Where to stand when playing darts?

TL;DR

This paper investigates the optimal standing distance in darts, analyzing how the distribution of the dart's landing point affects whether standing closer always yields better expected outcomes, with results depending on distribution properties.

Contribution

It characterizes conditions under which standing closer is always optimal, focusing on selfdecomposable distributions and the behavior of characteristic functions.

Findings

Selfdecomposable distributions always favor standing closer.

For cosine payoff, decreasing characteristic function implies closer is better.

Distributions with point masses or compact support can favor standing farther.

Abstract

This paper analyzes the question of where one should stand when playing darts. If one stands at distance $d>0$ and aims at $a\in \mathbb{R}^n$, then the dart (modelled by a random vector $X$ in $\mathbb{R}^n$) hits a random point given by $a+dX$. Next, given a payoff function $f$, one considers $$ \sup_a Ef(a+dX) $$ and asks if this is decreasing in $d$; i.e., whether it is better to stand closer rather than farther from the target. Perhaps surprisingly, this is not always the case and understanding when this does or does not occur is the purpose of this paper. We show that if $X$ has a so-called selfdecomposable distribution, then it is always better to stand closer for any payoff function. This class includes all stable distributions as well as many more. On the other hand, if the payoff function is $\cos(x)$, then it is always better to stand closer if and only if the characteristic function $|\phi_X(t)|$ is decreasing on $[0,\infty)$. We will then show that if there are at least two point masses, then it is not always better to stand closer using $\cos(x)$. If there is a single point mass, one can find a different payoff function to obtain this phenomenon. Another large class of darts $X$ for which there are bounded continuous payoff functions for which it is not always better to stand closer are distributions with compact support. This will be obtained by using the fact that the Fourier transform of such distributions has a zero in the complex plane. This argument will work whenever there is a complex zero of the Fourier transform. Finally, we analyze if the property of it being better to stand closer is closed under convolution and/or limits.