C ̄N− 1
r(θ)
√
1 +(N−^3 1) 2
θ
α
C ̄ 0
C ̄N− 2
1
1
N− 1
Figure 10: The radiusr(θ) along the path from fromC ̄N− 2 toC ̄ 0 has large deriva-
tive,O(N), nearθ= 0.
6 A tiny ball in most directions
A basic principle in probability theory asserts that while anything that might
happen will happen as the system gets large, certain features can become regular,
universal, and non-random [22]. AsNgets large,DN, in most directions, is a ball,
whose radius is
rt≈
r 0
2
(6.1)
Although the radius of the ball is small whenNis large, it is much larger than the
inscribed ball whose radius isr^20. The computation ofrtis a simple application of
random matrix theory [7].
6.1 Application of random matrix theory
Define a random directionθby a vector of iid Gaussian random variables:
θ= (θ 1 ,...,θN (^2) − 1 ), θα∼N
[
0 ,
1
N^2 − 1
]
(6.2)
whereN[μ,σ^2 ] denotes the normal distribution with meanμand varianceσ^2. θ
has mean unit length
E
(
|θ|^2
)
= (N^2 −1)E
(
θ^21