Waves, Crystals and Chaos
A Survey of State Symmetrical Cellular Automata

0. Contents

Introduction       top

This is a survey of all state symmetrical two state two dimensional cellular automata with rules based on cell counts using a Moore Neighbourhood in a closed torus universe. It finds three main classes of interesting behaviour, waves, crystals and chaos. Chaotic behaviour when the universe fills with ever changing patterns. Wave behaviour where the boundary between areas of different state supports complex waves. Crystallisation when the universe fills with unchanging patterns which often have low period oscillators trapped in them. In addition to minimal behaviour two small groups of rules show edge effects and the Structural properties. The edge effects are one dimensional oscillators on stable boundaries and the structural properties show up when the universe size has a major impact on the evolution of random patterns. The Fredkin replicator cellular automata being the prime example.

In more detail all rules where classified into 4 categories:

Some rules where found to have the following Boundary effects:

The term state symmetrical is used to describe Cellular Automata (CA) that have the property that the states are all equivalent. The initial condition of the universe of any CA is defined by the pattern made by the different states that the cells initially have. A Cellular Automata is state symmetrical if the value of the different states make no difference to the evolution of the pattern. More formally initial patterns are made by attributing each cell in the universe to one of n states P1 .. Pn each of which takes a value mapped on a 1:1 basis from the n states CA1 .. CAn. Of the CA. In a state symmetrical CA the outcome is the same for all 1:1 mappings.

This survey limits itself to 2 state 2D CAs with rules based on cell counts using a Moore Neighbourhood. There are 256 distinct state symmetrical CAs. The survey was done using Golly a tool for running CAs. A control panel StateSymCP was written in lua to aid this investigation. Statistical analysis was performed with the help of the R package. These CAs are traditionally defined by the number neighbourhood counts for a cell to be born and for a live cell to survive. The most studied state symmetrical CA is known as Day & Night which is defined as: B3678/S34678. The state symmetry is seen in that the list for survive is the opposite of the list for not survive (not Survive count = 8 - survive count) thus a state symmetrical CA can be defined just by the rules for being born. Following the practice of Wolfram we shall treat the counts on which cells are born as bits in a binary number bit 0 set 1 for born on count zero thus B3678 is the binary number 1 1100 1000, however to combine brevity with ease recovery of the bit positions we will use a hexadecimal notation giving H1C8 for this rule. The binary notation is used in some descriptions below adding X to mean either 1 or 0.

The number of gliders found in each rule is mentioned when listing gliders in each category. This information comes from David Eppstein's glider pattern data base.

Definitions       top

Initial Pattern

The square torus universe is initialised with half the cells of each state as far as possible. This is done in one of three ways:

Complimentary Rules

Every CA has a complimentary CA that creates the same patterns by one CA changing state when the other does. In binary notation the complimentary rule is the rules ones' compliment. E.g. Rule Day & Night is B3678/S34678 which in notation defined above is 1 1100 1000 or H1C8. Its compliment is B01245/S0125 which is 0 0011 0111 or H037. H1C8 + H037 = H1FF the definition for ones' compliment. In general we will use the even rule of the complimentary pair to identify a rule and where useful include the other in round brackets e.g. H1D0(H02F).

Golly always runs the even rule for both rules of a complimentary pair as it is not easy to run the odd rule in an open universe. Golly's RuleLoader was used to create three state definition to allow the odd rules to be run.


The seed is used to initialise the random sequence for the random initializations. When initialising with bands the seed pattern is used to modify one boundary by adding or removing cells in the following way. The seed is made up of the characters '0', '1' ,2' or '3' where each character dictates the state of successive pairs of cells one on each side of the boundary. These cell pairs normally define the boundary by being of opposite state. The characters have the following effects:

'0'  No change
'1'  The boundary cell perpendicularly above changes state
'2'  The boundary cell perpendicularly below changes state
'3'  Both boundary cells change state
Seeds on an orthogonal band
F#1 Seeds on an Orthogonal band
Seeds on an orthogonal band
F#2 Seeds on a Diagonal band

The seed can unbalance the proportion of cells in each state as will a universe whose size is an odd number. The sequence of patterns created by a seed can not be predicted with the exception of rules in the Structural Category.

Symmetrical seeds on initial bands can create symmetrical patterns. The symmetry of the seed does not effect the symmetry of a random pattern. Both mirror symmetry and rotation symmetry occur. On diagonal bands symmetry occurs when pairs of symbols align with the zigzag pattern of the diagonal boundary. Seed "12" is rotational symmetrical on both diagonal and orthogonal bands but seed "102" is only rotationally symmetrical on an orthogonal band. Symmetrical patterns generate half the variation of non symmetrical patterns and preserve the ratio of cells of each state from one generation to another.

Strobing, Reverse Strobing

Strobing is the flashing behaviour which one of the complementary pair of rules often exhibits. A rule and its complement both show the same sequence of patterns but the states which make up the patterns differ. If the state of a cell changes state under one rule then it would not under the complementary rule and visa versa. In a universe which has all cells with the same state (e.g. is empty) even rules cause no change and odd rules cause a complete state charge every generation.

A rule is said to exhibit reverse strobing if after running for a while the even rule typically strobes a pattern more than its complement. In this document where appropriate RS is appended to the rule name to indicate reverse strobing e.g. H072RS. It is easier to see the changes over one generation viewing the odd rule when reverse strobing occurs.

A test was performed of all the Chaos and Structural category rules to see how often an arbitrary cell changed state creating the Strobe Ratio Data. The measurement was made over 240 runs of 200 generations between each chaos CA and it's complementary CA. The 240 runs where made up of the combinations of 60 seeds and 4 widths. A strobe ratio of less than 1 indicates normal strobing and a ratio greater than 1 indicates reverse strobing. The strobing type of H016 and H116 are unexpected both look like they reverse strobe but the measured strobing ratios of 0.78 and 0.8 respectively imply otherwise.

Another measure taken is the State Change Ratio. This is calculated as the proportion of the 256 different neighbourhood pattern that cause a state change for the a rule compared to the number that the complimentary rule would have caused. In general this was similar to the strobe ratio but with the strobe ratio showing a larger range. At the low end is rule H088 which had a Strobe Ratio of 0.1 and a Sate Change Ratio of 0.33 while at the height end rule H076RS had a State Change ratio of 9.14 and a State Change Ratio of just 2.88.

Golly will accept the rule for either complementary pair but will actually run the one which does not strobe an empty universe. Golly's rule RuleLoader feature was used to create a three state rule where one state is not used but satisfies Golly's condition that there is a state that does not change when zero neighbours of the that state. This was embedded in a lua script StateSymCP developed to investigate state symmetrical cellular automata. This script allows swapping between complimentary pair rules and helps the investigation of reverse strobing.

Static Debris

Static Debris is the small still life and short period oscillators often left in islands of one state which only effect the pattern when a boundary moves over them.

Random Walk Speed/Drag

The speed a boundary round the universe moves is referred to as the Random Walk Speed and is a characteristic of Category Wave. An object moving at random will on average be a distance from its starting point which is proportional to the square of the time elapsed. A boundary made up of linked objects each of which is moving at random while remaining linked will not have moved so far.


A fluid is a chaotic mixture of the two states which has a characteristic proportion of cells of one type to the other. If the ratio is 50:50 this term is not useful but a ratio of 3:2 would imply one of 2:3 as well. No measurement of fluid ratio has been attempted to date.

End Rate Analysis

Patterns under rules in the Crystallising and Wave categories settle down to a final state after a time which tends to be longer in larger universes. For rules in the Crystallising category this is when the universe becomes crystallized. For Wave rules this is when peaks in the waves beak through the band and the band collapses. Crystallising was also observed in some rules in the Chaos category for small universes see Small Universe Crystallisation below.

The following algorithms are used is End Rate Analysis.

Blowoff       top

When the universe is initialised with two bands of opposite state some rules will generate lines of cells parallel to the band in sections which are not modified by the seed. This is termed Blowoff. The following describe the behaviour with seed "0".

Blowoff from Orthogonal bands

Two types of Blowoff occur. Stable Blowoff where parallel lines spread through the universe and Resonating Blowoff where lines spread through the universe and also change width each generation. In both cases newly created lines of cells change state next generation so stable Blowoff creates strobing parallel lines in even rules.

Blowoff from Diagonal bands

Types of Blowoff

F#1 B-R1/S1/1/1S Gen 1
53x53 seed "0"
F#2 B-R2/S2 Gen 1
53x53 seed "0"
F#3 B-R4/S4 Gen 1
53x53 seed "0"
B-R1 H00C-w53s0g500.png
F#4 B-R1 H00C Gen 500
53x53 seed "0"
B-R2 H006-w53s0g500.png
F#5 B-R2 H006 Gen 500
53x53 seed "0"
B-R4 H00A-w53s0g500.png
F#6 B-R4 H00A Gen 500
53x53 seed "0"
B-3C H03C-w53s0g12.png
F#7 B-S1 H03CRS Gen 12
53x53 seed "0"
B-S2 H182-w53s0g6.png
F#8 B-S2 H182 Gen 6
53x53 seed "0"
B-S4 H00A-w53s0g500.png
F#9 B-S4 H01A Gen 7
53x53 seed "0"

B-R1: The 1st generation is a single diagonal line. Subsequently generating strobing parallel bands of varying width. Rules with the binary patterns X X00X 1100, X X1X0 1100 and X X010 1100. See F#1 and F#4.

B-R2 The 1st generation is pair of diagonal lines meeting at the corners. Subsequently generating strobing parallel bands of varying width.
Rules with the binary patterns X XXXX 0X10 excluding X 000X 0010. See F#2 and F#5.

B-R4 The 1st generation is pair of diagonal lines meeting at 2 sides. Subsequently generating strobing parallel bands of varying width.
Rules with the binary patterns X XXX0 1X10. See F#3 and F#6.

B-S1 stable 1st generation is a single diagonal line
Rules with the binary patterns X XX11 1100 add additional diagonal lines each generation giving a strobing checkerboard pattern.
Rules with the binary patterns X X101 1100 result in strobing bands of varying width ending quickly in a P2 oscillator dominated by bands three cells wide.
See F#1 and F#7.

B-S2 The 1st generation is pair of diagonal lines meeting at the corners
Rules with the binary patterns X XX00 0010 add additional diagonal lines also joined at the corners each generation. See F#2 and F#8.

B-S4 stable 1st generation is pair of diagonal lines meeting on 2 sides
Rules with the binary patterns X XX00 0010 add additional diagonal lines each generation giving a pattern of strobing parallel lines. See F#3 and F#9.

B-1 1st generation is a single diagonal line, no change in subsequent generations.
>Rules with the binary patterns X XX0X 1000. See F#1.

B-1S 1st generation is a single diagonal line, no change in subsequent generations.
Rules with the binary patterns X XXX1 1X10 strobe the single line. See F#1.

Category Chaos       top

Chaos patterns expand to full the universe and are chaotic, these CAs show four distinct visual patterns Maze, Network, Flecks and Uniform. Maze pattern show vertical and horizontal lines with right angle corners. Network rules produce network patterns from the interleaving of two fluids each dominated by different a state. Flecks rules show small short lived islands of a single state that are appear and disappear quickly, some migrate as their boundaries change. Uniform patterns occur when a large portion of the cells change state almost at random like white noise forming no visible structures.

Chaos rules can be divided into the following categories:

Subcategory Maze

Chaos Maze Rules:
H072RS, H074RS, H076RS, H088, H08A, H08C, H172RS, H174RS, H176RS, H188, H18A,H18C

Maze Rules with Gliders:

Maze Rules patterns are made up of orthogonal lines one cell wide with right angle junctions. The universe is divided into stable areas with continually moving boundaries between them. In small universes the universe crystallizes. In principle some large universes could crystallize if a stable pattern can tessellate it completely. F#10 shows rule H088 as an example of a large maze with scattered small changes over one generation.

Subcategory Network

The division into Large, Medium and Small sub categories was done arbitrarily by observation.

Chaos Large Network Rules:
H016RS, H01A, H01CRS, H01ERS, H022, H02ERS, H04ERS, H05ERS, H062, H064, H08ERS, H09CRS, H09ERS, H0A2, H0AERS, H0CERS, H0D2, H0E2, H0E4, H116RS, H11ARS, H11CRS, H122, H12ERS, H14ERS, H162, H164, H19CRS, H1A2, H1AERS, H1CERS, H1D2, H1DCRS, H1E2, H1E4

Chaos Large Network Rules with Gliders:

Chaos Medium Network Rules:
H014, H02CRS, H036RS, H03ARS, H04C, H052, H056, H05ARS, H05CRS, H06ERS, H096, H09ARS, H0A4, H0ACRS, H0B2, H0C4, H0DCRS, H0EERS, H114, H12CRS, H136RS, H14C, H15ARS, H15CRS, H16ERS, H196, H19ARS, H1A4, H1ACRS, H1B2, H1EARS

Chaos Medium Network Rules with Gliders:

Chaos Small Network Rules:
H032, H066, H0CC, H0E6, H132, H13ARS, H152, H166, H1C4, H1D4, H1E6

Chaos Small Network Rules with Gliders:

The universe consists of two or more interlaced fluids. Within each network pockets expand and contract. The pockets often contain small islands of the other fluid which join and break off from the edge as the network connections continually break and reform. The interface between the fluids is often an area of chaos. The subdivision into Large Medium and Small was done visually. F#11 shows the reverse strobing rule H1E1(H016) as an example of a large network with considerable change over one generation. The changes for rule H016 are all the cells which did not change in H1E1.

(b) Plus 1 Generation of Change
(b) Plus 1 Generation of Change
F#10 H088 Maze F#11 H1E1(H016) Large Network

Subcategory Flecks

Chaos Flecks Rules:
H010, H018, H028, H048, H04A, H06CRS, H0A8, H0D4, H0D6RS, H0D8RS, H118, H128, H148, H156, H198, H1CC, H1D6, H1D8RS, H1EERS

Chaos Flecks Rules with Gliders:

Random patterns in which small short lived islands of a single state appear. In some rules the islands vanish very quickly in others persist for a few generations changing shape and migrating around. F#12 to F#15 show an example of a fleck in rule H0A8 lasting over 12 generations it is nearly invisible in static images but stand out from the background of change.

(b) Plus 4 Generations of Change
(b) Plus 4 Generations of Change
F#12 H0A8 Flecks Generation 5,760 F#13 H0A8 Flecks Generation 5,764

(b) Plus 4 Generations of Change
(b) Plus 4 Generations of Change
F#14 H0A8 Flecks Generation 5,768 F#15 H0A8 Flecks Generation 5,772

Small Universe Crystallisation is discussed further below. Rule H1D8RS has some wave behaviour which is described separately below.

Subcategory Random

H012, H024, H026, H02A, H034RS, H038RS, H044, H046, H058RS, H06ARS, H092, H094, H098, H0A6, H0B4RS, H0B6RS, H0B8RS, H0BARS, H0C6, H0CA, H0DARS, H0EARS, H0ECRS, H112, H124, H126, H12A, H134RS, H138RS, H144, H146, H14A, H158RS, H16ARS, H16CRS, H192, H194, H1A6, H1B4RS, H1B6RS, H1B8RS, H1BARS, H1C6, H1CA, H1DARS, H1ECRS,

Random Rules with Gliders:

The universe changes chaotically with no structure.

Category Crystallising       top

There are 77 CAs that crystallise to fill the universe will a fixed pattern. Gliders have been found in four of these CAs. An analysis was performed on how long it took for the universe to crystallize with universes of different sizes and with 60 different seeds The results are the FindEnd10-50 Data Set. This also identified the period of any low period oscillators that might have been trapped in the crystal. Some CAs where never observed with trapped oscillators while others had more that one narrow band round the universe with trapped oscillating patterns resulting in an oscillation period too long to measure. In some cases a narrow band round the universe is a one dimensional CA. It is noted that some of these the one dimensional CAs are not state symmetric see Trapped One Dimensional Cellular Automata.

The type of pattern made when the universe is initialised with a band is greatly affected by the type of Blowoff.

(a) Generation 0
(b) Generation 10
(c) Generation 33
(a) Generation 0
H0BE Ortho
(a) Generation 10
H0BE ortho
(a) Generation 39
F#16 H002 Universe 50 cells square seed "13" F#17 H0BERS Universe 50 cells square seed "13"

These CAs can be sub categorised by the way crystallisation progresses and by the type of pattern after crystallisation.

Crystallising Network Rules:
H03CRS, H042, H0BCRS, H0C2, H102, H13CRS, H142, H1BCRS, H1C2

Crystallising Network Rules with Gliders:

Interlaced networks connected diagonally. Some of these rules have solid Blowoff with initial diagonal bands but more interesting crystals with initial orthogonal bands. F#18 shows Rule H03CRS.

Crystallising Maze Rules:
H004, H006, H008, H00A, H00C, H00E, H078RS, H07ARS, H084, H086, H0F2RS, H0F4RS, H0F6RS, H0FARS, H104, H106, H108, H10A, H10C, H10E, H178RS, H17ARS, H184, H186, H1F2RS, H1F4RS, H1F6RS, H1FARS and H1FCRS

Crystallising Maze Rules with Gliders:

Alternate lines in maze like patterns. Two of these (H078RS & H178RS) are border line Minimum changes category as for some seeds the CAs crystallisation halts almost immediately. F#19 shows H008.

Crystallising Blobs Rules:
H020, H03ERS, H040, H060, H0A0, H0DERS, H0E0, H11ERS, H15ERS, H160, H18ERS, H19ERS, H1A0, H1BERS, H1C0, H1DERS and H1E0.

Crystallising Blobs Rules with Gliders:

Blobs of each state consolidate from the chaos with single or double cell islands dotted in them some of which are oscillators. In some rules the boundaries of the blobs support waves which die away. Similar in appearance to H09ERS in chaos where the boundaries do not stabilise. F#20 shows H18ERS.

Crystallising Minimal Rules:

Crystallising Minimal Rules with Gliders:

Patterns expand into empty space and crystallise with minimal change. H082, H07ERS, H0BERS and H182 have longer period oscillators than the others. H07ERS and H0BERS particularly from diagonal bands. The two images above of H07ERS in F#22 where made with seed "13" in a universe 50 cells square with an orthogonal band which has an oscillation period of 224 generations. F#21 and F#22 show H07ERS.

Crystallising Edge Rules:
H030, H050, H090, H0B0, H0D0, H110, H130, H150, H190, H1B0 and H1D0.

Crystallising Edge Rules with Gliders:

Edge behaviour is described in subsection 10. When initialised with a random pattern these rules behave in a similar fashion to the Wave Category rules.

The types of pattern after crystallisation from a random pattern are :

F#18 H03CRS Network
F#19 H008 Maze
F#20 H18ERS Blobs
H07E Random
F#21 H07ERS Minimal

The type minimal initialised from an orthogonal band :

H07E Random
(a) Generation 20
H07E Random
(b) Generation 21
F#22 H07ERS Minimal

Category Minimal       top

Crystallising Minimal Rules:
H000, H002, H070, H080, H0C0, H0F0, H0F8RS, H100, H120, H140 H170, H180, H1F0, H1F8RS

Crystallising Minimal Rules with Gliders:

The behaviour of these rules can be defined as rules where the evolution stops in less generations than the universe size They all have a final population ratio of 4:5 or closer in FindEnd10-50 Data Set.

Minimal rules generate the almost the same pattern whether a cell is born with 8 neighbours or not e.g. H0C0 and H1C0. The difference being limited to isolated single dead cells left when cells are not born with 8 neighbours.

Sub Category- No change: H000, H100.

Sub Category- Still Life after a few generations: H002, H080, H0C0, H0F8RS, H120, H140, H180, H1F0, H1F8RS

Sub Category- Edge: H070, H0F0, H170

Category Structural       top

Structural Rules:
H054, H0AA, H154 and H1AA

Structural Rules with Gliders:

All these rules have a State Change Ratio very similar to the Strobe Ratio which is almost 1.

Rule Strobe Ratio State Change Ratio

T#1 Structural Rules Strobe Ratio and State Change Ratio

The rule H0AA has special symmetry. The rule is called Fredkin and is B1357/S02468 in the B/S notation in the binary notation used here it is 0 1010 1010. CA Fredkin in an unbounded universe is a replicator. A patten size n × n becomes nine copies, one where the original was with 8 new copies centres m cells apart after m generations where m is larger then n and m = 22k and k is an integer. More generally if j is the smallest integer such that n22j the universe only contains copies of the original pattern at generations p×(22j) for every integer p > 0 and seemingly random pattern at other generations. Further details can be found in PEDRO and CHUA's work.

In a torus universe the result is dependent on the universe size. In a universe size 2n × 2n the original pattern reappears after 2n - 1 generations. In other sizes either the original reappears or one of the successor patterns reappears periodically.

The rules H154 and H1AA also exhibit Structural features. They crystallize in small universes and apart from very small universes the oscillation time of the final pattern is the same for all initial seeds in the same size universe.

(a) Generation 0
(b) Generation 1
(c) Generation 14
(d) Generation 15
(e) Generation 16
(f) Generation 17
F#23 Fredkin CA:H0AA Universe size 32 × 32 The original pattern re-appears from chaos. Seed "000000013"

Category Wave       top

Wave Rules with Gliders:
H068, H0C8, H0E8, H168, H1A8, H1C8 and H1E8

Wave Rules with Gliders:

Rule H1D8RS has some wave behaviour but is different from the others and not categorized as Wave but as Subcategory Flecks. It's wave behaviour is described separately below.

Category Wave rules support complex waves on boundaries between areas of different state. When seeded with a random pattern all rules in the Category Wave form blocks of similar state and convex boundaries shrink. This leads to either one state dominating containing islands of the other state or separate bands of each state forming round the universe. In the former case the result that the islands continue to shrink leading to an empty universe possibly with Static Debris. This is very similar to rules of the Edge subcategory of Category Crystallising. In the latter case complex waves form on the bands which perform a random walk ending either in a collapse of one of the bands when the boundaries touch or both boundaries becoming frozen probably with trapped oscillators in them.

An interest in checkerboard patterns instigated this survey and one feature of rules with interesting checkerboard patterns is that they also support walking diagonal boundaries. This survey identified eight CA rules that exhibit deep waves on the diagonal boundaries. The seven described in this section and H1D8RS described below Gliders have been found in all of theses CAs. All these rules generate Blowoff from straight orthogonal boundaries, B-R1 for H1A8 and B-S1 in the others. From diagonal boundaries H0C8 and H1C8 have B-1 and the others have B-1S Blowoff.

These rules are examined in detail using an initial pattern with half the universe one state and the other half the other state. This is set out as diagonal bands and complex waves form along the seeded boundary. Initialised with horizontal bands Blowoff occurs leading to unpredictable results which include islands shrinking to nothing, bands with all boundaries having complex waves, chaos with large islands forming and disappearing and checkerboard patterns with static horizontal boundaries and complex waves on vertical boundaries.

Diagonal Bands with Asymmetrical Seeds

Asymmetrical waves form along the seeded boundary and it performs a random walk until it meets another boundary, one area then collapses and shrinks leaving the universe all one state sometimes with Static Debris. The Wave Break data set contains the results of an analysis of how many generations it took for boundaries of a wave to touch an a gap appear in the band. All the rules in Category Wave have diagonal Blowoff of either B-1 or B-1S so the initial band is lined with a row of cells connected diagonally after one generation. This is static thereafter in B-1 and strobes in B-1S. This line of cells means that when a waving boundary touches a static boundary changes quickly run along the whole boundary setting it oscillating. The analysis identified how the time to collapse changes with universe size. The seeds used for the analysis where chosen to avoid generating symmetrical waves.

(a) Generation 6,498
(a) Generation 6,507
(a) Generation 6,518
(a) Generation 65,29
(a) Generation 6,538
F#50 H0E8 Universe 50 cells square seed "1113" Breaking at Generation 6,518

The (b)'s in F#26 to F#32 below show the statistical fit for the random walk step size = a × U b where U is the size of the universe is generated from the Wave Step Size data set. The (c)'s show the fitted line from the (b)'s over the initial part of the range. The vertical line indicate the start of the fitted section. Some of the rules had some results removed when it was known that the band had broken before the measurement was taken. In some rules small universes end in chaos rather than an empty universe and this has not been corrected for.


F#26 (a)
F#26 (b)
F#26 (c)

F#27 (a)
F#27 (b)
F#27 (c)

In rule H0C8 with initial diagonal band the band sometimes stabilises with a trapped oscillator e.g. Seed "1231" in a universe 60 cells square stabilised with oscillators of period 6 after 763 generations.


F#28 (a)
F#28 (b)
F#28 (c)


F#29 (a)
F#29 (b)
F#29 (c)

F#30 (a)
F#30 (b)
F#30 (c)

H1A8 has some similarities to H1D8 in that a layer of chaos develops between areas of different state however in this case the layer never extends into the solid state areas. It is does generate its own activity rather than be just a part of the wavy boundary.

The deep waves generated by H1A8 fill up most of small universe creating a universe full of chaos resulting in a distortion of the results for universes smaller than 45 cells square.


F#31 (a)
F#31 (b)
F#31 (c)

F#32 (a)
F#32 (b)
F#32 (c)

Rule H1E8 from a random initial pattern sometimes forms orthogonal bands which often stabilise with long period oscillators trapped, e.g. Universe size 75 seed "113" stabilised a band after 741 generations with a period of 1098.

Symmetrical Seeded Bands

When the seed on an initial orthogonal boundary has mirror symmetry the wave generated does not perform the usual random walk and the number of cells of each state is constant. The wave always passing through two fixed points one in the symmetrical centre of the seed and the other at the opposite side of the universe. If the wave amplitude exceed the width of the band the wave must breakdown however if the universe dimensions are even both bands are of equal width so the equality of the number of cells in each states is preserved. The result is that the wave alternates between horizontal and vertical passing through a mixture in between. Rule H0C8 usually forms stable orthogonal boundaries which stops the oscillation.

(a) Generation 0
(b) Generation 397
(c) Generation 5490 Period 6 Oscillator
F#33 H0C8 28 × 28 cell universe Seed "000000000000012"

Waves on Orthogonal Bands

After Blowoff islands of single state emerge from chaos. These can join up to form of bands or checkerboard patterns or one state may dominate. Bands and checkerboard patterns may have stable boundaries or walking boundaries. Walking boundaries will eventually meet and result in one state dominating a universe which is either empty or with small still life patterns or small oscillators. H1A8 has not been observed forming checkerboards.

In addition to the rotational symmetry found on diagonal bands an orthogonal band can have mirror symmetry. Any seed consisting of just '0' and '3' will create a pattern where both sides of the boundary are reflections of each other with opposite state. These rules can generate symmetrical patterns on an orthogonal boundary from other seeds as well. If the pattern develops mirror symmetry when the universe is an even height the original boundaries will remain anchored if they service Blowoff although the symmetry is not always easy to see through chaotic patterns.

(a) Universe height divisible by four
(b) Odd universe height
(c) Universe height divisible by two but not four
F#34 H0E8 an Orthogonal band with seed "11321" results in Strobing Blowoff

(a) H168 44 × 44
seed "112"
Checkerboard Pattern
(b) H1E8 45 × 45
seed "1120011"
"Orthogonal Band"
(c) H1E8 44 × 44
seed "113"
"Rotated Orthogonal Band"
(d) H0C8 44 × 44
seed "11321"
"Double Band"
(e) H068 44 × 44
seed "11321"
F#35 Various outcomes from an Orthogonal band and Blowoff

Wave Behaviour in Rule H1D8RS       top

Rule H1D8RS is classified as chaos with flecks It always goes to chaos in large universes when initialised with a random patten. It can crystallize in a small one. While in the other rules with Wave Behaviour any island of one state will shrink to Static Debris in H1D8RS a randomly initialised universe remains chaotic. An island of single state will become an island of chaos with some peculiar behaviour described below. Initialised with orthogonal bands the Blowoff goes directly to chaos. However if the orthogonal bands have jagged edges to prevent Blowoff they generate chaos which spreads through the universe at a constant speed.

Initialised with a diagonal band chaos spreads quickly along the seeded boundary forming a third band of chaos which has complex waves on both sizes. In large universes this band expands very slowly until both bands of single state collapse leaving the universe full of chaos. In small universes, particularly if the seed is mirror symmetric, after one single state band has collapsed the band of chaos itself collapses leaving Static Debris in an otherwise empty universe.


(a) Universe 50 × 50 seed "13" Generation 200
(b) Generations to Band Break
(c) Generations to Universe Chaos or Empty
F#36 The time for the universe to become chaotic
(a) Proportion of Chaos results in F#36 (c)
(b) Small Universe Crystallization
F#51 The time for the universe to become chaotic

The proportion of results that end in chaos changes with the size of the universe. This is related to the size of islands of chaos which expand rather than contract. The initial downward section in F#51 (a) is caused by small universe crystallization shown in the end rate analysis of F#51 (b) using the FindEnd10-30Chaos Data Set .

The H1D8RS Random Diamond Island data set analysed whether a diamond shaped random patterns of different sizes expanded or shrunk by 50%. The results in F#37 show that a diamond 125 cells square is the mid point and larger diamonds are most likely to expand and smaller diamonds are more likely to shrink. F#38 and F#39 show examples of a diamond island shrinking and one expanding.

F#37 Analysis of Shrinking or Expanding 50% by Universe Size

(a) Generation 0 Population 3,890
(b) Generation 484 Population 3,059
(c) Generation 968 Population 2,760
(d) Generation 1,452 Population 2,272
(e) Generation 1,936 Population 1,898
F#38 H1D8RS Random Diamond Island size 92 seed "10013" Shrinks by 50&percnt

(a) Generation 0 Population 11,660
(b) Generation 1,104 Population 12,814
(c) Generation 2,207 Population 14,535
(d) Generation 3,311 Population 16,307
(e) Generation 4,415 Population 17,755
F#39 H1D8RS Random Diamond Island size 160 seed "1113" Expands by 50&percnt

F#40 shows H1D8RS with a symmetric seed on a diagonal band in a universe 130 cells square. The symmetric flecks give the appearance of volcanoes. F#40 (d) shows the black state band breaking in two places leaving the chaos band. In this example the chaos band does not expand but shrinks and the band breaks.

(a) Generation 64
(b) Generation 7,188
(c) Generation 27,303
(d) Generation 44,954
(e) Generation 45,010
(f) Generation 90,069
(g) Generation 90,169
(h) Generation 93,432
Period 2 Oscillator
F#40 H1D8RS an Example of a Symmetrical Chaos Band with Complex Waves on Both Boundaries which Eventually Collapses

Edge Behaviour       top

Rules which show Edge Effects are mostly the Edge subcategory of Category Crystallising with H070, H0F0 and H170 from Category Minimal and H010 from Category Chaos. Initialised with a band the seed cause changes to oscillate round the boundary which otherwise remains stable. In addition is H1E8RS from the Blobs subcategory of Category Crystallising which fills the universe with chaos from a band but then crystallizes into blobs with small period oscillators round the boundaries. All the other rules with Edge Behaviour do this when initialised with a random pattern except for H010 which does not crystallize. The oscillators on the boundaries act as 1D CAs with the rules from the Minimal category taking very few generations to arrive at this state.

These rules are divided into two according to the type of oscillations that form on boundaries. The short period rules form simple period two period oscillators on boundaries while the long period rules form complex oscillators. Initialised with a random pattern the long period rules of the Category Crystallising form islands that often do not stop shrinking until either the universe is all one state or contains bands of alternate state sometimes with Static Debris in a similar fashion to the Wave Category rules.

Short Period Rules

In rules with the binary codes X XX11 0000 (H030, H070, H0B0, H0F0, H130, H170, H1B0 and H1F0) an seed on an initial band boundary spreads along the boundary forming a short period oscillator typically period 2. The strobing boundary cells act as a one dimensional CA. Rule 51 in Wolfram's classification Wolfram S. 2002 is one example that is generated.

Initialised with a random pattern these rules quickly form large islands with fixed boundaries lined with strobing cells.

Long Period Rules

The rules with the binary codes X XX01 0000 (H010, H050, H090, H0D0, H110, H150, H190 and H1D0) produce a oscillations that spread along a seeded boundary forming a long period oscillator often too long to measure. These act as a one dimensional CA, rule 150 in Wolfram's classification. Rule 150 is one of the eight additive elementary cellular automata. F#54 some example patterns F#54 (b) and (c) show the period 14 oscillator produced by H010, H090, H110 and H190 and (d) and (e) show the period 12 oscillator produced by H050, H0D0, H150 and H1D0. The changed cells are shown in pink.

(a) 20 × 20 Seed "000000000112"
(b) H010 Generations 12
(c) H010 Generations 12-26
(d) H050 Generations 11
(e) H050 Generations 11-23
F#54 Long Period Edge Oscillators

(a) Long Period Rules with Diagonal Bands
(b) Long Period Rules with Orthogonal Bands
F#56 Long Period Edge Oscillators Period with Universe Size

F#56 shows that the oscillation period is rising sharply with the universe size with a pronounced difference in odd and even universe sizes. T#2 shows the oscillation periods which occurred 10 or more times in the last 3 sizes for which all the seed got results, 60 for each of the 8 rules. It is noticeable that a lot of the oscillation periods are close to 2n.

Band Type Universe
No of
Oscillation Period (Occurrences > 9)
Orthogonal3016 30 (347)56 (23) 126 (12)210 (30) 1,022 (21)32,766 (10)
Orthogonal3515 30 (20)32 (22) 60 (37)62 (15) 112 (7)120 (12) 124 (7)240 (10) 4,095 (327)
Orthogonal4015 24 (346) 60 (13)168 (32) 1,022 (10)2,044 (22) 8,190 (10)58,254 (23)
Diagonal3032 4 (12) 28 (13)30 (173) 31 (103)42 (11) 56 (31)84 (22) 210 (15) 420 (19)1,022 (14)
Diagonal3535 28 (162)30 (23) 32 (13)62 (36) 112 (16) 210 (11) 420 (10) 4,095 (121)
Diagonal4036 24 (259)60 (15) 420 (12)1,022 (12) 1,023 (13)2,044 (34) 12,264 (31)58,254 (14)

T#2 The Oscillation periods which occurred 10 or more times in.

The behaviour of these trapped 1D CA's is described further below.

Initialised with a random pattern these rules quickly form large islands with oscillations along the boundaries which sometimes become fixed or form bands. In rules H050,H0D0,H150 and H1D0 the islands usually shrink to nothing so , rules H090,H110 and H190 the island boundaries usually become fixed. These rules are in the Crystallising Category Rule H010 is an exception it is chaotic with Small Universe Crystallisation.

An example of forming islands whose boundaries become fixed is rule H090 from a random initial patten for width and height of 200 and seed "1232". The oscillation period of this pattern has not yet been determined. It remains an oscillator after 107 generations contains many trapped oscillators. Oscillators of periods 8, 14 and 36 have been seen. The actual oscillation period is 1,718,640 generations so there must also be oscillators with periods 5,11 and 31 or multiples of these.

(a) Generation 0
(b) Generation 10
(c) Generation 20
(d) Generation 30
(e) Generation 107

F#24 H090 from a random initial patten for width and height of 200 and seed "1232" Crystallized as Generation 107 as a period 1,718,640 generation oscillator

An example of forming a band is rule H150 from a random initial patten for width and height of 200 and seed "1232" stopped arbitrarily at generation 4,185,098 oscillation period is greater than 300,000. The general shape of the boundary is maintained but the oscillation along curved section is 3 cells deep but only one cell deep on the straight sections.

(a) Generation 0
(b) Generation 50
(c) Generation 2000
(d) Generation 4,185,098
(e) Generation 5,118,331

F#25 H150 from a random initial patten for width and height of 200 and seed "1232"

Trapped One Dimensional Cellular Automata

Trapped One Dimensional Cellular Automata occur in rules of the Edge category which exhibit long period oscillation. They also occur in maze patterns trapped in straight maze sections. Maize patterns produce two different One Dimensional Cellular Automata, one with each state as the CA and the other providing the background. An example is H08D(H172) shown in F#55. In a universe 24 cells square with a random seed of "112" it has a period of 84 generations after 1,367 generations. It has 6 trapped oscillators with periods of 6, 12, 14 and 28 generations. These act like Rule 178. T#3 lists the rules and the State Symmetric rules that generate them. Seeds that create trapped One Dimensional Cellular Automata on a boundary can be constructed that create sort sections of known patterns so that adding seeds together the result can be predicted.

An orthogonal boundary can support a classical 2 state one dimensional CA. A diagonal boundary has a more complex neighbourhood structure. H150 also supports 60 degree diagonal bands which have even more complex neighbourhood structures on their boundaries see F#49.

1D RULE (in Binary)State SymmetricState Symmetric Rules
72 (0100 1000)NotH178
72 (0100 1000)NotH178
90 (0101 1010)State SymmetricH17A
94 (0101 1110)NotH072, H07A
126 (0111 1110)NotH07E
129 (1000 0001)NotH07E
133 (1000 0101)NotH072, H07A
146 (1001 0010)NotH086, H182
150 (1001 0110)State Symmetric H010, H050, H082, H082, H090, H0D0, H110, H150, H190, H1D0
160 (1010 0000)NotH184, H18C
164 (1010 0100)NotH084, H08C
165 (1010 0101)State SymmetricH172, H17A
178 (1011 0010)Not H070, H0B0, H0F0, H130, H170, H186, H186, H1B0
182 (1011 0110)NotH086, H182
218 (1101 1010)NotH084, H08C
237 (1110 1101)NotH178
250 (1111 1010)NotH184, H18C

T#3 One Dimensional Cellular Automata and the State Symmetric Rules that Generate them

(a) Generation 0
(a) Generation 1
F#49 H150 with a 60 degree Diagonal Band Seed "0" a Period 2 Oscillator

(a) Generation 1,367
(a) Plus Changes
F#55 H08D(H172) A Reverse Strobing Maize Rule with long period trapped oscillators

Small Universe Crystallisation       top

The rules in T#4 all show crystallization in small universes and Crystal Reworking in larger ones. Crystal Reworking is the growing and shrinking of stable areas within a chaotic background. F#42 to F#46 show images of the universe with the changes over a small number of generations hi lighted. Small Universe Crystallisation is distinct from very small universes which do not have a very large number of possible patterns so the length of the sequence leading to repeating of a pattern is shorter. A universe 10 by 10 has been chosen as the minimum having 2100 possible patterns so that any repeating sequence must be a very small fraction of this number to be detected.

An analysis of the time to crystallise with respect to universe size shows that the time to crystallise grows exponentially with the universe size to the point where most patterns will fail to crystallise within the experiment time limit F#47 is an example graph showing this also giving the maximum oscillation period of 5 for the usable section of the data. By way of comparison F#48 shows the average data for all non chaos rules and gives a linear result. The FindEnd10-30Chaos Data Set used 60 seed results for each universe size over the range 10 - 30 cells square. It detects a crystallized universe in the first 1 million generations if it contains oscillators with a combined period of less then 200 generations. T#4 contains the results. The rules H08C and H172RS crystallized to maze patterns with trapped oscillators that had very long periods.

The rules included in the FindEnd10-30Chaos Data Set are those that the FindEnd10-50 Data Set found more than 10 crystallization out of 60 seeds for the universe size 10 x 10

RULEChaos TypeCrystal TypeUniverse Size Range 90% Solved  RULEChaos TypeCrystal TypeUniverse Size Range 90% Solved
H010NetworkEmpty10-30  H0A8NetworkEmpty10-15 Limited by number of generations
H01ERSNetworkDiagonal Bands10-18 Limited by number of generations  H172RSMazeMaze16 Limited by Oscillation period
H022NetworkGel10-15 Limited by number of generations  H174RSMazeMaze10-24 Limited by number of generations
H05ERSNetworkDiagonal Bands10-18 Limited by number of generations  H176RSMazeMaze10-30
H072RSMazeMaze10-28 Limited by number of generations  H188MazeMaze10-30
H074RSMazeMaze10-28 Limited by number of generations  H18AMazeMaze10-25 Limited by number of generations
H076RSMazeMaze10-30  H18CMazeMaze10-30
H088MazeMaze10-30  H1AERSNetworkGel10-27 Limited by number of generations
H08AMazeMaze10-26 Limited by number of generations  H1CERSNetworkGel10-13 Limited by number of generations
H08CMazeMaze10-16 Limited by Oscillation Period  H1D8RSNetworkEmpty2
H08ERSNetworkOrthogonal Bands10-14 Limited by number of generations  H1DCRSNetworkGel10-14 Limited by number of generations
H09ERSNetworkIslands10-25 Limited by number of generations      

T#4 Chaos Rules that Crystallize in Small Universes

Figures F#42 to F#46 below shows example patterns for the different types of chaos and the types of crystals that result. In the chaos phase both states have a connected network throughout the universe which interact with each other in a chaotic fashion. When the chaos phase forms parallel lines of alternate state with right angle bends we describe this as a Maze pattern. Different type of crystal form in different ways:

Rule H051(H1AE) shows the chaos pattern in a larger universe then the crystallization as crystallization only occurs after a very large number of generation in universes where the chaos pattern is interesting. This supports the theses that in Crystal Reworking the crystal size is similar to the size of universes where crystallization is seen.

(a) Generation 500
(b) Changes Generation 500-501
(c) Crystallized at Generation 2906
Period 2 Oscillator
(d) Changes of the Oscillator
F#42 H010 Network → EMPTY, Universe 24 x 24 (Cells that changed state are in red)

(a) Universe 40 x 40
Generation 2000
(b) Changes Generation 2000-2001
(c) Crystallized at Generation 362
Period 4 Oscillator
(d) Changes of the Oscillator
F#43 H051(H1AE) Network → GEL, Universe 20 x 20 (Cells that changed state are in red)

(a) Generation 2000
(b) Changes Generation 2000-2001
(c) Crystallized at Generation 6,163
Period 12 Oscillator
(d) Changes of the Oscillator
F#44 H161(H09E) Network → Islands, Universe 24 x 24 (Cells that changed state are in red)

(a) Generation 500
(b) Changes Generation 500-501
(c) Crystallized at Generation 1011
Period 2 Oscillator
(d) Changes of the Oscillator
F#45 H18D(H072) Maze → Maze, Universe 24 x 24 (Cells that changed state are in red)

(a) Generation 500
(b) Changes Generation 500-501
(c) Crystallized at Generation 3,681
Period 2 Oscillator
(d) Changes of the Oscillator
F#46 H1E1(H01E) Network → Bands, Universe 20 x 20 (Cells that changed state are in red)

F#47 H18A Maze → Maze
A good example of Crystal Reworking. Crystallizing when Chaos is trapped.
F#48 Average Generations to Crystallization in non Chaos Rules.

The rules of the Chaos subcategory Network show shadowy patches that to move about the universe. These are areas of chaos with just enough of a distinctive proportion of each state for the eye to pick up. If one thinks of the areas of different state proportion being different fluids then the equivalent to crystallisation in small universes would be the complete universe being one type of fluid.

Summary       top

The survey found five categories of behaviour, Chaos, Crystallizing, Minimal, Structural and Waves.

The Chaos Category has interesting rules which form islands of crystals or fluids of different proportions of the two states with boundaries that are continually changing. In some rules they touch each and in others they are embedded in chaos. In small universes some will crystallize.

The Crystallising Category rules produce crystals with different forms, Networks, maze patterns and blobs.

The Minimal Category rules do very little.

The Structural Category rules perform with mathematical precision.

The Wave Category rules showed that the step size of the random walk changes according to a × U b where U is the size of the universe and b is close to -0.5.

Boundary Waves in H1D8RS (B34678/S3678) where a surprise and showed that while the two states behave symmetrically a mixture of the two can have an asymmetrical relationship with both. It has the unique property that leads to the tendency for small islands to shrink and large ones expand. The population size for a diamond shaped random filled island of around 7,500 cells gives a 50 50 change of shrinking or expanding.

The emergence of asymmetry is shown in the fact that the One Dimensional Cellular Automata that occur spontaneously are not all themselves state symmetric. It is also shown in the Chaos created by H1D8 is not symmetric with either state.

The motivation of this survey was to find rules that support checkerboard patterns first identified in H1E8 which is B35678/S4678 and has the name Holstein. These patterns show a random walk drag which is the square of the random walk drag of the waves described here. They are initialised as rectangles of alternate state filling a torus universe with seeds on one boundary which do not modify the boundary too close to a corner. They have the following properties:

Rules H0E8, H1A8 and H1E8 support checkerboard patterns. The boundaries in H0E8 stabilize quickly. H1A8 requires large universes to form checkerboards patterns which have deep complex waves on the boundaries in contrast to H1E8 where in larger universes large sections of the boundaries are stable but disturbed by oscillations emanating from the corners.

The properties of all the rules can be found here.

Appendix DataSets       top

FindEnd10-50 Data Set
An End Rate Analysis performed on all rules using 60 non symmetrical seeds with a random initial pattern for square universes sizes 10 to 50 cells in steps of 5 run for up to 10,000 generations or ending with an oscillating pattern with a maximum oscillation period of 100 generations.
FindEnd10-50Edge Data Set
This data set performs the same analysis as the FindEnd10-50 Data Set except that it uses diagonal and orthogonal bands with one seeded edge and cover rules with the Edge Behaviour that generate one dimensional CA's. The oscillation period is very dependent on the length of the any oscillating section of the seeded edge and can be very large. It collects the data for universe size 10 to 50 cells square in steps of 1.
FindEnd10-30Chaos Data Set
This data set performs the same analysis as the FindEnd10-50 Data Set over the size range 10 to 30 in steps of 1.
State Change Data
The State Change Ratio is calculated as the proportion of the 256 different neighbourhood pattern that cause a state change for the a rule compared to the number that the complimentary rule would have caused. This was calculated by a Python program.
Strobe Ratio Data
The Strobe Ratio measurement was made over 240 runs of 200 generations between each chaos CA and it's complementary CA counting the number of state changes. The 240 runs where made up of the combinations of 60 seeds and 4 widths. A strobe ratio of less than 1 indicates normal strobing and a ratio greater than 1 indicates reverse strobing.
State Symmetrical Control Panel for Golly
This ZIP file contains lua modules for the control panel and also a folder of Golly RuleLoader files for the odd complementary rules. Run the script controlPanel.lua contained in the folder. Golly's preferences must be modified to point to this file in order to run the odd rules.
Wave Break
Generation to wave break for universes size 10-100 in steps of 5 for rules with Wave Behaviour. It measures the number of generations taken for a band boundary with complex waves to touch the neighbouring boundary and for a band to start to collapse. This information is used to calculate the average time for a band to collapse for each size of universe. It gives a measure of the drag created by the rule slowing down the random walk of the wave. The seeds used for the analysis where chosen to avoid generating symmetrical waves.
Wave Step Size
Measurements of the change is population over one generation of a diagonal band. Five samples for each of 60 seed at different generations to give 300 independent measurements for each universe size from 10 to 500 cells square. This number divided by the universe size gives the random walk step size.
H1D8RS Random Diamond IslandThe number of generations it takes for an random island in H1D8RS to either grow or shrink. Used to determine the island size for which both outcomes are equally likely. The initial pattern is a diamond shape filled with a seed based random pattern on cells to be 50% of each state. It was run until the population was 50% larger or 50% smaller. SymCACPWave_ChaosIsle_Script.txt

T#5 DataSets

References       top

Day & Night Wikipedia entry
Wikipedia the free encyclopedia.
Drunkard's Walk Wikipedia entry
Wikipedia the free encyclopedia.
Eppstein D. Glider Pattern data base https://www.ics.uci.edu/~eppstein/ca/glider.db  
Eppstein D. Glider Rule Classification https://www.ics.uci.edu/~eppstein/ca/wolfram.html  
Golly's RuleLoader RuleLoader is Golly's system for defining complex rules for two dimensional cellular automata. The Golly Team
LUA The Programing Language LUA. https://www.lua.org/. Lua team at LabLua, a laboratory of the Department of Computer Science of PUC-Rio
Moore Neighbourhood. Wikipedia entry
Wikipedia the free encyclopedia.
Python. The Python is a programming language https://www.python.org/ The Python Software Foundation.
R Core Team (2016). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.
Trevorrow A. and Rokicki T. Golly is an open source, cross-platform application for exploring Conway's Game of Life and other cellular automata. The Golly Team
Wolfram S. Universality and complexity in cellular automata. Physica D: Nonlinear Phenomena, 1984 - Elsevier
Wolfram S. 2002 Wolfram, S. A New Kind of Science. Champaign, IL: Wolfram Media, 2002.

Site by Paul Rendell. Last Update 15/January/2023 Comments to Paul Rendell