Response and Relaxation Functions

Nudge a system in equilibrium with a weak external field and its observables shift, then evolve. Because the perturbation is small, only the term linear in the field survives, and the entire behaviour is carried by two intimately related causal Green functions. The response function $\Phi(t)$ is the system's reaction to an impulsive (delta-function) kick — its impulse response. The relaxation function $\Psi(t)$ is what happens the other way around: hold the system displaced with a static field, then suddenly release it, and $\Psi(t)$ is the decay back to equilibrium. Both are properties of the unperturbed ensemble; the field merely reads them off, which is why they can be written entirely in terms of equilibrium correlations. These two functions are not independent. Because a static field is a continuous stack of impulses switched on and never off, the free decay is the accumulated tail of the impulse response, and the two are tied by the single differential identity $\Phi(t)=-\dot\Psi(t)$ . In the frequency domain both feed the same complex susceptibility $\chi(\omega)$ ; the response function is its inverse Fourier transform on the half line, and the relaxation function is the after-effect (aftereffekt) function whose transform builds $\chi(\omega)$ through $\chi(\omega)=\chi(0)+i\omega\int_0^\infty\Psi(t)e^{i\omega t}\,\mathrm{d}t$ . The deep physical content is Onsager's regression hypothesis: a spontaneous thermal fluctuation of an observable decays, on average, by exactly the same law that governs the relaxation of a deliberately prepared, macroscopic displacement of that observable. You cannot tell from the decay whether the bump arose by chance or by your hand. This single statement, made quantitative through the classical Kubo relation $\Phi(t)=-\beta\,\dot C(t)\,\Theta(t)$ , is the seed of the entire fluctuation–dissipation framework and the reason spectroscopy, dielectric loss, viscosity, and spin relaxation all measure the same object.

Key takeaways

The response function $\Phi(t)$ is the impulse (delta-function) response of a system to a weak field; causality forces $\Phi(t)=0$ for $t<0$ .
The relaxation function $\Psi(t)$ is the free decay after a long-held static field is switched off; it is the integral of the response, $\Psi(t)=\int_t^\infty\Phi(\tau)\,\mathrm{d}\tau$ , so $\Phi(t)=-\dot\Psi(t)$ .
In equilibrium the classical Kubo relation ties the response to the correlation function, $\Phi(t)=-\beta\,\dot C(t)\,\Theta(t)$ , and the static limit recovers the isothermal susceptibility $\Psi(0)=\chi(0)=\beta\langle\delta A^2\rangle$ .
Onsager's regression hypothesis: spontaneous fluctuations decay by the same macroscopic law as driven displacements — the step-off relaxation exactly mirrors the step-on approach.
All three objects — $\Phi$ , $\Psi$ , and $\chi(\omega)$ — carry the same information: pulse, step, and sinusoidal responses are one function seen through different time boundary conditions.
Key idea: A weak field couples as $H'=-f(t)A$ , so $\delta\langle A(t)\rangle=\int_{-\infty}^t\Phi(t-t')f(t')\,\mathrm{d}t'$ with a causal kernel fixed by equilibrium dynamics; the relaxation function $\Psi(t)=\int_t^\infty\Phi\,\mathrm{d}\tau$ obeys $\Phi=-\dot\Psi$ , and $\chi(\omega)=\chi(0)+i\omega\int_0^\infty\Psi(t)e^{i\omega t}\mathrm{d}t$ — response, relaxation, and dissipation are one object.

The core insight

A weak field couples as $H'=-f(t)A$ , so $\delta\langle A(t)\rangle=\int_{-\infty}^t\Phi(t-t')f(t')\,\mathrm{d}t'$ with a causal kernel fixed by equilibrium dynamics; the relaxation function $\Psi(t)=\int_t^\infty\Phi(\tau)\,\mathrm{d}\tau$ obeys $\Phi=-\dot\Psi$ , and $\chi(\omega)=\chi(0)+i\omega\int_0^\infty\Psi(t)e^{i\omega t}\,\mathrm{d}t$ — response, relaxation, and dissipation are one object.

Where this lives today

Response and relaxation functions are the daily currency of experimental condensed-matter physics because they are literally what the instruments read. Broadband dielectric spectroscopy sweeps a field from millihertz to terahertz and records $\chi''(\omega)$ directly; the loss peak's position gives a relaxation time and its temperature dependence gives an activation energy — the primary probe of the glass transition, where $\tau$ swells by fifteen orders of magnitude as a liquid vitrifies. In magnetic resonance the spin-lattice and spin-spin rates $1/T_1$ and $1/T_2$ are values of $\chi''(\omega)$ at the Larmor frequency, so an MRI contrast image is a spatial map of relaxation functions. The same formalism now underwrites the search for departures from equilibrium. In aging glasses, sheared colloids, and active matter (bacterial baths, molecular motors) one measures the response $\Phi$ and the spontaneous correlation $C$ independently; when the classical Kubo relation $\Phi=-\beta\dot C$ fails, the mismatch defines a frequency-dependent effective temperature that quantifies how far the system sits from equilibrium. Onsager's regression hypothesis, a century old, is thus both the foundation of equilibrium transport theory and the yardstick against which non-equilibrium matter is measured.

The physical picture

Picture a Brownian particle held in a weak optical trap by a small steady force. Pull it aside and hold it, then release: the way its displacement drifts back to centre is the relaxation function $\Psi(t)$ . Instead give it a single sharp tap and watch the recoil ring down: that is the response function $\Phi(t)$ . Both describe the same microscopic dynamics — the same molecular collisions, the same trap stiffness — seen through two different boundary conditions in time. One holds the field for all past time and cuts it; the other applies it for an instant. Because the field is weak, the system never leaves the regime where thermal fluctuations dominate. So the relaxation of an externally imposed displacement and the spontaneous regression of a thermal fluctuation of the same observable follow identical decay laws. This is Onsager's regression hypothesis: nature does not distinguish a bump you made from one that happened by chance. The step-off decay curve is the mirror image of the step-on approach. The link between the two functions is pure bookkeeping. A static field switched on long ago and left on is the sum of infinitely many impulses, one at each past instant. So the post-switch-off decay is the integral over all those impulse responses that have not yet died — hence $\Psi(t)=\int_t^\infty\Phi(\tau)\,\mathrm{d}\tau$ , and differentiating gives $\Phi(t)=-\dot\Psi(t)$ . Sinusoidal drive is the third face: Fourier-transform the same $\Phi$ and you get the complex susceptibility $\chi(\omega)$ whose imaginary part is the absorption spectrum measured in the lab.

Figure

The response function Φ(t) turns on at the causal wall t = 0 and never rings before the tap

Figure

Onsager regression: the step-off relaxation Ψ(t) exactly mirrors the step-on approach — chance and design decay alike

The core equations

The whole framework starts from how a weak external field enters the Hamiltonian and how the induced change of any observable follows to linear order. A field $f(t')$ conjugate to an observable $A$ perturbs the system, and the average of a (possibly different) observable $B$ shifts by a convolution with a causal kernel.

\delta\langle B(t)\rangle=\int_{-\infty}^{\infty}\Phi_{BA}(t-t')\,f(t')\,\mathrm{d}t',\qquad \Phi_{BA}(t-t')=0\ \text{for}\ t<t' \qquad (1)

where: $\delta\langle B(t)\rangle$ — deviation of $B$ from its equilibrium value induced by the field; $\Phi_{BA}$ — causal (retarded) response function of $B$ to a field conjugate to $A$ ; $f(t')$ — weak external field entering as $H'=-f(t')A$ . The step condition $\Phi=0$ for $t<t'$ is causality: no response before its cause. Setting $B=A$ gives the autoresponse; this is the object measured whenever a single conjugate pair is driven and observed. The microscopic form of $\Phi$ is fixed by first-order perturbation of the equilibrium density matrix — the Kubo formula. Quantum-mechanically it is an equilibrium average of a commutator; classically it collapses to a time derivative of the correlation function.

\Phi_{BA}(t)=\frac{i}{\hbar}\,\Theta(t)\,\big\langle[B(t),A(0)]\big\rangle_{\rm eq}\;\xrightarrow{\hbar\to0}\;\Phi(t)=-\beta\,\Theta(t)\,\dot C(t) \qquad (2)

where: $\Theta(t)$ — Heaviside step enforcing causality; $[B(t),A(0)]$ — Heisenberg-picture commutator averaged over the unperturbed ensemble; $C(t)=\langle A(t)A(0)\rangle_{\rm eq}$ — equilibrium autocorrelation; $\beta=1/k_BT$ . The classical relation carries a minus sign, a time derivative, and a step — the three pieces students most often drop. It says the impulse response is (minus $\beta$ times) the slope of the correlation, so a slowly decaying fluctuation produces a slowly decaying response. The relaxation function is the integral of the response — the accumulated tail — and equals the correlation function measured from its long-time floor.

\Psi(t)=\int_t^{\infty}\Phi(\tau)\,\mathrm{d}\tau=\beta\,[\,C(t)-C(\infty)\,],\qquad \Phi(t)=-\dot\Psi(t) \qquad (3)

where: $\Psi(t)$ — relaxation (after-effect) function, the decay following removal of a static field; $C(\infty)=\langle A\rangle^2$ — the disconnected long-time value subtracted off. The differential identity $\Phi=-\dot\Psi$ is the exact statement that response and relaxation are two faces of one function: a pulse is the derivative of a step. Evaluating $\Psi$ at $t=0$ recovers a purely static, thermodynamic quantity — the isothermal susceptibility — from equal-time fluctuations. This is the static face of the fluctuation–dissipation link.

\Psi(0)=\chi(0)=\beta\,[\,C(0)-C(\infty)\,]=\beta\,\langle\delta A^2\rangle \qquad (4)

where: $\chi(0)$ — zero-frequency (DC, isothermal) susceptibility; $\langle\delta A^2\rangle$ — equilibrium variance of $A$ . This is the sum rule that ties a static thermodynamic response (magnetic susceptibility, compressibility) to the size of the fluctuations — no dynamics required, only the equal-time variance. The frequency-domain susceptibility is built from the relaxation function through the after-effect integral; its imaginary part is the dissipation spectrum, and the work a periodic drive feeds in is fixed by it.

\chi(\omega)=\chi(0)+i\omega\!\int_0^{\infty}\!\Psi(t)\,e^{i\omega t}\,\mathrm{d}t,\qquad P_{\rm abs}=\tfrac12\,\omega\,|f_0|^2\,\chi''(\omega)\ge0 \qquad (5)

where: $\chi(\omega)=\chi'+i\chi''$ — complex susceptibility; $\chi''$ — dissipative (absorptive) part; $P_{\rm abs}$ — time-averaged power absorbed from a drive $f(t)=f_0\cos\omega t$ . Passivity of an equilibrium system forces $\omega\chi''(\omega)\ge0$ : a system at equilibrium can only take up net energy from a weak drive, never supply it.

Derivation ARelaxation as the integrated response, and its static-limit sum rule

The goal is to prove that the free decay after switching off a static field is the running integral of the impulse response, then use the classical Kubo relation to express it through the equilibrium correlation function and land on the static-susceptibility sum rule. The strategy is to feed the general convolution $(1)$ a step field that is on for all negative time and off at $t=0$ , then convert the resulting time integral into the after-effect function $\Psi$ and evaluate its endpoints. Watch the causal limits of the integral carefully: the field is nonzero only for $t'<0$ , and causality restricts $\Phi$ to $t>t'$ , so the two constraints together set the integration range.

Step 1. Insert the step field $f(t')=f_0\Theta(-t')$ (on for $t'<0$ , off at $t'=0$ ) into the convolution: $\delta\langle A(t)\rangle=\int_{-\infty}^{0}\Phi(t-t')f_0\,\mathrm{d}t'$ . Why: the field is zero for $t'>0$ , so the upper limit drops to $0$ ; causality already guarantees $\Phi(t-t')=0$ when $t'>t$ , which is automatic here for $t>0$ .

Step 2. Change variable to $\tau=t-t'$ ; as $t'$ runs from $-\infty$ to $0$ , $\tau$ runs from $\infty$ down to $t$ , giving $\delta\langle A(t)\rangle=f_0\int_t^{\infty}\Phi(\tau)\,\mathrm{d}\tau\equiv f_0\,\Psi(t)$ . Why: this defines the relaxation function as the running integral of the response, so by construction $\Phi(t)=-\dot\Psi(t)$ — a pulse is the derivative of a step.

Step 3. Substitute the classical Kubo relation $\Phi(\tau)=-\beta\dot C(\tau)$ and integrate: $\Psi(t)=\int_t^\infty(-\beta\dot C(\tau))\,\mathrm{d}\tau=\beta[C(t)-C(\infty)]$ . Why: the integral of a total derivative is a boundary difference; the floor $C(\infty)=\langle A\rangle^2$ is the disconnected part that survives at long times.

Step 4. Evaluate at $t=0$ : $\Psi(0)=\beta[C(0)-C(\infty)]=\beta\langle\delta A^2\rangle=\chi(0)$ . Why: the equal-time connected correlation is the variance, so the zero-time relaxation amplitude is the isothermal susceptibility — a static, temperature-weighted fluctuation sum rule identical to $\chi(\omega=0)$ . *Verified: verify_aq-sm-response-relaxation-functions.py confirms $\Psi=\int_t^\infty\Phi\,\mathrm{d}\tau$ , the identity $\Phi=-\dot\Psi$ , and the static limit $\Psi(0)=\beta\langle\delta A^2\rangle$ on the exponential model.*

Go deeper · Derivation B · 1 minFrom the relaxation function to the complex susceptibility

Derivation BFrom the relaxation function to the complex susceptibility

The goal is to show that Fourier-transforming the relaxation function through the after-effect integral produces the complex susceptibility $\chi(\omega)=\chi(0)+i\omega\int_0^\infty\Psi(t)e^{i\omega t}\,\mathrm{d}t$ , and to extract the reactive and dissipative parts for the single-relaxation-time (Debye) model that describes the Ornstein–Uhlenbeck process and overdamped Langevin dynamics. The strategy is to integrate the response-function transform by parts once, using $\Phi=-\dot\Psi$ , and to insert the exponential relaxation $\Psi(t)=\chi(0)e^{-t/\tau}$ to read off the Debye loss peak.

Step 1. Start from the half-line transform of the response, $\chi(\omega)=\int_0^\infty\Phi(t)e^{i\omega t}\,\mathrm{d}t$ , and substitute $\Phi=-\dot\Psi$ : $\chi(\omega)=-\int_0^\infty\dot\Psi(t)e^{i\omega t}\,\mathrm{d}t$ . Why: the response is the time derivative of the relaxation function, so its transform relates to that of $\Psi$ by an integration by parts.

Step 2. Integrate by parts: $\chi(\omega)=-[\Psi e^{i\omega t}]_0^\infty+i\omega\int_0^\infty\Psi(t)e^{i\omega t}\,\mathrm{d}t=\Psi(0)+i\omega\int_0^\infty\Psi(t)e^{i\omega t}\,\mathrm{d}t$ . Why: $\Psi(\infty)=0$ kills the upper boundary term and the lower one gives $\Psi(0)=\chi(0)$ , so the susceptibility is the static value plus $i\omega$ times the transform of the after-effect function.

Step 3. Insert the exponential relaxation $\Psi(t)=\chi(0)e^{-t/\tau}$ : the integral is $\chi(0)\tau/(1-i\omega\tau)$ , so $\chi(\omega)=\chi(0)[1+i\omega\tau/(1-i\omega\tau)]=\chi(0)/(1-i\omega\tau)$ . Why: a single relaxation time yields the Debye form — the simplest causal susceptibility, and exactly the Ornstein–Uhlenbeck / overdamped-Langevin spectrum.

Step 4. Separate real and imaginary parts: $\chi'(\omega)=\chi(0)/(1+\omega^2\tau^2)$ and $\chi''(\omega)=\chi(0)\,\omega\tau/(1+\omega^2\tau^2)$ , which peaks at $\omega\tau=1$ . Why: the dissipative part is a loss peak whose maximum locates the relaxation frequency $1/\tau$ ; its height $\chi(0)/2$ and half-width fix the damping — the observable signature in dielectric and NMR spectra. *Verified: verify_aq-sm-response-relaxation-functions.py confirms $\chi(\omega)=\chi(0)/(1-i\omega\tau)$ , the split into $\chi',\chi''$ , and that $\chi''$ peaks at $\omega\tau=1$ with height $\chi(0)/2$ .*

Limiting cases

The single susceptibility $\chi(\omega)=\chi'+i\chi''$ collapses to familiar regimes:

Case	Result	Meaning
Static / thermodynamic	$\chi(0)=\Psi(0)=\beta\langle\delta A^2\rangle$	Zero-frequency response equals $\beta$ times the equilibrium variance — the isothermal susceptibility fixed by equal-time fluctuations.
Debye / single relaxation time	$\chi''(\omega)=\chi(0)\dfrac{\omega\tau}{1+\omega^2\tau^2}$	One loss peak at $\omega\tau=1$ ; the Ornstein–Uhlenbeck and overdamped-Langevin models live here.
High frequency	$\chi'(\omega)\to-\dfrac{\langle\dot A^2\rangle}{k_BT\,\omega^2},\ \chi''\to0$ faster	The system cannot follow a fast drive; the reactive part falls as an $f$ -sum-rule tail $\propto1/\omega^2$ .
Classical FDT	$\chi''(\omega)=\dfrac{\beta\omega}{2}\,S(\omega)$	The dissipative response is the correlation spectrum weighted by $\beta\omega/2$ — the high- $T$ limit of the quantum theorem.

Worked examples

Example 1 — The Debye triple for an overdamped Brownian particle in a trap. This computes the correlation, response, and relaxation functions of a real, concrete model and shows how the temperature drops out of the mechanical response. Take a particle of drag $\gamma$ in a harmonic trap of stiffness $k$ , so $\gamma\dot x=-kx+\xi(t)+f(t)$ with white noise $\langle\xi(t)\xi(t')\rangle=2\gamma k_BT\,\delta(t-t')$ and relaxation time $\tau=\gamma/k$ .

Step 1. Solve for the equilibrium correlation: the linear Langevin equation gives $C(t)=\langle x(t)x(0)\rangle=(k_BT/k)\,e^{-|t|/\tau}$ , with $C(0)=k_BT/k$ fixed by equipartition ( $\tfrac12 k\langle x^2\rangle=\tfrac12 k_BT$ ).

Step 2. Apply the classical Kubo relation for $t>0$ : $\Phi(t)=-\beta\dot C(t)=-\beta\cdot(k_BT/k)\cdot(-1/\tau)e^{-t/\tau}=(1/k\tau)e^{-t/\tau}$ . The $\beta=1/k_BT$ cancels the $k_BT$ in $C$ , leaving a purely mechanical amplitude $1/k\tau$ .

Step 3. Integrate to the relaxation function: $\Psi(t)=\int_t^\infty(1/k\tau)e^{-\tau'/\tau}\,\mathrm{d}\tau'=(1/k)e^{-t/\tau}$ , so $\Psi(0)=1/k$ .

Step 4. Read the static response: $\chi(0)=\Psi(0)=1/k$ . A constant force $f_0$ shifts the trap minimum by $f_0/k$ — exactly Hooke's law, temperature-independent, even though the fluctuation amplitude $C(0)=k_BT/k$ is not. The whole triple is $C(t)=(k_BT/k)e^{-t/\tau},\ \Phi(t)=(1/k\tau)e^{-t/\tau}\Theta(t),\ \Psi(t)=(1/k)e^{-t/\tau}$ with $\tau=\gamma/k$ : the Debye triple, the FDT in its simplest form.

Example 2 — Locating and sizing the dielectric loss peak. This turns the relaxation time into a measured frequency and shows the loss-peak height is fixed. Take a dipolar liquid with static susceptibility $\chi(0)=8.0$ (dimensionless) and molecular reorientation time $\tau=1.0\times10^{-11}\,\mathrm{s}$ (10 ps), a typical value near room temperature.

Step 1. Write the Debye susceptibility $\chi''(\omega)=\chi(0)\,\omega\tau/(1+\omega^2\tau^2)$ . Differentiate: the loss peak sits at $\omega_{\rm max}\tau=1$ , i.e. $\omega_{\rm max}=1/\tau=1.0\times10^{11}\,\mathrm{rad/s}$ .

Step 2. Convert to a measurement frequency: $f_{\rm max}=\omega_{\rm max}/2\pi=1.0\times10^{11}/(2\pi)\approx1.59\times10^{10}\,\mathrm{Hz}\approx\mathbf{15.9\ GHz}$ — squarely in the microwave band, which is why water's dielectric loss peaks around a few tens of GHz.

Step 3. Evaluate the peak height: at $\omega\tau=1$ , $\chi''_{\rm max}=\chi(0)\cdot1/(1+1)=\chi(0)/2=\mathbf{4.0}$ . The peak is exactly half the static susceptibility, independent of $\tau$ — a robust fingerprint of single-relaxation-time (Debye) response.

Step 4. Sanity-check the static sum rule via Kramers–Kronig: $\chi(0)=\tfrac{2}{\pi}\int_0^\infty\chi''(\omega)/\omega\,\mathrm{d}\omega=\tfrac{2}{\pi}\chi(0)\int_0^\infty\tau/(1+\omega^2\tau^2)\,\mathrm{d}\omega=\tfrac{2}{\pi}\chi(0)\cdot(\pi/2)=\chi(0)$ . The Debye pair is internally KK-consistent, closing the loop between response, relaxation, and dissipation.

Origins · 1 minresponse function and how this was discovered

response function

concept — the causal Green function of linear response · The response function Φ(t) is the impulse response of a system in equilibrium to a weak external field: the induced change of an observable is its convolution with the field. Its defining property is causality — Φ(t) vanishes for t < 0 — which alone forces the frequency-domain susceptibility to be analytic in the upper half-plane and to obey the Kramers–Kronig relations. The concept was crystallized by Ryogo Kubo's 1957 formulation of linear-response theory, which expressed Φ as an equilibrium average of a commutator, and by Herbert Callen and Theodore Welton's 1951 fluctuation–dissipation theorem, which tied its dissipative part to spontaneous fluctuations. Its classical incarnation Φ(t) = −β Ċ(t) Θ(t) makes Onsager's 1931 regression hypothesis quantitative, unifying the impulse, step, and sinusoidal responses of any equilibrium system into a single object.

History

1928

— Johnson measures and Nyquist explains thermal voltage noise in a resistor, an early response–fluctuation link

1931

— Onsager states the regression hypothesis: spontaneous fluctuations decay by the macroscopic relaxation law

1951

— Callen and Welton prove the general fluctuation–dissipation theorem, fixing the dissipative response by fluctuations

1957

— Kubo formulates linear-response theory, expressing the response function as an equilibrium commutator average

1966

— Kubo's review consolidates response, relaxation, and after-effect functions into one formalism

Going further

References. Kubo, Toda & Hashitsume, Statistical Physics II: Nonequilibrium Statistical Mechanics, Ch. 4 (linear response, relaxation functions) · Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 15 (linear response) · Kardar, Statistical Physics of Particles, Ch. 9 · Reif, Fundamentals of Statistical and Thermal Physics, Ch. 15 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 10 · Landau & Lifshitz, Statistical Physics Part 1, §123–126 · Chaikin & Lubensky, Principles of Condensed Matter Physics, Ch. 7.

Provenance. The response/relaxation identities are standard linear-response results. A verifier recomputes: (1) $\Psi=\int_t^\infty\Phi\,\mathrm{d}\tau$ and the identity $\Phi=-\dot\Psi$ ; (2) the static limit $\Psi(0)=\chi(0)=\beta\langle\delta A^2\rangle$ (Reif §15, Sethna Ch. 10); (3) the after-effect transform $\chi(\omega)=\chi(0)+i\omega\int_0^\infty\Psi e^{i\omega t}\mathrm{d}t$ and the Debye split $\chi''=\chi(0)\omega\tau/(1+\omega^2\tau^2)$ peaking at $\omega\tau=1$ (Kubo–Toda–Hashitsume §4.2); (4) the overdamped-Langevin triple $C,\Phi,\Psi$ with $\tau=\gamma/k$ , $C(0)=k_BT/k$ ; (5) the Example-2 numbers $f_{\rm max}\approx15.9$ GHz, $\chi''_{\rm max}=\chi(0)/2=4.0$ . No nonstandard constants asserted.

Builds on: The Fluctuation-Dissipation Theorem · The Langevin Equation · The Ornstein-Uhlenbeck Process · Brownian Motion. Leads to: The Einstein Relation · The Fokker-Planck Equation · The Master Equation.

Are the response function and the relaxation function really the same information?

Yes — they are two encodings of one causal Green function. The response function $\Phi(t)$ is the reaction to an impulse; the relaxation function $\Psi(t)$ is the free decay after a held field is released. They are tied exactly by $\Psi(t)=\int_t^\infty\Phi(\tau)\,\mathrm{d}\tau$ , so $\Phi=-\dot\Psi$ : one is the derivative of the other. Both build the same frequency susceptibility $\chi(\omega)$ . Knowing any one of $\Phi$ , $\Psi$ , or $\chi(\omega)$ over its full domain determines the other two; they differ only in the time boundary condition (pulse, step-off, or sinusoid) used to probe the identical microscopic dynamics.

Why does the classical Kubo relation carry a time derivative and a factor of β?

Quantum-mechanically the response is $\Phi=(i/\hbar)\Theta(t)\langle[A(t),A(0)]\rangle$ . In the classical limit the commutator becomes $i\hbar$ times a Poisson bracket, and integrating by parts over phase space against the Boltzmann weight (whose gradient brings down a factor $\beta$ ) converts the bracket into $-\beta$ times the time derivative of the correlation. So $\Phi(t)=-\beta\dot C(t)\Theta(t)$ . The derivative reflects that response comes from change in the correlation; the $\beta$ is the temperature that converts a mechanical perturbation into a statistical shift of populations.

Why is the static response 1/k for a trapped particle, independent of temperature, when the fluctuations scale with T?

The static response measures how far a constant force shifts the average position: a force $f_0$ moves the trap minimum by $f_0/k$ (Hooke's law), a purely mechanical fact with no $T$ in it. The fluctuation amplitude $C(0)=k_BT/k$ does scale with temperature, by equipartition. The classical Kubo relation $\Phi=-\beta\dot C$ reconciles them: the $\beta=1/k_BT$ exactly cancels the $k_BT$ in $C$ , so the response amplitude $1/k$ is temperature-free while the fluctuation it derives from is not. This cancellation is the fluctuation–dissipation theorem in miniature.

What is the physical meaning of the requirement ωχ″(ω) ≥ 0?

It is passivity — the statement that a system in thermal equilibrium can only absorb net energy from a weak periodic drive, never spontaneously deliver work to it. The time-averaged power fed in by a drive $f_0\cos\omega t$ is $\tfrac12\omega|f_0|^2\chi''(\omega)$ , so requiring this to be non-negative for every frequency forces $\omega\chi''\ge0$ . Because $\chi''$ is odd in $\omega$ , this means $\chi''>0$ for $\omega>0$ and $<0$ for $\omega<0$ . A violation would let you extract work by cycling a field — a perpetual-motion machine — so passivity is a thermodynamic stability condition, not a modelling choice.

How is Onsager's regression hypothesis actually used, rather than just stated?

Operationally it says: to predict how a driven displacement relaxes, you may compute the decay of a spontaneous fluctuation of the same observable in the undriven equilibrium system. This is enormously practical — it means transport coefficients and relaxation times can be extracted from equilibrium molecular-dynamics trajectories (via correlation functions) without ever applying a field, which avoids finite-field nonlinearities and driving artifacts. Conversely, measuring both the driven response and the spontaneous fluctuation and finding they disagree is the operational definition of a non-equilibrium state — the basis of effective-temperature measurements in glasses and active matter.

Does the relaxation function decay to zero or to a constant?

The relaxation function $\Psi(t)=\beta[C(t)-C(\infty)]$ is defined with the long-time floor $C(\infty)=\langle A\rangle^2$ already subtracted, so it decays to zero as $t\to\infty$ for an ergodic system that fully relaxes. If the system has a broken symmetry or a conserved quantity that prevents complete relaxation (e.g. a magnetization that does not fully decay), $C(t)$ approaches a value above $\langle A\rangle^2$ and $\Psi$ retains a residual plateau — a non-ergodicity parameter. In the glass literature this residual is precisely the order parameter that jumps at the ideal glass transition, so whether $\Psi\to0$ is itself physically diagnostic.