Maximal functions and weak-type bounds

For locally integrable $f$ on $\mathbb R^d$, define $Mf(x)$ as the supremum over positive radii of the average of $|f|$ on the Euclidean ball centered at $x$. The output may initially be infinite at some points. Taking absolute values before averaging prevents cancellation, and taking a supremum makes the operator sublinear rather than linear. Centered balls are part of this definition; allowing every ball containing the point yields a related uncentered maximal operator with comparable bounds, but different exact values.

A bounded input has maximal function bounded by its essential supremum, because each average has that bound. For integrable inputs a more subtle estimate uses a covering argument: points with large maximal value have witnessing balls with large average, and a suitably selected disjoint collection controls the measure of their union after enlargement. Constants depend on dimension and the geometry of the balls. Simply summing every witnessing ball would overcount without control, which is why the covering step is essential.

For $f\in L^1(\mathbb R^d)$ and every $\lambda>0$, the set where $Mf$ exceeds $\lambda$ has measure at most $C_d\|f\|_1/\lambda$. This is weak type (1,1), also described as a weak-L¹ estimate. It implies that an integrable input has finite maximal function almost everywhere, but it does not imply that the maximal function itself is integrable. A tail proportional to the reciprocal of distance in one dimension already shows the difference.

Combining the weak endpoint estimate with the L∞ averaging bound through an appropriate sublinear interpolation argument gives strong Lᵖ bounds for every exponent strictly greater than one and finite; the infinity bound is separate and direct. Constants need not remain bounded as the exponent approaches one. Do not substitute p=1 into the strong estimate or infer a uniform constant across exponents. Endpoint spaces record information that ordinary strong-norm statements lose, and their exceptional-set formulation is often exactly what an almost-everywhere argument needs.

The Lebesgue differentiation theorem states that for a locally integrable function, averages of its absolute difference from its value tend to zero over shrinking balls at almost every point. Such points are Lebesgue points. One proof approximates by continuous functions, where local recovery is straightforward, then uses maximal weak bounds to control the set where approximation errors can spoil the conclusion. The estimate controls the exceptional set before passing to the approximation limit; norm density alone would not give the needed pointwise statement.

Differentiation remains an almost-everywhere theorem. A single jump can have a local average different from any arbitrarily assigned value, and changing the input on a null set changes no ball integral. The maximal function also takes a supremum over large as well as small balls, so it is not the same as the shrinking-radius limit. In applications separate the operator that measures worst-case averages from the limiting process it helps control. Numerical inspection of finitely many radii can illustrate the idea but cannot replace the supremum or its measure estimate.