Complex Systems 1 (1987) 373-400

Candidates for the Game of Life in Three Dimensions'

Carter Bays
Department of Computer Sc ienc e, University of So uth Carolina,
Columbia, SC 29208, USA

Abstract. The game "Life" is defined in a strict sense and three

candidates for t hree-d imensional versions are presented. One of these
versions can be structured to contain an infinite number of parallel
t wo-d imens ion al universes, each of which allows for t he evolution of
Conway life objects. Various oscillators are described, and a few inter-
esting collisions between translating oscillator s ("g lide rs") and other
objects are mentioned.

1. Introduction-Conway's Game of Life

Most readers are probably familiar with John Conway 's two-dimensional
cellular automaton known as the "Game of Life" 13,4J. The game is "played"
by zero players on an arbitrarily large grid of square cells, where each cell
is either "alive" or "dead". Essentially, the game works as follows. Start at
generation one with some pa ttern of living cells (squares on the grid that
are filled in). To obt ain t he nex t generation , apply the following transition
rules concur rent ly to each cell, C, on the gr id, whet her filled in or not. Rule
On e: If C is living an d if it touc hes two or t hree living cells, it remains alive
for th e nex t generation; otherwise , C dies [i.e ., erase t he filled-in square for
next generation) . Rule Tw o: If C is not living and if it touches .exact ly
three living cells, C becomes alive [i.e., fill C in for next gen eration).
Readers familiar with t he game may recall t hat with ap propriate start-
ing patterns , we can obtain a host of stable and oscillating shapes, which
Conway and ot hers have given such whimsical names as "bee hive", "blink-
er", "clock", "pulsar", etc . Severa l oscillators translate across the grid
with successive generat ions; such oscillators are t rad it iona lly called gliders ,
a term which we shall use throughout this paper .

1.1 T he r ules of Life

We can formalize the rules for Life as follows. Define environment E as
t he number of living neighbors required to pr event a cu rrently living cell
..A p re lim inary report on so me of th is work appeared in [1,2].

@ 1987 Complex Systems P ub licaticne, Inc .

374 Carter Bays

from expiring, with El :5 E :5 Euo FertjJjty F is the number of neighbors

required to create a new liv ing cell, F, :5 F :5 F.... Define the transition rule
R as the 4-tuple (E,EvF1Fv ) ' For Conway's Life, R = (2333).
Naturally, we can construct other rules to apply to a two-dimensional
grid. For example , R = (3434) y ields a game known as "3-4" Life, which
exhibits a variety of osci llators which are totally different from those result-
ing from R = (2333) . Unfortunate ly, it is easy to produce starting patterns
in 3-4 Life t hat rapid ly expand forever . 'True, we can force (2333) to pro-
duce forms t hat grow without limit , but the int r iguing feature of (2333) is
that this can only be done with carefully cons tructed configurations.
We could also enlarge the neighborhood , or intro duce "aging", where
a living cell dies afte r a certain numb er of gen erations. Work has also
been done using a hex agonal instead of a square grid [5] . Conway 's rule is
much less compl ex than most of the hexagonal transition rules- and therein
lies its beauty. Our goal here is to describe similar elegant rules in three
dimensio ns which yield a large, interesting variety of stab le shapes and
osc illators, suppo rt one or more gliders, and when applied to any init ial
and relat ively haphazard configuration of cells, will ultim ately stabilize .
Hence, we restrict our use of the name "Life" to only those rules that are,
as Dewdney puts it, "worthy of the name" 121. T he following definit ion
formalizes this restriction .

D efinition 1 . A rule E1EuF,Fu defines a "Game of Life" if and only if both

of the following are true.

1. A glider must exist and must occur "nat urally" if we apply EIEuFiFu
repeatedl y to primordial soup configurations.
2. All primordial soup configurations , when subjec ted to EI EuF1Fu, must
exhibit bounded growth.

(Here we define primordi al soup as any finite mass of arbitrarily dense

randomly dispe rsed livin g cells.)
We have not specifie d in our definit ion jus t how many "soup ex peri-
ments" to perform before we conclude that a glider does not ex ist; this
question is best answered by considering the implicat ion of definition I-if
a glider does not con dense out of some haphazard arrangement of cells, then
there is little hope of creating one by bombarding some configuration with
a (man-made) glider. Thus, the "rarer" a glider is, the less likely that in-
terest ing configurations (e.g. a "glider gun"-a manufactured device which
produces an endless supply of gliders ) may exist.

2. Finding a rule for three-dimensional Life

In three dimensions, a cell can have from 0 to 26 living neighb ors; hence ,
we may construct a huge variety of rules of the form described above.
Specifica lly, we can have
Can didates for the Game of Life in Three Dimensions 375

(E" E u ) = (1,1), (1,2),(1,3), . .. (1,26);(2,2) ,(2,3) , ... etc.

for a total of (26 + 25 + . . . 1) = 351. A similar number of values for (P" Pu )
gives a total of 123,201 possible ru les. Fortunately, if we are looking for
a rule that behaves in a manner similar to Conway's rule, we can restrict
our scope cons iderably, since most of the possible combinations y ield forms
that expand rapid ly and indefi nitely or quick ly shrink and disappear. The
followin g theorems are of assistance.

Theorem 2. Any rule EIEuFIFu with FI ~ 10 cannot s upport a glide r.

T his is eas ily seen whe n one observes that a non- living cell adjacent to
a p lane of living neighborin g cells can h ave at most nine neighbors-s-an
upp er b ou nd for a cell adjacent to any plane grouping of cells . Thus, an y
for mation under ru le (X Y 10 Z) will ulti m at ely either shrink and disappear,
or will form a convex b lob whose ou ter surface of living cells may remain
in t u rmoil , but w ill never translate across the un iverse (see figure 1) .

Theorem 3. Any rule E1EuFlFu with Fi ::; 4 leads to unlimited grow th.

To prove theorem 3, simply start with a cluster of four neighboring cells

arranged in a square (see figure 2 for a more exotic example).
After testing several possibilit ies, it becomes obvious that rules dealing
with from four to seven neighbors have the mo st potential. Starting con-
figur ations t hat are operate d upon by rules (5767), (5777), (5566 ), (5755) ,
(4656) , (4655), (6767), (4567), (6766), etc., seem to eit her va nish qu ickly,
leavi ng little or no res idue, or grow ind efinit ely. Furthermore, none of these
ru les seem to supp ort a glider. Rule (5655) offers several int eresti ng small
osc illat ors (see figure 3), but its residue is rather sparse and an exhaustive
search h as revealed no glider. Ru le (5877), as well as other rules whose
environment range exceeds 3, leaves too much nameless debris and does
not seem to y ield particularly interesting configurations. Rule (4666) offers
several int eresting oscillators, but leaves much more res idue than (4555)
a nd does no t appear to support a glider. The same can be said about
(4566), (3455) , and (3566) . One shou ld observe at this point that we can
eas ily create ru les wh ich leave stable non-oscilla ting patterns just by uti-
lizing a small fertility range and a lar ge envi ronment range. Furthermore,
one can find an infinite supply of distinct osc illators with ar bi trarily long
periods simply by constructing dense "random" blobs and apply ing ru les
wit h F, > 10; for example, t he per iod of the oscillating blob in figure 1
un der rul e (10 2110 21) exce eds 100.

2. 1 The hest rules

Of a ll the ru les investigated, only R = (4555) and R = (5766) satisfy
defini t ion 1. (T hese games can b e denoted "Life 4555" and "Life 5766".)
376 Carter Bays

GENERAnON I
GENERATION a

GENERAliON 3 GENERATION 12 GENERATION 34

Figure 1: The rule (10 2110 21) frequently leads to ohjects that look
similar to the one shown. This object oscillates with a period in excess
of 100. By starting with large initial objects, we can create oscillators
with periods as long as we wish.
Candidates for the Game of Life in Three Dimensions 377

IT!

01':1 ~
GENERATION 0 2

Figure 2: The rule (4526) leads to immediate unbounded growth.

Here, the starting pattern was six centrally placed cells.
378 Carter Bays

Figure 3: Some of the osc illato rs for the rule (5655). Wit h initial ran-
dom sou p, (5655 ) produces very littl e residue . An exhaus tive search
has revealed that this rule supports no natu rally occ urring glider.
Candida tes for the Game of Life in Three Dimens ions 379

Ru le (5766) leads to stable and oscillat ing forms (see figure 4) t hat are
similar in many ways t o Conway's (2333) , as discussed below. One of the
characteristics of Life 5766 is t hat the t hree env ironment states a llow for a
large number of small stable asymmetric objects. For example, if we confin e
our scope to the stable forms t hat ca n be contained wit hin a 4 x 4 x 4 cube ,
t here are well over 100 varieties. (F igure 5 depicts just a few of t he m any
stab le shapes t hat ca n be created by removing from one t o four cells from
a 24-element stable object.)
Rule (4555), alt hough somewhat less pro lific than (5766), may ulti-
mately be a more int erest ing r ule. For one t h ing, (4555) requ ires mo re
t ime to "set t le down " t han (5766) ; hence, there is more of a poss ib ility for
interesting intermediate re acti on s. Moreover , it is formed simply by a dding
2 to Conway's rule, R = (2333). Perhaps the most fascinating feature of
(4555) is that there exist an abu ndance of small stable an d oscillating forms
that usu ally exh ibit symmetry of some sort. This r ule will be discu ssed in
more det ail in section 2.3.

2.2 A com p a r ison between Conway's L ife and t h r ee- d im ensional

L ife
Be fore proceeding further, we should examine t he re la t ions h ip between the
above t hree-dimensional Life r ules and Conway's two -d imensional r ule , R =
(2333). De fine a Conway obje ct as any con figuration of cells, stable or no t,
th at ex ists at some p oint during Co nway's ga me . Ce lls in Conway objects
w ill have coo rdinates (XhYi,O)j t hat is, t he object lies in t he Z = a pl an e.
We shall further employ the following de finit ions .
Defi nit ion 4. An expansion of a Conway object is form ed in three dim en-
sions by m aking copies of all livin g cells (Xi , Yi, 0) in to the adjacent Z plane,
i.e, (xi,Yi, I) .
Hence, the expansion h as twice as many living cells as t he original Conway
object. It mayor may not behave in an interest ing manne r when s ubjected
to one of t he th ree-dimensional Life rules: R = (5766) or R = (4555).
Definition 5. A projection of a three-dimens ional Life object into t wo di-
mensions ex ists if and only H both of the following are true.
1. A ll of the livin g cells (Xi, yi, Zi) lie in t wo adjacen t planes. For th e
sake of discussion , let these planes be Z = a an d Z = l.
2. T he p air of cells (Xi,yi,O) an d (xi, Yi,l ) are eit he r both alive or bot h
dead .
Definition 6. A n analog of a Conway object in th ree dimensions is an
expans ion which, wh en subject ed to the appropriate three-dim ensi onal Life
rul e, y ields after each and every generatio n a proj ect ion id entical t o the
original Conway object for the same generation un der th e two-dimens ional
rule, R = (2333) .
380 Carter Bays

GLIDER

Figure 4: A few of the many small stable forme under Life 5766.
Candidates for the Game of Life in Three Dimensions 381

Figure 5: Her e are just a few of t he many stable shapes under Life
5766 that can be created by removing one or more cells from a 24-
element symmetric stable object. There are too many such forms to
illustrate.
382 Carter Bays

The above definitions fac ilitate statement of the following theorem.

The orem 7. A Conway object has an analog under the three-dimensional

Life rule R = (5766) if and only if the Conway object has both the following
charactedstics at every (subsequent) generation:

1. A non-living cell in the neighborhood of the object cannot have six

living n eigh bors.
2. A living cell cannot have Bve neighbors.

The proof is obtained by examining tables 1 and 2.

C onway O bject (5 7 6 6) Expans ion

number of neighbors , N N) state of cells at N state of cells at
I

when cell at (Xi, Yi,O) is alive (Xi , Yi, O) and (x j, Vi,l ) (Xi l Yi, - I) and (Xi , Yi , 2)
N Next state N Next state N Next state
0 dead 1 dead 1 dead
1 dead 3 dead 2 dead
2 alive 5 alive 3 dead
3 alive 7 ali ve 4 dead
4 dead 9 dead 5 dead
5 dead 11 dead 6 alive
6 dead 13 dead 7 dead
7 dead 15 dead 8 dead
8 dead 17 dead 9 dead
Table 1: Comparison of the number of neighbors and the status for
Conway cells an d Life 5766 cells. For exa mple, if a Conway cell is alive
and h as four neighb ors (line 4 in th e table), then n ext generation it
will die, as will th e pair of cells in the Life 5766 expansion . Whe n
a cell in th e Conway object has five neighbors, the next generation
of 5766 expansion will have new live cells in plants adj acent to t he
expa nsion, t hereby destroy ing t he analog.

Notice t hat the behavior of t he ex pansion in Z = 0 and Z = 1 under

R = (5766) is iden ti cal to Conway's Life . A deviation only occurs in the
Z = - 1 an d Z = 2 planes; these deviations are the restrictions imposed by
theorem 1.
Upon further examination of tables 1 and 2, we obtain the following
corollary.

Corollary 8. T he three-dimensiona l Life rule R = (5766) y ields behavior

that is m ore analogous to Conway's Life than any other three-dimensional
rule that we may const ruct as R= (E1, Eu, F1, Fu).
T he implicat ions of t heo rem 7 are startling . If on e exam ines a ll t he
small stable an d osci lla t ing Conway forms , one not ices t hat a great many
Can dida tes for the Game of Life in Three Dimensions 383

Conway Object (5 766) Expans ion

numb er of neighbors, N, N, state of cells at N , state of ce lls at
whe n cell at (Xj,yj,O) is dead (Xj, Yi,O) and (Xi, Yi, 1) (x;,y,, -l ) an d (x;,y;,2 )
N Next stat e N Next state N Next state
0 dea d 0 dead 0 dead
1 dead 2 de ad 1 dead
2 dead 4 dead 2 dead
3 alive 6 aliv e 3 dead
4 dead 8 dead 4 dea d
5 dead 10 dead 5 dead
6 dead 12 dead 6 alive
7 dead 14 dead 7 dea d
8 dead 16 dead 8 dead

Table 2: Here we are concerned about next generation status for cells
t hat are not alive, but are in the immediate vicinity of live cells. When
the Conway object contains vacant cells with six live neighbors, t hen
the ne xt generation of t he Life 5766 expansion will have new live cells
in planes adjacent to t he expansion .

of t hem satisfy the cr iteria of t he above theorem (see figure 4). Co nway 's
glider h as an analog un der (5766) , as do some of the more comp licat ed
oscillators. Un fortunately, many of t he more int eresti ng and impor t an t
Conway objects do not h ave analogs: for example, there is no analog for
t he "glider gun" and ot he r so-called breeding oscillators.
P re liminary t esting has revealed t hat collisio ns b et ween gliders an d
ot he r objects, though occasionally analogous, usually yield non-analogous
res u lts. Sometimes t he first few generations after impact of analogous ob-
jects behave nicely, but sooner or lat er t he condit ions of theorem 7 are
usu ally violated; when this happens, the object, t heret ofore confined to
two planes, almost always forms a roundish t hree-dimensional mass that
usu ally dies rather quickly, but occasionally stabilizes.
Note that analogous behav ior is very narrow in scope-it takes place
entirely in two adjacent parallel pla nes. Obvious ly, we may alt er any t hree-
dime ns ional glider-object collision by shifting one of t he participants in t he
Z direct ion . T hus, if t he analogs lie in t he Z = 0, Z = 1 planes, we can
sh ift one of the objects by one, two, or three Z planes an d ach ieve entire ly
differen t collision results. Furthermore, we need not confine our objects t o
nearby Z planes-a glider can, after all, attack from a perpendicular plane
(see figure 6). Hence, analogous be havior is at most a small subset of Life
5766, which is replete with its own objects and collisions.
It is , of course, rather convenient to have an immediate supply of kn own
stable and osc illating Co nway analogs a lready available for Life 5766. Fur-
thermore, we will soo n see t hat it just mig ht be poss ible to cons t r uct a
three-dimensional glider gun by placing appropriate objects on either side-
384 Carter Bays

most likely in the Z = -2 and Z = 3 p lanes.

2.2.1 Time-space barriers

If we bu ild a stable p lana r for m where each living cell has seve n neighb ors,
then no bi rths can ever occ ur in the two parallel ad jacent planes; t hese
planes are "dead". A port ion of such a form, ca lled a time-spa ce barrier,
is shown in figure 7. If the flat part lies in t he Z = 0 pl ane (extending an
un sp ecified am ount in t he positive X and Y directi ons) , t hen no glider or
other form approach ing from a higher Z p lane can eve r penetrate into the
Z = 1 pl ane. Of course, t he sa me is t rue on the other side of t he barrier,
a nd we have not cons idered t he boundary, wh ich here h as bee n stabilized
with eigh t-element cubes. Nat urally, we m ight choose to have our bar r ier
extend to infini ty in t he X and Y dir ecti ons.

2.2.2 " Nearly 3-D" Life and Conway's game as a su bset

Construct t wo arb itrarily la rge parallel time-space barriers and place them
init ially qui t e some distance apart. Life forms under R = (5766) wou ld
behave in their usua l unrestricted fas hion insofar as our distant b arriers
would allow. But now we will move t he barriers closer toge t he r . As we do ,
evolving Life forms wou ld be "squeezed"; growth in t he Z directi on becomes
more and more inh ibited . For example, when t he barriers are separate d by
six pl an es, all life is confined to t he fOUI planes in the m iddle-what we have
her e is a "nearly 3_D" Life where eac h spacing of t he barri ers exhibits a
ver sion of the game whose behavior is disti nct from any ot her configurat ion.
Now consider what happens when t he barriers are four p lanes apart (figure
7). All life must t hen b e confined to two planes. Recall from theorem 7 that
the (5766) analog to Conway's Life breaks down on ly b ecause of growt h in
the Z direct ion . But now we have preve nted su ch growth ; hence, an analog
to th e entire Conway Life uni verse is contained between the barriers. For
that matter, we cou ld const ruct an infinite nu mb er of parallel Conway Life
universes .
Of course, noth ing can slip out (in the X or Y dir ecti ons) from b et ween
finite barriers-at leas t as t hey are constructed. For example, t he edges
would inte ract with any escap ing glider, thus ruling out a simp le glider gun.
Possibly, some osc illators cou ld be appropriate ly placed to allow a glider to
leave t he v icinity un impaired ; this is undoubtedly t he eas iest approach t o
gun construction .

2.3 The rule R = (4555)

We can build charts sim ilar to tables 1 and 2 for R = (4555), yie ld ing t he
following res u lts.

T heorem 9. A stable Conway object has an analog under R = (4555) if

and only if each living cell in the Conway object h as exactly two n eighb ors.
Candidates for the Game of Life in Three Dimensions 385

.'
A~
. L/
~E ..... ... . , .. ..
.,,,.
·.·.·" · · ··. ·.· ·+.··
oiL : :~:::
.". ,...... ,,,

2
m 5
6

GENERATION

~I';.'.: ' '

~~ :·i

..".
/ :
IPB
~
7 9 10

Figure 6: A collision b etween objects that are Conway analogs in Life

5766-a glider and a "clock" . Here, however, the glider is attack-
ing from a p erp endicular plane. M far as th e ent ire configura t ion
is concerned, only generations 7 through 10 are analogs of Conway
Life. At generation 10, theorem 1 part b, is violated; hence, growth
in th e Z direction is initiated. For this particular collision, the living
mass seemed t o remain confined to two planes for a short while before
suddenly "releasing" a stable 24-element object . The usual result of
a collision is a rather quick annihilation of both objects.
386 Carter Bays

Figure 7: The object at the left is a portion of a "t ime-space barrier".

No life can get within one cell of the flat portion of this form , which
can be made to extend indefinitely. The finite parallel barriers at the
right have four planes between. Hence, a "mini Conway universe"
analog can exist in the two middle planes as long as no shape wanders
too close to the barrier boundaries.
Can dida tes for the Game of Life in T hree Dimensions 387

Corollary 10. A Conway object that changes from one generation to the
next h as n o analog under the three-dimensiona l Life rule R = (4555).

The con clus ion is t hat a lt hough R = (4555) h as an occasional analogous

for m or two, it s "universe" is completely different from t hat of Conway-
and , for that matter, d ifferent from the universe of R = (5766). It is int er-
est ing that alt hough t he two t hree-dimensional Life games be have in totally
d iffere nt ways, t he y bo th seem t o stabilize rather qu ickly, with (4555) re-
quiring somewhat more ti m e. For example, if we start w ith a ran dom
configurat ion in, say, a 70 X 70 X 70 gr id , bo t h rules w ill stabilize after
about 30 t o 70 generations, dep ending upon the starting density. Colli-
sions between gliders and other sm all objects us ually die after about 5 to
20 generations-unless t hey hap pen to yield debris.
T he ea rly discovery of a t ot ally distinct glider (figur e 8) was the cata-
lyst that led to the extensive invest iga ti on of t his rule . The (4555) glider
cont ains ten elements and , like Co nway's glider (and its (5766) analog),
h as a p eriod of fou r afte r wh ich it has moved a distance of J2 in one of
twelve di rection s- per pendicular to one coo rdinate ax is and at an ang le of
45 degrees wi th the othe r two.

2.3 .1 Additional shapes of Life 4555

The re lative low den sity of stable life (result ing when we start with pseudo-
random primordial soup) is more th an compensated for by t he rich variety
and symmetry of small Life forms . Several of t hese "naturally" occ urri ng
forms a re shown in figure 9.
We may create primordia l soup in seve ral ways. Perhaps t he easiest
me thod is to initiali ze each cell in our un iverse according to t he ou tput
of a random number gen erator. For example, on ou r exhaustive j ourney
t hrough t he universe, as we pass eac h cell generate a random number, T.
Then, for some fixed constant, k, if r < k , let that cell be alive, ot he rwise
not.
We can get some idea of the re lative paucity of Life for ms by examining
t ab le 3. The entries were found by app lying the "soup st irring" rule (4512)g
repeatedly g t imes to a 70 X 70 X 70 space that had been filled with about
40 random living cells. T he rule (4555) was t hen emp loyed . Ten samples
were made with g set to various values be tween 14 and 27. All for ms wer e
allowed t o stab ilize; t his us ually occ urred after about 60 or 70 gene ratio ns .
The stable res id ue was then t alli ed . T he lone observed glide r was no t
counted . Per haps gliders are m ore common t han this table wou ld ind icat e,
as t hey may have gone off t he screen (or collided with something) befor e
being obs erved.

2.3 .2 Glider collisions in Life 4555

The huge number of reflect ions and rotations of the small stable forms
leads to a myriad of distinct possible collisions betwee n gliders and ot he r
388 Carter Bays

Fig ure 8: The (4555) glider, showing the four states. When state one
is encountered again , the glider will have moved one unit up and one
forwar d (i.e., in the positi ve Y Z direct ion).
 Candidates for the Game of Life in Three Dimensions 389

B3 @ m ffiB~
~~~
J
r-
In
11

Figure 9: A few of the small symmetric forms which occur "naturally" in

Life 4555.
390 Carter Bays

Object Number of elements Occurrences of

(see fig. 9) in the object object at stability
A 6 124
B 8 37
C (period = 4) 8 or 10 (ave. = 9) 36
D 6 28
E (period = 4) 8 or 10 (ave. = 9) 14
F 12 3
G 6 3
H 10 2
I (period = 2) 7 2
J 10 1
K 10 1
L (period = 4) 10 1
Glider 10 o (1 observed)
M 9 1
Total vo lume of residue objects = 1,761
Total vo lume of "universe" = 3,430,000
Approximate density of res idue = .00051

Table 3: The most common Life 4555 objects resulting from co n-

densation of primordial so up. T he app roximate density of t he live
residue was .0005. This value can vary considerably depending upon
the density of the original primordial soup .

objects. One would expect (and indeed one finds) the usual result of such
collisions to be the ann ihilat ion of both objects. However, preliminary
exploration has revealed a surp rising numb er of interesting interactions
(see (1) Appendix A for an extensive list). With the low (- 10-') density
of stable life ultimately settling out from "soup", the fact that so many
Life forms result when a glider (ten elements) collides wit h another small
object (abo ut ten elements) implies that some rather mysterious forces are
at work . A typical interesting collision resu lt is shown in figure 10.

2.3.3 Manufactured stable forms

Although most of the stab le Life 4555 shapes found "in nature" [i.e., as the
result of some evolving popu lation, randomly created or otherwise) rarely
contain more than about a dozen elements, it is possible to construct exotic
stable forms (see figure 11). These forms would be highly unl ikely to appear
as the resu lt of a primord ial soup exper iment. One should note that such
forms are harder to construct for Life 4555 than for Life 5766; this is due
to the more limited safe environment range.
Candidates for the Game of Life in Th ree Dimensions 391

GENERAT ION GENERAT ION 2

GENERAT ION 7 GENERATION 8

GENERAT ION 12 GENERATION 13

Figure 10: One of the more interestin g Life 4555 glider collisions.
Here, a glider collides with an object called a "blinker". Th e origin al
glider an d th e blinker are destroyed, but a new glider appears. If th e
original glider was traveling in th e (- Y - Z ) direct ion, then the new
one will be heading toward (-X + Z).
392 Carter Bays

g
l;]G:b:ID
'.. [JLciG
o

Qb . o
~
...
IMMOB ILIZED : ::c . . :.
GLI DERS "':":
. . ",. :.
,

Figure 11: Manufactured Life 4555 stable forms such as these would
never be found by conducting primordial soup experiments; they must
be carefully constructed.
Candidates for the Game of Life in Three Dimensions 393

3. Other poss ibilities

We may broaden our allowable configurations somewhat by treating cor-

ner, edge, and side ne ighbors as distinct typ es. Hence, we h ave six "face"
neighbo rs, twelve "edge" neighbors and eight "cor ner" neighb ors (see figure
12). For t he Life game (whether in two or three dimensions ), all ne ighbors
have equal weight, wheth er touch ing on a face or a corner. We int rod uce
the following notation. Let R denote any rul e of the form E1EuFiFu. Then ,
Life.100 R den ot es a neighborhood consist ing only of t he eight corner neigh-
bors; Life.010 R denot es a neighborhood cons ist ing onl y of the twelve edge
neighbors , and Life.DOl R similarly deals only with the six face neighbors.
Thus, Life 4555 can also be written Life.111 4555 , et c.
Two additional gliders ha ve been found -one in Life.Oll 4544 and an -
ot her in Life.110 4544. These are the on ly ru les t hat seem to support
gliders. (Here, we do not consider rules such as Life.x 1111, which provably
allows infinite growth .) Unfortunate ly, pr imordial soup experiments with
Life.110 4544 and Life 011.4544 usu ally lead t o unboun ded growth; hence,
these rules have not been investigated furt her .
It is very important t o note that if we expan d our rul es to con sider
the more gen eral neighborhood configuration (ther e a re 226 not counting
reflections and rotations), then other gliders probably exist and definition
1 can likely be sat isfied- but t he beauty of Life is its simp licit y. The next
section discusses the most elegant rul e of all.

4. Another Game of Life

Mu ch energy has been expended in an effort t o discover a Life game in

t he two-dimensional hexagonal grid, where each cell has six ne ighbors. Un-
fortunately, no worthy rul e exists, unless we consider exot ic configurations
such as Golay surroun ds 151.
But let us expand the two-dimensional hexagonal grid into three di-
mensions. We then obtain the hexahedral tessellation, a universe where
neighbors can be represented by the corner s of the 14-sided hexadecahedron
(figure 13). Here, there are 12 neighbors which line up in four intersecting
hexagons. These hexagons form four non-orthogonal planes which are par-
allel to the sides of a regu lar tetra hedron. This configuration can also be
represented by "densely packed spheres" an d conforms to certain natural
crystal struct ures .
Note tha t t here are two d ist inct universes , U and U,. (see figure 13). For
the remainder of this discussion, we shall dea l only with U j U,. is obtained by
reflecting U in a vertical plane. Also, we sh a ll use spheres to represent cells,
although we could also use hexadecahedrons, or , more s imp ly, points with
neighbors connected by lines of equal length. Aga in , although there are 2 12
possible neighborhood configurations, we sha ll only con sider the quantity
of neighbors and not their orientation. To narrow down the possible Life
ru les, note the following th eorems.
394 Carter Bays

1"-" )-_--.;..-4
6 FACES -,
(00 1) "-

r-... ............ ..
12 EDGES ....
( 0 10) /

6 CORNERS
( 10 0)

2
GLIDER FOR
L1 FE.O 1 1 4 5 44

GLIDER FDR
LIFE. 1 10 4544

Figure 12: Two glide rs have been discove red for life rules that would
be "worthy of the name ," except primordial growth unde r these rules
is unlimited .
Candidates [or the Game o[ Life in Three Dimensions 395

t he two stetes of the u t n e glider

~
. r. . . •, ,~
Yo • •,
r. I';' • . r. .
•/ . . I. ~ .1 ' '0 '
,
<
,
2 ,
,

r"_ ' ,i8iQ"

o

"-1< »:
.
'I;
r.
I.
7.
or.

.
•
r emove eny one
or t he fo ur
teotcet ec cell s
for 0 dif ferent
gl1der crtenteucn

~c
\

,@ "~ - ~ "

, 1 rr <' 2 1
@I,
I':'
r.
I.
r-
Littl e
Glhle r
\.-r

o ?~ r
.»

~'
.' .d Generetlon 0

qs)~
-,

\ ' (~r e /'

, ~~
O~)
~?
, ,
- r

10
.~ sene reu cn 17

Figure 13: (C lockwise from upper left) The skelet on of a tetrahedron

shows the four planes (A-D) that ar e parallel to the four hexagonal
neighborhood s in U. U,. is obtained by reflecting U in a vertical plane.
The two universes are distinct. The little glider has two states and
travels parallel to an edge (a-f) . By removing anyone of the cells
(r,s), we obtain a different orientation; hence, there are 24 different
orientat ions. The six glide rs at gene ra t ion zero w ill travel parallel to
edges [e- I]: at gene ration 17, t hey have advanced as indi cat ed . As it
advances, the 100element big glide r resemb les a frog. The 7-element
lit tl e glide r is the "tadpole».
396 Carter Bays

The o r em 1 1. Any rule EjEuF, Fu where .Fl ~ 4 canno t suppo rt a glider.

The proof is analogous t o that of t heore m 3, and t he theor em below
correspo n ds to t heo rem 2.

Theorem 12. Any rul e E,EuF,Fu wh ere Fl ::; 2 allows unlim it ed grow th .
T hus, if any Game of Life exists, it mu st be of the form E,E u3 3. The
only rul e th at seems t o exh ibit glide rs is (rather nicel y) R = (3333) (see
figure 13). T his rule a lso has bo unded growth; hence, ou r defini ti on of a
Life game "worthy of t he name" has been satisfied .

4 .1 Symmetry of obj ects in Life 3333

One of t he most in teres t in g features of Life 3333 is that every stab le or
oscillating object so far discovered exhibits symmetry in some form. For
convenience, cons ider t he tetrahed ron in figure 13. itself constructed of
cells in U . (By the way. t his t et rahedr on is not a st ab le str uct ure in Life
3333.) The four ax is planes correspond to sides A, B, C , an d D; the edges
me nt ioned b elow refer t o a, b , c, d , e, a nd f. So far , objects discovered have
exh ib ited symmetry abo ut:
1. a point (six-way)
2. a point (four-way)
3. a line p erpendicular to a side (six-way)
4. a line perpendicul ar to a side (three-way)
5. a line p erpendicular to a side (four-way)
6. a lin e p erpendicular to a side and the side
7. a line perpendicular to two opposing edges [e.g. edges a and e)
8. a lin e describ ed above and a plane p erpendicular to this line
9. a pl ane parallel to a side and perpendicular to the opp osing side
10. a plane described a bove and a plane perpendicular to this plane
Wh en figuring t he symmetry of a periodic obj ect, we mu st cons ider whether
obj ect ph ases ar e reflections of each other, rotations of each other, and so
on . We need t o find the symmetry type in order to determine how many
different orientations of the object exist. For exam ple, if an object exhibits
six-way symmetry a bout a point, then there is only one distinct ori entation.
On t he ot her hand, if a to tally asymmetric obj ect were found (so far it has
not b een), it would have 48 distinct orientations. Obj ects with distinct
ori entations numbering a ll the factors of 24 have been found (see figure
14).
Life 3333 is currentl y under intense investigation an d m ay ultimately
be revea led as the most interesting Life game of all .
 Candidat es for the Game of Life in Three Dim ensions 397

Q O?l A '
qp B ' r.r.
Qu
c

6} b 2
't:

1
z
r.

o (;A E £J
0& ~O
~0
'

~
r.
, r. O~r.
r.
r.
1 2

'M
r. t:
r«
t. r.
4
2

H~
r. ~
r.
~
G
r. r. r.
~'/
r: I.

r.r. 2 r
r. r. 2 r Y
.' I. r. t. r.
I r. r. r. r. r.
1

~~
J
t' r. 't:
r. f.

r. r I
I r. r.
r. I.

o~" g~
r.
Or.
'r. r. L r. r.
r.r. 1
~
~
00 M
o
~ 2
~
0
1 2

Fi gure 14: Here are just a few of th e many objects in Life 3333. All
t ha t have been discovered so far exhib it symmet ry of one form or
ano ther. Period two oscillators abou nd, but oscillators with other
periods have b een found . Mot ionless stable objects seem to be Quite
rare.
398 Carter Bays

5. Programming the Games of Life 4555 and Life 5766

Fo r the sm a ll co m puter, two types of a lgor it hms may be em p loyed. Mo re
exotic ap proaches pr ob abl y can b e found , but t he methods be low are rea-
sona b ly st ra ight forwa rd. The discussion here uses R = (4555); R = (5766)
could j ust as easily be em ployed. The non-orthogon al Life 3333 prese nts a
speci al programming problem and will be deferred to lat er work .

M ethod A. Tally living cells in the leading nine-element "plane" as it

sweeps across space. Thus, at a given ce ll, we need to make on ly 11
calculations instead of 26. Essentially, as we check neighbors for each
cell, we add the living cells found in leading n ine elements, subtract
the t ra iling n ine, an d ad just t he center. Th is concept can be ca rr ied
one d imens ion furt her, tallying a leading t hree-element line in each
ni ne-element plane; how ever, we only sav e about three calculations.
Method A is usefu l wh en developing s t art ing sh apes or creating pr i-
mord ial soup. It also can be used when t he density of living ce lls is
re latively h igh.
Method B. This method utilizes the fact that populations are relatively
sparse. Keep wit h in each cell: (a) whether it is dead or a live and (b)
t he number of living neighbors.

Option 1 . As we are examin ing each cell in the universe , if a ce ll is

nonvital an d has fewer than five neighbors, take no action . This
can be done wi th one test [i .e., rep resent "alive" by adding, say,
32 t o the number of neighbor s . Our loop then checks wh et her
a cell is less than five). We then check for other less freq uent
situation s. If a cell is to d ie, subtract one from eac h of its neigh-
bo rs, and if a new cell is bo rn, add one to each neighb or. This
method will speed u p or slow down as t he p opulation grows an d
shrinks.
Option 2. Keep the living cells so rt ed in some fas h ion (or in a hash
table for quick acc ess). Then, we need only examine ch an ged
cells and alter next-generation s tates wh ere appropriate while we
are tallying neighbors. Hen ce, if we find that a new ce ll is born,
we immedi ately place it in t he list of changes for the nex t gener-
ation. Similarly, if a ce ll dies , we place it on the list of cha nges
indicating that it is to be removed . Aft er t he entire universe has
been examine d , we merge the "changes" (adds and deletes) with
t he prior gen eration 's living ce ll list. This algo rit hm should use
hash t ables for t he changes and shou ld keep track of the neigh-
borho od s and wh ether a cell is alive, in a three-dimensional array
(sparse or otherwise). Not e that our three-dimensional universe
array can he as large as memory all ows, with little degradat ion
to the execution time. Furthermore, if we use sparse hash ar-
rays, an in finit e universe is poss ible. This a lgorithm is rathe r
Candidates for th e Game of Life in Three Dimensions 399

tricky to im plem ent, but can be exceedingly fast. It h as b een

implement ed in C on the Ridge 32 and the Macintosh compu t-
ers and evaluates mo st patt erns at a bo ut 4 t o 15 gen erations per
secon d, dep ending upon t he pattern size.

5 .1 Execution speed of the algorithms

(For all the cases ab ou t to be discussed, we have ignored the time required
to disp lay the cubes.) Let N be the size of the universe cube edge an d let
p be t he number of living objects. We should note t hat a "wors t case" but
trivial a lgorit hm simp ly looks at every ce ll and t a llies t he neighbors. This
algorithm runs in t ime 26kN 3 , where k is som e cons tant that depends upon
the t ime req uired on a p ar ticular computer to fetch an element from a three-
dimensional array. Method A reduces this t ime to l1kN s . M ethod B option
1 increases the spee d to ab ou t kN 3 + 26k t p, where k t ~ 2k . Sin ce p <
NS , we obt ain a significant increas e in speed . Method B option 2 further
enhanc es the speed by eliminat ing the NS te rm, achi eving approxima t ely
26k t p + kzp . Here, k z is somewh at lar ger t han k and de pe nds upon t he
hashing methods emp loyed . A fur ther characterist ic of t his method is t he
fact that a particular cu be may come and go in the "cha nges" list severa l
t imes during evalu ation of a given gen eration before its final st atus (alive or
dead) is de t ermined. Experience has in dicated t hat for Life 4555 t he re are
usually t hr ee times as m any references to t he "changes" hash table during
gen eration eva luation as there are actual changes that finally get m erged
and plo tted at the end of t ha t particular gen eration evalu ation. Som e sort
of exotic parallel architecture could avoid this.

6. Summary and conclusion

The two rules for three-dimensional Life "worthy of the name" in the or-
t hogonal univers e have been presented a nd investigated. The rul e (5766)
relates closely to Conway's two-dimensional Life-to t he exten t th at any
three-dimensi on al rul e could. Man y of Conway's shapes h ave an alogs un-
de r (5766) ; fur thermore, Conway's ent ire uni verse, or an infini t e number of
para llel universes, can be contained with t ime-space b ar riers .
The rul e (4555) yields a distinct r ich un iverse of small symmetric stable
an d osc illating for ms. T his rule has little in commo n with eit her Conway's
Life or t he rule (5766). Other rul es such as (5655) and (4666) are int erest-
ing in t hat they support small symmetric osc illators, bu t they do not fit
defin ition 1.

6 .1 Possible limitations
There are two important differ ences between the behavior-at-large of Life
forms in two-dimensional Life and three-dimensional Life. Stable residue
event ually resulting from appropriate ly primed "r andom primordi al soup"
occupies about 5 percent of Conway 's uni ver se, but only about .05 to .1
400 Carter Bays

percent in a t hree-dimensional Life univers e. This paucity of population

seems to be compensated for by the lar ge number of small stable forms,
t heir reflect ions and rotations; hence , interest ing interactions are abundant.
Life forms, t hough sparse, are intense.
The second difference, un fortunately, may ultimately be a factor that
prevents either three-dimensional Life rule from yield ing exotic constructs
similar t o Conway's huge "puffer train" or "sp aceship factory" . Many con-
figurations in Co nway 's Life a llow a large number of generations to elapse
before settling down . (For example, the seven-element "acorn" evolves to
a maximum of 1057 livi ng ce lls an d finally stabilizes at generat ion 5206 .)
He nce , the possibility of complicated inter med iate interactions. Most t hree-
dimens ional Life inter actions , on the other hand, converge rap idly-more
so under (5766) than (4555). (Moreover, t he line between conve rging an d
diverging rules is a fine one; e.g., R = (5755) exh ib its unlimi ted growth .)
T his is not meant to im ply that, for example, a glider gun doe s no t exist
in t hree di mensions . In fact , under (5766L it is likely that one mig ht be
constructed by stabilizing t he emerging glider from a two-dimensional gu n
analog t h at has been confined wit h a t ime-space barrier. Such a constr uc-
t ion would probably consist of carefully orchestrated osc illators sit uated at
t he barrie r edge near t he emerging glider.
On the ot he r hand, a "native" glider gun under either (5766) or (4555)
will b e difficult to discover- the task appears more difficult for (5766 ) Life
since a non-ana logous glider do es not appear to exist. Of course, a three-
d imensional gun of any type would open the uni vers e to the construction
of sparse b ut purposeful Life forms.
The recent discovery of Life 3333 in the hexahedral tessellation is cur-
rently undergoing int ense investigati on and, due to the hig hly sy mmetric
nature of its objects (and the relevance to inorganic crystal structure), may
eve ntually t urn out to be t he most in teresting three-d ime ns ional Game of
Life "wo rthy of t he n ame ."

References
[11 Carter Bays , The Game of Three -dimensional Life, unpublis hed document .

!21 A.K. Dewd ney, "T he game Life acquires some successors in th ree dimen-
sione," Scien tific American, (286) 2 (Feb ru ary 1987) 16-22.

[3] Martin Gardner , "On Cellular Automata, Self-reproduction, t he Garden of

Eden and the Game 'Life',n Scientific A merican, 224 (2 ) (1971) 112-118.

[4] William Poundstone, The Recursive Universe, (Morr ow and Company, Inc.,
New York, 1985).
[5] Kendall P reston, Mo dern Cellular Automata T heory and Applicat ions ,
(Plenum Press, New York , 1984) .