Cellular Automata (CA) is a family of algorithms featuring a regular grid of discrete entities —- *(cells)* -- that continuously interact with one another. Typically, the cells are organised as a fixed, two dimensional grid. At regular intervals *(time steps)*, each cell calculates its new value *(state)* using a mathematical or logical formula *(transition function)*, while consulting the states of cells around it *(neighbourhood)*. This causes a feedback loop from which complex, higher-order structures may emerge.

There are countless possible transition functions, most of which quickly drive the system towards either uniformity or noise. However, there are also (countless still) functions that result in the emergence of coherent structures that posses interesting, beautiful and potentially even useful patterns of behaviour.

utomata is Javascript/WebGL framework for exploration and study of novel cellular automata algorithms. It is designated for cross-disciplinary research with an emphasis on computational arts and artificial life. It's light weight, dependency free, hardware accelerated and easy to embed in any web page. utomata's minimalist functional syntax can express any algorithm up to outer totalistic, any-neighbour, 4D continuous state CA's.

utomata can work in any HTML page. To get started create a new text file using a code editor such as atom, paste the following snippet and save it as: uto1.html. Then open the file using a web browser. NB: you can also follow this guide using the sandbox, or in utomata.net

```
<!DOCTYPE html>
<html>
<body></body>
<script src="https://soogbet.github.io/utomata/utomata.js"></script>
<script>
// 1. create a system with 1M cells
var uto = new utomata(1024, 1024);
// 2. configure to random binary: 10% white
uto.setup("vec( stp(rand(), 0.1) )");
// 3. run Conway's game of life
uto.run("add(eql(3, V9), mlt(eql(4, V9), V))");
</script>
</html>
```

The run() command expects a transition function to be applied to every cell in the system. The transition can be thought ot as just an operation that returns a value. For example: 2+3 is an operation that returns 4. it consists of two parameters - 2 and 3, and an operator- addition. Note that the transition function is not a javascript function but a string containing instructions in a special syntax that runs on the graphics card. This is what allows utomata to calculate large systems at high frame rates.

`uto.run("vec( 1.0, 0.0, 0.0 )");`

utomata uses four dimensional vectors to describe cell states as RGBA color. The vec() operator can be used to return such a vector. Each component in it conforms to a corresponding color component - red, green and blue. So the above transition function asks for full red, no green and no blue, resulting in a static uniformly red system. For now, we will ignore the alpha channel and just consider RGB color. Note that even though the system appears static - each one of its cells actually chooses that same red color 60 times every second. Below are a few more examples of valid transition functions:

```
// Any RGB color. values above 1.0 or below 0.0 are clamped.
uto.run("vec(0.98, 0.5 , 0.34)");
```

```
// Provideing just one value to vec applies it to all 3 components
uto.run("vec( 1.0 )");
```

```
// Set to a random value at each frame
uto.run("vec( rand() )");
```

utomata uses an internal set of operators. Below are examples of simple transition functions that make use of them:

```
// calculate the gradient distance from the cursor
vec(dst(cell.xy, crsr.xy))
```

```
// create horizontal waves using the sine function and the time-step variable
vec(sin(time*0.05 + cell.x*20)*0.5 + 0.5)
```

```
// explore perlin noise using the mouse
vec( nois(cell.xy*2 + cursor.xy*100) )
```

utomata has a number of built in variables. Consider the following examples:

```
// Use the x and y coordinate of the cell to set its red and green values
vec(cell.x, cell.y, 0.)
```

```
//
vec(crsr.x, crsr.y, 0.)
```

```
// Use the cursor position to move along a perlin noise landscape
uto.run("vec(nois(cell.xy crsr.xy)))");
```

The examples in the previous section made use of simple operations to control the colors of individual cells procedurally. This was done by using static vectors, mathematical operators, built-in variables and a number of functions that are commonly used in procedural content generation, such as sine, random and perlin noise.

However, these approaches are not yet considered CA's because they lack a crucial ingredient - the **neighbourhood**. This section will introduce how to incorporate neighbourhoods in the transition function and will demonstrate how this can lead to emergent, as opposed to procedurally generated patterns.

The built-in variable **V** (Value) holds the cell's current state. That is, if the transition function sets a new value for each cell- **V** keeps its current value for us to use in that same transition. Consider the following:

`frc(add( V , 0.01 ))`

The transition function above takes the current value of each cell and adds 0.01 to all 4 components. The result is then stripped to its fractional part by using the unary operator - frc(), thus creating a looping function.

CA's are often chaotic systems; extremely sensitive to initial conditions. The initial state of a system is called the *configuration* and it plays a vital role in the system's behaviour. You can set the system's configuraion using the setup method of by setting the setup variable. Consider the following:

```
// configure to all black initial state:
uto.setup("vec(0.0)");
```

```
// configure to a random RGB color
uto.setup("rand(1.0, 2.0, 3.0)");
```

```
// configure to a random binary state with 10% white
uto.setup("stp(rand(), 0.1)");
```

```
// reset to a previously defined configuration
uto.setup();
```

```
// set the transition manually using the setup and update variables
setup = vec(rand()); // random greyscale
update = div(V4, 4.0); // blur filter
```

`add(eql(3, V9), mlt(eql(4, V9), V))`

Let us unpack the algorithm into its components:

```
add( // add up the following two operations
eql(3, V9), // return 1 if the Moore Neighbourhood (inclusive) is equal to 3, otherwise 0
mlt( // multiply the following two operations
eql(4, V9), // return 1 if the Moore Neighbourhood (inclusive) is equal to 4, otherwise 0
V // return the value of the cell in question
) // end multiply
) // end add
```

```
// configure to random binary: 10% white
setup = vec( stp(random(), 0.1) );
// use a 10px wide pen
pen.w = 10;
// run Conway's game of life
update = add(eql(3, V9), mlt(eql(4, V9), V));
```

Finally, add a configuration and increase the pen size

```
vec4 T =
stp(1, U(-1, 0)) +
stp(1, U(0, -1)) +
stp(1, U(1, 0)) +
stp(1, U(0, 1));
update =
vec( stp(dst(vec2(0.5), cell.xy), cell.w )) + // center dot
mlt(V, stp(V, 0.99)) + // collapse a full cell
mlt(0.25, T); // add from collapsed neighbour
```

```
// create variables
float diffR = 0.22;
float diffG = 0.05;
float F = 0.036;
float K = 0.061;
float rgg = V.r * V.g * V.g;
vec4 hood = V4 - V * 4.0;
update = add(frc(V), vec( diffR * hood.r - rgg + ( F*(1.0 - V.r)), diffG * hood.g + rgg - ( (K + F)* V.g), 0,0));
```

The constructor creates a new instance of utomata and appends a canvas element of the same size to the body HTML tag.

`var uto = new utomata(1024, 1024);`

The setup method configures the system to its initial state and accepts an optional configuration rule as a string. The configuration can use any operators but cannot access the neighbourhood (V values, U and val functions) because cells have no value at configuration.

```
// configure the system to a random 10% white
uto.setup("vec(stp(0.9, rand()))");
```

The run() method starts calculating the system at up to 60 frames per second. It accepts a transition function (in string format). The transition is then applied to each cell in the system at each time step. If no transition is provided the previously used transition will be used. The default transition is "update = V;"

```
// run a Game of life transition function
uto.run("add(eql(3., V9), mlt(eql(4., V9), V))");
```

The stop method pauses a currently running system. The same system can subsequently be resumed by calling run() with no parameters.

`uto.stop();`

The size method resets the system with the new width and height provided.

```
// resize the system
uto.size(256, 256);
```

The width method resets the system with a new width and current height.

```
// change width to 16 cells
uto.width(16);
```

The height method resets the system with a new height and current width.

```
// change height to 1024
uto.height(1024);
```

The fps() method determines the max framerate of utomata. It accepts any integer value between 0 (stop) and 60. Note that the actual framerate depends greatly on the size and type of the system and on your hardware.

```
// run utomata at 12 steps per second
uto.fps(12);
```

The zoom method changes the size of the canvas without affecting the system itself.

```
// set magnification to 16 pixels per cell
uto.zoom(16);
```

```
// set magnification to 4 cells per pixel
uto.zoom(0.25);
```

utomata has two ways to treat the cells which are at the egde of the system (for example the first row, which has no cells above it). "CLAMP" means that whenever a cell at the edge looks-up a neighbour that isn't there - the result will be 0. "REPEAT" means that the look-up will wrap around the other side of the system so that, for instance, the cell directly ABOVE the one at (0,0) is the cell at (0,1). Note that "REAPEAT" only works for systems whose width and height are powers of two. Calling the edge method will reset the system to its configuration.

`uto.edge("REPEAT");`

utomata accepts an optional static input image. It can be used both in the configuration rule and also in the program itself using the I value and I() function. The input must be a reference to an HTML element.

```
// set input image from the document
uto.input(document.getElementById("myStaticConfig"));
```

```
// stop using an image and set config rule intsead
uto.input(false);
uto.config("vec(0.0)");
```

utomata uses an internal pseudorandom number generator. The seed value can be set for getting deterministic results from the rand() function.

`uto.seed(4.2342562);`

The tile method divides the canvas into independent columns and rows, allowing the use of multiple transition functions side by side. the default is untiled (1,1).

```
// divide the system into 4 separate tiles (two columns and two rows)
uto.tile(2,2);
```

utomata can make use of external variables to allow input from UI elements, external API's, sensors, etc. These are known as uniforms because their value is the same for all cells in the system. Uniforms are entered as key-value pairs, and once entered can be used inside the transition function as float variables. Changing a value for an existing key works the same way as inserting it.

`<input type="range" min="0" max="1.0" step="0.01" value=".5" onchange="uto.setVar('redValue', this.value)" >`

```
// set an initial value
uto.setVar('redValue', 0.5);
// run the system
uto.run("V = vec(redValue,0,0);");
```

Download the canvas as a png file.

`uto.saveImg();`

returns a refernce to the image as a javascript object.

```
var currentState = uto.getImg();
document.getElementById("myImage").src = currentState;
```

Return an array of current errors in utomata.

`console.log(uto.errors());`

Return an object containing the key-value pairs for all uniforms defined on the system.

`console.log( uto.getUniforms() );`

Return a formatted string containing realtime system info

`console.log( uto.getInfo() );`

Return the current number of columns in the system

`console.log( uto.getWidth() );`

Return the current number of rows in the system

`console.log( uto.getHeight() );`

Return the current edge type

`console.log( uto.getEdgeType() );`

Return the current transition function

`console.log( uto.getTransition() );`

Return the current configuration function (only if set via the setup method)

`console.log( uto.getSetup() );`

Return the current normalized position of the cursor on the x axis

`console.log( uto.getCursorX() );`

Return the current normalized position of the cursor on the x axis

`console.log( uto.getCursorY() );`

`V, V1, V2, V3, V4, V5, V6, V7, V8, V9, V24, V25`

Many cellular automata algorithms use a *totalistic neighbourhood*. This means that each cell refers to its adjacent cells only by summing up their state values and using the total in its transition function.

utomata has a number of built-in values for commonly used totalistic neighbourhood types. You can may use any combination of these in the same transition function. The following diagram illustrates all available ones. all of these return a 4D vector representing the sum RGBA values of the corresponding neighbours. NB: the value of the cell itself is also considered as a neighbourhood type.

```
- - - | - + - | - + - | + + + | - + -
- + - | - - - | - + - | - - - | + - +
- - - | - - - | - - - | - - - | - + -
V V1 V2 V3 V4
- + - | + + + | + + + | + + + | + + +
+ + + | - - - | - + - | + - + | + + +
- + - | + + + | + + + | + + + | + + +
V5 V6 V7 V8 V9
+ + + + + | + + + + +
+ + + + + | + + + + +
+ + - + + | + + + + +
+ + + + + | + + + + +
+ + + + + | + + + + +
V24 V25
```

`U(x,y)`

A cell can also access the individual values of other cells using *relative* coordinates. The *U(x,y)* function returns the current state of any cell in the system relative to self, so that: U(0,0) is the same as V, U(0, -1.0) refers to the cell directly above, and U(0, 1.0) is the adjacent neighbour on the right. you can use the U(x,y) function to construct outer-totalistic transition functions or custom totalistic neighbourhoods.

```
// Von-Neumann Neighbourhood (same as V4)
vec4 VN = U(-1,0) + U(0, -1) + U( 1, 0) + U(0,1);
// only diagonal neighbours
vec4 Diag = U(-1., -1.) + U( 1., -1.) + U(-1., 1.) + U( 1., 1.);
```

`get(x,y)`

A cell can also access individual values of other cells using *absolute* coordinates. In this case the x and y coordinates signify cell positions between top-left (0, 0) and bottom-right (1,1).

```
// use the value of the cell pointed by the cursor
add(V, get(crsr.x, crsr.y))
```

`set(x, y)`

Set the value of any cell to vec(1.0). NB: You can multiply set(x,y) by any value to set the cell at x,y to it that value.

```
// set the center cell to a random value
update += set(0.5, 0.5) * rand(1., 2., 3.);
```

`rand(seed)`

Return a random number between 0 and 1. The rand function takes an optional numerical seed value. utomata uses a rudimentary pseudorandom number generator with a unique seed per cell. If multiple random numbers are needed - use a unique seed for each call.

```
// configure to a random RGB vector
setup = rand(1., 2., 3.);
```

`nois(x, y)`

Return the perlin noise z value for a given x and y coordinate.

```
// move along a perlin noise landscape using the cursor
update = vec(noise(cell.xy + crsr.xy*10.));
```

`grid.xy`

The grid variable holds the (non-normalized) width and height of the system, such that grid.x equals the number of columns and grid.y equals the number of rows.

`cell.xyw`

The cell variable holds the normalized x and y coordinates of each cell. cell.z and cell.w hold maximal size of each cell: max(1.0/grid.x, 1.0/grid.y)

`crsr.xyz`

The crsr variable holds the current normalized x and y coordinates of the mouse cursor. crsr.z is 1. when the mouse is pressed and 0. otherwise.

`pcrsr.xy`

The pcrsr variable holds the previous normalized x and y coordinates of the mouse cursor.

`time`

The time variable counts the number of steps since the system was initialized.

```
// create moving bars using sin and stp function
update = vec( stp(sin(cell.x*100.0 + time*0.1), 0));
```

utomata uses a functional programming paradigm. At its core are a set of operators that operate on floating point 4D vectors.

`vec(a, b, c, d)`

Return a 4D vector using parameters. This operator accepts 1,2,3 or 4 float values and always returns a 4D vector. Consider the following scheme:

```
vec(a) >> vec4(a, a, a ,a)
vec(a, b) >> vec4(a, a, a ,b)
vec(a, b, c) >> vec4(a, b, c, 1.0)
vec(a, b, c, d) >> vec4(a, b, c, d)
```

For binary operators, utomata will always return a 4D vector regardless of input. This may sometimes be confusing as the input may be a float, a vec or one float and one vec. Consider the following scheme:

```
op(f1, f2) >> vec(op(f1, f2), op(f1, f2), op(f1, f2), op(f1, f2))
op(V1, V2) >> vec(op(V1.r, V2.r), op(V1.g, V2.g), op(V1.b, V2.b), op(V1.a, V2.a))
op(f, V) >> vec(op(f, V.r), op(f,V.g), op(f,V.b), op(f,V.a))
op(V, f) >> vec(op(V.r, f), op(V.g, f), op(V.b, f), op(V.a, f))
```

`add(a, b)`

Return the sum of **a** and **b**.

`sub(a, b)`

Subtract **b** from **a**

`mlt(a, b)`

Multiply **a** by **b**

`div(a, b)`

Divide **a** by **b**

`pow(a, b)`

Raise **a** to the power of **b**

`mod(a, b)`

Return the remainder of **a** divided by **b**

`stp(a, b)`

Return **vec(1.0)** if **b** is larger than **a**, otherwise return **vec(0)**

`eql(a, b)`

Return **vec(1.0)** if **b** is equal to **a**, otherwise return **vec(0)**

`min(a, b)`

Return the minimal value of **b** and **a**

`max(a, b)`

Return the maximal value of **b** and **a**

`dot(a, b)`

Return the dot product **a** and **b**

`dst(a, b)`

Return the distance between **a** and **b**

`atn(a, b)`

Return the two component arc tangent of **a** and **b**

For Unary operators utomata uses a similar approach to the one above. It appplies the operator on each component of the input and returns the result as a 4D vector. It follows the following scheme:

```
op(f) >> vec(op(f), op(f), op(f) ,op(f))
op(f1, f2) >> vec(op(f1), op(f1), op(f1) ,op(f2))
op(f1, f2, f3) >> vec(op(f1), op(f2), op(f3) , 1.0)
op(f1, f2, f3, f4) >> vec(op(f1), op(f2), op(f3) , op(f4))
op(V) >> vec(op(V.r), op(V.g), op(V.b) ,op(V.a))
```

`rnd(a)`

Rounded values of **a**

`cil(a)`

Round up values of **a**

`flr(a)`

Round down values of **a**

`sqt(a)`

Return the Square root of **a**

`log(a)`

Return the natural logarithm of **a**

`sgn(a)`

Return the sign of **a** (-1. / 0. / 1. )

`frc(a)`

Return the fractional part of **a**

`nrm(a)`

Return the normalized vector of **a**

`sin(a)`

Return the sine function of **a**

`cos(a)`

Return the cosine function of **a**

`tan(a)`

Return the tangent function of **a**

`asn(a)`

Return the arc sine function of **a**

`acs(a)`

Return the arc cosine function of **a**

`atn(a)`

Return the arc tangent cosine function of **a**