Welcome to Scribd!
Academic Documents
Professional Documents
Culture Documents
Hobbies & Crafts Documents
Personal Growth Documents
1 an integer, satisfying Pp +1) +494 1)=n(n +1). Solution Rewrite the equation as P(p+1) = (n-a)(n+9+1). (1) Bither n—g or n+q-+1 must be divisible by the prime p. The first possibility would imply p 2. Equation (1) after reduction by p becomes p +1 = (kp — 2q— 1)k, i., 2kq = (k + 1)(kp— p- 1) (2) The prime g must divide one of the factors on the right. Assuming that q divides kp — p — 1, so that kp— p—1= ql, we get from (2) 2k = (k+ 1), which is impossible for k > 2. Therefore q is a divisor of k +1. Writing k-+1 =m (m> 1 an integer) we obtain from (2) 2k = m(kp — p— 1). (3) Obviously, k is coprime to k +1. Thus & divides the second factor on the right side of (2): kp—p—1 = kr (r an integer); so (3) becomes 2 = mr, which means that m is equal either to 1 or to 2. Accordingly, equation (3) yields either k = (p+ 1)/(p—2) or k = (p+1)/(p—1). The ratio has to be an integer. Hence: In the first case, p = 3 or p = 5. In the second case, p = 2 or p= 3. So there are four subcases possible: Subcase (m = 1, p = 3). Then k = (p+1)/(p—2) = 4, 9 = (k+1)/m and n= kp—q-1=6. Subcase (m = 1, p=5). Then k= 2,q9=3andn=6. ). Then k = (p+1)/(p-1) = 3, = (k+1)/m = 2 5 Subcase (m = 2, p andn=kp—q-1 Subcase (m = 2, p= 3). Then k = 2, q = (k+ 1)/m = 3/2 is not an integer. Thus the general solution is constituted by the following triples (p, 9, n) (3, 5,6), (5, 3,6), (2, 2,3). 6.3 A bounded planar region of area S is covered by a finite family F of closed discs. Prove that F contains a subfamily consisting of pairwise disjoint discs, of joint area not less than S/9. Solution We induct on n, the number of discs in F. For n = 1 there is nothing to prove. Fix n > 2 and assume the statement is true for every bounded planar region covered by a family of less than n discs. Let now F be a family of exactly n discs, covering a region R of area S. Let D be the disc in F of maximum radius r, Assume that D has area less than S/9 (if not, we are done). Then the disc D’ concentric with D, of radius 3r, has area less than S; hence there are points of R outside D’.SOLUTIONS 6.3 ~ 6.5 65 Partition the family F into two nonempty subfamilies F;, Fo, letting F, consist of those discs that have points in common with D and Fy consist of those that do not have (the disc D itself is in F;). Let R; be the union of discs in subfamily F; and let 5; be the area of R; (i = 1,2). Discs that belong to F; are contained in D’, by the maximality of r. Thus Ri C D! and Sy < 5! = area of D’; i.e., $1/9 < area of D. By inductive hypothesis one can select a number of pairwise disjoint discs of family Fo, of joint area not smaller than 52/9. Adjoining to Fo the disc D, we obtain a set of pairwise disjoint discs whose areas sum to a value T'such that T > Sp/9+ (area of D) > S2/9+ 1/9 > S/9, just as needed. This completes induction 6.4 The set N of positive integers has been partitioned into two disjoint sets A, B, their union exhausting all of N. Show that for every n € N there exist distinct integers a,b > n such that the three- element set {a,b,a +5} is contained either in A or in B. Solution At least one of the sets A, B — say, A — is infinite. Fix n € N. We can find in A three numbers ¢, y, z with z>n, y> 2x, z>2y. The six numbers 2, y, 2, y—z, 7—Yy, 2-2 exceed n and are pairwise distinct; this follows from 2 = 22-2 0 for §=1,2,3,4. Let u; =ai/2, v; = Yai, i= 1,2,3,4. The expressions in (1) (case n = 4) become 2 2 (ur + wave + wavs + ugva)” = (aii + aze2 + age3 +0424) = (uP + uh + uf + uf) (v2 + 0} + 03 + v2) = (ajay + ada, + a3r3 + a3x4)(r1 +22 +23+24) = 07, so that (1) turns into equality. Since the case v; = v2 = v3 = v4 = 0 (i.e., 21 = 22 = £3 = 24 = 0) must be excluded, we have aJe=rJ/e for i= 1,2,3,4, for a certain real constant A. Thus a; = ) for all i such that 2; #0. But a1, 2, ag, aq are all distinct, so three of the 2;s must be zero. According to the first equation, the remaining 2; is 1, and then the second equation yields that c equals the corresponding aj. In conclusion, the given system has the desired property if and only if c is equal to one of the four given numbers a1, a2, a3, a4. Note also that the a;s need not be positive. Remark We give a short proof of (1), implying also the condition for equality. If all vjs are zero, there is nothing to prove. If this is not the case, write SHU +--+ +Uprn, usupte tu, vate deb. Then ted if = uv? — 267v + s?v = v(uv ~ 8). os Dou = su)? = 0 ya - 29> i= i= Since v > 0, the claimed inequality s? < uv follows. Equality requires that 0, ie, weedy for i vu; — 504 L...yn, where A=s/v. 6.6 — There are given six straight lines in the space. In every triple of those lines, two lines are perpendicular. Show that the given lines can be labeled Jy,...,/6 in such a way that ty, lo,l3 are pairwise perpendicular, and so are l4, ls, le. Solution Choose one of the given lines ! and consider the following two (comple- mentary) cases:SOLUTIONS 6.6 ~ 6.7 67 (i) There are three other lines !’, !”, !”” perpendicular to !. (i) There are three other lines 1", !”, !’" nonperpendicular to l. In case (i), the triple (W', ",’") contains a pair of perpendicular lines, say, l'L1", Thus the lines J, !, !” are mutually perpendicular. In case (ii), each one of the triples (1,1’,”), (1,1",0"), (1%, 0") contains a pair of perpendicular lines. Since / is not perpendicular to any one of V1", 0”, it follows that U, 1”, l’” must be mutually perpendicular. In each case we have found a triple of pairwise orthogonal lines. Denote them by hy, la, ts, and let ly, Is, ls be the remaining lines. We have to show that IgLisLisLla. Assume that e.g. [, and ls are not perpendicular. Each one of the triples (11, la, ls); (lo, lays), (ls, 445) contains a pair of perpendicular lines. Thus {, is perpendicular to either l4 or {s, and the same can be said of ly and of I. Consequently, either ls or Js is orthogonal to some two lines of Ly, laylg. Lete.g. th, lal ly. Since also Ig lh, Ig. ly, it follows that ls[|l4. Hence the triple (Is, 4,5) contains no orthogonal pair; a contradiction. ‘Thus 41 Js, and similarly 141 le, Is-L le. 6.7 Let Pi, Pz, Ps, Pj be four distinct points in the plane. Suppose [1, In, Ig, Ig, Is, Ig are segments in the same plane, with the following property: every straight line passing through at least one of the points P; meets the union I; U Ip U Ig U 14 U Ig U Ie in exactly two points (we agree that the endpoints of each I; belong to J;). Prove or disprove that the segments I; necessarily form a hexagon. Solution ‘The following example disproves the statement.68 SOLUTIONS 6.7 - 6.8 In coordinates, let Py = (2,0), Pe EC rC—si—Csr—_—CisCSCizsKCzarsSN and look eo ce disjoint triangles ABC and DEF. Take for hh,..., Ie the segments AB, BC, CA, DE, EF, FD. They do not form a hexagon, yet they satisfy the condition of the problem, 6.8 (a) Prove that (2"+!—1)! is divisible by [[(2"**~! — 1)”, for every natural number n. io (b) Define the sequence (cn) by c1 = 1, ¢n = ((4n — 6)/n)cn—1 for n > 2. Show that each cp is an integer. Solution : (a) Write Np =(2"t2=1)!, Dn = PTi(2"t!-#- 1)", Qn = Nn/Dn- Thus Do = 1, D: =3, Dz = 3-7, D3 = 34-77-15, Dg = 38-74-15?-31, and in general, Dn = D,2,(2"+1 — 1). Hence Qn _ NyDn21 _ (art? — 1)! _ ee - ) Qn Nn? (@ =)" = 1) ar is an integer for each n > 1. Since Qo = No/Do = 1, all Qas are integers. (b) As in part (a), binomial coefficients are the clue. The initial cps are: > e=e=1, c=2, 4 =5, cs=14, c6=42, Have you ever met these numbers? Look closely into the Pascal’s triangle and try to see them: 6 dt, 21 | 35) 30) 7 1 de 5) 282 062 70 067 287 8) 1 9 36 84 126 126 84 36 9 1 1 10 45 120 210 252 210 120 45 10 1 ce ne os 42 = 252 — 210, and so on. In general, we guess that fi fan Qn ea =(T)- (0 for all n > 1. Once guessed, this is easily proved by induction, using the recurrence formula that defines the sequence. And since the binomial coefficients are integers, the claim follows.SOLUTIONS 6.8 - 6.9 69 Remark The cps of part (b) are called Catalan’s numbers. A more standard formula to define them is oe = 2” eel (but it does not so readily imply that they are integers). The Catalan numbers occur in many problems of combinatorics. For instance, ¢n41 is equal to the number of paths from (0,0) to (n, n) consisting of n unit steps north and n unit steps east, passing only through lattice points (2, y) with 2 > y. 6.9 To each side of the regular p-gon of side length 1 there is attached a 1x k rectangle, partitioned into k unit cells; k and p are given positive integers, p and odd prime. Let P be the resulting noncon- vex star-like polygonal figure, divided into kp +1 regions (kp unit cells and the p-gon). Each region has to be colored in one of three colors, adjacent regions having different colors. Furthermore, it is required that the colored figure should not have a symmetry axis. In how many ways can that be done? Solution = n Fix the color of the p-gon. Each one of the p cells adjacent to the p-gon can be colored in any one of the two remaining colors; the same regards the next bordering cells, and so on. We arrive at the number of 2? colorings, given the color of the p-gon, disregarding the condition of nonsymmetry. ‘The polygon P has p axes of symmetry. Let us fix one of them, call it 1, and count the number of colorings which are symmetric with respect to I. Once we color the cells of the rectangle bisected by / and the (p—1)/2 rectangles lying on one side of /, the coloring of the remaining rectangles is determined. This can be done in 2*°+1)/? ways. We have to multiply this outcome by p (the number of symmetry axes) and subtract repetitions.70 SOLUTIONS 6.9 ~ 7.1 Note, however, that if a certain configuration is symmetric with respect to two distinct lines, then it is invariant under rotation by some angle 2j7/p, 0 f(n) such that the greatest common divisor of n and f(n) is a maximum Solution Assume that the decimal representation of n uses digits « and y (x # ). Denote the greatest common divisor of n and f(n) by g(n). Then g(n) is also a divisor of n + f(n) = 1111-(e¢+y) =101-11-(2+y). ‘A number of type (eeey)io or (ezyz)10 or (zyez)10 or (zyyy)io or (ezyy)io or (2yye)io is not divisible by 101 (easy exercise). Thus if n is of one of these types then g(n) necessarily divides 11-(z+y) and hence does not exceed 11-17 = 187. Now let us examine the remaining type n = (zyzry):0. Let (zy)10 = a, (ye)io = 6, ged (a,b) =d. So n=101-a, f(n) = 101-6, g(n)=101-d. To maximize g(n), we have to maximize d. From n > f(n) we obtain z > y. Surely d must divide a— 6 = 9(2 — y). If 3 does not divide d, then d has to divide 2 — y; hence d < 9— 1 =8 in that case. Suppose 3 divides d. Then of course 3 divides « + y, so that a = (zy)i0 is one of the numbers 21, 51, 42, 81, 72, 63, 54, 93, 84, 75, 96, 87. The greatest common divisor d of a and 6 = (yz)1o equals accordingly 3, 3, 6, 9, 9, 9, 9, 3, 12, 3, 3, 3; the greatest value d = 12 occurs when a = 84, b= 48. This gives n = 8484, with f(n) = 4848 and g(n) = 101-12 = 1212, the maximum value available for g(n). 7.3 Let n > 2 be an integer and let a, 21,...,2, be positive real numbers. Prove that £ = Gaeaea a+r. 22423 and find conditions for equality. Solution Using (twice) the inequality between the arithmetic mean and the geo- metric mean we obtain (writing tn41 = 71) ins (SS 2 (2D +2) =72 SOLUTIONS 7.3 — 7.4 This is exactly the inequality we wish to prove. Equality requires that each time the means inequality is being used, the averaged terms should be equal. This yields the system of equations ati-®2 tees aen-1 = , qa +22 t2+23 rte ay t22=a2+23= tm ta. (2) If n is odd, equations (2) imply 21 = 22 =- (1) are also satisfied in that case. If n is even, equations (2) force zn; Clearly, equations @pSagS Seq Hie, 25 a45- and then, by (1), For a = 1 the last equation is automatic. When a # 1, we obtain a = f. The conclusion follows: the given inequality turns into equality if either all 2js are equal or, in the case of a = 1, n even, when all even-indexed as are equal and all odd-indexed 2;s are equal. 74 Ay A2AsAqAsAoAr is the regular heptagon and P is a point of its circumcircle, lying on the minor arc A7A;. Show that PA + PAs + PAs + PAr = PAz + PAa + PAs. Solution This standard equality can be justified without much trouble by simple manipulation on trigonometric expressions. A shorter proof is also available: Ag As AsSOLUTIONS 17.4 ~ 7.5 73 Let PA; = 2; for i=1,...,7. Denote the length of the side and the length of the shortest. diagonal of the heptagon by a and 6, respectively: AjAg = ApAg == ArAy =a, Ay A3 = ApAg = ++: = A7An = 6. Applying Ptolemy’s theorem to each of the quadrilaterals PAs—1A:Ai41 (where, cyclically, Ao = Ar, As = Ax) we arrive at the system of equalities = 2,3,4,5,6, br; =az,—-azz7, bry = are —ary. be; =azi-y+azi41 for Thus bay + bas + bas + bez = axe + 2axq + 2azg — az, — azz, bare + bag + beg = az; + 2ars + 2azs + azz. Subtracting these two equalities we obtain (2a-+ 6)(x, +23 + 25+ £7) = (2a+6)(r2 +24 +26), whence zy +egtas+ a7 = 22+ 24426. Obviously, the same argument works for every odd n-gon. 7.5 Let n > 3 be an integer and let a1,...,an, be given nonnegative integers, pairwise distinct. Find all (n + 1)-tuples of nonnega- tive integers (21,...,¢n,), with the greatest common divisor of 21,..+)2n equal to 1, such that 4), + 42%2+-+-++ Anta yr, at) + a3t2+++++ aitq Solution Let 21,...,2n,y satisfy the required conditions. Adding the equations of the system we obtain (a1 +++ + an)(ti +++ tn) = y(ei +--+ 20). The sum 21 +-++-+ tn is positive (otherwise all 2; would have to be zero, contrary to the condition that ged(21,...,2,) = 1). Thus y=arte-tan (a) Rewrite the given system of equations in the form Voatigie = ye, for k=1,...,n; a74 SOLUTIONS 7.5 ~ 7.6 of course the index i+1—k is to be understood (mod n). In view of (1), we conclude that Soa(eigi-e-2e)=0 for R= 1.0.0. (2) i Let zp = min{z1,...,2n}, ty = max{z1,...,2n}. Set in (2), first, k=p, and then k= q: Voaleisi-p— 2) =0, Yo ai(tiging — 2; isi i=1 In the first sum, all summands are nonnegative; in the second one, all summands are nonpositive. Thus all those summands are zero: ai(zigi-p—2p)=0 for #=1,...,0, (3) (4) Since n > 3, we can choose r € {1,...,n}, different from p and q. Setting in (3) i= p+r—1, and in (4) i=g+r—1, we obtain ai(tixi-g—2)=0 for i= Qpyr—1(@r— 2p) =0 and dg 4p—1(Bp — 22) = 0; and setting either in (3) or in (4) i= p+q—1: p4q—1(q — Zp) = ‘The numbers ap4r—1, Gg4r-1, Ap4q-1 are pairwise distinct (by assump- tion). Thus at most one of them can be zero, and therefore at least two of the three differences 2, — 2p, t+-q, t—2p must be equal to zero. Conse- quently, zp = 2, = zy. By the definition of zp and ¢,, this means that all zis are equal, their common value being 1 (because ged(z1,...,2n) = 1). So (21,..y2n)¥) = (1,-0.1, a1 -+-++-a@q) is the unique solution. 7.6 Ina folk dance, the dancers are standing in two rows, n boys facing n girls. Each dancer gives her/his left hand to the person opposite or to her/his left neighbor or to the person opposite the left neigh- bor. The analogous rule applies to right hands. Nobody gives both hands to the same person. Find the number of possible configura- tions. Solution Call the boys, from left to right, Alan, Bill, Charlie, Donald, ...; sup- pose they are facing Ann, Betty, Claudia, Doris, .... Alan joins his left hand with Ann’s right hand. We shall say that Alan’s right hand, Ann’s left hand, Betty’s right hand and Bill’s left hand form a “box” AB. Box BC is defined similarly, with reference to Bill, Betty, Claudia, Charlie; and so on. In each box we have three configurations available. Encode themSOLUTION 7.6 75 by symbols 0, 1, 2 as follows. In box BC: if Bill joins his right hand with Betty’s left hand and Charlie joins his left hand with Claudia’s right hand, we have configuration 0; if the two boys join their hands and the two girls join their hands, this is configuration 1; the remaining “diagonal” configuration is given code 2. In box AB, state 0 is not possible (Alan should not give both his hands to Ann); similarly, state 0 is not admitted in the rightmost box. State 0 in box BC excludes state 0 in box CD (Charlie should not give both his hands to Claudia); the conditions of the problem involve no other restriction concerning relations between these two “boxes”. As there are n — 1 boxes, the admissible configurations are represented by (n — 1)- tuples (“strings”) of entries 0, 1, 2 without any two Os in succession, entry 0 excluded on the outer positions. Denote the number of such strings by up. Obviously u; = 0, u2 = 2. Now consider a string of length n. It could have arisen from any (n—1)- string by attaching either 1 or 2 as the n-th entry; it could have also arisen from any (n — 2)-string by attaching 01 or 02 on the last two positions. The number of admissible (n — 1)-strings equals un; the number of admissible (n — 2)-strings equals u,—;. From each string of any one of these two types, we could have produced a “good” n-string in two ways, by choosing one of the actions described in the last paragraph. All this taken into account, we arrive at the recursive equation Unt = 2tn + 2ttn—1- qa There is a well-known method to handle equations of this type. One postulates tin = an" + by", (2) where (2) and (y") are geometric progressions satisfying the same recur- sion: 2"+1 = 22"422"-!) y+! = oy" 49y"-1, Thus x and y are the roots of the characteristic equation: )? = 21+2; ie, 2=1+¥3,y=1- V3, and (2) becomes Un = a(1 + V3)" +b(1- v3)". Setting n = 1,2 (and knowing the initial data u, = 0, u2 = 2) we obtain a system of two linear equations with unknowns a and 6, from which we determine a = 4(3- ¥3), 6= 4(3 + V3). Thus, finally, 3(3 - v3) (1+ v3)" + (3+ V3) (1 - v3)" BV ((1+- v3)" ~ (1 v8)""). Of course, one can verify directly that the last formula defines a (unique) sequence satisfying recursion (1) together with the initial conditions u; = 0, ug = 2. Un "76 SOLUTION 7.7 7.7 We are given a rectangular array ay Ain Gm +++ mn of real numbers with |a;;| < 1, such that each column sum aij +-+++ am; equals 0. Prove that by rearranging the elements in each column one can obtain an array bu bin bat ses Bonn such that each row sum satisfies |bj) +--+ + bin| <2. Solution Such rectangular arrays are usually called matrices Let A= (aij) cm, jcn be the given matrix and let B= (bis) cm be the matrix whose each column is a permutation of the corresponding column of A and such that, denoting the new row sums by bu ++ bin = 81, the sum |s;|+---+|Sm| is a minimum. (There are only finitely many rearrangements, so the minimum value is attained.) We will show that the required condition |s;| < 2, ..., |sm| <2 is then satisfied. Assume this is not the case and |s,| > 2 for a certain k; there is no loss of generality in assuming just s; > 2, as the conditions of the problem are not affected upon replacing each aj; by —aij. Each column sum (in A, hence also in B) is zero, and therefore $1 +-+-+8m = 0. So there exists an s; < 0. In view of the inequali- ties Se = be +--+ +bin 2>2>0, 81= br t---tbn <0 we have bi- > bir for some r. Write d = bk, — bir. Since all entries are not greater than 1 in absolute value, 0 2>d > 0, we see that |s;| = |sz| — d; and since s; <0 |s;|. Contradiction ends the proof. Remark The constant 2 is the best possible. (For specific values of m an n one can, in general, work out a better bound for the absolute values of row sums available; think of the trivial case m = n = 1 or, slightly less trivial, m =n =2). Yet, the number 2 is the least constant that works for every m and n. In fact, it is the least constant that works for n = 2 and an arbitrary m. To see this, take an odd m and consider the two-column matrix, each column consisting of (m+1)/2 entries equal to (m—1)/(m+1) and (m—1)/2 entries equal to —1. Any rearrangement of terms in columns will produce a matrix with two positive terms in at least one row. The corresponding row sum will be 2(m— 1)/(m-+ 1), approaching the bound 2 as m becomes large. 7.8 Consider the functions fole)=22, fle) = =a for z>1. Prove that for every real numbers a, b with 1 1 and a k-tuple of indices i, i,...,é € {0,1} such that a < fag (Fy y(---Uin())--)) 1 and assume the claim is true for q — 1;78 SOLUTIONS 7.8 — 7.9 that is to say, every dyadic fraction with denominator 2!~? can be obtained in the fashion as described. Take a v = p/2° € (0,1). Either v or 1—v equals half a number u € (0, 1) which is a dyadic fraction with denominator 29-1, Accordingly, either v = go(u) or v = gi(go(u)). Since by hypothesis u can be represented as a finite composition of g:s applied to the number 1/2, the same is true of v, and the assertion for q results; the lemma is proved. Returning to the problem in the original setting, choose an interval (a, ) with 1 R, other than f(z) = 1 and f(x) = z +1, satisfying that equation, if no regularity conditions are imposed on them?80. SOLUTIONS 7.9 - 8.1 The answer is no. The equation (1) in our solution turns out to be equivalent to the system of equations o(z+y)=9(z)+9(y) and —_g(zy) = o(z)a(y). (8) A proof of this fact can be found in the paper by J. G. Dhombres, Relations de dépendance entre les équations fonctionnelles de Cauchy in the journal Aequationes Mathematicae 35 (1988), pp. 186-212 (Th. 5). A function g which satisfies the second equation of (8) has the property that. g(2) = (9(2))’, (9) and therefore it assumes nonnegative values on nonnegative numbers. This, combined with the first equation of (8), implies that g is nondecreasing; and since the first equation leads (in a standard way) to the conclusion that 9(2) = g(1)z on the rationals, the same conclusion is forced on the reals. By (9), a(1) equals 0 or 1. Hence g(2) = 0 or g(z) = 2, hence f(z) must be equal to either 1 or z +1. 8.1 Let a, b, c be pairwise distinct numbers, different from zero and such that a+6+¢=0. Prove that (S52 SY Solution ‘The following identity is easily verified: rr ———=————C—t— + v z wyz Set c= b—c, y=c—a,z=a-—bin (I). In view ofatb+e=0 we have y—z=b+c—2a=~3a, z—2 = —30, r—y = —3c, and (1) becomes =3a bre (~8a)(~38)(—36) ~ b= oNe= a\a=8)' Yabe (b= c)(e— a)(a—b)* aX b_ b= oe b abe Multiplying the last two equalities we get the claimed one.SOLUTIONS 8.2 - 8.3 81 8.2 Suppose that n persons P;, P2,...,P, meet at a party. Assume that P, knows 4 persons, P, knows 5 persons, and so on, Pa—¢ knows n — 3 persons. Further assume that each of Pa_s, Pa—4; Pn3 knows n—2 persons, and each of P,—2, Pa—1, Pa knows n—1 persons. Find all integers n > 8 for which this is possible. (It is understood that “to know” is a symmetric nonreflexive relation: if P; knows P; then P; knows P,; to say that P; knows p persons means: knows p persons other than himself/herself.) Solution Assume n> 10. P; has four acquaintances, Pp has five acquaintances, and Py2, Pai, Pa know everybody. Thus P; knows exactly one person and P) knows exactly two persons other than Px», Pai, Pa. Consider the triple Pa_s, Pa—a, Pas. Each of them misses just one other person. It follows that exactly one person among Pn-s, Pa—4; Pas knows P;, but not P2, while the other two know P2, but not P;. Thus P, and P, have no acquaintances P, with k 10). Hence P3 knows all P,s with k > n — 6 (and there are seven of them). This, however, contradicts the assumption that P; knows exactly six persons at the party. Consequently n = 8 or n = 9. And for these two values of n the conditions of the problem can be realized as follows: Case (n= 8). Ps, Pr, Pa know everybody; Ps, Ps know everybody except Pi; Pa knows everybody except Pz. Then P, knows four people, Pz knows five people, Ps knows six, as needed. Case (n= 9). Pr, Ps, Po know everybody; Ps, Ps know all except Pi; P4 knows all except P,; Ps knows all except P; and P2. This is the required configuration. 8.3 In a convex quadrilateral of area 1, the sum of the lengths of all sides and diagonals is not less than 4 + V8. Prove this. Solution If ABCD is the quadrilateral, write a = AB, b = BC,c = CD, d= DA, p= AC, q = BD; assume that the diagonals intersect under an angle of size y. The (unit) area of the quadrilateral is equal to (pg/2)sing, and hence does not exceed pq/2. Thus pq > 2. ‘Triangles ABC and CDA have areas (ab/2)sin B and (cd/2) sin D, not exceeding ab/2 and cd/2, respectively. So ab d (area of ABCD) < F +5.82 SOLUTIONS 8.3 ~ 8.4 Similarly, 1 = (area of ABCD) = (area of BCD) + (area of DAB) < x + 2, and hence ab+be+ced+da> 4. Consequently, (p +9)? = 4pq + (p — 9)? > 4pq > 8, (atb+etd)? =4(ate)(b+d) + ((ate)—(b+))? 2 4(a+c)(b +d) = 4(ab + be + cd + da) > 16, and so a+bt+c+dt+p+q2>Vi6+ V8 =4+ VB. 8.4 Find all pairs of real numbers z, y satisfying the system of equations ai +y?—ay— fe=0, yt t2?— 25y- By=0. Solution Multiply the first equation by y and the second by z. Then add and subtract the two equations that result: { a +y¥ — fry =0, ( Qety +P — 2ny! —29=0. Equivalently, writing s = 2 +y,q=2y: 4s(s? — 39) —9q =0, ears @) (= P)Qr-1)=0. Ifc = y then the first equation of (1) becomes 823 = 92, yielding « = 0 or z= 9/8. If z # y then the second equation of (2) implies g = 1/2, which inserted into the first equation of (2) gives 8s? — 12s — 0, ie., (2s — 3)(4s? +65 + 3) = 0, with the unique real root s = 3/2. The system tty=s=%, sy=q=}h has solutions z = 1, y= 1/2 and z= 1/2,y=1. Thus the general solution of the original system is constituted by the four pairs (z, y): (0,0), (8,8), (443), G9).SOLUTIONS 8.4 ~ 8.5 83, Remark There are six further solutions in complex numbers, namely (2,58), (30,8), (8a, 8), where a = }(-1+i¥3), 6 = 4(-1—iV3) or vice versa; for details see T. H. Wang’s solution in the problem solving journal Cruz Mathematicorum 17 (1991), 5 (577 King Edward, Ottawa, Ontario, Canada KIN 6N5). 8.5 We are given a certain number of identical sets of weights; each set consists of four different weights expressed by natural numbers (of weight units). Using these weights we are able to weigh out every integer mass up to 1985 (inclusive). How many ways are there to compose such a set of weight sets given that the joint mass of all weights is the least. possible? Solution Let a, b, c, 1 be the weights, from largest to smallest (clearly, the smallest weight must be 1). Suppose there are k such sets. The given conditions imply: a>b>c>l, (1) e (k +1)a, then p =k, and hence m! = m—pa=m-—ka< 1985—ka 10, the equality in (3) forces k = 5,a+b-+c+1 = 397. Conditions (2) then yield ¢ < 6, b < 36, a < 216, whence a+b+c+1 < 259, a contradiction. ‘Try 1986 next, with prime factorization 2-3-331. The sum a+b+c+1 has to be equal to one of the four divisors of 1986 which are not less than 10, namely 331, 662, 993 and 1986. The three largest numbers lead to contradiction, in a manner similar to the case of 1985. So let us try a-+b-+¢+1 = 331, k = 6; conditions (2) become cS7, b<6c+7, a<6b+6c+7. (4) Assuming ¢ < 5 we obtain b < 37, a < 259 and a+b+c+1 < 302, a contradiction again. This leaves 6 and 7 as the possible values of c. Let c = 6. Then by (4) 5 < 43. The values 43, 42, 41 produce the following triples (a, b, c): (281, 43,6), (282, 42,6), (283, 41, 6), all of them satisfying (1), (2), (3) with k = 6. Any 6 < 40 would not do, for then a < 283 (by (4)) and a+b+e+1< 330. Finally, let c = 7. Then by (4) 6 < 49. Values 6 < 39 are eliminated like in the last case. The other ten values of 6 give rise to the ten triples (a, 6,): (323-6, 6,7), b= 40,41,...,49, all of which are good with k = 6 Thus there are 13 ways to compose the sets of weights with properties as required. 8.6 Let P be a point inside a tetrahedron ABCD and let Sa, Sp, Sc, Sp be the centroids (ie. centers of gravity) of the tetrahe- dra PBCD, PCDA, PDAB, PABC. Show that the volume of tetrahedron S4SpScSp equals 1/64 the volume of ABCD. Solution Denote by Ga, Ga, Gc, Gp the centroids of the respective faces BCD, CDA, DAB, ABC, and by G the centroid of ABCD. The centroid of a tetrahedron lies on each one of the four segments joining the vertices to the centroids of the opposite faces and partitions each of these segments in ratio 3: 1. So we have the vector equalities GGA = -} GA, GB, — = (1) GG = -4-GG, GD.SOLUTIONS 8.6 - 8.7 85. ‘The same argument applied to the tetrahedron PBCD yields the equality 5aGa = -4-5,P, which can be rewritten as PSa = 2-PGa. Considering the four tetrahedra PBCD, PCDA, PDAB, PABC we thus have PR =3-PGA, PSp = 3: PGs, (2) PSc =8-PGe, PSp = 8-PGp. Equalities (1) show that the tetrahedron GaGgGcGp is the homothetic image of ABCD; point G is the center of homothety, the ratio equals —}. Similarly, equalities (2) show that S4SgScSp is the homothetic image of GaGaGcGp, with center P and ratio 3. Consequently, tetrahedron S4SpScSp is similar. to ld) |-3|-|3] = 4. Hence the ratio of volumes equals (1)° = & 8.7 Find an upper bound for the ratio 2122 + 2rorg3 + 2324 ap +03+23 +23 over all quadruples of real numbers (21,22, 23,24) # (0,0, 0,0). Note. The smaller bound, the better the solution. Solution It is not hard to work out the exact upper bound for the quantity under consideration; denote this expression by F(21, 22, 23, 24) Write /23 + 23 =u, 2? +23 =0;s0u>0,v>0,ut+v>0. Then Qeorg =u? — (22-23)? < uv? qQ and, by the inequality of Cauchy-Schwarz (see the solution to problem 6.5), 2122 + 23rq < uv. (2) Hence e wun P(21,20,25,24) < BTM =: Olu). @) For u = 0 we have G(0,v) = 0. For u> 0, 1 wou?» du _ iy Btu uy Gun) = ule + u) vtu uo tote Setting w = (v+u)/u, 2 agg Tt S = BVE4 (Vw - Va)? > 23; (4) and hence, in view of (3), Ss 1 _ 1+ 22-20 0°«2 &) F(z1, 22, 23,24) < Gu, v) <86 SOLUTIONS 8.7 ~ 8.9 Equality in (5) is possible; it holds if and only if (1), (2) and (4) become equalities. Equality in (4) requires w = V2, ie., v = (V2 —1)u. Inequalities (1) and (2) turn into equalities when x2 = 3 and 2; = 24. ‘Thus the number 4(1+¥2) is the least upper bound of F(21, 22,3, 24); attained e.g. at 22 =23=1, 2) =2,=V2-1. 8.8 A convex n-gon ApA;...4An—1 has been partitioned into n — 2 triangles by certain diagonals not intersecting inside the n-gon. Prove that these triangles can be labeled Ai, Az,..., An—2 in such a way that A; is a vertex of Aj, for i = 1,...,n—2. Find the number of all such labelings. Solution The labeling with the given property is unique. We will prove this by induction. For n = 3 the claim is obvious. Fix n > 4 and assume the claim is true for any m-gon, 3< m isSOLUTIONS 8.9 - 9.1 87 a rectangle or there is no such side. In the first case, let C be any one of the endpoints Ci, C2. In the second case, the (open) strip limited by the two lines passing through By, By and perpendicular to B; By contains some vertices of the polygon; choose any one of them and denote it by C. Let £; be the portion of the boundary of the polygon, connecting C to By, and let Co be the portion of the boundary, connecting C to Bz (so that B, ¢ Lo, Bo ¢ £1). Let Dy € L; and Dz € Lz be the two vertices adjacent to C (it can happen that D, = By or Dy = Bz). On the internal bisector (6) of angle DiCD» choose a point Q with 0 < QC < min{QD,,QD2}, close enough to C in order that the foot F of the perpendicular dropped from Q to line B, Bz lies strictly between B, and Bo (this property of points on 6 near C is ensured by the definition of C). \. D, 4 BF By Now, let A; be the vertex on £; whose distance from Q is a maximum (Ai # C because QC < QD;). Then all vertices on £;, and also the point F, lie inside or on the circle with center Q and radius QA;. Con- sequently, ray AiQ makes acute angles with the two sides of the polygon adjacent to Ay. Similarly, the vertex Az on C2 with maximum distance from @ has the analogous property. We need a third such vertex, As. But this is easy: just take vertex C to be Ag. Since Q lies on the bisector of angle C, the angles QCD; and QCD2z are acute. 9.1 We are given a non-right-angled triangle A, AzA3. Points Oi, Oo, Os are the centers of circles Ii, In, I's, pairwise tangent (inter- nally or externally), circle I; passing through A» and As, circle passing through A3, Ai, circle [3 passing through Ay, Ap. Given that the triangles A; A7A3 and O;0203 are similar, determine their angles.88 SOLUTION 9.1 Solution Let £A;, £0; (i = 1,2,3) be the angles of the two triangles. Thus we have the set equality {2A,, ZA2, £43} = {201, O02, 403}. Two cases are possible: Case 1. The three circles are pairwise externally tangent. Then each A; lies on the corresponding segment O;—10;41 (numbering modulo 3) and therefore 180° = LO, A3A2 + £A3 + £4, A302 = 4(180° — 201) + 23 + 3(180° — 202), whence £O3 = 180° — 22.43; in general, 2O; =180°-22A;, i= 1,2,3. Assuming (without loss of generality) ZA: < LAz < LAs, we have the inequalities LO, > LO > £Oz, so that LA, = 203, LAr =LO2, LA = LO. This yields the system of equations 180° 224A; = Agi, 7 1,2,3, with the unique solution LA; = £Ag = £A3 = 60°SOLUTIONS 9.1 ~ 9.2 89 Case 2. Some circles are tangent internally. Assume that e.g. I, and T, mutually tangent externally, touch I's internally. Then Oz lies on the segment O3Ay, point O; lies on O3Az, and As lies on 0102, and so 180° = ZO, AgAz + £A3 + LA, A302 = $(201) + LAs + $(202), whence LO3 = 2443 — 180° (thus ZAs > 90° and £03 < ZAs). Now, the size of angle A1 420s can be evaluated in two ways: £AyA203 = (180° — £03) = 180° — ZAs, and £A1A203 = LA, AnAg + LA3A201 = LAz + 3(201). Equating the right sides we obtain £0, = 2(180° — 2A3 ~ ZA2) = 22Ai, and likewise LO) = 244. The obtuse angle ZAg is the largest angle in triangle A1A2A3. Since the roles of A, and Ap are symmetric, we may assume LA; < LAz. Then £0, < LO». Since LO3 < LAs, we infer 402 = LA > 90°. Consequently £02 > ZO, whence ZAz > ZA1. Also, 201 =2LA; > ZAj. It follows that £03 = LA, < LO, = LA2 < £02 = LAs. ‘This together with the former equalities yields the system of equations 224A, =2An, 2LA2= LAs, 24 As — 180° = Z Ai, with the unique solution 2A, =4-180°, LAg = 3- 180°, ZAg = 4 -180°. 9.2 — The monic polynomial P(z)= 2" +an-12"*) +++. +012 +49 of degree n > 1 has n distinct real negative roots. Prove that a1 P(1) > 2n?a9. Solution Denote the roots by —21,...,—@n, so that. P(t)=T]@+2:), 2 >0, iat By Vidte’s formulas, Ils. ag90. SOLUTIONS 9.2 ~ 9.4 Hence, by the arithmetic mean-geometric mean inequality, n>e(Iv)"=9(f8) "=r" the inequality is strict because the y;s are pairwise distinct. For every positive « we have from the binomial formula (1+ 2)" >1+nz+ 4n(n—1)2? > nz (2) (the last inequality is justified by subtracting 2nz from the expression be- tween the two “greater than or equal to” signs and examining the quadratic trinomial that results). Setting in (2) ¢ = 2; (i= we obtain n) and multiplying out the n inequalities P(l)= Te +2) > T=)" = 2nay!”. (3) ist a1 Now it remains to multiply (1) by (3) to establish the claimed inequality. 9.3 Each point in space is colored either blue or red. Show that there exists a square with edge length 1, having exactly 0, 1 or 4 blue vertices Solution Choose any red point P in the space (if there is no such point, there is nothing to do). Consider the sphere S with center P and radius V2. If S is entirely blue, we can easily find a unit square with 4 blue vertices. Thus suppose S has red points and let @ be one of them. Consider the regular octahedron PABCDQ with edge length 1, points P and Q being its antipodal vertices. The unit square ABCD either has 4 blue vertices (and we are done) or it has at least one red vertex; say, A. Then the unit square APCQ has either 0 or 1 blue vertices. 9.4 — Find all triples of positive integers (z, y, z) such that a Solution Suppose (2, y, z) is such a triple. Write = 2*u, y = 2!v, with u, v odd integers. Assuming k > 1, we have 9200 = (gy)? (aly) f#1 = get) where w = (2*~'y)"*? — y*44 is an odd integer greater than 1 (because z+1 > 2). This is impossible, and hence & < I. In a similar way we showSOLUTIONS 9.4 - 9.5 91 that k>l. Sok =1 and we have z = 2'u, y= 2'v, u>v, HED (y= 0) Sout, 50 9300 = (aby)? _ (pty) #2 The sum occurring in the last expression consists of z + 1 odd summands, hence z+ 1 must be even. Write z + 1 = 2t, to obtain 2100 — (atu) — (atv) = 27 (ut — ot) (ut + of) (1) ‘The last two factors have to be some powers of 2; thus ubf-of=2™, ub +ot'= 2", O2; un equivalently, at eget Now we can rewrite (1) as 2200 = 978 (y — v)(u + v) = 22H, Hence 2k +n = 99; and since n > 2, we get 0 << 48. Thus, finally, 99-2k, u= 29-7 41, 298-2k _ 1, and consequently ea y= 298-F 4 OF, y = hy = 298-# — ob , z=1, kaninteger, 0 r. On the other hand, tetrahedron T is also similar to 7’, in ratio 3. Sphere S’ is circumscribed about 7", hence its radius equals 1/3 the circumradius of T: r= R/3. So we get R/r > 3. It is not difficult to see that every number > 3 can be attained as a value of R/r for some tetrahedron T € M. To show this, consider the rectan- gular parallelepiped whose two parallel bases are congruent unit squares ABCD and A'B'C'D’, distance t apart (t = AA’ = BB’ = CC’ = DD’, a varying real parameter). Let T be the tetrahedron AB’CD’. Its circum- center and incenter coincide with the center of symmetry of parallelepiped ABCDA'B'C'D’; thus T is in class M. The circumradius and inradius are easily calculated: _V2+? tSOLUTIONS 9.6 ~ 9.7 93 ‘The ratio R/r = (1/t),/(2+)(1 +20) is a continuous function of the variable t, attaining value 3 at t = 1 (T' then becomes the regular tetra- hedron) and tending to infinity as + 0 or ¢ —+ 00; hence it attains every value > 3. So the range in question is the interval [3, 00). Remark It follows from the solution that the bound R/r > 3 is valid for every tetrahedron, not necessarily from class M. 9.7 Let k and n be integers with 0 < k < n?/4. Assume that k has no prime divisor greater than n. Prove that n! is divisible by k. Solution Let p be any prime divisor of k; then p < n by hypothesis and so p is also a divisor of n!. Suppose p enters k and n! in powers a and f, respectively (that means, k is divisible by p®, but not by p*t, and n! is divisible by p®, but not by p*t#). It will be enough to show that a < . Let m be defined by the condition p” 4k > 4p*, (2) and hence n > 2p%/? > 2p7. This in view of n < p™+} leads to the inequality prtinr 52, Therefore . if p=2, then y3, then y 2p, which, inserted into formula (1), implies >See] =0 46, 4) i=l Skee"), where94 SOLUTIONS 9.7 ~ 9.8 Note that (in view of (3)) s={j if p=2, 0 if p>3, while a= ado > 2c at 2 >; fa the last inequality is ee for every integer y > 2. Thus by (4) we have B22; (5) equality could only (perhaps) be possible when y < 1 and p> 3. If y= 0, then (5) is certainly strict (@ > 1 because p 2 unless y=1 and p>3. Now, if a is even (a = 27), then by (5) 8 > a, as needed; if a is odd 2y + 1), but strict inequality holds in (5), we are done, as well. It remains to consider the case where y = 1, p> 3 anda =2y+1=3. From (2) we have n? > 4p, i.e., n > 2p9/?, and hence n/p > 2p'/? > 3. Returning to (1) we obtain # > {n/p] > 3 = a and the proof is complete. Remark Here is a proof of the Legendre formula (1). Obviously, # is the sum of the exponents with which p enters 1,2,3,...,n. Write g; = [np7'] for ,m+1 (thus gm41 = 0). Note that, ‘for each i < m+ 1, the set ,n} contains exactly g; integers divisible by p'. So there are exactly Gi—4i41 integers divisible by p', but not by p'+? (in this set). Consequently, B= (q1 — 92) 1+ (92 — 93) 2+ +++ + (Gm — dmti)-m Sn tate + amy as claimed. 9.8 Pairwise distinct real numbers are arranged into a rectangular array with m rows and n columns. In each row the entries are arranged increasingly (from left to right). Each column is then rearranged in decreasing order (from top to bottom). Prove that in the reor- ganized array, all rows are again arranged increasingly. Solution Suppose, to the contrary, that there exist indices j € {1,...,m} and kl {1,...,n}, & < 1, such that the jth row of the new array forms a sequence (21,...,2n) with zz > a). Let () ~ ()SOLUTIONS 9.8 - 9.9 95. be the & th column and the /th column of the original array; by assumption, a > Bp(m) 5 Bacay > +++ > Byte) - Thus, if ie {1,. if ie sd}, then apc: > apy) = te > 215 ym}, then ayy < bai) < by(j) = 1 It follows that the m-element set {a1,...,@m} contains at least j numbers greater than z; and at least m—j +1 numbers smaller than 2). This is obviously a contradiction. 9.9 Determine all continuous monotonic functions f : R —+ R. which satisfy f(1)=1and f(f(c)) = (f(z) for all 2 eR. Solution Denote by J the set of all values taken by f. Since f is continuous and monotonic, J is an interval (finite or not). The equation imposed on f can be rewritten as f(z)=2? for zed. Clearly, 1 = f(1) € J. Suppose that J contains a negative number u. Then [u, 1] is a subinterval of J; a contradiction because the squaring function is not monotone on such an interval. Thus J C [0,0o), which means that f(c)20 forall 2eR. Choose any a € J. Then a? = f(a) € J, a4 = f(a?) € J, and so on, a” € J for n = 1,2,3,.... Ifa < 1, this sequence tends to zero (and thus 0 J, by the continuity of f); if a> 1, this sequence tends to infinity. In other words: if J contains numbers smaller than 1, then 0 is the left endpoint of J, and if J contains numbers greater than 1, then oo is the right endpoint of J. Thus J is one of the following four sets: {1}, (0, 1), [1, 09), [0, 00).96 SOLUTIONS 9.9 - 10.1 Accordingly, f is given by one of the following four formulas: f(@e)=1 forall ceR, 0, 2<0, see)= $2?) 0Se<1, aie ata 1 a 4. Write p instead of 1987. (In general, p may be any prime such that ged(n, p— 1) = 1; p = 1987 has this property, as 1986 = 2-3-331, while all prime divisors of n have an even number of digits by assumption.) Suppose, contrary to the assertion, that the equality P(x) = P(y) holds for some rationals z = a/c, y = b/d, # y; assume that a,b,c,d are integers with c,d > 0, ged(a,c) = ged(b,d) = 1. Equality P(x) = P(y) says that d°(a" — pac?) = c®(6" — pbd”~). Q) The expressions in the parentheses on the left side and on the right side of (1) are coprime to ¢ and d, respectively; so c = d, and hence a ¢ 8. Equality (1) becomes b" — a" = p(b— ajc"). (2) Division by 6 — a yields (3) It follows from (2) and (3) (a being coprime to c) that a and 6 are nondivisible by p. Thus be=1 (mod p) (4) holds for some e € {1,2,...,p— 1}. Hence by (2) (ae)" = a"e" = bre" = (mod p). (5) Two cases are now possible. Case (ae #1 (mod p)). By (5), ae # 0 (mod p), and hence, by virtue of Fermat’s theorem (see the Remark, below) (aeP"!=1 (mod p). (6)98 SOLUTION 10.2 Let @ be the least positive exponent for which (ae) = 1 (mod p); clearly, @ > 1, and it follows from (5) and (6) that a is a common divisor of n and p—1; acontradiction. Case (ae = 1 (mod p)). Then by (4) a=b#0 (mod p). (n Thus equality (3) yields na”~! = 0 (mod p), whence n = 0 (mod p). Since nis a perfect square, n=0 (mod p?) (8) Write, in agreement with (7), a = kp+r, 6 = Ip+r. By the binomial formula, a’ =ikpr’}4r! (mod p*), = jlpri-t +7 (mod p?), and hence abl So the sum on the left side of (3) is congruent modulo p? to n(n—1) 2 ik + j)prit3-) 499 (mod p?). act Yo (ikt (n= 1- lpr? turd = (k + prt? + nr? io In view of (8), this number is divisible by p; so the product on the right of (3) must be divisible by p?, which means that p divides c. Thus (2) becomes oa" =(b—a)p"w, —_w an integer. (9) Assume that p enters the difference b— a in a power m > 1 (see (7)): b-a=p™q, m2>1, q#0 (mod p). (10) Now, let s be a positive integer not divisible by p. We claim that for every integer z > 0 there exists an integer t, with ote? — gt?’ = p™t4,, — t, #0 (mod p). (11) We induct on z. For z = 0 we have by (10) and the binomial formula bf —a® = (a+ qp™)* —a* = sa‘~!qp™ + Ap’, Aan integer; so we can set to = sa*'qg-+ Ap™ #0 (mod p). Fix z > 0; assume ¢, exists and (11) holds. Consider z + 1: perth _ gspttt _ (ory? _ (a?*)? = (a°** +0™4,)? — (ay? P sy () aiF (Pi) lms) gh FeSOLUTIONS 10.2 ~ 10.3 99 In this sum we isolate the first term and the last term; all the other sum- mands are divisible by p!+2("+#) (because the binomial coefficients are divisible by p). So we have fe pat? (P-Lpmtz4, 4 Bpltomt2 4 port? ae , B an integer, and it suffices to set tog = at ODE, + Bp™* 4 l™*NP-“D-1 40 (mod p) to obtain representation (11) for z + 1 in place of z. This completes the proof of claim (11). Write the given integer n in the form = ope eee is 0, ceeesodl isie10.\(modap): (12) Apply equality (11) to these specific values of s and z (and the suitable t,). Equating the right sides of (9) and (11), and using (10), we obtain pt, =p" qu. Since t, # 0 (mod p), this implies n < z, while (12) forces n > p* > z. This is the desired contradiction, showing that the equality P(z) = P(y) cannot hold for any distinct rational numbers « and y. Remark Fermat’s “Little” Theorem says that if p is a prime and w is an integer nondivisible by p then u?-? = 1 (mod p). Here is an outline of a proof: it is enough to consider u € {1,...,p — 1}. Induct on u, expanding (u + 1)? binomially and using the fact that the binomial coefficients @), O0 0. Suppose there exists a positive integer m such that the difference tm — 20 equals k, also an integer. Show that the limit lim(z,/n) exists; express its value in terms of the given data. Solution ‘The functional equation satisfied by f implies f(et+)=f(z)+l for ceR, lez (1) (use induction on I = 0,1,2,...; the case of negative | follows easily from that for positive !). A next quick induction proof shows that tngm=@ntk for n=0,1,2,... (2) (induct on n starting from tm = 20 +k and using (1)).100 SOLUTIONS 10.3 ~ 10.4 The third claim is eee ny | toca Oy, (3) Considering r to be fixed, induct on q. For q = 0, (3) holds. Assume (3) for some q > 0 and pass tog +1: Lrpm(g41) = Lrtmgem = Cr¢mg th = ar thgtk=2p+k(q+1); we have used (2) with n = r+ mg, and the inductive assumption. Thus, (3) follows. From (3) we are able to derive the solution of the problem. Take any positive integer n; divide it by m, denote (the integer part of) the quotient by q(n) and the remainder by r(n). Then by (3) Zn _ _Tr(n)tmaln) _ Er(ny than) _ k 1+ an m n r(n)-+mg(n) r(n)+mg(n) m 1+f,’ where (n) = Een) = 2th w= Fain)? = agin As n goes to infinity, q(n) does so too, while the numerators take only a finite number of values. Hence a + 0, Bn — 0, and so 10.4 Does the set {1,2,...,3000} contain a subset A consisting of 2000 numbers such that 2 € A implies 22 ¢ A? Solution Call a set A with this property double-free. Consider the problem with 3000 replaced by arbitrary N. For every positive odd integer q consider the sequence % 24, 49, 84, .-- qa Out of each pair of consecutive terms of such a sequence, only one can belong to a double-free set. Thus in order to construct a double-free set Amax C {1,2,...,N} with a maximum number of elements, we can assign to Amax every second entry of each sequence (1), starting, for instance, from the least one. So we may define Amax = {n= 4g: n0 integers, g odd} (note that, in general, the set Amax thus defined is not the unique double- free set in {1,..., NV} of maximum cardinality).SOLUTIONS 10.4 - 10.5 101 For a fixed k, the set {1,...,.N} contains [(4-*N + 1)/2] numbers of the form 4*q, with q odd. Thus, if m is such that 4" < N < 4™+1, then the number of elements of Amax equals Solent + yy. @ es For N = 3000, this sum evaluates to 1999. So there is no 2000-element double-free subset of {1,2,...,3000}. Remark If N= oad (4 € {0,1}, =o of N, then the maximum cardinality of a double-free subset of {1,..., N} equals sy) is the binary representation f= 24 Ey ei(-1y (3) = To see that the formulas (2) and (3) indeed define the same number, denote the expression in (2) by g(N), assume (as before) 4" < N <4™+1, m>1, and write N=K-.4"+M with K € {1,2,3}, 0 dy, S contains pairs of points at mutual distance dj; in every such pair, at least one point does not belong to Ay. So we may find a point PE SM Aa Now consider the sphere $’ of radius dz, centered at P. According to condition (1), S’ C A, U Ag. Hence SAS'CAL (2) The intersection SS’ is a circle; denote its radius by r. For every point @ of this circle, OPQ is an isosceles triangle with OP = OQ = ds, PQ = do < dg. Hence 60° < OPQ < 90°, and so r= PQ-sinLOPQ > (V3/2)d2 > do/2, showing that the diameter of circle S'S’ exceeds dp. Thus we can find a pair of points of S175! at distance d;. And this is a contradiction to (1) and (2) 10.6 Let C be the circle with radius 1 and let n > 1 be a fixed integer. For any set A of n points Py,...,P, on C' define D(A) = max(min6(P,,d)) where 6(P,1) denotes the distance from point P to line | and the maximum is taken over all diameters d of circle C. Let F, be the family of all n-element subsets AC C and let Dn = min D(A). AOR, Calculate D, and describe all sets A € F, with D(A) = Dn. Solution For a point P € C denote by P’ the point antipodal to P; for a set ACC denote by A’ the antipodal image of A (ie. A’ = {P's P € A}).Se’ \7 7 Take aset A= {Pi,..., Pn} € Fn. The set AUA’ consists of 2m points, m sin(x/2m) > sin(x/2n). (2) The first inequality in (2) becomes an equality if and only if all ares in A are equal; i.e., when AU A’ is the set of vertices of a regular 2m-gon. The second inequality in (2) is an equality for m =n, ie., when A and A’ are disjoint. The conclusion follows: Dy = juin D(A) = sin(x/2n). The minimum is attained for every set A consisting of some n vertices of a regular 2n-gon inscribed in C, without a pair of antipodal vertices.104 SOLUTION 10.7 E 10.7 For any natural number n = J a;10° (where a; € {0,1,...,9}, i=0 a, # 0) write i Var-s10 i=0 & Sa, 2 iso Consider the set P={n: n=n*, p(n)/3=s(n)-1} and let Q be the set of all numbers in P with all digits greater than 1. (a) Show that P is infinite. (b) Show that @ is finite. (c) Write down alll the elements of Q. Solution When n= n*, we shall call na palindrome. (a) For any natural & > 2 let k’ = (3-1 — 6k + 1)/2 (an integer). ‘Write a string of k’ ones followed by 2k threes and k’ ones, and denote the resulting palindrome by n. Then s(n) = 6k + 2k’ = 3-141, p(n) = 37, son € P. Since k might be an arbitrary integer > 2, we obtain an infinite sequence of elements of P. (b) Suppose n € Q is an m-digit number. The product p(n) is divisible by 3; therefore at least one digit is > 3. Hence p(n) > 3-277, s(n) < 9m. The equation defining the set P now implies the inequality 2"-! < 9m—1, and so m < 6. (c) According to the conclusion of (b), every element of Q has one of the following forms: (zyzzyz)io, (zyzyz)io, (zyyz)o, (zur), (z)0, (2), with 2,y,z € {2,3,...,9} Case 1. n=(zxyzzyz)19. The equation of P takes the form 2y?2” = 6(z@ +y+2)—3. () Then «, y, z are odd integers; hence they exceed 2; the product on the left side of (1) is > 3° = 729, the right side of (1) is < 6(9 +9 +9) —3 = 159; no solutions.