Abstract
This research extends Yeadon’s 1977 noncommutative $L^1$-maximal ergodic inequality from the semifinite (tracial) setting to the more general $\sigma$-finite (non-tracial) framework. By utilizing Hilsum spatial $L^p$-spaces and Tomita-Takesaki theory, the author provides a crucial building block for establishing a complete noncommutative Dunford-Schwartz inequality for $L^p$ spaces where a traditional trace does not exist, such as those arising in quantum statistical mechanics.
Introduction: The Challenge of Noncommutative Supremums
In classical ergodic theory, the Dunford-Schwartz maximal ergodic inequality is a fundamental tool used to bound the long-term average behavior of dynamical systems. It relies on "maximal functions"—the supremum of a sequence of averages. However, transitioning this theory to the noncommutative world of von Neumann algebras introduces a significant obstacle: defining the "supremum" of a sequence of operators. Unlike functions, operators do not possess a pointwise maximum.
Previous work by Junge and Xu (2007) successfully addressed this for $L^p$ spaces ($1 < p < \infty$) by building upon Yeadon’s 1977 result for the $L^1$ case. However, Yeadon’s proof was restricted to semifinite von Neumann algebras—those with a well-defined trace. Many physically significant algebras, such as Type III factors, are non-tracial. This paper bridges that gap for $L^1$ spaces.
Methodology: Interpolation and Spatial Derivatives
To handle the absence of a trace, the author employs the framework of Hilsum spatial $L^p$-spaces, which are isometrically isomorphic to Haagerup’s $L^p$-spaces. These spaces are constructed using Tomita-Takesaki theory, specifically the "spatial derivative" ($d = \frac{d\phi}{d\psi}$), which serves as a functional substitute for the missing trace.
A key technical strategy involves treating $L^p$ spaces as interpolation spaces between the von Neumann algebra $M$ and its predual $M_*$. This structural relationship is central to the proof's logic.
The proof also relies on variational arguments to identify the necessary "bounding" projections and spectral decomposition to manage unbounded operators. Notably, the author demonstrates that for the $L^1$ case, simple positivity and contractivity of the operator $T$ are sufficient, bypassing the need for the stronger condition of complete positivity.
Key Results: The Non-Tracial $L^1$ Bound
The primary contribution is Theorem 1.3, which establishes a noncommutative extension of the Chebyshev inequality for $\sigma$-finite von Neumann algebras. For a positive linear contraction $T$ and a positive element $a$ in the $L^1$ space, the theorem proves:
- Existence of Projections: For any $\lambda > 0$, there exists a projection $e$ in the algebra.
- Bounded Ergodic Averages: On this projection, the ergodic averages $S_r(a)$ are bounded: $\int e S_r(a) e d\psi \leq 4\lambda$.
- Size Constraint: The "size" of the complement of this projection, measured by a state $\phi$, is inversely proportional to $\lambda$: $\phi(1-e) \leq \frac{2}{\lambda} \int a d\psi$.
This result shows that even without a trace, one can find a "subset" (projection) of the noncommutative space where ergodic averages remain controlled, with the excluded portion being appropriately small.
Future Directions: The Interpolation Hurdle
While this work completes the $L^1$ picture for non-tracial spaces, a major challenge remains: the lack of a noncommutative Marcinkiewicz interpolation theorem for these settings. In tracial algebras, one can interpolate between $L^1$ and $L^\infty$ results to derive bounds for all intermediate $L^p$ spaces. However, the absence of generalized singular numbers for non-tracial algebras prevents standard interpolation arguments from working. Overcoming this technical obstacle is the next frontier in noncommutative ergodic theory.
Summary updated