From the Littlewood-Paley-Stein Inequality to the Burkholder-Gundy Inequality

Let $\{\mathsf{T}_t\}_{t>0}$ be a symmetric diffusion semigroup on a $\sigma$-finite measure space $(\Omega, \mathscr{A}, \mu)$ and $G^{\mathsf{T}}$ the associated Littlewood-Paley $g$-function operator: $$G^{\mathsf{T}}(f)=\Big(\int_0^\infty \left|t\frac{\partial}{\partial t} \mathsf{T}_t(f)\right|^2\frac{\mathrm{d}}{t}\Big)^{\frac12}.$$ The classical Littlewood-Paley-Stein inequality asserts that for any $10}$ of $L_p(\Omega)$. Recently, Xu proved that $ \mathsf{L}^{\mathsf{T}}_{ p}\lesssim p$ as $p\rightarrow\infty$, and raised the problem abut the optimal order of $ \mathsf{L}^{\mathsf{T}}_{ p}$ as $p\rightarrow\infty$. We solve Xu's open problem by showing that this upper estimate of $\mathsf{L}^{\mathsf{T}}_{ p}$ is in fact optimal. Our argument is based on the construction of a special symmetric diffusion semigroup associated to any given martingale such that its square function $G^{\mathsf{T}}(f)$ for any $f\in L_p(\Omega)$ is pointwise comparable with the martingale square function of $f$. Our method also extends to the vector-valued and noncommutative setting.