Don't Try to Solve These Problems

Author(s): Richard K. Guy

Source: The American Mathematical Monthly, Vol. 90, No. 1 (Jan., 1983), pp. 35-38+39-41
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2975688 .
Accessed: 20/07/2013 08:22

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.

http://www.jstor.org

This content downloaded from 171.67.34.69 on Sat, 20 Jul 2013 08:22:21 AM

All use subject to JSTOR Terms and Conditions
1983] UNSOLVED PROBLEMS 35

48. C. D. Olds, ContinuedFractions,New MathematicalLibraryNo. 9. MathematicalAssociationof America,

Washington,D.C., 1963.
49. G. M. Phillips,Archimedesthenumericalanalyst,thisMONTHLY,88 (1981) 165-169.
50. C. W. Puritz,An elementary methodof calculatingr, Math. Gaz., 58 (1974) 102-108.
51. E. Salamin,Computationof v usingarithmetic-geometric mean,Math. Comp., 30 (1976) 565-570.
52. H. C. Schepler,The chronologyof pi, Math. Magazine,23 (1950) 165-170,216-228,279-283.
53. L. Schlesinger,Uber Gauss' Jugendarbeitenzum arithmetisch-geometrischen Mittel,Jber.Deutsch. Math.-
Verein.,20 (1911) 396-403.
54. D. Shanks,Review 15, Math. Comp., 31 (1977) 1044.
55. _ _, Quartic approximations for g, Abstract No. 80T-A183, Abstracts Amer. Math. Soc., 1 (1980) 558.
56. D. Shanksand J.W. Wrench,Jr.,Calculationof v to 100,000decimals,Math. Comp., 16 (1962) 76-99.
New York, 1969.
57. C. L. Siegel,Topics in ComplexFunctionTheory,vol. 1, Wiley-Interscience,
58. H. C. Thacher,Jr.,Iteratedsquare root expansionsfor the inversecosine and inversehyperboliccosine,
Math. Comp., 15 (1961) 399-403.
59. J.Todd, A problemon arc tangentrelations,thisMONTHLY,56 (1949) 517-528.
60. , The lemniscate constants, Commun. ACM, 18 (1975) 14-19, 462.
61. , Basic Numerical Mathematics, Academic Press, New York, 1981.
62. H. W. Turnbull,Editor,JamesGregoryTercentenary MemorialVolume,publishedfortheRoyal Societyof
Edinburgh,London, 1939.
Math. Gaz., 17 (1933) 5-17.
century,
63. G. N. Watson,The marquisand theland-agent;a tale of theeighteenth
century,ArchiveforHistoryof
64. D. T. Whiteside,Patternsof mathematicalthoughtin thelatterseventeenth
Exact Sciences,vol.1, 1960-1962.
65. J. W. Wrench,Jr.,The evolutionof extendeddecimal approximationsto v, Math. Teacher, 53 (1960)
644-650.

UNSOLVED PROBLEMS
EDITEDBY RICHARDGUY

In this departmentthe MONTHLYpresentseasily stated unsolvedproblemsdealing withnotionsordinarily

encountered mathematics.
in undergraduate Each problemshouldbe accompaniedby relevantreferences (if any
are knownto the author)and by a briefdescriptionof knownpartial results.Manuscriptsshouldbe sent to
RichardGuy,Department of Mathematics of Calgary,Calgary,Alberta,Canada
and Statistics,The University
T2N IN4.

DON'T TRY TO SOLVE THESE PROBLEMS!

RICHARDK. GUY
ofMathematics
Department ofCalgary,Calgary,Alberta,Canada T2N 1N4
and Statistics,The University

Such an exhortation but I'm serious,and I'll explain

will likelyproducetheoppositeeffect,
why.Thisarticlehas beenin mindforsometime,butitseruptionis triggeredby a proposalfrom
Department
SchmuelSchreiber, and Computer
ofMathematics
Ramat-Gan,Israel.
Science,Bar-IlanUniversity,
Problem0. For an integera definethesetSa inductively
by
(1) a E Sa, (2) if k E Sa, then2k + 2 E Sa, (3) if k E Sa, then3k + 3 E Sa
definea function
Equivalently, by
sa(n) on theintegers

(1) Sa(l) = a, (2) Sa(2k) = 2Sa(k) + 2, (3) Sa(2k + 1) = 3Sa(k) + 3.

For a <-3 Or does Sa containrepeatedelements?

or a > 2 is Sa injective?

This content downloaded from 171.67.34.69 on Sat, 20 Jul 2013 08:22:21 AM

All use subject to JSTOR Terms and Conditions
36 RICHARD K. GUY [January

Some of you are alreadyscribbling,

in spiteof thewarning!More cautiousreadersmayhave
been reminded
of otherproblems, perhapsone or moreof thefollowing.
Problem1. The diophantine equationa2 + b2 + c2 = 3abc has thesingularsolutions(1, 1,1)
and (2, 1,1). Othersolutionscan be generatedfromthese,because theequationis quadraticin
each variable,forexample,b = 2, c = 1 givesa2 - 6a + 5 = 0, a = 1 or 5 and (5,2,1) is a
solution.Each solution,apartfromthesingularones,is a neighbor ofjust threeothers,and they
forma binarytree.Is thisa genuinetree,or can thesamenumberbe generated by twodifferent
routesthrough it?
Problem2. Considerthesequencea,+1 = an,/2(an even),an+l = 3an+ 1 (an odd). For each
positiveintegera1 is therea valueof n suchthatan = 1?
Problem3. Considerthemapping
2m ---3m, 4m -1I-) 3m -1, 4m + I1-) 3m + 1.
This generatesthecycles(1), (2,3), (4,6,9,7,5) and (44,66,99,74,111,83,62,93,70,105,79,59).
Are thereothers?
Problem0 can be visualizedas a binarytree generatedby the pair of unaryfunctions
a -- 2a + 2, a -- 3a + 3. For example,ifa = 1, we have Figure 1.

4 '3

/10 /1>

FIG. 1. Binarytreegeneratedby two unaryfunctions.

The number66 appearstwicein Figure1,bymakingthreestepsto theright, or bymakingone

to theleft,one to therightand twoto theleft.A rightstepmultiplies
by 3 (roughly);a leftstep
multipliesby2 (roughly);thecoincidenceis roughly
explainedbytheapproximation: tworight
threeleft;32 23.Is thisanotherexampleofthestronglaw ofsmallnumbers[11]?Ifwe tryother
smallvaluesfora, we findsimilarcoincidences(Figure2).

73X) c3f-1 0

+ 3 (
, ~~~~~
~~14\- ~

FIG. 2. Small values of a lead to coincidences.

Let us look at a = -4. We'veomittedtheminussigns;alternatively,

changetheplus signsto

This content downloaded from 171.67.34.69 on Sat, 20 Jul 2013 08:22:21 AM

All use subject to JSTOR Terms and Conditions
1983] UNSOLVED PROBLEMS 37

minusesin each of conditions(2) and conditions(3) in Problem0. The binarytreeis now as in

Figure3.

10
s
I8 Z 28 \4Z 30\ 46 69
FI51 3. 54Biar1 82 123 58g e8 20
1bya 1 136204-
d

FIG. 3. Binarytreegeneratedby a --) 2 a - 2 and a3-3a - 3.

The numbersthatappear,whenarrangedin numerical

order,are
4, 6, 9, 10, 15, 16,18,24, 27,28, 30,34,42, 45, 46, 51, 52,54, 58, 66, 69, 78, 81, 82,
87, 88, 90, 99, 100,102,106, 114,123,130, 132,135, 136,150,153, 154,159, 160,
162,171,172,174,178,195,196,198,202,204,210,....
Does a numberever occur twice?I sent an earlierdraftof this articleto JohnLeech, who
disobeyedorders,startedscribbling,
and foundthathad I continuedFigure3, I would have
discovered258 and 402 on thenextrow,withbothof theserepeatedtworowsfurther down.In
fact,reinstating
theminussignsand usingL fora leftstepand R fora rightone,we can write
thesecoincidencesas
-4L7 -258 = -4RL2R2 and -4L2RL4
= = -402 = -4R2LR2
and thosestarting
from1 and 0 in Figures1 and 2 as
1LRL2 = 66 = 1R3 and OL4 = 30 = OLR2.
The last coincidencestartswitha leftstepin eithercase, so can be regardedas a coincidence
with2: 2L3 = 30 = 2R2. Leechusedhis computer
starting to findcoincidencesstartingwith3, 4
and 5:
3LRLRL'0 = 177150= 3R7LR2
4L4R8 = 626574= 4RL7RLRL4
SL2RL2R6= 241662= 5R2L12

Since3R = 12 = SL and 3L2 = 18 = SR, thislastgivesanothercoincidencestarting

from3. We
have seencoincidences from- 1, -2, - 3 and - 4; Leech also found
starting

-5L3R4= -1986 = -5R2L6

- 7LRL2RL3RL4= - 143250= -7R2L2R6
-8LRL8 = -9474 = -8R4LR2

-9L'0 =- 7170 = -9RL2R4

- IOL3RL6 = - 12354 = - IOR2LR4
-1 1L5R2L6 = -166146 = - 1IR4LRL2R2

This content downloaded from 171.67.34.69 on Sat, 20 Jul 2013 08:22:21 AM

All use subject to JSTOR Terms and Conditions
38 RICHARD K. GUY [January

(1, I,i)

(2,,1,1)

(521)

(29,5,2) (13,5,1)

FIG.4 .Binary treeo a94rk5) (tp,13,s )

FIG. 4. Binarytreeof Markofftriples.

and conjecturesthattherewillalwaysbe "duplicatefruitson thetrees"no matterwhereyou start.

If you wanta coincidencestartingwith- 6, note that - 6R = - 10,but it wouldbe cleanerto
findone of theform- 6L... = - 6R.... and I'm temptedto strengthen to:
Leech'sconjecture
C For eachinteger L... and R... suchthatnL... = nR...?
n, therearestrings

so don't botherto solve this

Stoppress! In yet anotherletter,Leech proveshis conjecture,
problem!
In problem1 a binarytreeis similarly
generatedby thepairof ternaryfunctions
(a, b, c) -- (3ab - c, a, b), (a, b, c) -- (3ac - b, a, c)
as in Figure4.

433 169
/\
+~~~-
m\ 5d /w\
5144 I1351?5184 907? 255

/\ ~ ~ /

44&L
1090

FIG. 5. Simplified tree.

Markoff

This content downloaded from 171.67.34.69 on Sat, 20 Jul 2013 08:22:21 AM

All use subject to JSTOR Terms and Conditions
1983] UNSOLVED PROBLEMS 39

,50
152, 25
A"

22+ 37 612
36At 36*"O

Is
B 116 M9
56 58
\2ss 29s 176 ISO 3
01,
* 176
>N0 15
Af 45 272
136 2K ?40
11s 68 \
3Z
192

48 52.5 2
\2+ a26 160 iet68

106/
1~20 21 128
3N'*0.0 ~ "'16+1110

\2.
FIG.6. Does theCollatzalgorithm
giveanymorecycles?

We can exhibitmoreofthetreeby simplifying it as in Figure5. To recapture

thetriplesfromthis,
chooseanyentryfora, and itsimmediate predecessor forb. Thenc is foundwhentravelling up
the tree,just afterthe firststep afterthe firstrightward step. For example,a = 985 has
predecessorb = 169.Whentravelling upwardsfrom985,thefirstrightward stepis from29 to 5.
The nextstep is from5 to 2, so c = 2.
Whether
or notthereare repetitions
in thesequenceof Markoff
numbers
1,2,5,13,29,34,89,169,194,233,433,610,985,1325,1597,2897,
4181,5741,6466,7561,9077,10946,14701,28657,37666,51641,...
has becomea notoriousproblem.Thereare occasionalclaimsto have proveduniqueness,but
none seem to hold water[19]. Don Zagier [23] has some resultson distribution,but none on
distinctness.
He can showthattheproblemis equivalentto theunsolvabilityofa certainsystemof
diophantineequations,so we maybe nearingtherealmofHilbert'stenthproblem.Hencethetitle
of thispaper.
Problem2 is associatedwithvariousnames:Collatz,Hasse, Kakutani,Syracuse.It is just as
notorious.LotharCollatztoldme thathe thought ofit whena student.One ofitsseveralwavesof
popularitystartedwhenhe mentioned it to severalpeopleat the1950InternationalMathematical
Congressin Cambridge(thewrongCambridge).Presumably somemathematicians fromSyracuse
(thewrongSyracuse)becameinterested in it; theboysfromSyracusecan perhapsfillin thatbitof
history.

This content downloaded from 171.67.34.69 on Sat, 20 Jul 2013 08:22:21 AM

All use subject to JSTOR Terms and Conditions
40 RICHARD K. GUY [January

Is thegraphof theCollatzsequenceunicycic?Figure6 includesall thenumbersup to 26; the

branchcontaining 27 is a muchlongerone,but stillcomesdownto 1 after111 steps.
Aftera long and inconclusive correspondence some yearsago, a claimantto have a proof
eventuallyadmitted that" Erdossaysthatmathematics is notyetripeenoughforsuchquestions."
Hence thetitleof thispaper.
Problem3 is one ofJohnConway'spermutation sequences.It is similarto theCollatzproblem,
butherethefunction has an inverse
3m -2m, 3m-1 -4m-1, 3m + 1 -4m +
(ifthenumber'sa multipleof 3, takea thirdoff;otherwise add a thirdon) so thesequencecan be
pursuedin eitherdirection.Its graphconsistsof a numberof disjointcyclesand doublyinfinite
chains.But it hasn'tevenbeen provedthatan infinitechain exists!Whatis the statusof the
sequencecontaining thenumber8?
.,41,31,23, 17,13,10,15,11,8,12,18,27,20,30,45,34,51,38,57,43,32.
Whatgivesa cycle?Each termis either3/2 timesthepreviousone,or approximately 3/4 ofit.
Our bestchanceof getting back to an earliervalue is to finda powerof 3 whichis close to a
powerof2. The knowncyclesoflengths1,2, 5 and 12 correspond to theapproximations of 31,32,
35 and 312 by 22, 23, 28 and 2'9. The last is the ratioof D sharpto E flat!In factin each of
problems0, 2 and 3, theconvergents
1 2 3 8 19 65 84 485 1054 24727 50508
1 1 2 5 12 41 53 306 665 15601 31867
to thecontinuedfractionforlog3 to thebase 2 are of significance.Note thatthereare cycles
corresponding 1,2, 5 and 12.It has been shownthatthereare noneoflength
to thedenominators
41, 53 or 306. Computerscan pushnumerical resultsquitea longway,butit'snotclearthatthey
can be of anyuse withsuchproblems.
In [5] Conway relatesfamiliesof sequencessimilarto that in Problem3 to the vector
reachabilityproblemand Minskymachines.Hence thetitleof thispaper.
AfterI'd written
Postscript. problemof David A. Klarnerreachedme from
this,thefollowing
twodifferentdirections. similarities.
Thereseemto be somesignificant
Problem4. Let S be thesmallestsetofnumberssuchthat1 E S and ifx E S, then2x, 3x + 2
and 6x + 3 eachbelongto S, i.e.,
S = {1,2,4,5,8,9,10,14,15,16,17,18,20,26,27,28,29,30,32,33,34,...}.
Thatis, as n -- oo,is
Does S havepositivedensity?
liminfIS n (1,2,..., n}I/n> 0?

References

1. Michael Beeler,WilliamGosper,and Rich Schoeppel,Hakmem,Memo 239, ArtificialIntelligenceLabora-

tory,M.I.T., 1972,p. 64.
2. CorradoB6hm and Giovanna Sontacchi,On theexistenceof cyclesof givenlengthin integersequenceslike
xn+ I = xn/2 ifxn even,and xn+ AttiAccad. Naz. Lincei Rend. Cl. Sci. Fiz. Mat. Natur.(8)
= 3xn + I otherwise,
64 (1978) 260-264.
3. J.W. S. Cassels, An Introductionto DiophantineApproximation, Cambridge,1957,27-44.
4. H. Cohn, Approachto Markoff'sminimalformsthroughmodularfunctions, Ann. of Math., Princeton(2)
61(1955) 1-12.
5. J. H. Conway,Unpredictableiterations,Proc. NumberTheoryConf.,Boulder,1972,49-52.
6. R. E. Crandall, On the "3x + 1" problem,Math. Comput., 32 (1978) 1281-1292; MR 58 #494; Zbl.
395.10013.
7. L. E. Dickson, Studiesin theTheoryof Numbers,Chicago Univ. Press,1930,Chap. 7.

This content downloaded from 171.67.34.69 on Sat, 20 Jul 2013 08:22:21 AM

All use subject to JSTOR Terms and Conditions
1983] NOTES 41

8. C. J. Everett,Iterationof thenumber-theoretic functionf(2n) = n,f(2 n + 1) = 3n + 2, Adv. in Math.,25

(1977) 42-45; MR 56 #15552; Zbl. 352.10001.
9. G. Frobenius,Uber die Markoffschen Zahlen, S.-B. Preuss.Akad. Wiss. Berlin(1913) 458-487.
10. MartinGardner,MathematicalGames, A miscellanyof transcendental problems,simpleto statebut not at
all easy to solve,ScientificAmer.,226 # 6(Jun1972) 114-118, esp. p. 115.
11. MartinGardner,MathematicalGames, Patternsin primesare a clue to the stronglaw of small numbers,
ScientificAmer.,243 #6 (Dec 1980) 18-28.
12. RichardK. Guy, Unsolved Problemsin NumberTheory,Springer,New York, 1981, ProblemsD12, E16,
E17.
13. E. Heppner,Eine Bemerkung zum Hasse-Syracuse-Algorithmus, Arch.Math. (Basel), 31 (1977/79) 317-320;
MR 80d: 10007; Zbl. 377.10027.
14. David C. Kay, Pi Mu Epsilon J.,5 (1972) 338.
15. A Markoff,Sur les formesquadratiquesbinairesindefinies, Math. Ann., 15 (1879) 381-409.
16. H. M6ller,Uber Hasses Verallgemeinerung der Syracuse-Algorithmus (Kakutani'sProblem),Acta. Arith.,34
(1978) 219-226; MR 57 # 16246; Zbl. 329.10008.
17. R. Remak,Uber indefinite binire quadratischeMinimalformen, Math. Ann.,92 (1924) 155-182.
18. R. Remak,Uber die geometrische Darstellungder indefinitivenbinirenquadratischenMinimalformen, Jber.
Deutsch Math.-Verein,33 (1925) 228-245.
19. GerhardRosenberger,The uniquenessof the Markoffnumbers,Math. Comp., 30 (1976) 361-365; but see
MR 53 #280.
20. Ray P. Steiner,A theoremon the Syracuse problem,CongressusNumerantiumXX, Proc. 7th Conf.
NumericalMath. Comput.Manitoba, 1977,553-559; MR 80g:10003.
21. Riho Terras,A stoppingtimeproblemon the positiveintegers,Acta Arith.,30 (1976) 241-252; MR 58
*27879 (and see 35 (1979) 100-102; MR 80h:10066).
22. L. Ja.Vulah,The diophantineequationp2 + 2q2 + 3r2 = 6pqr and theMarkoffspectrum(Russian),Trudy
Moskov. Inst. Radiotehn.Elektron.i Avtomat.Vyp. 67 Mat. (1973) 105-112, 152; MR 58 #21957.
23. Don B. Zagier,Distributionof Markov numbers,Abstract796-A37, Notices Amer.Math. Soc., 26 (1979)
A-543.
24. David A. Klarner,An algorithmto determinewhen certainsets have 0-density,J. Algorithms,2 (1981)
31-43; Zbl. 464.10046.

NOTES
EDITED BY SHELDON AXLER, KENNETH R. REBMAN,AND J.ARTHUR SEEBACH,JR.

Materialforthisdepartmentshouldbe sentto Professor

J. ArthurSeebach,Jr.,Department
of Mathematics,
St.
NM 55057.
Olaf College,Northfield,

WELL-DISTRIBUTED MEASURABLE SETS

WALTER RUDIN
Department
ofMathematics,
University
of Wisconsin,
Madison,WI 53706

setA c I = [0,1]suchthat
THEOREM.Thereis a measurable
0 < m(A n V) < m(V)
foreverynonempty
openset V c I.
Proof. Let CTDP mean:CompactTotallyDisconnectedsubsetofI, havingPositivemeasure.
Let {I,} be an enumeration of all segmentsin I whose endpointsare rational.Construct
sequences{A,,},{(B,} of CTDP's as follows:
StartwithdisjointCTDP's A1 and B1 in I,.
Once A1, B1,. .., An -,BnIi are chosen,theirunion Cn is CTD, hence In\Cn containsa

This content downloaded from 171.67.34.69 on Sat, 20 Jul 2013 08:22:21 AM

All use subject to JSTOR Terms and Conditions