Landauer Entropy: Thermodynamic Cost

• Landauer entropy is the decrease in an information-bearing system’s entropy that sets a fundamental lower bound on the thermodynamic cost of irreversible processes.
• The equality form refines the textbook Q ≥ kT ln2 by incorporating finite reservoir effects, mutual information, and deviations from equilibrium.
• Extensions of Landauer entropy span quantum thermodynamics, computational information theory, and even black-hole physics, highlighting its broad impact across physics.

to=arxiv_search 天天爱彩票怎么json
{"query":"id:1306.4352 OR all:\"An improved Landauer Principle with finite-size corrections\"","max_results":5}
to=arxiv_search აციოjson
{"query":"id:1901.10327 OR all:\"Physical Foundations of Landauer's Principle\"","max_results":5}
to=arxiv_search აციოjson
{"query":"id:2509.17060 OR all:\"Going beyond Landauer: Information-cost relations from inference based on the maximum entropy principle\"","max_results":5}
to=arxiv_search აციოjson
{"query":"id:2506.10876 OR all:\"Landauer Principle and Thermodynamics of Computation\"","max_results":5}
Landauer entropy denotes the entropy accounting that links information-theoretic irreversibility to thermodynamic cost. In the formulation of Reeb and Wolf, the “Landauer entropy” is the von Neumann entropy decrease of the system,
[
\Delta S_S := S(\rho_S)-S(\rho_S'),
\qquad
S(\rho):=-\operatorname{Tr}[\rho\ln\rho],
]
for a logically irreversible process implemented by a joint unitary on a system (S) and a thermal reservoir (R) [1306.4352]. In thermodynamics-of-computation treatments, the same informational quantity is the entropy reduction that must be compensated by an equal or greater thermodynamic entropy increase elsewhere, typically in a thermal environment, giving the canonical one-bit benchmark (Q_{\min}\ge k_B T\ln 2) and (\Delta S_{\mathrm{env}}\ge k_B\ln 2) [2506.10876]. Modern quantum treatments replace the textbook inequality by an exact balance law in which correlations, finite-reservoir effects, and nonequilibrium distortion of the bath contribute explicitly to the dissipation budget [1306.4352].

1. Definition, notation, and sign conventions

Landauer’s principle is usually stated for a system (S) initially uncorrelated with a thermal reservoir (R), with
[
\rho_{SR}=\rho_S\otimes \rho_R^\beta,
\qquad
\rho_R^\beta=\frac{e^{-\beta H_R}}{\operatorname{Tr}[e^{-\beta H_R}]},
\qquad
\beta=\frac{1}{k_B T},
]
and joint unitary dynamics
[
\rho'_{SR}=U(\rho_S\otimes \rho_R^\beta)U^\dagger.
]
The dissipated heat is identified with the reservoir energy increase,
[
\Delta Q:=\operatorname{Tr}[H_R(\rho_R'-\rho_R^\beta)].
]
With Reeb–Wolf’s convention, erasure corresponds to (\Delta S_S>0), because the system becomes more ordered [1306.4352].

A distinct but compatible sign convention is common in nonequilibrium thermodynamics: one writes (\Delta S_{\mathrm{sys}}=S(\rho_f)-S(\rho_i)), so erasure has (\Delta S_{\mathrm{sys}}<0), and the classical Landauer inequality takes the form
[
Q\ge -T\Delta S_{\mathrm{sys}}.
]
For one-bit erasure, (\Delta S_{\mathrm{sys}}=-\ln 2), hence (Q\ge k_B T\ln 2) when (k_B) is restored [2509.17060]. The two sign conventions differ only by whether the entropy change is written as a decrease of the system or as a signed final-minus-initial change.

The informational content that enters Landauer entropy is Shannon entropy for classical memories and von Neumann entropy for quantum memories. In the classical case,
[
H=-\sum_x p(x)\log_2 p(x),
]
while in the quantum case
[
S(\rho)=-\operatorname{Tr}(\rho\ln\rho).
]
The general content of the principle is that the reduction of information entropy in the information-bearing degrees of freedom constrains the minimum thermodynamic entropy production or heat dissipation required by the process [2506.10876].

2. Equality form in quantum statistical mechanics

In the Reeb–Wolf formulation, the basic structural identity is the “Second Law Lemma”
[
\left[S(\rho_S')-S(\rho_S)\right]+\left[S(\rho_R')-S(\rho_R)\right]=I(S':R')\ge 0,
]
equivalently
[
\Delta = \Delta S_S + I(S':R') \ge \Delta S_S,
]
where (\Delta=S(\rho_R')-S(\rho_R)) and
[
I(S':R'):=S(\rho_S')+S(\rho_R')-S(\rho'_{SR})
]
is the mutual information created between system and reservoir [1306.4352].

The sharpened Landauer principle is then an equality rather than merely an inequality:
[
\beta\Delta Q=\Delta S_S + I(S':R') + D(\rho_R'|\rho_R^\beta)\ge \Delta S_S,
]
with quantum relative entropy
[
D(\rho|\sigma):=\operatorname{Tr}[\rho(\ln\rho-\ln\sigma)].
]
This decomposition identifies three separate contributions. The term (\Delta S_S) is the Landauer entropy proper, namely the information decrease in (S). The mutual information term (I(S':R')) quantifies correlations created during the process. The relative-entropy term (D(\rho_R'|\rho_R^\beta)) measures the deviation of the final reservoir state from the initial thermal state and is thermodynamically the excess dimensionless free energy stored in the reservoir [1306.4352].

A central consequence is that the textbook equality (\beta\Delta Q=\Delta S_S) is attained only in trivial cases. Reeb and Wolf show that Landauer’s bound is tight iff
[
\rho'_{SR}=\rho_S'\otimes \rho_R',
\qquad
\rho_R'=\rho_R,
\qquad
\rho_S'=V\rho_SV^\dagger
]
for some unitary (V) on (S), equivalently (\Delta S_S=\Delta Q=0). Any nontrivial entropy change or any nonzero heat flow necessarily yields strict inequality because at least one of the nonnegative correction terms is positive [1306.4352].

This equality form also clarifies a common ambiguity in informal presentations. The minimal heat is not, in general, “(k_B T) times the erased entropy” simpliciter. That expression is a limiting case in which the bath remains effectively thermal and uncorrelated with the system. Finite disturbances of the reservoir and correlation buildup are not optional corrections; they are the generic structure of the exact balance law.

3. Finite reservoirs, effective size, and unattainability

For finite reservoirs, the equality immediately yields explicit strengthening of the Landauer bound. If (d=d_R<\infty), Reeb and Wolf prove that for (2\le d<\infty),
[
\beta\Delta Q \ge
\begin{cases}
\Delta S_S + M(\Delta S_S,d)\ \ge\ \Delta S_S + \dfrac{(\Delta S_S)^2}{2N}, & \Delta S_S\ge 0, \[6pt]
\Delta S_S + \Big[N-\Delta S_S-\sqrt{N^2-2N\Delta S_S}\Big], & \Delta S_S\le 0,
\end{cases}
]
for any (N\ge N(d)), where
[
N(d):=\max_{0<r<1/2}r(1-r)\left(\log\frac{1-r}{r}(d-1)\right)^2,
]
and (M(x,d)) is the tight dimension-dependent lower bound induced by
[
D(\sigma|\rho)\ge M(S(\sigma)-S(\rho),d).
]
A universal estimate is
[
M(x,d)\ge N e^{x/N}-N-x \ge \frac{x^2}{2N}+\frac{x^3}{6N^2}\ge 0.
]
The parameter (N(d)) plays the role of an effective size parameter and scales as
[
N(d)=\frac14\log^2 d + O(1),
]
so the corrections decay only logarithmically with reservoir dimension [1306.4352].

The physical interpretation is direct. Finite reservoirs cannot act as perfect equilibrium baths: bounded heat capacity and bounded variance of (\log\rho^\beta) force appreciable deviations from thermality, and those deviations produce strictly positive (D(\rho_R'|\rho_R^\beta)). At the same time, the unitary interaction generically creates correlations, contributing (I(S':R')). The extra heat above (k_B T\Delta S_S) is therefore not an implementation defect but a finite-size necessity [1306.4352].

For a one-qubit entropy reduction (\Delta S_S=\ln 2), the finite-size penalties can be substantial. Using the quadratic lower bound with natural logarithms:

Reservoir dimension (d)	Chosen (N)	Lower bound on (\beta\langle Q\rangle)
(4)	(\approx 1.302)	(\approx 0.877)
(16)	(\approx 2.835)	(\approx 0.778)

These values sit above the classic (\ln 2\approx 0.693), with the (d=4) case about (26\%) above the textbook Landauer limit and the (d=16) case about (12\%) above it [1306.4352].

The same analysis shows an unattainability feature resembling a Third-Law statement. Exact erasure of a qubit from maximally mixed to pure state with a finite reservoir at finite temperature is not physically achievable without infinite energies or (\beta=\infty). Nevertheless, the bound can be approached arbitrarily closely by making the reservoir large and using stepwise swap operations; in the (d\to\infty) limit, (M(\Delta S_S,d)\to 0), (I(S':R')\to 0), and (D(\rho_R'|\rho_R^\beta)\to 0), so the dissipated heat approaches (k_B T\Delta S_S) [1306.4352].

4. Computational information, correlated information, and common misconceptions

The physical foundations literature places Landauer entropy in a computational setting. The central point is not that “entropy moved from one subsystem to another,” but that controlled, correlated computational information is ejected into uncontrolled thermal degrees of freedom and then re-randomized. In that setting, the minimal environmental entropy increase is
[
\Delta S_{\mathrm{env}}\ge k\ln 2\cdot H_{\mathrm{lost}},
\qquad
Q_{\min}\ge kT\ln 2\cdot H_{\mathrm{lost}},
]
where (H_{\mathrm{lost}}) is the number of lost bits of controlled information [1901.10327].

A key structural relation is the Fundamental Theorem of the Thermodynamics of Computation,
[
S(\Phi)=H(C)+S(\Phi|C)=k\ln 2\cdot H(C)+S_{nc}(\Phi),
]
where (C) is the computational state, (\Phi) the underlying physical microstate, and (S_{nc}(\Phi)) the conditional non-computational entropy. Since microscopic dynamics are bijective—unitary in quantum theory and symplectic in classical Hamiltonian dynamics—a logically irreversible map that reduces (H(C)) must produce a compensating increase in non-computational entropy when the lost information is not preserved in controllable correlations [1901.10327].

This framework is used to correct several persistent misconceptions. First, Landauer’s principle is not a cost for arbitrary entropy transfer between subsystems. Entropy exchange can be thermodynamically reversible when correlations are preserved. Second, not every many-to-one logical map dissipates (kT\ln 2) per operation. The cost attaches to loss of correlated information to uncontrolled thermal degrees of freedom, not to many-to-one mapping per se. Third, re-randomizing an already-random and uncorrelated bit need not increase entropy. These points motivate reversible computing, decomputation, and conditional logical reversibility as the route to asymptotically avoiding Landauer dissipation [1901.10327].

Review literature on the thermodynamics of computation makes the same point operationally: Landauer entropy is the minimal thermodynamic entropy that must be produced in an environment when information is irreversibly erased, but practical erasure typically dissipates much more because of finite-time driving, coherence, control noise, finite baths, and non-Markovian effects [2506.10876]. This suggests that Landauer entropy should be understood as a lower-bound contribution within a broader nonequilibrium dissipation budget rather than as an empirical estimate of total device heating.

5. Beyond the textbook thermal-bath bound

Recent work extends Landauer-type relations beyond the single conserved quantity, weakly perturbed thermal bath scenario. An inference-based approach built on the maximum entropy principle introduces a reference state
[
\rho_r(t)=Z_r(t)^{-1}\exp!\left(-\sum_i \lambda_i(t)\mathcal O_i\right),
\qquad
-\frac{\partial \ln Z_r}{\partial \lambda_i}=\langle \mathcal O_i\rangle(t),
]
and proves the exact identity
[

\sum_i \lambda_i(t)\langle \mathcal O_i\rangle(t)-S(t)

-\ln Z_r(t)+D[\rho_S(t)|\rho_r(t)].
]
From it one obtains, for multiple conserved charges,
[
\Delta S(t)=\sum_i \lambda_i(0)\Delta C_i^S(t)+\mathcal R(t)-D[\rho_S(t)|\rho_r(t)],
]
hence the upper bound
[
-\sum_i \lambda_i(0)\Delta C_i^S(t)\le \mathcal U_M(t)\equiv \mathcal R(t)-\Delta S(t).
]
A parallel construction for Gaussian MaxEnt reference states yields a Landauer-type bound on fluctuation costs:
[
\lambda_v(0)\Delta\mathrm{Var}\mathcal O+\mathcal R_v(t)\ge \Delta S(t),
\qquad
\Delta\mathrm{Var}\mathcal O\ge \frac{\Delta S(t)-\mathcal R_v(t)}{\lambda_v(0)}.
]
In this framework, Landauer entropy—the system entropy change—acts as a universal information yardstick that constrains not only energetic costs but also higher-order fluctuation costs in finite-time quantum processes with multiple charges and non-thermal environments [2509.17060].

At very low temperature the classical form (Q\ge k_B T S_L) becomes trivial. An all-temperature refinement avoids this collapse by expressing the bound through the bath heat capacity:
[
Q \ge \mathcal Q(\mathcal S^{-1}(S_L)),
\qquad
\mathcal Q(T')=\int_T^{T'} C_B(\tau)\,d\tau,
\qquad
\mathcal S(T')=\int_T^{T'} \frac{C_B(\tau)}{\tau}\,d\tau,
]
with (S_L:=-\Delta S\ge 0). This refined bound is always tighter than the original one, tends to it at high temperature, and remains nontrivial at (T\to 0). For a (1)-dimensional waveguide,
[
Q \ge T S_L + \frac{3\hbar c}{\pi L}S_L^2,
]
so at (T=0),
[
Q \ge \frac{3\hbar c}{\pi L}S_L^2.
]
For low-temperature Debye heat capacity (C_B(T)=aT^3), the zero-temperature scaling is
[
Q \ge \frac{3^{4/3}}{4}a^{1/3}S_L^{4/3}.
]
This removes the standard bound’s zero-temperature degeneracy while preserving the assumption that the bath is initially thermal [1911.00910].

At the level of full statistics, repeated interaction systems provide a trajectory-wise refinement. In that setting one has
[
\Delta s_{S,T}(\omega)+\varsigma_T(\omega)=\Delta s_{E,T}(\omega),
]
where (\varsigma_T) is the entropy production random variable, defined as the log-likelihood ratio between forward and backward two-time measurement protocols. In the adiabatic regime, (\varsigma_T) obeys a large deviation principle and a central limit theorem; in the special case (X(s)\equiv 0), its limiting distribution collapses to a Dirac mass at zero, so the Landauer bound is saturated in distribution [1705.08281].

6. Generalizations and applications across physics

Landauer entropy has been extended far beyond binary erasure in a thermal bath. In continuous phase transitions with binary order parameter, ordering can be interpreted as erasure with error probability (p), and the generalized Landauer bound implies a universal thermodynamic entropy
[
s(p)=\frac{S}{N}=k_B H_b(p),
\qquad
H_b(p)=-p\ln p-(1-p)\ln(1-p).
]
For Ising-like and Hopfield systems,
[
p=\frac{1-m}{2},
\qquad
S=N k_B H_b!\left(\frac{1-m}{2}\right),
]
so the remaining thermodynamic entropy in the ordered phase coincides with the information-theoretic error entropy [1403.5416].

For analog variables, the corresponding erasure cost is geometric rather than binary:
[
\Delta S_{\mathrm{erase}}=k_B\ln(\Omega/\Omega_{\min}),
\qquad
Q_{\min}=k_B T\ln(\Omega/\Omega_{\min}),
]
where (\Omega) is the configurational volume of the variable and (\Omega_{\min}) the minimal physically distinguishable quantum of volume. This implies that infinite precision would require infinite entropy production and is therefore forbidden by the fundamental laws of physics [1607.01704].

Alternative entropy functionals change the formal Landauer cost. In a two-state erasure model based on Tsallis entropy,
[

\Delta E_{\min}^{(q)}=\Delta Q_{\min}^{(q)}=k_B T\ln_q 2

k_B T\,\frac{1-2^{1-q}}{q-1},
]
recovering (k_B T\ln 2) as (q\to 1). For (01), it is cheaper [2411.07897]. This suggests that “Landauer entropy” is not always tied to a unique entropy functional, although such extensions depend on adopting a non-Boltzmann entropy definition.

In black-hole applications, the elementary entropy step is identified with the cost of erasing one bit, (\Delta S=k_B\ln 2). For the Bekenstein–Hawking entropy this reproduces the Bekenstein–Mukhanov area spacing,
[
\gamma=4\ln 2,
\qquad
\Delta A=4\ell_p^2\ln 2.
]
The same prescription organizes area spectra for Barrow, modified Rényi, and modified Kaniadakis entropies, with level-dependent parameters or branch singularities depending on the model [2605.26386]. A distinct geometric proposal defines the Landauer entropy of a static, spherically symmetric spacetime by
[
S(r):=\sqrt{|g_{tt}|}=f(r),
\qquad
S_L[\Sigma]=\frac{1}{4G\hbar}\int_\Sigma f(r)\,dA,
]
so that for spherical surfaces (S_L(r)=f(r)S_{BH}(r)), with (S_{BH}(r)=A(r)/(4G\hbar)) [2605.22172].

Landauer transport ideas also appear in horizon thermodynamics. Modeling Hawking radiation as a one-dimensional Landauer transport channel gives the entropy current
[
J_S=\frac{\pi^2 k_B^2}{3h}\,T_H,
]
the energy current
[
J_E=\frac{\pi^2 k_B^2}{6h}\,T_H^2,
]
and therefore
[
\frac{J_S}{J_E}=\frac{2}{T_H},
]
which is larger than the (4/(3T_H)) ratio of standard three-dimensional blackbody emission [1009.3974].

Finally, the quantum equality version has been tested experimentally. In a trapped-(^{40}\mathrm{Ca}^+) system, with the internal two-level system as memory and a quantized axial motional mode as finite bath, the measured quantity (\Delta Q/(k_B T)) almost perfectly overlaps (\Delta S + I(S':R') + D(\rho_R'|\rho_R)), while showing a pronounced disparity from (\Delta S) alone. The experiment thereby verifies that, in a finite quantum environment, Landauer entropy by itself does not close the thermodynamic balance; the correlation and relative-entropy terms are operationally necessary [1803.10424].

Landauer entropy is therefore best understood as the informational core of irreversible thermodynamic cost. In its narrowest form it is the entropy decrease of the information-bearing system. In its broader physical role it is the entropy that must be exported, at minimum, to uncontrolled degrees of freedom when that information is discarded. Equality-based quantum formulations, finite-size corrections, generalized inference bounds, and applications from neural networks to black-hole thermodynamics all preserve the same structural lesson: the thermodynamic price of information processing is controlled not only by how much entropy is removed from the system, but also by how correlations are created, how far the environment is driven from equilibrium, and how the relevant entropy itself is defined.