Nonlocal Maxwell Demon in Thermodynamics

• Nonlocal Maxwell demon is an information-driven thermal mechanism where measurement and feedback occur remotely, ensuring compliance with the second law.
• It employs diverse models—remote-feedback controllers, detector-assisted setups, and topological ergotropy protocols—to simulate spatially separated work extraction.
• These frameworks integrate ergotropy teleportation, Landauer erasure, and infrastructure cost analysis to establish performance limits and thermodynamic horizons.

Searching arXiv for recent and foundational papers on nonlocal Maxwell demons and related variants. A nonlocal Maxwell demon is a thermodynamic or information-processing construction in which the acquisition or use of information is spatially separated from the location where work is extracted, entropy is reduced, or transport is rectified. In the literature, the term does not denote a single canonical model. It spans remote classical feedback controllers that replace bulk energy transfer by information transfer, nanoscale detector–system architectures in which kinetics are modified with negligible first-law back-action, protocols that simulate apparent quantum nonlocality by expending work, and finite-temperature quantum schemes that teleport ergotropy through a topologically protected resource (Hossenfelder, 2014, Strasberg et al., 2012, Hu et al., 2021, Abd-Rabbou et al., 14 May 2026). A separate but closely related line of work studies “demon” behavior in coherent scattering and demon-like sorting in nonreciprocal structures, often to delimit which forms of asymmetry do and do not qualify as genuine Maxwell-demon operation (Ruschhaupt et al., 7 Mar 2025, Manaselyan et al., 2019).

1. Operational meanings of nonlocality

The nonlocality in a nonlocal Maxwell demon is operational rather than uniquely ontological. In the remote-feedback setting, the demon and the work-producing machine are at different locations and exchange only classical signals; the output energy is drawn primarily from the machine’s local bath rather than transmitted from the controller (Hossenfelder, 2014). In stochastic nanoelectronics, the demon is nonlocal at the level of the observed subsystem because a detector modifies effective transition rates through capacitive detection and feedback, while the energetics of the system under observation can be made essentially unchanged to leading order in the relevant limit (Strasberg et al., 2012). In the surface-code construction, nonlocality is implemented by local operations and classical communication together with a shared topological resource, so that Alice injects energy locally into a logical degree of freedom and Bob reconstructs ergotropy in a distant battery using only classical syndrome information (Abd-Rabbou et al., 14 May 2026).

The literature also distinguishes nonlocal thermodynamics from quantum-nonlocality simulation. In the EPR-steering loophole construction, the demon is strictly local to Bob’s side, but it can simulate apparent steering correlations by basis-conditioned work and by exploiting the classical communication already allowed in steering tests; the result concerns simulated nonlocal correlations rather than a spatially separated work extractor (Hu et al., 2021). Conversely, some mesoscopic and scattering systems exhibit demon-like asymmetry or sorting without explicit measurement-and-feedback cycles, and these are often treated as adjacent rather than identical notions (Manaselyan et al., 2019, Ruschhaupt et al., 7 Mar 2025).

Paradigm	Nonlocal resource or channel	Thermodynamic role
Remote energy down-converter	Classical signals between demon and machine	Information selects thermal fluctuations for work extraction
Detector-assisted SET demon	Capacitive coupling and coarse-grained feedback	Information current changes effective affinities
Topological demon	Classical communication and shared surface code	Ergotropy is teleported and exponentially protected
Steering loophole	Local work plus allowed steering communication	Apparent nonlocal correlations are simulated
Scattering “demon”	Effective nonlocal single-channel description	Subject to a no-go bound in passive Hermitian parents

This multiplicity of meanings is central. A nonlocal Maxwell demon need not move energy over distance; it may instead move syndromes, side information, feedback commands, or effective control over transition rates.

2. Thermodynamic and informational structure

All nonlocal Maxwell-demon models in this literature are organized around the same basic resolution of the Maxwell-demon paradox: information has a thermodynamic cost, typically expressed through Landauer erasure, entropy production inequalities, or explicit infrastructure overheads. The canonical Landauer statement used across these works is that erasing one bit dissipates at least $k_B T \ln 2$ of heat (Abd-Rabbou et al., 14 May 2026, Hu et al., 2021). In remote classical architectures, the demon’s memory reset contributes the entropy cost that restores global second-law compliance, and the total efficiency of demon plus machine is bounded by a Carnot-like expression $\eta \le 1 - T_D/T$ once the demon’s erasure heat is included (Hossenfelder, 2014).

In the ergotropy-based formulation, the relevant payload is not bare internal energy but the unitary-extractable component of energy. For a state $\rho$ and Hamiltonian $H$, the ergotropy is

$$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$

where $\rho_{\mathrm{pass}}$ is the passive rearrangement of $\rho$ (Abd-Rabbou et al., 14 May 2026). The same work also distinguishes ergotropy from the nonequilibrium free energy $F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$, emphasizing that only part of the free-energy difference is unitarily extractable, while dissipative costs remain local and must be paid through entropy production and heat exchange.

In stochastic-thermodynamic implementations, the key object is a generalized entropy balance for the coarse-grained system. For the capacitively coupled detector–SET model, the effective SET entropy production in the fast-demon limit is

$\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$

with an information current $I_F$ that, in the strict Maxwell-demon limit, reduces to

$\eta \le 1 - T_D/T$0

(Strasberg et al., 2012). The associated modified local detailed balance is

$\eta \le 1 - T_D/T$1

This formulation makes explicit that the demon changes kinetics and effective affinities rather than directly supplying useful work to the observed subsystem.

A further information-theoretic bound appears in the topological ergotropy protocol, where the Sagawa–Ueda relation is written as

$\eta \le 1 - T_D/T$2

with $\eta \le 1 - T_D/T$3 the number of communicated syndrome bits (Abd-Rabbou et al., 14 May 2026). This bound is used there to show that the teleported ergotropy cannot exceed the information-processing cost once Landauer erasure and infrastructure costs are included.

3. Remote feedback and detector-mediated implementations

The classical remote demon analyzed as an energy down-converter is a paradigmatic nonlocal construction. The machine consists of $\eta \le 1 - T_D/T$4 cells, each containing $\eta \le 1 - T_D/T$5 identical four-level elements with a ground state $\eta \le 1 - T_D/T$6, a long-lived excited state $\eta \le 1 - T_D/T$7 with $\eta \le 1 - T_D/T$8, a short-lived upper level $\eta \le 1 - T_D/T$9 with $\rho$0, and a highest level $\rho$1 used to realize a four-level structure and noninvasive readout (Hossenfelder, 2014). The energy scales satisfy $\rho$2 and $\rho$3 with $\rho$4. The low-energy resonance at $\rho$5 is used to measure whether a cell is in the metastable excited manifold, and the higher-energy transition at $\rho$6 is used to stimulate emission and extract work.

The protocol has four stages: measurement, remote copying and decision, stimulated cooling/work extraction, and erasure. The demon sends a probe at energy $\rho$7 to identify cells in which all $\rho$8 elements occupy $\rho$9; it then sends actuation pulses at energy $H$0 only to those target cells; the machine extracts work from stimulated emission and refills from its local bath; finally, the demon erases its copied record (Hossenfelder, 2014). In the noiseless $H$1 limit the energy conversion factor is

$H$2

and for finite stimulation probability $H$3 it becomes

$H$4

In the general $H$5 case at $H$6, the work and down-conversion factor are

$H$7

with

$H$8

Latency degrades performance through

$H$9

so usable separation is bounded by metastable lifetime (Hossenfelder, 2014).

A later comment proposes an improved remote demon in which the computation and memory are implemented as a spectral “permutation machine,” so that internal processing proceeds by reversible permutations rather than erasure (Raptis, 2017). In that proposal, a machine step is $$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$0 with $$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$1, preserving the spectral density and, in the ideal limit, allowing the erasure heat $$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$2 to become negligible. The comment explicitly frames this as a way to raise the net work gain relative to the earlier remote-demon model.

At the nanoscale, a physically implemented Maxwell demon was constructed from a single-electron transistor capacitively coupled to a detector dot (Strasberg et al., 2012). The full system is described by the Hamiltonian

$$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$3

with no particle exchange between detector and SET. In the fast-demon limit, the detector equilibrates much faster than the SET, and the coarse-grained SET dynamics obey an effective rate equation with conditional probabilities slaved to the detector reservoir. In the regime $$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$4 and $$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$5, the detector’s action changes the kinetics while leaving the SET’s first-law energetics essentially unchanged to leading order. This is the sense in which the demon is nonlocal or quasi-nonlocal at the SET level: the control is mediated by information rather than appreciable energetic intervention in the observed subsystem (Strasberg et al., 2012).

4. Quantum nonlocality, steering, and work-assisted simulation

The relationship between Maxwell demons and quantum nonlocality is sharpened by the proposal of Maxwell demon-assisted EPR steering (Hu et al., 2021). In that construction, the demon is not spatially remote from the work extraction site; it is local to Bob’s side. Its significance lies instead in showing that apparent nonlocal correlations can be simulated by basis-conditioned work plus information processing. The demon learns Bob’s measurement setting, rotates Bob’s qubit into an eigenstate of the chosen basis, uses an internal random bit to determine which eigenstate to prepare, and then secretly informs Alice how to announce a consistent result using the classical channel permitted in steering tests (Hu et al., 2021).

For two measurement settings, the effective two-qubit state is modeled as

$$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$6

and the steering parameter is

$$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$7

With the local-hidden-state bound $$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$8, a violation occurs iff

$$W(\rho,H)=\mathrm{Tr}(\rho H)-\mathrm{Tr}(\rho_{\mathrm{pass}} H),$$9

(Hu et al., 2021).

The thermodynamic cost enters through erasure and reset. The average dissipated energy per acted run satisfies

$\rho_{\mathrm{pass}}$0

or, for superconducting-qubit reset by spontaneous emission,

$\rho_{\mathrm{pass}}$1

Over $\rho_{\mathrm{pass}}$2 runs,

$\rho_{\mathrm{pass}}$3

In this model, the threshold to fake steering with two settings is

$\rho_{\mathrm{pass}}$4

(Hu et al., 2021). The loophole can therefore be closed only by continuously monitoring the local heat or energy fluctuations at Bob’s side with sufficient sensitivity to detect the cumulative excess dissipation. The paper explicitly states that this loophole is specific to steering, not to Bell tests, because Bell scenarios disallow the classical communication exploited by the demon.

This construction is conceptually important for nonlocal-demon theory because it separates two issues that are sometimes conflated: nonlocal thermodynamic action and apparent quantum nonlocality. Here the demon does not nonlocally extract work; rather, it uses local work and hidden classical side information to mimic a nonlocal correlation witness.

5. Topological nonlocal demons and ergotropy teleportation

The most explicit recent realization of a nonlocal Maxwell demon is the surface-code protocol that teleports ergotropy at finite temperature using classical communication and a shared topological resource (Abd-Rabbou et al., 14 May 2026). The surface code is defined on a rectangular lattice

$\rho_{\mathrm{pass}}$5

with horizontal extent $\rho_{\mathrm{pass}}$6, vertical width $\rho_{\mathrm{pass}}$7, and stabilizer Hamiltonian

$\rho_{\mathrm{pass}}$8

where

$\rho_{\mathrm{pass}}$9

The code encodes one logical qubit. Alice and Bob hold identical two-level batteries with

$\rho$0

with Alice initially charged and Bob initially empty (Abd-Rabbou et al., 14 May 2026).

The protocol has five stages. First, Alice performs logical charging through

$\rho$1

which yields

$\rho$2

Because $\rho$3, the stabilizer energy is unchanged and the injected energy $\rho$4 is encoded in the logical degree of freedom. Alice then measures $\rho$5 boundary stabilizers at $\rho$6, sends the syndrome string $\rho$7 to Bob, and Bob decodes the logical class $\rho$8 using minimum-weight perfect matching on spacetime syndrome data. If $\rho$9, Bob applies $F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$0 and charges his battery (Abd-Rabbou et al., 14 May 2026).

The teleported ergotropy in Bob’s battery is determined by the decoding success probability:

$F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$1

and

$F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$2

for $F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$3, and zero otherwise (Abd-Rabbou et al., 14 May 2026). The physical error rate $F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$4 is related to temperature by

$F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$5

Below the threshold $F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$6, the logical error probability decays exponentially with $F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$7,

$F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$8

so the teleported ergotropy approaches $F(\rho)=\mathrm{Tr}(\rho H)-TS(\rho)$9 exponentially fast as the code distance increases (Abd-Rabbou et al., 14 May 2026).

The thermodynamic analysis introduces a net-work order parameter

$\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$0

which in the paper’s accounting becomes

$\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$1

with $\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$2 and $\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$3 (Abd-Rabbou et al., 14 May 2026). A continuous thermodynamic phase transition separates a profitable demon phase, $\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$4, from a thermal phase, $\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$5. For $\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$6, $\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$7, and $\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$8, the reported values are $\dot S_i^*=\dot S_i^{\rm SET}+I_F \ge 0,$9 and a thermodynamic critical point $I_F$0; the two thresholds are explicitly distinguished as information-theoretic protection loss versus thermodynamic unprofitability (Abd-Rabbou et al., 14 May 2026).

A central result is that causality enforces a quadratic infrastructure cost. Since syndrome data must traverse the separation before decoding, the number of syndrome rounds satisfies

$I_F$1

and the maintained spacetime volume scales as $I_F$2. If each stabilizer-round costs $I_F$3, then

$I_F$4

Together with Landauer erasure,

$I_F$5

this yields a strict second-law statement

$I_F$6

and an energy conservation law

$I_F$7

(Abd-Rabbou et al., 14 May 2026). The same reasoning defines a thermodynamic horizon,

$I_F$8

the largest separation for which $I_F$9. In the regime where $\eta \le 1 - T_D/T$00, the paper states that $\eta \le 1 - T_D/T$01 and is strictly independent of code distance once logical protection is effective (Abd-Rabbou et al., 14 May 2026).

This construction reinterprets quantum error correction as a thermodynamic resource. The payload is operationally meaningful work content, not a quantum state, and the nonlocality is realized without any physical operator or energy carrier traversing the channel in the usual sense; the extended logical string appears only computationally at the decoder (Abd-Rabbou et al., 14 May 2026).

6. Limits, misconceptions, and adjacent demon-like phenomena

Several recurring misconceptions are explicitly addressed in this body of work. The first is that nonlocal demons violate the second law by replacing energy transport with information transport. In the remote down-converter, the machine alone can appear highly efficient, but the combined demon-plus-machine efficiency is bounded by

$\eta \le 1 - T_D/T$02

after the demon’s memory entropy and erasure heat are included (Hossenfelder, 2014). In the detector–SET implementation, the full entropy production

$\eta \le 1 - T_D/T$03

remains nonnegative even when the coarse-grained SET description suggests information-assisted transport against bias (Strasberg et al., 2012). In the topological demon, second-law compliance is even stricter: the quadratic infrastructure term is always positive for $\eta \le 1 - T_D/T$04, so the demon never approaches perpetual motion at finite separation (Abd-Rabbou et al., 14 May 2026).

A second misconception is that any demon-like sorting mechanism is a Maxwell demon in the information-theoretic sense. The nonreciprocal quantum ring provides a clear counterexample (Manaselyan et al., 2019). In that system, Rashba spin–orbit coupling is applied only to one arm of a ring, while a perpendicular magnetic field controls whether the electron localizes in the Rashba arm or the non-Rashba arm. The result is arm-resolved differences in kinetic energy and effective spin “temperature,” together with suppression of Aharonov–Bohm oscillations in the ground state (Manaselyan et al., 2019). Yet the same source explicitly characterizes the mechanism as resembling the action of a demon rather than implementing full measurement, memory, erasure, and feedback. The total state remains pure at $\eta \le 1 - T_D/T$05, and no work is extracted from a single reservoir.

A third misconception concerns coherent scattering. One might expect that nonlocal, non-Hermitian effective single-channel potentials derived by Feshbach elimination could realize a perfect scattering Maxwell demon. The no-go theorem for multichannel scattering shows otherwise (Ruschhaupt et al., 7 Mar 2025). For a designated asymptotic channel, demon behavior would require

$\eta \le 1 - T_D/T$06

But block unitarity of the full Hermitian multichannel $\eta \le 1 - T_D/T$07 matrix implies the cross inequalities

$\eta \le 1 - T_D/T$08

which immediately forbid the demonic limit (Ruschhaupt et al., 7 Mar 2025). The same paper defines a distance-to-demon parameter

$\eta \le 1 - T_D/T$09

and proves the tight bound

$\eta \le 1 - T_D/T$10

Thus even an effective nonlocal, non-Hermitian single-channel description cannot approach a perfect demon when it descends from a passive Hermitian parent system.

Taken together, these results suggest a disciplined taxonomy. Genuine nonlocal Maxwell demons require an explicit information-to-work channel or a rigorously defined ergotropy-transfer protocol with complete thermodynamic accounting. Demon-assisted simulations of quantum correlations, passive spin sorting, and coherent scattering asymmetries are closely related because they expose the same interplay among information, feedback, asymmetry, and entropy production, but they are not interchangeable categories (Hu et al., 2021, Manaselyan et al., 2019, Ruschhaupt et al., 7 Mar 2025). The most developed current formulation is the topological finite-temperature demon, in which nonlocal thermodynamics is mediated by quantum error correction and bounded by a causality-induced thermodynamic horizon (Abd-Rabbou et al., 14 May 2026).