Technical Summary: "When Does Wave Memory Compute?"

Problem Statement

The paper addresses a fundamental ambiguity in unconventional and neuromorphic computing: the distinction between physical memory and autonomous physical computation. While physical systems like walking droplets, reservoir computers, and cortical waves often exhibit memory, feedback, and rich state-dependent dynamics, these properties alone do not constitute computation. The central question is: When does a wave-memory system transition from being a dynamical system with recoverable past traces to an autonomous physical computer that internally selects its own operations?

The authors argue that current frameworks often conflate observer-assigned computational interpretations with intrinsic physical computation. Specifically, they investigate the "wave–particle walker" (a droplet bouncing on a vertically vibrated fluid bath), which demonstrates writing, storage, reading, and erasing of information via a self-generated wave field. The paper seeks to determine if this system, as experimentally realized, qualifies as an autonomous computer or merely a memory-mediated dynamical system.

Methodology

The authors develop a rigorous, multi-level framework to classify physical systems, moving from dynamical systems theory to category theory.

1. Physical Modeling (Stroboscopic Reservoir):
   The walking droplet is modeled as a stroboscopic reservoir system with a discrete-time state $x_n = (r_n, v_n, \sigma_n, H_n)$.

   • $r_n, v_n$: Droplet position and velocity.
   • $\sigma_n$: Bouncing phase.
   • $H_n$: Wave-memory field, which stores an exponentially decaying trace of previous impacts ($H_{n+1} = \lambda H_n + \sigma_n \psi_{r_n}$).
     The model explicitly separates writing (impact), storage (decay), reading (local slope coupling $\nabla H_n(r_n)$), and feedback.

2. Three-Tiered Classification Hierarchy:
   The paper establishes a hierarchy of physical information processing:

   • Level 1: Physical Memory: The system state encodes information about past states (e.g., the wave field $H_n$).
   • Level 2: Transition-Preserving Computation: The system implements an abstract transition rule $G: A \to A$ via a coarse-graining map $\Pi: X \to A$ such that $\Pi \circ F = G \circ \Pi$. This requires that physical dynamics preserve the structure of abstract transitions under noise and finite resolution (robust basin separation).
   • Level 3: Closed Autonomous Physical Computation: The system must contain an internal physical readout state $y_n$ that causally selects the next physical operation. The evolution takes the form $x_{n+1} = F_{C(y_n)}(x_n)$, where $C$ maps the internal readout to an operation.

3. Categorical Formulation:
   To formalize these concepts, the authors employ category theory.

   • PhysProc: A category of physical processes (regions and morphisms).
   • AbsProc: A category of abstract processes.
   • Computation: Defined as a quotient functor $Q_\Pi: \text{PhysProc}_\Pi \to \text{AbsProc}$ that respects coarse-graining.
   • Autonomy (Closure): Defined as the internalization of morphism selection. In standard functorial formulations, the operation is external; in the closed case, the morphism is selected by the system's own state ($o = C(p_Y(x))$), making the diagram self-contained.

Key Contributions

1. The Closure Criterion for Autonomy:
   The primary theoretical contribution is the definition of closure as the necessary condition for autonomous physical computation. The paper posits that memory and feedback are insufficient; a system is only an autonomous computer if a physical readout variable selects the subsequent operation. This distinguishes a system that "computes" from one that is merely "controlled" or "interpreted" by an external agent.

2. Robust Coarse-Grained Transition Preservation:
   The paper refines the definition of physical computation by rejecting arbitrary observer labeling. It requires that abstract states correspond to separated physical basins and that transitions between them are stable under noise (probabilistic basin stability). This replaces the requirement for a one-to-one microstate-to-symbol mapping with a physically meaningful notion of robust symbolization.

3. Categorical Formalization of Autonomy:
   The authors provide a categorical proof that autonomy corresponds to the internalization of the morphism selection process. By defining the closed-loop map $\Phi = \text{Act} \circ \langle C \circ p_Y, \text{id}_X \rangle$, they demonstrate that the choice of the next physical process is generated by the system's internal state rather than supplied externally.

4. Application to the Walking Droplet:
   The framework is applied to the Perrard–Fort–Couder walker experiment. The analysis confirms the walker possesses:

   • Physical memory (wave field).
   • Writing, storage, and local reading mechanisms.
   • Externally triggered erasure (via a $\pi$-phase shift).
   • Finite-time trajectory reversal.
     However, the analysis concludes the walker fails the closure criterion because the $\pi$-shift (the operation triggering erasure/reversal) is imposed externally. There is no internal physical readout state $y_n$ that selects the phase shift based on the wave memory.

Results

• Classification of the Walker: The walking droplet is classified as a "wave-memory physical machine with Turing-like primitives" but not as a "closed autonomous physical computer" (or Turing machine). It possesses the physical primitives of a computer (write, read, erase, feedback) but lacks the architectural layer of internal operation selection.
• Distinction Validated: The paper successfully demonstrates that a system can exhibit complex, memory-mediated reversal and erasure without being an autonomous computer. The missing link is the internalization of the control signal.
• Constructive Extensions: The paper outlines specific physical architectures that would satisfy the closure criterion, transforming the walker into a closed computer. These include:
  • Multistable readout traps: A bistable potential driven by wave slope to trigger phase flips.
  • Thresholded wave detectors: Hysteretic detectors that switch the forcing phase when wave energy crosses a threshold.
  • Coupled walkers: Using a second walker as a physical controller for the first.
  • Boundary-controlled reservoirs: Where readout states modulate bath parameters (memory depth, confinement).

Significance and Claims

The paper claims to provide a sharpening, not a rejection, of the wave-based computing interpretation. It does not deny that the walker performs remarkable physical information processing; rather, it precisely locates the system within a hierarchy of physical computation.

• Theoretical Precision: The work resolves the "pancomputationalist" worry (that any physical system computes anything an observer maps onto it) by demanding transition preservation and internal operation selection. It asserts that memory is not computation until it is organized into robust, transition-preserving basins and controlled by internal readouts.
• Design Principle: The paper reframes the challenge of unconventional computing as a design problem: how to couple wave memory to intrinsic readout basins that select operations. This moves the field from metaphorical descriptions of "wave computers" to constructive architectural requirements.
• Broader Applicability: The closure criterion is presented as a general tool for evaluating neuromorphic hardware, reservoir computing, and biological systems (e.g., cortical traveling waves). It asks whether the readout in these systems is merely external (measured by an observer) or physically internalized (causally affecting future dynamics).

In summary, the paper argues that physical memory + robust coarse-grained transition preservation + internal operation selection = closed physical computation. The walking droplet satisfies the first two (partially) but fails the third, serving as a concrete test case for the necessity of the closure criterion.