This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Structure of Quantum Mean Force Gibbs States for Coupled Harmonic Systems

Joonhyun Yeo Department of Physics, Konkuk University, Seoul 05029, Korea    Haena Shim Department of Physics, Konkuk University, Seoul 05029, Korea
Abstract

An open quantum system interacting with a heat bath at given temperature is expected to reach the mean force Gibbs (MFG) state as a steady state. The MFG state is given by tracing out the bath degrees of freedom from the equilibrium Gibbs state of the total system plus bath. When the interaction between the system and the bath is not negligible, it is different from the usual system Gibbs state obtained from the system Hamiltonian only. Using the path integral method, we present the exact MFG state for a coupled system of quantum harmonic oscillators in contact with multiple thermal baths at the same temperature. We develop a nonperturbative method to calculate the covariances with respect to the MFG state. By comparing them with those obtained from the system Gibbs state, we find that the effect of coupling to the bath decays exponentially as a function of the distance from the system-bath boundary. This is similar to the skin effect found recently for a quantum spin chain interacting with an environment. Using the exact results, we also investigate the ultrastrong coupling limit where the coupling between the system and the bath gets arbitrarily large and make a connection with the recent result found for a general quantum system.

I Introduction

In the standard theory of thermodynamics the interaction between a system and its environment is assumed to be weak. However, for systems with decreasing sizes Jarzynski (2017) or for quantum mechanical systems Ford et al. (1985), the interaction cannot be neglected. This consideration has led to the recent development of strong coupling thermodynamics Campisi et al. (2009); Gelin and Thoss (2009); Subaşı et al. (2012); Seifert (2016); Jarzynski (2017); Strasberg and Esposito (2017); Aurell (2017); Miller (2018); Hsiang and Hu (2018); Perarnau-Llobet et al. (2018); Talkner and Hänggi (2020); Rivas (2020); Anto-Sztrikacs et al. (2023); Kaneyasu and Hasegawa (2023); Diba et al. (2024). The mean force Gibbs (MFG) state Trushechkin et al. (2022); Cresser and Anders (2021); Chiu et al. (2022) and the associated Hamiltonian of mean force Kirkwood (1935); Grabert et al. (1984); Hilt et al. (2011); Miller (2018); Timofeev and Trushechkin (2022) play a central role in the theory of strong coupling thermodynamics. The MFG state describes the equilibrium state of the system interacting with a thermal bath at given temperature. When the interaction between the system and the bath is not negligible, the MFG state deviates from the usual system Gibbs state which is described by the system Hamiltonian only Trushechkin et al. (2022), and the Hamiltonian of mean force is different from the system Hamiltonian Hilt et al. (2011).

Despite the importance in strong coupling thermodynamics, explicit evaluation of the quantum MFG state turns out to be quite difficult. Exact results are known only for simple limited cases Grabert et al. (1984); Hilt et al. (2011) of a damped harmonic oscillator described by the Caldeira-Leggett model Caldeira and Leggett (1983). Recently, the MFG state for a general quantum system interacting with a bosonic environment was obtained in both weak and ultrastrong coupling limits Cresser and Anders (2021). The ultrastrong coupling limit of the MFG state obtained in Ref. Cresser and Anders (2021) is quite interesting as it is characterized by a diagonal projection of the system Hamiltonian onto the eigenstates of the system operator which couples to the environment. In order to gain further understanding of quantum thermodynamics beyond weak coupling, it would be useful to have explicit expressions of the MFG state of a more general quantum system which covers the range of the coupling strength from weak to ultrastrong. One of the purposes of this paper is to provide such examples of the quantum MFG state.

Another aspect of the MFG state which has not been investigated as yet is its spatial structure. For an open quantum system with spatial extent, only a part of the system is directly coupled to the heat bath. Therefore, the MFG state as an equilibrium state is expected to exhibit a spatial structure reflecting this local coupling. Recently, in Ref. Burke et al. (2024), the Hamiltonian of mean force of a spin chain interacting with its environment was studied numerically. The results show a kind of skin effect where the effect of the coupling to the environment decays exponentially as one moves away from the system-environment boundary. This suggests that the effect of the local coupling to the environment on the equilibrium state of the system is confined to the vicinity of the point of contact with the environment.

In this paper, we present calculations leading to the MFG state of a coupled system of quantum mechanical harmonic oscillators interacting with multiple thermal baths maintained at given temperature. This is an exact result which covers the range of the coupling strength to the heat baths from weak to ultrastrong. We consider an open quantum system of coupled oscillators described by the Caldeira-Leggett type Hamiltonian Caldeira and Leggett (1983) where only parts of the system are coupled with the baths. By explicitly evaluating the path integrals involved in tracing out the bath degrees of freedom, we present a formalism of obtaining the exact MFG state. Being a Gaussian state, the MFG state is completely described by its covariances. We present a nonperturbative method to evaluate the covariances with respect to the MFG state. By studying the difference in the covariances obtained with respect to the MFG and the system Gibbs states, we show how the local nature of the coupling to the bath shows up in the MFG state. We find that there is a skin effect similar to that found in Ref. Burke et al. (2024), where the effect of the coupling to the heat bath decays exponentially as a function of the distance from the system-bath boundary. Using the exact expressions, we also explore the ultrastrong coupling regime. We find that the expression found in Ref. Cresser and Anders (2021) for this limit is still valid even in the present case where the coupling system operator is local and has a continuous spectrum.

In the following section, we present a general formalism for the calculation of the MFG state of coupled harmonic systems using the path integral method. In Sec. III, we focus on the case where the system is in contact with a single heat bath. Using a specific example of a ring of coupled oscillators, we present detailed expressions for the exact MFG covariances and study their ultrastrong coupling limit. We then generalize to the case where the system is in contact with two or more baths at the same temperature in Sec. IV. Finally. we conclude with discussion.

II MFG state for coupled harmonic systems from path integral

An open quantum system in contact with a thermal bath is expected to approach in the long time limit its stationary state described by the MFG state Subaşı et al. (2012). We consider a quantum mechanical system described by the Hamiltonian HSH_{\rm S} interacting with the bath given by HBH_{\rm B}. The total Hamiltonian is written as

H^=H^S+H^B+H^I,\displaystyle\hat{H}=\hat{H}_{\rm S}+\hat{H}_{\rm B}+\hat{H}_{\rm I}, (1)

where the interaction between the system and the bath is governed by HIH_{\rm I}. The MFG state is defined by

ρ^mfG=TrBρ^β1ZβTrB(eβH^),\displaystyle\hat{\rho}_{\rm mfG}=\mathrm{Tr_{B}}\hat{\rho}_{\beta}\equiv\frac{1}{Z_{\beta}}{\rm Tr}_{\rm B}(e^{-\beta\hat{H}}), (2)

which is obtained by the partial trace of the equilibrium state of the total system ρ^β\hat{\rho}_{\beta} with ZβTrSBexp(βH)Z_{\beta}\equiv\mathrm{Tr_{SB}}\exp(-\beta H). When the interaction between the system and bath is not negligible, the MFG state is in general different from the usual system Gibbs state

ρ^G=1TrSeβH^SeβH^S\displaystyle\hat{\rho}_{\rm G}=\frac{1}{\mathrm{Tr_{S}}e^{-\beta\hat{H}_{\rm S}}}e^{-\beta\hat{H}_{\rm S}} (3)

given by the system Hamiltonian only.

We consider a collection of NN quantum harmonic oscillators of equal mass MM. The coupling among these oscillators is described by a symmetric positive definite matrix 𝚲\bm{\Lambda}. The system is described by the Hamiltonian

H^S=i=0N1p^i22M+12i,j=0N1Λijx^ix^j,\displaystyle\hat{H}_{\rm S}=\sum_{i=0}^{N-1}\frac{\hat{p}_{i}^{2}}{2M}+\frac{1}{2}\sum_{i,j=0}^{N-1}\Lambda_{ij}\hat{x}_{i}\hat{x}_{j}, (4)

where x^i\hat{x}_{i} and p^i\hat{p}_{i} are the position and momentum operators of the ii-th oscillator, respectively.

We consider the case where only a part of the system is in contact with a collection of heat baths labelled by α\alpha. We set up a Caldeira-Leggett model Caldeira and Leggett (1983) where each bath is composed of harmonic oscillators with mass mn,αm_{n,\alpha} and frequency ωn,α\omega_{n,\alpha}. The Hamiltonian for the heat baths and their interaction with the system can be written as

H^B+H^I\displaystyle\hat{H}_{\rm B}+\hat{H}_{\rm I} (5)
=\displaystyle= αn[π^n,α22mn,α+12mn,αωn,α2(q^n,ακn,αmn,αωn,α2x^α)2],\displaystyle\sum_{\alpha\in\mathcal{B}}\sum_{n}\left[\frac{\hat{\pi}^{2}_{n,\alpha}}{2m_{n,\alpha}}+\frac{1}{2}m_{n,\alpha}\omega^{2}_{n,\alpha}\left(\hat{q}_{n,\alpha}-\frac{\kappa_{n,\alpha}}{m_{n,\alpha}\omega^{2}_{n,\alpha}}\hat{x}_{\alpha}\right)^{2}\right],

where π^n,α\hat{\pi}_{n,\alpha} and q^n,α\hat{q}_{n,\alpha} are the momentum and position operators for the oscillators in the bath α\alpha, respectively. Here the index α\alpha runs through a subset \mathcal{B} of {0,1,,N1}\{0,1,\cdots,N-1\} such that the oscillator described by the position x^α\hat{x}_{\alpha} is coupled with the oscillators in the α\alpha-th bath with the coupling constant κn,α\kappa_{n,\alpha}. The distribution of the α\alpha-th bath modes is conveniently described by the spectral function

Jα(ω)=nκn,α22mn,αωn,αδ(ωωn,α).\displaystyle J_{\alpha}(\omega)=\sum_{n}\frac{\kappa^{2}_{n,\alpha}}{2m_{n,\alpha}\omega_{n,\alpha}}\delta(\omega-\omega_{n,\alpha}). (6)

We first write the matrix element of ρβ\rho_{\beta} using Euclidian path integrals as

𝒙¯,{𝒒¯α}|ρ^β|𝒙¯,{𝒒¯α}\displaystyle\langle\bm{\bar{x}},\{\bm{\bar{q}}_{\alpha}\}|\hat{\rho}_{\beta}|\bm{\bar{x}}^{\prime},\{\bm{\bar{q}}^{\prime}_{\alpha}\}\rangle
=\displaystyle= 1Zβ𝒙(0)=𝒙¯𝒙(β)=𝒙¯𝒟𝒙(τ)α𝒒α(0)=𝒒¯α𝒒α(β)=𝒒¯α𝒟𝒒α(τ)\displaystyle\frac{1}{Z_{\beta}}\int_{\bm{x}(0)=\bm{\bar{x}}^{\prime}}^{\bm{x}(\beta\hbar)=\bm{\bar{x}}}\mathcal{D}\bm{x}(\tau)\prod_{\alpha\in\mathcal{B}}\int_{\bm{q}_{\alpha}(0)=\bm{\bar{q}}^{\prime}_{\alpha}}^{\bm{q}_{\alpha}(\beta\hbar)=\bm{\bar{q}}_{\alpha}}\mathcal{D}\bm{q}_{\alpha}(\tau)\
×exp[1(SS[𝒙]+SB[{𝒒α}]+SI[𝒙,{𝒒α}])],\displaystyle\times\exp\left[-\frac{1}{\hbar}\left(S_{\rm S}[\bm{x}]+S_{\rm B}[\{\bm{q}_{\alpha}\}]+S_{\rm I}[\bm{x},\{\bm{q}_{\alpha}\}]\right)\right], (7)

where the system Euclidian action is given by

SS[𝒙]=0β𝑑τ\displaystyle S_{\rm S}[\bm{x}]=\int_{0}^{\beta\hbar}d\tau [M2d𝒙T(τ)dτd𝒙(τ)dτ+12𝒙T(τ)𝚲𝒙(τ)]\displaystyle\;\Big{[}\frac{M}{2}\frac{d\bm{x}^{\rm T}(\tau)}{d\tau}\frac{d\bm{x}(\tau)}{d\tau}+\frac{1}{2}\bm{x}^{\rm T}(\tau)\bm{\Lambda}\bm{x}(\tau)\Big{]} (8)

with the matrix notation 𝒙T=(x0,x1,,xN1)\bm{x}^{\rm T}=(x_{0},x_{1},\cdots,x_{N-1}) and 𝒒αT=(q1,α,q2,α,)\bm{q}_{\alpha}^{\rm T}=(q_{1,\alpha},q_{2,\alpha},\cdots). The remaining actions for the bath and the interaction are given respectively by

SB[{𝒒α}]=αn\displaystyle S_{\rm B}[\{\bm{q}_{\alpha}\}]=\sum_{\alpha\in\mathcal{B}}\sum_{n} 0βdτ[mn,α2(dqn,αdτ)2\displaystyle\int_{0}^{\beta\hbar}d\tau\;\Big{[}\frac{m_{n,\alpha}}{2}\left(\frac{dq_{n,\alpha}}{d\tau}\right)^{2}
+12mn,αωn,α2qn,α2(τ)],\displaystyle+\frac{1}{2}m_{n,\alpha}\omega^{2}_{n,\alpha}q^{2}_{n,\alpha}(\tau)\Big{]}, (9)

and

SI[𝒙,{𝒒α}]\displaystyle S_{\rm I}[\bm{x},\{\bm{q}_{\alpha}\}] (10)
=α0β𝑑τ[nκn,αqn,α(s)xα(s)+μα2xα2(τ)]\displaystyle=\sum_{\alpha\in\mathcal{B}}\int_{0}^{\beta\hbar}d\tau\;\left[-\sum_{n}\kappa_{n,\alpha}q_{n,\alpha}(s)x_{\alpha}(s)+\frac{\mu_{\alpha}}{2}x_{\alpha}^{2}(\tau)\right]

with

μαnκn,α2mn,αωn,α2=20𝑑ωJα(ω)ω.\displaystyle\mu_{\alpha}\equiv\sum_{n}\frac{\kappa^{2}_{n,\alpha}}{m_{n,\alpha}\omega^{2}_{n,\alpha}}=2\int_{0}^{\infty}d\omega\;\frac{J_{\alpha}(\omega)}{\omega}. (11)

The matrix element of the MFG state is obtained by tracing over the bath variables. This type of path integral is well-known in the study of the quantum Brownian motion Grabert et al. (1988). The path integral over the bath variables results in the so-called Feynman-Vernon influence functional Feynman and Vernon (1963); Grabert et al. (1988); Weiss (2012). In our case, we can easily modify it to the imaginary time path integral. As a result, we have the influence functional obtained by tracing out the α\alpha-th bath as

Ψα[xα]=\displaystyle\Psi_{\alpha}[x_{\alpha}]= 0β𝑑τ0τ𝑑τKα(ττ)xα(τ)xα(τ)\displaystyle-\int_{0}^{\beta\hbar}d\tau\int_{0}^{\tau}d\tau^{\prime}\;K_{\alpha}(\tau-\tau^{\prime})x_{\alpha}(\tau)x_{\alpha}(\tau^{\prime})
+μα20β𝑑τxα2(τ),\displaystyle+\frac{\mu_{\alpha}}{2}\int_{0}^{\beta\hbar}d\tau\;x_{\alpha}^{2}(\tau), (12)

where

Kα(τ)0𝑑ωJα(ω)cosh(12βωωτ)sinh(12βω).\displaystyle K_{\alpha}(\tau)\equiv\int_{0}^{\infty}d\omega\;J_{\alpha}(\omega)\frac{\cosh(\frac{1}{2}\beta\hbar\omega-\omega\tau)}{\sinh(\frac{1}{2}\beta\hbar\omega)}. (13)

Now the matrix element of the MFG state with respect to the system position variables is given by

𝒙¯|ρ^mfG|𝒙¯=αndq¯n,α𝒙¯,{𝒒¯α}|ρ^β|𝒙¯,{𝒒¯α}\displaystyle\langle\bm{\bar{x}}|\hat{\rho}_{\rm mfG}|\bm{\bar{x}}^{\prime}\rangle=\int\prod_{\alpha\in\mathcal{B}}\prod_{n}d\bar{q}_{n,\alpha}\;\langle\bm{\bar{x}},\{\bm{\bar{q}}_{\alpha}\}|\hat{\rho}_{\beta}|\bm{\bar{x}}^{\prime},\{\bm{\bar{q}}_{\alpha}\}\rangle
=\displaystyle= 1Z𝒙(0)=𝒙¯𝒙(β)=𝒙¯𝒟𝒙(τ)exp[1(SS[𝒙]+αΨα[xα])],\displaystyle\frac{1}{Z}\int_{\bm{x}(0)=\bm{\bar{x}^{\prime}}}^{\bm{x}(\beta\hbar)=\bm{\bar{x}}}\mathcal{D}\bm{x}(\tau)\exp\left[-\frac{1}{\hbar}(S_{\rm S}[\bm{x}]+\sum_{\alpha\in\mathcal{B}}\Psi_{\alpha}[x_{\alpha}])\right], (14)

where Z=ZB/ZβZ=Z_{\rm B}/Z_{\beta} with ZB=TrBexp(βHB)Z_{\rm B}=\mathrm{Tr}_{\rm B}\exp(-\beta H_{\rm B}). We note that we can rewrite Eq. (12) as (see Appendix A)

Ψα[xα]=M20β𝑑τ0β𝑑τLα(ττ)xα(τ)xα(τ),\displaystyle\Psi_{\alpha}[x_{\alpha}]=\frac{M}{2}\int_{0}^{\beta\hbar}d\tau\int_{0}^{\beta\hbar}d\tau^{\prime}\;L_{\alpha}(\tau-\tau^{\prime})x_{\alpha}(\tau)x_{\alpha}(\tau^{\prime}), (15)

where Lα(τ)L_{\alpha}(\tau) is defined as a Fourier series expansion in the interval 0τβ0\leq\tau\leq\beta\hbar as

Lα(τ)=1βr=ζr(α)eiνrτ.\displaystyle L_{\alpha}(\tau)=\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}\zeta^{(\alpha)}_{r}e^{i\nu_{r}\tau}. (16)

with the Matsubara frequency νr=2πr/(β)\nu_{r}=2\pi r/(\beta\hbar) and

ζr(α)=1M0𝑑ωJα(ω)ω2νr2ω2+νr2.\displaystyle\zeta^{(\alpha)}_{r}=\frac{1}{M}\int_{0}^{\infty}d\omega\;\frac{J_{\alpha}(\omega)}{\omega}\frac{2\nu^{2}_{r}}{\omega^{2}+\nu^{2}_{r}}. (17)

Since the path integral in Eq. (14) is Gaussian, it can be evaluated by considering the stationary path and the fluctuation around it. The path integral over the fluctuation will give a constant independent of the endpoints, which can be determined later from the normalization of ρ^mfG\hat{\rho}_{\rm mfG} Grabert et al. (1988). The stationary path is determined from

0=\displaystyle 0= Md2xi(τ)dτ2+j=0N1Λijxj(τ)\displaystyle-M\frac{d^{2}x_{i}(\tau)}{d\tau^{2}}+\sum_{j=0}^{N-1}\Lambda_{ij}x_{j}(\tau)
+Mα0β𝑑τLα(ττ)xα(τ).\displaystyle+M\sum_{\alpha\in\mathcal{B}}\int_{0}^{\beta\hbar}d\tau^{\prime}\;L_{\alpha}(\tau-\tau^{\prime})x_{\alpha}(\tau^{\prime}). (18)

for i=0,1,,N1i=0,1,\cdots,N-1. This is to be solved with the boundary conditions 𝒙(0)=𝒙¯\bm{x}(0)=\bm{\bar{x}^{\prime}} and 𝒙(β)=𝒙¯\bm{x}(\beta\hbar)=\bm{\bar{x}}. The solution can be conveniently obtained in terms of the Fourier series expansion

xi(τ)=1βr=xi,reiνrτ.\displaystyle x_{i}(\tau)=\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}x_{i,r}e^{i\nu_{r}\tau}. (19)

The detailed procedure is given in Appendix B. The solution for the Fourier modes xi,rx_{i,r} in a matrix form are given by

𝒙r=iνr𝑮(νr)(𝒙¯𝒙¯)+12𝑮(νr)𝑭1(𝒙¯+𝒙¯),\displaystyle\bm{x}_{r}=i\nu_{r}\bm{G}(\nu_{r})(\bm{\bar{x}}-\bm{\bar{x}^{\prime}})+\frac{1}{2}\bm{G}(\nu_{r})\bm{F}^{-1}\left(\bm{\bar{x}}+\bm{\bar{x}^{\prime}}\right), (20)

where the matrices

𝑮(νr)[νr2𝟏+1M𝚲+𝚺(νr)]1,\displaystyle\bm{G}(\nu_{r})\equiv\left[\nu_{r}^{2}\bm{1}+\frac{1}{M}\bm{\Lambda}+\bm{\Sigma}(\nu_{r})\right]^{-1}, (21)

and

𝑭1βr=𝑮(νr).\displaystyle\bm{F}\equiv\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}\bm{G}(\nu_{r}). (22)

Here 𝚺(νr)\bm{\Sigma}(\nu_{r}) is a diagonal matrix whose diagonal element Σjj\Sigma_{jj} is equal to ζr(α)\zeta_{r}^{(\alpha)} given in Eq. (17) if jj belongs to the set \mathcal{B} and is equal to zero otherwise.

If we insert the solution Eq. (20) with Eq. (19) into the expression of the MFG state, Eq. (14), we obtain (see Appendix C)

𝒙¯|ρ^mfG|𝒙¯=Cexp[M2(𝒙¯𝒙¯)T𝑨(𝒙¯𝒙¯)\displaystyle\langle\bm{\bar{x}}|\hat{\rho}_{\rm mfG}|\bm{\bar{x}^{\prime}}\rangle=C\exp\Bigg{[}-\frac{M}{2\hbar}(\bm{\bar{x}}-\bm{\bar{x}^{\prime}})^{\rm T}\bm{A}\;(\bm{\bar{x}}-\bm{\bar{x}^{\prime}})
M2(𝒙¯+𝒙¯2)T𝑭1(𝒙¯+𝒙¯2)],\displaystyle~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}-\frac{M}{2\hbar}\left(\frac{\bm{\bar{x}}+\bm{\bar{x}^{\prime}}}{2}\right)^{\rm T}\bm{F}^{-1}\;\left(\frac{\bm{\bar{x}}+\bm{\bar{x}^{\prime}}}{2}\right)\Bigg{]}, (23)

where

𝑨1βr=[𝟏νr2𝑮(νr)].\displaystyle\bm{A}\equiv\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}\left[\bm{1}-\nu^{2}_{r}\bm{G}(\nu_{r})\right]. (24)

The overall constant CC is determined from the normalization

1=dN𝒙¯𝒙¯|ρmfG|𝒙¯,\displaystyle 1=\int d^{N}\bm{\bar{x}}\;\langle\bm{\bar{x}}|\rho_{\rm mfG}|\bm{\bar{x}}\rangle, (25)

which gives

C=(M2π)N21(det𝑭)12\displaystyle C=\left(\frac{M}{2\pi\hbar}\right)^{\frac{N}{2}}\frac{1}{(\det\bm{F})^{\frac{1}{2}}} (26)

The MFG state is Gaussian and characterized by the following covariance matrices, which can easily be calculated as

x^ix^jmfG=Tr[x^ix^jρ^mfG]=MFij,\displaystyle\langle\hat{x}_{i}\hat{x}_{j}\rangle_{\rm mfG}=\mathrm{Tr}[\hat{x}_{i}\hat{x}_{j}\hat{\rho}_{\rm mfG}]=\frac{\hbar}{M}F_{ij}, (27)
p^ip^jmfG=Tr[p^ip^jρ^mfG]=MAij,\displaystyle\langle\hat{p}_{i}\hat{p}_{j}\rangle_{\rm mfG}=\mathrm{Tr}[\hat{p}_{i}\hat{p}_{j}\hat{\rho}_{\rm mfG}]=M\hbar A_{ij}, (28)

and

{x^i,p^j}mfG=Tr[(x^ip^j+p^jx^i)ρ^mfG]=0.\displaystyle\langle\{\hat{x}_{i},\hat{p}_{j}\}\rangle_{\rm mfG}=\mathrm{Tr}[(\hat{x}_{i}\hat{p}_{j}+\hat{p}_{j}\hat{x}_{i})\hat{\rho}_{\rm mfG}]=0. (29)

III System in Contact with a single heat bath

In this section, using the general formalism developed above, we develop a nonperturbative method to evaluate matrix Green’s function in Eq. (21). To do this, we first focus on coupled harmonic oscillators one end of which is in contact with a heat bath at inverse temperature β\beta. In the notation of Eq. (5), we take ={0}\mathcal{B}=\{0\}, that is the α=0\alpha=0 oscillator is interacting with the heat bath. Then the only non-vanishing matrix element of 𝚺(νr)\bm{\Sigma}(\nu_{r}) which appears in Eq. (21) is given by Σ00(νr)=ζr\Sigma_{00}(\nu_{r})=\zeta_{r}. Here we have dropped the index α\alpha in the definition of ζr\zeta_{r} in Eq. (17) as there is only one bath.

We note that the calculation of the MFG state comes down to evaluating the Green’s function 𝑮\bm{G} in Eq. (21). A convenient way to do this to express 𝑮\bm{G} in terms of 𝑮(0)(νr)[νr2𝟏+𝚲/M]1\bm{G}^{(0)}(\nu_{r})\equiv[\nu^{2}_{r}\bm{1}+\bm{\Lambda}/M]^{-1}, which is the Green’s function in the absence of coupling to the bath. If we use 𝑮(0)\bm{G}^{(0)} in Eq. (23), then we can get the matrix element of the system Gibbs state ρ^G\hat{\rho}_{\rm G}. We note that 𝑮(0)\bm{G}^{(0)} can be obtained by diagonalizing 𝚲\bm{\Lambda}. Once we have 𝑮(0)\bm{G}^{(0)}, we can write

𝑮(νr)=\displaystyle\bm{G}(\nu_{r})= [(𝑮(0))1(νr)+𝚺(νr)]1\displaystyle\left[(\bm{G}^{(0)})^{-1}(\nu_{r})+\bm{\Sigma}(\nu_{r})\right]^{-1}
=\displaystyle= [𝟏+𝑮(0)𝚺]1𝑮(0)\displaystyle\left[\bm{1}+\bm{G}^{(0)}\bm{\Sigma}\right]^{-1}\bm{G}^{(0)} (30)
=\displaystyle= [𝟏𝑮(0)𝚺+𝑮(0)𝚺𝑮(0)𝚺+]𝑮(0).\displaystyle\left[\bm{1}-\bm{G}^{(0)}\bm{\Sigma}+\bm{G}^{(0)}\bm{\Sigma}\bm{G}^{(0)}\bm{\Sigma}+\cdots\right]\bm{G}^{(0)}.

Since the only non-vanishing element of 𝚺\bm{\Sigma} is Σ00=ζr\Sigma_{00}=\zeta_{r}, we can write

Gij(νr)=\displaystyle G_{ij}(\nu_{r})= Gij(0)ζrGi0(0)G0j(0)+ζr2Gi0(0)G00(0)G0j(0)\displaystyle G^{(0)}_{ij}-\zeta_{r}G^{(0)}_{i0}G^{(0)}_{0j}+\zeta^{2}_{r}G^{(0)}_{i0}G^{(0)}_{00}G^{(0)}_{0j}
ζr3Gi0(0)[G00(0)]2G0j(0)+.\displaystyle-\zeta^{3}_{r}G^{(0)}_{i0}[G^{(0)}_{00}]^{2}G^{(0)}_{0j}+\cdots. (31)

By summing the infinite series, we thus have

𝑮(νr)=𝑮(0)(νr)+Δ𝑮(νr),\displaystyle\bm{G}(\nu_{r})=\bm{G}^{(0)}(\nu_{r})+\Delta\bm{G}(\nu_{r}), (32)

where

ΔGij(νr)ζrGi0(0)(νr)G0j(0)(νr)1+ζrG00(0)(νr).\displaystyle\Delta G_{ij}(\nu_{r})\equiv-\zeta_{r}\frac{G^{(0)}_{i0}(\nu_{r})G^{(0)}_{0j}(\nu_{r})}{1+\zeta_{r}G^{(0)}_{00}(\nu_{r})}. (33)

For later purpose, we rewrite Eq. (33) as

Δ𝑮(νr)=𝑮(0)(νr)𝚪(νr)𝑮(0)(νr)\displaystyle\Delta\bm{G}(\nu_{r})=-\bm{G}^{(0)}(\nu_{r})\bm{\Gamma}(\nu_{r})\bm{G}^{(0)}(\nu_{r}) (34)

using the matrix 𝚪\bm{\Gamma} with only one nonvanishing element defined by

Γij(νr)=δi0δj0ζr1+ζrG00(0)(νr).\displaystyle\Gamma_{ij}(\nu_{r})=\delta_{i0}\delta_{j0}\frac{\zeta_{r}}{1+\zeta_{r}G^{(0)}_{00}(\nu_{r})}. (35)

Another way of writing the full Green’s function is

Gij(νr)\displaystyle G_{ij}(\nu_{r}) (36)
=Gij(0)(νr)+ζr{Gij(0)(νr)G00(0)(νr)Gi0(0)(νr)G0j(0)(νr)}1+ζrG00(0)(νr).\displaystyle=\frac{G^{(0)}_{ij}(\nu_{r})+\zeta_{r}\{G^{(0)}_{ij}(\nu_{r})G^{(0)}_{00}(\nu_{r})-G^{(0)}_{i0}(\nu_{r})G^{(0)}_{0j}(\nu_{r})\}}{1+\zeta_{r}G^{(0)}_{00}(\nu_{r})}.

We can then use this equation to obtain the expressions for the covariances with respect to the MFG state in Eqs (27) and (28) as

x^ix^jmfG=1Mβ[Gij(0)(0)+2r=1Gij(0)(νr)+ζr{Gij(0)(νr)G00(0)(νr)Gi0(0)(νr)G0j(0)(νr)}1+ζrG00(0)(νr)]\displaystyle\langle\hat{x}_{i}\hat{x}_{j}\rangle_{\rm mfG}=\frac{1}{M\beta}\left[G^{(0)}_{ij}(0)+2\sum_{r=1}^{\infty}\frac{G^{(0)}_{ij}(\nu_{r})+\zeta_{r}\{G^{(0)}_{ij}(\nu_{r})G^{(0)}_{00}(\nu_{r})-G^{(0)}_{i0}(\nu_{r})G^{(0)}_{0j}(\nu_{r})\}}{1+\zeta_{r}G^{(0)}_{00}(\nu_{r})}\right] (37)

and

p^ip^jmfG=Mβ[δij+2r=1{δijνr2Gij(0)(νr)+νr2ζr{Gij(0)(νr)G00(0)(νr)Gi0(0)(νr)G0j(0)(νr)}1+ζrG00(0)(νr)}],\displaystyle\langle\hat{p}_{i}\hat{p}_{j}\rangle_{\rm mfG}=\frac{M}{\beta}\left[\delta_{ij}+2\sum_{r=1}^{\infty}\left\{\delta_{ij}-\frac{\nu^{2}_{r}G^{(0)}_{ij}(\nu_{r})+\nu^{2}_{r}\zeta_{r}\{G^{(0)}_{ij}(\nu_{r})G^{(0)}_{00}(\nu_{r})-G^{(0)}_{i0}(\nu_{r})G^{(0)}_{0j}(\nu_{r})\}}{1+\zeta_{r}G^{(0)}_{00}(\nu_{r})}\right\}\right], (38)

where the summation is now over positive frequencies νr\nu_{r}.

III.1 Example: Chain of Harmonic Oscillators in Contact with a Heat Bath

In this subsection, we apply the method developed in the previous section to a chain of coupled oscillators one end of which is in contact with a heat bath. We consider a system described by Hamiltonian

H^S=i=0N1(p^i22M+12MΩ2x^i2Mλx^ix^i+1).\displaystyle\hat{H}_{\rm S}=\sum_{i=0}^{N-1}\left(\frac{\hat{p}_{i}^{2}}{2M}+\frac{1}{2}M\Omega^{2}\hat{x}^{2}_{i}-M\lambda\hat{x}_{i}\hat{x}_{i+1}\right). (39)

The parameter λ\lambda describes the coupling between the neighboring oscillators. For convenience, we assume that the system satisfies the periodic boundary condition, x^N=x^0\hat{x}_{N}=\hat{x}_{0}. The oscillators thus form a ring (see Fig. 1 (a) for schematic description). The matrix 𝚲\bm{\Lambda} in Eq. (4) is given by

𝚲=MΩ2𝟏Mλ𝑽,\displaystyle\bm{\Lambda}=M\Omega^{2}\bm{1}-M\lambda\bm{V}, (40)

where

𝑽=(010001101000010100100010)\displaystyle\bm{V}=\begin{pmatrix}0&1&0&0&\cdots&0&1\\ 1&0&1&0&\cdots&0&0\\ 0&1&0&1&\cdots&0&0\\ \vdots&\vdots&\vdots&\vdots&&\vdots&\vdots\\ 1&0&0&0&\cdots&1&0\end{pmatrix} (41)

As shown in Appendix D, the eigenvalues of 𝚲\bm{\Lambda} are given by MΩk2M\Omega^{2}_{k}, k=0,1,,N1k=0,1,\cdots,N-1, where

Ωk2Ω22λcos(2πNk).\displaystyle\Omega^{2}_{k}\equiv\Omega^{2}-2\lambda\cos\left(\frac{2\pi}{N}k\right). (42)

Note that there are two-fold degeneracies as ΩNk=Ωk\Omega_{N-k}=\Omega_{k}. For stability we require 2|λ|<Ω22|\lambda|<\Omega^{2}. The eigenvectors of 𝚲\bm{\Lambda} can be used to construct an orthogonal matrix 𝑹\bm{R}, which diagonalizes 𝚲\bm{\Lambda} as 𝑹T𝚲𝑹=diag(MΩ02,,MΩN12)\bm{R}^{\rm T}\bm{\Lambda}\bm{R}=\mathrm{diag}(M\Omega^{2}_{0},\cdots,M\Omega^{2}_{N-1}). We therefore have

𝑮(0)(z)=[z2𝟏+1M𝚲]1\displaystyle\bm{G}^{(0)}(z)=\left[z^{2}\bm{1}+\frac{1}{M}\bm{\Lambda}\right]^{-1} (43)
=𝑹diag(1z2+Ω02,,1z2+ΩN12)𝑹T.\displaystyle~{}~{}~{}~{}~{}=\bm{R}~{}\mathrm{diag}\left(\frac{1}{z^{2}+\Omega^{2}_{0}},\cdots,\frac{1}{z^{2}+\Omega^{2}_{N-1}}\right)\bm{R}^{\rm T}.

In Appendix D, we present explicit eigenvectors and consequently 𝑹\bm{R} which diagonalizes 𝑽\bm{V} and 𝚲\bm{\Lambda}.

Refer to caption
Figure 1: Schematic diagrams depicting a chain of harmonic oscillators interacting with (a) a single heat bath and (b) two heat baths.

Now the full Green’s function 𝑮\bm{G} in the presence of the coupling to the bath can be obtained using Eq. (32). The effect of coupling to the bath is characterized by ζr(α)\zeta^{(\alpha)}_{r} in Eq. (17) and the bath spectral density Jα(ω)J_{\alpha}(\omega) in Eq. (6). We will again drop the index α\alpha as there is only one bath. We consider an Ohmic bath with the cutoff ωD\omega_{D} and the coupling strength γ\gamma, which is given by

J(ω)=2MγπωωD2ω2+ωD2.\displaystyle J(\omega)=\frac{2M\gamma}{\pi}\omega\frac{\omega^{2}_{D}}{\omega^{2}+\omega^{2}_{D}}. (44)

Putting this into Eq. (17), we have

ζr=2γωD|νr||νr|+ωD.\displaystyle\zeta_{r}=2\gamma\frac{\omega_{D}|\nu_{r}|}{|\nu_{r}|+\omega_{D}}. (45)

In the following, we present explicit expressions for the MFG state for general values of N2N\geq 2. As we have seen above in Eqs. (27) and (28), the MFG state is completely characterized by its covariances. First, using the eigenvectors obtained in Appendix D and Eq. (43), we can write

Gij(0)(z)=1Nk=0N1cos(2πN(ij)k)z2+Ωk2.\displaystyle G^{(0)}_{ij}(z)=\frac{1}{N}\sum_{k=0}^{N-1}\frac{\cos\left(\frac{2\pi}{N}(i-j)k\right)}{z^{2}+\Omega^{2}_{k}}. (46)

We then obtain the covariances with respect to the system Gibbs state ρ^G\hat{\rho}_{\rm G}, which we denote by G\langle\cdots\rangle_{\rm G}. This can be done by using Eqs. (27) and (28) with 𝑮\bm{G} replaced by 𝑮(0)\bm{G}^{(0)}. Using the summation formula

r=1νr2+ω2=(β2ω)coth(βω2),\displaystyle\sum_{r=-\infty}^{\infty}\frac{1}{\nu^{2}_{r}+\omega^{2}}=\left(\frac{\beta\hbar}{2\omega}\right)\coth\left(\frac{\beta\hbar\omega}{2}\right), (47)

we obtain

x^ix^jG=2MNk=0N1coth(βΩk2)Ωkcos(2πN(ij)k).\displaystyle\langle\hat{x}_{i}\hat{x}_{j}\rangle_{\rm G}=\frac{\hbar}{2MN}\sum_{k=0}^{N-1}\frac{\coth\left(\frac{\beta\hbar\Omega_{k}}{2}\right)}{\Omega_{k}}\cos\left(\frac{2\pi}{N}(i-j)k\right). (48)

For the momentum covariances, we first note from Eq. (43) that

𝟏z2𝑮(0)(z)\displaystyle\bm{1}-z^{2}\bm{G}^{(0)}(z)
=𝑹diag(Ω02z2+Ω02,,ΩN12z2+ΩN12)𝑹T.\displaystyle=\bm{R}~{}\mathrm{diag}\left(\frac{\Omega^{2}_{0}}{z^{2}+\Omega^{2}_{0}},\cdots,\frac{\Omega^{2}_{N-1}}{z^{2}+\Omega^{2}_{N-1}}\right)\bm{R}^{\rm T}. (49)

We have from Eq. (28)

p^ip^jG=M2Nk=0N1Ωkcoth(βΩk2)cos(2πN(ij)k).\displaystyle\langle\hat{p}_{i}\hat{p}_{j}\rangle_{\rm G}=\frac{M\hbar}{2N}\sum_{k=0}^{N-1}\Omega_{k}\coth\left(\frac{\beta\hbar\Omega_{k}}{2}\right)\cos\left(\frac{2\pi}{N}(i-j)k\right). (50)

Now the covariances with respect to the MFG state can be obtained from the full Green’s function in Eq. (32) with Eq. (33). We therefore have

x^ix^jmfG=x^ix^jG+Δx^ix^j,\displaystyle\langle\hat{x}_{i}\hat{x}_{j}\rangle_{\rm mfG}=\langle\hat{x}_{i}\hat{x}_{j}\rangle_{\rm G}+\Delta\langle\hat{x}_{i}\hat{x}_{j}\rangle, (51)
p^ip^jmfG=p^ip^jG+Δp^ip^j,\displaystyle\langle\hat{p}_{i}\hat{p}_{j}\rangle_{\rm mfG}=\langle\hat{p}_{i}\hat{p}_{j}\rangle_{\rm G}+\Delta\langle\hat{p}_{i}\hat{p}_{j}\rangle, (52)

where the differences compared to those obtained with respect to the system Gibbs state are given in terms of Δ𝑮\Delta\bm{G} in Eq. (33) by

Δx^ix^j=1Mβr=ΔGij(νr)\displaystyle\Delta\langle\hat{x}_{i}\hat{x}_{j}\rangle=\frac{1}{M\beta}\sum_{r=-\infty}^{\infty}\Delta G_{ij}(\nu_{r}) (53)
Δp^ip^j=Mβr=νr2ΔGij(νr).\displaystyle\Delta\langle\hat{p}_{i}\hat{p}_{j}\rangle=-\frac{M}{\beta}\sum_{r=-\infty}^{\infty}\nu^{2}_{r}\Delta G_{ij}(\nu_{r}). (54)

We evaluate the infinite sums in Eqs. (53) and (54) numerically and calculate the covariances with respect to the MFG state. For the numerical calculations, we present x^ix^j\langle\hat{x}_{i}\hat{x}_{j}\rangle and p^ip^j\langle\hat{p}_{i}\hat{p}_{j}\rangle in units of /(MΩ)\hbar/(M\Omega) and MΩ\hbar M\Omega, respectively. In Figs. 2 and 3, we show the results for x^i2\langle\hat{x}^{2}_{i}\rangle and p^i2\langle\hat{p}^{2}_{i}\rangle calculated with respect to the MFG state and compare them with those obtained from the system Gibbs state for various values of the coupling strength γ\gamma (in the unit of Ω\Omega) to the bath and of the inter-oscillator coupling constant λ\lambda (in the unit of Ω2\Omega^{2}). The calculations were done for a ring of N=20N=20 coupled oscillators where j=0j=0 oscillator is coupled with the heat bath at inverse temperature β\beta. We take the Ohmic bath described by Eqs. (44) and (45) with ωD=100\omega_{D}=100 (in the unit of Ω\Omega).

At relatively high temperature, β=1\beta=1 (The inverse temperature is in the unit of 1/(Ω)1/(\hbar\Omega).), Fig. 2 (a) shows that x^j2mfG\langle\hat{x}^{2}_{j}\rangle_{\rm mfG} exhibits very little variation compared to those obtained from the system Gibbs state, which is independent of jj (see Eq. (48)). The small difference between the MFG and the system Gibbs averages occurs only near the contact point j=0j=0 between the system and the bath. On the other hand, the momentum covariance p^j2mfG\langle\hat{p}^{2}_{j}\rangle_{\rm mfG} shows a different behavior as one can see from Fig. 2 (b). The value p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG} right at the point of contact with the bath increases quite rapidly as the coupling strength γ\gamma to the bath increases. As we will see below in the next subsection, this is a characteristic behavior of a quantum system where the coupling between the position operators of the system and the heat bath gets stronger. As we can see in the inset of the same figure, away from the point of contact, the effect of the coupling to the bath quickly disappears as in the position covariance.

Refer to caption
Figure 2: The covariances (a) x^j2\langle\hat{x}^{2}_{j}\rangle and (b) p^j2\langle\hat{p}^{2}_{j}\rangle as functions of the position jj of the N=20N=20 oscillators. At j=0j=0, the system is in contact with the heat bath maintained at the inverse temperature β=1\beta=1. The solid horizontal lines are those calculated with respect to the system Gibbs (G) states. The data points and the broken lines are obtained from the averages with respect to the MFG states for various values of the inter-oscillator coupling λ=0.05\lambda=0.05 (squares) and λ=0.45\lambda=0.45 (circles) and the coupling strength to the bath γ=0.1,1\gamma=0.1,1 and 10. The inset in (b) is the same plot excluding j=0j=0 point.
Refer to caption
Figure 3: The covariances (a) x^j2\langle\hat{x}^{2}_{j}\rangle and (b) p^j2\langle\hat{p}^{2}_{j}\rangle as functions of the position jj of the oscillators at the inverse temperature β=100\beta=100. The other parameters are the same as in Fig. 2.

At lower temperatures, the effect of the coupling to the bath in x^j2mfG\langle\hat{x}^{2}_{j}\rangle_{\rm mfG} is more prominent as we can see in Fig. 3 (a) at the inverse temperature β=100\beta=100. The effect, however, disappears as we move away from the point of contact with the bath. In fact, at lower temperatures β10\beta\gtrsim 10, we find that the difference defined in Eq. (51) is well described by an exponential decay |Δx^j2|exp(j/ξ)|\Delta\langle\hat{x}^{2}_{j}\rangle|\sim\exp(-j/\xi) with the characteristic length ξ\xi. This is quite similar to the skin effect found in Ref. Burke et al. (2024), where the effect of coupling to the environment decays exponentially with the distance from the system-environment boundary. In Fig. 4 (a), we plot the skin depth ξ\xi as a function of the inter-oscillator coupling λ\lambda for various values of the temperature and the coupling strength γ\gamma to the bath. We note that ξ\xi is quite small (1\lesssim 1) even at low temperatures and strong system-bath couplings, and that it shows only a moderate increase as a function of λ\lambda and β\beta. A somewhat surprising result is that the effect of the bath measured in terms of ξ\xi is almost independent of the coupling strength γ\gamma to the bath for given values of β\beta and λ\lambda. The momentum covariance p^j2mfG\langle\hat{p}^{2}_{j}\rangle_{\rm mfG} at lower temperatures, on the other hand, behaves quite similarly to that at higher temperatures as can be seen from Fig. 3 (b). The main feature still is a sharp increase of p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG} as γ\gamma increases. The effect of the coupling to the bath quickly disappears as can be seen in the inset of Fig. 3 (b).

Refer to caption
Figure 4: The skin depth ξ\xi obtained from the exponential curve fitting of (a) |Δx^j2||\Delta\langle\hat{x}^{2}_{j}\rangle| and (b) |Δx^0x^j||\Delta\langle\hat{x}_{0}\hat{x}_{j}\rangle| (see main text) as a function of the inter-oscillator coupling λ\lambda for the coupling strength to the bath γ=0.1\gamma=0.1, 1 and 10, and for the inverse temperature β=10\beta=10 (circles) and 100 (squares).
Refer to caption
Figure 5: The differences (a) Δx^0x^j\Delta\langle\hat{x}_{0}\hat{x}_{j}\rangle and (b) Δp^0p^j\Delta\langle\hat{p}_{0}\hat{p}_{j}\rangle of covariances between those calculated with respect to the Gibbs and the MFG states as functions of the position jj of the N=20N=20 oscillators for various values of the coupling strength γ=0.1,1\gamma=0.1,1 and 10 to the bath and the inter-oscillator coupling λ=0.05\lambda=0.05 (squares) and 0.45 (circles). The system is in contact with the heat bath at j=0j=0. The inverse temperature is β=100\beta=100. The inset in (b) shows the same data from j=1j=1.

In order to study the effect of the coupling to the bath on the MFG state further, we look at another type of covariances, x^0x^jmfG\langle\hat{x}_{0}\hat{x}_{j}\rangle_{\rm mfG} and p^0p^jmfG\langle\hat{p}_{0}\hat{p}_{j}\rangle_{\rm mfG}, which connect operators at different sites, and compare them with those obtained with respect to the system Gibbs state. We again study the effect of the coupling to the bath using the differences, Δx^0x^j\Delta\langle\hat{x}_{0}\hat{x}_{j}\rangle and Δp^0p^j\Delta\langle\hat{p}_{0}\hat{p}_{j}\rangle between the MFG and the system Gibbs states, which are plotted in Fig. 5 at inverse temperature β=100\beta=100 for N=20N=20 oscillators. We note that in this case the system Gibbs averages as well as the MFG ones show a spatial dependence on jj as one can see from Eqs. (48) and (50). The differences between them, however, show a similar behavior to the previous case of Δx^j2\Delta\langle\hat{x}_{j}^{2}\rangle as can be seen from Fig. 5 (a). In fact, we find that |Δx^0x^j||\Delta\langle\hat{x}_{0}\hat{x}_{j}\rangle| is also well described by an exponential decay exp(j/ξ)\sim\exp(-j/\xi). We plot the skin depth ξ\xi for this covariance in Fig. 4 (b) for various values of the parameters. We note that the skin depths in this case are slightly bigger than those for Δx^j2\Delta\langle\hat{x}_{j}^{2}\rangle, but the overall behavior of ξ\xi as a function of the coupling constants and temperatures is quite similar to the previous case. The main point we make here is that the effect of the heat bath on the equilibrium state measured with ξ\xi in both cases is quite short-ranged reaching only over one site. The difference in the momentum covariance Δp^0p^j\Delta\langle\hat{p}_{0}\hat{p}_{j}\rangle connecting different sites also shows a similar behavior to Δp^j2\Delta\langle\hat{p}^{2}_{j}\rangle. Apart from the increasing Δp^02mfG\Delta\langle\hat{p}^{2}_{0}\rangle_{\rm mfG} at j=0j=0 as γ\gamma gets bigger, the effect of the coupling to the bath quickly disappears as we move away from the contact point as we can see in Fig. 5 (b).

III.2 The Ultra Strong Coupling (USC) Limit

In this subsection, we investigate the MFG state obtained above, which is an exact result, in the limit when the coupling between the system and the bath is arbitrarily large. This is to make a connection with the ultra strong coupling (USC) limit obtained recently in Ref. Cresser and Anders (2021) of the MFG state for a general system coupled with a bath. If the interaction Hamiltonian between system and bath is given by κX^B^\kappa\hat{X}\hat{B} with a system operator X^\hat{X} and a bath operator B^\hat{B}, and the coupling strength κ\kappa, it was shown in Ref. Cresser and Anders (2021) that the MFG state in the limit κ\kappa\to\infty is diagonal in the eigenstates of X^\hat{X}. More explicitly, the ultra strong coupling limit is given by Cresser and Anders (2021)

ρ^USCexp[βnP^nH^SP^n],\displaystyle\hat{\rho}_{\rm USC}\sim\exp[-\beta\sum_{n}\hat{P}_{n}\hat{H}_{\rm S}\hat{P}_{n}], (55)

where P^n\hat{P}_{n} is the projection operator on the eigenstates |Xn|X_{n}\rangle of X^\hat{X} with eigenvalue XnX_{n}. In other words, the USC limit is characterized by the effective Hamiltonian of mean force HUSCnP^nH^SP^nH_{\rm USC}\equiv\sum_{n}\hat{P}_{n}\hat{H}_{\rm S}\hat{P}_{n}, which is just the system Hamiltonian but projected onto the eigenstates of the system operator taking part in the coupling to the bath.

The reason we study our present system in this limit is twofold. First, for the case of harmonic oscillators, the system operator coupled to the bath is just the position operator x^0\hat{x}_{0} at j=0j=0 thus has a continuous spectrum. Therefore, the USC limit of the exact MFG state for the quantum harmonic oscillators can provide a generalization of Eq. (55) which was originally obtained in Ref. Cresser and Anders (2021) for the coupling system operator with a discrete spectrum to a continuous one.

To illustrate this point, we consider the simplest case of a single (N=1N=1) damped harmonic oscillator Grabert et al. (1984); Hilt et al. (2011) in the USC limit. Putting N=1N=1 in Eqs. (21), (27) and (28), we have

x^2mfG=1Mβr=1νr2+Ω2+ζr,\displaystyle\langle\hat{x}^{2}\rangle_{\rm mfG}=\frac{1}{M\beta}\sum_{r=-\infty}^{\infty}\frac{1}{\nu_{r}^{2}+\Omega^{2}+\zeta_{r}}, (56)
p^2mfG=Mβr=Ω2+ζrνr2+Ω2+ζr,\displaystyle\langle\hat{p}^{2}\rangle_{\rm mfG}=\frac{M}{\beta}\sum_{r=-\infty}^{\infty}\frac{\Omega^{2}+\zeta_{r}}{\nu_{r}^{2}+\Omega^{2}+\zeta_{r}}, (57)

Taking the coupling strength γ\gamma in ζr\zeta_{r} to infinity (see Eq. (45)), we see that the infinite sum in p^2mfG\langle\hat{p}^{2}\rangle_{\rm mfG} diverges. We also find that only the r=0r=0 term in Eq. (56) survives, which gives x^2mfG1/(MβΩ2)\langle\hat{x}^{2}\rangle_{\rm mfG}\to 1/(M\beta\Omega^{2}) as γ\gamma\to\infty. As can be seen from Eqs. (23), the diverging momentum covariance indicates that the matrix element x|ρmfG|x\langle x|\rho_{\rm mfG}|x^{\prime}\rangle is proportional to δ(xx)\delta(x-x^{\prime}) i.e. diagonal in eigenstates of x^\hat{x} in the USC limit. This point was recognized recently in Ref. Kumar and Ghosh (2024). The actual value of the position covariance in the USC limit indicates that ρUSC\rho_{\rm USC} is given by the Gibbs state of an effective Hamiltonian of the form H^USC=(1/2)MΩ2x^2\hat{H}_{\rm USC}=(1/2)M\Omega^{2}\hat{x}^{2}. This is exactly the system Hamiltonian projected onto the position eigenstate as done in Eq. (55). It is, however, different from another form of the USC limit suggested in Refs. Goyal and Kawai (2019); Orman and Kawai (2020), which is equal to nP^nρ^GP^n=𝑑x|xx|ρ^G|xx|\sum_{n}\hat{P}_{n}\hat{\rho}_{\rm G}\hat{P}_{n}=\int dx|x\rangle\langle x|\hat{\rho}_{\rm G}|x\rangle\langle x| in our case. Therefore it would give the position covariance as (/(2MΩ))coth(βΩ/2)(\hbar/(2M\Omega))\coth(\beta\hbar\Omega/2) as one can see from the diagonal element in Eq. (23) with G(0)G^{(0)} for ρ^G\hat{\rho}_{\rm G}. But this is clearly in disagreement with the actual USC limit obtained from the exact expression except for very high temperatures.

The second point why we study the USC limit in our model is the following. We note that the USC limit suggested in Eq. (55) is characterized by the effective Hamiltonian which is a diagonal projection onto the eigenstates of the system operator that couples to the bath. In our model, this system operator is only local as only a part of the system couples to the bath. Therefore, using our exact result, we can check whether the USC limit in Eq. (55) still holds for the case where the coupling operator is local in space. To do this we focus on the simplest case N=2N=2, where the positions of the oscillator are x^0\hat{x}_{0} and x^1\hat{x}_{1} and only x^0\hat{x}_{0} is coupled to the bath. The normal mode frequencies from Eq. (42) are Ω02=Ω22λ\Omega^{2}_{0}=\Omega^{2}-2\lambda and Ω12=Ω2+2λ\Omega^{2}_{1}=\Omega^{2}+2\lambda, and from Eq. (46), we have

G00(0)(z)=G11(0)(z)=12[1z2+Ω02+1z2+Ω12],\displaystyle G^{(0)}_{00}(z)=G^{(0)}_{11}(z)=\frac{1}{2}\left[\frac{1}{z^{2}+\Omega^{2}_{0}}+\frac{1}{z^{2}+\Omega^{2}_{1}}\right], (58)
G01(0)(z)=G10(0)(z)=12[1z2+Ω021z2+Ω12].\displaystyle G^{(0)}_{01}(z)=G^{(0)}_{10}(z)=\frac{1}{2}\left[\frac{1}{z^{2}+\Omega^{2}_{0}}-\frac{1}{z^{2}+\Omega^{2}_{1}}\right]. (59)

The USC limit of the two-oscillator system is obtained by taking γ\gamma\to\infty in Eqs. (37) and (38). We have

x^02mfGγG00(0)(0)Mβ=12Mβ(1Ω02+1Ω12),\displaystyle\langle\hat{x}^{2}_{0}\rangle_{\rm mfG}\xrightarrow[\gamma\to\infty]{}\frac{G^{(0)}_{00}(0)}{M\beta}=\frac{1}{2M\beta}\left(\frac{1}{\Omega^{2}_{0}}+\frac{1}{\Omega^{2}_{1}}\right), (60)
x^0x^1mfGγG01(0)(0)Mβ=12Mβ(1Ω021Ω12),\displaystyle\langle\hat{x}_{0}\hat{x}_{1}\rangle_{\rm mfG}\xrightarrow[\gamma\to\infty]{}\frac{G^{(0)}_{01}(0)}{M\beta}=\frac{1}{2M\beta}\left(\frac{1}{\Omega^{2}_{0}}-\frac{1}{\Omega^{2}_{1}}\right), (61)

and

x^12mfG\displaystyle\langle\hat{x}^{2}_{1}\rangle_{\rm mfG} γG11(0)(0)Mβ+2Mβr=1G00(0)(νr)2G01(0)(νr)2G00(0)(νr)\displaystyle\xrightarrow[\gamma\to\infty]{}\frac{G^{(0)}_{11}(0)}{M\beta}+\frac{2}{M\beta}\sum_{r=1}^{\infty}\frac{G^{(0)}_{00}(\nu_{r})^{2}-G^{(0)}_{01}(\nu_{r})^{2}}{G^{(0)}_{00}(\nu_{r})}
=12Mβ(1Ω02+1Ω12)+2Mβr=11νr2+Ω2.\displaystyle=\frac{1}{2M\beta}\left(\frac{1}{\Omega^{2}_{0}}+\frac{1}{\Omega^{2}_{1}}\right)+\frac{2}{M\beta}\sum_{r=1}^{\infty}\frac{1}{\nu^{2}_{r}+\Omega^{2}}. (62)

For the momentum covariances, we can easily see from Eq. (38) that p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG}\to\infty and p^0p^1mfG0\langle\hat{p}_{0}\hat{p}_{1}\rangle_{\rm mfG}\to 0 as γ\gamma\to\infty and

p^12\displaystyle\langle\hat{p}^{2}_{1}\rangle γmfGMβr=[1νr2G00(0)(νr)2G01(0)(νr)2G00(0)(νr)]{}_{\rm mfG}\xrightarrow[\gamma\to\infty]{}\frac{M}{\beta}\sum_{r=-\infty}^{\infty}\left[1-\nu^{2}_{r}\frac{G^{(0)}_{00}(\nu_{r})^{2}-G^{(0)}_{01}(\nu_{r})^{2}}{G^{(0)}_{00}(\nu_{r})}\right]
=Mβr=Ω2νr2+Ω2=MΩ2coth(βΩ2).\displaystyle=\frac{M}{\beta}\sum_{r=-\infty}^{\infty}\frac{\Omega^{2}}{\nu^{2}_{r}+\Omega^{2}}=\frac{M\hbar\Omega}{2}\coth\left(\frac{\beta\hbar\Omega}{2}\right). (63)

We now check whether the USC limit, Eq. (55) conjectured in Ref. Cresser and Anders (2021) is consistent with this large-γ\gamma limit obtained from the exact results. In the present two-oscillator case, the USC limit of Ref. Cresser and Anders (2021) is characterized by the effective Hamiltonian which is a diagonal projection of H^S\hat{H}_{\rm S} onto the eigenstates of x^0\hat{x}_{0}. As we have seen in the one-oscillator case, it amounts to suppressing in the system Hamiltonian the kinetic energy (p^02/(2M)\hat{p}^{2}_{0}/(2M)) of the oscillator in contact with the heat bath. The matrix element of ρ^USC\hat{\rho}_{\rm USC} in Eq. (55) is then given by

x¯0,x¯1|ρ^USC|x¯0,x¯1\displaystyle\langle\bar{x}_{0},\bar{x}_{1}|\hat{\rho}_{\rm USC}|\bar{x}_{0},\bar{x}^{\prime}_{1}\rangle (64)
=\displaystyle= 1ZUSCx1(0)=x¯1x1(β)=x¯1𝒟x1(τ)exp[1SUSC[x1;x¯0]],\displaystyle\frac{1}{Z_{\rm USC}}\int_{x_{1}(0)=\bar{x}_{1}^{\prime}}^{x_{1}(\beta\hbar)=\bar{x}_{1}}\mathcal{D}x_{1}(\tau)\exp\left[-\frac{1}{\hbar}S_{\rm USC}[x_{1};\bar{x}_{0}]\right],

where the system Euclidian action is given by

SUSC[x1;\displaystyle S_{\rm USC}[x_{1}; x¯0]=0βdτ[M2(dx1(τ)dτ)2\displaystyle\bar{x}_{0}]=\int_{0}^{\beta\hbar}d\tau\;\Big{[}\frac{M}{2}\left(\frac{dx_{1}(\tau)}{d\tau}\right)^{2}
+M2Ω2(x¯02+x12(τ))2Mλx¯0x1(τ)]\displaystyle+\frac{M}{2}\Omega^{2}\left(\bar{x}^{2}_{0}+x^{2}_{1}(\tau)\right)-2M\lambda\bar{x}_{0}x_{1}(\tau)\Big{]} (65)

with ZUSC=𝑑x¯0𝑑x¯1x¯0,x¯1|ρ^USC|x¯0,x¯1Z_{\rm USC}=\int d\bar{x}_{0}d\bar{x}_{1}\langle\bar{x}_{0},\bar{x}_{1}|\hat{\rho}_{\rm USC}|\bar{x}_{0},\bar{x}_{1}\rangle. We can evaluate this path integral by seeking the stationary path solution as done in Sec. II and in Appendices B and C. Detailed calculations are given in Appendix E. We obtain

x¯0,x¯1|ρ^USC|x¯0,x¯1=C~exp[M2{A(x¯1x¯1)2\displaystyle\langle\bar{x}_{0},\bar{x}_{1}|\hat{\rho}_{\rm USC}|\bar{x}_{0},\bar{x}^{\prime}_{1}\rangle=\tilde{C}\exp\Bigg{[}-\frac{M}{2\hbar}\Big{\{}A(\bar{x}_{1}^{\prime}-\bar{x}_{1})^{2}
+1F(x¯1+x¯122λΩ2x¯0)2+βΩ2(14λ2Ω4)x¯02}],\displaystyle+\frac{1}{F}\left(\frac{\bar{x}_{1}+\bar{x}_{1}^{\prime}}{2}-\frac{2\lambda}{\Omega^{2}}\bar{x}_{0}\right)^{2}+\beta\hbar\Omega^{2}\left(1-\frac{4\lambda^{2}}{\Omega^{4}}\right)\bar{x}^{2}_{0}\Big{\}}\Bigg{]}, (66)

where

A=1βr=Ω2νr2+Ω2=Ω2coth(βΩ2),\displaystyle A=\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}\frac{\Omega^{2}}{\nu^{2}_{r}+\Omega^{2}}=\frac{\Omega}{2}\coth\left(\frac{\beta\hbar\Omega}{2}\right), (67)
F=1βr=1νr2+Ω2=12Ωcoth(βΩ2),\displaystyle F=\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}\frac{1}{\nu^{2}_{r}+\Omega^{2}}=\frac{1}{2\Omega}\coth\left(\frac{\beta\hbar\Omega}{2}\right), (68)

and C~\tilde{C} is determined from 1=𝑑x¯0𝑑x¯1x¯0,x¯1|ρ^USC|x¯0,x¯11=\int d\bar{x}_{0}d\bar{x}_{1}\;\langle\bar{x}_{0},\bar{x}_{1}|\hat{\rho}_{\rm USC}|\bar{x}_{0},\bar{x}_{1}\rangle.

The covariances are now calculated by evaluating Gaussian integrals such as

x^02USC\displaystyle\langle\hat{x}^{2}_{0}\rangle_{\rm USC} =𝑑x¯0𝑑x¯1x¯02x¯0,x¯1|ρ^USC|x¯0,x¯1\displaystyle=\int d\bar{x}_{0}d\bar{x}_{1}\;\bar{x}_{0}^{2}\langle\bar{x}_{0},\bar{x}_{1}|\hat{\rho}_{\rm USC}|\bar{x}_{0},\bar{x}_{1}\rangle
=1MβΩ2Ω44λ2,\displaystyle=\frac{1}{M\beta}\frac{\Omega^{2}}{\Omega^{4}-4\lambda^{2}}, (69)

and

x^0x^1USC\displaystyle\langle\hat{x}_{0}\hat{x}_{1}\rangle_{\rm USC} =𝑑x¯0𝑑x¯1x¯0x¯1x¯0,x¯1|ρ^USC|x¯0,x¯1\displaystyle=\int d\bar{x}_{0}d\bar{x}_{1}\;\bar{x}_{0}\bar{x}_{1}\langle\bar{x}_{0},\bar{x}_{1}|\hat{\rho}_{\rm USC}|\bar{x}_{0},\bar{x}_{1}\rangle
=2λΩ2x^02USC=1Mβ2λΩ44λ2,\displaystyle=\frac{2\lambda}{\Omega^{2}}\langle\hat{x}^{2}_{0}\rangle_{\rm USC}=\frac{1}{M\beta}\frac{2\lambda}{\Omega^{4}-4\lambda^{2}}, (70)

where we have shifted the integration variable x¯1x¯1+2λx¯0/Ω2\bar{x}_{1}\to\bar{x}_{1}+2\lambda\bar{x}_{0}/\Omega^{2} in the second integral. Note that these are exactly the same as Eqs. (60) and (61), respectively. The remaining part is given by

x^12USC\displaystyle\langle\hat{x}^{2}_{1}\rangle_{\rm USC} =𝑑x¯0𝑑x¯1x¯12x¯0,x¯1|ρ^USC|x¯0,x¯1\displaystyle=\int d\bar{x}_{0}d\bar{x}_{1}\;\bar{x}^{2}_{1}\langle\bar{x}_{0},\bar{x}_{1}|\hat{\rho}_{\rm USC}|\bar{x}_{0},\bar{x}_{1}\rangle
=MF+4λ2Ω4x^02USC\displaystyle=\frac{\hbar}{M}F+\frac{4\lambda^{2}}{\Omega^{4}}\langle\hat{x}^{2}_{0}\rangle_{\rm USC} (71)

We can easily verify that this coincides with Eq. (62). For the momentum covariance, we can see that only the off-diagonal part (exp[MA(x¯1x¯1)2/(2)]\exp[-MA(\bar{x}_{1}^{\prime}-\bar{x}_{1})^{2}/(2\hbar)]) is relevant. Therefore, we have

p^12USC=MA,\displaystyle\langle\hat{p}^{2}_{1}\rangle_{\rm USC}=M\hbar A, (72)

which is just Eq. (63). In summary, we have verified that the USC limit given as Eq. (55) is still valid even if the projection is onto a local system operator.

As we have shown, the USC limit is characterized by the diverging momentum covariance p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG} at the point of contact with the heat bath. This is related to the projection of the effective Hamiltonian on the space of system local coupling operator x^0\hat{x}_{0} in the USC limit. The divergence of p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG} as γ\gamma\to\infty can be made more quantitative as follows. For the N=2N=2 case with the Ohmic bath given in Eq. (45), we can actually show that as γ\gamma\to\infty

p^02mfGMωDγ2.\displaystyle\langle\hat{p}^{2}_{0}\rangle_{\rm mfG}\sim M\hbar\sqrt{\frac{\omega_{D}\gamma}{2}}. (73)

The derivation is given in Appendix F. In fact we find that this large-γ\gamma behavior of γ12\sim\gamma^{\frac{1}{2}} continues to be true even for general number of oscillators. Figure 6 shows p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG} as a function of γ\gamma for N=20N=20 oscillators for various temperatures and inter-oscillator coupling constants. For large γ\gamma, it increases with γ12\gamma^{\frac{1}{2}}. Therefore, we conclude that in the USC limit p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG} goes to infinity, which results in the projection of the MFG state onto the eigenstate of x^0\hat{x}_{0} that couples to the heat bath.

Refer to caption
Figure 6: The momentum covariance p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG} with respect to the MFG state at the point of contact with the heat bath as a function of γ1/2\gamma^{1/2} for N=20N=20 coupled oscillators. For various temperatures and couplings between the oscillators, it increases as γ1/2\gamma^{1/2} in the large-γ\gamma limit.

IV System in contact with multiple heat baths at the same temperature

In this section, we apply the general formalism developed above to the case where there are multiple heat baths at the same inverse temperature β\beta. We first focus on the case where there are two baths and the system makes contact with them at the locations given by ={α,α}\mathcal{B}=\{\alpha,\alpha^{\prime}\}. We then generalize to the multiple-bath case. For the two-bath case, the diagonal matrix 𝚺\bm{\Sigma} in Eq. (21) has only two nonzero elements, namely Σαα(νr)=ζr(α)\Sigma_{\alpha\alpha}(\nu_{r})=\zeta^{(\alpha)}_{r} and Σαα(νr)=ζr(α)\Sigma_{\alpha^{\prime}\alpha^{\prime}}(\nu_{r})=\zeta^{(\alpha^{\prime})}_{r}. For later use, we denote by 𝚺¯\bm{\bar{\Sigma}} the 2×22\times 2 matrix composed of these nonzero elements, which belongs to the subspace spanned by {α,α}\{\alpha,\alpha^{\prime}\}.

As in the case of the single bath, we try to express the Green’s function in Eq. (21) in terms of 𝑮(0)\bm{G}^{(0)} which is obtained in the absence of the coupling to the bath and the correction to it. Now from Eq. (30), we can rewrite this as

𝑮(νr)=𝑮(0)(νr)𝑮(0)(νr)𝚪(νr)𝑮(0)(νr),\displaystyle\bm{G}(\nu_{r})=\bm{G}^{(0)}(\nu_{r})-\bm{G}^{(0)}(\nu_{r})\bm{\Gamma}(\nu_{r})\bm{G}^{(0)}(\nu_{r}), (74)

where

𝚪(νr)=\displaystyle\bm{\Gamma}(\nu_{r})= 𝚺𝚺𝑮(0)𝚺+𝚺𝑮(0)𝚺𝑮(0)𝚺.\displaystyle\bm{\Sigma}-\bm{\Sigma}\bm{G}^{(0)}\bm{\Sigma}+\bm{\Sigma}\bm{G}^{(0)}\bm{\Sigma}\bm{G}^{(0)}\bm{\Sigma}-\cdots. (75)

From the structure of these terms, we realize that all the elements of Γij\Gamma_{ij} are zero except when ii and jj are equal to α\alpha or α\alpha^{\prime}. We again denote this nonzero 2×22\times 2 matrix by 𝚪¯\bm{\bar{\Gamma}}. We note that Eq. (74) has the same structure as Eq. (34) for the one-bath case. The only difference is that 𝚪\bm{\Gamma} now contains a nonzero 2×22\times 2 sub-matrix instead of the single nonvanishing component. In order to calculate this 𝚪¯\bm{\bar{\Gamma}}, we notice that Eq. (75) can be considered as a relation among 2×22\times 2 matrices 𝚺¯\bm{\bar{\Sigma}}, 𝚪¯\bm{\bar{\Gamma}} and 𝑮¯(0)\bm{\bar{G}}^{(0)}, the last of which is a matrix composed of Gij(0)G^{(0)}_{ij} with i,j{α,α}i,j\in\{\alpha,\alpha^{\prime}\}. By summing the infinite series in Eq. (75), we can write

𝚪¯(νr)=\displaystyle\bm{\bar{\Gamma}}(\nu_{r})= (𝟏+𝚺¯𝑮¯(0))1𝚺¯\displaystyle(\bm{1}+\bm{\bar{\Sigma}}\bm{\bar{G}}^{(0)})^{-1}\bm{\bar{\Sigma}}
=\displaystyle= [𝚺¯1+𝑮¯(0)]1.\displaystyle\left[\bm{\bar{\Sigma}}^{-1}+\bm{\bar{G}}^{(0)}\right]^{-1}. (76)

So 𝚪¯(νr)\bm{\bar{\Gamma}}(\nu_{r}) is obtained by simply inverting the known 2×22\times 2 matrix. The correction term Δ𝑮\Delta\bm{G} can then be calculated using Eq. (34) with newly obtained 𝚪\bm{\Gamma}.

This procedure can easily be generalized to a system in contact with mm reservoirs with m>2m>2. All one has to do is to calculate the m×mm\times m kernel matrix 𝚪¯\bm{\bar{\Gamma}} from Eq. (76) and calculate the Green’s function for the MFG state from Eq. (74).

IV.1 Example: Chain of Harmonic Oscillators in Contact with Two Heat Baths at the Same Temperature

Here we apply this method to the ring of oscillators studied in Sec. III.1. We focus on the case where there are even number of oscillators and the two baths are connected to the oscillators at x0x_{0} and xN2x_{\frac{N}{2}}. In the present notation, we take α=0\alpha=0 and α=N2\alpha^{\prime}=\frac{N}{2}. See Fig. 1 (b) for a schematic diagram describing the situation. We assume that the two baths are described by the Ohmic spectral densities. In particular, we assume that the baths at α=0\alpha=0 and α=N2\alpha^{\prime}=\frac{N}{2} are described by ζr\zeta_{r} and ζr\zeta^{\prime}_{r}, respectively, which are given by

ζr=2γ1ωD|νr||νr|+ωD,ζr=2γ2ωD|νr||νr|+ωD.\displaystyle\zeta_{r}=2\gamma_{1}\frac{\omega_{D}|\nu_{r}|}{|\nu_{r}|+\omega_{D}},~{}~{}~{}\zeta^{\prime}_{r}=2\gamma_{2}\frac{\omega_{D}|\nu_{r}|}{|\nu_{r}|+\omega_{D}}. (77)

with in general two distinct coupling strengths γ1\gamma_{1} and γ2\gamma_{2}.

In order to calculate 𝚪¯\bm{\bar{\Gamma}} in Eq. (76), we first note that in the subspace of {0,N2}\{0,\frac{N}{2}\}

𝚺¯(νr)=(ζr00ζr).\displaystyle\bm{\bar{\Sigma}}(\nu_{r})=\begin{pmatrix}\zeta_{r}&0\\ 0&\zeta^{\prime}_{r}\end{pmatrix}. (78)

It is then straightforward to evaluate 𝚪¯(νr)\bm{\bar{\Gamma}}(\nu_{r}) from Eq. (76). From Eqs. (76) and (78), we have

𝚪¯(νr)=(G00(0)+1ζrG0N2(0)GN20(0)GN2N2(0)+1ζr)1.\displaystyle\bm{\bar{\Gamma}}(\nu_{r})=\begin{pmatrix}G^{(0)}_{00}+\frac{1}{\zeta_{r}}&G^{(0)}_{0\frac{N}{2}}\\ G^{(0)}_{\frac{N}{2}0}&G^{(0)}_{\frac{N}{2}\frac{N}{2}}+\frac{1}{\zeta^{\prime}_{r}}\end{pmatrix}^{-1}. (79)

From Eq. (46), we can see that only two of the components of 𝑮(0)\bm{G}^{(0)} are independent, which we denote by

G+(νr)G00(0)(νr)=GN2N2(0)(νr),\displaystyle G_{+}(\nu_{r})\equiv G^{(0)}_{00}(\nu_{r})=G^{(0)}_{\frac{N}{2}\frac{N}{2}}(\nu_{r}), (80)
G(νr)G0N2(0)(νr)=GN20(0)(νr).\displaystyle G_{-}(\nu_{r})\equiv G^{(0)}_{0\frac{N}{2}}(\nu_{r})=G^{(0)}_{\frac{N}{2}0}(\nu_{r}). (81)

By inverting the above 2×22\times 2 matrix, we obtain

Γ¯00(νr)=ζrD(νr)(1+ζrG+(νr)),\displaystyle\bar{\Gamma}_{00}(\nu_{r})=\frac{\zeta_{r}}{D(\nu_{r})}(1+\zeta^{\prime}_{r}G_{+}(\nu_{r})), (82)
Γ¯N2N2(νr)=ζrD(νr)(1+ζrG+(νr)),\displaystyle\bar{\Gamma}_{\frac{N}{2}\frac{N}{2}}(\nu_{r})=\frac{\zeta^{\prime}_{r}}{D(\nu_{r})}(1+\zeta_{r}G_{+}(\nu_{r})), (83)

and

Γ¯0N2(νr)=Γ¯N20(νr)=ζrζrD(νr)G(νr),\displaystyle\bar{\Gamma}_{0\frac{N}{2}}(\nu_{r})=\bar{\Gamma}_{\frac{N}{2}0}(\nu_{r})=-\frac{\zeta_{r}\zeta^{\prime}_{r}}{D(\nu_{r})}G_{-}(\nu_{r}), (84)

where

D(νr)=\displaystyle D(\nu_{r})= (1+ζrG+(νr))(1+ζrG+(νr))\displaystyle(1+\zeta_{r}G_{+}(\nu_{r}))(1+\zeta^{\prime}_{r}G_{+}(\nu_{r}))
ζrζrG2(νr).\displaystyle-\zeta_{r}\zeta^{\prime}_{r}G^{2}_{-}(\nu_{r}). (85)

Using these results and Eq. (34), we can write the correction to the system Gibbs state as

ΔGij(νr)=\displaystyle\Delta G_{ij}(\nu_{r})=- α=0,N/2α=0,N/2Giα(0)(νr)\displaystyle\sum_{\alpha=0,N/2}\sum_{\alpha^{\prime}=0,N/2}G^{(0)}_{i\alpha}(\nu_{r})
×Γαα(νr)Gαj(0)(νr).\displaystyle\times\Gamma_{\alpha\alpha^{\prime}}(\nu_{r})G^{(0)}_{\alpha^{\prime}j}(\nu_{r}). (86)

We now calculate the covariances using Eqs. (53) and (54) by evaluating the infinite sums numerically. In Fig. 7, we show results of these calculations for a system coupled to two heat baths maintained at some low temperature β=100\beta=100 with different coupling strengths. As in the one-bath case, the MFG state shows a skin effect for each heat bath, where the effect of the coupling to the bath quickly decreases as one moves away from the contact points. For a relatively large system size N=20N=20, Fig. 7 shows a separate skin effect for each heat bath.

Refer to caption
Figure 7: The covariances (a) x^j2\langle\hat{x}^{2}_{j}\rangle and (b) p^j2\langle\hat{p}^{2}_{j}\rangle as functions of the position jj of the N=20N=20 oscillators. As in Fig. 3, the solid horizontal lines are obtained using the system Gibbs (G) state. The system is in contact with two heat baths maintained at the same inverse temperature β=100\beta=100 at j=0j=0 and j=10j=10 with coupling strengths γ1=0.1\gamma_{1}=0.1 and γ2=1\gamma_{2}=1, respectively. The data points and the dashed lines are obtained using the MFG state. The covariances for two different values of the inter-oscillator coupling constants λ=0.05\lambda=0.05 (squares) and λ=0.45\lambda=0.45 (circles) are displayed.

V Discussion and Summary

We have presented a detailed procedure of obtaining the exact quantum MFG state of a coupled harmonic system interacting with multiple heat baths at the same temperature. We have provided a nonperturbative method to calculate the covariances with respect to the MFG state. For the specific examples of the chain of harmonic oscillators, we have calculated these covariances as we vary the physical parameters such as the temperature, the coupling strength to the bath and the inter-oscillator coupling constant and compare them with those calculated with respect to the system Gibbs state. We found that there is a skin effect similar to that found in Ref. Burke et al. (2024), where the effect of the coupling to the bath quickly disappears as a function of the distance of the system-bath boundary. We have also investigated quantitatively the skin depth and found that it is quite short-ranged and insensitive to the coupling strength to the bath. Using these exact results, we were also able to confirm that the ultrastrong coupling limit of the MFG state found recently in Ref. Cresser and Anders (2021) is valid even in the case where the system coupling operator is local and has a continuous spectrum.

An interesting aspect of the MFG state which has not been explored in this paper is its relation to the thermalization of an open quantum system. If we consider the dynamics of an open quantum system coupled with a heat bath, we expect the system reaches the MFG state as a steady state. The dynamics is often described in terms of quantum master equations, most of which are based on weak coupling approximations Breuer and Petruccione (2007). The Lindblad equation Lindblad (1976); Gorini et al. (1976) has been studied extensively as it guarantees the complete positivity. The standard derivation of the Lindblad equation involves a series of weak coupling approximations, which forces the local coupling system operator to be written in terms of eigenoperators with respect to the whole system Hamiltonian Breuer and Petruccione (2007) and in turn makes it global. The global Lindblad equation is often referred as a ultraweak coupling theory Trushechkin et al. (2022) as its steady state is given by the system Gibbs state, which is the zero coupling limit of the MFG state. One can keep the local nature of the coupling system operator in the Lindblad equation Landi et al. (2022). The local Lindblad equation is shown to be consistent with local conservation laws Tupkary et al. (2023) unlike the global one, but the steady state is no longer given by the Gibbs or MFG state Tupkary et al. (2022); Carlen et al. (2024). If we relax the condition of complete positivity, one has the Redfield equation, Redfield (1957) which can again be derived from the weak coupling approximations corresponding to the second-order perturbative expansion in the system-bath coupling. Now, the steady state of the Redfield equation is known to be equal to the perturbative MFG state valid up to the same order of perturbation theory Thingna et al. (2012); Tupkary et al. (2022); Lee and Yeo (2022). However, as in the global Lindblad equation, the coupling system operator becomes global. This seems to be in contrast to our finding here that the local nature of the system operator playing a role in the structure of the MFG state. It would be interesting to see how the structure of the MFG state found in this paper manifest in the steady states of various weak-coupling quantum master equations for an extended system.

Acknowledgements.
This work was supported by NRF grant funded by the Korea government (MSIT) (RS-2023-00276248).

Appendix A The Influence Functional

We first write Kα(τ)K_{\alpha}(\tau) in Eq. (13) which is periodic in the interval [0,β][0,\beta\hbar] as a Fourier series

Kα(τ)=1βr=gr(α)eiνrτ\displaystyle K_{\alpha}(\tau)=\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}g^{(\alpha)}_{r}e^{i\nu_{r}\tau} (87)

with νr=2πr/(β)\nu_{r}=2\pi r/(\beta\hbar). Then using Eq. (13) we have

gr(α)\displaystyle g^{(\alpha)}_{r} =0β𝑑τeiνrτKα(τ)=μαMζr(α),\displaystyle=\int_{0}^{\beta\hbar}d\tau\;e^{-i\nu_{r}\tau}K_{\alpha}(\tau)=\mu_{\alpha}-M\zeta^{(\alpha)}_{r}, (88)

where μ(α)\mu^{(\alpha)} and ζr(α)\zeta^{(\alpha)}_{r} are defined in Eq. (11) and Eq. (17), respectively. Putting this back into Eq. (87) and using the Poisson sum rule, we have

Kα(τ)=μαr=δ(τrβ)MLα(τ),\displaystyle K_{\alpha}(\tau)=\mu_{\alpha}\sum_{r=-\infty}^{\infty}\delta(\tau-r\beta\hbar)-ML_{\alpha}(\tau), (89)

where Lα(τ)L_{\alpha}(\tau) is given in Eq. (16). We therefore have

0β𝑑τ0τ𝑑τKα(ττ)xα(τ)xα(τ)=μα20β𝑑τxα2(τ)\displaystyle\int_{0}^{\beta\hbar}d\tau\int_{0}^{\tau}d\tau^{\prime}\;K_{\alpha}(\tau-\tau^{\prime})x_{\alpha}(\tau)x_{\alpha}(\tau^{\prime})=\frac{\mu_{\alpha}}{2}\int_{0}^{\beta\hbar}d\tau\;x_{\alpha}^{2}(\tau)
M20β𝑑τ0β𝑑τLα(ττ)xα(τ)xα(τ).\displaystyle-\frac{M}{2}\int_{0}^{\beta\hbar}d\tau\int_{0}^{\beta\hbar}d\tau^{\prime}\;L_{\alpha}(\tau-\tau^{\prime})x_{\alpha}(\tau)x_{\alpha}(\tau^{\prime}). (90)

Appendix B Stationary Path Solution

We follow closely the procedure described in Ref. Grabert et al. (1988). We solve the stationary path condition Eq. (18) using the Fourier series given in Eq. (19), which is defined outside the interval [0,β][0,\beta\hbar] as well. The derivatives of 𝒙(τ)\bm{x}(\tau) with respect to τ\tau then produce singularities at the endpoints τ=rβ\tau=r\beta\hbar, r=0,±1,±2,r=0,\pm 1,\pm 2,\cdots. In order to account for these singularities, we rewrite Eq. (18) as

d2xj(τ)dτ2+1Mk=0N1Λjkxk(τ)\displaystyle-\frac{d^{2}x_{j}(\tau)}{d\tau^{2}}+\frac{1}{M}\sum_{k=0}^{N-1}\Lambda_{jk}x_{k}(\tau) (91)
+α0β𝑑τLα(ττ)xα(τ)=ajδ~(τ)bjδ~(τ),\displaystyle+\sum_{\alpha\in\mathcal{B}}\int_{0}^{\beta\hbar}d\tau^{\prime}\;L_{\alpha}(\tau-\tau^{\prime})x_{\alpha}(\tau^{\prime})=-a_{j}\tilde{\delta}^{\prime}(\tau)-b_{j}\tilde{\delta}(\tau),

for some constants aia_{i} and bib_{i} to be determined later, where

δ~(τ)n=δ(τnβ)=1βr=eiνrτ\displaystyle\tilde{\delta}(\tau)\equiv\sum_{n=-\infty}^{\infty}\delta(\tau-n\beta\hbar)=\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}e^{i\nu_{r}\tau} (92)

is the discrete delta function. In terms of the Fourier mode, we have for j=0,1,,N1j=0,1,\cdots,N-1,

νr2xj,r+1MkΛjkxk,r+αζr(α)xα,r\displaystyle\nu^{2}_{r}x_{j,r}+\frac{1}{M}\sum_{k}\Lambda_{jk}x_{k,r}+\sum_{\alpha\in\mathcal{B}}\zeta^{(\alpha)}_{r}x_{\alpha,r}
=\displaystyle= iνrajbj.\displaystyle-i\nu_{r}a_{j}-b_{j}. (93)

This equation is to be solved with the boundary condition 𝒙(0+)=𝒙¯\bm{x}(0^{+})=\bm{\bar{x}^{\prime}} and 𝒙(β)=𝒙(0)=𝒙¯\bm{x}(\beta\hbar)=\bm{x}(0^{-})=\bm{\bar{x}}. The solution can be written as

xj,r=k=0N1Gjk(νr)[iνrak+bk],\displaystyle x_{j,r}=-\sum_{k=0}^{N-1}G_{jk}(\nu_{r})\left[i\nu_{r}a_{k}+b_{k}\right], (94)

where

[𝑮1(νr)]jk=νr2δjk+1MΛjk+Σjk(νr)\displaystyle[\bm{G}^{-1}(\nu_{r})]_{jk}=\nu^{2}_{r}\delta_{jk}+\frac{1}{M}\Lambda_{jk}+\Sigma_{jk}(\nu_{r}) (95)

and Σjk(νr)\Sigma_{jk}(\nu_{r}) is a diagonal matrix with non-vanishing elements ζr(α)\zeta^{(\alpha)}_{r} only for j=k=αj=k=\alpha, that is

Σjk(νr)=δjkαδjαζr(α).\displaystyle\Sigma_{jk}(\nu_{r})=\delta_{jk}\sum_{\alpha\in\mathcal{B}}\delta_{j\alpha}\zeta^{(\alpha)}_{r}. (96)

The discontinuity of xj(τ)x_{j}(\tau) at τ=0\tau=0 is responsible for term proportional to δ~(τ)\tilde{\delta}^{\prime}(\tau). In fact we can easily see that

aj=x¯jx¯j.\displaystyle a_{j}=\bar{x}^{\prime}_{j}-\bar{x}_{j}. (97)

The discontinuity at τ=0\tau=0 can also be read off from Eq. (94) by forming xj(τ)=rxj,rexp(iνrτ)x_{j}(\tau)=\sum_{r}x_{j,r}\exp(i\nu_{r}\tau) and taking the τ0\tau\to 0 limit. We first note that for large νr\nu_{r}, Gjk(νr)1/νr2G_{jk}(\nu_{r})\sim 1/\nu^{2}_{r}. Therefore the term proportional to aka_{k} in Eq. (94) contains a contribution which goes as 1/νr1/\nu_{r} for large rr. This kind of contribution when summed over all nonzero rr signals a discontinuity at τ=0\tau=0. More explicitly, we write a part of xj(τ)x_{j}(\tau) obtained from the first term on the right hand side of Eq. (94) as

1βr=k1iνr[𝒢jk+(νr2Gjk(νr)𝒢jk)]akeiνrτ,\displaystyle\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty\prime}\sum_{k}\frac{1}{i\nu_{r}}[\mathcal{G}_{jk}+(\nu^{2}_{r}G_{jk}(\nu_{r})-\mathcal{G}_{jk})]a_{k}e^{i\nu_{r}\tau}, (98)

where 𝒢jk=limzz2Gjk(z)\mathcal{G}_{jk}=\lim_{z\to\infty}z^{2}G_{jk}(z) and the prime indicates that the absence of the r=0r=0 term. When taking the τ0\tau\to 0 limit, the term in the round parenthesis vanishes as Gjk(νr)G_{jk}(\nu_{r}) is even in νr\nu_{r}. We also note that a sawtooth wave f(τ)=1/2τ/(β)f(\tau)=1/2-\tau/(\beta\hbar) defined in [0,β][0,\beta\hbar] has a Fourier expansion

f(τ)=1βr=1iνreiνrτ.\displaystyle f(\tau)=\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty\prime}\frac{1}{i\nu_{r}}e^{i\nu_{r}\tau}. (99)

Note that the sawtooth wave has a discontinuity as f(0+)=1/2f(0^{+})=1/2 and f(0)=1/2f(0^{-})=-1/2. Therefore, Eq. (98) becomes (1/2)k𝒢jkak(1/2)\sum_{k}\mathcal{G}_{jk}a_{k} as τ0+\tau\to 0^{+} and (1/2)k𝒢jkak-(1/2)\sum_{k}\mathcal{G}_{jk}a_{k} as τ0\tau\to 0^{-}. Finally by taking the τ=0±\tau=0^{\pm} limit of xj(τ)x_{j}(\tau), we have from Eq. (94)

xj(0+)=k[12𝒢jkakFjkbk]=x¯j\displaystyle x_{j}(0^{+})=\sum_{k}\left[\frac{1}{2}\mathcal{G}_{jk}a_{k}-F_{jk}b_{k}\right]=\bar{x}^{\prime}_{j} (100)
xj(0)=k[12𝒢jkakFjkbk]=x¯j,\displaystyle x_{j}(0^{-})=\sum_{k}\left[-\frac{1}{2}\mathcal{G}_{jk}a_{k}-F_{jk}b_{k}\right]=\bar{x}_{j}, (101)

where

Fjk1βr=Gjk(νr).\displaystyle F_{jk}\equiv\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}G_{jk}(\nu_{r}). (102)

Adding these two equations, we have

bj=12k=0N1(𝑭1)jk(x¯k+x¯k).\displaystyle b_{j}=-\frac{1}{2}\sum_{k=0}^{N-1}(\bm{F}^{-1})_{jk}\left(\bar{x}_{k}+\bar{x}^{\prime}_{k}\right). (103)

Inserting Eqs. (103) and (97) into Eq. (94), we have Eq. (20).

Appendix C Evaluation of the MFG State from the Stationary Solution

We write the actions inside the exponential of Eq. (14) using the stationary solution Eq. (20) with Eq. (94). We note that the derivative dxj/dτdx_{j}/d\tau expressed in terms of the Fourier modes contains the delta function singularities coming from the discontinuities. By removing these singularities and using the matrix notations 𝒂\bm{a} for Eq. (97) and 𝒃\bm{b} for Eq. (103), we can write

d𝒙(τ)dτ=1βr=iνr𝒙reiνrτ𝒂δ~(τ)\displaystyle\frac{d\bm{x}(\tau)}{d\tau}=\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}i\nu_{r}\bm{x}_{r}e^{i\nu_{r}\tau}-\bm{a}\tilde{\delta}(\tau) (104)
=\displaystyle= 1βr=[{𝟏νr2𝑮(νr)}𝒂+iνr𝑮(νr)𝒃)]eiνrτ.\displaystyle-\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}\Big{[}\left\{\bm{1}-\nu^{2}_{r}\bm{G}(\nu_{r})\right\}\bm{a}+i\nu_{r}\bm{G}(\nu_{r})\bm{b})\Big{]}e^{i\nu_{r}\tau}.

Then we have

M20β𝑑τd𝒙T(τ)dτd𝒙(τ)dτ=M21βr=\displaystyle\frac{M}{2}\int_{0}^{\beta\hbar}d\tau\;\frac{d\bm{x}^{\rm T}(\tau)}{d\tau}\frac{d\bm{x}(\tau)}{d\tau}=\frac{M}{2}\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty} [𝒂T{𝟏νr2𝑮(νr)}{𝟏νr2𝑮(νr)}𝒂\displaystyle\Big{[}\bm{a}^{\rm T}\left\{\bm{1}-\nu^{2}_{r}\bm{G}(\nu_{r})\right\}\left\{\bm{1}-\nu^{2}_{r}\bm{G}(\nu_{r})\right\}\bm{a}
+(iνr)(iνr)𝒃T𝑮(νr)𝑮(νr)𝒃],\displaystyle+(i\nu_{r})(-i\nu_{r})\bm{b}^{\rm T}\bm{G}(\nu_{r})\bm{G}(\nu_{r})\bm{b}\Big{]}, (105)

where we have used the fact that 𝑮(νr)\bm{G}(\nu_{r}) is even in νr\nu_{r} and that the terms odd in νr\nu_{r} vanish after the summation. Using Eq. (94), we have

120β𝑑τ𝒙T(τ)𝚲𝒙(τ)=12βr=[(iνr)(iνr)𝒂T𝑮(νr)𝚲𝑮(νr)𝒂+𝒃T𝑮(νr)𝚲𝑮(νr)𝒃],\displaystyle\frac{1}{2}\int_{0}^{\beta\hbar}d\tau\;\bm{x}^{\rm T}(\tau)\bm{\Lambda}\bm{x}(\tau)=\frac{1}{2\beta\hbar}\sum_{r=-\infty}^{\infty}\Big{[}(-i\nu_{r})(i\nu_{r})\bm{a}^{\rm T}\bm{G}(\nu_{r})\bm{\Lambda}\bm{G}(\nu_{r})\bm{a}+\bm{b}^{\rm T}\bm{G}(\nu_{r})\bm{\Lambda}\bm{G}(\nu_{r})\bm{b}\Big{]}, (106)

and from Eqs. (15) and (16)

αΨα[xα]=\displaystyle\sum_{\alpha\in\mathcal{B}}\Psi_{\alpha}[x_{\alpha}]= M2βr=αxα,rζr(α)xα,r=M2βr=𝒙rT𝚺(νr)𝒙r\displaystyle\frac{M}{2\beta\hbar}\sum_{r=-\infty}^{\infty}\sum_{\alpha\in\mathcal{B}}x_{\alpha,-r}\zeta^{(\alpha)}_{r}x_{\alpha,r}=\frac{M}{2\beta\hbar}\sum_{r=-\infty}^{\infty}\bm{x}^{\rm T}_{-r}\bm{\Sigma}(\nu_{r})\bm{x}_{r}
=\displaystyle= M2βr=[(iνr)(iνr)𝒂T𝑮(νr)𝚺(νr)𝑮(νr)𝒂+𝒃T𝑮(νr)𝚺(νr)𝑮(νr)𝒃],\displaystyle\frac{M}{2\beta\hbar}\sum_{r=-\infty}^{\infty}\Big{[}(-i\nu_{r})(i\nu_{r})\bm{a}^{\rm T}\bm{G}(\nu_{r})\bm{\Sigma}(\nu_{r})\bm{G}(\nu_{r})\bm{a}+\bm{b}^{\rm T}\bm{G}(\nu_{r})\bm{\Sigma}(\nu_{r})\bm{G}(\nu_{r})\bm{b}\Big{]}, (107)

Adding these three contributions in Eq. (14) and recalling the definition of 𝑮\bm{G} in Eq. (21), we have

𝒙¯|ρ^mG|𝒙¯=Cexp[M2β2r=[𝒂T{𝟏νr2𝑮(νr)}𝒂+𝒃T𝑮(νr)𝒃]],\displaystyle\langle\bm{\bar{x}}|\hat{\rho}_{\rm mG}|\bm{\bar{x}^{\prime}}\rangle=C\exp\Big{[}-\frac{M}{2\beta\hbar^{2}}\sum_{r=-\infty}^{\infty}\left[\bm{a}^{\rm T}\left\{\bm{1}-\nu^{2}_{r}\bm{G}(\nu_{r})\right\}\bm{a}+\bm{b}^{\rm T}\bm{G}(\nu_{r})\bm{b}\right]\Big{]}, (108)

Inserting Eqs. (97) and (103) into this, we finally arrive at Eq. (23).

Appendix D Diagonalization of Matrices for a Chain of Oscillators

We first note that the matrices 𝚲\bm{\Lambda} and 𝑽\bm{V} in Eq. (40) share the same eigenvectors. By direct application, we can check that the eigenvectors of 𝑽\bm{V} in Eq. (41) are given up to normalization constant in the form Serafini (2017)

𝒕k(1,e2πik/N,e2πi(2k)/N,,e2πi(N1)k/N)T,\displaystyle\bm{t}_{k}\sim(1,e^{2\pi ik/N},e^{2\pi i(2k)/N},\cdots,e^{2\pi i(N-1)k/N})^{\rm T}, (109)

for k=0,1,2,,N1k=0,1,2,\cdots,N-1 with eigenvalue λk=2cos(2πk/N)\lambda_{k}=2\cos(2\pi k/N). The eigenvalue of 𝚲\bm{\Lambda} is then given by Eq. (42). We note that there are two-fold degeneracies as λNk=λk\lambda_{N-k}=\lambda_{k}. Below, we construct the orthonormal set of eigenvectors for the cases of even and odd NN separately.

For NN odd, the space of λ0\lambda_{0} is one dimensional with normalized eigenvector

𝒕0=1N(1,1,,1)T.\displaystyle\bm{t}_{0}=\frac{1}{\sqrt{N}}(1,1,\cdots,1)^{\rm T}. (110)

On the other hand, the space of λk=λNk\lambda_{k}=\lambda_{N-k} with k=1,2,,(N1)/2k=1,2,\cdots,(N-1)/2 is doubly degenerate. We can form a linear combination of 𝒕k\bm{t}_{k} and 𝒕Nk\bm{t}_{N-k} which is proportional to (1,e2πik/N,e2πi(2k)/N,,e2πi(N1)k/N)T(1,e^{-2\pi ik/N},e^{-2\pi i(2k)/N},\cdots,e^{-2\pi i(N-1)k/N})^{\rm T} to make a real eigenvector for this eigenvalue. These are given by

𝒄k=2N(1,cos(2πNk),\displaystyle\bm{c}_{k}=\sqrt{\frac{2}{N}}\Big{(}1,\cos(\frac{2\pi}{N}k), cos(2πN2k),\displaystyle\cos(\frac{2\pi}{N}2k),\cdots
,cos(2πN(N1)k))T,\displaystyle\cdots,\cos(\frac{2\pi}{N}(N-1)k)\Big{)}^{\rm T}, (111)

and

𝒔k=2N(0,sin(2πNk),\displaystyle\bm{s}_{k}=\sqrt{\frac{2}{N}}\Big{(}0,\sin(\frac{2\pi}{N}k), sin(2πN2k),,\displaystyle\sin(\frac{2\pi}{N}2k),\cdots,
sin(2πN(N1)k))T\displaystyle\cdots\sin(\frac{2\pi}{N}(N-1)k)\Big{)}^{\rm T} (112)

We can easily show that the collection of 𝒕0,{𝒄k}\bm{t}_{0},\{\bm{c}_{k}\} and {𝒔k}\{\bm{s}_{k}\} for k=1,2,,(N1)/2k=1,2,\cdots,(N-1)/2 form an orthonormal set. Now we relabel these vectors by using {𝒖k}\{\bm{u}_{k}\} for k=0,1,2,,N1k=0,1,2,\cdots,N-1 such that 𝒖0=𝒕0\bm{u}_{0}=\bm{t}_{0} and 𝒖k=𝒄k\bm{u}_{k}=\bm{c}_{k} for k=1,2,,(N1)/2k=1,2,\cdots,(N-1)/2. For the remaining part k=(N+1)/2,(N+3)/2,,N1k=(N+1)/2,(N+3)/2,\cdots,N-1, we assign 𝒖k=𝒔Nk\bm{u}_{k}=\bm{s}_{N-k}. Then the set {𝒖k}\{\bm{u}_{k}\} with k=0,1,2,,N1k=0,1,2,\cdots,N-1 is the desired orthonormal set of eigenvectors of 𝑽\bm{V} corresponding to eigenvalue λk\lambda_{k}.

Therefore, we can form an orthogonal matrix 𝑹\bm{R} by putting the eigenvectors {𝒖k}\{\bm{u}_{k}\} on the columns as

𝑹=\displaystyle\bm{R}= (𝒖0|𝒖1||𝒖N1)\displaystyle\left(\bm{u}_{0}|\bm{u}_{1}|\cdots|\bm{u}_{N-1}\right)
=\displaystyle= (𝒕0|𝒄1||𝒄(N1)/2|𝒔(N1)/2||𝒔1).\displaystyle\left(\bm{t}_{0}|\bm{c}_{1}|\cdots|\bm{c}_{(N-1)/2}|\bm{s}_{(N-1)/2}|\cdots|\bm{s}_{1}\right). (113)

This matrix can then be used to diagonalize Λ\Lambda as 𝑹T𝚲𝑹=diag(MΩ02,,MΩN12)\bm{R}^{\rm T}\bm{\Lambda}\bm{R}=\mathrm{diag}(M\Omega^{2}_{0},\cdots,M\Omega^{2}_{N-1}).

Using this matrix 𝑹\bm{R}, we can calculate the zeroth order Green’s function in Eq. (43) as

Gij(0)(z)=\displaystyle G^{(0)}_{ij}(z)= t0it0jz2+Ω02+k=1(N1)/2ckickj+skiskjz2+Ωk2\displaystyle\frac{t^{i}_{0}t^{j}_{0}}{z^{2}+\Omega^{2}_{0}}+\sum_{k=1}^{(N-1)/2}\frac{c^{i}_{k}c^{j}_{k}+s^{i}_{k}s^{j}_{k}}{z^{2}+\Omega^{2}_{k}} (114)
=\displaystyle= 1N1z2+Ω02+2Nk=1(N1)/2cos(2π(ij)k/N)z2+Ωk2.\displaystyle\frac{1}{N}\frac{1}{z^{2}+\Omega^{2}_{0}}+\frac{2}{N}\sum_{k=1}^{(N-1)/2}\frac{\cos(2\pi(i-j)k/N)}{z^{2}+\Omega^{2}_{k}}.

This can be easily seen to be equivalent to Eq. (46).

The case of even NN is similar. We first note that the spaces corresponding to λ0\lambda_{0} and λN/2\lambda_{N/2} are both one dimensional with eigenvectors 𝒕0\bm{t}_{0} in Eq. (110) and

𝒕N/2=1N(1,1,1,1,,1,1)T,\displaystyle\bm{t}_{N/2}=\frac{1}{\sqrt{N}}(1,-1,1,-1,\cdots,1,-1)^{\rm T}, (115)

respectively. For the doubly degenerate spaces corresponding to k=1,2,,N/21k=1,2,\cdots,N/2-1, the eigenvectors are again given by 𝒄k\bm{c}_{k} and 𝒔k\bm{s}_{k} defined in Eqs. (111) and (112). Therefore, the orthonormal set {𝒖k}\{\bm{u}_{k}\} with k=0,1,,N1k=0,1,\cdots,N-1 is given by 𝒖0=𝒕0\bm{u}_{0}=\bm{t}_{0}, 𝒖N/2=𝒕N/2\bm{u}_{N/2}=\bm{t}_{N/2}, and 𝒖k=𝒄k\bm{u}_{k}=\bm{c}_{k} for k=1,2,,N/21k=1,2,\cdots,N/2-1. For k=N/2+1,N/2+2,,N1k=N/2+1,N/2+2,\cdots,N-1, we take 𝒖k=𝒔Nk\bm{u}_{k}=\bm{s}_{N-k}. In this case, the orthogonal matrix 𝑹\bm{R} we are after takes the form

𝑹=(𝒕0|𝒄1||𝒄N/21|𝒕N/2|𝒔N/21||𝒔1).\displaystyle\bm{R}=\left(\bm{t}_{0}|\bm{c}_{1}|\cdots|\bm{c}_{N/2-1}|\bm{t}_{N/2}|\bm{s}_{N/2-1}|\cdots|\bm{s}_{1}\right). (116)

We therefore have in this case

Gij(0)(z)\displaystyle G^{(0)}_{ij}(z) =1N1z2+Ω02+2Nk=1N/21cos(2π(ij)k/N)z2+Ωk2\displaystyle=\frac{1}{N}\frac{1}{z^{2}+\Omega^{2}_{0}}+\frac{2}{N}\sum_{k=1}^{N/2-1}\frac{\cos(2\pi(i-j)k/N)}{z^{2}+\Omega^{2}_{k}}
+1N(1)i+jz2+ΩN/22.\displaystyle+\frac{1}{N}\frac{(-1)^{i+j}}{z^{2}+\Omega^{2}_{N/2}}. (117)

Again this can equivalently be written as Eq. (46).

Appendix E Evaluation of the Density Matrix in the Ultra Strong Coupling Limit

We closely follow the procedure described in Appendices B and C to evaluate the matrix element of ρUSC)\rho_{\rm USC}) given in Eq. (64). The stationary path for the path integral satisfies

Md2x1(τ)dτ2+MΩ2x1(τ)2Mλx¯0=0.\displaystyle-M\frac{d^{2}x_{1}(\tau)}{d\tau^{2}}+M\Omega^{2}x_{1}(\tau)-2M\lambda\bar{x}_{0}=0. (118)

This equation is solved in terms of the Fourier component defined in Eq. (19) as

x1,r=g(νr)[2λx¯0(β)δr,0iνrab].\displaystyle x_{1,r}=g(\nu_{r})\left[2\lambda\bar{x}_{0}(\beta\hbar)\delta_{r,0}-i\nu_{r}a-b\right]. (119)

where

g(νr)=1νr2+Ω2.\displaystyle g(\nu_{r})=\frac{1}{\nu^{2}_{r}+\Omega^{2}}. (120)

Here aa and bb result from the discontinuities at the boundaries at τ=rβ\tau=r\beta\hbar and will be determined from the specific boundary conditions we require as we have done in Appendix B (see Eq. (93)). The discontinuity of x1(τ)x_{1}(\tau) at τ=0\tau=0 is responsible for aa, which is given by

a=x¯1x¯1.\displaystyle a=\bar{x}_{1}^{\prime}-\bar{x}_{1}. (121)

Now following the same argument leading up to Eqs. (100) and (101), we have

x¯1=a2+1βr=g(νr)[2λx¯0(β)δr,0b],\displaystyle\bar{x}_{1}^{\prime}=\frac{a}{2}+\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}g(\nu_{r})\left[2\lambda\bar{x}_{0}(\beta\hbar)\delta_{r,0}-b\right], (122)
x¯1=a2+1βr=g(νr)[2λx¯0(β)δr,0b].\displaystyle\bar{x}_{1}=-\frac{a}{2}+\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}g(\nu_{r})\left[2\lambda\bar{x}_{0}(\beta\hbar)\delta_{r,0}-b\right]. (123)

Adding these two, we have

b=1F[2λx¯0Ω2x¯1+x¯12],\displaystyle b=\frac{1}{F}\left[\frac{2\lambda\bar{x}_{0}}{\Omega^{2}}-\frac{\bar{x}_{1}+\bar{x}_{1}^{\prime}}{2}\right], (124)

where

F1βr=g(νr)=12Ωcoth(βΩ2).\displaystyle F\equiv\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}g(\nu_{r})=\frac{1}{2\Omega}\coth\left(\frac{\beta\hbar\Omega}{2}\right). (125)

We now put the stationary path solution Eq. (119) in the action in Eq. (65). Following the same steps leading to Eqs. (104) and (105), we have

M20β𝑑τ(dx1(τ)dτ)2\displaystyle\frac{M}{2}\int_{0}^{\beta\hbar}d\tau\;\left(\frac{dx_{1}(\tau)}{d\tau}\right)^{2} (126)
=\displaystyle= M21βr=[(1νr2g(νr))2a2+νr2g2(νr)b2].\displaystyle\frac{M}{2}\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}\Big{[}(1-\nu^{2}_{r}g(\nu_{r}))^{2}a^{2}+\nu^{2}_{r}g^{2}(\nu_{r})b^{2}\Big{]}.

We also obtain

MΩ220β𝑑τx12(τ)\displaystyle\frac{M\Omega^{2}}{2}\int_{0}^{\beta\hbar}d\tau\;x^{2}_{1}(\tau) (127)
=\displaystyle= MΩ22[1βr=g2(νr)(νr2a2+b2)4λx¯0Ω4b+4λ2βx¯02Ω4],\displaystyle\frac{M\Omega^{2}}{2}\Bigg{[}\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}g^{2}(\nu_{r})(\nu^{2}_{r}a^{2}+b^{2})-\frac{4\lambda\bar{x}_{0}}{\Omega^{4}}b+\frac{4\lambda^{2}\beta\hbar\bar{x}^{2}_{0}}{\Omega^{4}}\Bigg{]},

and

2Mλx¯00β𝑑τx1(τ)=2Mλx¯0Ω2[2λx¯0βb].\displaystyle-2M\lambda\bar{x}_{0}\int_{0}^{\beta\hbar}d\tau\;x_{1}(\tau)=-\frac{2M\lambda\bar{x}_{0}}{\Omega^{2}}[2\lambda\bar{x}_{0}\beta\hbar-b]. (128)

Adding these three terms together with (M/2)Ω2x¯02(M/2)\Omega^{2}\bar{x}^{2}_{0} term, we have the stationary path expression for the action in Eq. (65) as

SUSC\displaystyle S_{\rm USC} =M2[Aa2+Fb2+βΩ2{14λ2Ω4}x¯02]\displaystyle=\frac{M}{2}[Aa^{2}+Fb^{2}+\beta\hbar\Omega^{2}\{1-\frac{4\lambda^{2}}{\Omega^{4}}\}\bar{x}^{2}_{0}]
=M2[A(x¯1x¯1)2+1F(x¯1+x¯122λΩ2x¯0)2\displaystyle=\frac{M}{2}\Big{[}A(\bar{x}_{1}^{\prime}-\bar{x}_{1})^{2}+\frac{1}{F}\left(\frac{\bar{x}_{1}+\bar{x}_{1}^{\prime}}{2}-\frac{2\lambda}{\Omega^{2}}\bar{x}_{0}\right)^{2}
+βΩ2(14λ2Ω4)x¯02]\displaystyle~{}~{}~{}~{}~{}~{}~{}~{}+\beta\hbar\Omega^{2}\left(1-\frac{4\lambda^{2}}{\Omega^{4}}\right)\bar{x}^{2}_{0}\Big{]} (129)

where

A1βr=(1νr2g(νr))=Ω2coth(βΩ2).\displaystyle A\equiv\frac{1}{\beta\hbar}\sum_{r=-\infty}^{\infty}(1-\nu^{2}_{r}g(\nu_{r}))=\frac{\Omega}{2}\coth\left(\frac{\beta\hbar\Omega}{2}\right). (130)

Putting this expression for the action into Eq. (64), we obtain Eq. (66).

Appendix F The large-γ\gamma limit of p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG}

From Eq. (38), we have

p^02mfG=Mβ[12r=1{1νr2G00(νr)}].\displaystyle\langle\hat{p}^{2}_{0}\rangle_{\rm mfG}=\frac{M}{\beta}[1-2\sum_{r=1}^{\infty}\{1-\nu^{2}_{r}G_{00}(\nu_{r})\}]. (131)

If we use the Ohmic bath, we have from Eqs. (36) and (58)

G00(z)=(z+ωD)(z2+Ω2)P(z),\displaystyle G_{00}(z)=\frac{(z+\omega_{D})(z^{2}+\Omega^{2})}{P(z)}, (132)

where

P(z)=(z+ωD)(z2+Ω02)(z2+Ω12)+2γωDz(z2+Ω2)\displaystyle P(z)=(z+\omega_{D})(z^{2}+\Omega^{2}_{0})(z^{2}+\Omega^{2}_{1})+2\gamma\omega_{D}z(z^{2}+\Omega^{2}) (133)

for nonnegative frequency z=νr0z=\nu_{r}\geq 0. In order to evaluate p^02mfG\langle\hat{p}^{2}_{0}\rangle_{\rm mfG} from Eq. (131), we need to decompose

1z2G00(z)=i=15Diz+zi,\displaystyle 1-z^{2}G_{00}(z)=\sum_{i=1}^{5}\frac{D_{i}}{z+z_{i}}, (134)

where zi-z_{i} is the ii-th root of P(z)=i=15(z+zi)P(z)=\prod_{i=1}^{5}(z+z_{i}) and the coefficient DiD_{i} is determined from the method of residues. The large-γ\gamma behaviors of ziz_{i} are given by z1O(γ1)z_{1}\sim O(\gamma^{-1}), z2iΩ+O(γ1)z_{2}\sim-i\Omega+O(\gamma^{-1}), z3iΩ+O(γ1)z_{3}\sim i\Omega+O(\gamma^{-1}), and

z4i2ωDγ+O(γ1/2)\displaystyle z_{4}\sim-i\sqrt{2\omega_{D}\gamma}+O(\gamma^{-1/2}) (135)
z5i2ωDγ+O(γ1/2).\displaystyle z_{5}\sim i\sqrt{2\omega_{D}\gamma}+O(\gamma^{-1/2}). (136)

We can easily establish that in the large-γ\gamma limit, the leading order behavior of D1,D2D_{1},D_{2} and D3D_{3} are of O(1)O(1) and

D4iωDγ2,D5iωDγ2.\displaystyle D_{4}\sim-i\sqrt{\frac{\omega_{D}\gamma}{2}},~{}~{}D_{5}\sim i\sqrt{\frac{\omega_{D}\gamma}{2}}. (137)

Since D4+D5=0D_{4}+D_{5}=0 in this limit, the sum over the positive frequency in Eq. (131) can be written in terms of the digamma function ψ(x)\psi(x) Arfken and Weber (1995). If we write νr=rν\nu_{r}=r\nu with ν=2π/(β)\nu=2\pi/(\beta\hbar), we have

p02mfG\displaystyle\langle p^{2}_{0}\rangle_{\rm mfG} Mβ[2νi=45Diψ(1+ziν)]\displaystyle\sim\frac{M}{\beta}\left[-\frac{2}{\nu}\sum_{i=4}^{5}D_{i}\psi(1+\frac{z_{i}}{\nu})\right]
Mπi=45Dilnzi\displaystyle\sim-\frac{M\hbar}{\pi}\sum_{i=4}^{5}D_{i}\ln z_{i}
MωDγ2,\displaystyle\sim M\hbar\sqrt{\frac{\omega_{D}\gamma}{2}}, (138)

where we have used the asymptotic behavior of the digamma function Arfken and Weber (1995).

References