Entropy Pumping in Nonequilibrium Systems

• Entropy pumping is a nonequilibrium process where structured driving or feedback actively redistributes entropy beyond passive heat diffusion.
• In systems like quantum dots, feedback-cooled oscillators, and spin-pumping setups, the entropy balance is modified without eliminating the underlying work dissipation.
• The concept unifies multiple phenomena by showing how controlled protocols reassign entropy via phase-space reorganization, coherence enhancement, or hidden controller effects.

Searching arXiv for recent and foundational papers on entropy pumping across quantum thermodynamics, feedback cooling, and pumping transport. Entropy pumping denotes a class of nonequilibrium processes in which external driving, feedback, or mode conversion induces a directed entropy flow or an entropy-balance contribution that cannot be reduced to passive thermal relaxation alone. Across the literature, the term appears in several technically distinct settings: adiabatic quantum pumps in mesoscopic conductors, feedback-controlled Langevin systems, spin pumping in magnetic multilayers, optical pumping in open quantum systems, and entropy-flux analyses of continuous bosonic radiation. In each case, the central issue is how entropy transport, entropy reduction, or entropy production is modified by structured driving. In adiabatic quantum pumping through a resonant-level quantum dot, entropy pumping is a second-order effect in the driving speed and is closely tied to dissipation and quantized transport (Nello et al., 2023). In feedback cooling, “entropy pumping” is the explicit entropy-reduction term extracted by the controller from the observed subsystem, thereby modifying the second law at the apparent level (Munakata et al., 2013). Related formulations appear in spin-current thermodynamics (Taniguchi et al., 2014), NV-center optical pumping (Medina et al., 11 Mar 2025), and entropy-flux analyses of parametric amplifiers (Khlebnikov, 9 Jan 2025).

1. Conceptual scope and core definitions

The broadest common structure is an entropy balance for an open driven system in which the entropy change cannot be identified solely with heat exchange divided by temperature. Instead, an additional contribution appears because the driving protocol, feedback loop, or scattering geometry reorganizes phase-space volume, redistributes occupations, or exports correlations to external degrees of freedom.

In adiabatic quantum pumping, charge, energy, and entropy are transported when at least two parameters of a scatterer are varied periodically in time with period $T_0$ and no bias $\mu_L-\mu_R$ is applied. For the resonant-level model treated in "Thermodynamics of adiabatic quantum pumping in quantum dots" (Nello et al., 2023), the adiabatic limit is specified by $\Omega=2\pi/T_0\ll\Gamma$, with $\Gamma$ the dot broadening. In this regime, the pumped charge per cycle is given by Brouwer’s formula,

$$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$

where $x_{1,2}(t)$ are the two driving parameters, $S_{\alpha\beta}(\epsilon;x)$ the instantaneous scattering matrix, and $A$ the area enclosed in parameter space over one cycle (Nello et al., 2023). Entropy flow arises because the driven scatterer generates a nonequilibrium distribution on the dot that leaks back into the reservoirs as heat and entropy.

In stochastic thermodynamics with feedback cooling, the same phrase refers to a specific correction term in the entropy-production budget. For a harmonic oscillator subjected to a velocity-dependent feedback force $F_{\rm fb}=-\gamma' v$, the trajectory-level entropy production is

$$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$

with $\mu_L-\mu_R$0 in the controlled-system description (Munakata et al., 2013). Here entropy pumping represents the continuous contraction of momentum phase-space induced by the feedback damping.

These two usages are not identical. In the mesoscopic pump, entropy pumping refers to entropy transported or produced by slow cyclic driving; in feedback cooling, it denotes the entropy reduction extracted by the controller from the observed subsystem. This suggests that “entropy pumping” is best understood as a family of thermodynamic effects rather than a single universal observable.

2. Adiabatic quantum pumping in resonant-level quantum dots

The quantum-dot realization studied in (Nello et al., 2023) is a single-level quantum dot connected to two fermionic leads. The thermodynamic description is built by adiabatic expansion in small time derivatives of the control parameters, notably the dot energy $\mu_L-\mu_R$1 and the tunnelling rates to the reservoirs. The instantaneous spectral function is

$\mu_L-\mu_R$2

with $\mu_L-\mu_R$3.

A central result is that the entropy production rate first appears at second order in the driving: $\mu_L-\mu_R$4 The entropy pumped per cycle is then

$\mu_L-\mu_R$5

which is generically nonzero only at second order in the driving speeds (Nello et al., 2023).

The same adiabatic expansion yields the dissipated power,

$\mu_L-\mu_R$6

together with the compact relation

$\mu_L-\mu_R$7

showing that the second-order work term is directly proportional to the irreversibly produced entropy (Nello et al., 2023).

A major conclusion of the paper is the coexistence of two seemingly opposite features in the charge-quantization limit. When the dot level is swept far above and below the Fermi level so that each cycle loads one electron from one lead and unloads it into the other, the pumped charge satisfies $\mu_L-\mu_R$8, the charge noise vanishes, and simultaneously $\mu_L-\mu_R$9 and $\Omega=2\pi/T_0\ll\Gamma$0 everywhere along the cycle (Nello et al., 2023). Yet the dissipated work per cycle remains finite and saturates to a quantized value. For the peristaltic cycle,

$\Omega=2\pi/T_0\ll\Gamma$1

with $\Omega=2\pi/T_0\ll\Gamma$2 (Nello et al., 2023).

This establishes a specific thermodynamic pattern: charge quantization is accompanied by vanishing entropy production and vanishing noise, while the dissipated work approaches a quantized plateau. The paper states that these observations hold irrespective of the details of the cycle, provided the protocol isolates a limit in which the scatterer’s conductance is zero during the loading and unloading strokes so as to avoid leakage currents (Nello et al., 2023).

3. Feedback cooling and entropy pumping in stochastic thermodynamics

In the cold-damping problem analyzed in "Feedback cooling, measurement errors, and entropy production" (Munakata et al., 2013), the controlled system is a one-dimensional harmonic oscillator of mass $\Omega=2\pi/T_0\ll\Gamma$3, friction $\Omega=2\pi/T_0\ll\Gamma$4, and spring constant $\Omega=2\pi/T_0\ll\Gamma$5, coupled to a heat bath at temperature $\Omega=2\pi/T_0\ll\Gamma$6 and subject to a linear feedback force. With perfect velocity feedback, the underdamped Langevin equation is

$\Omega=2\pi/T_0\ll\Gamma$7

The defining stochastic-thermodynamic quantities are the system entropy $\Omega=2\pi/T_0\ll\Gamma$8, the medium entropy change

$\Omega=2\pi/T_0\ll\Gamma$9

and the trajectory entropy production

$\Gamma$0

At the path-probability level,

$\Gamma$1

which yields $\Gamma$2 in that convention (Munakata et al., 2013). The corresponding integral fluctuation theorem,

$\Gamma$3

implies the generalized second-law inequality

$\Gamma$4

At the ensemble level, if $\Gamma$5 obeys the Fokker–Planck equation with damping $\Gamma$6, then

$\Gamma$7

with

$\Gamma$8

for the idealized case (Munakata et al., 2013).

The paper emphasizes that entropy pumping here does not describe ordinary heat release into the reservoir. Rather, it is the entropy-reduction contribution associated with active feedback and reflects the presence of hidden degrees of freedom in the controller. Once those degrees of freedom are included, the joint process of oscillator plus controller becomes Markovian and obeys the conventional second law without a separate pumping subtraction: $\Gamma$9 with

$$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$0

Projecting out the controller underestimates dissipation, and one can show

$$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$1

(Munakata et al., 2013).

Measurement noise modifies the apparent pumping term. In model V, where the detector measures $$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$2 with white noise of spectral density $$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$3, the apparent feedback force becomes

$$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$4

The entropy balance keeps the same structure,

$$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$5

but now

$$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$6

In the nonequilibrium steady state with

$$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$7

one finds

$$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$8

(Munakata et al., 2013). The detailed summary further states that measurement noise raises $$Q_\alpha \;=\;\sum_\beta\int\frac{d\epsilon}{4\pi}\;(-f'(\epsilon))\iint_{A}\!\frac{dx_1\,dx_2}{i}\,\Bigl[ \partial_{x_2}S_{\alpha\beta}\,\partial_{x_1}S^\dagger_{\beta\alpha} - \partial_{x_1}S_{\alpha\beta}\,\partial_{x_2}S^\dagger_{\beta\alpha} \Bigr],$$9 and reduces the magnitude of the negative pumping rate $x_{1,2}(t)$0.

4. Quantization, reversibility, and entropy suppression

The relation between entropy pumping and reversibility is especially sharp in the adiabatic quantum-dot problem. The peristaltic cycle, consisting of loading one electron while the dot is coupled only to the left reservoir and unloading it after switching the coupling to the right reservoir, yields $x_{1,2}(t)$1, noise $x_{1,2}(t)$2, and $x_{1,2}(t)$3 in the quantization limit (Nello et al., 2023). A triangular cycle with fixed $x_{1,2}(t)$4 and a triangular path in $x_{1,2}(t)$5 produces a fractional plateau with

$x_{1,2}(t)$6

in the large-driving-amplitude limit, while the dissipated work per cycle tends to $x_{1,2}(t)$7 (Nello et al., 2023).

These results support a broader principle explicitly stated in (Nello et al., 2023): whenever the pumped charge is quantized to an integer multiple of $x_{1,2}(t)$8, the entropy produced per cycle vanishes and thermal and shot noise vanish as well. The paper interprets this as a signature of an effectively reversible, noise-free transport process. The reversible, geometric part of work and heat, linear in driving, integrates to zero over a closed cycle in the absence of net bias of $x_{1,2}(t)$9 or $S_{\alpha\beta}(\epsilon;x)$0, whereas irreversible dissipation and entropy production are pure second-order effects.

A related but conceptually different entropy-suppression mechanism appears in sequences of electric pulses driving Schwinger pair production (Dunne et al., 2022). There, for each momentum mode $S_{\alpha\beta}(\epsilon;x)$1, the reduced density matrix is diagonal,

$S_{\alpha\beta}(\epsilon;x)$2

with entropy

$S_{\alpha\beta}(\epsilon;x)$3

For antisymmetric pulse sequences, the occupation spectrum becomes

$S_{\alpha\beta}(\epsilon;x)$4

and the entropy is correspondingly modified to

$S_{\alpha\beta}(\epsilon;x)$5

The summary states that interference redistributes the modes without increasing the total particle number and lowers the entropy per particle, so that pulse parameters can be tuned to “pump” entropy out of the produced state (Dunne et al., 2022). This suggests an analogy with quantum-dot pumping only at the level of entropy suppression under structured driving; the microscopic mechanism is entirely different.

5. Spin pumping, heat pumping, and generalized entropy currents

In spintronics, pure spin-current generation by ferromagnetic resonance produces entropy through spin-dependent transport and interface conversion. The generalized thermodynamic treatment in "Dissipation due to pure spin-current generated by spin pumping" (Taniguchi et al., 2014) introduces spin-resolved electrochemical potentials $S_{\alpha\beta}(\epsilon;x)$6 and $S_{\alpha\beta}(\epsilon;x)$7, the spin accumulation $S_{\alpha\beta}(\epsilon;x)$8, and the spin-current density $S_{\alpha\beta}(\epsilon;x)$9. For a one-dimensional multilayer at uniform temperature $A$0, the bulk entropy-production density is

$A$1

For pure spin transport with $A$2, this reduces to

$A$3

or, in the collinear case,

$A$4

At an interface,

$A$5

(Taniguchi et al., 2014).

For a precessing ferromagnet, the pumped spin current is

$A$6

Its associated pumped energy flux is

$A$7

and the entropy flux is

$A$8

In the limit $A$9,

$F_{\rm fb}=-\gamma' v$0

while the spin-pumping enhancement of Gilbert damping is

$F_{\rm fb}=-\gamma' v$1

so that symbolically

$F_{\rm fb}=-\gamma' v$2

(Taniguchi et al., 2014). The theory therefore identifies an entropy current carried away from the interface by spin pumping, with dissipation directly proportional to the experimentally observed damping enhancement.

A different transport setting is the hydrodynamic, charge-neutral electron liquid studied in "Electronic pumping of heat without charge transfer" (Andreev, 2021). There the central object is the entropy current $F_{\rm fb}=-\gamma' v$3 and the heat current $F_{\rm fb}=-\gamma' v$4, generated by a time-dependent external potential $F_{\rm fb}=-\gamma' v$5 in the adiabatic regime. At charge neutrality and to leading order in $F_{\rm fb}=-\gamma' v$6, the flow velocity is uniform and the instantaneous pumping velocity is

$F_{\rm fb}=-\gamma' v$7

with

$F_{\rm fb}=-\gamma' v$8

For a traveling-wave potential $F_{\rm fb}=-\gamma' v$9, one obtains

$$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$0

which becomes

$$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$1

in the pristine limit (Andreev, 2021). This is explicitly described as pumping of entropy or heat without net particle transfer or voltage buildup. A plausible implication is that hydrodynamic entropy transport provides a macroscopic analogue of the entropy-current viewpoint that appears microscopically in mesoscopic pumps and spin pumping.

6. Open-quantum-system formulations and radiation-field entropy flow

Open quantum systems driven by incoherent pumping provide another setting in which entropy changes split naturally into heat-induced and work-induced components. In the eight-level NV-center model analyzed in "Thermodynamics of the optical pumping process in Nitrogen-Vacancy centers" (Medina et al., 11 Mar 2025), the dynamics obey a Lindblad master equation

$$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$2

with a time-independent Hamiltonian and dissipators representing laser pumping, fluorescence, inter-system crossing, and weak non-spin-preserving leaks.

Using Alicki’s partitioning, the first law takes the form

$$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$3

with

$$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$4

For the dominant channels,

$$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$5

where $$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$6 and $$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$7 are the populations of the ground and excited triplets (Medina et al., 11 Mar 2025).

The von Neumann entropy

$$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$8

satisfies

$$\sigma[\{x_s\}] = \Delta s_{\rm sys} + \Delta s_m - \Delta s_{\rm pu},$$9

and upon time integration,

$\mu_L-\mu_R$00

(Medina et al., 11 Mar 2025). The paper further relates the measurable fluorescence rate $\mu_L-\mu_R$01 directly to the heat current through

$\mu_L-\mu_R$02

It also reports that increasing the laser pump rate raises the entropy of the final state after laser-off relaxation and thereby hinders polarization efficiency (Medina et al., 11 Mar 2025). This is not called “entropy pumping” in the same formal sense as (Munakata et al., 2013), but it exemplifies a driven open-system decomposition in which the entropy change has distinct work- and heat-related pieces.

A more radical formulation of entropy flow appears in the continuous-spectrum setting of "Entropy flow in a parametric amplifier" (Khlebnikov, 9 Jan 2025). There, entropy flux in an output radiation field is defined by discretizing the field into Gabor atoms,

$\mu_L-\mu_R$03

The resulting Shannon entropy is

$\mu_L-\mu_R$04

or approximately

$\mu_L-\mu_R$05

with entropy flux

$\mu_L-\mu_R$06

For a driven parametric amplifier coupled to a zero-temperature Markovian bath, the late-time photon-number output flux and energy flux remain nonzero,

$\mu_L-\mu_R$07

yet

$\mu_L-\mu_R$08

(Khlebnikov, 9 Jan 2025). The summary attributes this to the buildup of off-diagonal coherences between distinct Gabor modes, which restore the purity of each window’s multimode state. This again separates energy transport from entropy transport, paralleling the quantum-dot result that nontrivial transport need not imply positive entropy pumping in the naive sense.

7. Common themes, distinctions, and misconceptions

A recurring misconception is that entropy pumping always means negative entropy production. The literature does not support that universal identification. In feedback cooling, the pumping term indeed enters with a sign that reduces the apparent entropy production of the controlled subsystem (Munakata et al., 2013). In adiabatic quantum pumping, by contrast, the relevant computed quantity is the entropy production rate $\mu_L-\mu_R$09, which is a second-order transport-induced entropy flow that vanishes in the quantized limit rather than becoming a large negative quantity (Nello et al., 2023). In spin pumping, entropy pumping refers to entropy carried away by spin-current-mediated energy transfer and subsequently dissipated through spin relaxation (Taniguchi et al., 2014). In radiation problems, entropy flow may even vanish while energy and particle fluxes stay finite (Khlebnikov, 9 Jan 2025).

A second misconception is that entropy pumping is synonymous with heat pumping. The hydrodynamic neutral-electron problem shows that heat current can be written as $\mu_L-\mu_R$10, making the two closely related (Andreev, 2021), but the feedback-cooling formulation demonstrates that an entropy-pumping term can arise from phase-space contraction and hidden controller degrees of freedom rather than from a directly measurable heat current (Munakata et al., 2013). Conversely, the NV-center analysis separates entropy change into $\mu_L-\mu_R$11 and $\mu_L-\mu_R$12, indicating that entropy modification by pumping need not be reducible to a thermal channel alone (Medina et al., 11 Mar 2025).

A third misconception is that reversible or quantized pumping must eliminate all work cost. The quantum-dot study shows the opposite: in the charge-quantization limit, entropy production and noise vanish, yet the dissipated work per cycle saturates to a finite quantized value proportional to the speed of the control parameter (Nello et al., 2023). This suggests that vanishing entropy production in the pumped subsystem does not imply zero energetic overhead in the full driven process.

Across these disparate settings, several common principles emerge. First, entropy pumping is fundamentally tied to nonequilibrium coarse-graining: scattering reservoirs, projected-out controllers, spin accumulations, incoherent optical drives, or discretized field modes. Second, the sign and interpretation of the entropy contribution depend on which degrees of freedom are retained. Third, regimes of suppressed entropy flow are often associated with high coherence, quantization, or optimized control. In mesoscopic charge pumps this appears as $\mu_L-\mu_R$13 with $\mu_L-\mu_R$14 (Nello et al., 2023); in feedback cooling as maximal cold-damping efficiency with $\mu_L-\mu_R$15 and $\mu_L-\mu_R$16 (Munakata et al., 2013); and in parametric amplification as the emergence of off-diagonal mode coherences with asymptotically vanishing entropy flux (Khlebnikov, 9 Jan 2025).

Taken together, these results indicate that entropy pumping is not a single phenomenological law but a unifying thermodynamic lens for analyzing how driven systems export, suppress, or reassign entropy under controlled nonequilibrium conditions.