1 und b=ra=r(1 + 2 r2). Dies ist jedoch nur
m¨ oglich wenn rselbst in Nliegt, was man wie folgt einsehen kann. Ist r= 0 so ist das evident. Sei also
r■= 0. Die Voraussetzungen a∈Nund a= 2r2+ 1 implizieren m:= 2r2∈N. Wir zeigen dass mvon der
Form m= 2n2ist f¨ ur ein n∈Nund damit ist r∈N. Es sei r=p/q, mit ggt ( p, q) = 1. Sodann mq2= 2p2. Ist
m= 2i gerade, so erhalten wir iq2=p2. Da jeder Primfaktor von qnunp2teilt, also auch p, muss wegen
ggt ( p, q) = 1 nun q= 1 sein. D.h. mhat die gew¨ unschte Form. Ist m= 2i+1 ungerade, so muss wegen
2 p2=mq2die Zahl qauch gerade sein. Sagen wir q= 2ju, wobei j∈N, j■= 0, unduungerade. Folglich ist
p2= (mu2)22j−1. Weil 2 kein gemeinsamer Faktor von pundqist, erhalten wir den Widerspruch, da die
linke Seite von p2= (mu2)22j−1ungerade ist, die rechte aber gerade. Damit haben alle L¨ osungen von
(57) notwendigerweise die Form ( a, b) = (1+2 r2, r(1+2 r2)), r∈N. Umgekehrt, ist auch jedes solche
Paar Lo¨ sung der Gleichung (57): 2r2−(1 + 2 r2) =−1. Beispiele f¨ ur ( a, b): (1,0), (3 ,3), (9 ,18), (19
,57), (33 ,132). Damit hat man mit ( a, b) = (1 ,0) den einzigen Schnittpunkt M= (0,1) von Γ+mit Γ−im 90
Grad Winkel f¨ ur ( s, t) = (1 ,−1) bei Γ+(s) =■ s−1 s, s2■ , s > 0,Γ−(t) =■ t−1 t, t2■ , t <0,
156 Figure 11. r= 0,a= 1,b= 0 oder mit ( a, b) = (3 ,3) den Schnittpunkt M= (−1,4) von Γ+mit Γ−im 90
Grad Winkel f¨ ur (s, t) = (−1+√ 13 2,−1−√ 13 2) bei Γ+(s) =■ s−3 s, s2+3 s■ , s > 0,Γ−(t) =■ t−3 t, t2+3
t■ , t <0, oder mit ( a, b) = (9 ,18) den Schnittpunkt M= (−2,13) von Γ+mit Γ−im 90 Grad Winkel f¨ ur (s,
t) = (√ 10−1,−1−√ 10) bei Γ+(s) =■ s−9 s, s2+18 s■ , s > 0,Γ−(t) =■ t−9 t, t2+18 t■ , t <0, Das
¨Uberraschendste f¨ ur uns bei dieser proposition: der Schnittpunkt Mhat auch ganz- zahlige
Komponenten: M= (−r,1 + 3 r2). Figure 12. r= 1,a= 3,b= 3
157 4.2.Die restlichen F¨ alle. Fall 2.1 :b2+ 4a3<0, ¨ aquivalent a <−■ b2 4■1/3 . In diesem Fall hat die
quadratische Gleichung (54) keine reellen L¨ osungen, und folglich sind Γ+und Γ− Jordanb¨ ogen die
sich nicht schneiden (siehe Grafik 13). Figure 13. a=−1,b= 1 Fall 2.2 :−■ b2 4■1/3 < a < 0. In diesem
Fall ist b2+ 4a3>0 und es liegen zwei verschiedene L¨ osungen s, tder quadratische Gleichung (54) vor,
welche aber wegen st=−a >0 dasselbe Vorzeichen haben. Damit schneiden sich die Kurven Γ+und
Γ−nicht, aber genau eine von denen hat einen Selbst¨ uberschneidungspunkt (siehe Grafik 14). N¨
amlich Γ+fallsb >0 und Γ− fallsb <0. Anmerken m¨ ochten wir noch, dass die Gleichung (56), 2
b2+a2−a3=−1, keine L¨ osung hat falls a <0. Folglich ist diese Selbst¨ uberschneidung nie senkrecht.
Figure 14. a=−1,b= 3
158 Fall 2.3 :b2+ 4a3= 0. Auch hier schneiden sich die Kurven Γ+und Γ−nicht, und beide sind wieder
Jordanb¨ ogen. Siehe Grafik (15) zum Beispiel Γ+(s) =■ s+1 s, s2+2 s■ , s > 0,Γ−(t) =■ t+1 t, t2+2 t■ ,
t <0. Figure 15. a=−1,b= 2 Fall 3 :a >0,b≤0. Da ist prinzipiell kein Unterschied zum Fall a >0, b≥0; es
liegt nur eine Spiegelung der Kurven an der y-Achse vor. Z.B. hat f¨ ur b <0 die Kurve (x(t), y(t)) =■ t−a
t, t2+b t■ , t >0, mit der Transformation t→ −tauch die Parameterdarstellung ■ −(t−a t), t2−b t■ , t <0.
Die Spiegelung an der yAchse ist dann gegeben durch ■ t−a t, t2+(−b) t■ , t <0 Figure 16. a= 9,b= 18
blau, b=−18 rot
159 Fazit : Die Kurve Γ+= Γ+(a, b) schneidet f¨ ur a, b∈Zdie Kurve Γ−(a, b) in einem rechten Winkel
genau dann wenn (a, b) =■ (1 + 2 r2),±r(1 + 2 r2)■ mitr∈N={0,1,2,3. . .}. Oder in der Formulierung der
proposition: Die unstetige ”Kurve” c=c(a, b) besitzt eine ”Selbst¨ uberschneidung” mit senkrechtem
Schmittwinkel genau dann wenn (a, b) =■ (1 + 2 r2), r(1 + 2 r2)■ mitr∈Z.
160 Solution to problem 1442 Elem. Math. 78 (2023), 180 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Letf(x, y, z ) :=1 x+1 y+2 z. Using Lagrange
multipliers, we show that ■ ■ inf x2+y2+z2=1 x>0,y>0,z>0f(x, y, z ) = min x2+y2+z2=1 x,y,z≥1 4√ 3f(x,
y, z ) = (2 + 22/3)3/2∼6.794693902 ··· First we recall the version of Lagrange’s theorem, we will use:
Theorem 12. LetG:={ξ= (x, y, z )∈R3:x > 0, y > 0, z > 0}be the first octant and g(x, y, z ) :=x2+y2+z2−1.
Then fandgbelong to C1(G). Ifζ∈Gis a local extremum of fon the set N:={(x, y, z )∈G:g(x, y, z ) = 0}for
which∂ ∂zg(ζ)■= 0, then there exists λ∈R such that (ζ, λ)is a stationary point of Lagrange’s function
L(x, y, z, λ ) :=f(x, y, z ) +λg(x, y, z ). Next, we prove the existence of such a local extremum. Let η:=
(3−1/2,3−1/2,3−1/2). Then η∈N. Moreover, f(η) = 4√ 3. Also, if ( x, y, z )∈Nis such that at least one of its
coordinates is strictly bigger than (4√ 3)−1, then f(x, y, z )>4√ 3. Hence inf Nf= min■ f(x, y, z )
:x2+y2+z2= 1, x, y, z ≥1 4√ 3■ . Finally we solve Lagrange’s equations for ( x, y, z )∈Gandλ∈R: (1)∂
∂xL(x, y, z, λ ) =−1 x2+λ(2x)!= 0 (2)∂ ∂yL(x, y, z, λ ) =−1 y2+λ(2y)!= 0 (3)∂ ∂zL(x, y, z, λ ) =−2 z2+λ(2z)!=
0 (4)∂ ∂λL(x, y, z, λ ) =x2+y2+z2−1!= 0 (1) and (2) yield that x=yand (2) and (3) yield that2 z3=1 y3,
equivalently z= 21/3y. Due to (4), 1 =x2+x2+ 22/3x2, hence x=y= (2 + 22/3)−1/2, z = 21/3(2 + 22/3)−1/2.
Consequently, the unique stationary point of LonG×Ris P=■1√ 2 + 22/3,1√ 2 + 22/3,21/3 √ 2 + 22/3,(2
+ 22/3)3/2 2■
161 Letζbe the point formed with the first three coordinates of P, which are of course bigger than (4√
3)−1. Then f(ζ) = 2p 2 + 22/3+ 2√ 2 + 22/3 21/3= (2 + 22/3)p 2 + 22/3= (2 + 22/3)3/2. Of course, this
point ζmust now be that unique point on Nwhere inf Nis taken (note that supfN=∞).
162 Solution to problem 1438 Elem. Math. 78 (2023), 135 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Three solutions for (1) in [12, Appendix 27],
one for (2) in [12, Appendix 31]. (1) Man beachte zun¨ achst ∆ ={t1z1+t2z2+t3: 0≤tj≤1, t1+t2+t3= 1}. Ist
nun z∈∆, so gilt z=t1z1+t2z2+ (1−t1−t2) =t1(z1−1) +t2(z2−1) + 1 . Daher |1−z|=|t1(z1−1) +t2(z2−1)|
≤t1|z1−1|+t2|z2−1|. Andererseits |z| ≤t1|z1|+t2|z2|+ 1−t1−t2. Somit 1− |z| ≥t1(1− |z1|) +t2(1− |z2|).
Folglich gilt |1−z| 1− |z|≤t1|z1−1|+t2|z2−1| t1(1− |z1|) +t2(1− |z2|)≤max{Z1, Z2}:=κ, wobei Zj=|1−zj| 1−
|zj|. (2) Ist f¨ ur κ >0,D∗ κ={reit∈D:|t| ≤κ(1−r)}, so ist der Teil der Randkurve welche im 1-ten Quadraten
liegt, in Polarkoordinaten gegeben durch r(t) = 1−t κ,0≤t≤min{π/2, κ}. Diese Kurve ist Graph einer
konkaven Funktion (Bilder und Details wie im Buch [12, Ap- pendix 31] unter Beachtung dass dort alles
auch f¨ ur 0 < κ < 1 gilt). Folglich ist die Menge {(x, y) : 0≤x=x(t)≤1,0≤y≤y(t)} konvex, und somit auch
S(κ) :=D∗ κ∩ {z∈C: Rez≥0}. Unter Beachtung dass D∗ κ1⊆D∗ κ2 f¨ ur 0 < κ1≤κ2, schliessen wir, dass
mit z1, z2,1∈S(κ) auch ∆ in S(κ) liegt falls κ= max■|t1| 1−r1,|t2| 1−r2■ .
163 Solution to problem 1437 Elem. Math. 78 (2023), 135 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - This is well known since ”hundreds of
years” (see [40]). (a) First note that f(x) =−´1 0(1−x)sds. In fact −ˆ1 0(1−x)sds=−ˆ1
0eslog(1−x)ds=−■eslog(1−x) log(1−x)■1 0 =−■1−x log(1−x)−1 log(1−x)■ =x log(1−x). Now (1−x)s=∞X
k=0(−1)k■s k■ xk. Since´P=P´ (due to uniform convergence in sfor every fixed x∈]−1,1[; note that
|(−1)k■s k■ | ≤1 ), we obtain f(x) =∞X k=0■ (−1)k+1ˆ1 0■s k■ ds■ xk. Hence ak= (−1)k+1ˆ1 0■s k■
ds= (−1)k+1ˆ1 0s(s−1)(s−2)···(s−k+ 1) k!ds =ˆ1 0s(1−s)(2−s)···(k−1−s) k!ds =1 kˆ1 0s■ 1−s 1■■ 1−s 2■
···■ 1−s k−1■ ds. Since every factor is less than 1, we obtain 0≤ak≤1 k. Moreover, as 0 ≤s≤1, and for
k≥2, ak≥ˆ1 0s(1−s)1·2·3·4. . .(k−2) k!ds=■1 2−1 3■1 k(k−1)=1 6k(k−1). The right-hand side, though, is
smaller than (3 k2)−1. So we need a more careful estimate. Instead of ”breaking” after the second
factor, we brake after the fifth factor. Noticing that ˆ1 0s(1−s)(2−s)(3−s)(4−s)ds=9 4,
164 we obtain for k≥5, ak≥9 44. . .5. . .(k−2) k!=9 41 3!1 k(k−1)=3 81 k2≥1 31 k2. The estimate ak≥1
3k2now holds also for k= 1, . . . , 4, due to the following explicit repre- sentation of the Taylor sums for
f(x): (b) We first show that ( ak) is decreasing (to 0): ak+1=1 k+ 1ˆ1 0s■ 1−s 1■■ 1−s 2■ ···■ 1−s
k−1■■ 1−s k■ |{z} ≤1ds ≤k 1 +kak≤ak. The alternating series test of Leibniz now yields the
convergence ofP∞ k=0(−1)kak. Finally, by Abel’s rule (see [12, p. 1415]), ∞X k=0(−1)kak= lim
x→−1+f(x) =−1 log 2. Next we show that kak→0. In fact, due to 1 −x≤e−xforx≥0, kak=ˆ1 0s■ 1−s 1■■
1−s 2■ ···■ 1−s k−1■ ds ≤ˆ1 0exp −sk−1X j=11 j ds=− exp −sk−1X j=11 j  k−1X j=11 j
1 0 ≤1 k−1X j=11 j≤1 ˆk 1dx/x=1 logk. Note that lim x→1−f(x) = 0. Hence, by Tauber’s Theorem
([42, p. 52]), F:=P∞ k=0ak is convergent andP∞ k=0ak= 0. Very funny! By the way, Fis called Fontana’s
series (Gre- gorio Fontana 1735–1803), see [41, (formula 20)], and the akare the (moduli) of the
Gregory coefficients (James Gregory 1638–1675), see [41] and [40].
165 Solution to problem 1434 Elem. Math. 77 (2022), 85 Raymond Mortini, Rudolf Rupp - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let S3:=∞X n=01 (3n+ 3)3andT3:=∞X
n=0(−1)n (3n+ 3)3. Then S3=1 27ζ(3) and T3=1 27η(3) where η(3) is the Dirichlet η-function. It is
well-known that η(3) = (3 /4)ζ(3). Thus S1+S2+S3=ζ(3) and T1−T2+T3=η(3) imply that S1+S2=26
27ζ(3) and T1−T2=13 18ζ(3). On the other hand, S1−T1=∞X n=01−(−1)n (3n+ 1)3= n=2k+12∞X k=02
(3(2k+ 1) + 1)3 = 2∞X k=01 (6k+ 4)3=1 4∞X k=01 (3k+ 2)3 =1 4S2. Moreover, S2+T2=∞X n=01 + (−1)n
(3n+ 2)3= n=2k2∞X k=01 (6k+ 2)3 = =1 4∞X k=01 (3k+ 1)3 =1 4S1. This yields the linear system  1
1 0 0 1−1 4−1 0 −1 41 0 1 0 0 1 −1 · S1 S2 T1 T2 = 26 27ζ(3) 0 0 13 18ζ(3) . The
determinant is, unfortunately, zero. The null-space is the one dimensional vector space generated by (4
,−4,5,5)⊥. We may write the system as
166  1 1 0 0 1−1 4−1 0 −1 41 0 1 0 0 1 1 · S1 S2 T1 T2 = 26 27ζ(3) 0 0 13 18ζ(3) + 2
T2 . In other words,  S1 S2 T1 T2 = 1 22 5−2 52 5 1 2−2 52 5−2 5 3 8−1 2−1 21 2 −3
81 21 21 2 · 26 27ζ(3) 0 0 13 18ζ(3) + 2 T2 . That is (58)  S1=R1+4 5T2
S2=R2−4 5T2 T1=R3+T2 T2= T2 where the Rjare rational multiples of η(3). More precisely, R1=1 2·26
27ζ(3) +2 5·13 18ζ(3) =104 5·27·ζ(3), R2=1 2·26 27ζ(3)−2 5·13 18ζ(3) =26 5·27·ζ(3), R3=3 8·26 27ζ(3)
+1 2·13 18ζ(3) =13 18·ζ(3). Next we use that for 0 < a < 2π(see below) (59) h(a) :=∞X n=0sin(n+ 1)a
(n+ 1)3=a3−3πa2+ 2π2a 12, and put a= 2π/3. Then, since sin(2 π/3) =√ 3/2, S1−S2=2√ 3h(2π/3) =2√
3·2π3 81=4π3 81√ 3. By the formula (58) above, S1−S2=4 5·13 18ζ(3) +8 5T2. Hence T2=5 8■4π3 81√
3−4 5·13 18ζ(3)■ =■ ■ 5π3 2·81√ 3−13 36ζ(3)∼0.11843 ···. Finally, by (58) again, ■ ■ S1=13 27ζ(3)
+2π3 81√ 3= 3−5(117ζ(3) + 2√ 3π3)∼1.02078 ···, ■ ■ S2=13 27ζ(3)−2π3 81√ 3= 3−5(117ζ(3)−2√
3π3)∼0.13675 ···, ■ ■ T1=13 36ζ(3) +5π3 2·81√ 3=1 1944■ 702ζ(3) + 20√ 3π3■ ∼0.98659 ···.
Addendum
167 The value in (59) for h(a) is given as follows (we had developed this in solving the problem 12388
in AMM). Note that h′(a) =∞X n=0cos(n+ 1)a (n+ 1)2. Since1 3π2+ 4P∞ n=1cosnx n2is the Fourier
series of the function ( x−π)2, 0≤x <2π, extended 2π-periodically, we see that for 0 < a < 2π, h′(a)
=(a−π)2 4−π2 12. Ash(0) = 0, we deduce that for 0 < a < 2π, h(a) =(a−π)3 12−π2 12a+π3 12=a3−3πa2+
2π2a 12.
168 Solution to problem 1435 Elem. Math. 77 (2022), 85 Raymond Mortini, Rudolf Rupp - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let A(x, y) :=p x2+ 3 +p y2+ 3 +p xy+ 3.
Since by assumption x+y= 2, we just have to prove that 6 is the minimal value of the function f(x) :=p
x2+ 3 +p (2−x)2+ 3 +p x(2−x) + 3,0≤x≤2, which is attained at x= 1. First we note that fis symmetric with
respect to x= 1; that is f(1 +x) =f(1−x) for 0 ≤x≤1. We show that fdecreases on [0 ,1]. Note that f(0) = 2√
3 +√ 7>6 =f(1). It is sufficient to prove that f′≤0 on [0 ,1]: f′(x) =x√ x2+ 3−2−xp (2−x)2+ 3+1 22−2xp
x(2−x) + 3. Butf′≤0 for 0 ≤x≤1 if and only if (60) L(x) :=x√ x2+ 3+1−xp x(2−x) + 3≤2−xp (2−x)2+ 3. Next
note that, due to 2 −x≥x, 1−xp x(2−x) + 3≤1−x√ x2+ 3. Hence L(x)≤1√ x2+3. Thus (60) holds for 0 ≤x≤1
if (61)1√ x2+ 3≤2−xp (2−x)2+ 3, or equivalently (62)p (2−x)2+ 3√ x2+ 3≤2−x.
169 This holds, though, due to the following equivalences for 0 ≤x≤1: (2−x)2+ 3≤(2−x)2(x2+ 3)⇐⇒
3≤(2−x)2(x2+ 2) =: R(x). The latter is true, since min 0≤x≤1R(x) =R(1) = 3 ( note that the derivative of
Requals R′(x) =−4(2−x)(x2−x+ 1), so R′≤0 on [0 ,1].)
170 Partial solution to problem 1431, Elem. Math. 78 (2023), 44 Raymond Mortini, Rudolf Rupp - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let S:=∞X k,n=0(−1)k+n+1 (2n+ 1)(2
k+ 1)(2 n+ 2k+ 3)2■−1/2 n■ . Then Sis the value of the integral ˆ1 0arcsin xarctan xlogx dx. Just take
the series arcsin x=∞X n=0(−1)n 2n+ 1■−1/2 n■ x2n+1 arctan x=∞X n=0(−1)n 2n+ 1x2n+1 and use
thatˆ1 0xmlogx dx =−(m+ 1)−2. For a solution see [43].
171 Partial solution to problem 1428 Elem. Math. 77 (2022), 196 Raymond Mortini, Rudolf Rupp - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let f(a, b, c ) :=■a b+c+b c+a+c
a+b■ab+bc+ca a2+b2+c2. Then, by using that2xy x+y≤√xy≤x+y 2, f(a, b, c ) =1 a2+b2+c2■ abc
b+c+bca c+a+cab a+b■ + 1 ≤1 a2+b2+c2■ ab+c 4+bc+a 4+ca+b 4■ + 1 =1 a2+b2+c22ab+ 2bc+ 2ca
4+ 1 =1 a2+b2+c2(a+b+c)2−(a2+b2+c2) 4+ 1 ≤1 a2+b2+c2(a2+b2+c2)(1 + 1 + 1) −(a2+b2+c2) 4+ 1 ≤3
2. If we let a=b=c, then f(a, a, a ) = 3 /2 and so■ ■■ ■k2= 3/2 . To determine k1, let g(a, b, c ) :=■a
b+c+b c+a+c a+b■2ab+bc+ca a2+b2+c2. Then, by using two of the estimates above, namley f≥1, and
Cauchy-Schwarz, g(a, b, c ) =a2+b2+c2 ab+bc+caf(a, b, c )2>a2+b2+c2 ab+bc+ca≥1. We guess k1=√
2. In fact, we may restrict to triples ( x,1, c) (homogeniety). Then it remains to prove that fc(x) :=■1
x+c+x c+ 1+c 1 +x■2■x+cx+c 1 +x2+c2■ ≥2. Now lim c→0fc(x) =x+1 x=f0(x)≥2. Graphical evidence
seems to indicate that mc:= min x>0fc(x)≥2 and lim c→0mc= 2. As it is customn with this type of
questions, the infimum of the two-variable function f(x, c) := fc(x) is taken on the boundary of the first
quadrant; that is when c= 0. We have no proof though of this last claim.
172 problem 1383 in Elem. Math 74 (2019), 38, by Raymond Mortini and Rudolf Rupp - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Theorem 13. Let0< α≤1. Then σ(α) :=
sup■|(1−z)α−(1−w)α| |z−w|α:|z|,|w| ≤1, z■=w■ = max {1,21−αsin(απ/2)} =( 1 if0< α≤1/2
21−αsin(απ/2)if1/2≤α≤1. Moreover, max 0<α≤1logσ(α) =■ 1−2 πarctan■π 2 log 2■■ log 2 + log πp π2+
4(log 2)2! . 101 αoσ(α )oq(α) α0.5 Figure 17. The H¨ older-Lipschitz constant σ(α) See [7] for this best
H¨ older-Lipschitz constant associated with (1 −z)α.
173 Solution to problem 1350 in Elem. Math 71 (2016), 84 Raymond Mortini - - - - - - - - - - - - - - - - - - - -
----------------------------------
174 Solution to problem 1339 Elem. Math. 70 (2015), 82. Raymond Mortini - - - - - - - - - - - - - - - - - - - - -
---------------------------------
175 Solution to problem 1281 Elem. Math. 65 (2010), 127, by Raymond Mortini, J´ erˆ ome No¨ el - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - a) Exponentiating, we have to
calculate the value of the infinite product P=∞Y n=2■ e2■n−1 n+ 1■n■ . We claim that P=4π e3; soS=
log(4 π)−3. LetPN=QN n=2■ e2■ n−1 n+1■n■ . Then by Stirlings formula, telling us that n!∼e−nnn√
2πn, we obtain PN=1 e2e2NQN n=2(n−1)n QN n=2(n+ 1)n= 1 e2e2NQN n=2(n−1)n QN n=2nn+1QN
n=2nn+1 QN n=2(n+ 1)n= 2 e2e2N(N!)2 NN+1(N+ 1)N=2 e2■eNN! NN■2NN (N+ 1)N1 N∼ 2 e2(√
2πN)2 N1 (1 +1 N)N→4π e3. b) To determine S∗, we use the same method and calculate the value of
P∗=∞Y n=1e■n n+ 1■n+1 2 We claim that P∗=√ 2π eand so S∗=1 2log(2π)−1. In fact P∗ N=NY n=1e■n
n+ 1■n+1 2 =eNN! (N+ 1)N1√ N+ 1. Using Stirling’s formula we obtain PN∼NN (N+ 1)N√ 2πN1√ N+ 1=√
2π1 (1 +1 N)N√ N√ N+ 1→√ 2π e.
176 . Solution to problem 901 Elem. Math. 38 (1983), 128. Raymond Mortini
177 .5.Crux Mathematicorum Solution to problem 4988 Crux Math. 50 (9) 2024, 466 Raymond Mortini,
Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that
S:=∞X n=1 (2n−1)∞X j=01 (j+n)2−2 =−1 2. The series∞X j=01 (j+x)2, denoted by ψ(1)(x) or ζ(2, x),
is sometimes called the trigamma function. It is a special Hurwitz-zeta function. It can be represented
as an integral: if x >0, then ψ(1)(x) =ˆ∞ 0te(1−x)t et−1dt. Now, via induction (thanks to the software
wolframalpha; attention: that software says that the series diverges, sic!) NX n=1[(2n−1)ψ(1)(n)−2]
=−N+N2ψ(1)(N+ 1). Thus it remains to show that IN:=■ −N+N2ˆ∞ 0t et−1e−Ntdt■ → −1 2. This is done
using twice partial integration: IN = −N+N2" −e−Nt Nt et−1 ∞ 0+1 Nˆ∞ 0e−Ntet−1−tet (et−1)2dt# =
−N+N2■1 N+1 Nˆ∞ 0e−Ntet−1−tet (et−1)2dt■ =Nˆ∞ 0e−Ntet−1−tet (et−1)2dt = N −e−Nt Net−1−tet
(et−1)2 ∞ 0+1 Nˆ∞ 0e−Nt−(et−1)tet−(et−1−tet)2et (et−1)3 | {z } :=b(t)dt  t→0= l′Hospital−1 2+ˆ∞
0e−Ntb(t)dt=−1 2+εN. Since bis bounded on ]0 ,∞[ and´∞ 0e−Ntdt=1 N, we have that εN→0.
178 Solution to problem 4965 Crux Math. 50 (7) 2024, 368 Bikash Chakraborty, Raymond Mortini,
Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We observe that
the sign in a) is false. Step 1 Reduction to the computation of an integral We use that for a >0, 1 a=ˆ∞
0e−axdx. Since the moduli of the partial sumsPN k=0e−nx(−1)ke−2kxare bounded by the L1([0,∞[)-
function 2 e−nx, we haveP´ =´P, and so Sn:=1 n−1 n+ 2+. . . =∞X k=0(−1)k n+ 2k=∞X k=0(−1)kˆ∞
0e−(n+2k)xdx =ˆ∞ 0e−nx∞X k=0(−1)ke−2kxdx=ˆ∞ 0e−nx 1 +e−2xdx. Now let A:=∞X n=1(−1)n−1 n■1
n−1 n+ 2+. . .■ and B:=∞X n=11 n■1 n−1 n+ 2+. . .■ . Note that Snis decreasing, so the series
Aconverges (Leibniz rule). But actually, 0≤Sn≤ˆ∞ 0e−nxdx=1 n. So the series AandBconverge
absolutely as the general term is O(1/n2). Due to the boundedness of the partial sums of the seriesP∞
n=11 ne−nxby the L1(]0,∞[)- function |log(1−e−x)|, we haveP´ =´Pin both cases. Hence B=∞X n=11 nˆ∞
0e−nx 1 +e−2xdx=ˆ∞ 01 1 +e−2x∞X n=11 ne−nxdx =−ˆ∞ 0log(1−e−x) 1 +e−2xdx = x7→e−x−ˆ1
0log(1−x) x(1 +x2)dx. A=∞X n=1(−1)n−1 nˆ∞ 0e−nx 1 +e−2xdx=ˆ∞ 01 1 +e−2x∞X n=1(−1)n−1 ne−nxdx
=ˆ∞ 0log(1 + e−x) 1 +e−2xdx = x7→e−xˆ1 0log(1 + x) x(1 +x2)dx.
179 Step 2 Calculating the integrals To evaluate the integrals, we calculate I:=A+BandJ:=A−Bgiving A=
(I+J)/2 and B= (I−J)/2. I=ˆ1 0log■ 1+x 1−x■ x(1 +x2)dx1−x 1+x=t = x=1−t 1+tˆ1 0−logt ■1−t 1 +t■■ 1
+■1−t 1 +t■2■2dt (1 +t)2 =−2ˆ1 0(1 +t) logt (1−t)■ (1 +t)2+ (1−t)2■dt=−2ˆ1 0(1 +t) logt (1−t)(2 + 2 t2)dt
=−ˆ1 0(1 +t) logt (1−t)(1 + t2)dt=−■ˆ1 0logt 1−tdt+ˆ1 0tlogt 1 +t2dt■ s=t2 =−ˆ1 0logt 1−tdt−1 4ˆ1 0logs 1
+sds. J=ˆ1 0log(1−x2) x(1 +x2)dxx2=s=1 2ˆ1 0log(1−s) s(1 +s)ds =1 2ˆ1 0log(1−s) sds−1 2ˆ1 0log(1−s)
1 +sds =1 2ˆ1 0logt 1−tdt−1 2ˆ1 0logt 2−tdt. Now ˆ1 0logt 2−tdtt=2u=ˆ1/2 0log(2u) 2−2u2du=ˆ1/2 0log 2
+ log u 1−udu =−(log 2)h log(1−u)i1/2 0+ˆ1/2 0logu 1−udu = log22 +ˆ1/2 0logu 1−udu. Hence, by using
that for 0 < u≤1,−ˆu 0log(1−x) xdx= Li 2(u) =∞X n=1un n2(di-logarithm), we obtain with 0 < v < 1, ˆv
0logx 1−xdx = x=1−sˆ1 1−vlog(1−s) sds=ˆ1 0log(1−s) sds−ˆ1−v 0log(1−s) sds =−Li2(1) + Li 2(1−v).
Consequently, by additionally using that ˆ1 0logx 1 +xdx=∞X n=0ˆ1 0(−1)n(logx)xndx=∞X n=0(−1)n+1 1
(n+ 1)2=−π2 12,
180 2A=I+J=■ −ˆ1 0logx 1−xdx−1 4ˆ1 0logx 1 +xdx■ +" 1 2ˆ1 0logx 1−xdx−1 2 log22 +ˆ1/2 0logx 1−x!#
=−1 2log22 +1 2Li2(1)−1 2■ −Li2(1) + Li 2(1 2)■ −1 4∞X n=0(−1)n+1 (n+ 1)2 =−1 2log22 +π2 6−1 2■π2
12−log22 2■ +π2 48 =−1 4log22 +7 48π2, and 2B=I−J=■ −ˆ1 0logx 1−xdx−1 4ˆ1 0logx 1 +xdx■ −" 1 2ˆ1
0logx 1−xdx−1 2 log22 +ˆ1/2 0logx 1−x!# =1 2log22 +3 2Li2(1) +1 2■ −Li2(1) + Li 2(1 2)■ −1 4∞X
n=0(−1)n+1 (n+ 1)2 =1 2log22 +π2 6+1 2■π2 12−log22 2■ +π2 48 =1 4log22 +11 48π2. Consequently
A=−1 8log22 +7π2 96∼0.659602 . . . and B=1 8log22 +11π2 96∼1.19095 . . .
181 Solution to problem 4959 Crux Math. 50 (6) 2024, 314 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Forα >0, let Sn:=2nX k=1(−1)k■k 2n■α .
We show that■ ■■ ■limn→∞Sn=1 2. We will use the higher order Euler-MacLaurin formula applied to
the function f(x) =xα. This formula tells us that for n≥m, nX i=mf(i)−ˆn mf(x)dx=f(m) +f(n) 2+1 6f′(n)−f′(m)
2!−ˆn mf′′(x)P2(x) 2!dx where P2is the periodized Bernoulli function P2(x) =B2(x− ■x■), andB2(x)
=x2−x+1 6, the second Bernoulli polynomial. Let us first rewrite Sn, splitting the sum into odd and even
indices: Sn=2nX k=1(−1)k■k 2n■α =nX j=1■j n■α −nX j=1■2j−1 2n■α =1 nαnX j=1jα−1 (2n)αnX
j=1(2j−1)α. Case 1 α= 1. Then Sn=2nX k=1(−1)kk 2n=1 nnX j=1j−1 2nnX j=1(2j−1) =1 nnX j=1j−1 nnX
j=1j+1 2n·n=1 2. Case 2 α■= 1. By choosing m= 1 in the Euler-MacLaurin formula, and estimating ˆn
1f′′(x)P2(x) 2!dx ≤Cˆn 1xα−2dx=  C α−1■ nα−1−1■ ifα >1 Clogn ifα= 1 C 1−α■ 1−nα−1■ if 0< α
< 1, we obtain nX j=1jα=■nα+1 α+ 1−1 α+ 1■ +nα+ 1 2+1 12■ αnα−1−α■ +( O(nα−1) if α >1 O(1) if 0
< α < 1.
182 Hence 1 nαnX j=1jα=■n α+ 1+1 2+α 121 n■ +O(n−α) +( O(n−1) if α >1 O(n−α) if 0 < α < 1 =n α+
1+1 2+O(1). Using, additionally, the formula 1 (2n)αnX j=1(2j−1)α=1 (2n)α 2nX j=1jα−2αnX j=1jα 
=1 (2n)α2nX j=1jα−1 nαnX j=1jα we conclude that Sn=■n α+ 1+1 2■ −■2n α+ 1+1 2−■n α+ 1+1 2■■
+O(1) =1 2+O(1).
183 Solution to problem 4953 Crux Math. 50 (6) 2024, 313 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that under the assumption a∈R\
{0,1}, the number■ ■■ ■a= 3 is the only one for which the functional equation (63) (
x+y)f(x+y)−xf(x)−yf(y) =a(yf(x) +xf(y)) equivalently (64) ( x+y)f(x+y) =f(x)(x+ay) +f(y)(y+ax) has
non-trivial solutions. These are given by■ ■■ ■f(x) =cx2forc■= 0. It is obvious that the set of solutions
of (63) is a vector space; may be the trivial one {0}. Moreover, if fis a solution then f(0) = 0: put y= 0,
and so, xf(x) =xf(x) +f(0)ax. As a■= 0,f(0) = 0. Note that if a= 3, then f(x) =x2is a solution since
(x+y)3−x3−y3−3yx2−3xy2= 0. Now suppose that fis a solution whenever a= 3. Then fis even. In fact,
putting y=−x, and using that f(0) = 0, yields 0 = f(x)(−2x) +f(−x)(2x) and so f(−x) =f(x). Next assume that
f(1) = 1 and let us introduce the auxiliary function h(x) :=f(x)−x2. Then his a solution, too, satisfying h(0)
= h(1) = 0. We show that h≡0 (yielding the assertion that f(x) =x2, and so, if f(1) = c■= 0,f(x) =cx2, and if
f(1) = 0, f≡0). To see this, take y= 1 and x→ −x. Then (−x+ 1)h(−x+ 1) + xh(−x) = 3 h(−x), hence, by
using that his even, (65) h(1−x) =3−x 1−xh(x), x■= 1. On the other hand, if x+y= 1, then by (64), 0
=h(x)(x+ 3(1−x)) +h(1−x)(1−x+ 3x). That is (66) h(1−x) =2x−3 2x+ 1h(x), x■=−1 2. Consequently, 2x−3
2x+ 1h(x) =3−x 1−xh(x). Since2x−3 2x+1=3−x 1−x⇐⇒ − 3 = 3 is impossible, h(x) = 0 for x■∈ {1,−1/2}.
By (65), 0 =h(1/2) = h(−1/2). As h(1) = 0 we are done. It remains to see that anecessarily is 3 for the
existence of nontrivial solutions. So let fbe a solution to (63), f■≡0.
184 •Puty=x. Then 2 xf(2x)−2xf(x) = 2 axf(x), implying that f(2x) = (1 + a)f(x), valid also for x= 0. •Puty=
2x. Then 3 xf(3x)−xf(x)−2xf(2x) =a(2xf(x) +xf(2x)). Hence 3f(3x) = (1 + 2 a)f(x) + (2 + a)f(2x), valid also
for x= 0. Pulling in f(2x) gives 3f(3x) = (2 a+ 1)f(x) + (2 + a)(1 + a)f(x) = (3 + 5 a+a2)f(x). •Puty= 3x. Then
4 xf(4x)−xf(x)−3xf(3x) =a(3xf(x) +xf(3x)). Hence 4f(4x) = (1 + 3 a)f(x) + (3 + a)f(3x), valid also for x= 0.
Since f(4x) =f(2(2x)) = (1 + a)f(2x) = (1 + a)2f(x), we obtain with the formula for f(3x) that 4(1 + a)2f(x) =
(1 + 3 a)f(x) +1 3(3 +a)(3 + 5 a+a2)f(x). Consequently, as f■≡0, 3(3 + 5 a+ 4a2)−(3 +a)(3 + 5 a+a2) = 0
, or equivalently a(a−1)(a−3) = 0 . Thus a= 3. Remark What happens for a= 0 or a= 1? If a= 0, then the
solutions fare those functions for which xfis additive. If a= 1 then the solutions are exactly the additive
functions: in fact, by (64), ( x+y)f(x+y) =f(x)(x+y)+f(y)(y+x). Hence, if x+y■= 0,f(x+y) =f(x)+f(y). Now let
x+y= 0. Since f(2x) =f(x+x) =f(x) +f(x) = 2 f(x) we conclude that for x■= 0, f(x) +f(−x) = f((x−1) + 1) +
f((−1−x) + 1) = f(x−1) +f(1) + f(−1−x) +f(1) =f((x−1) + (−1−x)) + 2 f(1) = f(−2) + 2 f(1) = 2( f(−1) +f(1)). In
particular, if x= 1, we deduce that f(−1) +f(1) = 0, and so f(x) +f(−x) = 0 = f(0) = f(x+ (−x)). Remark It is
quite astonishing for us that in the these two exceptional cases there are so many solutions, as the set
of additive functions is in a one to one correspondance with the Q-linear functions on the Q-vector
space Vdetermined by the real numbers.
185 Solution to problem 4951 Crux Math. 50 (6) 2024, 313 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that the common value is
■ 1 + (−1)n+1 n+ 1. Let P1(x) :=nX k=1(−1)k−1 k■n k−1■ xk−1. To be calculated is P1(1). To this end,
note that (xP1(x))′=nX k=1(−1)k−1■n k−1■ xk−1=n−1X j=0(−1)j■n j■ xj =−(−x)n+nX j=0■n j■ (−x)j=
(1−x)n−(−x)n. Hence xP1(x) =ˆx 0((1−t)n−(−t)n)dt, from which we conclude that P1(x) =1 x(n+ 1)■
−(1−x)n+1+ (−1)n+1xn+1+ 1■ . Consequently P1(1) =1 + (−1)n+1 n+ 1. To calculate P2:=nX
k=1(−1)k−1 k■n k■−1 , we use that (see below) 1 k■n k■−1 =β(k, n−k+ 1) =ˆ1 0(1−x)k−1xn−kdx=:I.
Hence P2=ˆ1 0nX k=1(x−1)k−1xn−kdx=ˆ1 0n−1X j=0(x−1)jxn−j−1dx =−ˆ1 0■ (x−1)n−xn■ dx=1 +
(−1)n+1 n+ 1. Addendum A classical formula tells us that for µ, ν∈None has Γ(µ+ 1) Γ( ν+ 1) Γ■ (µ+ 1)
+ ( ν+ 1)■=β(µ+ 1, ν+ 1) :=ˆ1 0(1−x)µxνdx=µ!ν! (µ+ν+ 1)!. Hence, for µ=k−1 and ν=n−k, I=β(k, n−k+ 1)
=(k−1)! (n−k)! n!=1 kk! (n−k)! n!=1 k■n k■−1 .
186 Solution to problem 4937 Crux Math. 50 (4) 2024, 200 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetF(y) =ˆy af(t) tdtbe a primitive of f(t)/ton
]0,∞[. Note that F(y) may be unbounded near 0 (for f≡1 for instance). But one can control its behaviour
near 0: R(y) :=y(F(a+y)−F(y))→0 asy→0. In fact, as fis continuous, |f| ≤Mon [0, a]. Hence, for 0 < y≤a,
y|F(y)| ≤Myˆa y1 tdt=Myloga−Mylogy→0, and so R(y)→0 asFis continuous at a. Using the
transformation x+y=twith y≤t≤a+y, we obtain for δ >0 close to 0, ˆb δ■ˆa 0f(x+y) x+ydx■ dy =ˆb δ■ˆa+y
yf(t) tdt■ dy=ˆb δ(F(a+y)−F(y))dy =h y(F(a+y)−F(y))ib δ−ˆb δy(F′(y+a)−F′(y))dy =
b(F(a+b)−F(b))−δ(F(a+δ)−F(δ))| {z } =R(δ)−ˆb δy■f(a+y) a+y−f(y) y■ dy −→ δ→0b(F(a+b)−F(b))−ˆb
0a+y−a a+yf(a+y)dy+ˆb 0f(y)dy = b(F(a+b)−F(b))−ˆb 0f(a+y)dy+aˆb 0f(a+y) a+ydy+ˆb 0f(y)dy f
a-periodic= s=a+yb(F(a+b)−F(b)) +aˆa+b af(s) sds = b ˆa+b af(t) tdt−ˆb af(t) tdt! +aˆa+b af(s) sds =
bˆa+b bf(t) tdt+aˆa+b af(s) sds.
187 Solution to problem 4924 Crux Math. 50 (3) 2024, 148 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that for all B >0 and n≥2■ ■■ ■1
+n B≤n−1Y j=0■ 1 +1 B+jx■ if 0≤x≤1 and ■ ■■ ■1 +n B≥n−1Y j=0■ 1 +1 B+jx■ ifx≥1. In particular, we
have equality for x= 1. Moreover, equality holds for all B > 0 and all x≥0 ifn= 1. LetLnbe the left hand
side and Rnthe right hand side. •n= 1. Then L1−R1= 1 +1 B−(1 +1 B) = 0 . •n≥2. We show the assertion
above via induction on n. So let n= 2. L2−R2= 1 +2 B−■ 1 +1 B■■ 1 +1 B+x■ =x−1 B(B+x). Hence the
assertion is true for n= 2. •n→n+1. Assume that the assertion is correct for some n∈N. Then, for
x≥1,Ln≥Rn, and so Ln+1−Rn+1= 1 +n+ 1 B−nY j=0■ 1 +1 B+jx■ ≥1 +n+ 1 B−■ 1 +n B■■ 1 +1 B+nx■
=1 B■ 1−B+n B+nx■ =1 Bn(x−1) B+nx≥0. The same estimates replacing ≥by≤show that we also get
the assertion for 0 ≤x≤1.
188 Solution to problem 4930 Crux Math. 50 (3) 2024, 149 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - First we note that for x, y≥0,xy≤1 2(x2+y2).
Hence ab+bc+ca≤1 2(a2+b2) +1 2(b2+c2) +1 2(c2+a2) =a2+b2+c2. As log xis concave, we know that
log(Pn j=1εjxj)≥Pn j=1εjlogxjwheneverPn j=1εj= 1, εj≥0. Hence b a+b+cloga+c a+b+clogb+a
a+b+clogc≤log■b a+b+ca+c a+b+cb+a a+b+cc■ ≤log■a2+b2+c2 a+b+c■ . Hence
abbcca≤■a2+b2+c2 a+b+c■a+b+c .
189 Solution to problem 4925 Crux Math. 50 (3) 2024, 148 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that the functional equation (67)
f(x+y) +xf(f(y)) =f(f(x)) +f(y) +axy admits a solution if and only if a∈ {0,1}. Ifa= 1, then the identity is the
only solution and ifa= 0, then the zero function is the only solution. Putx=y= 0. Then f(0) + 0 = f(f(0)) +
f(0) implies that f(f(0)) = 0. Now put y= 0 in (67). Then (68) f(x) =f(f(x)) +f(0). This yields the new
equation (69) f(x+y) +x■ f(y)−f(0)■ =f(x)−f(0) + f(y) +axy. Now put x= 1. Then (70) f(1 +y) +f(y)−f(0) =
f(1)−f(0) + f(y) +ay, from which we conclude that f(y+ 1) = f(1) + ay, or in other words, (71) f(u) =f(1) +
a(u−1) =: au+b, that is, fis linear-affine. Since f(f(0)) = 0, we have a(a·0 +b) +b= 0, and so ab+b= 0.
Thus b= 0 or a=−1. Due to (68), ax+b=a(ax+b) +b+b. That is ax=a2x+ (ab+b) =a2x, from which we
deduce a∈ {0,1}. Hence, as a■=−1,b= 0. Consequently, f(x) =x(ifa= 1) orf≡0 ifa= 0. It is straightforward
to check that these are actually solutions.
190 Solution to problem 4929 Crux Math. 50 (3) 2024, 149 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that■ ■ I:=ˆ1 0log(1 +√ 1−u2) 1
+udu=π2 24∼0.4112335167 ···. In fact, I=ˆ1 0log■ 1 +p 1−u2■∞X n=0(−1)nundu=∞X n=0(−1)nˆ1 0log■
1 +p 1−u2■ undu, where the interchanging lim n´ Sn=´ limSnis allowed as the partial sums Sn=nX
j=0log■ 1 +p 1−u2■ (−1)juj are bounded by the L1[0,1]-functionlog(1+√ 1−u2) 1+u. Now put In:=ˆ1
0log■ 1 +p 1−u2■ undu, and use partial integration with f(u) = log■ 1 +√ 1−u2■ andg′(u) =un. Note that
g(u) = un+1 n+1and f′(u) =−u√ 1−u2 1 +√ 1−u2=−u(1−√ 1−u2)√ 1−u2u2=−1 u√ 1−u2+1 u. Hence In =−1
n+ 1ˆ1 0■ un−un √ 1−u2■ du = u=sint−1 (n+ 1)2+1 n+ 1ˆπ/2 0sinnt dt:=−1 (n+ 1)2+Jn. Method 1 . By
Lemma 14 below, I=−∞X n=0(−1)n 1 (n+ 1)2+∞X n=0(−1)nJn=−π2 12+∞X m=0J2m−∞X m=0J2m+1
=−π2 12+∞X m=0■2m m■ (2m+ 1)4m·π 2−∞X m=04m (2m+ 1)(2 m+ 2)■2m m■ =−π2 12+π2 4−π2
8=π2 24. Here we have used the Taylor series (see e.g. [44] and [45]): arcsin x=∞X n=0■2n n■
4nx2n+1 2n+ 1
191 arcsin2x= 2∞X n=04nx2n+2 (2n+ 1)(2 n+ 2)■2n n■ =∞X n=022n+1(n!)2 (2n+ 2)!x2n+2=∞X
n=122n−1 n2■2n n■x2n. evaluated at x= 1. Lemma 14. LetIn:=´π/2 0(sinx)ndx. Then I0=π/2,I1= 1and
for n∈N∗, (1)I2n=1 2·3 4·5 6···2n−1 2nπ 2=(2n)! 4n(n!)2·π 2=(2n n) 4n·π 2. (2)I2n+1=2 3·4 5·6 7···2n
2n+1=4n(n!)2 (2n+1)!=4n (2n+1)(2n n). Proof. (1)I2n=2n−1 2nI2n−2forn∈N∗andI0=π 2, because
2nI2n−(2n−1)I2n−2=ˆπ/2 0(sinx)2n−2■ 2nsin2x−(2n−1)■ dx =−ˆπ/2 0(sinx)2n−2■ (2n−1) cos2x−sin2x■
dx =−■ (sinx)2n−1cosx■π/2 0= 0. (2)I2n+1=2n 2n+1I2n−1forn∈N∗andI1= 1, because (2n+
1)I2n+1−2nI2n−1=ˆπ/2 0(sinx)2n−1■ (2n+ 1) sin2x−2n■ dx =−ˆπ/2 0(sinx)2n−1■ 2ncos2x−sin2x■ dx
=−■ (sinx)2ncosx■π/2 0= 0. ■ Next we note that for 0 ≤x <1, arcsin x= arctan■ x√ 1−x2■ . In fact
Lemma 15. For0≤x <1we have (1) arcsin x= arctan■x√ 1−x2■ . (2)arcsin x√ 1−x2=∞X n=0x2n+1ˆπ/2
0(sint)2n+1dt. (3) (arcsin x)2=∞X n=0 ´π/2 0(sint)2n+1dt n+ 1! x2n+2. Proof. (1) Obvious, as tan(arcsin
x) = sin(arcsin x)/(cos arcsin x). (2) As in [44]): we shall use that for 0 ≤y <1 y 1−y2=1 2■1 1−y−1 1 +y■
=1 2∞X n=0(yn−(−1)nyn) =∞X n=0y2n+1. Now let I(t) :=1√ 1−x2arctanxsint√ 1−x2. Then, with 0 < x < 1,
arcsin x√ 1−x2=I■π 2■ −I(0) =ˆπ 2 0∂I ∂tdt=ˆπ 2 0xcost 1−x2cos2tdt =ˆπ 2 0∞X n=0(xcost)2n+1dt=∞X
n=0x2n+1ˆπ 2 0(cost)2n+1dt.
192 HereP´ =´P, as all the terms are positive. (3) Just use thatd dx(arcsin x)2= 2arcsin x√ 1−x2. ■
Method 2 Note that I=−∞X n=0(−1)n 1 (n+ 1)2+∞X n=0(−1)nJn. Consider the auxiliary function h(x) =∞X
n=0 ˆπ 2 0(sint)n n+ 1dt! (−1)nxn+1. Then h′(x) =∞X n=0 ˆπ 2 0(sint)ndt! (−1)nxn =ˆπ 2 0∞X
n=0(−xsint)ndt=ˆπ 2 01 1 +xsintdt(∗)=arccos x√ 1−x2 Hence, h(1) = h(1)−h(0) =ˆ1 0arccos x√ 1−x2dx=■
−1 2arccos2x■1 0=π2 8. We conclude that I=−π2 12+π2 8=π2 24. To prove (*), just use the
transformation tan( t/2) = yto get ˆπ 2 01 1 +xsintdt= 2ˆ1 01 1 +y2+ 2xydy=2 1−x2ˆ1 01 1 +(y+x)2
1−x2dy = 2"arctanx+y√ 1−x2√ 1−x2#1 0= 2arctanx+1√ 1−x2−arctanx√ 1−x2√ 1−x2 =arccos x√ 1−x2.
The latter is verified by calculating the derivatives of the numerators and by using that for x= 0, 2 arctan
1 = π/2 = arccos 0. Method 3 f(1) =ˆ1 0ˆπ 2 01 1 +ξsintdtdξ =ˆπ 2 0ˆ1 01 1 +ξsintdξdt =ˆπ 2 0log(1 + sin
t) sintdt =ˆ1 0log(1 + y) yp 1−y2dy =π2 8. From [44]
193 Solution to problem 4920 Crux Math. 50 (2) 2024, 84 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let I:=ˆ1 0log(1 + xk+x2k+···+xnk) xdx. We
show that ■ ■ I=π2 6n k(n+ 1). For the proof, we use the power series representation log(1−x) =−∞X
j=1xj j,|x|<1.
194 So, if 0 < x < 1 we have f(x) :=log(1 + xk+x2k+···+xnk) x=1 xlog■1−xk(n+1) 1−xk■ =−∞X
j=1xk(n+1)j−1 j+∞X j=1xkj−1 j. Hence, a primitive is given by −∞X j=1xk(n+1)j j2k(n+ 1)+∞X j=1xkj j2k.
We conclude that I=ˆ1 0f(x)dx=∞X j=1■ −1 k(n+ 1)1 j2+1 k1 j2■ =π2 6■1 k−1 k(n+ 1)■ =π2 6n k(n+ 1).
195 Solution to problem 4918 Crux Math. 50 (2) 2024, 83 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let 0≤a < b . a) Suppose that a≤x < b .
Then for a≤t≤bwe have 1 ≥t/b, and so, for λ >1, 0≤I(λ) :=λx2 ´b aλt2dt≤bλx2 ´b atλt2dt=2b(logλ)λx2
λb2−λa2=2b(logλ)λx2−b2 1−λa2−b2. Since (logλ)e(logλ)(x2−b2)=logλ e(logλ)(b2−x2)→0 asλ→ ∞ , we
have that lim λ→∞I(λ) = 0. b) Suppose that x=b. Then for a≤t≤bwe have 1 ≤t/aand so I(λ) =λb2 ´b
aλt2dt≥aλb2 ´b atλt2dt=2a(logλ)λb2 λb2−λa2 ≥2a(logλ)λb2 λb2 = 2alogλ → ∞ asλ→ ∞ .
196 Solution to problem 4915 Crux Math. 50 (2) 2024, 82 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that a= log 2 , b=−1 2−log 2 and
c= log 2 +5 4; that is ■ ■ lim n→∞■ n3Sn−■ (log 2) n2+■ −1 2−log 2■ n+ log 2 +5 4■■ = 0. Sn:=∞X
k=1(−1)k+1 k(k+n+ 1)=1 n+ 1∞X k=1(−1)k+1n+ 1 + k−k k(k+n+ 1) =1 n+ 1∞X k=1(−1)k+1■1 k−1 n+k+
1■ =1 n+ 1∞X k=1 (−1)k+11ˆ 0xk−1dx−(−1)k+11ˆ 0xn+kdx  (∗)=1 n+ 11ˆ 0■∞X k=1(−1)k+1xk−1−∞X
k=1(−1)k+1xn+k■ dx =1 n+ 11ˆ 0 1 1 +x−xn+1∞X k=1(−1)k+1xk−1! dx=1 n+ 11ˆ 01−xn+1 1 +xdx
=log(2) n+ 1−1 n+ 11ˆ 0xn+1 1 +xdx =log(2) n+ 1−1 n+ 1 xn+2 n+ 21 1 +x 1 0+1 n+ 21ˆ 0xn+2 (1
+x)2dx  =log(2) n+ 1−1 2(n+ 1)( n+ 2)−1 (n+ 2)( n+ 1)1ˆ 0xn+2 (1 +x)2dx =log(2) n+ 1−1 2(n+ 1)( n+
2)−1 (n+ 2)( n+ 1) xn+3 n+ 31 (1 +x)2 1 0+2 n+ 31ˆ 0xn+2 (1 +x)3dx  =log(2) n+ 1−2n+ 7 4(n+ 1)( n+
2)( n+ 3)−2 (n+ 1)( n+ 2)( n+ 3)1ˆ 0xn+2 (1 +x)3dx.
197 Here the interchanging of´P=P´ in (∗) is possible since the partial sums NX k=1(−1)k+1(xk−1−xn+k)
are bounded. Hence14 n3Sn=n3 n+ 1log(2) −n3(2n+ 7) 4(n+ 1)( n+ 2)( n+ 3)−2n3 (n+ 1)( n+ 2)( n+ 3)|
{z } ≤21ˆ 0xn+2 (1 +x)3dx |{z } ≤1/(n+3) = (n2−n+ 1 + O(1/n)) log(2) −n 2+5 4+O(1/n) +O(1/n) =n2log(2)
+ n■ −1 2−log(2)■ + log(2) +5 4+O(1/n) =an2+bn+c+O(1), where the asymptotics are obtained by
calculating the partial fraction decomposition of the rational functions in n. 14HereOandOdenote the
Landau symbols: O(1/n) is a function n7→h(n) satisfying|h(n)| 1/n≤C andO(1) is a function n7→g(n)
with lim n→∞g(n) = 0. In particular, O(1/n) implies O(1).
198 Solution to problem 4914 Crux Math. 50 (2) 2024, 82 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -■ ■■ ■We show that the identity is the only
monotonically increasing solution on [0 ,∞[. First we note that f(x) =xtrivially satisfies the functional
equation on [0 ,∞[. Now suppose thatfis a solution and that fis increasing on [0 ,∞[. •Puty= 0,
respectively x= 0. Then f(x2+ 1) = xf(x) +f(0) + 1 and f(y+ 1) = f(y) + 1. Hence, with y=x2, respectively
y= 0, f(x2+ 1) = f(x2) + 1 and f(1) = f(0) + 1 and so (72) f(x2) + 1 = f(x2+ 1) = xf(x) +f(1). This implies
that, with x= 1, f(1) + 1 = 1 ·f(1) + f(1) = 2 f(1). Hence f(1) = 1 and so f(0) = 0. Consequently (73) f(x2)
=f(x2+ 1)−1 =xf(x) +f(1)−1 =xf(x). •Next we note that fis right-continuous at x1= 1. In fact, since fis
increasing, fis bounded in a right-neighborhood of x0= 015. Hence, if x→0, we have that xf(x)→0.
Consequently, by the second identity in (72) lim x→0f(x2+ 1) = f(1) = 1 . •Via induction we obtain from
f(x2) = xf(x), or equivalently f(x) =√x f(√x) (just replace x2byx), that f(2n√x) 2n√x=f(x) x. Now2n√x→1 for
x >0 and2n√x≥1 for x≥1. Thus, the right-continuity of fatx1= 1 yields that f(1) 1=f(x) x, from which we
conclude that f(x) =xforx≥1. Now let 0 ≤x≤1. Then x+ 1≥1 and so, due to f(y+ 1) = f(y) + 1, x+ 1 = f(x+ 1)
= f(x) + 1. Consequently f(x) =x, too. 15For this, it is important that fis defined at 0.
199 Remark 1 Our proof shows that instead of fbeing increasing, we may have assumed merely thatfis
bounded in a right neighborhood of the origin. Remark 2 A small modification of the proof (see below)
shows that any solution to the functional equation (E) f(x2+y+ 1) = xf(x) +f(y) + 1,(x, y≥0) actually is
additive on [0 ,∞[; that is satisfies f(u+v) =f(u)+f(v) with f(0) = 0 and f(1) = 1. Thus we obtain from the
well known fact on the Cauchy functional equation (restricted to the non-negative reals) that actually
every measurable solution of (E) coincides with the identity. In fact, f(x2)(73)=xf(x) and f(y+ 1) = f(y) + 1
imply that (E) becomes (74) f(x2+y+ 1) = f(x2) +f(y) + 1 = f(x2) +f(y+ 1),(x, y≥0), and so f(u+v) =f(u) +f(v)
for u, v≥0, due to the following reason: since for y≥1,f(y−1) = f(y)−1, f(u) +f(v) = f(u) +f■ (v+ 1)−1■ =f(u)
+f(v+ 1)−1 = (74)f(u+ (v+ 1))−1 =f(u+v) + 1−1 = f(u+v).
200 Solution to problem 4913 Crux Math. 50 (2) 2024, 82 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Leta∈Rand, for f∈C(R), let If(x, a) :=ˆxa
0f(tx2a−1)dt. We show that■ ■■ ■f(7) = 42 whenever If(x, a) = 3 ax4. Proof. Using for x■= 0 the linear
substitution t→uwith u:=tx2a−1anddt=ax−2du, we obtain If(x, a) = ax−2ˆx3 0f(u)du. Now If(x, a) = 3
ax4if and only ifˆx3 0f(u)du= 3x6. Differentiating yields 3x2f(x3) = 18 x5, equivalently f(x3) = 6 x3.
Asx7→x3is a bijection from RontoR, we obtain f(u) = 6 u. Conversely, it is straightforward to check that
this fsatisfies for every athe given integral equation. So f(7) = 42 independently of a.
201 Solution to problem 4910 Crux Math. 50 (1) 2024, 38 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - It is sufficient to consider the case n= 0,
otherwise write ˆSm−Sn x2dx=ˆSm−1 x2dx+ˆ1−Sn x2dx, where S=sinx x.Since|S| ≤1, we see that´∞
1Sm−1 x2dxconverges. Now use that 1−S=x2 3!−x4 5!± ··· =x2■1 6+O(x)■ asx→0, and |Sm−1|=|S−1|
m−1X j=0Sj ≤m|S−1|, to conclude thatˆ1 0Sm−1 x2dxconverges, too. Hence I(m) :=ˆ∞ 0Sm−1 x2dx
converges. Next we write J:=Sm−1 x2=(sinx)m−xm xm+2=:f(x) xm+2. Now we apply Apostol’s method
(see [47]). Integration by parts´ uv′=uv−´ u′vwith u=fand v′=x−m−2yields: I(m) =1 m+ 1ˆ∞ 0f′(x) xm+1dx,
since lim x→0f(x) xm+1= 0 (and lim x→∞f(x) xm+1= 0) because f(x) xm+1 ≤m|S−1| x=mx■1 6+O(x)■
asx→0. Similarily, since 0 is a zero of order 1 of the analytic16function J(z) :=(sinz)m−zm zm+1, we
have that for all j= 0,1, . . . , m lim x→0f(j)(x) xm+1−j= 0. Hence, by repeating this procedure another
m-times, we obtain I(m) =1 (m+ 1)!ˆ∞ 0f(m+1)(x) xdx. Now f(m+1)(x) =dm+1 dx(sinx)m−0. Next we
”linearize” the sinus-power: 16Note that J(z) =z(sinz z)m−1 z2=z■ −1 3!+z2 5!+···■ R(z), where lim
z→0R(z) = lim z→0m−1X j=0■sinz z■j = m.
202 (sinx)m=■eix−e−ix 2i■m =■1 2i■mmX j=0(−1)m−j m j! eijxe−i(m−j)x = (−1)m1 (2i)mmX j=0(−1)−j m
j! eix(2j−m). Since the ”constant” term (appearing for j=m/2 when mis even) is annihilated by the
derivative, we find dm+1 dx(sinx)m= (−1)m(−i)m+11 (2i)mX 0≤j<m 2(m−2j)m+1(−1)j m j! e−ix(m−2j)
+(−1)mim+11 (2i)mX m 2<j≤m(2j−m)m+1(−1)j m j! eix(2j−m). As the left hand side is real, we may take
the real part on the right hand side and get (by observing Reiz=−Imz) dm+1 dx(sinx)m=−1 2mX 0≤j<m
2(m−2j)m+1(−1)j m j! sin((m−2j)x) +(−1)m+11 2mX m 2<j≤m(2j−m)m+1(−1)j m j! sin((2 j−m)x). Finally,
asˆ∞ 0sin(px) xdx=ˆ∞ 0sin(x) xdx=π 2 whenever p >0, we deduce that I(m) =1 (m+ 1)!ˆ∞ 0f(m+1)(x) xdx
=π 21 (m+ 1)!" −1 2mX 0≤j<m 2(m−2j)m+1(−1)j m j! +(−1)m+11 2mX m 2<j≤m(2j−m)m+1(−1)j m j!#
which surely is a rational multiple of π. Making in the second summand the substitution k=m−j, then we
obtain I(m) =−π 2m(m+ 1)!X 0≤j<m 2(m−2j)m+1(−1)j m j! . For instance I(1) =−π 4, I(2) =−π 3, I(3) =−13π
32.
203 Solution to problem 4909 Crux Math. 50 (1) 2024, 38 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - First we note that Pnis continuous on [0
,∞[,Pn(0) = 1 and that lim n→∞Pn(x) =∞. Now P′ n(x) = x(x−1)2n■ (2n+ 3)x−(4n2+ 6n+ 4)■
=x(x−1)2n(2n+ 3)■ x−■ 2n+4 2n+ 3■■ . Hence Pnis strictly decreasing on■ 0,2n+4 2n+ 3■ and strictly
increasing on■ 2n+4 2n+ 3,∞■ . Com- bining all this, and thanks to the intermediate value theorem, we
deduce there exists a unique xn∈]0,∞[ with Pn(xn) = 1. Next we discuss the assymptotics of the
sequence ( xn). Since xn>2n+4 2n+3, we see that xn→ ∞ . •As 1 =Pn(xn) = ( xn−1)2n+1(x2 n−(2n+
1)xn−1) =xn(xn−1)2n+1(xn−2n−1−x−1 n) we get (75)1 xn(xn−1)2n+1+1 xn=xn−2n−1. Butxn→ ∞ . Thus
xn−2n−1→0, from which we conclude that■ ■■ ■xn−2n→1 . In particular, xn n−2 =xn−2n n→0. •By
(75), n(xn−2n−1) =n xn+n xn(xn−1)2n+1 =n xn■ 1 +1 (xn−1)2n+1■ →1 2(1 + 0) =1 2.
204 Solution to problem 4905 Crux Math. 50 (1) 2024, 37 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that
■ (x, y) =■π 12,5π 12■ , or in terms of degrees■ ■■ ■15■and 75■. We may assume that 0 ≤x≤y≤π/2
and x+y=π/2. Now tany= tan■π 2−x■ = cot( x) =1 tanx. So we have to solve for t= tan xthe equation
(76) t+t2+t3+1 t+1 t2+1 t3= 70 (or equivalently 1 + t+t2−70t3+t4+t5+t6= 0). Such symmetric equations
are solved using the substitution s:=t+ 1/t. Now s2= (t+1 t)2=t2+1 t2+ 2; hence t2+1 t2=s2−2. Moreover,
s3=■ t+1 t■3 =t3+ 3t+3 t+1 t3 and so t3+1 t3=s3−3s. This yields the equation s+s2−2 +s3−3s= 70, or
equivalently s3+s2−2s−72 = 0. As s= 4 is a solution, we obtain the factorization 0 = ( s−4)(s2+ 5s+ 18) =
( s−4)■■ s+5 2■2 +47 4■ . Sos= 4 is the only real solution. The equation 4 = t+1 tnow is equivalent to
t2−4t+ 1 = 0, which has 2±√ 3 as solutions. Now we have to calculate the values xfor which tan x= 2±√
3. As is well known, arctan(2 −√ 3) = π/12 and arctan(2+√ 3) = 5 π/12. This can be verified by using the
formulas sinx=r 1−cos 2x 2and cos x=r 1 + cos 2 x 2.
205 Solution to problem 4904 Crux Math. 50 (1) 2024, 37 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - This problem has no solution17. In fact,
suppose that x+yis prime whenever xandyare prime, y≤x. Since the sum of two odd prime numbers is
even, it cannot be prime. Hence y= 2. Let n:=x 18. Then we have to discuss the property 4 + nnis
divisible by 2 + n. 4 +nn 2 +n=4 + (( n+ 2)−2)n 2 +n=4 +nX j=0(n+ 2)j n j! (−1)n−j2n−j n+ 2 =4 +
(−1)n2n+nX j=1(n+ 2)j n j! (−1)n−j2n−j n+ 2 =4 + (−1)n2n n+ 2+nX j=1(n+ 2)j−1 n j! (−1)n−j2n−j =:4 +
(−1)n2n n+ 2+m, where m∈Z(note that the binomial coefficients belong to N). Ifn≥3 is odd, then 2 + nis
odd and therefore 2 + ncannot divide (in Z) the even number 4 + ( −1)n2n. Hence the primeness of
nimplies that n= 2. Since x+y= 2 + 2 = 4 is not prime, the pair (2 ,2) is not a solution either. 17Under the
usual assumption that the number 1 is not considered as a prime number. 18The symbol xfor a natural
number hurts my eyes ©.
206 Solution to problem 4903 Crux Math. 50 (1) 2024, 37 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let S:=∞X n=1 −1 4n+∞X k=0(−1)k 2n+
2k−1! which is the concise form of the sum in the problem. We claim that
■ S=log 2 2+π 8. Solution Forn≥1, let Fn(x) :=∞X k=0(−1)k 2n+ 2k−1x2n+2k−1 be the generating
function. Then Fn(x) converges for 0 ≤x≤1 (Leibniz rule for the alternating series atx= 1), and by Abel’s
rule, Fnis continuous on [0 ,1]. Now F′ n(x) =∞X k=0(−1)kx2n+2k−2=x2n−2 1 +x2. Since Fn(0) = 0, we
obtain Fn(1) =∞X k=0(−1)k 2n+ 2k−1=ˆ1 0x2n−2 1 +x2dx. Note that S=P∞ n=1■ Fn(1)−1 4n■ . Partial
integration´ u′v=uv−´ uv′with u′=x2n−2andv= (1 + x2)−1 yields ˆ1 0x2n−2 1 +x2dx =x2n−1 2n−11 1 +x2
1 0+1 2n−1ˆ1 0x2n−12x (1 +x2)2dx =1 2(2n−1)+2 2n−1ˆ1 0x2n (1 +x2)2dx. Hence, by using thatP´
=´Pas all factors are positive, and the fact that ∞X n=1■1 2(2n−1)−1 4n■ = lim N→∞NX n=1■1
2(2n−1)−1 4n■ =1 2lim N→∞2NX j=1(−1)j−1 j=log 2 2, we obtain S=∞X n=1■1 2(2n−1)−1 4n■ + 2ˆ1 0
∞X n=1x2n−1 2n−1! x (1 +x2)2dx =log 2 2+ˆ1 0■ log■1 +x 1−x■x (1 +x2)2■ dx =:log 2 2+ˆ1 0I(x)dx. To
calculate a primitive of I(x), we use partial integration with u= log■ 1+x 1−x■ andv′=x (1+x2)2. Hence
207 ˆ I(x)dx =−1 21 1 +x2log■1 +x 1−x■ +ˆ1 1−x4dx =−1 21 1 +x2log■1 +x 1−x■ +1 4log■1 +x 1−x■
+1 2arctan x =1 4x2−1 x2+ 1log■1 +x 1−x■ +1 2arctan x =:R(x). Hence ˆ1 0I(x)dx= lim x→1R(x)−R(0)
=π 8.
208 Solution to problem 4900 Crux Math. 49 (10) 2023, 541 Raymond Mortini, Rudolf Rupp - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We first show that (77) Hp+Hq≤1 +Hpqfor
1≤p≤q. In fact, if 1 = p=q, then nothing has to be shown. So let us assume that q≥2. Then 1 +Hpq= 1 +
Hp+■1 p+ 1+···+1 2p■ +■1 2p+ 1+···+1 3p■ +···+■1 (q−1)p+ 1+···+1 qp■ ≥1 +Hp+p·1 2p+p·1
3p+···+p·1 qp =Hp+Hq. Now let 1 ≤m≤n≤p≤q(of course this is without loss of generality). Then by (77),
(Hm+Hn) + (Hp+Hq)≤(1 +Hmn) + (1 + Hpq) = 2 + Hmn+Hpq ≤2 + 1 + H(mn)(pq)= 3 + Hmnpq. Remark
1 More generally, one can show that nX j=1Hnj≤(n−1) +HQn j=1nj. Remark 2 Solutions to the special
case (77) above also appeared in Amer. Math. Monthly 56 (2) 1949, 109-110, Problem E819 Euler’s
constant. Remark 3 A different proof of Hp+Hq≤1+Hpq,p, q∈N={1,2, . . .}, can be given via induction
onq: for q= 1, Hp+H1=Hp+ 1≤1 +Hp·1. Now for q→q+ 1, we use that Hp(q+1) =Hpq+1 pq+ 1+···+1
pq+p≥Hpq+p pq+p=Hpq+1 1 +q. Hence Hp+Hq+1=Hp+Hq+1 q+ 1≤1 +Hpq+1 q+ 1≤1 +Hp(q+1).
209 Solution to problem 4896 Crux Math. 49 (10) 2023, 540 Raymond Mortini, Rudolf Rupp - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that all solutions f:R→Rof the
functional equation (78) f■ f(x) +yf(z)−1■ +f(z+ 1) = zf(y) +f(x+z) (x, y, z ∈R) are given by■ ■■ ■f≡0
orf(x) =xforx∈R. Obviously f≡0 and the identity are solutions. Now let fbe a solution to (78) with f(y0)■=
0 for some y0∈R. We claim that fis surjective. In fact, put x= 1 and y=y0in (78). Then, for all z∈R. (79)
f■ f(1) + y0f(z)−1■ +f(z+ 1) = zf(y0) +f(1 +z)⇐⇒ f■ f(1) + y0f(z)−1■ =f(y0)z. As the function
z7→f(y0)zis surjective, the function z7→f■ f(1) + y0f(z)−1■ is surjective, too. Hence x7→f(x) is
surjective. Next put y= 0 and z= 0 in (78). Then, for all x∈R (80) f■ f(x)−1■ +f(1) = f(x). Now put u:=f(x).
Note that if xruns through R, the surjectivity of fimplies that uruns through R, too. In particular, f(u−1) =
u−f(1) for every u∈Rand so, with v:=u−1,f(v) =v+ 1−f(1). Now v= 1 yields that f(1) = 2 −f(1) and so f(1) =
1. Hence f(v) =vfor every v∈R.
210 Solution to problem 4894 Crux Math. 49 (10) 2023, 539 Raymond Mortini, Rudolf Rupp - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that ■ ■ ∞X n=1Hn−1Hn+1
n(n+ 1)= 3. For the proof we shall decompose the series into several telescoping series. Hn−1Hn+1
n(n+ 1)=Hn−1Hn+1 n−Hn−1Hn+1 n+ 1=Hn−1Hn+1 n−(Hn−1 n)(Hn+2−1 n+2) n+ 1 =Hn−1Hn+1
n−HnHn+2−1 nHn+2−1 n+2Hn+1 n(n+2) n+ 1 =■Hn−1Hn+1 n−HnHn+2 n+ 1■ +Hn+2 n(n+ 1)+Hn (n+
1)( n+ 2)−1 n(n+ 1)( n+ 2). Now Hn+2 n(n+ 1)= Hn+1 n+1 n+2 n! −Hn+2 n+ 1=■Hn+1 n−Hn+2 n+ 1■ +1
n(n+ 2) and Hn (n+ 1)( n+ 2)=■Hn−1 n+ 1+1 n n+ 1■ −Hn n+ 2=■Hn−1 n+ 1−Hn n+ 2■ +1 n(n+ 1).
Moreover, 2 n(n+ 2)=1 n−1 n+ 2=■1 n−1 n+ 1■ +■1 n+ 1−1 n+ 2■ 2 n(n+ 1)( n+ 2)=1 n(n+ 1)−1 (n+ 1)(
n+ 2). Hence Hn−1Hn+1 n(n+ 1)=■Hn−1Hn+1 n−HnHn+2 n+ 1■ +■Hn+1 n−Hn+2 n+ 1■ +■Hn−1 n+
1−Hn n+ 2■ +3 2■1 n−1 n+ 1■ +1 2■1 n+ 1−1 n+ 2■ −1 2■1 n(n+ 1)−1 (n+ 1)( n+ 2)■ . Consequently
lim N→∞NX n=1Hn−1Hn+1 n(n+ 1)=■ H0H2−lim N→∞HNHN+2 N+ 1■ +■ H2−lim N→∞HN+2 N+ 1■
+■H0 2−lim N→∞HN N+ 2■ +3 2+1 4−1 4. Since γ= lim n→∞(Hn−logn), we have that lim
N→∞HNHN+2 N+1= 0 as well as lim N→∞HN+2 N+1= 0 and limN→∞HN N+2= 0. Consequently, by
noticing that H0= 0, ∞X n=1Hn−1Hn+1 n(n+ 1)=H2+3 2= 3. We thank Roberto Tauraso for confirming
the result via Maple and wolframalpha.com, the latter though using a different representation:
211 Calculating the sum beginning with index n= 2, this software obtains the wrong result (the actual
sum is of course 3, too as the first summand(H1−1)(H1+0.5) 1∗2= 0): Very strange, too, is that the
software does not give the correct value of the original sum but only very rough approximations:
212 Solution to problem 4893 Crux Math. 49 (10) 2023, 539 Raymond Mortini, Rudolf Rupp - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The statement of the problem is a bit
ambiguous, as problems arise for x= 0. Note that for 0< x≤1 and 1 ≤t≤1/xone has 0≤x2t≤x21 x=x≤1, so
that the integral´1/x 1f(x2t)dtis well defined for 0 < x≤1. Moreover, for −1≤x < 0 and 1/x≤t≤0, one has
−1≤x=x21 x≤x2t≤0, and so the integral´1/x 1f(x2t)dt=−´1 0f(x2t)dt−´0 1/xf(x2t)dtis well defined for −1≤x
<0, too. Ifx= 0, though, then the symbol´1/x 1is not well defined as 1 /0+=∞and 1 /0−=−∞. Actually no
function can be a solution to x2+´1/x 1f(x2t)dt= 1 also at this point, as´±∞ 1f(0)dtis divergent if f(0)■= 0,
and if f(0) = 0, then´±∞ 10dt= 0 but 0 + 0 ■= 1. Thus we need to interprete at x= 0 this functional
equation as lim x→0 x2+ˆ1/x 1f(x2t)dt! = 1. We show that ■ ■■ ■f(x) = 2 x is the only continuous
function f: [−1,1]→Rsatisfying for x∈[−1,1]\{0}=:Xthe integral equation (81) x2+ˆ1/x 1f(x2t)dt= 1, and
(82) lim x→0 x2+ˆ1/x 1f(x2t)dt! = 1. Proof . For x∈Xandf∈C[−1,1], let F(x) :=x2+´1/x 1f(x2t)dt. By the
change of variable u:=x2t we obtain x2F(x)−x4=ˆx x2f(u)du. SoF≡1 onXif and only if´x
x2f(u)du=x2−x4onX, hence also on [ −1,1]. Hence, if f∈C[−1,1] is a solution on Xto (81) then, by taking
derivatives, f(x)−f(x2) = 2 x−4x3on [−1,1]. From this, we guess that f(x) = 2 x. To this end, let g(x) :=
f(x)−2x,x∈[−1,1]. Then g∈C[−1,1] and g(x) = 2 xg(x2) for x∈[−1,1]. Since g(−x) =−2xg(x2), it suffices to
determine gforx∈[0,1]. By induction, for each x∈[0,1], g(x) = 2nx2n−1g(x2n). Now, for 0 ≤x <1 we may
let n→ ∞ and conclude (due to the continuity of gat 0 and mym→0 for 0≤y <1) that g(x) = lim
n→∞2nx2n−1g(0) = 0. As gis continuous at 1, we deduce that g≡0 on [0 ,1], hence on [ −1,1], and so
f(x) = 2 xforx∈[−1,1] whenever fsatisfies (81) on X. Now it is straightforward to show that 2 xalso
satisfies (82), that is lim x→0 x2+ˆ1/x 1f(x2t)dt! = 1.
213 Solution to problem 4889 Crux Math. 49 (9) 2023, 491 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Consider the functional equation (83)1
y−xˆy xf(g(t))dt=f■x+y 2■ , x■=y. We show that all nonconstant continuous functions
f:R→Randg:R→Rsatisfying (1) are given by■ ■■ ■f(x) =ax+bfora■= 0, b∈Randg(x) =x. In fact, for
fixed x, lety→x. Then (1) implies (for instance via l’Hospital’s rule) that (84) f(g(x)) =f(x) for every x∈R.
Hence (85)1 y−xˆy xf(t)dt=f■x+y 2■ , x■=y. Next we observe that any continuous fsatisfying (85),
necessarily is C∞. In fact, for all x, ˆx+1 x−1f(t)dt= 2f(x). As the function on the left obviously is
differentiable by the fundamental theorem of calculus, we do have the same for the function on the
right. A calculation gives f(x+ 1)−f(x−1) = 2 f′(x). Hence, the continuity of fimplies that fis continuously
differentiable. Inductively, we now conclude that f isC∞. Now let H(x, y) :=ˆy+x y−xf(t)dt. Then, by
assumption, H(x, y) = 2 xf■y+x+ (y−x) 2■ = 2x f(y). Therefore, Hy=f(y+x)−f(y−x) = 2 xf′(y), and
Hx=f(y+x) +f(y−x) = 2 f(y). Addition yields f(y+x) =xf′(y) +f(y). Now f′(y+x) =∂ ∂yf(y+x) =xf′′(y) +f′(y)
f′(y+x) =∂ ∂xf(y+x) =f′(y). Hence, for all x, we must have xf′′(y) = 0. As f′′is continuous, f′′≡0, and so f(x)
=ax+bwith a■= 0 (since fis not constant) and b∈R. Moreover, as we know from (84) that f(g(x)) = f(x),
the injectivity of the linear function fimplies that gis the identity. We note that an equivalent for (85), the
mid-point mean value theorem for derivatives, was dealt with in [48].
214 Solution to problem 4862 Crux Math. 49 (7) 2023, 375 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let Lm(n) :=1 2n1 nmnX k=0 m+k k! m+n+
1 n−k! . We prove that
■ lim n→∞Lm(n) =2 m!. To this end, note that m+k k! m+n+ 1 n−k! =(m+k)! m!k!(m+n+ 1)! (m+k+ 1)!(
n−k)! =(m+n+ 1)! m!1 k!(m+k+ 1)( n−k)!=(m+n+ 1)! m!n! n k! 1 m+k+ 1. Hence nX k=0 m+k k! m+n+ 1
n−k! =(m+n+ 1)! m!n!nX k=0 n k! 1 m+k+ 1. Put f(x) :=nX k=0 n k! 1 m+k+ 1xm+k+1. Then f′(x) =nX k=0
n k! xm+k=xm(1 +x)n. Consequently, as´1 0f′(x)dx=f(1)−f(0) and f(0) = 0, nX k=0 m+k k! m+n+ 1 n−k!
=(m+n+ 1)! m!n!ˆ1 0xm(1 +x)ndx, and so (86) Lm(n) =1 nm(m+n+ 1)! m!n!ˆ1 0xm■1 +x 2■n dx. Case 1
Ifm= 0, then L0(n) =(n+ 1)! n!ˆ1 0■1 +x 2■n dx= (n+ 1)" 2 n+ 1■1 +x 2■n+1#1 0= 2■ 1−1 2n+1■ →2.
Case 2 m≥1. We claim that Rn:= (n+ 1)ˆ1 0xm■1 +x 2■n dx→2 as n→ ∞ . In fact, partial integration
yields
215 Rn=" xm2■1 +x 2■n+1#1 0−ˆ1 0mxm−12■1 +x 2■n+1 dx = 2−mˆ1 0xm−12■1 +x 2■n+1 dx | {z }
:=In. Since 0≤In≤2ˆ1 0■1 +x 2■n+1 dx=1 n+ 2"■1 +x 2■n+2#1 0≤1 n+ 2, we conclude that In→and so
Rn→2. Together with (86), this finally yields that Lm(n) =1 nm(m+n+ 1)! m!n!ˆ1 0xm■1 +x 2■n dx =1
m!(m+n+ 1)( m+n). . .(n+ 2) nm(n+ 1)ˆ1 0xm■1 +x 2■n dx =1 m!m+n+ 1 nm+n n. . .n+ 2 nRn →1m·2
m!=2 m!.
216 Solution to problem 4870 Crux Math. 49 (7) 2023, 377 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We will show the following: Letq, c > 0 and
define the sequence (not series like it is mentioned in the problem statement) (an)
bya1=candan+1=an+1 q·anforn≥1. Then, with c(q) = 2 /q lim n→∞ an−p c(q)n = 0. Remark If one starts
with c <0, then all the anare negative too, and one obtains an+p c(q)n →0. (This is done by considering
bn:=−an). Solution Leth(x) :=x+1 qx. Then h >0 on ]0 ,∞[ and so an+1=h(an) is well defined. Taking
squares a2 n+1=a2 n+2 q+1 q2a2n, or equivalently a2 n+1−a2 n=2 q+1 q2a2n, we obtain the finite
telescoping series: a2 n+1−a2 1=nX k=1(a2 k+1−a2 k) =2 qn+1 q2nX k=11 a2 k. Hence (87) a2
n+1=c2+2 qn+1 q2nX k=11 a2 k. This allows us to estimate an+1: a2 n+1≥c2+2 qn, and so by using
this, (88) a2 n+1≤c2+2 qn+1 q2nX k=11 c2+2 q(k−1). Next consider the decreasing function f(x) =1
c2+2 qx,x≥0. Then nX k=11 c2+2 q(k−1)=1 c2+n−1X k=1f(k)≤1 c2+ˆn−1 0f(x)dx =1 c2+■q 2log■ c2+2
qx■■n−1 0 =1 c2+q 2log■ 1 +2 qc2(n−1)■ .
217 Thus we have arrived at the following estimates: (89) c2+2 qn≤a2 n+1≤c2+2 qn+1 q21 c2+1
2qlog■ 1 +2 qc2(n−1)■ . Hence ∆n:= an+1−r 2 qn = a2 n+1−2 qn an+1+q 2 qn ≤c2+1 q21 c2+1 2qlog■
1 +2 qc2(n−1)■ q c2+2 qn+q 2 qn −→ n→∞0. Finally an−r 2 qn ≤ an−r 2 q(n−1) + r 2 q(n−1)−r 2 qn = ∆
n−1+r 2 q·1√n−1 +√n −→ n→∞0.
218 Solution to problem 4866 Crux Math. 49 (7) 2023, 376 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - This is entirely trivial. We claim that■ ■■
■the identity is the only solution f:R→Rto (90) f(xy+f(f(y))) = xf(y) +y. We first show that f(0) = 0. In fact,
if x= 0, then f(f(f(y))) = y. Now take y= 0 in (90). Then 0 =f(f(f(0))) = xf(0). Hence f(0) = 0. Next, we take
y= 1 in (90). Then (91) f(x+f(f(1))) = xf(1) + 1 . Putu:=x+f(f(1)). This yields (92) f(u) = (u−f(f(1))) f(1) + 1 =:
au+b In other words, fnecessarily is an affine function. As we already know, f(0) = 0. Thus b= 0. Now
(90) yields a(xy+a2y) =xay+y. Hence a= 1.
219 Solution to problem 4857 Crux Math. 49 (5) 2023, 323 Raymond Mortini - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Since the function log xis concave on ]0 ,∞[, we have
log■A+B+C 3■ ≥logA+ log B+ log C 3=:R. Here we take A:=aabb, B:=bbcc, C=ccaa. Now the function
f(x) := 2 xlogxis convex on ]0 ,∞[, since f′′(x) = 2 /x≥0. Hence 2aloga+ 2blogb+ 2clogc=f(a) +f(b)
+f(c)≥3f■a+b+c 3■ = 3 f■1 2■ =−3 log 2 . Hence 3 R≥ −3 log 2, equivalently R≥log(1 /2) from which we
deduce that A+B+C≥3/2. In other words aabb+bbcc+ccaa≥3 2.
220 Solution to problem 4855 Crux Math. 49 (5) 2023, 323 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that all solutions ( a, b)∈N×Nare
given by■ ■■ ■(1, v),(u,1),(t, t),(2,3),(3,2) where u, v, t∈N:={1,2,···} can be arbitrarily chosen. It is
easily seen that these are solutions. Now let ( a, b) be a solution. Then ( b, a) is a solution, too. Ifb=a, or
if b= 1, then nothing remains to be shown. So we may assume that a > b > 1. Let log x be the natural
logarithm. Now the function f:x7→x/logxis strictly increasing for x≥eand strictly decreasing for 1 <
x≤ewith min x>0f(x) =e. So if a > b≥3> e, a loga>b logb or equivalently, ba> ab. Hence ab−ba<0, but
a−b >0. So this case, where a > b ≥3, does not occur. So it remains to consider the case a > b = 2. If a=
3, then we actually have the solution (3 ,2). If a≥4, then a loga≥4 log 4=2 log 2, and so a2−2a≤0< a−2.
Thus this case a≥4>2 =bdoes not occur, either. As all cases have been considered, we obtain the
assertion.
221 Solution to problem 4854 Crux Math. 49 (5) 2023, 323 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We prove that for 1 ≤r, s≤n, ■ ■ S:=nX
j=1■ sin■ jrπ n+ 1■ + sin■ jsπ n+ 1■■2 =( n+ 1 if r■=s 2(n+ 1) if r=s We first show that (93)nX
j=1cos■ j2ρπ n+ 1■ =( −1 if ρ∈Z\(n+ 1)Z n ifρ∈(n+ 1)Z and that for odd ρ∈Z (94)nX j=1cos■ jρπ n+ 1■
= 0. To see this, we will use that cos x= Re( eix), and that (95)nX j=1eijt=eitn−1X j=0eijt=eit1−eint
1−eit=eit−ei(n+1)t 1−eit. Now put t= 2ρπ/(n+ 1) whenever ρ∈Z\(n+ 1)Z. The latter guarantees that the
denominator does not vanish. Hence nX j=1eij2ρπ n+1=ei2ρπ n+1−1 1−ei2ρπ n+1=−1. Now if ρ∈(n+
1)Z, then, nX j=1eij2ρπ n+1=n. Thus (93) holds. If ρis odd, then, by putting t=ρπ/(n+ 1) in (95), we obtain
nX j=1eijρπ n+1 =eiρπ n+1+ 1 1−eiρπ n+1=icot■1 2ρπ n+ 1■ . This is a purely imaginary number, so its
real part is 0. This yields (94). From (93) we easily deduce that for r∈ {1,2, . . . , n } (96)nX j=1sin2■ jrπ
n+ 1■ =n+ 1 2.
222 In fact, using that sin2x=1−cos 2x 2, we obtain from (93) nX j=1sin2■ jrπ n+ 1■ =nX j=11−cos■
j2rπ n+ 1■ 2 =n 2−1 2nX j=1cos■ j2rπ n+ 1■ =n+ 1 2. We are now ready to calculate the value of S.
•Case 1■ ■■ ■r=s. Then S=nX j=1■ sin■ jrπ n+ 1■ + sin■ jrπ n+ 1■■2 = 4nX j=1sin2■ jrπ n+ 1■ =
(96)4n+ 1 2= 2(n+ 1). •Case 2■ ■■ ■r■=s. Since r, s∈ {1,2, . . . , n },randsdo not belong to ( n+ 1)Z.
Note that due to sinxsiny=1 2(cos(x−y)−cos(x+y)), (97) (sin x+ sin y)2= sin2x+ sin2y+ cos(
x−y)−cos(x+y). Hence S=nX j=1sin2■ jrπ n+ 1■ +nX j=1sin2■ jsπ n+ 1■ +nX j=1cos■ jπr−s n+ 1■ −nX
j=1cos■ jπr+s n+ 1■ = (96)n+ 1 +nX j=1cos■ jπr−s n+ 1■ −nX j=1cos■ jπr+s n+ 1■ = n+ 1 + S1−S2.
Several cases have to be analyzed now: a)r−sis even, say r−s= 2ρ, where ρ∈Z. Then r+sis even, too.
Since 0 <|r−s| ≤n−1 and 0< r+s≤2n <2(n+ 1), we again have two subcases: a1)r+s■∈Z(n+ 1)
(equivalently r+s■=n+ 1): Then by (93), S=n+ 1 + ( −1)−(−1) = n+ 1. a2)r+s=n+ 1∈Z(n+ 1): Then nis
odd, say n= 2m+ 1 for some m∈ {0,1,2, . . .}, and so S2=2m+1X j=1cos(jπ) = (−1) + (+1) +···+ (−1) +
(+1) + (−1) =−1. Hence S=n+ 1 + ( −1)−(−1) = n+ 1. b)r−sis odd. Then r+sis odd, too. Again we have two
subcases: b1)r+s■=n+ 1: Then by (94), S=n+ 1 + 0 −0 =n+ 1. b2)r+s=n+ 1. Then nis even, say n=
2mwith m∈ {1,2, . . .}, and so S2=2mX j=1cos(jπ) = (−1) + (+1) +···+ (−1) + (+1) = 0. Hence S=n+ 1 + 0
−0 =n+ 1.
223 Solution to problem 4844 Crux Math. 49 (5) 2023, 273 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Forn≥1, let In:=ˆ∞ 0xn−1e−x √x n−1X k=0
2k k! x−k 22k(n−k−1)!! dx. We show that ■ ■■ ■In=√π . We use the following well-known formulas,
where Γ is the Gamma function: (98)ˆ∞ 0xs−1e−xdx= Γ(s),Γ(s+ 1) = sΓ(s), s > 0 (99) Γ■ m+1 2■ = Γ■1
2■1 23 25 2···2m−1 2=√πQm k=1(2k−1) 2m=(2m)! m!4m√π So, with m=n−k−1, In=n−1X k=0 2k k! 1
22k(n−k−1)!ˆ∞ 0x(n−k−1/2)−1e−xdx =n−1X k=0 2k k! 1 22k(n−k−1)!Γ(n−k−1/2) =n−1X k=0 2k k! 1
22k(n−k−1)!(2(n−k−1))! (n−k−1)!4n−k−1√π =1 4n−1n−1X k=0(2k)! (k!)2(2(n−k−1))! ((n−k−1)!)2√π=1
4n−1n−1X k=0 2k k! 2(n−k−1) n−k−1! √π. This is related to the coefficient in the Cauchy product of ∞X
n=0 2n n! xn=∞X n=0 −1/2 n! (−1)n4nxn=1√1−4x, with itself and which converges for |x|<1/4, or if we
take x=y/4, ∞X n=01 4n 2n n! yn=1√1−y. In fact, for |y|<1, ∞X m=0ym=1 1−y=1√1−y1√1−y=∞X m=0 mX
k=0■2k k■ 4k■2(m−k) m−k■ 4m−k! ym The coefficients being unique, we deduce that for every m=
0,1,··· mX k=0■2k k■ 4k■2(m−k) m−k■ 4m−k= 1. Hence, with m=n−1, we conclude that In=√π.
224 Solution to problem 4850 Crux Math. 49 (5) 2023, 274, first version19 Raymond Mortini, Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Letn≥2. We show that
for anyfinite field the sum Sof all invertible n×nmatrices is the n×n zero matrix On. Forn≥1, let Mnbe the
set of all n×nmatrices and let Unbe the set of all invertible n×n matrices. Since the field has only a finite
number of elements, Unhas only a finite number of elements. SoS:=P U∈UnUis a well defined element
in Mn. We will show that for every ˜U∈ Un, S·˜U=S. Fix an invertible matrix ˜U∈ Unand consider the
map ι:( Mn→ M n X 7→X·˜U. Then ιis a bijection of Mnonto itself. The inverse is given by ι−1(Y)
=Y·˜U−1, since ι■ι−1(Y) =ι(Y·˜U−1) = (Y·˜U−1)·˜U=Y and ι−1■ι(X) =ι−1(X·˜U) = (X·˜U)·˜U−1=X.
Moreover, and this is the main point here, ιmapsUnbijectively onto itself. Thus (and here we have not
yet used that n■= 1) (100) S=X U∈Unι(U) =ι(X U∈UnU) =ι(S) =S·˜U. Now we use that n≥2. Take for
˜Uand 1 ≤i < j≤nthe elementary matrices Eij= (■ e1, . . . , ■ e j|{z} i-th col, . . . , ■ e i|{z} j-th col, . . . ,■ e
n), which interchange for X·Eijthei-th and j-th column of X. Thus S·Eij=Simplies that all the columns of
Sare the same. Say S= (■ s, . . . ,■ s ). Next we consider the matrix E= 1 1 0 0 1 0 ...... 0 ··· 1 .
Note that the action X·EofEon a matrix Xis to replace the second column of Xby the sum of the first and
second column. Since Eis invertible, we obtain from (100) that S·E=Sand so ■ s+■ s=■ s. Hence ■
s=■0. Consequently S=On. Remark We may also consider the case n= 1. Note that the smallest field is
given by F2:={0,1}, with 1 ■= 0, where 0 is the neutral element for addition and 1 the one for
multiplication. This necessarily has characteristic 2. Here S= 1. If the finite field is not field-isomorphic to
F2, it has more than two elements, and so there is an (invertible) element udifferent from 1. Now by
(100), S=Su, hence S(1−u) = 0. Since 1 −u■= 0, hence invertible,we conclude that S= 0. 19This was
tacitly replaced by another problem later on.
225 Solution to problem 4835 Crux Math. 49 (4) 2023, 213 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - This is a standard result/exercise in old
monographs on function theory/complex analysis and is for instance in [49, p. 70] Using a not so
sophisticated wording, we will show that four distinct points zj(j= 1, . . . , 4) in the plane belong to a
circle or a line if and only if their cross-ratio (bi-rapport, Doppelverh¨ altnis) DV(z1, z2, z3, z4) :=z1−z2
z1−z4.z3−z2 z3−z4 is a real number. In particular, being real, will be independent of the ”order” of the
points on the circle, respectively line. Our proof will be done in the extended complex plane, bC:=C∪
{∞} (also called the one-point compactification of C). Let us recall some terminology here. If Lis a line in
C, then L∪ {∞} is called an extended line. As usual we call the elements of the set of circles and
extended lines in bC”generalized circles”. We also use an extension of the definition of the cross-ratio
to points in bC. This is done by taking limits. For instance (101) D(z1, z2, z3,∞) =z1−z2 z3−z2. Finally,
let us recall the following results: i) There is a unique linear-fractional map (or in modern terminology, a
M¨ obius transform) T(z) := (az+b)/(cz+d),ad−bc■= 0, viewed as map from bCtobCmapping three
distinct points z2, z3, z4in bCto 0,1,∞, namely T(z) =DV(z, z2, z3, z4). ii) The cross ratio is invariant
under linear-fractional maps: DV(T(z1), T(z2), T(z3), T(z4)) =DV(z1, z2, z3, z4). Note that the latter is
an immediate consequence of i). iii) The class of generalized circles is invariant under M¨ obius
transforms. Now we are ready to confirm the statement above: Given four distinct points z1, z2, z3,
z4∈C, consider the map S(z) := DV(z, z2, z3, z4). Suppose that these zjbelong to a generalized circle
E. Now Smaps Eto the extended real line R∪ {∞} , since z2→0,z3→1 and z4→ ∞ . In particular,
wj:=S(zj)∈R∪ {∞} forj= 1, . . . , 4. Since DV(w1, w2, w3, w4) is real, the invariance result shows that
DV(z1, z2, z3, z4) is real. Conversely, suppose that DV(z1, z2, z3, z4) is real. Note that S(zj)∈ {0,1,∞}
⊆ R∪ {∞} for j= 2,3,4. Now the image of the extended real line by the inverse M¨ obius transform S−1is
a generalized circle, E. Of course Econtains the points z2, z3andz4. But, by (101), and the assumption,
we have S(z1) =DV(S(z1), S(z2), S(z3), S(z4)) =DV(z1, z2, z3, z4)∈R. Hence z1=S−1(S(z1))∈E. In
other words, all the zjbelong either to a circle or a line.
226 This can be shortened, without the explicit use of the cross ratio. Actually, just iii) is relevant here:
Consider the M¨ obius transform M(z) :=z−z4 z−z2z3−z2 z3−z4. Then z4, z3, z2are mapped to 0 ,1,∞,
and so the (unique) generalized circle Edetermined by z4, z3, z2 is mapped to the extended real line.
Thus the point z1belongs to Eif and only if M(z1)∈R. In other words, all the zjbelong either to a circle or
a line if and only ifz1−z4 z1−z2z3−z2 z3−z4∈R.
227 Solution to problem 4836 Crux Math. 49 (4) 2023, 214 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We first calculate the missing part P:=∞Y
n=1 2n+1∈P4n(n+ 1) (2n+ 1)2. Putp:= 2n+ 1. Then n= (p−1)/2 and so, in view of the Euler formula π2
6=∞X n=11 n2=Y p∈P1 1−p−2, we have P=Y p∈P p■=2p2−1 p2=4 36 π2=8 π2. To calculate R:=∞Y
n=14n(n+ 1) (2n+ 1)2, we use partial products and Stirling’s formula lim n→∞nne−n√ 2πn n!= 1.
PN:=NY n=14n(n+ 1) (2n+ 1)2=4NN!(N+ 1)!■ (2N+1)!QN n=1(2n)■2=4NN!(N+ 1)! (2N+ 1)!2■ 2NN!■2
1 =42NN!4(N+ 1) (2N+ 1)!2 ∼42NN4Ne−4N4π2N2(N+ 1) (2N+ 1)4N+2e−4N−22π(2N+ 1)
=πe2(2N)4NN2(N+ 1) (2N+ 1)4N(2N+ 1)3 = 2 πe2 1h■ 1 +1 2N■2Ni2N2(N+ 1) (2N+ 1)3 →2πe21 e21
8=π 4. Hence∞Y n=1 2n+1/∈P4n(n+ 1) (2n+ 1)2=π/4 8/π2=π3 32. A second way to derive the value of
Pis as follows:
228 Forz∈Cwe have sin(πz) = πz∞Y n=1■ 1−z2 n2■ =πz∞Y n=1■ 1−z n■ ez n∞Y n=1■ 1 +z n■ e−z n
=πz(1−z)ez∞Y n=1■ 1−z n+ 1■ ez n+1∞Y n=1■ 1 +z n■ e−z n =πz(1−z)ez∞Y n=1■ 1−z n+ 1■■ 1 +z
n■ ez n+1−z n =πz(1−z)ezeP∞ n=1(z n+1−z n)∞Y n=1■ 1−z n+ 1■■ 1 +z n■ =πz(1−z)∞Y n=1■ 1−z n+
1■■ 1 +z n■ . Hence P=∞Y n=14n(n+ 1) (2n+ 1)2=∞Y n=12n 2n+ 12n+ 2 2n+ 1 =∞Y n=11 1 +1 2n1
1−1 2(n+1)=1 ∞Y n=1■ 1 +1 2n■■ 1−1 2(n+ 1)■ =πz(1−z) sin(πz) z=1/2 =π 4.
229 Solution to problem 4828 Crux Math. 49 (3) 2023, 157 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Let I:=ˆπ/4 0ˆπ/4 0cosxcosy cos(x+y) cos(
x−y)dy | {z } :=I(x)dx. Now fix the variable x. Since cos(x+y) cos( x−y) = cos2y−sin2x, we obtain I(x) =
cos xˆπ/4 0cosy (1−sin2x)−sin2ydy u:= sin y= cos xˆ√ 2/2 0du cos2x−u2 =1 2(log(cos x+u)−log(cos
x−u)) √ 2/2 u=0 =1 2log■cosx+ 1/√ 2 cosx−1/√ 2■ . Hence (using Fubini), (102) I=1 2ˆπ/4 0log■√ 2 cos
x+ 1√ 2 cos x−1■ dx. The value of this integral is known to be the Catalan number C(see formula (18)
in [50]). An inde- pendent proof is below: using that cos a+cos b= 2 cos(a+b 2) cos(a−b 2) and cos(
a−b) =−2 sina+b 2sina−b 2, we obtain log■√ 2 cos x+ 1√ 2 cos x−1■ = log■cosx+ cos π/4
cosx−cosπ/4■ =−log tan■x+π/4 2■ −log tan■−x+π/4 2■ . A change of the variable x+π/4 = 2 y,
respectively −x+π/4 = 2 y, and a standard integral representation of Cyields I=−1 2ˆπ/4 π/8log tan
y(2dy)−1 2ˆπ/8 0log tan y(2dy) =−ˆπ/4 0log tan y dy=C. A proof of this standard representation can be
given for instance by using power series or Fourier series: h(z) :=1 2log■1 +z 1−z■ =∞X n=01 2n+
1z2n+1. Its Taylor coefficients belong to ■2and so the associated Fourier series h∗(eit) :=∞X n=01 2n+
1ei(2n+1)t
230 converges in the L2(]0, π[)-norm to h(eit) =1 2log(icot(t/2)) = iπ 4−1 2log tan( t/2) (Actually the
series h∗(eit) converges pointwise for z=eitwith 0 < t < π by the Abel-Dirichlet rule, but we do not need
this.) Taking real parts, and using that´P=P´ (note that Fourier series converge in the L2-norm, hence in
the L1norm), we may conclude that −ˆπ/4 0log tan y dy=ˆπ/2 0∞X n=01 2n+ 1cos(2 n+ 1)t dt=∞X
n=0(−1)n 1 (2n+ 1)2.
231 Solution to problem 4830 Crux Math. 49 (3) 2023, 158 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - First we claim that on [0 ,1/2] the function
f(x) =r 1−x 1 +xis convex. In fact, f′(x) =−1q 1−x 1+x(1 +x)2 and f′′(x) =1−2x (1−x)(x+ 1)3q 1−x 1+x≥0.
Since the graph of a convex function lies below the secant determined by ( a, f(a)),(b, f(b)), we obtain
that f(x)≤1−2(1−3−1/2)x, where a= 0 and b= 1/2. Since 1 −3−1/2≥1/3, we deduce that for 0≤x≤1/2
f(x)≤1−(2/3)x , and so nX i=1f(ai)≤n−(2/3)nX i=1ai=n−2/3. But for n≥2, we have n−2/3<(n+ 1)r n−1 n+
1=p n2−1, since n2−1−(n−2/3)2= 4/3n−13/9≥8/3−13/9 = 11 /9>0. This upper bound in the problem
appears to be artificial. We did not see a way to derive this in a natural way. To prove the reverse
inequality, we use Jensen’s inequality and obtain 1 nnX i=1f(ai)≥f■Pn i=1ai n■ =f(1/n). Hence nX i=1r
1−ai 1 +ai≥ns 1−1 n 1 +1 n=nr n−1 n+ 1.
232 Solution to problem 4826 Crux Math. 49 (3) 2023, 157 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that
■ S=π2 12−1 2. Just write Hk k(k+ 1)( k+ 2)=1 2■ Hk■1 k(k+ 1)−1 (k+ 1)( k+ 2)■■ =1 2■Hk k(k+
1)−Hk+1 (k+ 1)( k+ 2)+1 (k+ 1)2(k+ 2)■ . Now 1 (k+ 1)2(k+ 2)=1 k+ 2−k+1−1 (k+ 1)2=■1 k+ 2−1 k+ 1■
+1 (k+ 1)2. Since the Cesaro means of the sequences (1 /k) converge to 0, that is Hk/k→0, we
conclude that S=1 2H1 2−1 21 1 + 1+1 2∞X k=11 (k+ 1)2=π2 12−1 2.
233 Solution to problem 4825 Crux Math. 49 (3) 2023, 157 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We prove that■ ■ I:=∞X n=1On n(n+ 1)=
log 4. First we note that On n(n+ 1)=On■1 n−1 n+ 1■ =On n−On+1 n+ 1+1 (2n+ 1)( n+ 1). Since the
Cesaro means of the null sequence (1 /(2n+ 1)) converge to 0, we obtain I=O1 1+∞X n=11 (2n+ 1)( n+
1)= 1 + 2 log 2 −1 = log 4 . The value of the series S:=∞X n=11 (2n+ 1)( n+ 1)can be determined as
follows: NX n=11 (2n+ 1)( n+ 1)=NX n=1■2 2n+ 1−1 n+ 1■ splitting into even an odd =NX n=1■1 2n+
1−1 2n+ 1■ +NX n=1■1 2n+ 1−1 2n■ +2N+1X n=N+11 n =−1 +2N+1X n=1(−1)n+11 n+2N+1X n=N+11
n −→ N→∞−1 + log 2 + log 2 . Note that the well-known assertion lim N→∞P2N n=N+11 n= log 2 is a
direct consequence of the fact that the Euler-Mascheroni constant γis given by γ= lim( Hn−logn), where
Hn:=Pn i=11 i, since H2N−HN= (H2N−log(2 N)−γ) + (log N+γ−HN) + log 2 →log 2.
234 Solution to problem 4822 Crux Math. 49 (3) 2023, 156 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Substituting x= cosh twe obtain Tn(cosh t)
= cosh( nt). In particular, Tnhas no zeros on [1 ,∞[. Hence I:=ˆ∞ 1dx Tn(x)2/n=ˆ∞ 0sinht ■ cosh(
nt)■2/ndt=ˆ∞ 0et−e−t 2■ent+e−nt 2■2/ndt = 2−1+2/nˆ∞ 01−e−2t et■ 1 +e−2nt■2/ndt. Hence I <
2−1+2/nˆ∞ 01−e−2t etdt= 2−1+2/n■ −e−t+1 3e−3t■∞ 0 = 2−1+2/n2 3=1 3n√ 4. Moreover I > 2−1+2/nˆ∞
01−e−2t et(1 + 1)2/ndt= 2−1ˆ∞ 01−e−2t etdt= 2−12 3=1 3.
235 Solution to problem 4816 Crux Math. 49 (2) 2023, 101 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that for a, b, k ≥0 (knot necessary
an integer)
■ In:=ˆ1 0xkr a x+bn2x2ndxn→∞−→√ b+√a k+ 1/2. Write fn(x) =xk−1/2p a+bn2x2n+1. Ifa= 0, then In=ˆ1
0√ bnxn+kdx=n√ b n+k+ 1→√ b. Fora >0, let dn(x) :=xk−1/2■p a+bn2x2n+1−√ bn2x2n+1■ . Then
0≤dn(x) =xk−1/2 a√ a+bn2x2n+1+√ bnxn+1/2≤a√axk−1/2. Hence dnis dominated by an L1[0,1] function
and so, by using that nxn→0 for 0 < x < 1, lim nˆ1 0dn(x)dx=ˆ1 0lim ndn(x)dx=ˆ1 0√axk−1/2=√a k+ 1/2.
Consequently, ˆ1 0fn(x)dx =ˆ1 0dn(x)dx+√ bˆ1 0nxk−1/2xn+1/2dx =ˆ1 0dn(x)dx+√ bn k+n+ 1 −→
n→∞√a k+ 1/2+√ b.
236 Solution to problem 4819 Crux Math. 49 (2) 2023, 102 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Note that the harmonic mean x0:=
2ab/(a+b) is less than or equal to the arithmetic mean y0:= (a+b)/2. We show that the inequality holds
for arbitrary x0, y0with 0 < x0< y0<1. So let F be that primitive of fon [0 ,1] with F(0) = 0. We shall prove
that
■ 2ˆy0 x0tf(t)dt≥(F(y0)−F(x0))(F(y0) +F(x0)), from which the desired inequality immediately follows. By
partial integration, (103) 2ˆy0 x0t f(t)dt= 2ˆy0 x0tF′(t)dt= 2(y0F(y0)−x0F(x0))−2ˆy0 x0F(t)dt. For 0≤x,
y≤1, put H(x, y) := 2 yF(y)−2xF(x)−2ˆy xF(t)dt−(F(y)2−F(x)2). We have to show that H(x0, y0)≥0. Since 0
≤f≤1,F(x)≤´x 01dt=x. Hence ∂H ∂x(x, y) =−2■ F(x) +xf(x)■ + 2F(x) + 2F(x)f(x) = 2■ F(x)−x■ f(x)≤0.
Consequently, by using that H(y, y) = 0, we obtain ξ∈]x0, y0[ with H(x0, y0) =H(x0, y0)−H(y0, y0) =∂H
∂x(ξ, y0) |{z} ≤0(x0−y0)|{z} ≤0≥0.
237 Solution to problem 4817 Crux Math. 49 (2) 2023, 102 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LetE:=]0,∞[×]0,∞[×]0,∞[ and let H:E→]0,∞[
be given by H(a, b, c ) =a7+a3+bc a+bc+ 1+b7+b3+ca b+ca+ 1+c7+c3+ab c+ab+ 1. PutL:={(a, b, c
)∈E:abc= 1}. To be shown is that inf LH= 3 and that this lower bound is obtained exactly at (1 ,1,1). To
this end, consider for x >0 the function f(x) :=x7+x3+x−1 x+x−1+ 1=x8+x4+ 1 x2+x+ 1=x6−x5+x3−x+ 1.
Then fis convex on [0 ,∞[. In fact, f′(x) = 6 x5−5x4+ 3x2−1 and f′′(x) = 30 x4−20x3+ 6x= 2x(15x3−10x2+
3). Now f′′(x) = 2 x■ 5x2(3x−2) + 3■ . Then, clearly, f′′(x)≥0 ifx≥2/3. Since max [0,2/3]x2(2−3x) = 32
/35≤3/5, we deduce that f′′(x)≥0 on [0 ,2/3], too. Due to Jensen’s inequality, for ( a, b, c )∈L H(a, b, c ) =
f(a) +f(b) +f(c) = 3f(a) +f(b) +f(c) 3≥3f■a+b+c 3■ Since fis convex for x≥0,f(x)≥f(1) + f′(1)(x−1) = 1 + 3(
x−1) =−2 + 3 x. Why we take evaluation at 1? Because it works! It is an a posteriori choice, since the
minimal value is taken at (a, b, c ) = (1 ,1,1). Thus we obtain the estimate H(a, b, c )≥3 (−2 + (a+b+c))
=−6 + 3■ a+b+1 ab■ . We can even avoid Jensen’s inequality: H(a, b, c ) =f(a) +f(b) +f(c)≥(−2 + 3 a) +
(−2 + 3 b) + (−2 + 3 c) =−6 + 3( a+b+c). Since a+b+1 ab≥3 (see below) we deduce that for abc= 1 we
have H(a, b, c )≥ −6 + 9 = 3. As H(1,1,1) = 3, we are done. The inequality g(a, b) := a+b+1 ab≥3 is well
known. It can for instance be shown by using differential calculus: ga(a, b) = 1−1 ab2= 0⇐⇒ ab2= 1
and gb(a, b) = 1−1 ba2= 0⇐⇒ ba2= 1. In other words, ab(a−b) = 0. Hence a=b= 1 is the only stationary
point. Thus g(1,1) = 3 is the minimum, since the limit of gat the boundary ab= 0 is ∞.
238 Solution to problem 4811 Crux Math. 49 (2) 2023, 101 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show that n= 2 is the only solution. In
fact√ 23+ 1 +√2 + 2 = 3 + 2 = 5. Now, for x, y≥0, one has√x+√y∈Nif and only if xandyare perfect
squares. To see this, just note that √x+√y=x−y√x−√y implies that√x+√y∈Qif and only√x−√y∈Qand so,
by adding (respectively substracting),√x and√yare rational. Thus√x=p/qfor some p, q∈Nwith no
common divisor. Hence x2=p2/q2∈N, and so q= 1. Due to a classical result by L. Euler, the
Diophantine equation n3+ 1 = m2has in N={0,1,2, . . .} only the solutions ( m, n) = (1 ,0) and ( m, n) = (3
,2) (see for instance [51], a reference provided to the first author by Amol Sasane). Thus n= 2 is the
only positive integer also satisfying√n+ 2∈N.
239 Solution to problem 4810 Crux Math. 49 (1) 2023, 45 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We first show that wheneverPn j=1a2 j= 1,
then (104)nX j=1 j■=ia2 j ■nX j=1 j■=iaj■3≥1p 1−a2 i√n−1 (n−1)2. In fact, using Cauchy-Schwarz, we
immediately obtain nX j=1 j■=ia2 j ■nX j=1 j■=iaj■3≥nX j=1 j■=ia2 j ■■nX j=1 j■=ia2 j■
(n−1)■3/2=1p 1−a2 i√n−1 (n−1)2. Next we prove that wheneverPn j=1a2 j= 1, then (105)nX i=11p 1−a2
i≥nrn n−1. In fact, consider the convex function f(x) =1√1−x. By Jensen’s inequality (or one of the
possible defintions of convexity), ifPn j=1tj= 1 where (0 ≤tj≤1), then f■nX j=1tjxj■ ≤nX j=1tjf(xj). Here we
choose xi=a2 i, and tj= 1/n. Note thatPn i=1a2 i n= 1/n. Hence nX i=11p 1−a2 i=nnX i=11 nf(xi)≥nf■1
nnX i=1xj■ =nq 1−1 n=nrn n−1. Now putting (104) and (105) together yields nX i=1nX j=1 j■=ia2 j ■nX
j=1 j■=iaj■3≥√n−1 (n−1)2nrn n−1=n√n (n−1)2.
240 Solution to problem 4803 Crux Math. 49 (1) 2023, 44 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - It turns out that the triple (3 ,4,5) satisfying
32+ 42= 52is relevant here. Only one solution to the problem with p≤qexists: p= 2, q= 3 and a= 2, b= 1,
c= 2. To sum up: ■ ■■ ■22·2+ 32·1= (2·2 + 1)2 To see this, we use of course the well known
parametrizations of the solutions to A2+B2=C2, which are given by (∗)A= 2mn, B
=m2−n2andC=m2+n2, m, n ∈N. The conditions to be dealt with are i)
2mn=pa,ii)m2−n2=qbandiii)m2+n2= 2c+ 1. •First we note that ( a, b) = (0 ,0) is not admissible as 1 + 1 =
2 is even. Now if b= 0 and a■= 0, then by i) pnecessarily must be an even prime, that is p= 2. Hence
22a+ 1 = (2 c+ 1)2. By (*), 1 = m2−n2and 22a= 2mn. Consequently mandnare powers of 2. Hence
m2−n2is an even number; and not 1. Thus ab■= 0. •So let ab > 0. Since pis prime, mandncan only be
powers of 2 by (i). Due to iii), telling us that m2+n2is an odd number, not both mandncan be proper
powers of 2. Since m≥n(by ii)), we necessarily have n= 1 and m= 2xwith x■= 0. By ii), qb=m2−1 =
(2x)2−1 = (2x−1)(2x+ 1). This implies that q■= 2 (as the right hand side is odd). Since the difference of
the factors is 2, q≥3 cannot divide both factors. Thus we can only have that the factor 2x−1 equals 1.
Hence x= 1 and qb= 3, yielding b= 1 and q= 3. Finally by i), pa= 2mn= 2·21·1 = 22. So p= 2 and a= 2.
Finally, c= 2 as 32+ 42= 52.
241 Solution to problem 4809 Crux Math. 49 (1) 2023, 45 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We show, more generally, that whenever f,
g: [0,1]→[0,∞[ are continuous and f(1)g(1)■= 0, then■ ■ lim n→∞1 n3nX n=1■ˆ1 0xkf(x)dx■−1■ˆ1
0xkg(x)dx■−1 =1 31 f(1)1 g(1). Hence the limit in the problem is ( a+b)2/3. Proof LetM:= max {|f(x)|:
0≤x≤1}. Given εwith 0 < ε <1 2min{f(1), g(1)}, choose δ >0 so that |f(x)−f(1)| ≤εforδ≤x≤1. Moreover, let
n0be so large that δk+1≤ε/(2M) for k≥n0. Then ˆ1 0xkf(x)dx−1 k+ 1f(1) = ˆ1 0xk(f(x)−f(1))dx ≤2Mˆδ
0xk+ˆ1 δxk|f(x)−f(1)|dx ≤2Mδk+1 k+ 1+εˆ1 0xkdx ≤ε k+ 1+ε k+ 1. Therefore ■1 k+ 1f(1) +2ε k+ 1■−1
≤■ˆ1 0xkf(x)dx■−1 ≤■1 k+ 1f(1)−2ε k+ 1■−1 . We conclude that nX k=n0k+ 1 f(1) + 2 εk+ 1 g(1) + 2
ε≤nX k=n0■ˆ1 0xkf(x)dx■−1■ˆ1 0xkg(x)dx■−1 ≤nX k=n0k+ 1 f(1)−2εk+ 1 g(1)−2ε. Hence, by using
thatPn j=1j2=n(n+1)(2 n+1) 6, we deduce that lim sup n→∞1 n3nX n=1■ˆ1 0xkf(x)dx■−1■ˆ1
0xkg(x)dx■−1 ≤1 31 g(1)−2ε1 f(1)−2ε and lim inf n→∞1 n3nX n=1■ˆ1 0xkf(x)dx■−1■ˆ1 0xkg(x)dx■−1
≥1 31 f(1) + 2 ε1 g(1) + 2 ε. from which we conclude that lim n→∞1 n3nX n=1■ˆ1 0xkf(x)dx■−1■ˆ1
0xkg(x)dx■−1 =1 31 f(1)1 g(1). Remark The lower estimate show that the limit is infinite if f(1)g(1) = 0.
242 Solution to problem 4805 Crux Math. 49 (1) 2023, 44 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - First we note that ab+bc+ca= 4abcis
equivalent to (∗) ■(a, b, c ) :=1 a+1 b+1 c= 4. Ifa=b=c, then this condition is satisfied if a= 3/4. Let g(a,
b, c ) =a−1/a+b−1/b+c−1/c. It suffices to show that the minimum Mofgunder condition (*) is obtained for
a=b=c. Note that M:=g(3/4,3/4,3/4) = 3(3 /4)−4/3= 4(4 /3)1/3∼4.402. The gradient of the Lagrange
function H(a, b, c, λ ) =g(a, b, c ) +λ(■(a, b, c )−4) is zero if λ=a−1/a(1−loga) =b−1/b(1−logb)
=c−1/c(1−logc). Since the function x7→x−1/x(1−logx) is strictly decreasing on ]0 ,∞[, the only solution is
where a=b=c. The existence of the minimum is shown as follows (note that E:={(a, b, c ) :a, b, c > 0,
■(a, b, c ) = 4} is not compact. Condition (*) implies that a, b, c≥1/4. Let L:= inf Eg. Then L≥3 min
[1/4,∞[x−1/x= 3e−1/e≥3×0.692 = 2 .076. If this infimum is not taken on E, then there is an→ ∞ (orbn→ ∞
,orcn→ ∞ ) such that (an, bn, cn)∈Eandg(an, bn, cn)→L. In particular a−1/ann →1. We may assume
that bn→b0 andcn→c0(since otherwise bn→ ∞ and so cn→1/4, as well as L= 1 + 1 + 44> M , a
contradiction). Hence L= inf E′(1 +b−1/b+c−1/c), where E′={(b, c) :b, c > 0,1/b+ 1/c= 3}. In particular,
b≥1/3. Thus (by using Lagrange again, yielding x= 2/3) L= inf ]1/3,∞[1 +x−1/x+■3x−1 x■3x−1 xx=2/3= 1
+ 2(3 /2)3/2∼4.674> M. A contradiction. Consequently ( an, bn, cn)→(α, β, γ )∈Eand so the infimum is a
minimum. Hence g(a, b, c )≥g(α, β, γ ) =L=M. Here is a second proof, based on the article [52] (which
unfortunately contains many typos (poor proofreading? Poor referee job?). The function f(x) := xxis
convex. Let Tu(x) := f′(u)(x−u) +f(u) be the tangent to the graph of fat the point ( u, f(u)). Then
f(x)≥Tu(x). Next, let x1= 1/a,x2= 1/bandx3= 1/c. Then with u:=S= (x1+x2+x3)/3, 3X j=1f(xj)≥3X j=1TS(xj)
=3X j=1(f′(S)(xj−S) +f(S)) =f′(S)3X j=1(xj−S) + 3f(S)
243 =f′(S)3X j=1xj−3Sf′(S) + 3f(S) = 3 f(S). Since S= (1/a+ 1/b+ 1/c)/3 = 4 /3, we obtain with 1 /a+ 1/b+
1/c= 4 that (1/a)1/a+ (1/b)1/b+ (1/c)1/c=3X j=1f(xj)≥3f(4/3) = 3(4 /3)4/3= 4(4 /3)1/3.
244 Solution to problem 4801 Crux Math. 49 (1) 2023, 44 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
--------------------------------------------
■ We show that all solutions are given by f(x) =C 1 +xforC∈R. •It is straightforward to check that these
are solutions: f■ x+1 y■ =c 1 +x+1 y=cy y+yx+ 1=yf(xy+y). •Suppose that f: ]0,∞[→Ris a solution. Let
y=1 1+x. Then (106) f(2x+ 1) = f(x+1 y) =yf(y(1 +x)) =1 1 +xf(1). Next, let y=1 x. Hence, by using (106),
f(2x) =f(x+1 y) =yf(y(1 +x)) =1 xf(1 +1 x) =1 xf(1 + 21 2x)(106)=1 xf(1) 1 +1 2x=2f(1) 2x+ 1. Now let X:=
2xandC:= 2f(1). Then f(X) =2f(1) X+ 1=C 1 +X.
245 Solution to problem 4772 Crux Math. 48 (8) 2022, 483 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Fork= 1, this problem was given for
instance in the Middle European Mathematical Olympiad (MEMO) in 2012 in Switzerland (see [54] and
[53]) and we follow those published solutions. We claim that for a >0, all solutions f:R+→R+of20 (107)
f(ax+f(y)) =y af(xy+ 1) are given by f(x) =a/x. First, it is straightforward to see that this is a solution. Now
we proceed as in [54, 53]. Let fbe a solution. Step 1 Consider for y >0,y■=a, the auxiliary function g(y)
:=a−yf(y) a−y (this function is formally obtained by solving in R×R+the equation ax+f(y) =xy+1, which
gives x=xy=1−f(y) a−yfory■=a, and so ax+f(y) =a−yf(y) a−y=g(y). It will turn out that x=−1/yandg≡0).
Now for every y >0 with y■=aandxy>0, we have that g(y)≤0, since otherwise fis well-defined at g(y)>0
and so f(g(y)) =y af(g(y)), yielding that y=a, a contradiction. Step 2 Case 1 If there would exist y0>1
such that f(y0)< a/y 0, then with x0:= 1−1 y0>0 we have x0y0+ 1 = y0, u0:=ax0+f(y0) =a−a y0+f(y0)< a,
and f(u0) =f(ax0+f(y0)) =y0 af(y0)<1. Then xu0:=1−f(u0) a−u0>0 and so g(u0) =axu0+f(u0) =xu0u0+
1>0. But by Step 1, g(u0)≤0, a contradiction. Case 2 If there would exist y1>1 such that f(y1)> a/y 1,
then by the same reasoning as above, with x1:= 1−1 y1and u1:=ax1+f(y1)> a, we have f(u1)>1 and so
g(u1)>0, again. A contradiction. We conclude that f(y) =a/yfor every y >1. To deal with the remaining
case, take x= 1/a and 0 < y≤1. Then by (107), (108) f(1 +f(y)) =y af■y a+ 1■ . 20We prefer to use the
letter ainstead of k, as for us kalways belongs to N.
246 As both 1 + f(y) andy a+ 1 are bigger than 1, we deduce from (108) that a 1 +f(y)=y aa y a+ 1=ay
y+a. Hence f(y) =a/y.
247 Solution to problem 4779 Crux Math. 48 (8) 2022, 483 Raymond Mortini - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Iffis constant, then f′≡0 and we may choose any numbers
a < c 1< c2< bto satisfy (109)√ bf′(c1)+√af′(c2) = 0 . Otherwise, ftakes its distinct extremal values on [ a,
b]. We may assume that M:= max [a,b]f > f(a) (if not, M=f(a) and so min [a,b]f < f (a) and we consider
−f). Say M=f(x0) for some x0∈]a, b[. Then f′(x0) = 0, and due to continuity of f′, there are a < x 1< x
2≤x0with f′(x)>0 for x∈]x1, x2[, but f′(x2) = 0; we may choose x2= inf{t≤x0:f′≡0 on [ t, x0]}. By a similar
argument, there are x0≤y2< y 1such that f′(y2) = 0, but f′(x)<0 for x∈]y2, y1[. By the intermediate value
theorem for continuous functions, here for f′, there exists a small ε >0 such that f′takes every value from
[0 , ε] on ] x1, x2] and every value from [−ε,0] on [ y2, y1[. Now choose c1∈]x1, x2[ so that√
b√af′(c1)∈]0, ε[ (this is possible since limx■x2f′(x) = 0). Hence there exists c2∈]y2, y1[ with f′(c2) =−√
b√af′(c1). Thus√ bf′(c1)+√af′(c2) = 0 and c1< c2. Remark I do not see the role played by the special
coefficients√aand√ b. The whole works for any 0 < s1< s2<∞.
248 Solution to problem 4771 Crux Math. 48 (8) 2022, 483 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The problem is a bit ambiguous, due to an
undefined ∼symbol. Let Ln:=ˆ1 0f(x)2n+1dx and Rn=a·bn n Is it Ln−Rn→0? Or Ln/Rn→1? Or
cLn≤Rn≤CLnfor almost every nand some positive constants c, C? Note that, a priori, it is not even clear
that Ln>0. We are going to show the following: Example 16.■ ■ lim n→∞n+ 1 f(1)2n+1ˆ1
0f(x)2n+1dx=f(1) 2f′(1). Hence, with a:=f(1)2 2f′(1)andb=f(1)2we get that Ln/Rn→1. Proof. Since fis
assumed to be increasing, we see that f′(x)≥0 for 0 ≤x≤1. To exclude that for some points
x0∈]0,1],f′(x0) = 0, we need the convexity21off: in fact, let Tbe the tangent to the graph of fat (x0, f(x0));
then T(x) =f(x0) +f′(x0)(x−x0). The convexity of fimplies that the graph of flies above T. In particular, if
f′(x0) = 0, then, due to fbeing strictly increasing, f(x0−ε)< f(x0)< f(x0+ε) would contradict this fact. We
conclude that f′(x)>0 for every x∈]0,1]. To calculate our limit, we let 0 < s < 1 and write the integraln+ 1
f(1)2n+1LnasIn(s) +Jn(s), where In(s) =n+ 1 f(1)2n+1ˆs 0f(x)2n+1dx and Jn(s) =n+ 1 f(1)2n+1ˆ1
sf(x)2n+1dx. Claim 1 There is a function h(s) with 0 < h(s)<1, such that (110)f(1) 2f′(1)■ 1−h2n+2(s)■
≤Jn(s)≤f(1) 2f′(s). To see this, note that fconvex and C1imply that f′is increasing (by the way, a fact
equivalent to fbeing convex). By the mean-value theorem, and for s < x ≤1, there is cx∈]s,1[ with f′(cx)
=f(1)−f(x) 1−x. Hence f′(s)≤f′(cx)≤f′(1) and so f′(s)≤f(1)−f(x) 1−x≤f′(1). 21Note that fmerely being strictly
increasing, does not exclude the existence of zeros of f′:f(x) = (x−1/2)3.
249 In other words (111) f(1)−f′(1) + f′(1)x≤f(x)≤f(1)−f′(s) +f′(s)x. Now for f(x) =Ax+Bwith A■= 0 we have
ˆ1 s(Ax+B)2n+1dx=(A+B)2n+2−(As+B)2n+2 A(2n+ 2). Applying this to (111) yields n+ 1 f(1)2n+1ˆ1
sf(x)2n+1dx≤n+ 1 f(1)2n+1(f′(s) +f(1)−f′(s))2n+2−(f′(s)s+f(1)−f′(s))2n+2 f′(s)(2n+ 2) =1
2f′(s)f(1)2n+2−(f′(s)s+f(1)−f′(s))2n+2 f(1)2n+1 =f(1) 2f′(s) 1−■ 1−f′(s) f(1)(1−s)■2n+2! ≤f(1) 2f′(s)
because 0 ≤1−f′(s) f(1)(1−s)<1 for s∈[s1,1]. Similarily, n+ 1 f(1)2n+1ˆ1 sf(x)2n+1dx≥n+ 1 f(1)2n+1(f′(1)
+ f(1)−f′(1))2n+2−(f′(1)s+f(1)−f′(1))2n+2 f′(1)(2n+ 2) =1 2f′(1)f(1)2n+2−(f′(1)s+f(1)−f′(1))2n+2 f(1)2n+1
=f(1) 2f′(1) 1−■ 1−f′(1) f(1)(1−s)■2n+2! =:f(1) 2f′(1)■ 1−h(s)2n+2■ , with h(s) := 1 −f′(1) f(1)(1−s). Note
that 0 < h(s)<1 for s∈[s2,1]. This finishes the proof of Claim 1. Claim 2 limn→∞In(s) = 0 for every 0 < s
< 1. To this end, we need to show that max [0,1]|f|=f(1) and that the maximum is onlyobtained at 1 (note
that fmay take negative values). In fact, since fis increasing, f(0)≤f(x)≤f(1) for every x∈[0,1]. If f(0)≥0,
nothing has to be proven. So let f(0)<0. Then, by the mean value theorem on [0 ,1] there is 0 < cx<1
such that f(x) =f(0) + f′(cx)x≤f(0) + f′(1)x≤f(0) + f′(1) (note that f′is increasing). Using that 0
≤f′(1)<2f(1)22, we obtain f(1)< f(0) + 2 f(1). Hence f(0)>−f(1). As fis strictly increasing, we also have
f(0)< f(1), and so |f(0)|< f(1). Moreover, |f(x)| ■=f(1) for any x∈[0,1[. We conclude that |In(s)|= n+ 1
f(1)2n+1ˆs 0f(x)2n+1 ≤(n+ 1)s■max [0,s]|f(x)| f(1)■2n+1 =: (n+ 1)M2n+1, where 0 < M =M(s)<1. AsP∞
n=1(n+ 1)M2n+1converges, In(s)→0 asn→ ∞ . We are now ready to determine the limit ofn+1
f(1)2n+1´1 0f(x)2n+1dx. To this end, fix ε >0 and choose s3=s3(ε)∈]0,1[ so that for all s∈[s3,1] f(1)
2f′(s)−f(1) 2f′(1) < ε. 22It is only here that we use this assumption.
250 Now for s0:= max {s1, s2, s3}, depending on ε, we obtain from Claim 1 that f(1) 2f′(1)■
1−h2n+2(s0)■ ≤Jn(s0)≤f(1) 2f′(s0)≤f(1) 2f′(1)+ε. Since 0 < h(s0)<1, there is n0=n0(ε, s0) such that 0<
h(s0)2n+2< εfor all n≥n0. Thus, for n≥n0 f(1) 2f′(1)(1−ε)≤Jn(s0)≤f(1) 2f′(1)+ε. By Claim 2, there is
n1≥n0(depending on ε) such that |In(s0)|< ε forn≥n1. We conclude that for these n≥n1 n+ 1
f(1)2n+1Ln=In(s0) +Jn(s0)( ≤ε+f(1) 2f′(1)+ε ≥ − ε+f(1) 2f′(1)(1−ε). Hence n+ 1 f(1)2n+1Ln−f(1) 2f′(1)
≤max■ 2ε, ε■ 1 +f(1) 2f′(1)■■ . ■ Remark The function f(x) =x−1/2 shows that the assertion may fail if
f′(1) = 2 f(1), since in this case Ln= 0. On the other hand, it may hold, too if f′(1) = 2 f(1). In fact, if f(x)
=e2x, then f′(1) = 2 f(1) and Ln=e4n+2−1 4n+ 2and Rn=e4 4e2·e4n n=e4n+2 4n, nevertheless
Ln/Rn→1. What is the reason for this? Well, an analysis of the proof shows that the condition f′(1)<2f(1)
can be replaced by the assumption that the maximum of |f| isonly obtained at 1. This makes the class of
functions with the wished assymptotic behavior of the integrals´1 0f(x)2n+1dxmuch larger.
251 Solution to problem 4780 Crux Math. 48 (8) 2022, 484 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The assertion is not compatible with the
hypotheses. So we prove the following two results: Example 17. Leta >0andf∈C1[−a, a]. Iff(0) = 0 , then
ˆa −a(f′(x))2dx≥3 2a3■ˆa −af(x)dx■2 . Example 18. Let0< a, b < ∞andf∈C1[a, b]. Iff((a+b)/2) = 0 , then,
with C=12 (b−a)3, ˆb a(f′(x))2dx≥C ˆb af(x)dx!2 . Proof of Example 1. Letpbe a polynomial. Then, using
The Cauchy-Schwarz inequality I:=■ˆa 0(f′p)(x)dx■2 ≤■ˆa 0(f′(x))2dx■ ■ˆa 0p(x)2dx■ Using partial
integration, I=■ (f(x)p(x) a 0−ˆa 0f(x)p′(x)dx■2 Now choose p(x) =x−a. Then´a 0p(x)2dx=1 3(x−a)3 a
0=1 3a3. Hence, by noticing that p(a) =f(0) = 0, I=■ˆa 0f(x)dx■2 ≤■ˆa 0(f′(x))2dx■1 3a3 If we choose
p(x) =x+a, then p(−a) = 0, and we similarily obtain the appropriate estimation for´0 −af(x)dx. Hence,
using that ( x+y)2≤2(x2+y2), ■ˆa −af(x)dx■2 ≤2 3a3ˆa −a(f′(x))2dx ■ Proof of Example 2. Just use the
affine transformation ϕgiven by ϕ(x) = x+a+b 2. Then ϕ(−b−a 2) = aandϕ(b−a 2) = b, as well as ϕ(0)
=a+b 2. Let c:= (b−a)/2. Hence, with F(t) :=f(ϕ(t)) for−c≤t≤cwe obtain ˆb a(f′(x))2dx=ˆc −c(F′(t))2dt≥3
2c3■ˆc −cF(t)dt■2 =12 (b−a)3 ˆb af(x)dx!2 . ■
252 Of course Example 1 is a special case of Example 2. Is Cbest possible? Let q(x) =  (x−a)2
2−(b−a)2 8ifa≤x≤(a+b)/2 (x−b)2 2−(a−b)2 8if (a+b)/2≤x≤b. Then qis continuous on [ a, b],q((a+b)/2) = 0
and ˆb a(q′(x))2dx=12 (b−a)3 ˆb aq(x)dx!2 . Unfortunately, qis not C1. How to modify?
253 Solution to problem 4777 Crux Math. 48 (8) 2022, 484 Raymond Mortini, Rudolf Rupp - - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The assertion is not correct. In fact, let x:=
(x1, . . . , x n),R+={x∈R:x >0}, S:=n (x1, . . . , x n)∈(R+)n:nX i=11 xi= 1o , and f(x) :=nX i=11 1 2+x2 i.
We prove that for n≥2,
■ n 1/2 +n2= min x∈Sf(x)<sup x∈Sf(x) =2 3, and that for n= 1,S={1}and so ■ ■ f(x) =1 1 2+x2=f(1) =2
3. Proof Wlog n≥2. First we note thatPn i=11/xi= 1 for xi∈R+implies that xi≥1 for every i. Now max
1≤x<∞x 1 + 2 x2=1 3, since the function is decreasing on [1 ,∞[. Hence, for x∈S, nX i=11 1 2+x2 i=nX
i=1xi 1 + 2 x2 i2 xi<2 3nX i=11 xi=2 3, since for n≥2, no xican be 1. If for k > n xk=■ x(k) 1, . . . , x(k) n■
:=■1 1−(n−1)/k, k, . . . , k■ , thenPn i=1(1/x(k) i) = 1, xk→(1,∞, . . . ,∞) and f(xk)→2/3. Hence supSf=
2/3. To prove the assertion on the minimum, we use Lagrange. It is preferable to work with the new
variable yj:= 1/xj(to get a compact definition set, guarantying the existence of the global extrema). So
let S′=n (y1, . . . , y n)∈Rn, yj≥0 :nX j=1yj= 1o and g(y1, . . . , y n) :=nX i=1y2 i 1 +1 2y2 i. Then S′is
compact and inf fS= inf gS′= min gS′=:m. Say g(x′) =mfor some x′∈S′. In order to apply Lagrange, we
need to show that x′is an interior point of S′(in symbols, x′∈(S′)■). Let y′:= (1 /n, . . . , 1/n). Then
y′∈(S′)■. Now on ∂S′at least one of the
254 coordinates of these points y:= (y1, . . . , y n)∈∂S′is 0. Say, yn= 0. But thenPn−1 i=1yi= 1 and (via
induction on n, starting with the trivial case of one-tuples) g(y)≥n−1 1/2 + (n−1)2>n 1/2 +n2=g(y′).
Hence the absolute minimum of gonS′does not belong to the boundary. By Lagrange’s theorem, there
exists λ∈Rand ( y1, . . . , y n)∈S′such that ∇■ g(y1, . . . , y n) +λ(1−nX i=1yi)■ =0. That is, for every i∈
{1, . . . , n }, (112) λ=2yi (1 +1 2y2 i)2. Unfortunately, the function y7→q(y) :=2y (1+1 2y2)2is not
injective on [0 ,1] (note that the derivative vanishes at y=±p 2/3). So we must discuss several cases
(see figure 18): Figure 18. Non injectivity of qon [0,1] (i) If 8 /9 =q(1)≤λ <max [0,1]q, then the equation
q(y) =λhas two solutions 0 < y1, y2≤1. (ii) If λ= max [0,1]qor if 0 ≤λ < q (1) = 8 /9, then the equation q(y)
=λhas exactly one solution 0 ≤y0≤1. (iii) In all other cases, there is no solution with y≥0. We first show
that the case (i) does not yield minimal solutions. In fact, for fixed λ∈ [q(1),max [0,1]q[, equation (112)
has 2nsolutions of the form P:= (a, . . . , a|{z} k-times, b, . . . , b|{z} (n−k)-times) and their permutations,
where k= 0, . . . , n and 0 ≤a≤b≤1. Note that (113) q(1/n) =8n3 (1 + 2 n2)2≤q(1/2)< q(1) =8 9< q(p 2/3).
Hence 1 /n≤1/2<min{a, b}(see figure 18). LetA:= (1 /n, . . . , 1/n). Then A∈S′. Since the function
y7→y2/(1 +1 2y2) is increasing on [0,∞[, we deduce that g(P) =ka2 1 +1 2a2+ (n−k)b2 1 +1 2b2> g(A),
soPdoes not yield a minimum. Thus only the second case occurs. That is, we need to consider only a
solution of (112) of the form ( y1, . . . , y n) = ( a, . . . , a ) with 0 < a≤1. Using the constraint conditionPn
i=1yi= 1, we obtain that a= 1/n, hence ( y1, . . . , y n) = (1 /n, . . . , 1/n).
255 Consequently, x′= (1/n, . . . , 1/n) is the unique point where gtakes its absolute minimum on S′. We
conclude that mingS′=n 1 2+n2. For completeness, we observe that M:= max S′gnecessarily is
obtained on the boundary of S′(for instance, M=g(1,0, . . . , 0) = 2 /3), as Lagrange only yields a single
stationary point of the Lagrange function in ( S′)■. A second way to see that case (i) does not occur
goes as follows: We first show that the case (i) does not yield minimal solutions. In fact, for fixed λ∈
[q(1),max [0,1]q[, equation (112) has 2nsolutions of the form P:= (a, . . . , a|{z} k-times, b, . . . , b|{z}
(n−k)-times) and their permutations, where k= 0, . . . , n and 0 ≤a≤b≤1. Note that q(1/2) = (8 /9)2and that
n≥2. Thus (114) q(1/n)≤q(1/2)< q(1)≤λ < q (p 2/3). Hence 1 /n≤1/2<min{a, b}=a(see figure 18). Since for
such a point P= (y1, . . . , y n) we havenX i=1yi=ka+ (n−k)b > k1 2+ (n−k)1 2=n 2≥1, Pdoes not belong
to S′; that is such a solution of the system (112) of equations does not satisfy the constraint P∈S′.
256 Partial Solution to problem 4763 Crux Math. 48 (7) 2022, 421 Raymond Mortini, Rudolf Rupp - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Here we give our thoughts on this
not very precisely formulated problem. First we note that S⊆Knecessarily is an additive subgroup of the
field K. Note that {0,1} ⊆K. In particular 0 = x−x∈Sand with x∈Swe have −x= 0−x∈S. If (FE) f(x−y)
=f(x)f(y) for all x, y∈S then we get the following: (1)y=x=⇒f(0) = f(x)2 (2)y= 0 =⇒f(x) =f(x)f(0)
=⇒f(x)(1−f(0)) = 0 Case 1 There exists x0∈Swith f(x0) = 0. Then, by (1), f(0) = 0 and so f(x) = 0 for all
x∈S. Case 2 fhas no zeros. Then (2) implies that f(0) = 1. We claim that f(2x) = 1 for every x∈S(note
that ZS⊆S). In fact, f(x) =f(2x−x) =f(2x)f(x), hence f(2x) = 1. We conclude that for S=Re.g., the constant
function f(y) = 1 is the only solution, as every y∈Rwrites as y= 2xfor some x. Next we show that fis even
and that f(x)∈ {− 1,1}. In fact, by (FE), for x= 0, f(−y) =f(0)f(y) =f(y) for every y∈S. Hence 1 = f(2u)
=f(u−(−u)) =f(u)f(−u) =f(u)2for any u∈S. IfS=Z, then we have three solutions: f≡0,f≡1 but also f(n) =( 1 if
neven −1 if nodd. In fact by the claim above, f(2m) = 1 for every m∈Z. Now let σ:=f(1). We already know
that σ=±1. Now for every m∈Z, σ=f(1) = f((2m+ 1)−2m) =f(2m+ 1)f(2m) =f(2m+ 1). LetP:=Pf:={x∈S:f(x)
= 1}andR:={x∈S:f(x) =−1}. Then Pis a subgroup of Ssince x, y∈Pimplies that x−y∈P, because f(x−y)
=f(x)f(y) = 1·1 = 1. As shown above, 2 S⊆P⊆Sand 2 Sis a subgroup of S. Here S= 2Sif and only if all
the translation operators τx:S→S, y7→x−yhave a fixed point. Also note that Rhas the following
property: (PR) ( R−R)⊆Pand ( R−P)∪(P−R)⊆R. Conversely, if Pis a proper subgroup of
SandR:=S\Psuch that (PR) holds, then the function ggiven by g(x) =( 1 if x∈P −1 if x∈R, satisfies the
functional equation (FE) g(x−y) =g(x)g(y) for x, y∈S.
257 Note that Pmay be strictly bigger than 2 S: in fact, let K=C,S:=Z+iZ,P= 2Z+iZ andR=S\P. Then S,
P, R satisfy (PR), but P:= 2Sdoes not satisfy (PR). IfS=Kis a field of characteristic 2, then Pf=R=S(note
that 1 = −1), and so only the constant functions 1 and 0 satisfy (FE).
258 Solution to problem 4747 Crux Math. 48 (2022), 282 by Raymond Mortini, Rudolf Rupp - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that all solutions of the
functional equation f(x2f(x) +f(y)) =f(f(x3))) + y, x, y ∈R are given by f(x) =xandf(x) =−x. Claim1 fis
injective: Putx= 0. Then (115) f(f(y)) =f(f(0)) + y. Now if f(y1) =f(y2), then by (115) f(f(y1)) =f(f(0)) +
y1andf(f(y2)) =f(f(0)) + y2 Hence y1=y2. Claim 2 fis surjective: Letw∈R. Then, by (115), w=f(f(0)) + (
w−f(f(0))) = f(f(w−f(f(0))). Claim 3 f(0) = 0: Take y= 0: then f(x2f(x) +f(0)) = f(f(x3)). Since fis bijective, we
conclude that x2f(x) +f(0) = f(x3). Now put x= 1: then 12f(1) + f(0) = f(1). Hence f(0) = 0. Claim 4
f■f=id(that is, fis an involution). This follows from (115). Hence our equation becomes (116) f(x2f(x)
+f(y)) =x3+y(x, y∈R). In particular, for y= 0, (117) f(x2f(x)) =x3or equivalenty x2f(x) =f(x3). Claim 5 fis
additive: In fact, the surjectivity of fandx7→x3now imply that x7→x2f(x) is surjective, too. Hence
f(x2f(x)|{z} =a+f(y)|{z} =b) =x3+y=f(x2f(x)) +y=f(a) +f(b) yields the additivity of f. Claim 6 f(−x) =−f(x). Just
use that with f(0) = 0 and fadditive, 0 =f(0) = f(x+ (−x)) =f(x) +f(−x). Claim 7 Letf(a) = 1. Then a=±1.
259 Recall that by (5) and (6), f(a+b) =f(a) +f(b) for a, b∈Randf(mx) =mf(x) for every m∈Z. Hence, by
(117), for x=a+b, (a+b)2f(a+b) =f((a+b)3). Expansion yields: (a2+b2+ 2ab)(f(a) +f(b)) = f(a3) + 3f(a2b) +
3f(ab2) +f(b3)⇐⇒ a2f(a)+b2f(a) + 2abf(a) +a2f(b) +b2f(b)+ 2abf(b) = f(a3)+ 3f(a2b) + 3f(ab2)
+f(b3)⇐⇒ b2f(a) + 2abf(a) +a2f(b) + 2abf(b) = 3 f(a2b) + 3f(ab2). •Letb= 1 and note that a=f(1). Then 1
+ 2 a+a3+ 2a2= 3f(a2) + 3 = 3 a3+ 3⇐⇒ 2a3−2a2−2a+ 2 = 0 ⇐⇒ a2(a−1)−(a−1) = 0 ⇐⇒ (a−1)(a2−1)
= 0 ⇐⇒ a∈ {− 1,1}. Claim 8 If the additive function fsatisfies x2f(x)) = f(x3), then f(x) =f(1)x. To see this,
we consider four cases: •Leta= 2,f(1) =±1 and b=x. Then (118)■ ■■ ■±x2±4x−4f(x) + 2xf(x)−3f(x2) = 0
. •Leta= 1,f(1) =±1 and b=x. Then, (119)■ ■■ ■±x2±2x−2f(x) + 2xf(x)−3f(x2) = 0 . Calculating
(118)-(119), yields ±2x−2f(x) = 0. Hence f(x) =±x=f(1)x. One can also prove Claim 8 without using Claim
7, and then deducing Claim 7 from Claim 8 if additionally we assume that fis an involution. In (120)
b2f(a) + 2abf(a) +a2f(b) + 2abf(b) = 3 f(a2b) + 3f(ab2). choose a= 1, resp. a= 2 and b=x. Then f(2) =
f(2·1) = 2 f(1) and so (121) x2f(1) + 2 f(1)x+f(x) + 2xf(x)−3f(x)−3f(x2) = 0 (122) 2 x2f(1) + 8 f(1)x+ 4f(x) +
4xf(x)−12f(x)−6f(x2) = 0 Hence, by calculating (121)-1 2(122), we obtain (123) −2f(1)x−f(1)x+ 3f(x) = 0 .
Hence f(x) =f(1)x. Using (117), that is f(x2f(x)) =x3, we have f(1)x2f(1)x=x3. Hence f(1)2= 1 and so f(1)
=±1.
260 Solution to problem 4657 Crux Math. 47 (2021), 301 by Raymond Mortini, Rudolf Rupp - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - a) Suppose that the function fsatisfies f(x)
+f(2x)≡0 onR. Then the continuity of fatx= 0 implies that f≡0. In fact, fix x∈R\ {0}. By induction, f(x/2n) =
(−1)nf(x). By taking limits, the continuity at 0 implies that for neven we get f(0) = f(x) and for nodd, we
get f(0) =−f(x). Hence 2f(x) =f(0)−f(0) = 0, and so f≡0. b) Define the function f:R+→Rbyf(0) = 0, f(x) = 1 if
1 ≤x <2 and xrational, f(x) =−1 if 1< x < 2 and xirrational. If n∈Nand 2n≤x <2n+1, put f(x) = (−1)nf(x/2n).
If1 2n+1≤x <1 2n, putf(x) = (−1)nf(2nx). If x <0, then let f(x) =f(−x). Then fis discontinuous everywhere
and, by construction, f(x) +f(2x) = 0. c) All solutions to f(x) +f(2x)≡0 onR: Letg: [−2,−1[∪[1,2[→Rbe an
arbitrary function. Put f(x) =  0 if x= 0 (−1)ng(x/2n) if 2n≤ |x|<2n+1 (−1)ng(2nx) if1 2n+1≤ |x|<1 2n.
This functional equation and its companion f(x) =f(2x) appear multiple times, see [55, 56, 57, 58, 59, 60,
61, 62, 63, 64]
261 Solution to problem 4636 Crux Math. 47 (2021), 200 by Raymond Mortini, Rudolf Rupp - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - The equation (3x+ 7)log43−(4x−7)log34=
4x−3x−14 has on Rthe unique solution x= 2. In fact, first note that a:= log43 =log 3 log 4>0 and log34 =
1 /a. Then with A:= 3x+ 7 and B:= 4x−7 we have to solve Aa−B1/a=B−Aor equivalently, Aa+A=
(B1/a)a+B1/a. Since the function x7→xa+xis strictly increasing, we deduce that A=B1/a. In other words,
3x+ 7 = (4x−7)1/a, or equivalently (124) log 4 log(4x−7) = log 3 log(3x+ 7). The curve y(x) = log 4
log(4x−7)−log 3 log(3x+ 7) is defined for x > log 7/log 4 := x0with limx→x0y(x) =∞and its derivative y′(x)
= log241 1−7x−4−log231 1 + 7 x−3 is strictly decreasing with lim y→x0y′(x) =∞and lim x→∞y′(x) =
(log24−log23). Note that the asymptote at infinite is the line y= (log24−log23)x. In particular, y′>0 and so
the curve is strictly increasing and its unique zero is x1= 2 (observe that log(4) log(9) = 4 log(2) log(3) =
log(3) log(16), so (124) holds).
262 Solution to problem 4634 Crux Math. 47 (2021), 200 by Raymond Mortini, Rudolf Rupp - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Foran>0, let Gn:=n(a1···an)1/n. Then lim
n→∞Gn= 0 wheneverP∞ n=1anis convergent. In fact, given ε >0, choose Nso big thatP∞ n=Nan< ε.
Due to the arithmetic-geometric inequality, for n > N , Gn= (a1···aN)1/n n n−N■
(n−N)(aN+1···an)1/(n−N■n−N n (n−N)1−n−N n ≤σn■ nX j=N+1aj■n−N n , where σn:= (a1···aN)1/n n
n−N(n−N)N/n. Since lim nσn= 1, we have lim supnGn≤lim supnεn−N n=ε, from which we deduce that
Gn→0.
263 Solution to problem 4615 Crux Math. 47 (2021), 301 by Raymond Mortini, Rudolf Rupp - - - - - - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ifpis a polynomial, we have (due to
Cauchy-Schwarz) ˆ1 0f′′pdx 2 ≤■ˆ1 0(f′′)2dx■■ˆ1 0p2dx■ . Now, by using twice integration by parts,ˆ
f′′pdx= (f′+c)p−■ (f+cx+c′)p′−ˆ (f+cx+c′)p′′dx■ Now let p(x) =x(x−1). Evaluation at the end-points and
using the hypothesis that´1 0fdx=f(1)/2, yieldsˆ1 0f′′pdx=−f(0). Since´1 0p2dx=´1 0(x4+x2−2x3)dx=
1/30, we deduce that ˆ1 0(f′′)2dx≥30f(0)2. Equality is given if f′′=pandf(1) = 2´1 0fdx; for instance if f(x)
=1 12x4−1 6x3−1 30. Here f(1) =−7/60. Generalizations appear in [8].
264 6.EMS Newsletter
265 7.Math. Gazette Solution to problem 108.H, Math. Gazette 108, Issue 572 (2024), p. 364 by
Raymond Mortini and Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
- - - - (a) We first consider the Fourier series of the function g(x) =( x(π−x) if 0 ≤x≤π x(π+x) if−π≤x≤0 (the
odd extension of x(π−x)), and extend it 2 π-periodically. Then g(x) =8 π∞X n=1 noddsin(nx) n3. Hence
I1:=ˆπ 0x(π−x) sinxdx =8 π∞X n=1 nodd1 n3ˆπ 0sin(nx) sinxdx =8 π∞X n=1 nodd1 n3π= 87 8ξ(3) = 7
ζ(3). Note that the integrand coincides with Un−1(cosx), where Unis the Chebyshev polynomial of the
second kind. To see that for odd n, Jn:=ˆπ 0sin(nx) sinxdx=π (a well known result), it suffices to show
that Jn=Jn+2. This holds, though, since Jn+2=ˆπ 0sin(nx) cos 2 x+ cos( nx) sin 2 x sinxdx =ˆπ 0■sin(nx)
sinx(1−2 sin2x) + 2 cos nxcosx■ dx =Jn+ 2ˆπ 0(−sin(nx) sinx+ cos( nx) cosx)dx =Jn+ 2ˆπ 0cos(n+ 1)x
dx =Jn.
266 (b) We shall work with suitable variable substitutions to regain the integral in (a). We use arctan x
instead of the ambiguous notation tan−1x. Note that arctan(1 /y) + arctan y=π/2 for y >0. I2:=ˆ∞
−∞arctan( ex) arctan( e−x)dxex=y= x=log yˆ∞ 0arctan yarctan1 y ydy y=tan s= arctan y=sˆπ/2 0s(π 2−s)
tans1 cos2sds = 2ˆπ/2 0s(π 2−s) sin(2s)ds 2s=t=ˆπ 0t 2■π 2−t 2■ sintdt =1 4I1.
267 Solution to problem 108.G, Math. Gazette 108, Issue 572 (2024), p. 364 by Raymond Mortini and
Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We suppose that
the lkin the integrand should be the Ik, that is Ik:=ˆπ/2 0(sinx)kdx. So let Sn:=n nY k=1Ik!2 n . We claim
that
■ lim nSn=π 2e . We shall use that Ik=√π 2Γ(k+1 2) Γ(k+2 2). Now Snis a telescopic product; hence
Sn=n■√π 2■2Γ(1) ■ Γ(n 2+ 1)■2/n. Using (the non-discrete) Stirling formula, which tells us that lim
x→∞Γ(x)q 2π x■x e■x= 1, we obtain with Γ■n 2+ 1■ =n 2Γ■n 2■ that Sn∼π 4n ■ n 2q 2π n/2■ n/2
e■n/2■2/n=π 4n ■n 2■2/n■4π n■1/nn 2e→π 2e.
268 Solution to problem 108.F, Math. Gazette 108, Issue 572 (2024), p. 364 by Raymond Mortini and
Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - This is entirely
trivial: Note that an ellipse with focii ±r,r >0, in the z-plane is given by |z−r|+ |z+r|= 2cfor some positive
constant c. Hence, as |z2−r2|=|w2|, respectively |w2−r2|=|z2|, and through squaring, |z−r|2+|z+r|2+
2|z−r| |z+r|= 2|z|2+ 2r2+ 2|z2−r2| = 2|z|2+ 2r2+ 2|w|2 As the latter is symmetric in zandw, we obtain
(|z−r|+|z+r|)2= (|w−r|+|w+r|)2. As the terms are positive, it follows that |z−r|+|z+r|=|w−r|+|w+r|.
269 Solution to problem 108.E, Math. Gazette 108, Issue 572 (2024), p. 364 by Raymond Mortini and
Rudolf Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that
only n= 5 yields the desired prime quintuplets. Suppose that n, n+ 2, n+ 6, n+ 8, n+ 14 are prime
numbers. Say n=a0+Pm k=1ak10k, with 0≤ak≤9. Obviously a0∈ {1,3,7,9}ora0= 5 and all ak= 0 for k≥1.
Case 1 a0= 1. Then n+ 14 is not prime since n+ 14 = 5 + ( a1+ 1)10 +Pm k=1ak10kis divisible by 523.
Case 2 a0= 3. Then n+ 2 is not prime since n+ 2 = 5 +Pm k=1ak10kis divisible by 5. Case 3 a0= 7.
Then n+ 8 is not prime since n+ 8 = 5 + ( a1+ 1)10 +Pm k=1ak10kis divisible by 5. Case 4 a0= 9. Then
n+ 6 is not prime since n+ 6 = 5 + ( a1+ 1)10 +Pm k=1ak10kis divisible by 5. So it remains a0= 5: (5
,7,11,13,19), the only prime quintuple of this form. 23Note that a1+ 1 may be equal to 10; that does not
alter the divisibility property, though.
270 Solution to problem 108.D, Math. Gazette 108, Issue 571 (2024), p. 167 Raymond Mortini, Rudolf
Rupp - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We give for (a) two
proofs. (a1) This is done via the Laplace transform F(s) :=L[f](s) :=´∞ 0f(t)e−stdtfor certain admissible
functions f. First note that1−e−2x x=ˆ2 s=0e−sxds.Then, due to the convergence of the integrals, and
Fubini’s theorem, I:=ˆ∞ 0(1−e−2x)sin2x x3dx =ˆ∞ 01−e−2x xsin2x x2dx=ˆ∞ x=0ˆ2 s=0e−sxsin2x x2dsdx
=ˆ2 s=0■ˆ∞ x=0e−sxsin2x x2dx■ ds. Starting with L[sin2t](s) =L[1−cos(2 t) 2](s) =1 2s−1 2s s2+4and
using twice the formula L[f(t)/t](s) =ˆ∞ sF(σ)dσ, we see that the Laplace transform of the function S(x)
:=sin2x x2is given by slogs 2−slog(s2+ 4) 4+ arctan(2 s) |{z} =π 2−arctans 2. Hence I=■sπ 2+s2logs
4−s2+ 4 8log(s2+ 4) +1 2−sarctans 2+ log■s2 4+ 1■■2 0=π 2. (a2) Note that I=1 2ˆ∞
0(1−e−2x)1−cos(2 x) x3dx. So we need to apply the residue theorem to the function f(z)
:=(1−e−2z)(1−e2iz) z3 and the contour
271 Γ(r, R) =γ1(r, R)⊕γ2(R)⊕γ3(r, R)⊕γ4(r, R), where γ1(n) = [r, R],γ2(R) =Reit,0≤t≤π/2,γ− 3(r)
=it,r≤t≤R,γ− 4(r, R) =reit, 0≤t≤π/2. Observe that´ Γ(r,R)f(z)dz= 0. Since ˆ γ− 3(r,R)f(z)dz=ˆR rf(it)idt=ˆR
r(1−e−2it)(1−e−2t) (−i)t3idt =−ˆ γ1(r,R)f(z)dz, we haveˆ γ1(r,R)f(z)dz+ˆ γ3(r,R)f(z)dz= 2ˆ γ1(r,R)Ref(z)dz=
4ˆR r(1−e−2x)sin2x x3dx. Now I(R) :=ˆ γ2(R)f(z)dz→0 asR→ ∞ . In fact, |I(R)|= ˆπ/2 0f(Reit)iReitdt ≤ˆπ/2
0(1 +e−2Rcost)(1 + e−2Rsint) R2dt≤4π 2 R2. Moreover, J(r) :=ˆ γ4(r)f(z)dz→ −2πasr→0. In fact, since
the Taylor expansion of zf(z) = (1−e−2z)(1−e2iz) z2 at the origin equals −4i+ (4 + 4 i)z−4z2+O(z3), we
have J(r) = −ˆπ/2 0(1−e−2reit)(1−e2ireit) r2e3itieitdtr→0−→ − (−4i)iˆπ/2 01dt=−2π (note that lim r→0´ =´
limr→0since the integrand is uniformly continuous for ( r, t)∈]0,1]×[0,2π].) Now 0 = lim R→∞ r→0ˆ
Γ(r,R)f(z)dz = lim R→∞ r→0 ˆ γ1(r,R)f(z)dz+ˆ γ3(r,R)f(z)dz! + lim R→∞ˆ γ2(R)f(z)dz+ lim r→0ˆ γ4(r)f(z)dz
= 4 I+ 0−2π. Consequently, I=π/2. (b) We first observe that due to the majorant 1 /x2for|x| ≥1, the
integral exists as Lebesgue integal as well as improper Riemann integral (note the discontinuity points
at −π/2 +kπ, k∈Z). Since f(x) := cos2(tanx) isπ-periodic and even, we may use the Lobashevski integral
formula ˆ∞ 0f(x)sin2x x2dx=ˆπ/2 0f(x)dx (see e.g. [66]24) to conclude that ˆ∞ −∞cos2(tanx)sin2x x2dx =
2ˆπ/2 0cos2(tanx)dx s:=tan x= dx=ds 1+s22ˆ∞ s=0cos2s 1 +s2ds=ˆ∞ s=01 + cos(2 s) 1 +s2ds = arctan
s ∞ 0+1 2ˆ∞ −∞cos(2 s) 1 +s2ds =π 2+πiRes■e2iz 1 +z2;z=i■ =π 2+πie−2 2i=π 2■ 1 +1 e2■ . 24The
proof there is also valid for the case of all even Riemann-integrable π-periodic functions.
272 Here we have applied the residue theorem to evaluate the classical integral´∞ −∞cos(2 s) 1+s2ds,
which used to be an exercise in many complex analysis courses (see e.g. [65, p. 210]). Now we have
the following identities: ˆ∞ −∞sin2(tanx)cos2x x2dx−ˆ∞ −∞cos2(tanx)sin2x x2dx =ˆ∞ −∞■
sin2(tanx)cos2x x2+ sin2(tanx)sin2x x2■ dx−ˆ∞ −∞■ sin2(tanx)sin2x x2+ cos2(tanx)sin2x x2■ dx =ˆ∞
−∞sin2(tanx) x2dx−ˆ∞ −∞sin2x x2dx. The latter difference, though, vanishes. In fact, sinceP´ =´P(note
that all terms are positive),we obtain in view of the classical formula (see e.g. [66]), 1 sin2x=∞X k=−∞1
(kπ+x)2 that ˆ∞ −∞sin2(tanx) x2dx =∞X k=−∞ˆπ 2+kπ −π 2+kπsin2(tanx) x2dx x=−π 2+kπ+u=∞X k=−∞ˆπ
0sin2(tan( u−π 2)) (−π 2+kπ+u)2du =ˆπ 0sin2(−cotu)∞X k=−∞1 (kπ+ (u−π 2))2du =ˆπ 0sin2(−cotu)1
sin2(u−π 2)du=ˆπ 0sin2(−cotu) cos2udu −cotu=s= du=sin2udsˆ∞ −∞sin2s cos2usin2u ds=ˆ∞ −∞sin2s
s2ds. Amazing! Hence the second formula ˆ∞ −∞sin2(tanx) cos2x x2dx=π 2■ 1 +1 e2■ holds, too.
273 Solution to problem 108.C, Math. Gazette 108, Issue 571 (2024), p. 167 by Raymond Mortini - - - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - We claim that all solutions
fcontinuous on [0 ,∞[ are given by■ ■■ ■λ eβx, where λ, β∈R, or equivalently■ ■■ ■f≡0 or±eα+βx,
where α, β∈R. To verify this, we first note that, trivially, all these functions are solutions, since for 0
≤a≤x λeβx·λeβ(a−x)=λ2eβa, Conversely, let fbe a solution to the problem, that is, f(x)·f(a−x) =:c(a) is
independent of x whenever 0 ≤x≤a, and this for any a >0. •Letx= 0. Then f(0)f(a) =c(a). Now let x=a/2.
Then f(a/2)f(a/2) = c(a), too. Hence, for every a >0, (125) f(a/2)2=f(0)f(a). Thus fhas everywhere the
sign of f(0) or f≡0 iff(0) = 0. Since fis a solution if and only λfis a solution, we may assume wlog that f(0)
= 1. Hence, by (125), f >0 on [0 ,∞[. •LetF(x) := log f(x). Then, for any a >0,Fsatisfies on [0 , a] the
functional equation (126) F(x) +F(a−x) =C(a) for some (continuous) function C(a). Now, by (125), C(a) =
2 F(a/2) = F(0) + F(a) =F(a). In particular, for b=a/2, (127) 2 F(b) =F(2b) Moreover, if we take x=a/3,
F(a/3) +F((2/3)a) =C(a) =F(a). Thus, for b=a/3, 0 =F(b) +F(2b)−F(3b) =F(b) + 2F(b)−F(3b), and so (128)
3 F(b) =F(3b). We conclude that for every x >0, and n, k∈N:={0,1,2, . . .}, F■2n 3kx■ =2n 3kF(x), since,
by induction, F(2nx) = 2nF(x) and F(x/3k) =1 3kF(x). Now replace xby 2nx. Since the set {2n 3k:n,
k∈N}is dense in [0 ,∞[ (see e.g. [12, p. 1879]), continuity of Fin [0,∞[ yields that F(µx) =µF(x) for every µ
>0 and x >0. Therefore, for x= 1, F(µ) =µF(1). Consequently, with β:=F(1), f(µ) =eβµ.
274 The following list is ordered according to the ”chronological” appearance in this text. References [1]
R. Mortini, P. Pflug, R. Rupp, Zero sets of a sum of exponential functions Comput. Methods Funct.
Theory (2025) 2, 20 [2] R. Mortini, R. Rupp, All solutions to a Schr¨ oder type functional equation,
Aequat. Math. 98 (2024), 1503–1525. 2, 36 [3] R. Mortini, A nice asymptotic reproducing kernel, Cubo
25 (2023), 441–446. 2, 50 [4] R. Mortini, P. Pflug, A. Sasane, Polynomial Cauchy functional equations:
A report. Elemente der Math. (2025) 2, 57 [5] R. Mortini, When is the product of two planar harmonic
maps harmonic? Lecturas Matem´ aticas 23 (2002), 5–10. 2, 90 [6] R. Mortini, R. Rupp, Sums of infinite
series involving the Riemann zeta function, Archiv. Math. 123 (2024), 163–172. 2, 134 [7] R. Mortini, R.
Rupp, The best H¨ older-Lipschitz constant associated with the function (1 −z)α, Comput. Methods
Funct. Theory 20 (2020), 667–676. 2, 172 [8] R. Mortini, R. Rupp, Some best possible integral
estimates involving Bernoulli polynomials, Archiv Math. 117 (2021),411–422. 2, 263 [9]
https://mortini.perso.math.cnrs.fr/publications2.html 3 [10] Basu, A., Apostol, T. A new method for
investigating Euler sums, The Ramanujan J. 4 (2000), 397–419. 4 [11] Generalized Euler sumP∞
n=1Hn nq, https://math.stackexchange.com/questions/469023/generalized-
euler-sum-sum-n-1-infty-frach-nnq 4 [12] R. Mortini, R. Rupp, Extension Problems and Stable Ranks, a
Space Odyssey, pp. 2198, Birkh¨ auser, Cham 2021. 11, 16, 17, 162, 164, 273 [13] J. Bruna, J. Cufi
Complex Analysis , EMS, Z¨ urich, 2013, 11 [14] R. Burckel, An Introduction to Classical Complex
Analysis , Birkh¨ auser , Basel, 1979 11 [15] R. Burckel, Classical Analysis in the Complex Plane ,
Springer-Birkh¨ auser, New-York, 2021. 11 [16] J.B. Garnett Bounded Analytic Functions Springer,
Springer, New York 2007. 11 [17] G. Polya, G. Szeg¨ o, Problems and Theorems in Analysis I,
Springer, Heidelberg, 1972 11 [18] Th. Ransford, Potential Theory in the Complex Plane Cambridge
Univ. Press. Cambridge 1995. 11 [19] P. Tauvel, Exercices d’analyse complexe Masson, Paris ,1994.
11 [20] Show that holomorphic f1, . . . , f nare constant ifPn k=1|fk(z)|is constant.
https://math.stackexchange.com/questions/289114/show-that-holomorphic-f-1-f-n-are-constant-
if-sum-k-1n-left-f 11 [21] N. Liron, Some infinite sums. SIAM J. Math. Anal. 2, 105-112 (1971). 17 [22]
D. Livshits, W. B. Jordan, Sums of Powers of Roots of a Transcendental Equation: Problem 85-13
SIAM Review, Vol. 28, No. 3 (Sep., 1986), pp. 400-402 17 [23] John William Strutt, Baron Rayleigh
(Lord Rayleigh), The Theory of Sound Second edition, Two volumes in one, Dover Pub. New York,
1945, vol I, original: 1894. 17 [24] Maximon, L.C. The dilogarithm function for complex argument, Proc.
R. Soc. Lond., Ser. A, Math. Phys. Eng. Sci. 459, No. 2039, 2807-2819 (2003). 23 [25] How to use
Laplace Transforms to solve HARD integrals. https://www.youtube.com/watch?v=p8ok5QNNlsc 54 [26]
F.W. Carroll A polynomial in each variable separately is a polynomial, Amer. Math. Soc. 68 (1961), 42
57 [27] K.Knopp, Infinite sequences and series Dover. Publ., New York (1956). 119 [28] R. Remmert,
Classical Topics in Complex Function Theory Springer New York (1998) 119 [29] Representations of
Catalan’s constant D.M. Bradley, 2001 Manuscript on the web. 124, 126 [30] D.H. Lehmer Interesting
series involving the central binomial coefficient, Amer. Math. Monthly 92 (1985), 449–457. 126 [31]
Solving the integral´π/2 0log■ 2+sin 2 x 2−sin 2x■ dx.
https://math.stackexchange.com/questions/3044687/solving-the-integral-int-0-pi-2-log-left-frac2-
sin2x2-sin2x-right-m/3044892#3044892 127 [32] Integral:´π/12 0ln(tan x)dx.
https://math.stackexchange.com/questions/983044/integral-int-0-pi-12-ln-tan-x-dx 127, 128
275 [33] Proving´1 0tanh−1√ x(1−x)√ x(1−x)dx=1 3(8C−πln(2 +√ 3)) for an identity of Srinivasa
Ramanujan.
https://math.stackexchange.com/questions/3800387/proving-int-01-frac-tanh-1-sqrtx1-x-sqrtx1-
xdx-frac13?noredirect=1&lq=1 127 [34] Computing´π 0ln(sin x+ 2)dxand´π 0ln(2−sinx)dx.
https://math.stackexchange.com/questions/4513080/computing-int-0-pi-ln-sin-x2-d-x-and-int-0-
pi-ln-2-sin-x 127 [35] Is it hard to evaluate´∞ 0ln(1−x+x2) 1+x2dx?
https://math.stackexchange.com/questions/4512956/is-it-hard-to-evaluate-int-0-infty-frac-ln-
left1-xx2-right1x 129 [36] https://de.wikipedia.org/wiki/Catalansche Konstante 129 [37] W. Fischer, I.
Lieb, Funktionentheorie, Vieweg&Sohn Braunschweig 1988 132 [38] B. Chakraborty, R. Mortini,
Representations of various power sums, Results Math. 79 (2024), paper No. 186, 16 pages. 151 [39]
S.M. Stewart, A Catalan constant inspired integral odyssey, Math. Gaz. 104, No. 561 (2020), 449–459.
153 [40] Gregory coefficients, https://en.wikipedia.org/wiki/Gregory coefficients 163, 164 [41] I.V.
Blagouchine, Two series expansions for the logarithm of the gamma function involving Stirling numbers
and containing only rational coefficients for certain arguments related to π−1, J. Math. Anal. Appl. 442
(2016), 404–434. 164 [42] E. Landau, D. Gaier. Darstellung und Begr¨ undung einiger neuerer
Ergebnisse der Funktionenthe- orie, Berlin, Springer-Verlag, (1986). 164 [43] A. Stadler. L¨ osung zur
Aufgabe 1431, Elem. Math. 79 (2024), 39–42. 170 [44] Deriving Maclaurin series forarcsin x√ 1−x2.
https://math.stackexchange.com/questions/549028/deriving-maclaurin-series-for-frac-arcsin-x- sqrt1-x2
190, 191, 192 [45] Taylor expansion for arcsin2x.
https://math.stackexchange.com/questions/1448822/taylor-expansion-for-arcsin2x 190 [46] Integral´π/2
0ln(1 + asinx) csc( x)dx.
https://math.stackexchange.com/questions/3262856/integral-int-0-frac-pi2-ln1a-sin-x-cscxdx [47] T.
Apostol, Solution to Problem 1064, Famous Formula, Math. Magazine 53 (1980), 183. 201 [48] Carter,
P., Lowry-Duda, D. On functions whose mean value abscissas are midpoints, with connec- tions to
harmonic functions, Amer. Math. Monthly 124 (2017), 535–542. 213 [49] K. Knopp, Elemente der
Funtionentheorie Sammlung G¨ oschen, Berlin, Leipzig 1937 225 [50] Se´ an M. Stewart, A Catalan
constant inspired integral odyssey, Math. Gaz. 104, No. 561, 449-459 (2020), 229 [51] Is there an
elementary way to find the integer solutions to x2−y3= 1?
https://mathoverflow.net/questions/39561/is-there-an-elementary-way-to-find-the-integer-
solutions-to-x2-y3-1 238 [52] Manasseh Ahmed, The Tangent Line Trick, CRUX 49 (2023), 38–43 242
[53] Mohammad Hossein Rezaei, Solving functional equations in R+using inequalities and sequences
https://www.awesomemath.org/wp-pdf-files/math-reflections/mr-2022-03/ mr32022 solving functional
equations inr+.pdf 245 [54] Problems and Solutions, MEMO 2012 (Switzerland)
https://promathmedia.files.wordpress.com/2013/04/ memo2012 solutions-memo-2012.pdf 245 [55] If
f(x) +f(2x) is continuous, is fcontinuous or not?
https://math.stackexchange.com/questions/3144431/if-fx-f2x-is-continuous-is-f-continuous-or-not 260
[56] Limit question unknown function.
https://math.stackexchange.com/questions/3374236/limit-question-unknown-function 260 [57] Finding
continuous functions. https://math.stackexchange.com/questions/1046961/finding-continuous-functions
260 [58] Continuous functions satisfying f(x) +f(2x) = 0?
https://math.stackexchange.com/questions/1039622/continuous-functions-satisfying-fxf2x-0 260 [59]
Find an example of function wherelim x→0)f(x) +f(2x)) = 0 but lim x→0f(x)■= 0.
https://math.stackexchange.com/questions/3480985/find-an-example-of-function-where-lim-x-to-
0fxf2x-0-but-lim-x-to 260 [60] Find f′(0) if f(x) +f(2x) =x∀x.
https://math.stackexchange.com/questions/2757365/find-f0-if-fxf2x-x-space-space-forall-x 260
276 [61] Function staisfying lim( f(x)−f(2x)) but doesn’t have lim x→0f(x).
https://math.stackexchange.com/questions/2579482/function-satisfying-lim-limits-x-to-0fx-f2x-
but-doesnt-have-lim-li 260 [62] If f(x) =f(2x), then fis differentiable.
https://math.stackexchange.com/questions/3524310/if-fx-f2x-then-f-is-differentiable 260 [63] Proving a
function is constant in Riff(x) =f(2x) and if fis continous at zero.
https://math.stackexchange.com/questions/277313/proving-a-function-is-constant-in-mathbbr-if-
fx-f2x-and-f-is-continuo 260 [64] Functional equation satisfying f(2x) =f(x),
https://math.stackexchange.com/questions/2821984/functional-equation-satisfying-f2x-fx 260 [65] S.D.
Chatterij, Cours d’Analyse, 2 Analyse complexe, Presses polytech. et univ. romandes, Lau- sanne
1997. 272 [66] H. Jolany, An extension of the Lobachevsky formula. Elem. Math. 73, No. 3, 89–94
(2018). 271, 272