You are on page 1of 4

An Ancient Unfair Game

Author(s): Robert Feinerman


Source: The American Mathematical Monthly, Vol. 83, No. 8 (Oct., 1976), pp. 623-625
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2319887
Accessed: 11-12-2015 16:07 UTC

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 138.73.1.36 on Fri, 11 Dec 2015 16:07:10 UTC
All use subject to JSTOR Terms and Conditions
MATHEMATICAL NOTES 623

(1) shouldbe contrasted


to thefamiliar
result

(z
dZ2logr(z + 1)= ( 1) +2)2+

(1) can,ofcourse,be obtainedindependently.


Forexample,itfollows
fromNorlund's
series[1,p.
261]

0(a - z) = +(a) -() + 2 (a) 3 a)

validforRe(a - z)>O and a O0,-1, -2,. . ., inwhich+(z)= (dldz)logF(z), bydifferentiation at


z = 0, and can also be obtainedbya slightmodification
of standardtheoryforthedevelopment of
functionsin seriesof inversefactorials and Watson[2, pp. 142-1441).(We remarkin
(Whittaker
passingthatthelastmethodis hereratherclumsy, andthecoefficientsof 1I(z + 1)j are obtainedas
linearfunctions of the Bernoullinumbers.)Nevertheless,the proofof thisnote is simpleand
suggestive.

References
1. N. E. Norlund,Vorlesungeniiber Differenzenrechnung,
Springer,Berlin, 1924.
2. E. T. Whittakerand G. N. Watson, A Course of Modern Analysis,4th ed., Cambridge,1946.

DEPARTMENT OF MATHEMATICS, MCGILL UNIVERSITY, MONTREAL, QUEBEC.

AN ANCIENT UNFAIR GAME

ROBERT FEINERMAN

The gameofdreydel hasbeenplayedbyJewsformanyyearson thefestival ofChanukah.(Some


dateitbackabout2000yearsandsomedateitfromtheearlymiddleages.)Whatwe have
historians
discoveredis that,surprisingly,
suchan ancientgameis also an unfairgame.
The dreydelis a four-sided top whose sides we'll denoteby the lettersN, G, H, and S
(corresponding to the HebrewlettersNun,Gimel,Hay, and Shin).The gameis playedwithany
number ofplayers, eachofwhomcontributes one unitto thepotto startthegame.The playerstake
turnsspinning thedreydeluntileach playerhas had histurn,at whichtimetheplayersspinagainin
thesameorder.The gamecontinues untilsomemutually agreedstoppingpoint.
Thepayoffs (tothespinning toeachofthefourequallylikelyoutcomes
player)corresponding are:
N: no payoff H: halfthepot
G: entirepot S: putsone unitintothepot.

Whenone playerspinsa G, he collectstheentirepot,andall theplayersthencontribute


one unitto
formthenewpot.
We willprovethefollowing:
MAIN THEOREM. LetXNbe thepayoff
on then-thspinand letp be thenumber
ofplayers.Then,the
valueofXn,E(Xn), is
expected

P+ 0 n-1 2

This content downloaded from 138.73.1.36 on Fri, 11 Dec 2015 16:07:10 UTC
All use subject to JSTOR Terms and Conditions
624 ROBERT FEINERMAN [October

We thussee thatifp > 2 (i.e.,ifthereare morethantwoplayers), theexpectedpayoff on thenth


spin is a strictlymonotonicdecreasingfunction of n. Therefore, the firstplayer(whose spins
correspond to n = 1, p + 1,2p + 1, 3p + 1,. . .) has a term-for-term
greaterexpectedpayoff thanthe
secondplayer(whosespinscorrespond to n = 2, p + 2, 2p + 2, 3p + 2,. . .) and who in turnhas a
greater expectedpayoff thanthethirdplayer,etc.Thus,thefirst playerhas an unfair advantageover
thesecondplayer,whoin turnhas an unfairadvantageoverthethirdplayer,etc.Furthermore, this
.unfairnessis accentuated ifa stoppingruleis usedwhichdoesnotguarantee an equalnumber ofturns
to each player(e.g.,untilthe20thG).
A moment's thought convincesus thaton anyspinoneofthemostimportant factors
is thenumber
of spinssincethelast G. We shalltherefore need thefollowing:
inthepotat thestartofthek-thspin,giventhattherewereno G's
LEMMA. Let Yk betheamount on
thefirstk - 1 spins. Then, E(Yk), the expectedvalue of Yk, is
2+ (5)k-l(p - 2).
Proof:The lemmais obviouslytruefork = 1 as E(Y1) = p.
Nowassumethelemmais truefork andthattherewereno G's on thefirst k spins.Then,on the
kthspintherewere3 equallylikelyoutcomes(N,H, and S) whichwouldproducepotswhichare,
respectively:
thesame as, halfof,and one morethanthepot at thestartof the kthspin.Thus,

E(Yk+l) =3 (E(Yk)+2E(Yk)+E(Yk)+ 1)

= 6E(Yk) + 3

=5(2+ (6)k l(p - 2))+ 3


2+ (6)k (p - 2)
andthelemmais provenbyinduction (wecouldalso haveproventhelemmabysolvingtherecursion
formulaE(Yk?+) = 5E(Yk) + 3 withtheinitialcondition
E(Y,) = p).
We are nowreadyforthe
ProofofMain Theorem:The expectedpayoff, E(Xn), dependson thelastoccurrence ofa G (if
therewereanypreviousG's).
IfthelastG wason the(n - 1)stspin,thentherewereE(Y1) unitsinthepotatthestartofthenth
spin.Thuswithprobabilityl thepayoffwouldbe eachof0, E(Y1), 2E(Y1), and - 1. Sincethereis a
of 1 thattherewas a G on the(n - 1)st spin,we get a contribution
probability (to E(Xn)) of

4[1(O+ E(Y1) + 2E(Y1) - 1)] =(32E(Y1) - 1).

IfthelastG wason the(n - 2)ndspin,therewasno G on the(n - 1)stspinandtherefore E(Y2) in


thepotat thestartofthenthspin.Therefore, 4thepayoff
withprobability wouldbe eachof0, E(Y2),
'E(Y2) and - 1. Sincethereis a probability
of4 4 ofthelastG beingon the(n - 2)ndspin,we geta
contribution(to E(Xn)) of
1
4[4(O +E(Y2) + 2E(Y2) - 1)] = 34(322E(Y2)- 1).

In general,if the last G was on the (n - k)th spin (k = 1,2,..., n - 1) whichoccurswith


(to E(Xn)) of
1(3)k-1 we get a contribution
probability
4 421(:41)[4(O+ E( Yk)+ 2E (Yk)-1)] = 6(3-12 E (Yk) 1).

Finally,iftherewereno previousG's, whichoccurswithprobability


(4)n-1, we geta contribution

(to E(Xn)) of

This content downloaded from 138.73.1.36 on Fri, 11 Dec 2015 16:07:10 UTC
All use subject to JSTOR Terms and Conditions
1976] MATHEMATICAL NOTES 625

The sumof all theseis, of course,E (Xn). Therefore,


n-2
n
(X,n) =6 E (4 -(2 +3 2-() (p -2)) + 3)-1(2 +3_ (Sn-1 (p -2))
k =O

DAM M + Ce H. L C

DEPARTMENT OF MATHEMATICS, HERBERT H. LEHMAN COLLEGE (CUNY), BRONX, NY 10468.

CROWDED POSETS AND RAMSEY'S THEOREM

J. M. WEINSTEIN

groups,MukhinandStarukhina
In theirrecentpaperon topological givea basicresulton lattices
([1],Lemma1; restated inourProposition resultandto supply
3). Ouraimhereis to generalize'their
an improved proofbased on thecombinatorialtheoremof Ramsey.
Posets.Our discussionuses littlemore than the basic definitions and standardnotations
concerning posets(orpartiallyorderedsets).Twoelements x,y ofa posetA (withpartialorder' ) are
comparableif eitherx ' y or y ' x. A chain is a set of pairwisecomparableelementsand an
antichain is a setofpairwiseincomparable elements.By w and w* we meantheusualposetsformed
bythenonnegative andbythenonpositive
integers TheposetA has (respectively
integers. omits)the
p'set C ifsome(respectively no) isomorphic ofA. (Posetswhichomit
copyofC occursas a sub-poset
w* arevariously called"well-founded" well-ordered"
or "partially or "withminimum condition" or
"withdescending chaincondition".)GiventwoposetsA andB, theirdirectproductA x B is theposet
whoseelementsare the orderedpairs(a, b) witha in A and b in B, and suchthatthe relation
(a, b) ? (c,d) holdsbetweentwosuchpairsinA x B ifandonlyifbotha -' c (inA) andb ' d (inB).
Ramsey's theorem.A versionof the well-known "pigeonholeprinciple"assertsthatwhena
"large"setS is coveredbya "few"setsQ 1,Q2, , Qk,thenat leastone Qi coversa "ratherlarge"
subsetT of S. Ramsey'stheorem([2], or cf.chapter1 of [3] or chapter4 of [4]) generalizesthis
fromcoverings
principle of S itselfto coverings of S (i.e.,thesubsetsof S ofsome
ofther-subsets
fixedfinitesize r). WhenS is infinitewe have:
1. THEOREM (Ramsey).Supposegiven an infinite set S and positiveintegersk,r and sets
ofS belongsto at leastone ofthesetsQi. Thenthereis an
Q1, Q2, *, Qk suchthateveryr-subset
subsetT ofS suchthatsomeone Qi containseveryr-subset
infinite of T.
Thistheorem has an important directapplication posetA, let S be a
to posets.Givenan infinite
denumerable subsetandletf be a one-onefunction on S ontothenonnegative Fori = 1,2 or
integers.
3,defineQi tobe thesetofallpair-subsets {a, b} ofS forwhichthefollowing (i) holds:
condition
(1) a, b are incomparable in A;
(2) a < b (in A) and f(a) < f(b), orb < a and f(b) < f(a);
(3) a < b and f(a) > f(b), or b < a and f(b) > f(a).
Choose T and Qi by Ramsey'stheorem.Then (accordingas i is 1,2 or 3) thesub-posetof A
inducedby T is eitheran infiniteantichain oris isomorphic withw*. Thatis,
withw or is isomorphic
we have:
2. posethas an
COROLLARY.Everyinfinite antichainor has w or co
infinite
Crowding.In [1] a posetis called crowdedif everyantichainis finite.The authorsof [1] are
inclassifying
interested groupshavinga crowdedlatticeofclosedsubgroups,
topological buttheyfirst

This content downloaded from 138.73.1.36 on Fri, 11 Dec 2015 16:07:10 UTC
All use subject to JSTOR Terms and Conditions

You might also like