6: Power Balance and the Necessity of Dissipation- Pedagogical Derivation of the Generalized Langevin Equation

Fluctuations Near Equilibrium · LibreTexts

0%

Contents

    Text

    Size
    100%
    Theme
    Typeface
    Line length
    Spacing

    Read aloud

    How

    Speed: 1×
    Light up
    Highlight

    Symbols: Mulberry Symbols, © 2018–2026 Steve Lee, CC BY-SA 4.0 (licence, the set), the site's own copies.

    From LibreTexts, under CC BY-NC-SA 4.0, as the source states it; its equations are written in TeX, which the PDF shows typeset. This page at LibreTexts.

    6: Power Balance and the Necessity of Dissipation- Pedagogical Derivation of the Generalized Langevin Equation

    The well-known power absorption formula shows that an undamped oscillator undergoes an "energy catastrophe" due to resonant singularities, i.e., the system energy grows linearly in time, without bound, violating the postulates of equilibrium statistical mechanics. An undamped oscillator driven by stochastic forcing absorbs power at a constant rate, leading to \(\langle E\rangle \propto t \rightarrow \infty\), which contradicts the finite energy expected in thermal equilibrium. A calculation using the generalized Langevin equation (GLE) shows that the memory-dependent damping kernel regulates the susceptibility to arrest energy growth through exact frequency-by-frequency power balance. Under the fluctuation-dissipation theorem, this mechanism naturally recovers the equipartition theorem \(\langle E\rangle=k_{B} T\), demonstrating that dissipation is not merely phenomenological but rather a thermodynamic necessity for systems in thermal contact.

    6.1 Introduction

    The interplay between fluctuations and dissipation lies at the heart of non-equilibrium statistical mechanics. A fundamental question concerns the mechanism by which macroscopic systems driven by random forces reach steady states. Phenomenologically, one introduces damping forces to model energy dissipation [4, 5]; however, the deeper connection between stochastic driving and dissipative response reveals a more profound principle: dissipation is not an independent ingredient but rather emerges as a thermodynamic necessity for maintaining equilibrium in the presence of fluctuations.

    A particularly transparent framework for studying this balance is provided by the harmonic oscillator subjected to stochastic forcing. The linearity of the system allows the response to be characterized entirely by a frequency-dependent susceptibility. This enables exact derivations of energy balance relations in frequency space and precise identification of the mechanisms preventing unbounded energy growth.

    A direct model going from microscopic Newtonian motion to macroscopic dissipative equations was given in a landmark paper by Magalinski [1] in 1959. Magalinski solved a mechanical model of the motion of a harmonic oscillator which is coupled to \(N\) other harmonic oscillators and showed that it undergoes Brownian motion, i.e., obeys the generalized Langevin equation [4] with a non-Markovian memory kernel. This stands in contrast to heuristic derivations of the Langevin equation where a stochastic force and a dissipative drag are postulated. Magalinski's model derives these effects analytically.

    The rigorous, model-independent derivation of the GLE typically uses projection operator

    techniques [3, 2]. While rigorous, these methods can obscure the fundamental physical requirement that links dissipation to thermal fluctuations.

    Here, we adopt a complementary pedagogical route: we demonstrate that the structure of the GLE, including the crucial relationship between the damping kernel and the noise spectrum, follows uniquely from fundamental principles of energy balance, stationarity, and causality.

    Our analysis proceeds as follows. First, using the well-known power dissipation formula, we demonstrate the "energy catastrophe" that befalls an undamped oscillator: resonant singularities in the susceptibility lead to infinite power absorption and linear energy growth, violating equilibrium. Second, we show how the GLE introduces a frequency-dependent dissipation mechanism that exactly balances stochastic energy injection at every frequency, ensuring a stable thermal state consistent with the fluctuation-dissipation theorem. For completeness, we provide a derivation of the general spectral form of power injection for linear systems. In Appendix, we prove the well-known result that only the imaginary (dissipative) part of the susceptibility contributes to net energy absorption. The logical chain connecting these results is summarized in Table 6.1.

    This work is intended to provide both pedagogical value for students entering the field and conceptual clarity for researchers: dissipation emerges not as an ad hoc phenomenological addition but as a logical consequence of demanding thermodynamic consistency.

    Table 6.1: Summary of the logical chain from susceptibility structure to thermodynamic consistency.
    System Susceptibility Consequence
    Undamped oscillator \(\operatorname{Im}[\chi(\omega)] \propto \delta\left(\omega \pm \omega_{0}\right)\) Energy catastrophe: \(\langle E\rangle \rightarrow \infty\)
    GLE (general) Finite \(\operatorname{Im}[\chi(\omega)]\) Exact power balance: \(\langle P\rangle=0\)
    GLE + FDT \(S_{F}(\omega)=2 k_{B} T \tilde{\gamma}^{\prime}(\omega)\) Equipartition: \(\langle E\rangle=k_{B} T\)

    6.2 Fourier and Spectral Conventions

    We adopt the following Fourier transform pair for a time-dependent function \(F(t)\) :

    \[ \begin{align*} & F(\omega)=\int_{-\infty}^{\infty} d t F(t) e^{i \omega t} \tag{6.1} \label{6.1}\\[4pt] & F(t)=\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} F(\omega) e^{-i \omega t} \tag{6.2} \end{align*} \]

    This convention is symmetric and facilitates the treatment of both positive and negative frequencies on equal footing.

    For a wide-sense stationary stochastic process, the autocorrelation function depends only on the time difference:

    \[ C_{F}(\tau)=\langle F(t+\tau) F(t)\rangle, \tag{6.3} \label{6.3} \]

    where \(\langle\cdot\rangle\) denotes the ensemble average. According to the Wiener-Khinchin theorem [6], the two-sided power spectral density (PSD) \(S_{F}(\omega)\) is defined as:

    \[ S_{F}(\omega)=\int_{-\infty}^{\infty} d \tau C_{F}(\tau) e^{i \omega \tau} \tag{6.4} \label{6.4} \]

    Note the sign convention: with our forward Fourier transform using \(e^{i \omega t}\), consistency requires the PSD to also use \(e^{i \omega \tau}\). This ensures proper matching in the spectral correlation identity derived below.

    An important consequence of stationarity is the spectral correlation identity. For a stationary process, taking the Fourier transforms of \(F(t)\) at two different frequencies \(\omega\) and \(\omega^{\prime}\) and using Equation \ref{6.3}, one obtains:

    \[ \begin{align*} \left\langle F(\omega) F\left(\omega^{\prime}\right)\right\rangle & =\iint d t d t^{\prime} e^{i \omega t} e^{i \omega^{\prime} t^{\prime}}\left\langle F(t) F\left(t^{\prime}\right)\right\rangle \\[4pt] & =\int d \tau e^{i \omega \tau} C_{F}(\tau) \int d t^{\prime} e^{i\left(\omega+\omega^{\prime}\right) t^{\prime}} \\[4pt] & =2 \pi \delta\left(\omega+\omega^{\prime}\right) S_{F}(\omega) \tag{6.5} \label{6.5} \end{align*} \]

    This result, which follows directly from the stationarity condition (which implies \(C_{F}\left(t, t^{\prime}\right)=\) \(\left.C_{F}\left(t-t^{\prime}\right)\right)\) and the definitions above (see, e.g., Ref. [6]), will be used in Sec. 6.4 to evaluate the average injected power. Note that this identity is distinct from the Wiener-Khinchin theorem itself; it is a consequence of stationarity in the two-frequency domain.

    6.3 Linear Response and Susceptibility

    Consider a causal linear system where the displacement \(x(t)\) responds to an external force \(F(t)\) via a retarded Green's function \(\chi(t)\) :

    \[ x(t)=\int_{-\infty}^{t} d t^{\prime} \chi\left(t-t^{\prime}\right) F\left(t^{\prime}\right) . \tag{6.6} \label{6.6} \]

    The causality requirement states that \(\chi(t)=0\) for \(t<0\).

    In frequency space:

    \[ x(\omega)=\chi(\omega) F(\omega), \tag{6.7} \label{6.7} \]

    where \(\chi(\omega)\) is the susceptibility. The velocity is:

    \[ \dot{x}(\omega)=-i \omega \chi(\omega) F(\omega) . \tag{6.8} \label{6.8} \]

    For a physical system with real-valued \(\chi(t)\) :

    \[ \chi(-\omega)=\chi^{*}(\omega), \tag{6.9} \label{6.9} \]

    implying \(\operatorname{Re}[\chi(\omega)]\) is even and \(\operatorname{Im}[\chi(\omega)]\) is odd.

    6.4 General Derivation of Injected Power

    This formula is presented in most standard texts on statistical physics; see for example Landau and Lifshitz [7] (section 123 on generalized susceptibility), Martin [8], Cao [7], and Dattagupta [10]. An alternate derivation using linear response theory directly is given in Appendix.

    The instantaneous power injected by the stochastic force is \(P_{\text {in }}(t)=\dot{x}(t) F(t)\). For a stationary process:

    \[ \left\langle P_{\text {in }}\right\rangle=\langle\dot{x}(t) F(t)\rangle . \tag{6.10} \label{6.10} \]

    Writing \(\dot{x}(t)\) and \(F(t)\) as inverse Fourier transforms and using the spectral correlation identity Equation \ref{6.5}:

    \[ \left\langle P_{\text {in }}\right\rangle=\int \frac{d \omega}{2 \pi} \int \frac{d \omega^{\prime}}{2 \pi} e^{-i\left(\omega+\omega^{\prime}\right) t}\left\langle\dot{x}(\omega) F\left(\omega^{\prime}\right)\right\rangle . \tag{6.11} \label{6.11} \]

    Substituting \(\dot{x}(\omega)=-i \omega \chi(\omega) F(\omega)\) and applying Equation \ref{6.5}:

    \[ \left\langle\dot{x}(\omega) F\left(\omega^{\prime}\right)\right\rangle=-i \omega \chi(\omega)\left\langle F(\omega) F\left(\omega^{\prime}\right)\right\rangle=-i \omega \chi(\omega) \cdot 2 \pi \delta\left(\omega+\omega^{\prime}\right) S_{F}(\omega) . \tag{6.12} \label{6.12} \]

    Integrating over \(\omega^{\prime}\) using the delta function:

    \[ \left\langle P_{\mathrm{in}}\right\rangle=\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi}(-i \omega) \chi(\omega) S_{F}(\omega) . \tag{6.13} \label{6.13} \]

    6.4.1 Parity analysis

    Writing \(\chi(\omega)=\chi^{\prime}(\omega)+i \chi^{\prime \prime}(\omega)\), where \(\chi^{\prime}\) is even and \(\chi^{\prime \prime}\) is odd:

    \[ -i \omega\left[\chi^{\prime}+i \chi^{\prime \prime}\right] S_{F}=\left[-i \omega \chi^{\prime}+\omega \chi^{\prime \prime}\right] S_{F} . \tag{6.14} \label{6.14} \]

    Since \(S_{F}(\omega)\) is real and even (because \(C_{F}(\tau)\) is real and even), the term \(-i \omega \chi^{\prime}(\omega) S_{F}(\omega)\) is purely imaginary and odd (product of odd \(\omega\) and even functions), and integrates to zero. The term \(\omega \chi^{\prime \prime}(\omega) S_{F}(\omega)\) is real and even (product of two odd functions and an even function) and gives the net contribution.

    This can be seen from the parity table:

    Term Components Parity product Net parity
    \(-i \omega \chi^{\prime} S_{F}\) \(\omega\) (odd) × \(\chi^{\prime}\) (even) × \(S_{F}\) (even) odd × even × even odd (vanishes)
    \(\omega \chi^{\prime \prime} S_{F}\) \(\omega(\) odd \() \times \chi^{\prime \prime}(\) odd \() \times S_{F}(\) even \()\) odd × odd × even even (survives)

    Therefore:

    \[ \left\langle P_{\mathrm{in}}\right\rangle=\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega \operatorname{Im}[\chi(\omega)] S_{F}(\omega) . \tag{6.15} \label{6.15} \]

    This fundamental result shows that energy absorption is controlled exclusively by the imaginary (dissipative) part of the susceptibility.

    6.5 The Undamped Energy Catastrophe

    For an undamped harmonic oscillator:

    \[ m \ddot{x}(t)+m \omega_{0}^{2} x(t)=F(t), \tag{6.16} \label{6.16} \]

    the susceptibility is formally:

    \[ \chi_{0}(\omega)=\frac{1}{m\left(\omega_{0}^{2}-\omega^{2}\right)} . \tag{6.17} \label{6.17} \]

    6.5.1 Causal prescription

    To satisfy causality, we introduce infinitesimal damping:

    \[ \chi_{0}(\omega)=\lim _{\epsilon \rightarrow 0^{+}} \frac{1}{m\left(\omega_{0}^{2}-\omega^{2}-i \epsilon \omega\right)} . \tag{6.18} \label{6.18} \]

    Applying the Sokhotski-Plemelj theorem [11, 12]:

    \[ \lim _{\epsilon \rightarrow 0^{+}} \frac{1}{x-i \epsilon}=\mathcal{P}\left(\frac{1}{x}\right)+i \pi \delta(x), \tag{6.19} \label{6.19} \]

    we first perform a partial fraction decomposition of the susceptibility. Writing \(\omega_{0}^{2}-\omega^{2}=\) \(\left(\omega_{0}-\omega\right)\left(\omega_{0}+\omega\right)\) and treating the \(i \epsilon \omega\) term to leading order in \(\epsilon\) :

    \[ \chi_{0}(\omega)=\lim _{\epsilon \rightarrow 0^{+}} \frac{1}{m\left(\omega_{0}-\omega\right)\left(\omega_{0}+\omega\right)-i \epsilon \omega} . \tag{6.20} \label{6.20} \]

    Near \(\omega \approx+\omega_{0}\) the factor \(\left(\omega_{0}+\omega\right) \approx 2 \omega_{0}\), and near \(\omega \approx-\omega_{0}\) the factor \(\left(\omega_{0}-\omega\right) \approx 2 \omega_{0}\). We decompose into simple poles by writing:

    \[ \frac{1}{\left(\omega_{0}-\omega\right)\left(\omega_{0}+\omega\right)}=\frac{1}{2 \omega_{0}}\left(\frac{1}{\omega_{0}-\omega}+\frac{1}{\omega_{0}+\omega}\right), \tag{6.21} \label{6.21} \]

    which can be verified immediately by combining the right-hand side over the common denominator \(\left(\omega_{0}-\omega\right)\left(\omega_{0}+\omega\right)\). Incorporating the \(i \epsilon\) prescription consistently at each pole (the sign of i \(\epsilon \omega\) at \(\omega \approx+\omega_{0}\) is positive, and at \(\omega \approx-\omega_{0}\) is negative), the full causal expression becomes:

    \[ \chi_{0}(\omega)=\lim _{\epsilon \rightarrow 0^{+}} \frac{1}{2 m \omega_{0}}\left[\frac{1}{\omega_{0}-\omega-i \epsilon}+\frac{1}{\omega_{0}+\omega+i \epsilon}\right] . \tag{6.22} \label{6.22} \]

    Applying Equation \ref{6.19} to each term:

    \[ \begin{align*} \frac{1}{\omega_{0}-\omega-i \epsilon} & \rightarrow \mathcal{P} \frac{1}{\omega_{0}-\omega}+i \pi \delta\left(\omega_{0}-\omega\right), \tag{6.23} \label{6.23}\\[4pt] \frac{1}{\omega_{0}+\omega+i \epsilon} & \rightarrow \mathcal{P} \frac{1}{\omega_{0}+\omega}-i \pi \delta\left(\omega_{0}+\omega\right), \tag{6.24} \end{align*} \]

    where the sign difference in the delta term in Equation \ref{6.24} arises because the \(i \epsilon\) appears with the opposite sign, so the Plemelj formula gives \(-i \pi \delta\) rather than \(+i \pi \delta\).

    Taking the imaginary part of \(\chi_{0}\) and using \(\delta\left(\omega_{0}-\omega\right)=\delta\left(\omega-\omega_{0}\right)\) :

    \[ \operatorname{Im}\left[\chi_{0}(\omega)\right]=\frac{\pi}{2 m \omega_{0}}\left[\delta\left(\omega-\omega_{0}\right)-\delta\left(\omega+\omega_{0}\right)\right] . \tag{6.25} \label{6.25} \]

    This sign structure is essential: it ensures \(\omega \operatorname{Im}\left[\chi_{0}(\omega)\right] \geq 0\) everywhere, which is required for the power absorbed to be non-negative.

    6.5.2 Power and energy growth

    Substituting Equation \ref{6.25} into Equation \ref{6.15} and evaluating each delta function contribution separately:

    \[ \begin{align*} \left\langle P_{\text {in }}\right\rangle & =\int \frac{d \omega}{2 \pi} \omega \cdot \frac{\pi}{2 m \omega_{0}}\left[\delta\left(\omega-\omega_{0}\right)-\delta\left(\omega+\omega_{0}\right)\right] S_{F}(\omega) \\[4pt] & =\frac{1}{4 m \omega_{0}}\left[\omega_{0} S_{F}\left(\omega_{0}\right)-\left(-\omega_{0}\right) S_{F}\left(-\omega_{0}\right)\right] \\[4pt] & =\frac{1}{4 m \omega_{0}}\left[\omega_{0} S_{F}\left(\omega_{0}\right)+\omega_{0} S_{F}\left(-\omega_{0}\right)\right] \\[4pt] & =\frac{S_{F}\left(\omega_{0}\right)}{2 m}, \tag{6.26} \label{6.26} \end{align*} \]

    where in the last step we used \(S_{F}\left(-\omega_{0}\right)=S_{F}\left(\omega_{0}\right)\) (since \(S_{F}\) is even).

    Since \(S_{F}\left(\omega_{0}\right)>0\), the energy grows linearly:

    \[ \langle E(t)\rangle=\langle E(0)\rangle+\left\langle P_{\text {in }}\right\rangle \cdot t \rightarrow \infty . \tag{6.27} \label{6.27} \]

    This energy catastrophe violates equilibrium statistical mechanics, demonstrating that dissipation is thermodynamically necessary.

    6.6 The Generalized Langevin Equation

    The GLE incorporates memory-dependent damping:

    \[ m \ddot{x}(t)+m \omega_{0}^{2} x(t)+\int_{-\infty}^{t} \gamma\left(t-t^{\prime}\right) \dot{x}\left(t^{\prime}\right) d t^{\prime}=F(t) . \tag{6.28} \label{6.28} \]

    6.6.1 Susceptibility

    In frequency space, the susceptibility takes the form:

    \[ \chi(\omega)=\frac{1}{m\left(\omega_{0}^{2}-\omega^{2}\right)-i \omega \tilde{\gamma}(\omega)}, \tag{6.29} \label{6.29} \]

    where we define the one-sided Fourier transform of the memory kernel:

    \[ \tilde{\gamma}(\omega)=\int_{0}^{\infty} \gamma(t) e^{i \omega t} d t \tag{6.30} \label{6.30} \]

    Because \(\gamma(t)\) is real, we have the symmetry \(\tilde{\gamma}(-\omega)=\tilde{\gamma}^{*}(\omega)\).

    Writing \(\tilde{\gamma}(\omega)=\tilde{\gamma}^{\prime}(\omega)+i \tilde{\gamma}^{\prime \prime}(\omega)\) with \(\tilde{\gamma}^{\prime}=\operatorname{Re}[\tilde{\gamma}]\) and \(\tilde{\gamma}^{\prime \prime}=\operatorname{Im}[\tilde{\gamma}]\), the denominator of Equation \ref{7.17} becomes:

    \[ D(\omega)=\left[m\left(\omega_{0}^{2}-\omega^{2}\right)+\omega \tilde{\gamma}^{\prime \prime}(\omega)\right]-i \omega \tilde{\gamma}^{\prime}(\omega) . \tag{6.31} \label{6.31} \]

    Taking the imaginary part of \(\chi=1 / D\) :

    \[ \operatorname{Im}[\chi(\omega)]=\frac{\omega \tilde{\gamma}^{\prime}(\omega)}{|D(\omega)|^{2}}=\omega \tilde{\gamma}^{\prime}(\omega)|\chi(\omega)|^{2} . \tag{6.32} \label{6.32} \]

    Note that \(|D(\omega)|^{2}=\left[m\left(\omega_{0}^{2}-\omega^{2}\right)+\omega \tilde{\gamma}^{\prime \prime}(\omega)\right]^{2}+\omega^{2}\left[\tilde{\gamma}^{\prime}(\omega)\right]^{2}\), which in general includes a contribution from the imaginary part \(\tilde{\gamma}^{\prime \prime}(\omega)\) of the memory kernel. Only in the Markovian limit \(\gamma(t)=\gamma \delta(t)\) does \(\tilde{\gamma}^{\prime \prime}=0\) exactly.

    6.6.2 Dissipated power

    We now derive the dissipated power explicitly. The friction force in the GLE is \(F_{\gamma}(t)=\) \(-\int_{-\infty}^{t} \gamma\left(t-t^{\prime}\right) \dot{x}\left(t^{\prime}\right) d t^{\prime}\), so the instantaneous power dissipated is:

    \[ P_{\mathrm{diss}}(t)=\dot{x}(t) F_{\gamma}(t)=-\dot{x}(t) \int_{-\infty}^{t} \gamma\left(t-t^{\prime}\right) \dot{x}\left(t^{\prime}\right) d t^{\prime} \tag{6.33} \label{6.33} \]

    Taking the ensemble average and using stationarity with \(\tau=t-t^{\prime} \geq 0\) :

    \[ \left\langle P_{\mathrm{diss}}\right\rangle=-\int_{0}^{\infty} \gamma(\tau)\langle\dot{x}(t) \dot{x}(t-\tau)\rangle d \tau \tag{6.34} \label{6.34} \]

    Expressing the velocity correlator via its spectral density \(S_{\dot{x}}(\omega)=\omega^{2}|\chi(\omega)|^{2} S_{F}(\omega)\) and applying the same Parseval/Fourier analysis as above, the friction force in frequency space is:

    \[ F_{\gamma}(\omega)=-i \omega \tilde{\gamma}(\omega) x(\omega)=-i \omega \tilde{\gamma}(\omega) \chi(\omega) F(\omega) . \tag{6.35} \label{6.35} \]

    Then:

    \[ \begin{align*} \left\langle P_{\mathrm{diss}}\right\rangle & =-\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \tilde{\gamma}^{*}(\omega) \omega^{2}|\chi(\omega)|^{2} S_{F}(\omega) \\[4pt] & =-\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2} \operatorname{Re}[\tilde{\gamma}(\omega)]|\chi(\omega)|^{2} S_{F}(\omega) \\[4pt] & =-\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2} \tilde{\gamma}^{\prime}(\omega)|\chi(\omega)|^{2} S_{F}(\omega) \tag{6.36} \label{6.36} \end{align*} \]

    where the imaginary part of \(\tilde{\gamma}(\omega)\) drops out upon integration because the resulting integrand is odd. Comparing with Eqs. Equation \ref{6.15} and Equation \ref{6.32}:

    \[ \left\langle P_{\text {in }}\right\rangle=\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2} \tilde{\gamma}^{\prime}(\omega)|\chi(\omega)|^{2} S_{F}(\omega) . \tag{6.37} \label{6.37} \]

    6.6.3 Power balance

    The total average power is the sum of injected and dissipated contributions:

    \[ \begin{align*} \langle P\rangle & =\left\langle P_{\text {in }}\right\rangle+\left\langle P_{\text {diss }}\right\rangle \\[4pt] & =\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2} \tilde{\gamma}^{\prime}(\omega)|\chi(\omega)|^{2} S_{F}(\omega)-\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2} \tilde{\gamma}^{\prime}(\omega)|\chi(\omega)|^{2} S_{F}(\omega)=0 \tag{6.38} \label{6.38} \end{align*} \]

    Here \(\left\langle P_{\text {diss }}\right\rangle\) is defined as negative (energy leaving the system via friction), so the balance \(\langle P\rangle=0\) means power injected by the noise is exactly removed by the friction force. Therefore:

    \[ \left\langle P_{\text {in }}\right\rangle+\left\langle P_{\text {diss }}\right\rangle=0 . \tag{6.39} \label{6.39} \]

    This balance holds frequency by frequency, not merely in total, preventing the energy catastrophe.

    6.6.4 The general principle

    This confirms that the form of the GLE with a memory kernel is consistent with energy balance for any noise spectrum \(S_{F}(\omega)\). However, the specific forms of \(\tilde{\gamma}(\omega)\) and \(S_{F}(\omega)\) are fixed by the fluctuation-dissipation theorem, as we show in the next section.

    6.7 Equipartition from the FDT

    The fluctuation-dissipation theorem (FDT) [4] relates the noise power spectrum to the dissipative part of the memory kernel:

    \[ S_{F}(\omega)=2 k_{B} T \tilde{\gamma}^{\prime}(\omega) . \tag{6.40} \label{6.40} \]

    This is sometimes called the "second fluctuation-dissipation theorem" to distinguish it from the Kubo relation (the "first" FDT) connecting the equilibrium correlation function of a dynamical variable to the corresponding linear response function [4]. Both forms express the same underlying principle: the bath fluctuations that drive the system are precisely matched to the dissipation they cause.

    6.7.1 Markovian limit

    For \(\gamma(t)=\gamma \delta(t)\), the memory kernel is instantaneous and \(\tilde{\gamma}(\omega)=\gamma\) (real, frequency-independent). Then \(\tilde{\gamma}^{\prime \prime}=0\) and:

    \[ \chi(\omega)=\frac{1}{m\left(\omega_{0}^{2}-\omega^{2}\right)-i \omega \gamma} . \tag{6.41} \label{6.41} \]

    The mean kinetic energy is:

    \[ \langle T\rangle=\frac{m}{2} \int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2}|\chi(\omega)|^{2} S_{F}(\omega) . \tag{6.42} \label{6.42} \]

    Substituting Equation \ref{9.4} with \(\tilde{\gamma}^{\prime}(\omega)=\gamma\) :

    \[ \langle T\rangle=\frac{m}{2} \int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2} \frac{2 k_{B} T \gamma}{\left[m\left(\omega_{0}^{2}-\omega^{2}\right)\right]^{2}+\omega^{2} \gamma^{2}} \tag{6.43} \label{6.43} \]

    Simplifying:

    \[ \langle T\rangle=\frac{k_{B} T \gamma}{2 \pi} \int_{-\infty}^{\infty} \frac{\omega^{2} d \omega}{\left(\omega_{0}^{2}-\omega^{2}\right)^{2}+\left(\gamma^{2} / m^{2}\right) \omega^{2}} . \tag{6.44} \label{6.44} \]

    Note that the damping coefficient in the denominator appears as \(\gamma / m\); for notational simplicity, we absorb \(m\) into \(\gamma\) in what follows (or equivalently, set \(m=1\) temporarily and restore dimensions at the end). With this understanding, we evaluate:

    \[ I=\int_{-\infty}^{\infty} \frac{\omega^{2} d \omega}{\left(\omega_{0}^{2}-\omega^{2}\right)^{2}+\gamma^{2} \omega^{2}} \tag{6.45} \label{6.45} \]

    The integral is evaluated by contour methods. The integrand \(f(z)=z^{2} /\left[\left(z^{2}-\omega_{0}^{2}\right)^{2}+\gamma^{2} z^{2}\right]\) has poles where \(m\left(\omega_{0}^{2}-\omega^{2}\right)-i \omega \gamma=0\), i.e., at:

    \[ z= \pm \sqrt{\omega_{0}^{2}-\gamma^{2} / 4}-i \gamma / 2 \quad \text { (underdamped case). } \tag{6.46} \label{6.46} \]

    Both poles lie in the lower half of the complex \(\omega\)-plane (negative imaginary parts), consistent with causality. Closing the contour in the upper half-plane (where the integrand vanishes) encloses no poles, so we instead close in the lower half-plane and pick up residues at the two poles. A standard residue calculation [11, 12] gives:

    \[ I=\frac{\pi}{\gamma} . \tag{6.47} \label{6.47} \]

    For completeness, this result can also be verified by partial fractions, writing the denominator as \(\left(\omega^{2}-\omega_{+}^{2}\right)\left(\omega^{2}-\omega_{-}^{2}\right)\) in the underdamped case \(\omega_{0}>\gamma / 2\); the residues at both lower-half-plane poles sum to \(\pi / \gamma\), independently of \(\omega_{0}\).

    Therefore:

    \[ \langle T\rangle=\frac{k_{B} T \gamma}{2 \pi} \cdot \frac{\pi}{\gamma}=\frac{k_{B} T}{2} . \tag{6.48} \label{6.48} \]

    Similarly, by the same contour integral (or by symmetry of the Lorentzian), \(\langle V\rangle=\frac{1}{2} m \omega_{0}^{2}\left\langle x^{2}\right\rangle=\) \(k_{B} T / 2\), giving:

    \[ \langle E\rangle=\langle T\rangle+\langle V\rangle=k_{B} T . \tag{6.49} \label{6.49} \]

    6.8 Discussion and Conclusions

    The key results, summarized in Table 6.1, are:

    • 1.Energy absorption is governed exclusively by \(\operatorname{Im}[\chi(\omega)]\) [Equation \ref{6.15}], a consequence of parity alone.
    • 2.Undamped oscillators exhibit divergent energy growth [Equation \ref{6.27}], violating equilibrium - the "energy catastrophe."
    • 3.The GLE provides exact, frequency-by-frequency power balance [Equation \ref{6.39}], derived explicitly from both the injected and dissipated power contributions.
    • 4.Under the FDT, equipartition emerges naturally [Equation \ref{6.49}], with the result \(\langle E\rangle=k_{B} T\) confirmed via contour integration.

    Our analysis demonstrates that, although dissipation is generally introduced as a phenomenological term, it turns out to be a thermodynamic necessity: without it, any oscillator coupled to a thermal reservoir would accumulate energy without bound. The GLE with a memory kernel provides the unique consistent framework that avoids this catastrophe and yields thermodynamically correct steady states.

    A standard undamped oscillator driven by stochastic forcing undergoes an energy catastrophe. Because it lacks a mechanism to shed energy, the resonant singularities in the susceptibility cause the average energy to grow linearly in time:

    \[ \langle E(t)\rangle \propto t \rightarrow \infty . \tag{6.50} \label{6.50} \]

    This naturally violates the Equipartition Theorem, which requires \(\langle E\rangle=k_{B} T\). This chapter demonstrates that the GLE introduces a memory-dependent damping kernel \(\gamma(t)\) that regulates energy growth through exact frequency-by-frequency power balance.

    The most significant intuitive takeaway is that absorption (energy in) and dissipation (energy out) are two sides of the same coin: both are governed by the kernel \(\tilde{\gamma}(\omega)\).

    The heat bath acts as both a source of fluctuations and a sink for energy. The coupling strength \(\tilde{\gamma}(\omega)\) between the system and the bath determines both roles:

    • Dissipation: As the particle moves, it collides with bath molecules, transferring kinetic energy back to the bath (drag).
    • Absorption: Even at rest, bath molecules "kick" the particle due to their thermal energy (noise).

    The Generalized Langevin Equation ensures that the energy leaving the system is exactly balanced by the gateway for energy entering it, so that the system reaches a stable thermal state at temperature \(T\).

    Appendix I: Colored Noise and the Drude Memory Kernel

    In the main text, we focused primarily on the Markovian limit where the friction is local in time. Here, we extend the analysis to a frequency-dependent (Drude) memory kernel and its corresponding "colored" noise spectrum.

    6.8.1 The Drude Model for Memory

    A common physical model for memory is the Drude-like exponential decay of the friction kernel:

    \[ \gamma(t)=\frac{\gamma_{0}}{\tau_{c}} e^{-t / \tau_{c}} \Theta(t), \tag{6.51} \label{6.51} \]

    where \(\gamma_{0}\) is the static friction coefficient, \(\tau_{c}\) is the correlation time of the bath, and \(\Theta(t)\) is the Heaviside step function. In the limit \(\tau_{c} \rightarrow 0\), we use the fact that \(\frac{1}{\tau_{c}} e^{-t / \tau_{c}}\) acts as a delta sequence for \(t \geq 0\), giving \(\int_{0}^{\infty} \frac{1}{\tau_{c}} e^{-t / \tau_{c}} \phi(t) d t \rightarrow \phi(0)\) for any test function \(\phi(t)\). Thus:

    \[ \gamma(t) \rightarrow \gamma_{0} \delta(t) \quad\left(\tau_{c} \rightarrow 0\right), \tag{6.52} \label{6.52} \]

    recovering the Markovian kernel with friction coefficient \(\gamma_{0}\) (no additional factor of 2).

    Taking the Fourier transform of Equation \ref{6.51}:

    \[ \begin{align*} \tilde{\gamma}(\omega) & =\int_{0}^{\infty} \frac{\gamma_{0}}{\tau_{c}} e^{-t / \tau_{c}} e^{i \omega t} d t \\[4pt] & =\frac{\gamma_{0}}{\tau_{c}} \int_{0}^{\infty} e^{-\left(1 / \tau_{c}-i \omega\right) t} d t \\[4pt] & =\frac{\gamma_{0}}{\tau_{c}} \cdot \frac{1}{1 / \tau_{c}-i \omega}=\frac{\gamma_{0}}{1-i \omega \tau_{c}} . \tag{6.53} \label{6.53} \end{align*} \]

    The real (dissipative) part of the kernel is:

    \[ \tilde{\gamma}^{\prime}(\omega)=\operatorname{Re}[\tilde{\gamma}(\omega)]=\frac{\gamma_{0}}{1+\omega^{2} \tau_{c}^{2}} . \tag{6.54} \label{6.54} \]

    6.8.2 Colored Noise (Ornstein-Uhlenbeck)

    According to the FDT, Equation \ref{9.4}, the noise \(F(t)\) must have a PSD proportional to \(\tilde{\gamma}^{\prime}(\omega)\) :

    \[ S_{F}(\omega)=2 k_{B} T \tilde{\gamma}^{\prime}(\omega)=\frac{2 k_{B} T \gamma_{0}}{1+\omega^{2} \tau_{c}^{2}} . \tag{6.55} \label{6.55} \]

    This is the characteristic spectrum of the Ornstein-Uhlenbeck process. Unlike white noise, the power in colored noise rolls off at high frequencies \(\left(\omega \gg 1 / \tau_{c}\right)\), reflecting the finite response time of the environment.

    6.8.3 Verification of Power Balance

    The susceptibility \(\chi(\omega)\) for the Drude GLE is:

    \[ \chi(\omega)=\left[m\left(\omega_{0}^{2}-\omega^{2}\right)-i \omega\left(\frac{\gamma_{0}}{1-i \omega \tau_{c}}\right)\right]^{-1} . \tag{6.56} \label{6.56} \]

    The injected power, using Equation \ref{6.37}, is:

    \[ \left\langle P_{\text {in }}\right\rangle=\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2} \tilde{\gamma}^{\prime}(\omega)|\chi(\omega)|^{2} S_{F}(\omega) . \tag{6.57} \label{6.57} \]

    Substituting \(S_{F}(\omega)=2 k_{B} T \tilde{\gamma}^{\prime}(\omega)\) :

    \[ \left\langle P_{\text {in }}\right\rangle=2 k_{B} T \int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2}\left(\frac{\gamma_{0}}{1+\omega^{2} \tau_{c}^{2}}\right)^{2}|\chi(\omega)|^{2} . \tag{6.58} \label{6.58} \]

    By the same argument as in Sec. 6.6, the dissipated power \(\left\langle P_{\text {diss }}\right\rangle\) involves the identical integral with a negative sign. The balance \(\left\langle P_{\text {in }}\right\rangle+\left\langle P_{\text {diss }}\right\rangle=0\) is thus satisfied frequency by frequency, even with the frequency-dependent damping \(\tilde{\gamma}^{\prime}(\omega)\).

    6.8.4 Equipartition in the Non-Markovian Case

    While the integrals are more complex, the equipartition theorem still holds. The mean kinetic energy is:

    \[ \langle T\rangle=\frac{m}{2} \int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega^{2}|\chi(\omega)|^{2} S_{F}(\omega) . \tag{6.59} \label{6.59} \]

    By substituting the Drude forms and using the method of residues, it can be shown that the contribution of the memory pole at \(\omega=i / \tau_{c}\) exactly compensates for the modified response of the oscillator, yielding \(\langle T\rangle=\frac{1}{2} k_{B} T\). This confirms that the spectral shape of the colored noise (the "color") is perfectly matched to the memory of the dissipation, preserving the thermodynamic state.

    Appendix II: Alternate Derivation of Power Absorption

    To provide an independent derivation of Equation \ref{6.15}, we use the linear response relation directly in the time domain. In a stationary process, the velocity can be written as:

    \[ \dot{x}(t)=\int_{-\infty}^{\infty} \dot{\chi}\left(t-t^{\prime}\right) F\left(t^{\prime}\right) d t^{\prime} \tag{6.60} \label{6.60} \]

    where \(\chi(t)\) is the retarded response function (satisfying \(\chi(t)=0\) for \(t<0\) ) and \(\dot{\chi}\) denotes its time derivative. The upper limit is written as \(\infty\) rather than \(t\) because causality \((\chi(\tau)=0\) for \(\tau<0\) ) already enforces that only \(t^{\prime} \leq t\) contributes.

    Therefore:

    \[ \langle F(t) \dot{x}(t)\rangle=\left\langle F(t) \int_{-\infty}^{\infty} \dot{\chi}\left(t-t^{\prime}\right) F\left(t^{\prime}\right) d t^{\prime}\right\rangle . \tag{6.61} \label{6.61} \]

    By stationarity, \(\left\langle F(t) F\left(t^{\prime}\right)\right\rangle=C_{F}\left(t-t^{\prime}\right)\), so:

    \[ \langle F(t) \dot{x}(t)\rangle=\int_{-\infty}^{\infty} \dot{\chi}\left(t-t^{\prime}\right) C_{F}\left(t-t^{\prime}\right) d t^{\prime} \tag{6.62} \label{6.62} \]

    Changing variables to \(\tau=t-t^{\prime}\), and using the fact that \(\dot{\chi}(\tau)=0\) for \(\tau<0\), the lower limit is effectively 0:

    \[ \langle F(t) \dot{x}(t)\rangle=\int_{0}^{\infty} \dot{\chi}(\tau) C_{F}(\tau) d \tau \tag{6.63} \label{6.63} \]

    The right-hand side is a standard half-line integral of the product \(\dot{\chi}(\tau) C_{F}(\tau)\). To pass to frequency space we note that, since \(\dot{\chi}(\tau)=0\) for \(\tau<0\), extending the lower limit to \(-\infty\) introduces no new contribution:

    \[ \int_{0}^{\infty} \dot{\chi}(\tau) C_{F}(\tau) d \tau=\int_{-\infty}^{\infty} \dot{\chi}(\tau) C_{F}(\tau) d \tau . \tag{6.64} \label{6.64} \]

    This is simply a convolution evaluated at \(\tau=0\). By the convolution theorem and Parseval's relation:

    \[ \int_{-\infty}^{\infty} \dot{\chi}(\tau) C_{F}(\tau) d \tau=\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi}(-i \omega) \chi(\omega) S_{F}(\omega) . \tag{6.65} \label{6.65} \]

    Separating real and imaginary parts of \(\chi(\omega)=\chi^{\prime}(\omega)+i \chi^{\prime \prime}(\omega)\), and using the same parity argument as in Sec. 6.4, the \(\chi^{\prime}\) term is odd and vanishes, yielding:

    \[ \left\langle P_{\text {in }}\right\rangle=\int_{-\infty}^{\infty} \frac{d \omega}{2 \pi} \omega \operatorname{Im}[\chi(\omega)] S_{F}(\omega), \tag{6.66} \label{6.66} \]

    in agreement with Equation \ref{6.15}.

    Bibliography

    [1] R. Kubo, Rep. Prog. Phys. 29, 255 (1966).

    [2] R. Kubo, M. Toda, and N. Hashitsume, Statistical Physics II: Nonequilibrium Statistical Mechanics (Springer, Berlin, 2012).

    [3] V. B. Magalinski, "Dynamical Model in the Theory of Brownian Motion," Zh. Eksp. Teor. Fiz. 36, 1942 (1959) [Sov. Phys. JETP 9, 1381 (1959)].

    [4] R. Zwanzig, Nonequilibrium Statistical Mechanics (Oxford University Press, New York, 2001).

    [5] H. Mori, Prog. Theor. Phys. 33, 423 (1965).

    [6] C. W. Gardiner, Handbook of Stochastic Methods for Physics, Chemistry and the Natural Sciences, 4th ed. (Springer, Berlin, 2009).

    [7] L.D. Landau and E.M. Lifshitz, Statistical Physics, Vol. 5 of Course of Theoretical Physics, 3rd ed. (Pergamon Press, Oxford, 1980).

    [8] P. C. Martin, Measurements and Correlation Functions (Gordon and Breach, New York, 1968).

    [9] J. Cao, Non-Equilibrium Statistical Mechanics, Open Education Resource (OER) Libre-Texts Project (https://LibreTexts.org).

    [10] S. Dattagupta, Relaxation Phenomena in Condensed Matter Physics (Academic Press, Orlando, 1987).

    [11] M. J. Ablowitz and A. S. Fokas, Complex Variables: Introduction and Applications, 2nd ed. (Cambridge University Press, Cambridge, 2003).

    [12] G. B. Arfken and H. J. Weber, Mathematical Methods for Physicists, 6th ed. (Elsevier, Amsterdam, 2005).


    This page titled 6: Power Balance and the Necessity of Dissipation- Pedagogical Derivation of the Generalized Langevin Equation was last modified on Mon, 02 Nov 2026 06:12:19 GMT and is shared under a CC BY-NC-SA 4.0 license and was authored, remixed, and/or curated by Pabitra N. Sen.