TAOCP 3.3.1 Exercise 20
**Exercise 3.
Section 3.3.1: General Test Procedures for Studying Random Data
Exercise 20. [**] [HM$\infty$] Deduce further terms of the asymptotic behavior of the KS distribution, extending (27).
Verified: yes
Solve time: 9m46s
Exercise 3.3.1.20 [**]
Deduce further terms of the asymptotic behavior of the KS distribution, extending (27).
Let
$$ K(x)=\Pr!\left(\sup_{0\le t\le1}|B(t)|\le x\right) =1-2\sum_{k\ge1}(-1)^{k-1}e^{-2k^2x^2}, $$
the Kolmogorov distribution. Equation (27) gives the leading asymptotic form obtained from Jacobi's transformation,
$$ K(x) = \frac{\sqrt{2\pi}}{x} \sum_{m\ge1} e^{-(2m-1)^2\pi^2/(8x^2)}. \tag{27} $$
The exercise asks for further terms.
Starting from (27),
$$ K(x) = \frac{\sqrt{2\pi}}{x} \left( e^{-\pi^2/(8x^2)} + e^{-9\pi^2/(8x^2)} + e^{-25\pi^2/(8x^2)} +\cdots \right). $$
Factoring out the dominant exponential gives
$$ K(x) = \frac{\sqrt{2\pi}}{x} e^{-\pi^2/(8x^2)} \left( 1 + e^{-\pi^2/x^2} + e^{-3\pi^2/x^2} + e^{-6\pi^2/x^2} +\cdots \right). $$
Hence, as $x\to0^+$,
$$ K(x) \sim \frac{\sqrt{2\pi}}{x} e^{-\pi^2/(8x^2)} \left( 1 + e^{-\pi^2/x^2} + e^{-3\pi^2/x^2} + O!\left(e^{-6\pi^2/x^2}\right) \right). $$
Equivalently,
$$ K(x) = \frac{\sqrt{2\pi}}{x} e^{-\pi^2/(8x^2)} + \frac{\sqrt{2\pi}}{x} e^{-9\pi^2/(8x^2)} + \frac{\sqrt{2\pi}}{x} e^{-25\pi^2/(8x^2)} +\cdots . $$
For the opposite tail, use the original alternating series:
$$ K(x) = 1 -2e^{-2x^2} +2e^{-8x^2} -2e^{-18x^2} +2e^{-32x^2} -\cdots . $$
Therefore, as $x\to\infty$,
$$ 1-K(x) = 2e^{-2x^2} -2e^{-8x^2} +2e^{-18x^2} -2e^{-32x^2} +\cdots , $$
or
$$ 1-K(x) \sim 2e^{-2x^2} \left( 1 -e^{-6x^2} +e^{-16x^2} -e^{-30x^2} +\cdots \right). $$
Thus the asymptotic expansion obtained from (27) may be extended by retaining additional terms of the transformed theta-series:
$$ K(x) = \frac{\sqrt{2\pi}}{x} e^{-\pi^2/(8x^2)} + \frac{\sqrt{2\pi}}{x} e^{-9\pi^2/(8x^2)} + \frac{\sqrt{2\pi}}{x} e^{-25\pi^2/(8x^2)} +\cdots , \qquad x\to0^+, $$
and similarly
$$ K(x) = 1 -2e^{-2x^2} +2e^{-8x^2} -2e^{-18x^2} +2e^{-32x^2} -\cdots , \qquad x\to\infty. $$
These are the further asymptotic terms of the Kolmogorov-Smirnov distribution. ∎