Curse of Dimensionality


The Number of Points Needs to Increase Exponentially to Maintain a Given Accuracy.


The requirement in direct-sampling* Monte Carlo simulation that the number of samples per variable must increase exponentially with the number of variables to maintain a given level of accuracy. This is the "Curse of Dimensionality."


Ten observations covers 10% of a \(10 \times 10\) area.
Ten observations covers only 1% of a \(10 \times 10 \times 10\) volume.

* Direct-sampling methods attempt to sample from the entire probability space and thus from the joint probability density of interest indirectly, usually inversely through the marginal cdfs.


Let the complete probability space for one variable be represented by the unit interval \((0, 1)\), and imagine drawing ten samples along that interval. Each sample would then have to represent 10% of the probability space (on average). Now consider a second variable defined on another, orthogonal, \((0, 1)\) interval, also being represented by ten samples. We have produced 10 \((x_1, x_2)\) points on a plane defined by the orthogonal \(x_1, x_2\) lines, and representing a new probability space.

But the new space has \(10 \times 10 = 100\) area units, so each of the ten points now represents only 1% of the probability space. It would require 100 points for each point to represent the same 10% of the probability space that was represented by 10 points in only one dimension. This is illustrated below, where the 10 points are not drawn at random to illustrate their diminished coverage.

Now, although there are 100 \((x_1, x_2)\) points, neither axis has more accuracy than was provided by the previous 10, because there are 10 values of \(x_1\) required for the 10 different values of \(x_2\) necessary to represent the new probability plane. Because in practice the samples are taken at random it appears as though there are more samples, but this isn’t the case in terms of probability space coverage which is related to the density of points, not the number of observations per axis.

Next consider adding another dimension creating a probability space represented by a cube with ten units on a side, and 1000 probability units within. It now requires stacking 10 of the previous \((x_1, x_2)\) planes to construct the new \((x_1, x_2, x_3)\) cube, and of course 10 times the number of samples per axis to maintain the former level of accuracy, or probability coverage. It is obvious that \(10 n\) samples would be required for a \(n\)-dimension problem.

This is referred to in the Bayesian literature as the “curse of dimensionality” and places a practical limit of about a half dozen dimensions on direct-sampling Monte Carlo. Fortunately there are other MC methods that aren’t encumbered by this impediment.