Opening the book…
The generalized Langevin equation (GLE) provides the exact equation of motion for a central coordinate coupled bilinearly to a harmonic bath, with the bath eliminated in favor of a memory kernel and a fluctuating force tied together by the fluctuation-dissipation theorem (FDT). While the standard textbook presentation specializes immediately to a harmonic central oscillator, the bath-elimination procedure itself never uses the harmonicity of the central potential-only its bilinear coupling to the bath. This appendix therefore derives the GLE for a completely general potential \(V(X)\) from the outset, and then specializes the single general result to three cases of increasing physical interest: an arbitrary smooth potential, a quartic (Duffing) correction that arises generically from the Taylor expansion of any smooth, symmetric potential minimum, and a cosine (Josephson-junction/pendulum) potential whose own Taylor expansion is shown to reproduce the quartic case with a fixed, universal softening coefficient. Using the quantummechanical FDT, we derive the self-consistent (Hartree) equation for the thermally and quantummechanically renormalized frequency \(\Omega_{\text {eff }}(T)\) of the quartic problem, and give a unified treatment of the two qualitatively different outcomes set by the sign of the quartic coefficient \(\beta_{D}\) : a hardening quartic \(\left(\beta_{D}>0\right)\) produces a frequency that rises and saturates smoothly \(\left(\Omega_{\text {eff }} \propto T^{1 / 4}\right.\) at high \(T\) ), while a softening quartic ( \(\beta_{D}<0\), as realized exactly by the cosine potential) produces a frequency that falls and the self-consistent solution terminates altogether at a finite Hartree instability temperature. We show that this instability has a precise dynamical meaning already visible in the linear theory: it is exactly the point at which the self-consistency condition on the renormalized pole of the retarded Green's function loses its solution. Finally we extend the self-consistency to the driven response amplitude itself, recovering the cubic equation responsible for Duffing bistability, and discuss which of its roots are dynamically stable and why.
B. 1 Hamiltonian and equations of motion
Consider a central coordinate \(X\) with mass \(M\), moving in an arbitrary potential \(V(X)\), bilinearly coupled to a bath of \(N\) independent harmonic oscillators labeled \(n=1, \ldots, N\), with coordinates \(q_{n}\), masses \(m_{n}\), and frequencies \(\omega_{n}\). Following Magalinski [1] (in the form popularized by Caldeira and Leggett [6, 7]), it is convenient to write the bath term with a built-in counterterm that removes an otherwise unphysical, instantaneous renormalization of the curvature of \(V\) :
\[ H=\frac{P^{2}}{2 M}+V(X)+\sum_{n=1}^{N}\left[\frac{p_{n}^{2}}{2 m_{n}}+\frac{1}{2} m_{n} \omega_{n}^{2}\left(q_{n}-\frac{c_{n}}{m_{n} \omega_{n}^{2}} X\right)^{2}\right], \tag{B.1} \label{B.1} \]
where \(c_{n}\) are coupling constants (dimensions of force). Expanding the square shows this is equivalent to the more commonly written bilinear-coupling form,
\[ H=\frac{P^{2}}{2 M}+V(X)+\sum_{n}\left(\frac{p_{n}^{2}}{2 m_{n}}+\frac{1}{2} m_{n} \omega_{n}^{2} q_{n}^{2}\right)-X \sum_{n} c_{n} q_{n}+X^{2} \sum_{n} \frac{c_{n}^{2}}{2 m_{n} \omega_{n}^{2}}, \tag{B.2} \label{B.2} \]
where the last term is exactly the counterterm needed to cancel the instantaneous frequency shift that the bath coupling would otherwise induce. Notice that \(V(X)\) enters Eq. (B.1) as a completely separate additive term: it plays no role whatsoever in the structure of the bath part of the Hamiltonian. Hamilton's equations give
\[ \begin{align*} M \ddot{X}(t)+V^{\prime}(X(t))-\sum_{n} c_{n} q_{n}(t)+X(t) \sum_{n} \frac{c_{n}^{2}}{m_{n} \omega_{n}^{2}} & =0, \tag{B.3} \label{B.3}\\[4pt] m_{n} \ddot{q}_{n}(t)+m_{n} \omega_{n}^{2} q_{n}(t) & =c_{n} X(t) . \tag{B.4} \end{align*} \]
Equation (B.4)-a driven harmonic oscillator for every bath mode, forced by the central coordinate- does not contain \(V(X)\) at all: the bath modes only ever see \(X(t)\) itself, never the force \(X\) experiences. This one observation is the crux of the entire appendix and is what allows Sec. B. 7 to specialize the result derived below to three completely different potentials without repeating any of the algebra.
B. 2 Eliminating the bath: the causal Green's function
The scalar operator for mode \(n\) is \(L_{n}=d^{2} / d t^{2}+\omega_{n}^{2}\), whose causal (retarded) Green's function, \(L_{n} g_{n}(t)=\delta(t)\) with \(g_{n}(t)=0\) for \(t<0\), is
\[ g_{n}(t)=\frac{\sin \left(\omega_{n} t\right)}{\omega_{n}} \Theta(t) . \tag{B.5} \label{B.5} \]
The general solution of Eq. (B.4), given initial data \(q_{n}(0), \dot{q}_{n}(0)\), is
\[ q_{n}(t)=q_{n}(0) \cos \left(\omega_{n} t\right)+\frac{\dot{q}_{n}(0)}{\omega_{n}} \sin \left(\omega_{n} t\right)+\frac{c_{n}}{m_{n}} \int_{0}^{t} g_{n}(t-s) X(s) d s . \tag{B.6} \label{B.6} \]
Substituting this into Eq. (B.3) and integrating the convolution piece by parts (using \(u=X(s)\), \(\left.d v=\left[\sin \omega_{n}(t-s) / \omega_{n}\right] d s\right)\) converts the memoryless convolution with \(X(s)\) into a convolution with \(\dot{X}(s)\); the resulting instantaneous term proportional to \(X(t) / \omega_{n}^{2}\) is exactly canceled by the counterterm already built into Eq. (B.1). Because none of this manipulation ever touched \(V(X)\)-it is purely a statement about how the bath responds to a given trajectory \(X(t)\)-it goes through for any \(V(X)\) exactly as it would for a harmonic potential. What survives is
\[ M \ddot{X}(t)+V^{\prime}(X(t))+\sum_{n} \frac{c_{n}^{2}}{m_{n} \omega_{n}^{2}} \int_{0}^{t} \cos \left[\omega_{n}(t-s)\right] \dot{X}(s) d s=\xi(t) \tag{B.7} \label{B.7} \]
with the fluctuating force collecting all the bath-initial-condition pieces,
\[ \xi(t) \equiv-\sum_{n} c_{n}\left[q_{n}(0)-\frac{c_{n}}{m_{n} \omega_{n}^{2}} X(0)\right] \cos \left(\omega_{n} t\right)-\sum_{n} c_{n} \frac{\dot{q}_{n}(0)}{\omega_{n}} \sin \left(\omega_{n} t\right) . \tag{B.8} \label{B.8} \]
B. 3 Memory kernel and the generalized Langevin equation
Defining the memory (damping) kernel, normalized by the system mass,
\[ \gamma(t) \equiv \frac{1}{M} \sum_{n} \frac{c_{n}^{2}}{m_{n} \omega_{n}^{2}} \cos \left(\omega_{n} t\right), \tag{B.9} \label{B.9} \]
Eq. (B.7) becomes the central result of this appendix, the classical GLE for an arbitrary central-oscillator potential \(V(X)\) :
\[ M \ddot{X}(t)+V^{\prime}(X(t))+M \int_{0}^{t} \gamma(t-s) \dot{X}(s) d s=\xi(t) \tag{B.10} \label{B.10} \]
This equation is exact for the bilinear bath model, Eq. (B.1), for any central-oscillator potential \(V(X)\) and any number of bath oscillators \(N\) (finite or, after the continuum limit below, infinite); it is causal, since the memory integral runs only over the system's own past; and, crucially, the memory kernel \(\gamma(t)\) and the noise \(\xi(t)\) depend only on the bath parameters \(\left\{c_{n}, m_{n}, \omega_{n}\right\}\)-not on \(V(X)\) at all. Section B. 7 exploits exactly this fact.
B. 4 Continuum limit, spectral density, and the Markovian limit
For a dense bath one introduces the spectral density \(J(\omega) \equiv \frac{\pi}{2} \sum_{n}\left(c_{n}^{2} / m_{n} \omega_{n}\right) \delta\left(\omega-\omega_{n}\right)\), in terms of which
\[ \gamma(t)=\frac{2}{M \pi} \int_{0}^{\infty} \frac{J(\omega)}{\omega} \cos (\omega t) d \omega . \tag{B.11} \label{B.11} \]
For an Ohmic bath with exponential cutoff, \(J(\omega)=M \eta \omega e^{-\omega / \omega_{c}}\),
\[ \gamma(t)=\frac{2 \eta}{\pi} \int_{0}^{\infty} e^{-\omega / \omega_{c}} \cos (\omega t) d \omega=\frac{2 \eta}{\pi} \frac{\omega_{c}}{1+\left(\omega_{c} t\right)^{2}}, \tag{B.12} \label{B.12} \]
plotted in Fig. \(\PageIndex{B1}\) for several cutoffs \(\omega_{c}\). As \(\omega_{c} \rightarrow \infty\) this family of functions approaches \(2 \eta \delta(t)\), the memory becomes instantaneous, and Eq. (B.10) reduces to the ordinary (Markovian) Langevin equation for a general potential,
\[ M \ddot{X}(t)+\gamma_{0} M \dot{X}(t)+V^{\prime}(X(t))=\xi(t), \quad \gamma_{0} \equiv 2 \eta . \tag{B.13} \label{B.13} \]
This Markovian limit is used throughout the rest of the appendix.
Figure \(\PageIndex{B1}\): The Ohmic memory kernel, Eq. (B.12), for several bath cutoffs \(\omega_{c}\). As \(\omega_{c}\) increases the kernel narrows and grows taller while its area stays fixed, converging to \(2 \eta \delta(t)\) : the Markovian, memoryless limit used for the rest of this appendix. Nothing in this figure depends on the form of \(V(X)\).
B. 5 Fluctuation-dissipation theorem
If the bath starts in thermal equilibrium at temperature \(T\), independent of the system, the fluctuating force \(\xi(t)\) of Eq. (B.8) has zero mean, and its correlator is fixed-not free to choose- by the very kernel \(\gamma(t)\) that damps the system, since both descend from the same coupling constants \(c_{n}\); this is Kubo's FDT [10, 11, 12]. Classically,
\[ \langle\xi(t)\rangle_{0}=0, \quad\left\langle\xi(t) \xi\left(t^{\prime}\right)\right\rangle_{0}=M k_{\mathrm{B}} T \gamma\left(t-t^{\prime}\right) . \tag{B.14} \label{B.14} \]
Quantum-mechanically, the symmetrized correlator instead involves the full Bose-thermal factor,
\[ S_{\xi}(\omega)=M \gamma(\omega) \hbar \omega \operatorname{coth}\left(\frac{\hbar \omega}{2 k_{\mathrm{B}} T}\right), \tag{B.15} \label{B.15} \]
where \(\gamma(\omega)\) is the Fourier transform of the memory kernel (for the Markovian limit, \(\gamma(\omega) \rightarrow \gamma_{0}\), a constant); the range of validity of this quantum Nyquist-type formula, and its relation to the Callen-Welton result, is discussed concisely by Ginzburg and Pitaevskiĭ [13] and at length by Weiss [9]. Equation (B.15) reduces to Eq. (B.14) whenever \(\hbar \omega \ll k_{\mathrm{B}} T\) over the frequencies that matter, and to a temperature-independent zero-point plateau, \(S_{\xi}(\omega) \rightarrow M \gamma_{0} \hbar|\omega|\), in the opposite limit. Like \(\gamma(t)\) itself, neither Eq. (B.14) nor Eq. (B.15) makes any reference to \(V(X)\) : they are generic consequences of the bath model of Sec. B.1, independent of any particular physical embodiment of the bath or of the potential the central coordinate happens to sit in.
B. 6 Linear stability: the retarded Green's function and its poles
For a general \(V(X)\), Eq. (B.10) is nonlinear and has no closed-form Laplace solution. However, linearizing about a stable equilibrium \(X=0\) of \(V(X)\) (i.e. \(V^{\prime}(0)=0, V^{\prime \prime}(0) \equiv M \Omega^{2}>0\) ) reduces Eq. (B.10) to the linear GLE
\[ M \ddot{X}(t)+M \Omega^{2} X(t)+M \int_{0}^{t} \gamma(t-s) \dot{X}(s) d s=\xi(t) \tag{B.16} \label{B.16} \]
which is Magalinski's original problem, and is also exact for the strictly harmonic case of Sec. B.7.1 below. Equation (B.16) can be solved formally in the Laplace domain, \(\tilde{f}(s)=\) \(\int_{0}^{\infty} e^{-s t} f(t) d t\). Using the convolution theorem,
\[ M\left[s^{2} \tilde{X}(s)-s X(0)-\dot{X}(0)\right]+M \Omega^{2} \tilde{X}(s)+M \tilde{\gamma}(s)[s \tilde{X}(s)-X(0)]=\tilde{\xi}(s), \tag{B.17} \label{B.17} \]
which defines the retarded Green's function \(\tilde{G}(s)=1 /\left\{M\left[s^{2}+\Omega^{2}+s \tilde{\gamma}(s)\right]\right\}\), satisfying \(M \ddot{G}+\) \(M \Omega^{2} G+M \int_{0}^{t} \gamma(t-s) \dot{G}(s) d s=\delta(t), G(t)=0\) for \(t<0\). The system's response is then \(X(t)=X_{\mathrm{hom}}(t)+\frac{1}{M} \int_{0}^{t} G(t-s) \xi(s) d s\), with \(X_{\text {hom }}\) the initial-condition transient.
The poles of \(\tilde{G}(s)\)-the roots, in \(s\), of \(s^{2}+\Omega^{2}+s \tilde{\gamma}(s)=0\)-are the normal-mode frequencies of the damped system, and their real parts determine stability: a physically sensible, damped, oscillatory system has all poles in \(\Re(s) \leq 0\). In the Markovian limit this quadratic is elementary, \(s=-\gamma_{0} / 2 \pm \sqrt{\Omega^{2}-\gamma_{0}^{2} / 4}\), manifestly stable (damped) for any \(\Omega^{2}>0\) and \(\gamma_{0}>0\). This single fact-that the pole structure requires \(\Omega^{2}>0\) for a bounded, oscillatory, causal response- is the thread that connects the purely linear theory of this section to the stability discussion of Sec. B.8.4 below: once thermal fluctuations renormalize \(\Omega^{2} \rightarrow \Omega_{\text {eff }}^{2}(T)\) through the quartic nonlinearity of Sec.B.7.2, the identical pole criterion, applied now to the renormalized frequency, is what decides whether the mean-field, self-consistent description remains a stable, physically meaningful oscillator at all.
B. 7 Case studies: general potential, quartic, and cosine
Equation (B.10), together with the memory kernel (B.9) and the FDT of Sec. B.5, holds for any \(V(X)\). We now specialize it to three choices of increasing physical richness, without repeating any bath-elimination algebra: only the force term \(V^{\prime}(X)\) changes from one case to the next.
B.7.1 General potential
The most general choice is an arbitrary smooth potential \(V(X)\) with a nondegenerate minimum at \(X=0\) [Fig. B.2(a)], for which the force \(V^{\prime}(X)\) is left unspecified and Eq. (B.10) is the general nonlinear GLE
\[ M \ddot{X}(t)+V^{\prime}(X(t))+M \int_{0}^{t} \gamma(t-s) \dot{X}(s) d s=\xi(t) \tag{B.18} \label{B.18} \]
This is the master equation from which all subsequent cases follow by choosing \(V(X)\). If \(V(X)\) is harmonic, \(V(X)=\frac{1}{2} M \Omega^{2} X^{2}\), then \(V^{\prime}(X)=M \Omega^{2} X\) and Eq. (B.18) reduces to the linear GLE (B.16) of Sec. B.6, with no approximation required: the Laplace/pole analysis given there is exact for this case. If \(V(X)\) is quartic or cosine, the force is nonlinear and the analysis of Secs. B.7.2 B.7.3 applies. In every case the memory kernel \(\gamma(t)\) and the noise \(\xi(t)\) are identical to those of the harmonic problem: the bath does not know what potential the central coordinate sits in.
B.7.2 Quartic (Duffing) oscillator
Any smooth potential with a nondegenerate minimum at \(X=0\) can be Taylor expanded there; if the potential is symmetric enough that no cubic term appears (e.g. by an inversion symmetry \(X \rightarrow-X\), as in both examples of this appendix), the leading anharmonic correction beyond the harmonic term is quartic,
\[ V(X)=V(0)+\frac{1}{2} V^{\prime \prime}(0) X^{2}+\frac{1}{24} V^{\prime \prime \prime \prime}(0) X^{4}+O\left(X^{6}\right) . \tag{B.19} \label{B.19} \]
Writing \(M \Omega^{2} \equiv V^{\prime \prime}(0)\) and \(M \beta_{D} \equiv V^{\prime \prime \prime \prime}(0) / 6\) gives the Duffing potential
\[ V(X)=\frac{1}{2} M \Omega^{2} X^{2}+\frac{1}{4} M \beta_{D} X^{4}, \tag{B.20} \label{B.20} \]
shown in Fig. B.2(b) for both signs of \(\beta_{D}\) : hardening \(\left(\beta_{D}>0\right)\) or softening \(\left(\beta_{D}<0\right)\), depending on the physical system. Since Eq. (B.10) depends on \(V(X)\) only through the force \(V^{\prime}(X)=\) \(M \Omega^{2} X+M \beta_{D} X^{3}\)-and, as emphasized in Sec. B.1, the bath elimination never touches this term at all-specializing Eq. (B.10) to the quartic potential is immediate, giving the noisy Duffing generalized Langevin equation,
\[ M \ddot{X}(t)+M \Omega^{2} X(t)+M \beta_{D} X^{3}(t)+M \int_{0}^{t} \gamma(t-s) \dot{X}(s) d s=\xi(t) \tag{B.21} \label{B.21} \]
with exactly the same memory kernel \(\gamma(t)\), Eq. (B.9), and the exactly same fluctuating force \(\xi(t)\), Eq. (B.8), obeying the exactly same classical and quantum FDTs, Eqs. (B.14) and (B.15), as in the strictly harmonic case. Taking the Markovian limit of Sec. B. 4 reduces Eq. (B.21) to
\[ \ddot{X}(t)+\gamma_{0} \dot{X}(t)+\Omega^{2} X(t)+\beta_{D} X^{3}(t)=\eta(t), \quad \eta(t) \equiv \xi(t) / M, \tag{B.22} \label{B.22} \]
with \(\left\langle\eta(t) \eta\left(t^{\prime}\right)\right\rangle_{0}=2 D \delta\left(t-t^{\prime}\right)\) classically \(\left(D=\gamma_{0} k_{\mathrm{B}} T / M\right)\), or the full quantum-mechanical noise spectrum of Eq. (B.15) divided by \(M^{2}\) in general. Equation (B.22) is the noisy Duffing-Langevin equation whose thermal renormalization and stability occupy the rest of this appendix (Secs. B. 8 B. 9 ).
B.7.3 Cosine (Josephson-junction / pendulum) potential
A third, physically important case is the cosine (washboard/pendulum) potential,
\[ V(X)=E_{J}[1-\cos X], \tag{B.23} \label{B.23} \]
shown in Fig. B.2(c), where \(E_{J}\) sets the energy scale (for a current-biased Josephson junction, \(E_{J}\) is the Josephson coupling energy and \(X\) is the gauge-invariant phase difference across the junction; for a simple pendulum, \(E_{J}=M g L\) and \(X\) is the angular displacement). Here \(V^{\prime}(X)=\) \(E_{J} \sin X\), and Eq. (B.10) specializes immediately, again with no new derivation, to
\[ M \ddot{X}(t)+E_{J} \sin X(t)+M \int_{0}^{t} \gamma(t-s) \dot{X}(s) d s=\xi(t) \tag{B.24} \label{B.24} \]
the full noisy pendulum (or RCSJ-Josephson) generalized Langevin equation. Expanding \(\sin X \approx\) \(X-X^{3} / 6\) for small \(X\) recovers exactly the quartic case of Sec. B.7.2,
\[ E_{J} \sin X \approx E_{J} X-\frac{E_{J}}{6} X^{3} \quad \Longrightarrow \quad \Omega^{2}=\frac{E_{J}}{M}, \quad \beta_{D}=-\frac{E_{J}}{6 M}=-\frac{\Omega^{2}}{6}, \tag{B.25} \label{B.25} \]
so that the cosine potential is not merely an example of a softening quartic oscillator: it fixes the softening coefficient to the specific, universal ratio \(\beta_{D}=-\Omega^{2} / 6\), independent of the overall energy scale \(E_{J}\). Every result derived below for a general softening \(\beta_{D}<0\) therefore applies to the Josephson/pendulum problem with this particular substitution, and Fig. \(\PageIndex{B2c}\) shows explicitly how good the quartic approximation is (and where it fails) by overlaying the truncated Taylor series on the exact cosine.
Figure \(\PageIndex{B2}\): The three potentials of Sec. B.7. (a) General smooth potential with a nondegenerate minimum at \(X=0\), Sec. B.7.1. (b) Quartic (Duffing), Sec. B.7.2, for both signs of \(\beta_{D}\); the softening curve \(\left(\beta_{D}<0\right)\) is the generic shape produced by truncating any cosine-like potential. (c) The cosine (Josephson/pendulum) potential of Sec. B.7.3, with its quartic (Taylor) approximation overlaid: the two agree closely near \(X=0\) and diverge once \(|X|\) is no longer small, marking the boundary of validity of the Duffing approximation to the full problem.
B. 8 Thermal (Hartree) renormalization of the vibrational frequency
We now specialize to the quartic case of Sec. B.7.2-which, by Sec. B.7.3, includes the cosine potential's small-oscillation regime as the particular case \(\beta_{D}=-\Omega^{2} / 6\)-and ask how the thermal bath renormalizes \(\Omega\).
B.8.1 Mean-field decoupling of the quartic term
Split \(X=X_{\text {coh }}+\delta X\) into a slowly varying coherent part and a rapidly fluctuating thermal part, and average the cubic restoring force over the noise at fixed \(X_{\text {coh }}\). Since \(\delta X\) is a zero-mean Gaussian process to this order, its odd moments vanish and \(\left\langle X^{3}\right\rangle_{\eta}=X_{\text {coh }}^{3}+3 \sigma_{\text {th }}^{2} X_{\text {coh }}+O\left(\beta_{D}^{2}\right)\), with \(\sigma_{\mathrm{th}}^{2} \equiv\left\langle\delta X^{2}\right\rangle\). The coherent equation of motion therefore acquires a mean-field-shifted linear coefficient,
\[ \ddot{X}_{\mathrm{coh}}+\gamma_{0} \dot{X}_{\mathrm{coh}}+\Omega_{\mathrm{eff}}^{2} X_{\mathrm{coh}}+\beta_{D} X_{\mathrm{coh}}^{3}=f \cos \left(\omega_{d} t\right), \quad \Omega_{\mathrm{eff}}^{2} \equiv \Omega^{2}+3 \beta_{D} \sigma_{\mathrm{th}}^{2} . \tag{B.26} \label{B.26} \]
The sign of \(\beta_{D}\) decides everything that follows: thermal and quantum fluctuations raise the vibrational frequency when the quartic term is hardening, and lower it when the quartic term is softening-which, by Eq. (B.25), is always the case for a cosine potential.
B.8.2 Quantum thermal variance from the FDT
For weak Markovian damping \(\left(\gamma_{0} \ll \Omega\right)\), the narrow-resonance form of the quantum FDT (Sec. B.5) gives the standard result for the thermal variance of a weakly damped quantum oscillator [11, 9],
\[ \sigma_{\mathrm{th}}^{2}=\frac{\hbar}{2 M \Omega} \operatorname{coth}\left(\frac{\hbar \Omega}{2 k_{\mathrm{B}} T}\right), \tag{B.27} \label{B.27} \]
with the familiar classical \(\left(k_{\mathrm{B}} T \gg \hbar \Omega\right)\) and zero-point \(\left(k_{\mathrm{B}} T \ll \hbar \Omega\right)\) limits,
\[ \sigma_{\mathrm{th}}^{2} \rightarrow \frac{k_{\mathrm{B}} T}{M \Omega^{2}} \quad(\text { classical }), \quad \sigma_{\mathrm{th}}^{2} \rightarrow \frac{\hbar}{2 M \Omega} \quad(\text { zero-point }) . \tag{B.28} \label{B.28} \]
B.8.3 Self-consistent gap equation; hardening vs. softening
Promoting Eq. (B.26) to self-consistency (evaluating \(\sigma_{\text {th }}^{2}\) at \(\Omega_{\text {eff }}\) rather than the bare \(\Omega\) ) gives the Hartree gap equation
\[ \Omega_{\mathrm{eff}}^{2}=\Omega^{2}+\frac{3 \beta_{D} \hbar}{2 M \Omega_{\mathrm{eff}}} \operatorname{coth}\left(\frac{\hbar \Omega_{\mathrm{eff}}}{2 k_{\mathrm{B}} T}\right) . \tag{B.29} \label{B.29} \]
In the classical regime this reduces to the algebraic equation \(\Omega_{\text {eff }}^{4}-\Omega^{2} \Omega_{\text {eff }}^{2}-3 \beta_{D} k_{\mathrm{B}} T / M=0\). Writing \(y \equiv \Omega_{\text {eff }}^{2} / \Omega^{2}\) and the dimensionless temperature \(\tau \equiv 3\left|\beta_{D}\right| k_{\mathrm{B}} T /\left(M \Omega^{4}\right)\), the physical root is
\[ \begin{align*} & \text { hardening }\left(\beta_{D}>0\right): y(\tau)=\frac{1+\sqrt{1+4 \tau}}{2}, \tag{B.30} \label{B.30}\\[4pt] & \text { softening }\left(\beta_{D}<0\right): y(\tau)=\frac{1+\sqrt{1-4 \tau}}{2}, \quad \tau \leq \frac{1}{4} . \tag{B.31} \end{align*} \]
Figure B. 3 plots both branches side by side. The hardening branch grows without bound, saturating at high temperature as \(y \sim \sqrt{\tau}\), i.e. \(\Omega_{\text {eff }} \propto T^{1 / 4}\) : a smooth, physically unremarkable crossover. The softening branch, in stark contrast, terminates: the square root in Eq. (B.31) becomes imaginary beyond \(\tau_{\text {max }}=1 / 4\), where \(y \rightarrow 1 / 2\). Beyond this point no real, self-consistent solution of Eq. (B.29) exists at all-the Hartree instability already anticipated in the Introduction, and, via Eq. (B.25), a generic feature of every cosine-potential (Josephson or pendulum) problem once it is heated enough.
B.8.4 Connecting back to the linear theory: what the Hartree instability actually is
Section B. 6 showed that the Markovian GLE, linearized about \(X=0\), has stable, causal, oscillatory poles only when the coefficient multiplying \(X\) is strictly positive: \(\Omega^{2}>0\). Equation (B.26)
Figure \(\PageIndex{B1}\): Self-consistent (classical-regime) thermal renormalization of the vibrational frequency, Eqs. (B.30)-(B.31). (a) Hardening quartic: the frequency rises and saturates smoothly, \(\Omega_{\mathrm{eff}} \propto T^{1 / 4}\), with no upper bound on \(\tau\). (b) Softening quartic-including the cosine potential of Sec. B.7.3, for which \(\beta_{D}=-\Omega^{2} / 6\) exactly: the frequency falls and the self-consistent branch terminates cleanly at \(\tau_{\text {max }}=1 / 4\), where \(y=\Omega_{\text {eff }}^{2} / \Omega^{2} \rightarrow 1 / 2\)-the Hartree instability discussed in Sec. B.8.4. shows that the Hartree-renormalized problem is, at the level of the coherent (mean-field) dynamics, an ordinary linear oscillator with \(\Omega^{2}\) replaced by \(\Omega_{\text {eff }}^{2}\); the pole criterion of Sec. B. 6 therefore applies verbatim, with \(\Omega \rightarrow \Omega_{\text {eff }}(T)\). As \(\tau \rightarrow \tau_{\text {max }}^{-}\)for a softening quartic, \(\Omega_{\text {eff }}^{2} \rightarrow \Omega^{2} / 2>0\) still-the branch shown in Fig. B.3(b) never itself passes through zero frequency. What actually happens at \(\tau_{\text {max }}\) is that the self-consistency condition itself loses its solution: the equation \(\Omega_{\text {eff }}^{2}=\Omega^{2}+3 \beta_{D} \sigma_{\text {th }}^{2}\left(\Omega_{\text {eff }}\right)\), read as a fixed-point problem, has a solution branch that folds back on itself and disappears in exactly the same saddle-node (fold) sense as the disappearance of a root pair in the bistability cubic of Sec. B. 9 below. In plain terms: as the softening well is heated, each increment of thermal spread \(\sigma_{\text {th }}^{2}\) lowers \(\Omega_{\text {eff }}\), which itself raises \(\sigma_{\text {th }}^{2}\) further (since \(\sigma_{\text {th }}^{2} \propto 1 / \Omega_{\text {eff }}^{2}\) classically)-a positive feedback loop that a hardening quartic does not have (there, softening feeds back negatively: lower \(\sigma_{\mathrm{th}}^{2}\) growth per unit \(\Omega_{\text {eff }}\) increase, which is why that branch saturates rather than runs away). Beyond \(\tau_{\text {max }}\) the feedback loop has no fixed point left to converge to, and the mean-field, small-oscillation description of the well breaks down-the physical system must be re-described either by the full cosine potential of Sec. B.7.3 or, physically, by thermally activated escape over the now-shallow barrier. The Hartree instability is thus not a failure of the linear stability criterion of Sec. B. 6 applied naively; it is a breakdown of the assumption, made in this section, that the quartic term can be treated as a small, self-consistently determined correction to an otherwise-good harmonic approximation-exactly the regime, visualized in Fig. B.2(c), in which the quartic curve has already begun to visibly depart from the true cosine.
B. 9 Stability of the driven, bistable response
The renormalization \(\Omega \rightarrow \Omega_{\text {eff }}(T)\) carries through into the driven problem unchanged, exactly as it did for the memory kernel and the FDT in Sec. B.7.2. Writing the driven, self-consistent fundamental-response amplitude as \(u \equiv\left|A_{1}\right|^{2}\), gives the self-consistency condition
\[ \left\{\left(\Omega_{\mathrm{eff}}^{2}-\omega_{d}^{2}+3 \beta_{D} f^{2} u\right)^{2}+\gamma_{0}^{2} \omega_{d}^{2}\right\} u=1, \tag{B.32} \label{B.32} \]
APPENDIX B. GENERALIZED LANGEVIN DYNAMICS FOR AN ARBITRARY POTENTIAL: THERMAL
an algebraic cubic equation in \(u\) (expand the square). A cubic with real coefficients has either three real roots or one real root together with a complex-conjugate pair; \(u=\left|A_{1}\right|^{2}\) must be real and non-negative to describe an actual oscillation amplitude, so a complex pair, when present, simply plays no physical role, and the single real positive root left over is the unique, unambiguously monostable response. When three real positive roots occur, only the smallest and the largest are dynamically stable; the middle root is the classic unstable branch of a folded (S-shaped) response curve, exactly as for the ordinary (non-thermally-renormalized) Duffing oscillator, and its presence is what produces hysteresis as the drive frequency or amplitude is swept back and forth. Because Eq. (B.32) depends on temperature only through \(\Omega_{\text {eff }}(T)\), the entire bistable window-its location, its width, and (for a softening quartic) its eventual disappearance-is carried along rigidly by the same thermal renormalization derived in Sec. B.8, with the two qualitatively different outcomes of Fig. B. 3 now manifesting as a bistable window that either drifts smoothly to higher frequency without limit (hardening) or drifts to lower frequency and is extinguished altogether at a finite temperature (softening)-a purely thermal quenching of bistability with no counterpart in the zero-temperature theory.
B. 10 Summary
- 1.Eliminating a bath of \(N\) harmonic oscillators bilinearly coupled to a central coordinate of arbitrary potential \(V(X)\), exactly following Magalinski's construction, gives the generalized Langevin equation (B.10): a memory kernel \(\gamma(t)\) and a fluctuating force \(\xi(t)\) tied together by the fluctuation-dissipation theorem, both fixed entirely by the bath's spectral density and completely independent of \(V(X)\) (Sec. B.1).
- 2.Specializing this one general result-with no new bath-elimination algebra-produces the GLE for a general smooth potential, the noisy Duffing GLE for a quartic well, and the noisy pendulum/Josephson GLE for a cosine potential; the latter's own Taylor expansion fixes the quartic coefficient to the universal softening value \(\beta_{D}=-\Omega^{2} / 6\) (Sec. B.7).
- 3.The quantum-mechanical FDT gives the thermal variance \(\sigma_{\mathrm{th}}^{2}=(\hbar / 2 M \Omega) \operatorname{coth}\left(\hbar \Omega / 2 k_{\mathrm{B}} T\right)\), and Hartree self-consistency turns the frequency shift into a gap equation whose solution hardens smoothly \(\left(\propto T^{1 / 4}\right)\) for \(\beta_{D}>0\) but softens and terminates at a finite Hartree instability temperature for \(\beta_{D}<0\) (Sec. B.8, Fig. B.3).
- 4.That instability is not a violation of the linear stability criterion \(\Omega_{\text {eff }}^{2}>0\) of Sec. B.6; it is the disappearance, via a saddle-node fold, of the self-consistency condition's own solution branch, driven by a positive feedback between frequency softening and thermal spread that has no counterpart in the hardening case (Sec. B.8.4).
- 5.The same renormalization propagates into the driven response, where a cubic self-consistency equation for the amplitude produces bistability with two stable and one unstable branch; the entire bistable window is carried along by \(\Omega_{\text {eff }}(T)\), either advected indefinitely (hardening) or extinguished at finite temperature (softening) (Sec. B.9).
Bibliography
- [1] V. B. Magalinski, "Dynamical Model in the Theory of the Brownian Motion," Sov. Phys. JETP 9, 1381-1382 (1959).
- [2] P. Ullersma, "An exactly solvable model for Brownian motion," Physica 32, 27-55, 56-73, 74-89, 90-96 (1966).
- [3] G. W. Ford, M. Kac, and P. Mazur, "Statistical Mechanics of Assemblies of Coupled Oscillators," J. Math. Phys. 6, 504-515 (1965).
- [4] R. Zwanzig, "Nonlinear generalized Langevin equations," J. Stat. Phys. 9, 215-220 (1973).
- [5] R. Zwanzig, Nonequilibrium Statistical Mechanics (Oxford University Press, New York, 2001).
- [6] A. O. Caldeira and A. J. Leggett, "Influence of Dissipation on Quantum Tunneling in Macroscopic Systems," Phys. Rev. Lett. 46, 211-214 (1981).
- [7] A. O. Caldeira and A. J. Leggett, "Quantum tunnelling in a dissipative system," Ann. Phys. 149, 374-456 (1983).
- [8] A. Ghosh, M. Bandyopadhyay, S. Dattagupta, and S. Gupta, "Independent-oscillator model and the quantum Langevin equation for an oscillator: a review," J. Stat. Mech.: Theory Exp. 2024(7), 074002 (2024).
- [9] U. Weiss, Quantum Dissipative Systems, 4th ed. (World Scientific, Singapore, 2012).
- [10] R. Kubo, "Statistical-Mechanical Theory of Irreversible Processes. I.," J. Phys. Soc. Jpn. 12, 570-586 (1957).
- [11] R. Kubo, "The fluctuation-dissipation theorem," Rep. Prog. Phys. 29, 255-284 (1966).
- [12] R. Kubo, M. Toda, and N. Hashitsume, Statistical Physics II: Nonequilibrium Statistical Mechanics, 2nd ed. (Springer, Berlin, 1991).
- [13] V. L. Ginzburg and L. P. Pitaevskiĭ, "Quantum Nyquist formula and the applicability ranges of the Callen-Welton formula," Sov. Phys. Usp. 30(2), 168-171 (1987).
This page titled 17: Generalized Langevin Dynamics for an Arbitrary Potential- Thermal Renormalization and Stability was last modified on Mon, 02 Nov 2026 05:51:20 GMT and is shared under a CC BY-NC-SA 4.0 license and was authored, remixed, and/or curated by Pabitra N. Sen.
The book could not be loaded just now.