Search papers, labs, and topics across Lattice.
This paper explores the distinctions between full-state and reduced-moment encodings in equilibrium quantum many-body theory by framing them as different types of encoders mapping admissible states to represented variables. The authors establish that an exact decoder for a specified task exists only when the task remains constant across the encoder fibers, thereby linking representation choices to task performance. Key findings reveal that various mathematical structures, such as variational principles and correlation functions, can be leveraged to optimize the task-relevant content derived from reduced representations, enhancing our understanding of quantum state representations.
An exact decoder for quantum many-body tasks hinges on the constancy of the task across encoder fibers, revealing a critical link between representation and performance.
Equilibrium quantum many-body methods differ not only in approximation, but in which information they represent explicitly. We formulate this distinction by fixing an equilibrium specification and viewing every representation as an encoder from admissible states to represented variables. The identity encoder gives a full-state representation, whereas a non-injective encoder gives a reduced representation whose value labels a fiber of compatible states. For a specified task, an exact decoder exists on a state class if and only if the task is constant on the encoder fibers within that class. Variational principles, reconstruction correspondences, functionals, kernels, and closures are different realizations of additional structure used to select, restrict, or approximate the task-relevant content of a fiber when the retained variable alone is insufficient. Static moments and imaginary-time correlation functions are unified as restrictions of a complete equilibrium readout functional to different probe families. Within the same principle, quantum embedding can be viewed as consistency or replacement between global and local descriptions through reduced interface encoders and their conjugate fields.
Stanford HAI