On the discrepancy function in arbitrary dimension, close to L 1

Abstract

Let A N to be N points in the unit cube in dimension d, and consider the discrepancy function $$ D_N (\vec x): = \sharp \left( {\mathcal{A}_N \cap \left[ {\vec 0,\vec x} \right)} \right) - N\left| {\left[ {\vec 0,\vec x} \right)} \right| $$ Here, $$ \vec x = \left( {\vec x,...,x_d } \right),\left[ {0,\vec x} \right) = \prod\limits_{t = 1}^d {\left[ {0,x_t } \right),} $$ and $$ \left| {\left[ {0,\vec x} \right)} \right| $$ denotes the Lebesgue measure of the rectangle. We show that necessarily $$ \left\| {D_N } \right\|_{L^1 (log L)^{(d - 2)/2} } \gtrsim \left( {log N} \right)^{\left( {d - 1} \right)/2} . $$ In dimension d = 2, the ‘log L’ term has power zero, which corresponds to a Theorem due to [11]. The power on log L in dimension d ≥ 3 appears to be new, and supports a well-known conjecture on the L 1 norm of D N . Comments on the discrepancy function in Hardy space also support the conjecture.