Skip to content
← Back to research
PublishedBioSystems (2025)

Timing Inaccessibility and the Projection Bound: Resolving Maxwell's Demon for Continuous Biological Substrates

What's this about?

This paper asks whether continuous biological coupling can postpone some costs associated with irreversible record formation. It does not imply that all pre-output neural dynamics are reversible or that digital hardware pays the Landauer bound on every clock cycle.

The paper traces the Maxwell's demon lineage (Maxwell → Szilard → Landauer → Bennett) and asks how its bookkeeping changes for continuous substrates and path-degenerate dynamics. Landauer and Bennett did not require memory to be digitally implemented: the relevant issue is whether a physical protocol reduces logical alternatives, stabilizes a record, or resets memory. Structural correlations and written bits are both physical and must be analysed at the level of the actual operation.

Landauer's bound applies when a physical device performs a logically irreversible operation such as resetting a stabilized memory. The paper's two-stage model separates transient correlation from durable registration: the first stage need not incur a per-bit erasure bound, while the second may. Actual energy use still depends on coupling, noise, maintenance, error correction, and the implementation.

Key findings

  • Projection-bound hypothesis: registration costs depend on which distinctions are physically stabilized or reset

  • Two-stage model: transient coupling versus durable stabilization/registration

  • Calls for operation-level accounting of correlations, durable records, and reset rather than headline path-count estimates

  • Digital efficiency depends on projection boundary placement, not "digital vs analog"

  • Proposes a testable contribution to biological–digital efficiency differences

Citation

Todd, I. (2025). Timing Inaccessibility and the Projection Bound: Resolving Maxwell's Demon for Continuous Biological Substrates. BioSystems.
doi: 10.1016/j.biosystems.2025.105632

Workflow: Claude 4.5 Sonnet (Anthropic) for drafting; GPT-5 (OpenAI) and Grok (xAI) for review. Author reviewed all content and takes full responsibility.