return

Celebratio Mathematica

Elwyn Berlekamp

Elwyn Berlekamp

by Joe Buhler

Re­search of­ten feels like play to math­em­aticians, and El­wyn Ber­lekamp ex­em­pli­fied this in many dif­fer­ent ways. Any­one who talked to him about in­form­a­tion the­ory, al­gorithms, prob­ab­il­ity, codes, fin­ance, and sig­nal pro­cessing cer­tainly saw this twinkle-in-the-eye spir­it in his joy, love of clev­erness and nov­elty, and oc­ca­sion­al hints of com­pet­it­ive­ness.

All of this was greatly amp­li­fied whenev­er games or puzzles were the ac­tu­al sub­ject! His work on com­bin­at­or­i­al games (with John Con­way and Richard Guy) [1], [2] covered a vast ar­ray of games and puzzles, and in­cluded deep dives in­to the un­der­ly­ing math­em­at­ics that emerged.1 El­wyn (and Dav­id Wolfe) ap­plied this body of work to en­dgames in the game of Go [3] (pre­ced­ing Al­phaGo by 20 years in do­ing bet­ter than the best pro­fes­sion­al play­ers in an as­pect of this no­tori­ously deep game). Later, par­tially mo­tiv­ated by ideas from eco­nom­ics, he in­tro­duced the gen­er­al idea of “coupon col­lect­ing,” which forces play­ers to as­sess the value of cur­rently be­ing on move.

It would be pos­sible to ex­pound at length on any of the above, or on his many oth­er activ­it­ies — phil­an­thropy, ser­vice on the boards of na­tion­al sci­ence so­ci­et­ies, fin­an­cial com­pan­ies (e.g., with Jim Si­mons), sup­port for MSRI over the years, sup­port for STEM edu­ca­tion and re­cre­ation­al math­em­at­ics, etc. However, here I will just fo­cus on my own ex­per­i­ences with El­wyn, pre­faced by some re­marks on jug­gling. We each (sep­ar­ately) spent time hon­ing the sort of phys­ic­al clev­erness re­quired to be­come good at this skill, and the in­sight it gave me in­to El­wyn’s per­son­al­ity was borne out in the sub­sequent years of our work to­geth­er in con­nec­tion with MSRI.

Be­fore I met him in Berke­ley in the early 1980s, I knew that Ber­lekamp had writ­ten a highly re­garded book on er­ror-cor­rect­ing codes, and that he was writ­ing a mul­tivolume tome on com­bin­at­or­i­al games with Con­way and Guy. I had also learned (from Ron Gra­ham) something truly re­mark­able, which speaks to El­wyn’s core traits. Namely, he mastered the re­verse five-ball cas­cade jug­gling pat­tern be­fore learn­ing the more nor­mal (“in­side”) five-ball cas­cade.

The cas­cade is a jug­gling pat­tern with an odd num­ber of balls, each tra­vers­ing a fig­ure-eight pat­tern (see fig­ure be­low); the two ver­sions are time-re­versals of each oth­er, but the nor­mal ver­sion is in­trins­ic­ally more nat­ur­al for, at the very least, 95% of the pop­u­la­tion (for, it is some­times con­jec­tured, subtle kin­emat­ic or ana­tom­ic­al reas­ons).

El­wyn is the only per­son that I’ve met who learned the re­verse five-ball cas­cade first be­fore learn­ing the usu­al one. Some people learn the re­verse three-ball cas­cade first, though nowadays this is rare be­cause the usu­al one is taught in any class-like en­vir­on­ment. Either five-ball pat­tern re­quires months (or more, or much more) of per­sist­ent prac­tice. El­wyn clearly had some com­bin­a­tion of tal­ent, ex­treme di­li­gence, and the sort of in­de­pend­ent single-minded­ness (i.e., stub­born­ness) re­quired to ig­nore ad­vice, which he surely re­ceived, to do things the usu­al way.

Figure 1. In the figure of the usual five-ball cascade (on the left) and the reverse five-ball cascade (on the right) the arrows indicate the direction of the balls. Two possible reasons that have been adduced for why the latter is “less natural” are (a) the throws in the reverse cascade on the outside require a slightly greater supination (twisting outward) of the hand, and (b) the figure-eight crossing point is a potential collision point: the balls go through that point just after the throw in the usual cascade, but they do so much later in the reverse cascade (after the throws have peaked), so greater accuracy of throws is required.

Meet­ing him, and then talk­ing to him about any num­ber of things dur­ing my vis­its to UC Berke­ley, did not dis­ab­use me of any of my pri­ors.

When I moved to Berke­ley for two years in 1999 to serve as Deputy Dir­ect­or at MSRI, El­wyn was help­ful in quickly bring­ing me up to speed on the Berke­ley cam­pus and MSRI’s place in it. We talked fre­quently, and it was nat­ur­al for Dav­id Eis­en­bud, MSRI’s Dir­ect­or, to draft the two of us in­to writ­ing a puzzle column for the Emis­sary, MSRI’s news­let­ter.

This was a blast. I could count on hav­ing twice-yearly bursts of emails and phone con­ver­sa­tions about prob­lems and solu­tions. This was some­times a de­mand­ing ex­er­cise, but we both en­joyed hear­ing what the oth­er had to say. And he was ex­traordin­ar­ily fair-minded — for ex­ample, in con­ced­ing a point on those rare oc­ca­sions when he was wrong.

These ex­changes con­tin­ued for 20 years, right up to a few days be­fore his death when he signed off on the fi­nal draft of the Spring, 2019 column from a hos­pit­al bed.

My wife, Danalee, has a num­ber of re­col­lec­tions of El­wyn (bey­ond watch­ing us chat­ter and/or squabble about the puzzle column) from our years in Berke­ley. One was see­ing El­wyn’s joy back­stage after an event or­gan­ized by MSRI, in which Steve Mar­tin and Bob Os­ser­man had an amus­ing con­ver­sa­tion on vari­ous top­ics, most vaguely re­lated to math­em­at­ics. This was (by plan) in­ter­rup­ted by Robin Wil­li­ams bar­ging on­stage out of the audi­ence. The idea of “or­gan­iz­ing” Robin Wil­li­ams is a non sequit­ur, and the en­su­ing zani­ness was greatly amus­ing to the audi­ence. El­wyn was ec­stat­ic af­ter­wards both be­cause of the qual­ity of the banter, but also be­cause it was all such a mar­velous show­case for math­em­at­ics.2

Danalee’s oth­er re­col­lec­tion was of the time when El­wyn gave me a chess puzzle, in her pres­ence, that he had de­vised. He was astoun­ded when she came back the next day with a solu­tion. Danalee felt that the re­quired moves were ali­en to chess play­ers per­haps be­cause they did not fol­low stand­ard pat­terns, so that in fact it was an ad­vant­age to barely know the rules. In any case, El­wyn was amazed.

My fond­est memor­ies of El­wyn over the last 20 years re­late to the puzzle column. He was ab­so­lutely in­defatig­able in work­ing on puzzles and their state­ments, and we spent lots of time de­bat­ing their suit­ab­il­ity for the column. He was de­lighted when a prob­lem was cited in an aca­dem­ic pa­per (or when it was men­tioned, in one case, in The New York Times). It would be fun to go through such high­lights over those 20 years, but this would take a lot of space, and it feels more ap­pro­pri­ate to fo­cus on one spe­cif­ic ques­tion that arose in our work.

El­wyn no­ticed that a vot­ing puzzle, that ap­peared in one of the earli­est “hat prob­lems” in our columns, seemed to have a re­mark­ably clean an­swer. The res­ult­ing con­jec­ture says that something nice hap­pens that does not hap­pen in the more usu­al con­text of bin­ary er­ror-cor­rect­ing or cov­er­ing codes.

If you’ve read this far, then I feel no guilt in veer­ing off in­to ex­pli­citly stat­ing this con­jec­ture, be­cause I know that El­wyn would ap­prove. Some nota­tion is needed. Let \( Q_n \) be the graph of the \( n \)-di­men­sion­al hy­per­cube, whose ver­tices are the \( 2^n \) \( n \)-bit bin­ary strings, and two such ver­tices are con­nec­ted by an edge ex­actly when they dif­fer in pre­cisely one po­s­i­tion. A Ham­ming ball (of ra­di­us one) con­sists of a ver­tex (called the cen­ter) to­geth­er with its \( n \) neigh­bors. The prob­lem of find­ing a max­im­al dis­joint col­lec­tion of Ham­ming balls is equi­val­ent to find­ing bin­ary er­ror-cor­rect­ing codes, and the dual prob­lem of find­ing a min­im­al set of Ham­ming balls that cov­er every ver­tex is an ex­ample of find­ing op­tim­al cov­er­ing codes. Un­less \( n \) is of the form \( 2^k-1 \) or \( 2^k \), the prob­lem of find­ing the num­ber of ele­ments of either kind of op­tim­al code is in­tract­able as soon as \( n \) is at all large.

Mo­tiv­ated by the afore­men­tioned vot­ing puzzle, which is equi­val­ent to find­ing a cov­er­ing by a gen­er­al­iz­a­tion of Ham­ming balls, define an oc­topus in \( Q_n \) to be a ver­tex (called its cen­ter) to­geth­er with \( n \) edge-dis­joint paths start­ing at the cen­ter. (Note that a Ham­ming ball is an oc­topus whose paths all have length 1.) An oc­topus cov­er­ing is a set of edge-dis­joint oc­topi such that every ver­tex of \( Q_n \) is either a cen­ter of some oc­topus, or the en­d­point of one of the paths that em­an­ates from an oc­topus cen­ter.

Con­jec­ture: (Elwyn Berlekamp) There is an oc­topus cov­er­ing of \( Q_n \) with \( \lceil 2^{n}/(n+1)\rceil \) oc­topi.

Note that each oc­topus cov­ers at most \( n+1 \) points, so this is say­ing that the smal­lest pos­sible num­ber of oc­topi is al­ways suf­fi­cient. In oth­er words, El­wyn was con­jec­tur­ing that, in the case of oc­topus cov­ers, the hap­pi­est pos­sible situ­ation al­ways holds. I am sure that El­wyn would love it if someone would prove this\( \ldots \)

Joe Buhler re­ceived a PhD in 1977 from Har­vard Uni­versity in al­geb­ra­ic num­ber the­ory, taught at Penn State and Reed Col­lege, be­came Deputy Dir­ect­or at MSRI in 1999, and was Dir­ect­or of the Cen­ter for Com­mu­nic­a­tions Re­search in La Jolla, CA, from 2004–2017.

Works

[1] E. R. Ber­lekamp, J. H. Con­way, and R. K. Guy: Win­ning ways for your math­em­at­ic­al plays, vol. 1: Games in gen­er­al. Aca­dem­ic Press (Lon­don and New York), 1982. A Ger­man trans­la­tion of this volume was pub­lished in two parts as Gewinnen: Strategi­en für math­em­at­ische Spiele Band 1: Von der Pike auf (1985) and Band 2: Bäumchen-wechsle-dich (1985). Like­wise, a second Eng­lish edi­tion was pub­lished in two parts in 2001 (“Volume 1”) and 2003 (“Volume 2”). MR 654501 Zbl 0485.​00025 book

[2] E. R. Ber­lekamp, J. H. Con­way, and R. K. Guy: Win­ning ways for your math­em­at­ic­al plays, vol. 2: Games in par­tic­u­lar. Aca­dem­ic Press (Lon­don and New York), 1982. A Ger­man trans­la­tion of this volume was pub­lished in two parts as Gewinnen: Strategi­en für math­em­at­ische Spiele Band 3: Fall­stud­i­en (1986) and Band 4: Solit­air­spiele (1985). Like­wise, a second Eng­lish edi­tion was pub­lished in 2003 (“Volume 3”) and 2003 (“Volume 4”). MR 654502 book

[3] E. Ber­lekamp and D. Wolfe: Math­em­at­ic­al Go: Chilling gets the last point. A K Peters (Welles­ley, MA), 1994. With a fore­word by James Dav­ies. MR 1274921 Zbl 0852.​90149 book

[4] E. R. Ber­lekamp, J. H. Con­way, and R. K. Guy: Win­ning ways for your math­em­at­ic­al plays, 2nd edition, vol. 1. A K Peters (Nat­ick, MA), 2001. Ex­pan­ded re­pub­lic­a­tion of (ap­prox­im­ately) the first half of the ori­gin­al volume 1 (1982). MR 1808891 Zbl 1005.​00004 book

[5] E. R. Ber­lekamp, J. H. Con­way, and R. K. Guy: Win­ning ways for your math­em­at­ic­al plays, 2nd edition, vol. 2. A K Peters (Nat­ick, MA), 2003. Ex­pan­ded re­pub­lic­a­tion of (ap­prox­im­ately) the first half of the ori­gin­al volume 1 (1982). MR 1959113 Zbl 1011.​00009 book

[6] E. R. Ber­lekamp, J. H. Con­way, and R. K. Guy: Win­ning ways for your math­em­at­ic­al plays, 2nd edition, vol. 3. A K Peters (Nat­ick, MA), 2003. Ex­pan­ded re­pub­lic­a­tion of (ap­prox­im­ately) the first half of the ori­gin­al volume 2 (1982). MR 2006327 Zbl 1083.​00003 book

[7] E. R. Ber­lekamp, J. H. Con­way, and R. K. Guy: Win­ning ways for your math­em­at­ic­al plays, 2nd edition, vol. 4. A K Peters (Welles­ley, MA), 2004. Ex­pan­ded re­pub­lic­a­tion of (ap­prox­im­ately) the second half of the ori­gin­al volume 2 (1982). MR 2051076 Zbl 1084.​00002 book