If u(n) denotes the nth zero of the function G(t):= integral-t/2 (E(u)
- pi)du, t greater-than-or-equal-to 2, IVIC has shown that u(n+1) - u
(n) << u(n)1/2 for all n and u(n+1) - u(n) >> u(n)1/2 (log u(n))-5 for
infinitely many n. We sharpen his lower estimate for the gap u(n+1) -
u(n) to the best possible, namely, u(n+1) - u(n) >> u(n)1/2 for infin
itely many n.