Homogeneous Coordinates for 3D Transformations

Introduction

In our last lesson, we saw how to combine individual scaling, rotation, and translation matrices into a single "model" matrix. We established a standard order of operations (Translate * Rotate * Scale) to correctly place objects in a scene.

Throughout that process, we took for granted that we were using 4x4 matrices and 4-component vectors. We briefly mentioned that this was necessary to make translation work with matrix multiplication, but we didn't explore why. This lesson fills that gap.

Today's learning outcome is to use homogeneous coordinates to uniformly represent 3D transformations with 4x4 matrices. We will explore:

• The fundamental limitation of 3x3 matrices that prevents them from handling translation.
• How adding a fourth dimension—introducing homogeneous coordinates—solves this problem mathematically.
• The elegant geometric intuition behind this "4D hack."
• The practical benefit of this system: representing all affine transformations in a single, unified format.

This concept is the bedrock of modern 3D graphics, enabling the efficient and flexible manipulation of objects, cameras, and entire scenes.

The Problem with Translation

As we've seen, scaling and rotation are linear transformations. They can be fully described by a 3x3 matrix. A key property of any linear transformation represented by a matrix is that the origin (0, 0, 0) always maps to itself. No matter what 3x3 matrix you multiply it by, the result is still (0, 0, 0).

Translation, however, is different. It's an affine transformation. Its goal is precisely to move the origin and every other point along with it. A translation is fundamentally an addition operation (new_position = old_position + offset), not a multiplication. You simply cannot represent an addition operation using only a 3x3 matrix multiplication.

The Mathematical Solution: Homogeneous Coordinates

To solve this, mathematicians came up with a clever trick. Instead of representing our 3D points as (x, y, z), we add a fourth component, w, making them (x, y, z, w). These are called homogeneous coordinates.

For representing a 3D position, we set w = 1. So, our 3D point (x, y, z) becomes the 4D vector (x, y, z, 1).

This extra dimension allows us to construct a 4x4 matrix that, through multiplication, effectively performs an addition. Let's see how.

As the video demonstrated, the 4x4 translation matrix looks like this:

When we multiply this by our homogeneous point [x, y, z, 1]ᵀ:

And there it is: we've achieved translation via matrix multiplication. This is the core reason we use 4x4 matrices.

The Geometric Intuition: Translation as a Shear

The math works, but why does it work? There's a beautiful geometric interpretation.
A translation in an N-dimensional space can be visualized as a shear transformation in an (N+1)-dimensional space.

• A 2D translation is a 3D shear.
• A 3D translation is a 4D shear.

This is a non-trivial concept to visualize, but a couple of videos do a great job of building the intuition.

The takeaway from this is that by "lifting" our 3D world into a 4D space (by adding the w=1 coordinate), we turn the non-linear translation operation into a linear shear operation, which can be represented by a matrix multiplication.

The Ultimate Benefit: A Unified System

By moving to 4x4 matrices, we can now represent scaling, rotation, AND translation in the same format. This means we can combine them all into a single matrix, just as we did in the last lesson. We can pre-calculate the final Model = Translate * Rotate * Scale matrix once, and then apply that one matrix to all vertices of an object. This is immensely more efficient than applying three separate operations.

What is 'w' Really?

So far, we've only seen w=1. But what does the w component actually represent, and what happens when it's not 1?

Formally, a point with homogeneous coordinates [x, y, z, w] corresponds to the 3D Cartesian point [x/w, y/w, z/w]. This division by w is often called the perspective divide.

For the affine transformations we have covered (scale, rotate, translate), w starts at 1 and the transformed point also has w=1. So, the division x/w has no effect.

However, this changes when we talk about projection matrices, which are used to create the illusion of perspective (objects further away appear smaller).

As you read, you saw that a projection matrix is constructed in such a way that the w coordinate of a transformed vertex ends up storing a value proportional to its distance from the camera. The final division by w (the perspective divide), which is automatically done by the GPU, is what scales the vertices to create the illusion of depth. We will build this matrix ourselves in a later module.

Conclusion

This lesson unpacked one of the most fundamental "tricks" in 3D graphics. You now have a much deeper understanding of the 4D math that powers every modern game engine.

Key Takeaways:

• 3x3 matrix multiplication cannot represent translation because it's an additive operation, and linear transformations preserve the origin.
• By adding a 4th component (w=1), we move to homogeneous coordinates. This allows us to express translation as a 4x4 matrix multiplication.
• This trick has a geometric interpretation: a 3D translation is equivalent to a 4D shear.
• The primary benefit is uniformity: all affine transformations (Scale, Rotate, Translate) become 4x4 matrices and can be combined into one.
• The w coordinate is crucial for perspective. While it remains 1 for affine transformations, projection matrices manipulate w to encode depth, and the final division by w creates the 3D perspective effect.

In our next and final lesson of this module, we will apply everything we've learned about transformations to construct a view matrix, which defines the position and orientation of the camera in our 3D world.