Skip to content

Latest commit

 

History

History
120 lines (102 loc) · 9.34 KB

README.md

File metadata and controls

120 lines (102 loc) · 9.34 KB

4D Rendering Engine

Introduction

This OpenGL program in C++ is capable of rendering edges of the "shadow" of arbitrary 4-dimensional polytopes to the screen. The program utilizes no builtin matrix operations, quaternion transforms, etc. All math that has been implemented was derived by myself with limited outside knowledge of linear algebra or graphics programming, other than the basic concepts of spherical coordinate systems and the parametric equations of lines and planes in n-space. This was written as a personal excercise and proof of concept, rather than to create an efficient and optimized rendering system.

How It Works

In 4-dimensional space, a fourth coordinate axis exists that is orthogonal to the x, y, and z axes. As 3-dimensional beings, it's not really feasible to perceive what this would look and feel like, but since the applicable math works out so nicely when extrapolating up a dimension, a program like this one can give a glimpse into how geometries in higher dimensions would behave. Just about anything regarding 4 dimensions can be understood, if not intuitively then logically, by creating an analogy that applies to 3 or fewer dimensions. Going through this thought process might be helpful in seeing how this program works.

Thought Excercises

Suppose you want to gain a better understanding of how rotation works in 4 dimensions. Let's start by understanding rotation in relation to the number of dimensions. No matter how a 3D object rotates, it rotates about a 1D = (3 - 2)D pivot (a line). In this case, we can confirm the analogy checks out by working out the same math in 2D, in which objects always rotate about a 0D = (2 - 2)D pivot (a point). Now extrapolating up gives insight on 4D rotation—a 4D object must rotate about a (4 - 2)D = 2D pivot (a plane).

To take it a step further, imagine a 2D square lying in a 2D reality. If a 3D being could manipulate that square like a playing card on a table, it could flip it into the 3rd dimension, keeping one of its edges in the 2D plane, and lay it down flat again, with the side originally facing "up" (a dimension and concept that doesn't exist to the 2D viewer) now facing the surface. From the 2d viewer's perspective, it would look like every part of that object spontaneously disappeared except the 1 dimensional edge about which it was rotating, then all at once reappeared on the other side of that face as a mirror image of the original object. Similarly, a 4D being could rotate a 3D cube "out" (for lack of a four-dimensional rotational verb) into the 4th dimension, keeping one of its six 2D faces in our 3D slice of reality, and make it reappear on the other side of that face as its mirror image, without the face ever moving.

(An even weirder side note about this analogy is that, assuming the 2D object was originally bound to the 2d reality and not in fact a cross section of a higher-dimensional object, the one-dimensional "side" of the cube that is left in their plane of existence would necessarily be truly 1D with a thickness of 0. If 2D laws of physics are analogous to those in 3D, a truly one-dimensional object could only exist in a 2D reality through manipulation by a higher-dimensional observer, the same way a two-dimensional object fundamentally cannot exist in 3D. However, a four-dimensional observer lifting a 3D cube off the 3D hyperplane of our reality would produce a truly 2D plane—if atoms are spheres, only the tangent points of those spheres would be touching our reality. Therefore, whether that would look like any physical object at all is questionable.)

Another example is understanding how a 4D being might see. Whereas a 3D human's retina is 2D, and a 2D being could have no more than a 1D retina, a 4D being's retina would be 3D. It could perceive every 2D cross-section of our 3D reality at once, the same way a human could see every 1D cross-section of a 2D reality. If that sounds crazy, consider a 1D being (living on a line) that would probably think it's equally crazy (if a stick is capable of thought) that a 2D being would be capable of seeing its entire reality at once. (This is a critical insight into how this program's Camera4d really works.)

Camera3D

It is not difficult then to understand how this program's Camera4D projects points from 4D space down into 3D space (and subsequently to 2D space to be rendered on the screen) by first understanding how a Camera3D works. Objects of the Camera3D class have a location, which is an (x, y, z) coordinate that represents where the camera is. However, this gives no information about the direction the camera is facing. A spatialVector (a mathematical vector, with a 3D direction and a magnitude) called the normal specifies this direction. The camera's focus is a special point that is always located away from the camera's location in the direction that faces opposite the normal by the camera's focalDistance. The Camera3D class has methods getUnitUpVector() and getUnitRightVector() that return unit spatialVectors pointing up and right relative to the camera. (A side note: the 3D camera cannot roll, as is standard in most video games, so the rightward-facing vector is always orthogonal to the vertical z axis.)

The Camera3D is able to render point3ds—that is, cast a point3d to a point2d based on the location of the point and the camera's location, focus, and normal. In short, Camera3D's projectPoint() method takes a point3D (call it P) and solves for the intersection point I of the line between the focus and P and the 2D plane orthogonal to the camera's normal that passes through its location. This still gives a point in 3-space. To find the (x, y) coordinate that represents the point's location on the camera's rendering plane, analygous to the screen's (x, y) pixel location where the point should be rendered, the vector from the camera's location to I (which is always parallel to the rendering plane) is scalar-projected onto the vectors given by getUnitRightVector() (to give the x coordinate) and getUnitUpVector() (to give the y coordinate).

Since scalar projections give positive or negative values depending on the angle between the vectors, the resulting point2d is treated as being relative to the screen's center as (0, 0) when an edge containing the point is rendered. Additionally, the (x, y) coordinates to render are scaled by ORTHO_ZOOM, a constant in utils.cpp that represent how many screen pixels equal one unit in the Scene's 3D space.

Camera4D

Everything about how the Camera4D works is exactly analogous to Camera3D. It has a location (point4d), focus (point4d), and normal (4D spatialVector). To render a 4D point P to 3D space, it first finds the 4D intersection point I of the line between the focus and P and the 3D hyperplane orthogonal to the normal that passes through the location. It then scalar-projects the 4-dimensional I into 3D space by creating a vector from the location to I and scalar-projecting it onto the vectors given by getUnitRightVector() (to give the x coordinate), getUnitUpVector() (to give the y coordinate), and getUnitOutVector() (to give the z coordinate).

Rendering Different Objects

The main() function, found in graphics.cpp, currently has hard-coded values for the vertices of the x, y, and z axes, as well as those for a 3d cube and 4d cube. To create a new Object, instantiate a new Object3D using a vector<point3d> and vector<edge3d>, or an Object4D using a vector<point4d> and a vector<edge4d>. An Object must have at least one edge to render on the screen, and will not render any points not contained in at least one edge.

Currently, there is no better way than to hard-code vertex locations before instantiating an Object. However, an extrude() method is nearly implemented (found in utils.cpp) that will extrude new vertices in the direction of a given spatialVector and add new edges to the Object appropriately.

For now, to add a custom Object, follow the format as utilized in main().

Note: Datatypes such as edge, point, and spatialVector are structs implemented in utils.cpp.

Movement and Rotation

A Scene, which has a vector of all Objects being rendered, has both a Camera3D and a Camera4D, whose movements are controlled independently. When the program begins, the movement and rotation keybinds control the 3D camera. To view a 3D object from a different perspective, move using the 3D camera. To change how the vertices of a 4D object are cast into 3 dimensions, move using the 4D camera. To toggle which camera is being controlled, press t.

The keyboard controls are set up as follows.

3D Camera

Movement keys are standard WASD, with rotation keys being IJKL.

  • Forward: w
  • Left: a
  • Back: s
  • Right: d
  • Up: space
  • Down: left shift
  • Rotate up: i
  • Rotate left: j
  • Rotate down: k
  • Rotate right: l

4D Camera

Four dimensional movement needs additional keys for moving and rotating into the 4th dimension—moving and rotating "in" and "out". The same keybinds as above apply, with these additional ones:

  • In: q
  • Out: e
  • Rotate in: u
  • Rotate out: o

Copyright © 2020 Jackson Hall. All rights reserved.