in C#, Maths, Tutorial, Unity

Unity 4D #2: Extending Unity to 4D

This article will show how to extend Unity to support four-dimensional geometry. This is the second article in a series of four, and the first one which will probably start discussing the Mathematics and the C# code necessary to store and manipulate 4D objects in Unity.

You can find all the articles in this series here:

A link to download the Unity4D package can be found at the end of this article.


Most of the readers following my blog are familiar with Pikuniku, a whimsical game I worked on in 2019. Not many of you, however, remember all of the teaser trailers that were posted prior to its release. Back in 2017, the official Pikuniku Twitter account posted a short video showing a 4D version of the game.

Many were quick to assume that was just a joke, ignoring what they saw was an actual 4D version of the titual character, Piku, rendered in four dimensions.

Five years later, this tutorial will finally explain how that video was created, and how Unity can be extended from its canonical three dimensions, to support four. In this instalment will focus on implementing the backbones of 4D geometry; the next article will focus on the rendering instead.


There are countless ways in which Unity could be extended to support four-dimensional objects. The solution proposed in this series is to create analogous 4D classes to Unity’s existing ones. For instance, a Mesh4D class will mirror the role of Unity’s Mesh. The table below maps the main components used in this project, and their analogous to the “traditional” Unity 3D.

UnityUnity 4D
MeshFilter(not implemented)
Mesh Mesh4D πŸ†•
A scriptable object that stores the 4D vertices that are making the 4D object, prior to any translation, rotation and scaling. Also includes a list of all edges (Edge) and triangles (Triangle). The data it contains should not be modified.
Transform Transform4D πŸ†•
Stores position, rotation and scale to be applied to the vertices of a Mesh4 object.
MeshRenderer MeshRenderer4D πŸ†•
Renders the 4D mesh as it intersects the 3D space.
MeshWireframeRenderer4D πŸ†•
Renders the 4D mesh as a wireframe, which allows projecting the parts outside of the 3D space.

The familiar MeshFilter component has not been implemented, as Mesh4D objects are linked directly.

On top of the principal components and scriptable objects seen above, this project also requires the introduction of data types that can support 4D calculations. In some cases, Unity already contains classes that can be used. For instance, Unity has its own definition of a Vector4, which already comes with everything needed. Unity also supports 4 by 4 matrices with Matrix4x4. Unfortunately, this class does not have feature parity with its 3D counterpart (Matrix3x3), as it does not implement basic operations such as the matrix product. In this case, extension methods will be used to seamlessly extend its capabilities.

Lastly, there will be some completely new classes that need to be created. For instance: Unity stores the rotations in the Transform class, using Vector3 variables. This is not really possible in 4D, since there are 6 Euler angles in 4D; this requires a new type called Euler4.

UnityUnity 4D
Used for all coordinates in 4D.
Euler4 πŸ†•
Stores the rotation along the 6 rotational axes in 4D.
Edge πŸ†•
Defines an edge between two vertices. Traditional meshes do not explicitly contain this information, but it is very useful for certain representations.
Used to represent rotation matrices, necessary to rotate the objects in 4D.
Matrix4x4Extension πŸ†•
Used to provide basic functionalities to Matrix4x4 class, including: matrix multiplication, component-wise product and division, inner and dot product.

On top of these classes, we will need a few more to physically build the Mesh4D scriptable objects representing hypercubes and hyperspheres, and to arbitrarily “extrude” three-dimensional meshes into four-dimensional ones.

The Mathematics of the Fourth Dimension

A traditional 3D model in Unity is represented by the Mesh class, which contains a list of its vertices along with the triangles that connect them. Together, they form the scaffolding of every 3D object. In four dimension, we will do pretty much the same. The main difference is that vertices will be stored using Vector4s, rather than Vector3s.

In this section, we will see how to represent them via code, and also how to extend the translation, scale and rotation from 3D to 4D. This will allow us to recreate the functionalities offered by the Transform component.


The class that contains the information about the 4D geometry is Mesh4D. Similarly to Unity’s Mesh, it contains a list of vertices; but unlike Mesh, it stores a list of edges, not triangles.

public class Mesh4D : ScriptableObject
    public Vector4[] Vertices;
    public Edge[] Edges;
❓ MonoBehaviour vs ScriptableObject

The Mesh4D class has been defined as a ScriptableObject, rather than a MonoBehaviour. Scriptable objects are assets that can be saved and re-used, and are much more similar to files exported from Maya or Blender.

An edge is a connection between two vertices. It is stored using the Edge struct, which simply contains the indices of the respective vertices in the Vertices array.

public struct Edge
    public int Index0;
    public int Index1;

    public Edge(int index0, int index1)
        Index0 = index0;
        Index1 = index1;

Both 3D and 4D meshes are built out of triangles. The only difference is that in 3D the vertices of those triangles are Vector3, while in 4D they should be Vector4. In this implementation, however, we are not storing triangles. The reason is simple: when visualising a 4D mesh, we need to calculate its intersection with the 3D world. Intersecting four-dimensional triangles with the 3D space is way more complex than intersecting edges.

By storing edges, we are still able to define a 4D mesh, and the overall code to bring it into the 3D world will be much simpler. As a drawback, unfortunately, this technique only works with convex geometries. This is not really an issue, as even many 3D algorithms (such as the ones related to physics and collisions) only work on convex meshes. Ultimately, working with convex meshes is not a limitation as concave ones can be built by composition.


The Mesh4D class works like an actual 3D model. The information contained inside is not supposed to be changed at runtime. Translation, rotation and scaling are applied by the Transform4D component, which serves as a 4D analogous to Unity’s Transform.

To make the class more computationally efficient, the positions of the transformed vertices are stored, alongside the rotation matrix and its inverse (which will be very helpful later on).

public class Transform4D : MonoBehaviour
    public Mesh4D Mesh;
    private Vector4[] Vertices;

    public Vector4 Position;
    public Euler4 Rotation;
    public Vector4 Scale = new Vector4(1,1,1,1);

    private Matrix4x4 RotationMatrix;
    private Matrix4x4 RotationInverse;

Both Position, Rotation and Scale have to account for the fact that four dimensions are now available. This means using Vector4 for Position and Scale, and a hypothetical Vector6 or Rotation. In fact, while there are 3 rotation axes in 3D, there are 6 rotation planes in 4D; Unity does not contain a Vector6 struct, so a custom type has to be created. For the occasion, it is called Euler4, as it represents Euler angles in 4D:

public struct Euler4
    [Range(-180, +180)]
    public float XY; // Z (W)
    [Range(-180, +180)]
    public float YZ; // X (w)
    [Range(-180, +180)]
    public float XZ; // Y (W)
    [Range(-180, +180)]
    public float XW; // Y Z
    [Range(-180, +180)]
    public float YW; // X Z
    [Range(-180, +180)]
    public float ZW; // X Y

Understanding how rotations work in 4D is fairly complex, so a later section will expand on the topic, and clarify why 4D dimensions have 6 rotation planes, and not just 4 rotation axes.

The responsibility of the Transform4D component is to update the vertices based on the desired position, rotation and scale. To do so, the component calculates the current rotation matrix, and updates the vertices using the Transform method that effectively maps a Vector4 point from object space to world space.

private void Update()

private void UpdateVertices ()
    for (int i = 0; i < Mesh.Vertices.Length; i++)
        Vertices[i] = Transform(Mesh.Vertices[i]);

At this point in the article is worth reminding that both rotation and scaling are typically performed through the same mechanism: matrix multiplication. A 3D point can be rotated and scaled using a 3×3 matrix; likewise, the same can be obtained in 4D using a 4×4 matrix. Translation, unfortunately, cannot be done like this. If you are familiar with how 3D graphic works, you might have heard of affine transformations and homogenous coordinates. In 3D, this means representing coordinates as \left[x,y,z,1\right], and using 4×4 matrices. By using this “trick”, it is possible to combine translation, rotation and scaling into a single matrix (sometimes referred to as TRS matrix).

Affine transformations work in 4D as well, and we could technically encode a 4D vertice in a 5D vector \left[x,y,z,w,1\right], performing all operations using 5×5 TRS matrices. However, matrix operations become progressively more expensive as the dimensions increase. For this reason, the solution proposed in this article to transform 4D points is to rely on 4×4 matrices for rotations only, and to perform translation and scaling separately. Both translations and scaling in 4D can be performed by component-wise addition and product, respectively.

// Takes a 4D point and translate, rotate and scale it
// according to this transform
public Vector4 Transform (Vector4 v)
    // Rotates around zero
    v = RotationMatrix.Multiply(v);
    // Scales around zero
    v.x *= Scale.x;
    v.y *= Scale.y;
    v.z *= Scale.z;
    v.w *= Scale.w;

    // Translates
    v += Position;

    return v;

What is now missing is to understand how to create the rotation matrix.

❓ Rows or columns?

Depending on the source, you may find the positions encoded as either column or row vectors. Generally speaking, this is nothing more than a convention, and you can derive working equations and code from both variants.

However, it is important to notice that this will have an impact on the final equations. Matrix multiplication is, in fact, not commutative. This means that the order in which elements are multiplied matters, and generally speaking changing it will produce an incorrect result.

There are also very strict rules when it comes to matrix multiplication.

  • If positions are stored as column vectors, the correct order to rotate them is R \times p;
  • If positions are stored as row vectors, the correct order to rotate them is p \times R.

In the context of game development, column vectors are more commonly used. This aligns with the convention in linear algebra of having matrices on the left of a multiplication. On top of that, Graphics APIs such as OpenGL and DirectX typically expect column vectors for position data.


Understanding rotations in 2D and 3D comes naturally to us, since we have evolved to manipulate complex objects in space. However, anyone who has studied the mathematics behind rotations can verify how messy it gets. What is geometrically intuitive for us, becomes impossibly counterintuitive when we start formalising it mathematically. It does not help that there are several different ways to model both orientations and rotations. Unity supports three of them: Euler angles, rotation matrices and quaternions. The last ones are used internally by the engine. Despite their popularity, quaternions are deemed among the most technically challenging subjects in geometry. So much so that in the past they have even been labelled as “evil” by Lord Kelvin:

Β«Quaternions came from Hamilton after his really good work had been done; and, though beautifully ingenious, have been an unmixed evil to those who have touched them in any way, including Clerk Maxwell.Β»

Lord Kelvin, 1892.

πŸ”„ Quaternions in 4D

Quaternions are the de-facto standard to represent and manipulate rotations in modern game engines. There are many reasons for their adoption, like numerical stability and compact form. A 3×3 matrix has 9 dimensions, while quaternions only need 4 to represent the same information. On top of that, quaternions are not affected by the infamous gimbal lock, in the same way that Euler angles can.

While quaternions only work in 3D, the concept of rotation can be extended in any dimension. The mathematical object that can represent a rotation in an arbitrary dimension is called a rotor. A 4D rotor is akin to a quaternion for 4D objects, and requires eight dimensions. Marc ten Bosch, the developer of “Miegakure” and “4D Toys” wrote a short article explaining rotors in 4D which might be a good read to get you started on the topic: 4D Rotations and the 4D equivalent of Quaternions.

It is also worth noticing that quaternion-like systems can be derived for all dimensions that are powers of 2, through a process known as the Cayley–Dickinson construction.

DimensionLatin nameMicheal Carter
1Real numbers
2Complex numbers

While the terms quaternions, octonions, sedenions and trigintaduonions are pretty much established, the naming conventions for dimensions above the 32th are not-so-clear and differ between different sources.

Some libraries, like hypercomplex in Python, have adopted the names suggested by Micheal Carter in the paper titled Visualization of the Cayley-Dickson Hypercomplex Numbers Up to the Chingons (64D).

In this article, we will expose the orientation of a 4D mesh using Euler angles, which is Unity’s method of choice to display them in the inspector. The “Rotation” field of the Transform component in every game object is, in fact, displaying Euler angles. Euler angles are a way to visualise the orientation of an object by decomposing it as three successive rotations around different axes. In Unity, these rotations are performed around the Z axis, the X axis, and the Y axis. The order in which these are performed is important, as rotations are not commutative: doing them in a different order might result in a different final orientation.

One common misconception that needs to be clarified is that there are three rotation axes in 3D because there are 3 dimensions: this is not correct. In fact, there are 6 rotation planes in 4D, not 4. The root of this misconception is that in 3D there are as many rotation axes as dimensions; but that is a coincidence, and does not occur in other dimensions. For instance, there is only one rotation axis in 2D, not 2.

As explained by Steven Richard Hollasch in “Four-Space Visualization of 4D Objects“, rotations […] are more properly thought of not as rotations about an axis, but as rotations parallel to a 2D plane. There is only one rotation axis in 2D, because there is only one 2D plane. Such rotation can be defined by the plane in which it takes place (XY) or by the normal to that plane (Z axis). Incidentally, all points on the rotation axis are unchanged. Another way to see this is to imagine the normal as a handle that rotates the plane it is attached to.

There are three rotation axes in 3D, because there are three 2D planes: XY, YZ and XZ, which normals correspond to Z, X and Y axes.

Likewise, there are six rotation axes in 4D, because there are six 2D planes: XY, YZ, XZ, XW, YW and ZW. While 2D and 3D rotations leave the points on their rotation axes unchanged, in 4D there is an entire plane of points unaffected by the rotation.

Generally speaking, in an n-dimensional space, there are exactly {n \choose 2} rotation axes, which correspond to the number of unique 2D planes available (without counting repetitions, as XY and XY are the same plane).

XY plane (Z axis fixed)XY plane (Z axis fixed)XY plane (ZW plane fixed)
YZ plane (X axis fixed)YZ plane (XW plane fixed)
XZ plane (Y axis fixed)XZ plane (YW plane fixed)
XW plane (YZ plane fixed)
YW plane (XZ plane fixed)
ZW plane (XY plane fixed)

πŸ”„ A more detailed explanation

The section above explained that the number of rotation axes in a n-dimensional space, is given by the number of 2D planes available. That turns out to be {n \choose 2} (to be read as n choose 2), which is the number of ways in which 2 elements can be chosen from a collection of n.

However, such a statement was presented as a fact. There is a more detailed (and hopefully clearer) procedure that helps identify the number of rotation axes in a given space.

Let’s start with 3D. In a three-dimensional space, there are three axes: X, Y and Z. An arbitrary orientation can be taught as the combination of a rotation on these axes. So what are the possible combinations?

  • Rotations around one axis:
    • X
    • Y
    • Z
  • Rotations around a plane (rotation on one axis, followed by a rotation on another axis):
    • XY (or YX)
    • YZ (or ZY)
    • XY (or YX)
  • Rotations around the space:
    • XYZ (or XZY, YXZ, YZX, ZXY, ZYZ)

Here we are not counting the permutations (i.e.: both XY and YX are counted as one). This is because we are trying to count the unique types of rotations, and while XY and YX are different, they take place on the same plane.

Overall, we have 7 possible combinations. However, some of them are equivalent to each other (i.e.: they identify the same space), collapsing the number to just three.

The same reasoning can be used to calculate the total combination of axes permutations in 4D: that is 15, but once symmetries are removed, the total number goes down to 6.

Now that we understand that there are 6 rotation planes in 4D, and that rotation can be performed using matrix multiplication, the next step is to define them. In this article, we will not derive them as this is outside the scope. However, If you are interested I suggest reading the following articles which provide a detailed explanation of how rotation matrices are derived:

The proposed solution for this problem is to have a static function that can produce the rotation matrix for each separate rotation plane. For instance, RotateXY(Mathf.PI/2f) will return the rotation matrix that performs a 90Β° rotation around the XY plane. Once we have that, we can chain all rotations by multiplying together their respective rotation matrices:

private Matrix4x4 UpdateRotationMatrix()
    RotationMatrix =
        .RotateXY(Rotation.XY * Mathf.Deg2Rad)
        .RotateYZ(Rotation.YZ * Mathf.Deg2Rad)
        .RotateXZ(Rotation.XZ * Mathf.Deg2Rad)
        .RotateXW(Rotation.XW * Mathf.Deg2Rad)
        .RotateYW(Rotation.YW * Mathf.Deg2Rad)
        .RotateZW(Rotation.ZW * Mathf.Deg2Rad);
    RotationMatrixInverse = RotationMatrix.inverse;

    return RotationMatrix;

In the function above, we use Mathf.Deg2Rad since Euler angles are expressed in degrees, while the various Rotate-- functions take radians as input.

Below, you can find the definition for all the various rotation matrices.

πŸ”„ Rotations in 4D

Rotate XY (Z axis)

(1)   \begin{equation*}  R_{XY}\left(\alpha\right)=\begin{bmatrix} \cos{\alpha} & \sin{\alpha} & 0 & 0 \\ -\sin{\alpha} & \cos{\alpha} & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{bmatrix} \end{equation*}

public static Matrix4x4 RotateXY(float a)
    float c = Mathf.Cos(a);
    float s = Mathf.Sin(a);
    Matrix4x4 m = new Matrix4x4();
    m.SetColumn(0, new Vector4(c, -s, 0, 0));
    m.SetColumn(1, new Vector4(s, c, 0, 0));
    m.SetColumn(2, new Vector4(0, 0, 1, 0));
    m.SetColumn(3, new Vector4(0, 0, 0, 1));
    return m;

public static Matrix4x4 RotateXY(this Matrix4x4 m, float a)
    return m * RotateXY(a);

Rotate YZ (X axis)

(2)   \begin{equation*}  R_{YZ}\left(\alpha\right)=\begin{bmatrix} 1 & 0 & 0 & 0\\ 0 & \cos{\alpha} & \sin{\alpha} & 0\\ 0 & -\sin{\alpha} & \cos{\alpha} & 0 \\ 0 & 0 & 0 & 1 \end{bmatrix} \end{equation*}

public static Matrix4x4 RotateYZ(float a)
    float c = Mathf.Cos(a);
    float s = Mathf.Sin(a);
    Matrix4x4 m = new Matrix4x4();
    m.SetColumn(0, new Vector4(1, 0, 0, 0));
    m.SetColumn(1, new Vector4(0, c, -s, 0));
    m.SetColumn(2, new Vector4(0, s, c, 0));
    m.SetColumn(3, new Vector4(0, 0, 0, 1));
    return m;
public static Matrix4x4 RotateYZ(this Matrix4x4 m, float a)
    return m * RotateYZ(a);

Rotate XZ (Y axis)

(3)   \begin{equation*}  R_{XZ}\left(\alpha\right)=\begin{bmatrix} \cos{\alpha} & 0 & -\sin{\alpha} & 0\\ 0 & 1 & 0 & 0\\ \sin{\alpha} & 0 & \cos{\alpha} & 0 \\ 0 & 0 & 0 & 1 \end{bmatrix} \end{equation*}

public static Matrix4x4 RotateXZ(float a)
    float c = Mathf.Cos(a);
    float s = Mathf.Sin(a);
    Matrix4x4 m = new Matrix4x4();
    m.SetColumn(0, new Vector4(c, 0, s, 0));
    m.SetColumn(1, new Vector4(0, 1, 0, 0));
    m.SetColumn(2, new Vector4(-s, 0, c, 0));
    m.SetColumn(3, new Vector4(0, 0, 0, 1));
    return m;
public static Matrix4x4 RotateXZ(this Matrix4x4 m, float a)
    return m * RotateXZ(a);

Rotate XW

(4)   \begin{equation*}  R_{XW}\left(\alpha\right)=\begin{bmatrix} \cos{\alpha} & 0 & 0 & \sin{\alpha} \\ 0 & 1 & 0 & 0\\ 0 & 0 & 1 & 0 \\ -\sin{\alpha} & 0 & 0 & \cos{\alpha} \end{bmatrix} \end{equation*}

public static Matrix4x4 RotateXW(float a)
    float c = Mathf.Cos(a);
    float s = Mathf.Sin(a);
    Matrix4x4 m = new Matrix4x4();
    m.SetColumn(0, new Vector4(c, 0, 0, -s));
    m.SetColumn(1, new Vector4(0, 1, 0, 0));
    m.SetColumn(2, new Vector4(0, 0, 1, 0));
    m.SetColumn(3, new Vector4(s, 0, 0, c));
    return m;
public static Matrix4x4 RotateXW(this Matrix4x4 m, float a)
    return m * RotateXW(a);

Rotation YW

(5)   \begin{equation*}  R_{YW}\left(\alpha\right)=\begin{bmatrix} 1 & 0 & 0 & 0 \\ 0 & \cos{\alpha} & 0 & -\sin{\alpha} \\ 0 & 0 & 1 & 0 \\ 0 & \sin{\alpha} & 0 & \cos{\alpha} \end{bmatrix} \end{equation*}

public static Matrix4x4 RotateYW(float a)
    float c = Mathf.Cos(a);
    float s = Mathf.Sin(a);
    Matrix4x4 m = new Matrix4x4();
    m.SetColumn(0, new Vector4(1, 0, 0, 0));
    m.SetColumn(1, new Vector4(0, c, 0, s));
    m.SetColumn(2, new Vector4(0, 0, 1, 0));
    m.SetColumn(3, new Vector4(0, -s, 0, c));
    return m;
public static Matrix4x4 RotateYW(this Matrix4x4 m, float a)
    return m * RotateYW(a);

Rotate ZW

(6)   \begin{equation*}  R_{ZW}\left(\alpha\right)=\begin{bmatrix} 1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & \cos{\alpha} & -\sin{\alpha} \\ 0 & 0 & \sin{\alpha} & \cos{\alpha} \end{bmatrix} \end{equation*}

public static Matrix4x4 RotateZW(float a)
    float c = Mathf.Cos(a);
    float s = Mathf.Sin(a);
    Matrix4x4 m = new Matrix4x4();
    m.SetColumn(0, new Vector4(1, 0, 0, 0));
    m.SetColumn(1, new Vector4(0, 1, 0, 0));
    m.SetColumn(2, new Vector4(0, 0, c, s));
    m.SetColumn(3, new Vector4(0, 0, -s, c));
    return m;
public static Matrix4x4 RotateZW(this Matrix4x4 m, float a)
    return m * RotateZW(a);

What’s Next…

This article explained in details the mathematics of four-dimensional objects, as a direct extension of the more traditional Euclidean geometry. We also created a new set of classes capable of storing and manipulating 4D meshes, in a way that is not dissimilar to how Unity stores and manipulates conventional 3D meshes.

The next instalment in this series will explore three different techniques to render 4D meshes.

You can read the remaining articles in the series here:

Additional Resources

If you are interested in learning more about the fourth dimension and the hidden beauty of the objects it contains, I would suggest having a look at the following articles and books:

πŸ“¦ Download Unity4D Package

All of the diagrams and animations seen in this tutorial have been made with Unity4D, the unity package that extends support for 4D meshs in Unity.

The Unity4D package contains everything needed to replicate the visual seen in this tutorial, including the shader code, the C# scripts, the 4D meshes, and the scenes used for the diagrams and animations. It is available through Patreon.

πŸ’– Support this blog

This website exists thanks to the contribution of patrons on Patreon. If you think these posts have either helped or inspired you, please consider supporting this blog.

Patreon Patreon_button

πŸ“§ Stay updated

You will be notified when a new tutorial is released!

πŸ“ Licensing

You are free to use, adapt and build upon this tutorial for your own projects (even commercially) as long as you credit me.

You are not allowed to redistribute the content of this tutorial on other platforms, especially the parts that are only available on Patreon.

If the knowledge you have gained had a significant impact on your project, a mention in the credit would be very appreciated. β€οΈπŸ§”πŸ»

Write a Comment



  • Unity 4D #4: Creating 4D Objects - Alan Zucconi

    […] Part 2: Extending Unity from 3D to 4D […]

  • Unity 4D #3: Rendering 4D Objects - Alan Zucconi

    […] Part 2: Extending Unity from 3D to 4D […]

  • η»ŸδΈ€4DοΌšε°†Unityζ‰©ε±•εˆ°η¬¬ε››η»΄ - εζ‰§ηš„η ε†œ

    […] 详情参考 […]

  • Unity 4D #1: Understanding the Fourth Dimension - Alan Zucconi

    […] Part 2: Extending Unity from 3D to 4D […]