Article
On Folk Theorems
David itarel
IBM Thomas J Watson Research Center
Yorktowr~ Heights, New York
- - ~ i f e d a are sug\gesied fbr gg a
statement is a folk theorem~ The ideas are t:he~
illustrated wi~h a detailed example from the theory of
program.mi~g.
Key Words and Phrases: flowchart, folk theorem,
stn~¢tured pr~gammlng, white-programs
CR Categories: 2,1, 4,2, 5~:N
1. InmMucdnn
l;>lkl~)ee: The ~radi~or~ai ~Iiefs, legen& and
c~ioms c~rercn~amor~g c o m m o n peopie.
The Oxfogd gng~isk Dictionary
......
7;~eorem: A general conclusion which has been proved,
.......Mad, ema~i~::s Dtctiona<y, Van Nes~rand Reinhoid
In view of these quotatior~s one might be tempted m
conclude simply than a {blk theorem is a ,general conclu-
sion which has been proved and which is a traditional
belief, ~egend, or custom current among common ~ople.
The purpose of t,his paper is to refine this defimtion
somewhaL adapting it to the purposes of the research
community in computer science, Accordingly, we shall
attempt to provide a reasonable definition o f or, rather,
criteria fbr ~btk theorems, followed by a detailed example
illustrating the ideas, The latter endeavor might take one
of two ~ : ~ i b t e forms, We could take a piece of ~blklore
and show thal it is a theorem, or take a theorem and
show that it is ff~lklore. As an example of the first tbrm
we could have shown that the statement P ~ NP, which
is iblkiore, is also a theorem~ However, since we have
resolved to inUoduce no new technical material in this
pxaper~ and moreover, since researchers in our community
seem to be less tamiliar with folklore than with theorems,
Permi:.~km ~o vopy without (¥e nit o;r part of ~,hi~ ma{eria} is
gr~m~ei~p~ovide~l ~ha~~he ~oNe,:~are nor made of disaffected lk~rdirect
¢~m:~er~,l adva,mag~, d~ ACM copyright ~oti~,¢ ~md ~t~e~i~leof the
puNica~ioe and ink ease @peal arid ,'a~ice/:, given ~Jha~~p}/mg i~ by
perimi~ioa of ~he A~,~iali~m iSot Ccu~pud~g M~zgi,~ef~y, To copy
o~hef'wiae~ar ~ 0 re'pub/inn, req,,sires a Ne and/or ~g~cifie N:fmk~aiee,
A~aNoFs ad/dfe~ a~ of Aagas~ 15: LX HareL ~>ept of Applied
Ma~hema~ics~ The Weizman~ lt~ti{~ae of Scie/~c:e, Rehovo~ israel,
379
we prefer and wit~ adopt the second approach. Accord-
ingly, in Secdon 3 we pro;sent strong evidence ~o the
effect that a particular theorem about flowcharts is a t~lk
theorem.
2. What is a Folk Theorem?
We woutd like the reader m try' and recall the tast
time he received a referee's report in which the referee
dismissed his latest achievement as being % piece of
foNtore which haas been around for ten years," or, Nice-
natively, the last time the reader himself drastically
reduced a studenfs overflowing enlhusiasm about, a
resutt, labetfng it % good try but umS~rtunatety a folk
theorem2' The high frequency of such rather unhappy
occasions cedis for extra investigation. What makes a
statement a fokk theorem? What facts prompted the
careful reviewer m make his devastating comment, or by
virtue of what did t;he reader himself dare turn a seI.f-
confident student or colleague into a temporarily crushed
individual?
Let us quote three more definitions of f d k objects:
f?dk song: A so:ng made and handed down m~eng ~,hecommon people;
fnlk song.s are usually of anowmous anNam:hip and often have many
veirsiot'13.
Folk tale {orfMk ~n<v): A aory ~ua|ty of ~ o n y m o u s authorship and
legeedau at mythical elements, made and handed down orally among
d~e common pc/opec.
--.. YV~/b~wr~ New g%~rtd D&'gi~e~z~y ~3[fke A n~eica~ Long.age
FeN tune: A N~pula~ story handed dm~,'n~-dly from past generations.
..-,,o-7?~e O.~ford A~,~c~Jd Learners DicgionaO~o f c%gren¢ En~ish
We wilt not ~ry our hand at providing a similar clear-cut
definition o f a ~blk theorem suited for a dictionary.
though ~t is s~mewhat amusing m see whN happens
when the word "theorem" is substituted #or "song" or
"story-" in the aNwe. In parficulan we are certain that
the (by' now annoyed~ reader wouN require an expN-
nation of the phrase -'common people" i~t our comext.
On closet- inspecuon, however, it seems that there are
three centrM properties which we can abstract from these
definitions: an(tortuous and age~
Note that these phenomena are a|so implicit in the
c~f Volume '23
~he ACM N~mbef 7
4 e 6 n m o n of/}d£~o~: quo:ed s: ~h¢ s:a:":: k~af:e:.uamong
c o m m o n peo[:@" imphe> p,%Aarny. "b<:heA+ }e~:ends
a:~d :;us:os:~s ~ ~nd:ca:es a n o n y tne:.~,s ~::::h w'q::p+ a n d ~'m:.+ : ~::~,:s, b~{CC{:q. +++ :~+>~: :%:' H14:1~1 : hc :>4::: ot+~ )y. :{
d@onaF+ adds. {he d:memo~on o} :H::¢ V:¢ CAM~P{ IH:d +::: $::':i:U+: ::lC{:/h H: {}Wig ~:¢ ~k, ,+, {~
v~ he:: :q, :~g :o da+J,, @e::~e:hr::e:~:~pc<{>,<d {:a>&:: } n
@era f:om @ e :eMma: o{ :ell:rig:ale:.~4rid rag:rig ~,"ng>
:o :ha: of p:o~:::g @eo:en::,: :{ :shou:d h< nosed :h~:~:
neHhe: one <q' {h¢:~: :> :o be t & c n m at', s¢n<: :::(::4:
:,once The p<>p:Aari:) of a thecwen: ::ex:d so: mx<.,~<:rd~,
b~: ¢s~abhshed b) ;: he,ad ~,oun: ,of ~:e~'a:cher~ :h,> hay{:
used {',rquo:ed H. :ho~:h {~::~::h:s£:a:ed m :he :~e::
se<:th2:) s:a:~sdcs can be h d p G i R4:he< :{ usua: b sho::~
::p when o:e ::eat ask:n: ~ hm~dfi.A ofco:}e;:guex whe:h~::
{he s:a:eme~:: + ~nm. WiJ~ ~he "+:w+ he+wd :<>{ne~hmg "+:'c.'~] {(:k ::::'<::<qy we:: b: :cpRv, cd ,..'~¢'~:he :,e:~o a~L~H:
hke :ha+ heR,re" :<x>k on h~ f:H~< :he :x:r:~on que~noned
wi{~ ~s~m::v be ab}e ~o gave :.he <or:w{ :nss~e~ a n d mo:q <on:~p}c:¢~ +u>{~fied .:,~d ~hou{d tw +v~a:-dcd, ,,> par: o+
ofle:a he w{l be a b k ~o ~<x, mDw:y ~{ ,a'i~ha p~.w+f Rx'x :he C::~:::f~ :}Hg a m a : c : : r hl~{o~:a¢1 :n:\ {h~:% ::oco;:::e
Ofkon ~he :heo+'em is qe#~ed by v+~ue of h b<'~s~ ~er y
s:mila~ +x>~nother £a{eme{~: which > he:rag p:o~ed+ This hob{> ~:~ch ~+o, ~:an:p coHeca+n}{
bring:+ +a.s~o :he +R4{,en have many vem:oa::s++phoneme+ }{esc~ ~h:ie :he p~>:>,,au(h-+:£¢ pro{~cr::es +ecru :v ~,
non.. which :s ck+,-se~yre:aRx+ ~o pc~p@ar+:y Sure:y+ w::h ~e<ess.~ws :rid a4~[l~cie::I . t ~ + n d i h o ~ , - ~%.>~' o t~he~w:;s:.:n {0 b<
~he size of our :eseareh comma.bait: k~@'~gwha: ~: as.+sheL
co:::pac{ s:a:ement's of {heovems :end :o be ~e.memb~+red
even by pmop:e no: +aorkmg in {he a d u a : area ++~ +e~ea+~.,h

g++'arn:s<omp+~:e e~eo1h/in+ ~" :s {n::ch more ca:ch}+ amd

ap,~eah:g :~a::::hs EbvK>~5 :go:+:>:~s ce:~n:erpart S:a+Jng A A Folk f'beorem
s:~ch a +heore: i : :his foam ~+e.a::) { a e r i e s ~:,:~pop:+++
h'~ the inkwe£ e:: ma£in g {his ~,x:cnon rcadab:¢ ~br as
~a.:hy aad w+:~+:hh cha:ces of ff,s:b~ecommg a +1ege::d ~
Iaqge a~ a s d eece as po~Abh:+ we h a ~ c o l c a ~ d i:s more
or %,.replya {:elk thee;era The vague~m~+ m a :~::+~em%
+~a~emer+: can th~s g+ve rise ~o more :hart one {legal)
oriented ~ H e + is suc+ngly ~:rg+xJ :o :ega+d as p a n c.q":he
: of +¢+ amd ~ a i : h ~ a m e s t o o + f::m amd mo:e :ext +~se~L However, since :he reade+' wou:d p:esumab~y
wbe:: e ~ h ++:'e:~i<>::equi:es a ~'. ~:ngIy di£b:es:
n<x ~>e ~eadmg ~h~s ]ou:~ai o~/he~waH+ we car~ safe:y
p:re&>£ ('The >cadet > advi:~sd no: {0 coRf~/:se s~c~ {b{k
se,s:u~me :ha: he is {i~miha+' wi~h >sch gee:era: ~e:ms as
ff:eo~erns whh +::a~eme~+ which are ge~e~ic Le.+ sta:e+
e:+ pvo£ram, and ,fl:~k~a:< s a d a b e w~h moee
me~>s thag are e+x de+..ibe~a:e~y m a ~%>>mf:~orn
gpeem~:;~:ed oae:. such air ~'e~g~%qe pee£~aen-:o~,ge< and
w:hkh :maav.+, ~ea.: :he-orem++ {:re given a~ m<+:{.ances+ }ike
~£~te~te+ W+~ho~£ fl.anh¢~ ado, le~ :as row+ re:reduce ~he
"+paralk~ :ime eq~ab spa+so.+~~.e+ e.g, chap~e+ 2 of' {36}.
s~b.~e,c:of th> ~0ik sod.ion+
Thi:+ we do ao~ waa: ~o a.s ~ m g ~>l~+)
A~on}~o~c~ a~:hoe~.h-@ +ho{~/Id also rio{ ~ ~a.kea
comgk':e{y h{e:aHy :~deeA since :he fmrh~ds of ~ime we
are ~a:kinig abs+a: are :o{ e~>~ e a : (see ~'mk:>w),i: {>.:+:>4r~+-¢(?
::he caste tha{ no o++e ka:x>w+ wi:h whom :he thee+era.+ or
+:s rigoro+a::y ~.va:;ed var~+.~,:+rots+on.>, o:ig+r+a:ed R.aghe~.
very start ofan mves~gan<m oith> :heorem Spe¢~fic~A~y
wha: ~enck~~o h a ~ r n {., :ha: :he effbr+ 'in+a>bed m ~+acmg
m < ~ peopk% g~s{ ~eae~.m~ i:~ ~¢:~aHdhu:e :he :heorem ~o
back s~,eh a ~heo~em is f}a: uea+e~ th<a~ %at mw~ved i~
~he ::9"~ paper o f g o h m and Sacopmi {+5} which a p
~\pK:vmg }< ~ > ? g {oT ¢~e ~/ea~: b e h ~ i :he pn~g@q of
g~ta~ed iS %£s )o.~r,ai :~ £~cL ~>ome Of dee p~bhshed
~Jk @eo~em/+ :end to tm +~ha:nded down" :ogethe{ w+:h
~x:vs of +he theorem+ which we men~ion ::~ef+ ate
a::::ib;~ed by :heap :~{ho~s :o {:5:~ ahhot+gh, as we +hal
deep :heo~vm with a iea:ed ~ o f w~:H ::+:+a{b~ non
q~+a:if} a,;+a fo:k :he~>:em)+ Th+ :o{x+~of a 6;Ak ~hec+:em
might axe bmr:ie+] im + m e o ~ + m m kc~+:e r+o+e£ m a +e::e: (:~e~ ~+,s ~ema+~ here d:a+ :h~;~ mad o~het phenomena
{o a a +~dm>:+o+ +or+e+ im a +priva:e c+mm+niva:FmF + h d~ed m CD: se%~el are {0 ~ {a~e~ a5 ~a{he, amm+i~g
is ~ a a l y ~ s i B k , if' <ms i.:+ '+if:rag m p+a: <am h+~+A+ex mgmi >~)ms of ~h+ b e e and k g e n d Of our yO~mg gekL
+{a:e)+++++?++ +:i++ :ho+< ++:+:+{m ks:+: ~ + , e a~:{ by r~o ::veaa+ a:e ~hey m~c.i:~m of ~h+ h
m prm:)+ :ho+gh ++mid++++ +he ~ : > f :+~(h/: ++<~
h,~t+e ++era +1~)m~~}e:e+y ~ +~:+ +h~ +mf~+@dt ' {)++' ~{he
ca:~+~ +,.+~¢h~'x:. ~ x L a s d h-ere is+ where #~e age e l
 inwAved~ We witt do o,~r best to convey this spirit to the
reader as we go along.)
P,et~:rmng to o~.~r topic, we note that the theorem
proved m the first part of'{~ 5] asserts that, with additior~al
Boolea~~ va~iabtes :~, ever), flowchart is equivatem to a
whib~program {with, in get, oraL more than one occur-
re~ce of wNle~de)~ Having proved the first "structt~ring
res~/f' of its kind~ Bohm a~d Jacopini's work has become
immortalized as comaining "the mathematical justifica-
ticm ti)r strtictured programming," to paraphrase many
at~thor~ ors the subject. The resulting tmiversal popularity
of {15], the ~i~ct that due to its rather technical style it is
apparea.tly more ot~eta cited than read in detail and, of
course the similarity of the theorem proved in I15] to
the one we are interested in seem to be some of the
reasons ~k?rthis common reaction.
Bohm and Jacopini prove their result by .providing
~local" trans%rmations on the various kinds of flow-
charts possible, leading to the final structured while-%mx,
which involves only sequencing (;), conditionals (if*then-
etseL and iteration (while-do). Put another way, their
proof is by induction on the structure of a flowchart. The
Boolean. variables are used to "mark" paths taken by the
computation, m order to remember them later.
We can point to three subsequent papers in which
similar proofs of the B6hm and Jacopini result appear;
namely, Mills {67], and Cutik [24, pp. t%t9; 25, pp. 11-
t4]. Indeed, it seems that H.D. Mills who, in his unpub-
lished lecture notes, termed the result of[151 "The Struc-
ture Theorem" was one of the driving tbrces beNnd the
glorification of B6hm and Jacopini's result, which he
reproved and discusvsed in highly attended lectures and
seminars in the foBowing years.
tt is interesting that prmsely what is needed in order
to extend this velT proof to one proving our Folk Theo-
rem can be %and in Mirkowska's 1972 thesis in Polish
about algorithmic logic [70].~ The "normal form theo-
rem" of [70] states that with additionN Boolean variables
(of the kind allowed in [15]) every while-progam is
equivalent to one in which whil~do occurs only once.
The proof, which is also by induction, provides local
transformations which serve to eliminate nested and
neighboring while's and m distribute while-do over if-
then-else. A sample tr~stbralation appears in the Ap-
pendix. To Mirkowska~s proof of what we might call the
second half of our Theorem, a complete paper by Per-
kowska [77, p. 441t was devoted two years later. A third
exposition of this proof appears in a paper by Krecz~nar
[55, pp. 23-24} which deals mainly with interesting but
completely different issues~
Thus, we have found that Bohm and Jacopini |15|
are not to be det~vched ~¥om our investigation, since
besides being the first to prove a "structuring result" and
the first, it seems, to illustrate some of the power of
These can take on one o f two distract val~es~ distingms:hable by
a test
:~N~t having 170~ ir~ {?o~t o f ~s, we rely here ot~ p~zrse~aI eom-
muaicatio~ wi~h Mirkowska a~d on the refe~mce in [55/.
3St
Boolean variables, their work togeIher with that of Mir-
kowska [70] can be interpreted as a complete proof of
our own Theorem. If one prefers symbols over words
and is willing, for the moment, to overlook priorities in
~avor of documenting proofs appearing in print, one
might assc~ziate this proof of the Folk Theorem with
([|51 V {67] ,,./{24] .,/I251)/'-. ([7O] V [771 '¢ [551),
where "V" and "A" stand, respectively, %r ~br" and
*'and."
Now, all this soun& like complicated mathematical
research concerning a rather deep result %r whose proof
the efforts and partial rose,Its of at least two researchers,
working separately and, indeed, in different countries,
were needed. The real finn in investigating this Theorem
starts when one realizes that there is a completely differ-
ent, Mmost trivial way of' proving the Theorem and that,
furthermore, it is with this proof that the names of B6hm
and Jacopini are otlen erroneously associated. This
"global" proof, given more rrigorousty in the Appendix,
involves constructing a simple one-loop program which,
starting with the first "box" in the oriNnal flowchart.
executes one such box each time around the loop. Upon
completing the execution of a box, a variable is set to the
index of the following box in the original flowchart. (For
a test box, an if-then-else accomplishes this for the two
possiNe outcomes.) Control then returns to the beginmng
of the loop. The body of the loop, which is executed until
the index of the original STOP box is detected, starts
with a nested set of conditionals which test the value of
this one new variable and branch to the appropriate box
accordingly. Since any flowchart contains only a finite
number of boxes, the new (numerical) variable can be
simulated by a finite number of Boolean ones, giving the
Theorem.
It is interesting to note that both the local transfor-
mation of B6hm and Jacopini [l 5] and Mirkowska [701,
and the aforementioned globM one fail, by their very
nature, to preserve the structure of the original flowchart.
Ironically, though, they are proofs of a so-called "struc-
turing result."
But let us continue our investigation, being curious
as to whether we ourselves, by virtue of having just
presented the global proof, are to play a central role in
this little tale. We shall clearly be deprived of this if the
proof has appeared in print earlier~ "'Appeared," did we
say? Yes. it has appeared. But that would seem to be a
rather mild word to associate with the situation we now
set out to describe.
Textbooks are a good place to start. And. indeed, if
one browses through the boo.ks published in the general
area of structured programming, one finds two which
contain the above proof: McGowan and KeUy in 1975
[63, pp. 62-64] and more recently, in 1979. Linger, Mills.
and Witt [60, pp. 118-120l~ A similar scan of books of a
more theoretical nature reveals two more occurrences of
the prcu3~? Greibach's 1975 monograph on program
schemes [38, pp. 4.52-4.53] and C|ark and Cowel/'s f976
Communkali~r~s JM~ tqS0
of Vole-me 23
lhe ACM Number 7
inm::~daclor?¢ ~x~ {20, pp. 61.~.,63}. t~ is ~:~o~ewonhythan m
all fb~r ca:des~he theorem s~a{ed ~s ~he B o h m and Jaco*
pmi one {and hence i~ am°ibt~tior~ in [(~31 and {38] 4o
{151L bm ~he global pro~:~fmakes ~:~e of only one ~hik,~
~ and so proves met stronger Folk Theorem,
Since an poimers ~o ~he origin of ~he pr<~~fare given
in ~hese references, oar b~~k search has b,ce~ only partly
suc~e:ssft~i The idea of she global proof seems ~o be
simpk arm pop~flar enough a> have ~ e n es~ar~tiaf~y
reinvemed by :the autho:cs of ~hese books. Reinvemed,
that is, if we can find i~ ooc:~arringin prim even earlier.
Leaping from hooks ~o e~aoste~ers, a reader leafing
through ~he Februa U 1975 i~s~e of S I G P L A N Ne~ces
noticvs a two-page paper by Plum {79 pp~ 32-331, m
whwh the a ~ h o r , terming the use of ~he Bohm aq~d
1acnpini {{5] mmh~l *~ma~hema~ica{overkilL" presem:s
a %;imp{e proof of this simpk ~:sutt'---4he global preen{
abcwe One finds i~ hard {o ~{ieve, though, that 1:4 ~ook
Nmos~ rune years fi?r (his observaaon ~o find its way ~o
the printer.
We rnuN turn, i~ seems, ~o one of compmer science's
s.taadard oracles: D, Km~.h~s ~d~oNrty sur~ey p a ~ r
pubt:iahed m 1974 [52}, k~side:s containi~g much smnu-
latmg ma4e~ial concerning ~he (non-) vNue of re~satts
sach as o~ar T ~ o r e m , coafirrcc~ o:ar s~spicinn by reducing
she ainu years ~o saghdy over one; in {52 po 274t we find
our sixth ~>ccurrer~ceof{h> prone b ~ here is is a~tdbmed
primarily ~o *'~he commems made by Cooper m {%7" m
his terser {221 m ~he editor of ff~is journal Indeed, D~Co
Cooper {22! (eight years, we m ~ h t add, ~ f o r e ~he N 1 n ~
identica{ {79~) notes ~hat ~*a red{radon [thaa~ thai
of B6hm and J~oDed} is ~.siNe0" and s ~o
prove our Folk Theorem ~he globN way.
And so, we are exp~vsed ~o what seem ~o be the fins,;
stages i.n ~he evolution of' arm authem~ J!;aik theo~m: As
a reaction ~o ff~e journal pubtica~mn o f a proof of half
~he theorem, an acqmost ~fiv.iat proof of ~he whole is
ted m a loner m Ihe edkon and many subse-qeen~
ambers, not ihavmg seen {~heaforemen~med te~er (in-
dee& one does not expect them m have stun i 0 then
pro>cued ~o remvem ~he proof themselves
Thin situation is enihanced by ~he hallowing addi~maal
five papers, all o f which p~e~m C'.ca~pefs proof: A.shcrof{
and M~naa's t97/ iFtPS p a ~ f on r%rqacing gate's by
({2~ p. {48/; see also {3, po 140t),, @~e~imitar b~a~
mdeper, dem I972 paper of Brunn and Steigtitz {17,
p. 52l}, Wulf°s 1972 <~c~ again:st {he gore" [87~ pp. 64--
67]~ Mill's {975 journal version of OReS~r'uc,~re Theorem
I6& p. 45]. and the appendix ~o (~e ~977 paper on retiaNe
~;~og~am.sof L i n e r arid Mils ~59, pp~ ~3,6-139]. Of these,
.[2, 68, 59t credi~ C o o e r with ~he 1:man{, {{7~ Nves the
p r ~ f without cred.i~ bm the ~heo~em ~ga{ed ia Ihe v~aker
g6hm arid J versioe, ruth credit m ~hem, and [87]
in a way kypicai of/he Mkishm:~s o f {he e.i~a.a~ioecr~i;gs
[ { 5] with ~he ver3io~a~
Similarly tyDcaL by ~:e way, g Defining's, S~ptem~a.r
{975 ¢nmp4aim {27] ag~ou~ Pb~m'~ Febma U t975 N ~ r
t791: " ly dh4s cor~macdor~ was suggested by Br,ano
3gg~
and S4{eg{az ia I972.- Denmng':, arg~mem cars ~c~.~al~y
made 6re years stretcher by re{Erring ~o ('oo~:~:r!
It is also wovd~ remarking tha4 m I2, p t49; 3, pp~
14t. 142] Ashcrofl a~d Mam~a. being imerested m s4rt~c,~
4urir~g rest~lts which preserve some of the sm~c~re of ~}~e
original flowchart, provide m add:~ion a slighdy more
cfficiem vartam of ~hc globat proof:, which they atmbme
atso ~o Cooper, in which whal one might ca//a ~maxima{-
loop~free'" component of die flowchart (a~ opposed so a
single box} is executed each ~{me amur~d ~he kmp.
Now. ever, ff we were to end om° s~ory here, there is
~o dnub~ that we could quite sa{~ty ~erm dds reatdt a
{bik theorem. As i~ ~urns ouL 4hough, ot~r tmle game ~ is
f~r from being over.
A~ this, poim we vemure o~r opinion that the global
prooL and he~ce ore Theorem iuseiL is actually roo4ed
m {he early work of Jobm vo~ Neumann on ~he structme
of digi~M <x>mpu~ers~ In Sec~io~ 6 of his 1946 (!) joim
p a ~ r wi~.h B~rks and Odds{inn {18/, ~:he idea of execut-
ing a p r o g a m on a compmer by means of a Nrge k>op
which cames ou~ an instruction a~ a ~ime is described. A
program contour is u ~ a ~ e d 1o comaia she addres.~ of die
~vext ms{ruction befbre returning ~o the sta:r~ of the tonp,
Admittedly, this is no~ a proof of a theorem, but ~he idea
is certainly there, prompting one ~o wonder whether
C<:~opaefs observation should m:~t be taken simply' as
providing an operatimaal semantics %r flowcharts, ralher
than being the pro>of of a theorem.
As if M1 this were ant enemas.h,this same pnx~f appears
i~ a h u m o r of additional p a n t s , in which its detection
is com~ideraNy harder. In the first, Meyer and Ri~chie
I65, Sec~.mn 61, erie can find the idea in a pnyof of the
fact {has prim.{dye recursive functions of certain time
comNexA~y can be written with :rome fixed depth of
~x>~nded l~x~ps,. The pro<l{ involves one potentially un-
bnuaded loop {corresponding m our single ~htteMoL ~he
body of which executes one s:ep of O:e computation a:
a time wida the aid of numerical program counters.
NexL ~he idea can be fb~nd in Norner {t 1] as the
~'Canonical Program" on page 438~ wNch si:mula~es any
given ~ machine" rasing a ~×t-stale fh~ction.
A third such o4:currence is m Bramerd and l.andwe-
bet {16: pp. 122,-123} where the idea is incorporated iron
the prc~f ~ha~ N ~ l e d Marker a{g~mth.ms comptge par-
da/ reoamve Nnc~ions~ TheN, a?~3. numerical program
contours are used.
name{her occ~rreace can ~ extracted fVom O~e proof
of m i6 m PraWs 1976 paper on dynamic iogic
tgL pp. ~19--t20t in which a ~mversN c~,umer machine
is simulated by a one-loop regular program whmh em-
ploys a numerical program counter.
A most irgeresting and ~uNte ~:xcc~rre:nceof this proof
can ~ fo~and in the fifth gaper, Efgnt {32I, also from
/976, which deals wit.h ORe N n ~ g e of m u N e n t u mul~
d V~:~h~m~23
~heACN Number 7
 tiexit flowcharts. Theorem 3.1 of [32, p. 441 states that
every such flowchart is equivalent to one in the sublan-
gt~age analogous to conventional whi|e-programs, but
with no additional variables at all! In view of the negative
res~tlts of Knuth and IF/oyd [53], Ashcrot~ and Manna
[2, 31, Peterson, Kasami, and Tokura [78], Indermark
[47], Kosarajt~ I54t, and Kasai {491, from which it follows
that Ibr conventional flowcharts this statement is false,
a care%l reading of glgot's proof becomes necessary.
The evidence of the fo/kishness of our Theorem is con-
siderably reinik~rced when one discovers that this proof
is also essentially the global one in disguise! Boolean
variables are cleverly replaced by what are called "trivial
schemes" m [321, which we might call "cables," in which
onty one "wire" of each is connected. Such a cable forces
control to proceed to one designated box i)r execution
next time around the loop. Some details and an illustra-
tion can be tbund in the Appendix.
Digging slightly deeper into the earlier papers in
algebraic semantics ti'om which Elgot's paper [32] on
structured programming grew, one discovers some inter-
esting ti~cts. A normal tbrm result for certain kinds of
operations on certain kinds of uninterpreted algebras
occurs in various papers differing quite drastically from
one another in notation and terminology. Thus, in a
t973 paper of Elgot [31, pp. 213--..222] (published in 1975)
we find a theorem about the "normal description of a
morphism over an iterative theory/' In Wand's 1972
paper [85, p. 335] (published in 1973) we find a normal
tbrm theorem for "/x-clones of operations over a lattice
algebra." I n a 1978 paper by Tiuryn [83, pp. 21-22] we
find a normal tbrm theorem for "regular polynomials
over regular algebras." Finally, it was as early as 1969
when Bekid [9, pp. 12-151 proved a normal form theorem
tbr "definable operations in general algebras."
All these four results can be shown to give use to
versions of our theorem for the more general case of
"multiwire" recursive program schemes (as opposed to
flowcharts). Without attempting to describe the technical
details of these papers, which are all far beyond the
scope and intention of this paper, we remark that in our
terminology the proofs in Bekid [9] and Elgot [31] are
local and in Tiuryn [831]and Wand [851 global. Further-
more, the proofs in [9, 31] start with schemes which are
analogous to while-programs, and all four proofs use.
essentially, the "trivial scheme" mechanism for mimtck-
ing Boolean variables. These observations make the task
of fitting these proofs into our framework somewhat
easier. Accordingly, our decisions on these were to clas-
sify Bekid [9] and Elgot [31] as additional proofs of the
Mirkowska part [70] of the local prooll and to add Wand
[85] and Tiuwn [831 to the list of Cooper-like global
proot~. Nevertheless, as the algebraic approach is so
different fi'om that of most of the other proofs we have
mentioned, we choose to assign priority credit [br the
second part of the local proof to Bekid and Mirkowska
independently, although the tbrmer precedes the latter
by three years.
Returning to the global proof, the learned reader can
probably see now where these findings are leading ins.
The search for occurrences of this proof in the (obviously
relevant) literature on structured programming and flow-
charts led to Cooper's operational version [22], which, in
turn, served to attract our attention to Burks, Goldstine,
and von Neumann [t8]. But now we have gradually been.
lured into considering theorems in algebraic semantics
(e.g, 183, 851]) and in recursive function theory (e.g, [16,
65]), which, at first sight, seem quite unrelated to our
naive flowcharts. And so, there .seems to be no escape
from considering the grand common ancestor of all such
results--Kleene's t936 normal Ibrm theorem for partial
recursive functions [5/]!
Indeed, Kteene [51, p. 736] (see also Kleene [50, p.
288] and Rogers [82, pp. 2%30]) showed that every
partial recursive function f can be described as the
application of a primitive recursive function g to pk,
where h is primitive recursive and #, the "minimization"
operator, acts, in essence, like a while loop. The "body"
of that loop, h, can be loosely described as simulating
one step m the computation o f f using coordinates for
the "current value" of the function and the "label" of
the next step. The function g simply isolates the final
value of the computation by projecting on the appropri-
ate coordinate.
N o w . Kleene's theorem appears repeatedly m count-
less papers on recursive function theory. However, since
we did, after all, start out with flowcharts, and since the
line must be drawn somewhere, we have decided to draw
it right here: no attempt shall be made to search for all
occurrences of Kleene's theorem, and the ones we have
mentioned [50, 82] will not qualify as proofs of our
Theorem.
To summarize this part of our tale. the global proof
has been traced down two orthogonal paths, each of
which has led to a pioneer~J, yon Neumann in his 1946
work on designing computers [18] and S.C. Kleene in
his 1936 work in recursive function theory [51]. Although
Cooper [22] provided the first explicit proof of our
Theorem as stated, we feel it is reasonable to make the
modest assumption that he. as well as M1 subsequent
provers, was either directly or indirectly influenced by
[181 and [5l]. Consequently, we assign credits for the
global proof of our Theorem to Kleene [51] and Burks.
Goldstine, and yon Neumann [18] independently. So, in
fact. the Theorem w a s "known" to Kleene and yon
Neumann.
Again. blurring credits m the interest o f enumeration
and begging the reader's pardon for inserting an annoy-
ing self-reference to the present paper [401 in which, no
doubt, the proof appears again. ~ we associate the global
proof of tlhe Theorem with
~"This is the " r e f e m n g m as m a n y of o~te's own papers as possible"
syndrc~me, taken to a new extrerae: a reference to the very paper being
read! In the a b ~ t c e of any other no~ab{e c~ntr~butions o f the present
paper, and in view of the amount o f work ~hat went into writing it, we
cannot resist ~he tempmtkm m claim priority ~or this first~
C~mmunicatmRs July t980
of V o | u m e 23
the ACM Number 7
 ({51] V {18]} A ({65I V [22] V [li] V {2] V {17{ v
[871 v {as] v {52t v {16] 'v [68] v {, >1 ,,~
{38] V {63] V [32] V {20] V [8l] V [59] V
is31 v I@t v {@]).
As memioaed earlier, statistics are perhaps not cru-
cial tbr ~he classification of a theorem as fbtklore., bu~
they are impassive never~hebess. The version of the
Theorem we have talked abom so far ~ has essemiatty
~wo pr(×~£s. Of ~he h>cal proof we have &rand 4 A 5
~>ccurrences {four of the first pa~% five of the second~
norie of both) and of the gIobal proof 2 A 20. While it is
nor d e a r whether 4 A 5 shmdd evaluate ~o 4, 5.9. or 20~
~here are certainly m ~ ~occurrences.
tn *he process of exposing these f%~ctswe have also
~burid numerous references to the 7 h e o ~ m or its Bohm
and Jacopini part~ which do nut comaia prri~xg~ mainly
in connection with the issue of sm~aured programming
For a sample of fifty see Arsac {1~ p. 3I~ Baker
{4, p. 991, Baker and KosarNu {A p~ 555], Banachowski
{& p. {15], Baaachowski e~ al. {7, p. 22], Bates
{8, p. 1%], Benson {10, pp~ 145-N6], Blorim arid Tiadell
{i2. p. 27~], Boehm {13. p. t13{, Bohl [|4, p. t40],
Chandra {19, p. 1] Cohen and Levi {2l, pp. 26% 225],
Cutik {26, p. 54], Dermmg {27, p. 10; 28, p. 216]. D~kstra
{29, p. 148], Donaldson 130~ p. 531, Engelfrie~ {34~ p.
2C~t, Fischer and Fischer {35, p. 461t, Goodman and
Hedemiemi [37 pp. I9-2@, Hard {39, p. 89], Hard,
Norvg, R o o d arid To D l, p. 2 I81, Hopkir~s {~, p. 59 I,
Hugbms [45, p. 591, Hughes arid Michtom {~, p. 611,
Jensen arid Toriies lath, ppo 228~ 236}, Kasai {49, p. 1771,
Knuth and F I w d {53, p. 31], Kosaraju {54, po 252I.
Leavenworth [56, p. 55], Ledgard and Marcx~tU {57, p.
6321, Lee and Chang {58, p. 65], Martin I6t, p. 51~
McGowan I62, p. 25], MiIler and Liridame<xt {~, p.
561, MiIls {69, p. 901, Myers ~71~ p~ tl0; 72. p. 5], N a z i
and Shaeidermari [73~ pp. 13-~4]~ Neety {74, p. 120],
NichoHs {75, pp. 4C~9-410f Partch {76, p. 12S21 Peterson,
Kasami, arid Tokura {78, p~ 51 l{, Prather {80, pp, t:59,
70f Van Ge|der {84, pp. 3, 5], Wise, Friedmam Shapiro,
and Wand {86, p. ~ / ] , Yourdon [g8, pp. |46-147; 89, p
107], Yo~rdori and Coristamir~e i90, p. 66], and Zdkow~
itz. Shaw, and Gannoa ~9I, p. 60!.
Maay Of"~he~ references c~ontain interesting relevam
material, but iri order ~o keep this irivesfigafion f%om
get{trig ou~ of" haad we wilt aot describe ~hem aN, Let us
just remark thai Bohl {14] infbrms as that B o h m and
Jacopini {~5] was '~mi~ially p~btb~hed in Italian ia 1%5,"
~hat Cohera and Levi {2 ! {extend fl~e Theorem *o parallel
programs, a.~d that Pray.her [gO[ proves a similar resuh ~
fbr Tuft:rig maclhines,
OBr story is given a final asd qm~e unexpected
frilkish twist by the e×islerice of a second versi<m of (he
Fheorem. ta preparation £)r thi£ we were carefu~ lo
choose ffm adiectiw~ %dditionaF' rather tha~ *~BooleaxF'
in its statemem~ In ~l~is version, auxiliary rather ~han
Be<~leaa variables are allowed, which are sm@y ,mew
variables of ~he types used ia ~he origimd flowchart.
Although not guarameed the ability to distinguish a~ wit1
between ~wo distinct vahms as in ~he Boolean version,
we are able ¢o freeze currem values m new variables, a~d
by re{esting them la{er are able ~o "~remember" the
outcome. (See the sample traasforma,qo~ ia ~he Apww.
dix.)
The reader is ~rged to Uy his hand at proving this
version, which is apparemty harder thar~ the first, tn face,
we are not aware of asy sirigle global proof of it. Curi-
ously (or should we say, lblkishlyL a fhll proof of this
version does not seem to be available in a~y single paper
and. as we shall see. cart be found only by juxtaposing
three papers (published, incidemally, o~ ~hree diflerem
coatmems .... ).
In 197 t, Cooper [23, pp. 47.....49] arid Engeler {33, pp.
93--.941 isdeperidentty pointed om that any flowchar~ can
be pm iri~o a ~*bkyck fbrm," which is basically a I reelike
flowchart in which branches are allowed to bend back-
war& but to ancestors orily b~ the same year, Ashcrofl
and Manna showed that riot any flowchart in ~his form
there is an equivalent while-program with additional
auxiliary variables. {This result is stated in {2, p. t48 t,
relying on a pr<×~f appearing in a preceding technical
report The full proof appears in their final version {3,
pp~ 135-t38], and also., with reference to [21, in Grei-
bach's m o n o g a p h {38, pp. 4.534.5.4].} Being unimter..
es~ed m mining the smmture of Ihe original program,
Ashcrofl sad Manna make no attempt to push their
vechnN~es arty further~
However~ m a paper published m t972, Hirose and
Oya {42, p p 369-370; 43, pp, 65--681, apparently wi~hom
having seen f21, lay down the final brick by proving tha,
eve©, whibe-program can be mmsfbrmed irito orie with
a single , using auxiliary variables. (In [43], fl~e
pr(yof starts from arbitrary flowcharts but the part leading
~o whik-programs is rather skeichy compared to [2],
whereas the transfbrma~Lion ~o single white4to fbrm is
de'ta:il~J .arid precise}
Att three comp~oaents of ~;his proof are '%col," Le,
p r ~ g d by iriductioa ori the sm,~c~ure of ~he program A
s-ample transfbrmation fl~om {42] appears iri the Ap~a..
dix. Here, ~oo, e:n~meratio~ {m¢~t~l~ difDarent versio~ss
of" the same paper) Nves ([23] V [33]) A ([2] V [381} A
~42I.
Arid ~} our folk ~ate er~ds arid barri~g ~r~expected
comp~ica~mn~, our Folk Theor~'ri has good chances of
livisg happily ever aflen We have {0m~d ~ha~ it has two
versior~s, the ~rs~ of which has two differem pr<~:~£~,o~e
of ~hese ~sentiNty going back m the cave-dwelling days
af Kkeene (1936} a~d ",'on No,mann (1946} The e~.xli~s
are ilk~s~rated ia Figure l, Besides over fifty refere~ce~
{be ACM Num~:~ 7
 {B~,ib,m a~d ~ac~,pif~i) {Beki~ / Mirkowska)

(K~¢cf~ / ~B~rk:~ GcdJs~me and yon Neum~mS)

flowch~rt ........................................................................................................................................................................................................................... ~ singqe ~hi~e4o

(Enge{e~ / ( ~ p e r } (A*hc*oti and Ma*ma) {Hir{x~e and O?'a)

~oc;d prcu.~f flowchar~ ................................................................ ~bk~& -l'orm .......................................................................... * ~hile-pr~ygram~ ................................................. "*single w h i t ~

to the Theorem or its parts, we have found (4 A 5) + (2 outgoing edge and test nodes have two, labeled 1 and 0.
A 2(;)) + (2 A 2 A l) printed occurrences of its proof; '~ There is one START node with no incoming edges and
which evaluates to 27, 36, or 64, depending, respectively, one outgoing edge, and at least one STOP node with no
upon whether "A" is interpreted as "max", " + ' , or outgoing edges. An interpretation I of a flowchart F
" × ' . These proof!s, which span forty-four years, also consists of a set of domains, an appropriate association
span the complete spectrum of recognized scientific lit- of domains with the variables of F, and an appropriate
erature: textbooks, monographs, survey articles, journal association of f~nctions and predicates over various
papers, conference proceedings, newsletters, theses, tech- cross-products of the domains with the fimction and
nical reports, lecture notes, letters to editors, and self- predicate symbols of F. Given an interpretatmn I and
referemial folk tales~ initiM values for the variables appearing in F, the way in
Quile a fblk theorem, it seems. which F proceeds to compute its values is straightforward
and is assumed to be known m the reader. Denote by' F~
the lunction, associating with each set of input vMues for
4. The Future F its output values, or a special "undefined" symbol if F
We have postulated the appropriately adapted prop- does not terminate. Also, let ash(F) and test(F) be the
erties of popularity, age, and anonymous authorship as sets of assignments and tests appearing in F.
criteria fbr determining if a statement is a tblk theorem, Next, we define an appropriate set ofwMle-programs,
and have illustrated by exhibiting a theorem:, the folkish- relative to sets of assignments and tests A and T. Define
hess of which seems to be beyond doubt. WH(A, T) as the least set such that:
As lbr the ff~ture, we envision three possible direc- (1) AC WH(A, T)
tions for further research: (2) if W~, V¢~ E WH(A, T) and P E T, then
(1) Compiling an encyclopedic list of folk theorems in (i) (W~; W~) E WH(A, T),
computer sdence. (it) ff P then W:1 else W~. E WH(A, T), and
(2) Investigating the related concepts of folk definition (iii) while P do W~ E WH(A, T).
andfi)lk technique (that is, proof=technique)~ Again, given an appropriate interpretation and initial
(3) Showing that folk facts such as P # NiP are fotk values for the variables appearing in a while-program W,
theorems~ the standard method of defining the computation of" W
is assumed to be knowm We use Wi simila@ as with
5. Appetldix flowcharts. Let WHdA, T) be the set of those elements
of WH(A, T) which contain at most one occurrence of
First, some definitions. Aflowehart is a finite directed wMie-do,
graph with nodes l a d l e d either with an assignment of Now. to be able to rigorously state the Boolean
the tbrm x ~- f(~/7) for vafiaNe x, function symbol t: and version of our Theorem. we let
vector of variables j~ or with a test which is a Boolean- Basra F) = ash(F) U { p~ ~- true. p~ ~-. r i s e }. I ~ t < w.
combination of e×pressions of the fom~ p ( f ) fi:~rpredicate
symbol p and vector j~\ Assignment nodes have one and

we would greatly appreciate poimer~ te ~ho~se¢~t~r~aces of preo£~ Btest ~ the set of Boolean combinations of tests m test(Fj
we might ha;'e mitred ia our ignor~nce~ia oui' haste, or e~herw~se, and expressions of the tbrm p~?. for t ~ i < ~.

Commurficafi~ns J u l y ~980
of V o l t , m e 23
the ACM Nu,mbe~ 7
 Fig. 2. Cooper's Global Proof.

/ /

[ --true,n 1)~ [B--false(n+l~

Fig. 3. Elgot'sProof.

where pl, p2 . . . are new variables and F is an arbitrary In order to illustrate the B6hm and Jacopini [15] A
flowchart. An interpretation is called nice if the pi range Mirkowska [70] "local" proof, we provide one of the
over the domain (true, false} and the test pi? is true (i.e., transformations of Mirkowska [70]:
evaluates to 1) iff the value of pi is true. Denote the set
while Pl do (Wg while P2 do W2)
of nice interpretations by NICE. Then the Boolean ver-
sion o f our Folk Theorem states:
(*) (VF) (3W E WH, (Basn(F), Btest(F)))
"pl ~-- Px"; pz ~-- false; while (pl V p2) do (if p2 then W2 else W1; "p2
(VI ~ NICE) (F, = WO. ~- e2"; ','pi "~- Pi"),
386 Communications , July 1980
of Volume 23
the ACM Number 7
where, e.go "'p: .......
P " >4ands tbr tf P~ ¢hen p~ ........tn, e
else p: ..... false
(~.×~per':, global proof I22] proceeds a/~ tbllows: (}ive~
a llowcharl {;~ denote the axs,ignmen: r~odes of F in m:'ne
arbitral T, b:a~ fixed order by o,,'n . . . . a ; < , and the test
node> s~mda:'ly by rose,,, . . . . to.re..... Think of alt {he
S T O P taxies a~ b~:ing n:m:bered rn + ]. }"or each I :~; a
5 n te~ ne.~¢(i) be she ~::bscrip: cor:'esponding So the
node adjaceru So ax< (by it5 outgoing edge) and f'or each
~ + t ~:: i ~ m let wue(i) and/}d.~e(i) be she ~ubscrip:~s
correG~onding ~o {he node:~, adjacent So to.< (by the i and
0 outgoing edges, respectively). Let i~ be ~he s:.~v;cript
corresp,.,r}di::g :o the node a<:tjacem So she S T A R T node,
As:tutor now thin a new variable B rar~ges over file set
{l, 2 .... m + 1}, and that we have assignments g ...... i
and tests g +~ i? fbr each i in that so< with the meaning
obvious, is should be clear that B can be replaced by
[ log m ] + I Boolean vafiable:~ encoding the val:~e o f B,
with B ~...-i and B = i? being replaced by the appropriate
sequences of m<signments and conjunctions o f te~:~
The white.-program W G:: W H d B a x n ( F l , Bge':g{F))
which is equivalent to F in the sense of (~), is Nve~
{using :he obvious translation into wbile-lbrm) by the
flowchart in Figure 2,
To illustrate Elgofs proof' {32]. we will ::or provide a
rigorous definition of his multiwire flowchart.s, but :rust
that the reader will be able to gather aa much as is
aee;ded from the folIowiag. For positive integers j and p
where j :~ p, the 'errtrial scheme j / ' o f [32] is simNy she
"'cable" of width p wish all but :he j'th compvment
disconnected:
: i+: j j+: p (;Jan. ]977), 9$-,{20.
Elgofs pr(x)f is illustrated by the muldwire flowchar~ of
Figure 3 in it, e 4 . nex~(i)~,,+.~, which follows a..~'~, is a
cable o f width m + t in which control, so to s~ak~ may
only flow through the next(i)~{h wire, which is ~himatety
cormected to a:rm,.,,.~, as expected in other words,
Co.owr's variable B g ~ h a N wired ~ rote Ne muRiwire
cable,5 o f Elgot% program.
In order to .scale the second version of ~he Theorem
rigorously, we let An<re(F) be the set of a~ignments of
she fbrm x +- f(f), where x and the componems o f fi
come from the union of {,J,, ua . . . . } and the se{ of
variables appearing in F. Here ~he u~ are eew va.riabte~,
and f i~ a function symbot a p ~ a f i : n g m F. Aees~jF) is
defined as the sol of Bc~Jean combiaatkm~; o f ~ g s of ~he
%rm N / a , fbr y as above and p i.n F The % t ~ l i a r y "
version of the Folk Theorem s, ate~:
{**) (VF) (SW e WH~{Aasn(F), A t ~ ( F ) )
(YI)(F, ,~ W 0.
As aa ilh>{ration of the {[331 ,./ {231) /' ([2~ V [38]) A
{42] proof of tbi.~ver~don+ we provide the ~rans%rmafion
of Iliro e ~nd Oya {42] analogow, ~o that of Mirkowe, ka
[70] given a~xvee:
~ ....f; g P~ th¢:t~ Wi; ~hlfe ¢J4~}{ P~ ~:h~g~W e~*~ iii .....2 if P~
{hera W~+
where k is the vector o f aI{ va:lable:~ appearing in the
original F,/~ is a vector of new variable:< and P~ is ~he
result of sub:stimting she ~/s for ~he x% in P Here° fbr
clarity, the clause else 2 +.--.-2 has b<ve~ omit~e,d.
Received tW7% ~evi,~d a n d accepted 2gJ0
Acknow[edgmems We are hTdebted ~0 N Pippenge~-
t\;',r helpful discussions aN?ut ffAk theorems i~ getters]
and about the one described in Section 3 m par~ic~ta~
Ak~:~,, thaeks go to M.I the people we have pestered
recently with questions aba:mt folk (~eorerms and trans-
lations l¥om flowcharts to ~bt|e's; m a y this. work help
erase the p~?sded lc~oks we often saw o~ their facee~ as a
rea~flL
R
L A : ~ J. U~ t:m~ge de pr%ramma::~n ~msasb~ch,eme.a~. Re~'.
~o "wbi~e- p~cVg~am~~a ICgbe~: ~rg~: 7L }97L pp. ~47,-i52,
ak Baker~8.5;. An algorithm 1:orsm~ae~nng~%weEam.J. ACM 2~
5. Baker.B.S. amJ Kes~r~ju, SR. A e v m ~ n ~ m of m~al;~k*:~¢~
break ~ d ~mx{~a{emerds..st A CM 26, ~ {3~ty ~¢d%,55S.-Pg%.
6. B oL {~v~Bgm~oa~o(peo~r~m of ipmg.~'am~by
V. ganachow.kL L.,e~ at,, Ar~i~ttgx~e~ion~o a~{~g~q~.bm};h~g~< 1~
Paw~ak_ Ed&..Ba>ac'hC:r Pabhcat,on&Wz¢~awoPv/[a:a&{977
2L_a~}
& {{a:{es~L1 ~ h ,g;c for ~:onwcl proWam ¢ievek~pmen~. P h D {h...
Cornet1 umv.. ~ag ~:479
9, B.ek~£.H DefiaaNe aWvta~i~m~m geee~a~aAcge;bt~,End~l~e
I~t4t~ ~e+OS~
I{}, gvm~O~. J P. S~r~ac.Ba::ed proS;staining teghm~a~e~. Pr~z [E.EE
Syrup o a (.;~'e@gf_ :5<fftm~re R e { ~ a N R B ~ New Y o ~ . Apr~ 1973~ p~
I L B~wr~e< {). },;kwecB:&:~machine& BIZ ~0 [970h 4 ~ - 4 4 2
Ilk Bhxm~. S.L... arid t i n d e i [ R A/gebra~c a ~ J g,eaph ~.e,~re,~,e
~aa~ae~e~w:<a~m:m o f ~4fuc,~zed ~,eme;s.
A@d-i:m#-WeMe) Reading, M~,,~ o pp. ~03--/28
{2i~ N J. ~V78
ASPcaWib~ ~#d ShT~mNew Y{~k, ~&~74
ef V~m~e 23
~.b~ACM N~mb~ "r
 18. Burks, A.W., Goldstine, H.H., and von Neumann, J. Preliminary
discussion of the logical design of an electronic computing
instrument, 1946. In J. yon Neumann, Collected Works, Vol. V, A.H.
Taub, Ed., MacMillan, New York, 1963, pp. 34-79.
19. Chandra, A.K. Degrees of translatability and canonical forms in
program schemas. Proc. 6th ACM(SIGACT) Syrup, on Theory of
Comptng., 1974, pp. 1-12.
20. Clark, K., and Cowell, D. Programs, Machines. and Computations:
An Introduction to the Theory of Computing. McGraw-Hill, New
York, 1976.
21. Cohen, A.T., and Levi, L.S. Structured flowcharts for
multiprocessiug. Comptr. Languages 3, 4 (1978), 209-226.
22. Cooper, D.C. B6hm and Jacopini's reduction of flow charts.
Comm. ACM 10, 8 (Aug. 1967), 463,473.
23. Cooper, D.C. Programs for mechanical program verification.
Machine Intell. 6, Edinburgh Univ. Press, 1971, pp. 43-59.
24. Culik, K. Structured algorithms and structured programming.
Rep. CS-79-40, Dept. ofComptr. Sci., Penn. State Univ., Aug. 1979.
25. Culik, K. Entry strong components and their applications (in
computer science). Rep. TR Nov. 79-01, Dept. of Comptr. Sci.,
Wayne State Univ., Nov. 1979.
26. Culik, K. What is a flowchart loop and about structured
programming. SIGPLAN Notices (ACM) 15, 1 (Jan. 1980), 45-57.
27. Denning, P.J. Comments on mathematical overkill. SIGPLAN
Notices (ACM) 10, 9 (Sept. 1975), 10-11.
28. Denning, P.J. Two misconceptions about structured
programming. Proc. Ann. ACM Conf., Minneapolis, Minn., Oct.
1975, pp. 214-215.
29. Dijkstra, E,W. Go to statement considered harmful. Comm.
A C M 11, 3 (March 1968), 147-148.
30. Donaldson, J.R. Structured programming. Datamation 19, 12
(Dec. 1973), 52-54.
31. Elgot, C.C. Monodic computation and iterative algebraic
theories. In Logic Colloquium '73, North-Holland Pub. Co.,
Amsterdam, 1975, pp. 175-230.
32. Elgot, C.C. Structured programming with and without go to
statements. IEEE Trans. Software Eng. SE-2, 1 (March 1976), 41-54.
33. Engeler, E. Structure and meaning of elementary programs. Proc.
Syrup. Semantics of Algorithmic Languages, Lecture Notes in Math.,
Vol. 188, Springer-Verlag, New York, 1971, pp. 89-101.
34. Engelfriet, J. Simple Program Schemes and b~rmal Languages.
Lecture Notes in Comptr. Sci., Vol. 20, Springer-Verlag, New York,
1974.
35. Fischer, B., and Fischer, H. Structured Programming in P L / I and
PL/C. Marcel Dekker, Inc., New York and Basel, 1976.
36. Goldschlager, L.M. Synchronous parallel computation. Ph.D. th.,
TR- 114, Univ. of Toronto, Dec. 1977.
37. Goodman, S.E., and Hedetniemi, S.T. Introduction to the Design
and Analysis of Algorithms. McGraw-Hill, New York, 1977.
38. Greibach, S.A. Theory of Program Structures: Schemes,
Semantics, Verification. Lecture Notes in Comptr. Sci., Vol. 36,
Springer-Verlag, New York, 1975.
39. Harel, D. And/or programs: A new approach to structured
programming. Proc. IEEE Specifications for Reliable Software Conf.,
Cambridge, Mass., April 1979, pp. 80-90.
40. Harel, D. On folk theorems. Comm. ACM 23, 7 (July 1980).
41. Harel, D., Norvig, P., Rood, J., and To, T. A universal
flowcharter. In Proc. A I A A / I E E E / A C M / N A S A Comptrs. in
Aerospace Conf. II, Los Angeles, Calif., Oct. 1979, pp.
218-224.
42. Hirose, K., and Oya, M. Some results in general theory of flow
charts. Proc. of the First USA-Japan Comptr. Conf., Sponsored by
AFIPS, Tokyo, Japan, Oct. 1972, pp. 367-37l.
43. Hirose, K., and Oya, M. General theory of flow charts. Comment.
Math. Univ. St. Pauli, XXI-2 (1972), 55-71.
44. Hopkins, M.E. A case for goto. SIGPLAN Notices (ACM) 7, 11
(Nov. 1972), 59-62; Proc. ACM Ann. Conf., Boston, 1972.
45. Hughes, J.K. P L / I Structured Programming. John Wiley and
Sons, New York, 1973.
46. Hughes, J.K., and Michtom, J.I. A Structured Approach to
Programming. Prentice-Hall, Englewood Cliffs, N.J., 1977.
47. Indermark, K. On a class of schematic languages. Technical rep.
82, Inst. for Res. and Programming, Gesellschaft fur Mathematik und
Datenverarbeitung mbH, Bonn, Germany, Nov. I974.
48. Jensen, R.W., and Tonics, C.C; Software Engineering. Prentice-
Hail, Engelwood Cliffs, N.J., 1979.
49. Kasai, T. Translatability of flowcharts into while programs. Z of
Comptr. and Syst. Sciences 9, 2 (Oct. 1974), 177-195.
388
50. Kleene, S.C. Introduction to Metamathematics. Van Nostrand
Co., New York, 1952.
51. Kleene, S.C. General recursive functions of natural numbers.
Math. Annalen 112 (1936), 727-742.
52. Knuth, D.E. Structured programming with go to statements.
Comptng. Surv. 6, 4 (Dec. 1974), 261-301.
53. Knuth, D.E., and Floyd, R.W. Notes on avoiding "go to"
statements. Inform. Processing Letters' 1, (t971), 23-3 i.
54. Kosaraju, S.R. Analysis of structured programs. Z of Comptr.
and Syst. Sciences 9, (1974), 232-255.
55. Kreczmar, A. Effectivity problems of algorithmic logic. In Annal.
Soc. Math. Pol. Series IV," b)mdamenta lnformaticae l, 1 (1977), 19-
32.
56. Leavenworth, B.M. Programming with(out) the goto. SIGPLAN
Notices (ACM) 7, 11 (Nov. 1972), 54-58; Proc. ACM Ann. Conf.,
Boston, 1972.
57. Ledgard, H.F., and Marcotty, M. A genealogy of control
structures. Comm. ACM 18, 11 (Nov. 1975), 629-639.
58. Lee, R.C,T., and Chang, S,K. Structured programming and
automatic program synthesis. SIGPLAN Notices (ACM) 9, 4 (April
1974), 60-70; Proc. Syrup. on Very High Level Languages, Santa
Monica, Calif., March 1974.
59. Linger, R.C., and Mills, H.D. On the development of large
reliable programs. In Current Trends in Programming Methodology,
Vol. 1, R.T. Yeh, Ed., Prentice-Hall, Englewood, Cliffs, N.J., 1977,
pp. 120-139.
60. Linger, R.C., Mills, H.D., and Witt, B.I. Structured Programming."
Theory and Practice. Addison-Wesley, Reading, Mass., 1979.
61. Martin, J.J. The "natural" set of basic control structures.
SIGPLAN Notices (ACM) 8, 12 (Dec. 1973), 5-14.
62. McGowan, C. Structured programming: A review of some
practical concepts. Computer 8, 6 (1975) 25-30.
63. McGowan, C.L., and Kelly, J,R. Top-Down Structured
Programming Techniques. Petrocelli/Charter, New York, 1975.
64. Merton, R.K. On the Shoulders of Giants: A Shandean Postscript.
Harcourt, Brace and World, New York, 1965.
65. Meyer, A.R., and Ritchie, D.M. The complexity of loop
programs. IBM Res. Rep. RC-1817, 1966.
66. Miller, E.F., Jr., and Lindamood, G.E. Structured programming:
Top-down approach. Datamation 19, 12 (Dec. 1973), 55-57.
67. Mills, H.D. Mathematical foundations for structured
programming. IBM rep. FSC 72-6012, Fed. Syst. Div., Gaithersburg,
Md., 1972.
68. Mills, H.D. The new math of computer programming. Comm.
ACM 18, 1 (Jan. 1975), 43-48.
69. Mills, H.D. How to write correct programs and know it. Proc.
IEEE Tutorial on Structured Programming, Washington, D.C., Sept.
1975, pp. 84-91.
70. Mirkowska, G. Algorithmic logic and its applications. Doctoral
diss., Univ. of Warsaw, 1972 (In Polish).
71. Myers, G.J. Reliable Software through Composite Design. Van
Nostrand Reinhold Co., New York, 1975.
72. Myers, G.J. Composite~Structured Design. Van Nostrand
Reinhold Co., New York, 1978.
73. Nassi, I., and Shneiderman, B. Flowchart techniques for
structured programming. SIGPLAN Notices (ACM) 8, 8 (Aug. 1973),
12-26.
74. Neely, P.M. On program control structure. Proc. Ann. ACM
Conf., Atlanta, Ga., 1973, pp. 119-125.
75. Nicholls, J.E. The Structure and Design of Programming
Languages. Addison-Wesley, Reading, Mass., 1975.
76. Partch, B. Improved technology for application development
management overview. Proc. SHARE XLI, Miami, Florida, Aug.
1973, pp. 1281-1300.
77. Perkowska, E. Theorem on the normal form of a program. Bull.
Acad. Pol. Sci., Ser. Sci. Math. Astr. Phys. 22, 4 (1974), 439-442.
78. Peterson, W.W., Kasami, T., and Tokura, N. On the capabilities
of while, repeat, and exit statements. Comm. ACM 16, 8 (Aug. 1973),
503-512.
79. Plum, T. W-S. Mathematical overkill and the structure theorem.
SIGPLAN Notices (ACM) 10, 2 (Feb. 1975), 32-33.
80. Prather, R.E. Structured Turing machines. Inform. and Control
35 (1977), 159-171.
81. Pratt, V.R. Semantical considerations on Floyd-Hoare logic.
Proe. 17th Syrup. on Foundations of Comptr. Sci., Houston, Texas,
Oct. 1976, pp. 109-129.
82. Rogers, H., Jr. Theory of Reeursive Functions and Effective
Computability. McGraw-Hill, New York, 1967.
Communications July 1980
of Volume 23
the ACM Number 7
KL { i ~ p ~ ] fa~ed poiv~s at@ a~gebr;~, w~*,h mih~imly king

i~> N ( r~p{* g<~, Pch';;h A<:a~d ~,fSci, i97g

V<YTL 2 !2
g~, W~ld, M A <:~f~cge~¢ aipr~ach ta~ b,,~r;~c{recur'sire dcfinhion3 Mann emem Science/ D, F, S ha a rio
H~>liam:lP~b Co~Am~e~dam, ~97? p p 331 3 4 1
Operations Research Edhor
g~ W~:/.:. {) S, F m q ~ ar~ i) P Sha?~rc,, S < , and Wand, M
g o , ~ e a ~ > v 0 l u e d k x p ::~::'tS V~75L431 4 5 i
87. W~A~ W A A Ca~< aga~ns~ @¢ p>~,:~ SI(]PIAN Nogice~' {A(M} 7o
Gamma Variate
1/ {Nov 1~)72),61 M< }>f>< ACM Aw< C o n f ;Bose{m, !972~
gag. Y~ urd<,~< ~,', levk~*N~e~ %tPe<~ge~um5.w~cmre a~d Deq.gtfn
P~eB~cedlall. Er~glewood Ctift%, N.J, t<&eS,
Gene Fa/ors
gJ)o Y'ou~'don E ,4fae~a~i#~ggke ,%'~r~.~ctur<'d7%'eh*uqmq. Scco~@ ed{{ior~,
Yovafdor'~ Pre:~< New 'Yo~k, :~29 {[ irsI edigiom How iv &f:a~age
with Increased Shape
S~rue~red Pr~Nr~mm*~g, 1~76}

l"~md,;~mee*t,A~<fie D~v*pf~ewo/' Compuge~ P~og~'ae.end Sy:~wm.~

Parameter Range
RC.H, Cheng and G,M, Feast
S¢~a'¢~ee Ft~:,Vt**~','mR amt De,~4gs. Pren~ice-}lalL Englewocd ( ~ift%, University o(Wates Ins@urn of Science
N J,, I~?79
and Technology

devdoped gamma ge~erator:s are compact.,

nasty p r o g r a m ~ , and ~ e m s s a saifnrm stoned over
the operathg range, t~t are o a b ~vNtd for a > 1. A
tran~ormation of variable together with a technique
suggested by" Kinder.man and Monakan of generating
random ,eariates u~ing the ratio of mniform variates are
cnmbiaed to produce a fatuity of generators r a i d far all
a > l / n where n is an arbitrary integer. Thus if s ks
greater than uNty, variates with a tess than unity can
be sam#ed. The cases n = 2 and n = 4 are considered
explicitly aeM are shown to retain the feaVares of
compactness, ease ~f and u~fform sDeed.
Key Words aud P h r a ~ gamma variates~ r m n ~
numbers, simulation
CR Categories: 5~5, 8,t

1, Introduction

Since the publication of the review paper of A~ki.nson

and Pearce {2], there have appeared a large number o f
gamma variate generators covering the case ~ > I where
~ is the shape parameter. One of the more noteworthy
methods is based on the generat technique of generating
random varia~es using the ratio of uniform variates
proposed originally by Kindermaa and Monahan {6].
Kinderman and Mona.ban [7] and Cheng and Feast [5]
;have suggested Mgorithms using this technique to gen-
erate gamma variates, The ~brme~ paper also comains
some sound criteria for the choice of an optima1 alg{~
Pem~i<'m h'~ copy wtho~t~ fee a ] ~ p ~ i ~
gra~te~ p~x~vided that ~.hec,epies are not made or d~tr/ba~e~ &~ dh'e¢~

~NU~ UK~

Cornm~m~c~mB~s h d ) igN)

~*e ACM Nu.r~*be~' 7