A method is introduced for the simultaneous study of the square function and the maximal function of a martingale that can yield sharp norm inequalities between the two. One application is that the expectation of the square function of a martingale is not greater than
3
\sqrt 3
times the expectation of the maximal function. This gives the best constant for one side of the Davis two-sided inequality. The martingale may take its values in any real or complex Hilbert space. The elementary discrete-time case leads quickly to the analogous results for local martingales
M
M
indexed by
[
0
,
∞
)
[0,\infty )
. Some earlier inequalities are also improved and, closely related, the Lévy martingale is embedded in a large family of submartingales.