Start with how many numbers you need
A point on a line needs one coordinate. A point in a plane needs two independent coordinates. Familiar space needs three. Each added dimension introduces a direction of variation that the existing coordinates cannot fully replace.
Move from one to four dimensions. Sweeping a line in a new direction makes a square; sweeping a square makes a cube. Repeating the construction produces a four-dimensional hypercube, or tesseract.
The screen has not become four-dimensional. The experiment projects the object into 3D and then onto the 2D screen. The apparently different sizes of the inner and outer boxes are effects of projection, not two differently sized rooms in the original object.
Rotate the object, or change your viewpoint
- Turn the viewing angle, then change the four-dimensional rotation. Compare the two changes.
- Slice a sphere from bottom to top. Watch the circle appear, grow and disappear.
- Compare projection with slicing. Which maps many positions, and which keeps only an intersection?
Do high-dimensional data need extra spatial directions?
Further controls and numerical checks
In Explore, drag the 4D projection. This changes the three-dimensional viewpoint. Now adjust rotation in the x–w plane: this rotates the original four-dimensional object before projecting it.
Edges may seem to cross, or the inner box may grow. Projection compresses and overlaps information. Such changes do not necessarily mean a collision or deformation in the original object.
Switch to slicing. As a plane moves through a sphere, the intersection grows from a point into a circle and then shrinks again. An observer confined to that plane would see a circle appear, grow and disappear.
A projection maps information from many locations; a slice keeps only the intersection. The sphere lets you compare those ideas directly.
Feature dimensions
No. A fruit can be described by mass, sugar content and water content. Those numbers form a feature vector. Each component is a dimension without needing a new physical direction.
Changing the plotted feature moves the same fruits relative to one another. Different questions call for different coordinates. These illustrative values are not nutritional measurements.
Features can also be correlated. The number of coordinates is not always the same as the number of independent degrees of freedom. High-dimensional data may lie near a lower-dimensional structure, an idea used by dimensionality reduction.
Count vertices and edges
Each coordinate can be −1 or +1, so n coordinates produce 2ⁿ vertices. Every vertex has n edges, but each edge is counted at both ends. There are therefore n × 2ⁿ⁻¹ edges.
In four dimensions, that means 16 vertices and 32 edges. You can study a precise construction mathematically without being able to picture an extra spatial direction directly.