Commutator estimates for normal operators in factors with applications to derivations

Abstract

For a normal measurable operator $a$ affiliated with a von Neumann factor $\mathcal{M}$ we show: If $\mathcal{M}$ is infinite, then there is $\lambda_0\in \mathbb{C}$ so that for $\varepsilon>0$ there are $u_{\varepsilon}=u_{\varepsilon}^*$, $v_{\varepsilon}\in \mathcal{U}(\mathcal{M})$ with $$v_\varepsilon|[a,u_\varepsilon]|v_\varepsilon^*\geq(1-\varepsilon)(|a-\lambda_0\textbf{1}|+u_\varepsilon|a-\lambda_0\textbf{1}|u_\varepsilon).$$ If $\mathcal{M}$ is finite, then there is $\lambda_0\in\mathbb{C}$ and $u,v\in\mathcal{U}(\mathcal{M})$ so that $$v|[a,u]|v^*\geq \frac{\sqrt{3}}{2}(|a-\lambda_0\textbf{1}|+u|a-\lambda_0\textbf{1}|u^*).$$ These bounds are optimal for infinite factors, II$_1$-factors and some I$_n$-factors. Furthermore, for finite factors applying $\|\cdot\|_{1}$-norms to the inequality provides estimates on the norm of the inner derivation $\delta_{a}:\mathcal{M}\to L_1(\mathcal{M},\tau)$ associated to $a$. While by [3,Theorem 1.1] it is known for finite factors and self-adjoint $a\in L_1(\mathcal{M},\tau)$ that $\|\delta_{a}\|_{\mathcal{M}\to L_1(\mathcal{M},\tau)} = 2\min_{z\in \mathbb{C}}\|a-z\|_{1}$, we present concrete examples of finite factors $\mathcal{M}$ and normal operators $a\in \mathcal{M}$ for which this fails.