Information thermodynamics of near-equilibrium computation
Information thermodynamics of near-equilibrium computation
复制标题
DOI：
10.1103/physreve.91.062143
复制
发表时间：
2015-06-29
影响因子：
2.4
通讯作者：
Einav, Itai
中科院分区：
文献类型：
--
作者：
Prokopenko, Mikhail;Einav, Itai
在研究基本的物理极限和计算过程的性质时，人们面临着通过定义明确的信息论和热力学来解释原始信息处理函数的挑战。特别是，表征计算传输函数及其可预测性的传递熵在临界区域附近达到峰值。我们专注于传递熵的热力学解释，旨在通过将计算传输所固有的信息流与特定的物理通量联系起来来解释潜在的临界行为。具体地说，在热力学平衡附近的等温系统中，平均传递熵的梯度与Fisher信息和系统的熵曲率动态相关。这种关系明确地将复杂系统固有的计算过程的可预测性、敏感性和不确定性联系在一起，并允许我们考虑几个重要极端情况和权衡的热力学解释。
In studying fundamental physical limits and properties of computational processes, one is faced with the challenges of interpreting primitive information-processing functions through well-defined information-theoretic as well as thermodynamic quantities. In particular, transfer entropy, characterizing the function of computational transmission and its predictability, is known to peak near critical regimes. We focus on a thermodynamic interpretation of transfer entropy aiming to explain the underlying critical behavior by associating information flows intrinsic to computational transmission with particular physical fluxes. Specifically, in isothermal systems near thermodynamic equilibrium, the gradient of the average transfer entropy is shown to be dynamically related to Fisher information and the curvature of system's entropy. This relationship explicitly connects the predictability, sensitivity, and uncertainty of computational processes intrinsic to complex systems and allows us to consider thermodynamic interpretations of several important extreme cases and trade-offs.