Mean-Force Gibbs State in Thermodynamics

Updated 5 July 2026

Mean-Force Gibbs State is the exact reduced equilibrium state of a subsystem, incorporating interaction-induced renormalization and bath-specific corrections.
It redefines thermodynamics by using an effective Hamiltonian that depends on temperature, coupling strength, and bath structure, diverging from the bare Gibbs state.
Analytical, perturbative, and numerical approaches reveal its role in preserving equilibrium coherences and capturing nontrivial boundary effects in extended systems.

The mean-force Gibbs state is the exact reduced equilibrium state of a subsystem that is coupled to an environment with non-negligible interaction energy. For a total Hamiltonian $H_{\mathrm{tot}}=H_S+H_B+H_{SB}$ , it is defined by reducing the global Gibbs state,

$\rho_S^{\mathrm{MF}}=\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{\operatorname{Tr}_{SB} e^{-\beta H_{\mathrm{tot}}}} =\frac{e^{-\beta H_{\mathrm{MF}}}}{Z_S^{\mathrm{MF}}}, \qquad H_{\mathrm{MF}}=-\frac{1}{\beta}\ln\!\left(\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{Z_B}\right),$

with $Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ fixing the additive normalization. In contrast to the ordinary Gibbs state $e^{-\beta H_S}/Z_S$ , the mean-force Gibbs state incorporates interaction-induced renormalization, temperature dependence, bath-specific structure, and, in general, equilibrium coherences and effective many-body terms. It reduces to the bare Gibbs description only in limiting regimes such as vanishing coupling and, in several formulations, infinite temperature (Burke et al., 2023, Trushechkin et al., 2021).

1. Formal structure and thermodynamic meaning

The Hamiltonian of mean force $H_S^*$ is not a microscopic Hamiltonian of the isolated system, but an effective equilibrium generator obtained after integrating out the bath. The reduced state is unique, whereas $H_S^*$ is defined only up to an additive scalar unless one chooses the convention

$e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$

With this normalization, the corresponding free energy of mean force is $F_S^*=-\beta^{-1}\ln Z_S^*=F_{\mathrm{tot}}-F_B$ , so the reduced equilibrium thermodynamics is expressed directly in terms of the interacting composite rather than the isolated subsystem. The deviation from the bare system Hamiltonian is commonly written as $\Delta H_S=H_S^*-H_S$ (Trushechkin et al., 2021, Hilt et al., 2011).

This formalism alters standard thermodynamic identities because $H_S^*$ generally depends on $\rho_S^{\mathrm{MF}}=\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{\operatorname{Tr}_{SB} e^{-\beta H_{\mathrm{tot}}}} =\frac{e^{-\beta H_{\mathrm{MF}}}}{Z_S^{\mathrm{MF}}}, \qquad H_{\mathrm{MF}}=-\frac{1}{\beta}\ln\!\left(\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{Z_B}\right),$ 0, the coupling strength, and bath properties. In the quasistatic switch-on picture discussed for damped quantum systems, the correction $\rho_S^{\mathrm{MF}}=\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{\operatorname{Tr}_{SB} e^{-\beta H_{\mathrm{tot}}}} =\frac{e^{-\beta H_{\mathrm{MF}}}}{Z_S^{\mathrm{MF}}}, \qquad H_{\mathrm{MF}}=-\frac{1}{\beta}\ln\!\left(\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{Z_B}\right),$ 1 modifies the standard Clausius form through

so $\rho_S^{\mathrm{MF}}=\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{\operatorname{Tr}_{SB} e^{-\beta H_{\mathrm{tot}}}} =\frac{e^{-\beta H_{\mathrm{MF}}}}{Z_S^{\mathrm{MF}}}, \qquad H_{\mathrm{MF}}=-\frac{1}{\beta}\ln\!\left(\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{Z_B}\right),$ 3 quantifies the energetic and entropic effect of finite system-bath coupling. In this sense, the mean-force Gibbs state is the natural equilibrium reference state of strong-coupling thermodynamics rather than a mere reparametrization of the bare Gibbs state (Hilt et al., 2011).

2. Weak-coupling and high-temperature structure

In weak coupling, the mean-force Gibbs state admits a perturbative expansion around the bare Gibbs state,

For bosonic reservoirs with linear coupling, the odd orders vanish, and the second-order correction is controlled by imaginary-time bath correlations and by the decomposition of the system coupling operator into Bohr-frequency components. The corresponding weak-coupling Hamiltonian of mean force is likewise

with explicit operator-valued corrections built from bath correlation functions $\rho_S^{\mathrm{MF}}=\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{\operatorname{Tr}_{SB} e^{-\beta H_{\mathrm{tot}}}} =\frac{e^{-\beta H_{\mathrm{MF}}}}{Z_S^{\mathrm{MF}}}, \qquad H_{\mathrm{MF}}=-\frac{1}{\beta}\ln\!\left(\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{Z_B}\right),$ 6. A central consequence is that the reduced equilibrium state need not commute with $\rho_S^{\mathrm{MF}}=\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{\operatorname{Tr}_{SB} e^{-\beta H_{\mathrm{tot}}}} =\frac{e^{-\beta H_{\mathrm{MF}}}}{Z_S^{\mathrm{MF}}}, \qquad H_{\mathrm{MF}}=-\frac{1}{\beta}\ln\!\left(\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{Z_B}\right),$ 7: equilibrium coherences can persist already at order $\rho_S^{\mathrm{MF}}=\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{\operatorname{Tr}_{SB} e^{-\beta H_{\mathrm{tot}}}} =\frac{e^{-\beta H_{\mathrm{MF}}}}{Z_S^{\mathrm{MF}}}, \qquad H_{\mathrm{MF}}=-\frac{1}{\beta}\ln\!\left(\frac{\operatorname{Tr}_B e^{-\beta H_{\mathrm{tot}}}}{Z_B}\right),$ 8, so bare-energy diagonality is not a generic property of finite-coupling equilibrium (Timofeev et al., 2022, Cresser et al., 2021).

At high temperature, the same object simplifies but does not usually revert to the bare Gibbs state unless renormalization is absent or has been compensated. For linear bosonic environments, the high-temperature mean-force Gibbs state takes the renormalized form

where $Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 0 is the reorganization energy. This expression clarifies the role of counter terms and Lamb shifts: in weak-coupling master equations, the secular GKLS generator relaxes to the Gibbs state of whatever Hamiltonian is inserted, whereas the correct high-temperature finite-coupling target is Gibbsian with respect to the renormalized Hamiltonian $Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 1. In the large-cutoff regime, the Lamb shift approximately cancels the coherent effect of the counter term, which explains why the common “shift the Hamiltonian and drop the Lamb shift” prescription can reproduce the correct classical/high- $Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 2 mean-force limit under the paper’s stated perturbative conditions (Correa et al., 2023).

3. Ultrastrong coupling and interaction-selected equilibrium

In the ultrastrong-coupling regime, the equilibrium structure changes qualitatively. For a general system coupled to a bosonic reservoir through an operator $Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 3, the mean-force Gibbs state tends to

$Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 4

The state is therefore diagonal in the eigenbasis of the interaction operator $Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 5, not in the eigenbasis of $Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 6. This is a strong-coupling, equilibrium analogue of pointer-basis selection: the interaction partitions the system Hilbert space, and the mean-force Hamiltonian becomes the projected Hamiltonian $Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 7 (Cresser et al., 2021).

This interaction-basis diagonality survives far beyond the harmonic Caldeira-Leggett model. For generalized Caldeira-Leggett environments with

$Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 8

the ultrastrong-coupling state remains block diagonal in the eigenspaces of $Z_B=\operatorname{Tr}_B e^{-\beta H_B}$ 9, but the block weights are modified by bath-dependent terms. In the generic GCL case,

$e^{-\beta H_S}/Z_S$ 0

where $e^{-\beta H_S}/Z_S$ 1 is determined by $e^{-\beta H_S}/Z_S$ 2. For the translationally structured subclass $e^{-\beta H_S}/Z_S$ 3, however, the Caldeira-Leggett result remains unchanged. In the Zwanzig model, the continuous-variable ultrastrong limit instead yields an effective Hamiltonian

$e^{-\beta H_S}/Z_S$ 4

showing that interaction-basis diagonality is robust while the detailed ultrastrong-coupling weights are model dependent (Kumar et al., 2024).

At very low temperature, the mean-force Hamiltonian also connects to entanglement-Hamiltonian physics. If the global Gibbs state collapses onto the ground-state projector as $e^{-\beta H_S}/Z_S$ 5, the reduced state becomes the reduced ground-state density matrix $e^{-\beta H_S}/Z_S$ 6, conventionally written as $e^{-\beta H_S}/Z_S$ 7. The comparison with the mean-force Gibbs form gives the asymptotic relation

$e^{-\beta H_S}/Z_S$ 8

so the mean-force Hamiltonian interpolates between high-temperature subsystem thermodynamics and zero-temperature entanglement structure (Burke et al., 2023).

4. Locality, quasi-locality, and the boundary “skin effect”

For extended systems with local interactions, the mean-force Hamiltonian is not arbitrarily delocalized. In spin chains with a subsystem $e^{-\beta H_S}/Z_S$ 9 coupled locally to an environment $H_S^*$ 0, the correction $H_S^*$ 1 is boundary-localized: when expanded in a Pauli-string basis $H_S^*$ 2, the coefficients

$H_S^*$ 3

decay exponentially with the distance of the operator support from the $H_S^*$ 4- $H_S^*$ 5 boundary,

$H_S^*$ 6

This “skin effect” implies that deep bulk observables are effectively thermal with respect to the bare subsystem Hamiltonian, even though the reduced state is exactly mean-force Gibbs. In the XXZ-chain analysis, the skin depth is well fitted by

$H_S^*$ 7

for small and moderate $H_S^*$ 8, and the corrections are traced to the locality constraints on building boundary-to-bulk operator strings (Burke et al., 2023).

The same work identifies detailed operator-content restrictions. In field-free XXZ chains, a Klein-four-group sign argument forbids single-body Pauli terms and mixed two-body terms $H_S^*$ 9 with $H_S^*$ 0 from appearing in the Hamiltonian of mean force at all, whereas local fields remove this restriction. For an $H_S^*$ 1-body Pauli operator $H_S^*$ 2 at distance $H_S^*$ 3, the first nonzero contribution appears no earlier than order

$H_S^*$ 4

so the coefficient of $H_S^*$ 5 starts at $H_S^*$ 6. This ties the spatial decay directly to a high-temperature expansion and to the necessity of constructing connected boundary-to-bulk strings of local terms (Burke et al., 2023).

An exact harmonic counterpart has been established for coupled oscillator networks. In harmonic rings or chains with only boundary oscillators attached to baths, the difference between mean-force and bare-Gibbs covariances decays exponentially with distance from the contact point, for example

$H_S^*$ 7

The effect is again strongest at the boundary and remains short-ranged even at strong coupling, showing that the boundary-localized dressing of equilibrium is not restricted to spin systems (Yeo et al., 2024).

5. Exact solutions, approximations, and numerical constructions

Several classes of systems permit either exact mean-force Gibbs states or controlled approximations. For the damped quantum harmonic oscillator, the reduced equilibrium state is exactly Gaussian, with Wigner function determined by $H_S^*$ 8 and $H_S^*$ 9. It can be rewritten as the thermal Wigner function of an effective oscillator

$e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 0

so that

$e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 1

The coefficients $e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 2 and $e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 3 are negative, the deviation from the bare Gibbs form is maximal at low temperature and strong coupling, and the overdamped semiclassical regime yields the approximate general-potential form

$e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 4

with the notation of that work (Hilt et al., 2011).

For continuous-variable systems beyond harmonicity, the local harmonic approximation provides an analytic construction of the mean-force Gibbs state for a particle in an arbitrary one-dimensional potential linearly coupled to a bosonic bath. The method expands the potential locally,

$e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 5

turning the path integral into a locally displaced damped harmonic problem. It is exact for harmonic potentials, accurate when the coupling or temperature is large or when higher derivatives are sufficiently small, and it recovers the ultra-strong-coupling and high-temperature limit

$e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 6

The same framework has been used for quartic oscillators, asymmetric double wells, and a DNA proton-tunneling model, where the equilibrium mutation probability inferred from the mean-force Gibbs state was argued to be much smaller than a previously reported nonequilibrium steady-state estimate (Kumar, 2024).

Nonperturbative numerical evaluation is possible through imaginary-time tensor-network methods. Imaginary-time TEMPO computes the unnormalized reduced operator

$e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 7

directly from an influence-functional path sum, and real-time TEMPO verifies that the long-time dynamics converges to the same reduced equilibrium state. In generalized spin-boson models, this approach demonstrates that the exact asymptotic state is the mean-force Gibbs state rather than the bare Gibbs state, and that the ultrastrong-coupling limit approaches the projected Gibbs state in the coupling-operator basis (Chiu et al., 2021).

A complementary comparison between perturbative mean-force expansions and reaction-coordinate thermal states yields simple empirical validity criteria. For the second-order perturbative mean-force Gibbs expansion, the practical condition reported is

$e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 8

while for the reaction-coordinate thermal steady-state approximation at high temperature the criterion is

$e^{-\beta H_S^*}=\frac{\operatorname{Tr}_B(e^{-\beta H_{\mathrm{tot}}})}{Z_B}, \qquad Z_S^*=\operatorname{Tr}_S(e^{-\beta H_S^*})=\frac{Z_{\mathrm{tot}}}{Z_B}.$ 9

These criteria are model dependent in origin but operationally useful for locating the regimes in which perturbative or embedding-based approximations remain reliable (Latune, 2021).

6. Dynamical equilibration, applications, and open issues

The static definition of the mean-force Gibbs state does not by itself guarantee dynamical convergence, so a central question is whether microscopic open-system dynamics relaxes to that state. In the rigorous return-to-equilibrium paradigm, sufficiently small coupling, suitable bath decay, and a Fermi Golden Rule condition imply convergence of the full interacting system to the global Gibbs state, hence convergence of reduced observables to the mean-force Gibbs state. However, the same review emphasizes that low-temperature rigor, infinite-dimensional systems, intermediate coupling, and general all-orders constructions remain open problems (Trushechkin et al., 2021).

Approximate master equations do not all reproduce the same equilibrium target. Secular weak-coupling GKLS equations relax to the Gibbs state of the Hamiltonian used in their construction, whereas nonsecular Bloch-Redfield theory captures finite-coupling corrections only perturbatively. For the generic spin-boson model, the time-convolutionless fourth-order master equation resolves a key limitation: a perturbative generator at order $F_S^*=-\beta^{-1}\ln Z_S^*=F_{\mathrm{tot}}-F_B$ 0 determines coherences through $F_S^*=-\beta^{-1}\ln Z_S^*=F_{\mathrm{tot}}-F_B$ 1 but populations only through $F_S^*=-\beta^{-1}\ln Z_S^*=F_{\mathrm{tot}}-F_B$ 2. Accordingly, TCL2/Bloch-Redfield gives the second-order coherence correction but misses the second-order population shift, while TCL4 yields a steady state whose full $F_S^*=-\beta^{-1}\ln Z_S^*=F_{\mathrm{tot}}-F_B$ 3 expansion is exactly identical to the corresponding mean-force Gibbs state at arbitrary temperature (Kumar et al., 2024).

This dynamical issue underlies a persistent practical controversy about counter terms and Lamb shifts. One strand of work shows that the standard secular equation built from the physical Hamiltonian may converge to the wrong equilibrium whenever finite coupling produces a non-negligible renormalization, while a “shift-and-drop-Lamb-shift” recipe can recover the correct classical/high-temperature mean-force target if the reorganization energy is small but non-negligible and the bath cutoff is large. The central point is not that all weak-coupling equations converge to the mean-force Gibbs state, but that the choice of coherent corrections determines whether the stationary state is the bare Gibbs state or its finite-coupling renormalized counterpart (Correa et al., 2023).

Applications of the mean-force Gibbs state extend beyond equilibrium populations. In two-qubit common-bath models, the Hamiltonian of mean force acquires bath-mediated two-qubit terms even though the bare system Hamiltonian contains no direct interaction. Analytic approximations then show that the equilibrium reduced state can become entangled, with negativity strongest at low temperature, non-monotonic in coupling strength, and, for the structured spectral densities studied there, enhanced by moderate reservoir broadening. The resulting picture is that strong system-reservoir coupling can generate equilibrium correlations rather than merely destroy them, but the precise balance between induced interactions and loss of subsystem purity is highly regime dependent (Williamson et al., 29 Apr 2026).