Curse of Dimensionality: Distance Concentration

Point Cloud

Dimensions d 3

Points N 600

Distribution

Live Statistics

Inside inscribed sphere

—%

Relative contrast

—

Mean pairwise distance (± std)

—

Nearest / farthest distance ratio

—

How it works

Take N points with d independent coordinates, each drawn uniformly from [−1, 1] (a d-dimensional hypercube of side 2). Two things happen as d grows that make high-dimensional data hostile to distance-based methods (k-NN, clustering, kernel methods):

Volume ratio (ball inscribed in cube):
  V_ball(r=1) / V_cube(side=2) = π^(d/2) / (Γ(d/2+1) · 2^d)  →  0 superexponentially

Distance concentration (relative contrast):
  RC = (d_max − d_min) / d_min  →  0 as d → ∞
  because Var(‖X‖) / E[‖X‖]²  ~  1/d  for i.i.d. bounded-variance coordinates

Almost all the cube's volume ends up in its corners, outside the inscribed sphere — the "inside sphere" readout collapses toward 0% within a few dozen dimensions. Meanwhile every pair of points drifts toward the same distance: the nearest and farthest neighbor of any point become almost indistinguishable, which is exactly why "nearest neighbor" stops being a meaningful concept in raw high-dimensional space and why PCA / t-SNE / UMAP exist — they project back down to a handful of dimensions where distance is meaningful again.

Dimensions d — the true dimensionality the statistics are computed in (the 3D view always shows a projection of it).
Points N — sample size for the Monte-Carlo pairwise-distance and inside-sphere estimates.
Uniform cube / Gaussian — switches the per-coordinate distribution; both concentrate, at slightly different rates.
Re-roll random projection — draws a fresh random 3×d orthonormal matrix (a Johnson–Lindenstrauss-style random projection) used only to place the points on screen; it never affects the d-dimensional statistics, which are always computed on the full data.
Point color encodes normalized distance from the centroid in the full d-dimensional space — as d grows, the cloud converges to a nearly uniform color, a direct visual signature of concentration.

Inside inscribed sphere Outside (in the "corners")

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.