Kramers-Kronig Relations

A linear, time-translation-invariant system responds to a perturbation $f(t)$ through a convolution $x(t)=\int_{-\infty}^{\infty}\chi(t-t')f(t')\,\mathrm{d}t'$ , where $\chi(t)$ is the response function. One physical requirement — that an effect cannot precede its cause — forces $\chi(t)=0$ for $t<0$ . The Kramers–Kronig relations are the exact translation of this single statement into the frequency domain, and they carry a surprising punchline: the dispersion of a material and its absorption are not independent quantities you can specify separately. Measure how strongly a system absorbs at every frequency, and its refractive (reactive) response everywhere is already determined. The mechanism is analyticity. Because $\chi(t)$ vanishes for negative times, its Fourier transform $\chi(\omega)=\int_0^{\infty}\chi(t)e^{i\omega t}\,\mathrm{d}t$ converges and is analytic throughout the upper half of the complex $\omega$ -plane, where $\mathrm{Im}\,\omega>0$ makes $e^{i\omega t}$ decay. Analyticity plus the requirement that $\chi(\omega)\to0$ at high frequency lets Cauchy's theorem run on a contour closed in that half-plane, and out drops a pair of Hilbert-transform integrals linking $\chi'(\omega)=\mathrm{Re}\,\chi$ to $\chi''(\omega)=\mathrm{Im}\,\chi$ . The two functions are shadows of one analytic object; fix one over the whole axis and the other is uniquely determined. These relations are model-independent and exact, and they are ruthlessly practical. In optics the refractive index $n(\omega)$ and the extinction $\kappa(\omega)$ are Kramers–Kronig partners, so a broadband absorption spectrum yields the index it was too hard to measure directly. In AC transport, dielectric relaxation, and the retarded Green's functions of many-body theory the same machinery enforces internal consistency — any proposed response that violates the relations is acausal and unphysical. Within statistical mechanics the relations dovetail with the fluctuation–dissipation theorem: the dissipative part $\chi''$ is the spontaneous fluctuation spectrum up to thermal factors, and Kramers–Kronig then reconstructs the entire complex response from that noise, closing the circle between response, dissipation, and fluctuation.

Key takeaways

The Kramers–Kronig (KK) relations express the real part of a causal response function as a Hilbert transform of its imaginary part, and vice versa — one determines the other over all frequencies.
They follow purely from causality: $\chi(t)=0$ for $t<0$ forces $\chi(\omega)$ to be analytic in the upper half of the complex-frequency plane.
Physically they tie dispersion (the reactive, energy-storing response $\chi'$ ) to absorption (the dissipative response $\chi''$ ): you cannot have loss without dispersion.
Combined with the fluctuation–dissipation theorem, $\chi''$ encodes equilibrium fluctuations, so KK links the full complex response to spontaneous noise spectra.
Sum rules, such as the dielectric $f$ -sum rule fixing $\int_0^\infty\omega\,\varepsilon''\,\mathrm{d}\omega=(\pi/2)\omega_p^2$ , are integral consequences of KK plus high-frequency asymptotics.
Key idea: Causality $\Rightarrow$ analyticity in the upper half-plane $\Rightarrow$ Hilbert-transform relations $\chi'(\omega)=\tfrac1\pi\mathcal{P}\!\int\chi''(\omega')/(\omega'-\omega)\,\mathrm{d}\omega'$ between the real and imaginary parts of any physical linear response. Dispersion and absorption are two shadows of one analytic function.

The core insight

Causality $\Rightarrow$ analyticity in the upper half-plane $\Rightarrow$ Hilbert-transform relations $\chi'(\omega)=\tfrac1\pi\mathcal{P}\!\int\chi''(\omega')/(\omega'-\omega)\,\mathrm{d}\omega'$ between the real and imaginary parts of any physical linear response. Dispersion and absorption are not independent — each is fully fixed by the other over all frequencies.

Where this lives today

Kramers–Kronig analysis is the routine backbone of optical and spectroscopic metrology. Reflectivity measured over a broad band is inverted by KK to the complex dielectric function of a solid, from which conductivity, plasmon energies, and interband transitions are read; ellipsometry and terahertz time-domain spectroscopy lean on the same consistency check to flag non-causal artifacts in noisy data. In accelerator and synchrotron science the optical constants of every mirror coating are KK-reconstructed from absorption edges. Any measured or modelled response that fails KK is, by theorem, acausal — so the relations double as an error detector. The reach extends far beyond optics. In many-body theory retarded Green's functions are the analytic continuation of imaginary-time (Matsubara) correlators, and KK is what connects computed spectral functions to measurable response; the same dispersion relations constrain scattering amplitudes in particle physics (forward dispersion relations for pion–nucleon scattering were an early triumph) and underlie modern bounds on the analytic structure of scattering in the "S-matrix bootstrap." Wherever causality meets a response function — a resistor's impedance, a black hole's quasinormal ringdown, a material's refractive index — the Kramers–Kronig relations are the law that dispersion and dissipation must obey.

The physical picture

Strike a system with a sharp kick at $t=0$ . A causal system can only ring afterwards, never before — there is no pre-echo. This one-sidedness in time is precisely what makes the frequency response analytic in the upper half $\omega$ -plane, where the transform integral converges. Everything in Kramers–Kronig is the accounting of that analyticity: the response, being zero for negative times, hides no singularities above the real axis, and Cauchy's theorem then chains together the values of $\chi$ at different real frequencies. The consequence is a locked relationship between absorption and dispersion. Wherever a system absorbs energy — a peak in $\chi''$ at some resonance $\omega_0$ — it must also have a frequency-dependent reactive response $\chi'$ that swings up below the resonance and down above it, passing through its background value right at the peak. This "anomalous dispersion" that opticians see flanking every absorption line is not a separate physical effect; it is the mathematically forced partner of the absorption, and its shape is dictated to the last detail by the absorption curve. Think of $\chi'$ and $\chi''$ as two projections of a single analytic object rotating in the complex plane. You do not get to choose them independently: specify the absorption over the entire frequency axis and the dispersion is fixed, with no remaining freedom. That is why a single Lorentzian absorption peak forces one and only one dispersion curve — the S-shaped reactive response whose zero-crossing sits exactly at the absorption maximum.

Figure

Absorption χ″ locks dispersion χ′: the reactive S-curve crosses zero exactly at the absorption peak

Figure

Numerical proof: KK-reconstructed χ′ from the damped-oscillator χ″ lands on the analytic curve to under 1%

The core equations

The starting point is the general causal linear response: a system's output at time $t$ can depend only on inputs from earlier times.

x(t)=\int_{-\infty}^{\infty}\chi(t-t')\,f(t')\,\mathrm{d}t',\qquad \chi(\tau)=0\ \text{for}\ \tau<0 \qquad (1)

where: $x(t)$ — observed response (polarization, displacement, current); $f(t)$ — applied perturbation; $\chi(\tau)$ — retarded response (after-effect) function; $\tau=t-t'$ — elapsed time since the perturbation. The causal condition $\chi(\tau)=0$ for $\tau<0$ is the entire physical input; everything else is complex analysis. Fourier-transforming the causal kernel over the half line gives the complex susceptibility, which is analytic wherever the transform converges.

\chi(\omega)=\int_{0}^{\infty}\chi(\tau)\,e^{i\omega\tau}\,\mathrm{d}\tau=\chi'(\omega)+i\chi''(\omega) \qquad (2)

where: $\chi(\omega)$ — complex susceptibility; $\chi'$ — reactive (dispersive) part; $\chi''$ — dissipative (absorptive) part; $\omega$ — (possibly complex) angular frequency. For $\mathrm{Im}\,\omega>0$ the factor $e^{-(\mathrm{Im}\,\omega)\tau}$ damps the integrand, so $\chi(\omega)$ is analytic in the whole upper half-plane — the single fact that generates the relations. The Kramers–Kronig relations are the reciprocal Hilbert transforms tying the two parts. They are exact for any causal response that decays at infinity.

\chi'(\omega)=\frac{1}{\pi}\,\mathcal{P}\!\int_{-\infty}^{\infty}\frac{\chi''(\omega')}{\omega'-\omega}\,\mathrm{d}\omega',\qquad \chi''(\omega)=-\frac{1}{\pi}\,\mathcal{P}\!\int_{-\infty}^{\infty}\frac{\chi'(\omega')}{\omega'-\omega}\,\mathrm{d}\omega' \qquad (3)

where: $\mathcal{P}$ — Cauchy principal value, which handles the singularity at $\omega'=\omega$ by symmetric excision. The relations are non-local in frequency: the value of $\chi'$ at one $\omega$ needs $\chi''$ at all $\omega'$ , which is why sum rules and asymptotics matter. For a real time-domain response, $\chi'$ is even and $\chi''$ is odd, so the full-axis integral folds onto positive frequencies — the form used with experimental data, which lives at $\omega>0$ .

\chi'(\omega)=\frac{2}{\pi}\,\mathcal{P}\!\int_{0}^{\infty}\frac{\omega'\,\chi''(\omega')}{\omega'^2-\omega^2}\,\mathrm{d}\omega' \qquad (4)

where: $\omega'$ — positive integration frequency; the kernel $2\omega'/(\omega'^2-\omega^2)$ results from combining the $\pm\omega'$ contributions using $\chi''(-\omega')=-\chi''(\omega')$ . This is the optical/dielectric working form — for instance $n(\omega)-1=\tfrac{c}{\pi}\mathcal{P}\!\int_0^\infty\alpha(\omega')/(\omega'^2-\omega^2)\,\mathrm{d}\omega'$ reconstructs the refractive index from the absorption coefficient $\alpha$ . Combining KK with the exact high-frequency behaviour of a physical response produces sum rules — global constraints on spectral weight independent of microscopic detail.

\int_{0}^{\infty}\omega\,\varepsilon''(\omega)\,\mathrm{d}\omega=\frac{\pi}{2}\,\omega_p^2,\qquad \omega_p^2=\frac{n\,e^2}{\varepsilon_0\,m} \qquad (5)

where: $\varepsilon''$ — imaginary part of the relative permittivity; $\omega_p$ — plasma frequency; $n$ — electron number density; $e,m$ — electron charge and mass. The rule follows because at very high frequency all bound electrons respond as free particles, $\varepsilon(\omega)\to1-\omega_p^2/\omega^2$ , which fixes the first moment of the absorption. Spectral weight can migrate in frequency but the integrated $\omega\varepsilon''$ is conserved. The dissipative part is sign-constrained by passivity and, in equilibrium, is the thermally weighted fluctuation spectrum — the bridge to the fluctuation–dissipation theorem.

\omega\,\chi''(\omega)\ge0\ \ \text{(passivity)},\qquad \chi''(\omega)=\frac{\omega}{2k_BT}\,S(\omega)\ \ (\text{classical FDT}) \qquad (6)

where: $S(\omega)$ — power spectrum of equilibrium fluctuations; $k_BT$ — thermal energy. Passivity forbids net energy delivery from a weak drive; the FDT link means that measuring the noise fixes $\chi''$ , and Kramers–Kronig then rebuilds the full complex $\chi$ .

Derivation AThe Kramers-Kronig relations from causality and a closed contour

The goal is to derive the dispersion relations directly from the single physical input $\chi(\tau<0)=0$ , by proving that the frequency response is analytic in the upper half-plane and then applying Cauchy's theorem to a contour that skirts the real-axis pole. The strategy is to write the half-line transform, argue its analyticity from the exponential damping for $\mathrm{Im}\,\omega>0$ , close the contour with a large upper semicircle that contributes nothing when $\chi\to0$ , and pick up the half-residue from the small indentation around the singular point. Keep careful track of the sign of the half-residue — a clockwise indentation contributes $-i\pi\chi(\omega)$ — because that sign is what distinguishes the two KK relations.

Step 1. Note that $\chi(\omega)=\int_0^\infty\chi(\tau)e^{i\omega\tau}\,\mathrm{d}\tau$ converges for $\mathrm{Im}\,\omega\ge0$ , so $\chi(\omega)$ is analytic in the open upper half-plane and continuous on the real axis. Why: only $\tau\ge0$ contributes because $\chi$ is causal, and for $\mathrm{Im}\,\omega>0$ the factor $e^{-(\mathrm{Im}\,\omega)\tau}$ damps the tail — analyticity is a direct gift of causality.

Step 2. Form the contour integral $\oint_C\chi(\omega')/(\omega'-\omega)\,\mathrm{d}\omega'=0$ , with $C$ running along the real axis (indented above the pole at $\omega'=\omega$ ) and closed by a large semicircle in the upper half-plane. Why: no singularities are enclosed — $\chi$ is analytic inside and the pole is excluded by the indentation — so by Cauchy's theorem the closed integral vanishes.

Step 3. Discard the large arc: assuming $\chi(\omega')\to0$ as $|\omega'|\to\infty$ fast enough, the semicircle at infinity contributes nothing, leaving the principal-value integral plus the small indentation. Why: physical responses vanish at high frequency because no medium can follow an infinitely fast drive, so the arc term is legitimately dropped (or subtracted if $\chi_\infty\ne0$ ).

Step 4. Evaluate the small clockwise semicircle around $\omega'=\omega$ ; it yields $-i\pi\chi(\omega)$ , so $\mathcal{P}\!\int\chi(\omega')/(\omega'-\omega)\,\mathrm{d}\omega'-i\pi\chi(\omega)=0$ , i.e. $\chi(\omega)=\tfrac{1}{i\pi}\mathcal{P}\!\int\chi(\omega')/(\omega'-\omega)\,\mathrm{d}\omega'$ . Why: a half-circle of radius $\to0$ picks up half the residue with a sign set by its orientation; this master relation is complex and contains both KK relations.

Step 5. Multiply by $1/i=-i$ and separate real and imaginary parts of $\chi=\chi'+i\chi''$ ; equating pieces gives $\chi'=\tfrac1\pi\mathcal{P}\!\int\chi''/(\omega'-\omega)$ and $\chi''=-\tfrac1\pi\mathcal{P}\!\int\chi'/(\omega'-\omega)$ . Why: the two shadows of one analytic function are Hilbert transforms of each other, so any absorption band ( $\chi''\ne0$ ) mandates frequency-dependent dispersion ( $\chi'$ ) — loss without dispersion is impossible. *Verified: verify_aq-sm-kramers-kronig-relations.py confirms the master relation's sign and reconstructs $\chi'$ from $\chi''$ numerically for the damped oscillator to under 1%.*

Go deeper · Derivation B · 1 minThe dielectric f-sum rule from KK plus the free-electron high-frequency limit

Derivation BThe dielectric f-sum rule from KK plus the free-electron high-frequency limit

The goal is to derive the Thomas–Reiche–Kuhn-type $f$ -sum rule $\int_0^\infty\omega\,\varepsilon''\,\mathrm{d}\omega=(\pi/2)\omega_p^2$ by feeding the folded KK relation its own high-frequency asymptotics, where every electron responds as if free. The strategy is to expand the KK kernel at large $\omega$ , identify the leading $1/\omega^2$ tail of the dispersion with the exact free-electron result $\varepsilon(\omega)\to1-\omega_p^2/\omega^2$ , and read off the first moment of the absorption. The physical content is that spectral weight is conserved: however bound-state resonances redistribute absorption across frequency, the integrated $\omega\varepsilon''$ counts the total electron density and nothing else.

Step 1. Take the folded KK relation for $\varepsilon'(\omega)-1$ and expand the kernel for $\omega\to\infty$ : $1/(\omega'^2-\omega^2)\approx-1/\omega^2$ , giving $\varepsilon'(\omega)-1\simeq-\tfrac{2}{\pi\omega^2}\int_0^\infty\omega'\varepsilon''(\omega')\,\mathrm{d}\omega'$ . Why: at frequencies far above every resonance the integration variable $\omega'\ll\omega$ , so the kernel simplifies and the whole absorption spectrum collapses into a single moment.

Step 2. Independently, at very high frequency bound electrons cannot respond and behave as free charges, giving the exact limit $\varepsilon'(\omega)-1\to-\omega_p^2/\omega^2$ with $\omega_p^2=ne^2/\varepsilon_0 m$ . Why: Newton's equation for a free electron in an oscillating field gives a $-1/\omega^2$ polarizability whose coefficient is the plasma frequency squared — this is model-independent.

Step 3. Match the two $1/\omega^2$ coefficients: $\tfrac{2}{\pi}\int_0^\infty\omega'\varepsilon''(\omega')\,\mathrm{d}\omega'=\omega_p^2$ , hence $\int_0^\infty\omega\,\varepsilon''(\omega)\,\mathrm{d}\omega=\tfrac{\pi}{2}\omega_p^2$ . Why: two expansions of the same physical $\varepsilon'$ must agree term by term, and the leading tail fixes the first moment of the absorption uniquely.

Step 4. Interpret: the sum rule counts electrons. Since $\omega_p^2\propto n$ , the total absorption-weighted spectral area is set by the electron density alone, independent of how the resonances are arranged. Why: this is why "missing spectral weight" is a diagnostic — e.g. the superconducting condensate in the Ferrell–Glover–Tinkham rule removes weight from finite $\omega$ into a $\delta$ -function at $\omega=0$ while the total is conserved. *Verified: verify_aq-sm-kramers-kronig-relations.py checks the $f$ -sum-rule constant $(\pi/2)\omega_p^2$ against a numerical integral of $\omega\varepsilon''$ for a Lorentz-oscillator dielectric.*

Limiting cases

KK holds for any causal response, but its precise form depends on the low- and high-frequency behaviour:

Case	Result	Meaning
Standard, $\chi(\infty)=0$	$\chi'(\omega)=\tfrac1\pi\mathcal{P}\!\int_{-\infty}^{\infty}\dfrac{\chi''(\omega')}{\omega'-\omega}\,\mathrm{d}\omega'$	The basic Hilbert-transform pair, valid when the response decays at high frequency.
Nonzero background $\chi_\infty$	$\chi(\omega)-\chi_\infty=\tfrac1{i\pi}\mathcal{P}\!\int\dfrac{\chi(\omega')-\chi_\infty}{\omega'-\omega}\,\mathrm{d}\omega'$	Subtract the constant (e.g. $\varepsilon_\infty$ in a dielectric) before applying KK.
DC pole (metallic conductivity)	$\sigma(\omega)=\dfrac{\sigma_0}{1-i\omega\tau}$ ; a $\delta(\omega)$ Drude term adds to KK	A pole at $\omega=0$ contributes a separate delta piece the standard relation misses.
Reality / parity	$\chi'(-\omega)=\chi'(\omega),\ \chi''(-\omega)=-\chi''(\omega)$	Real $\chi(t)$ makes $\chi'$ even, $\chi''$ odd — permitting the folded positive-axis form.

Worked examples

Example 1 — Reconstructing dispersion from absorption for a damped oscillator, numerically. This proves KK is a computable identity, not a slogan, by recovering $\chi'$ from $\chi''$ and checking against the exact answer. Take the standard damped-oscillator susceptibility $\chi(\omega)=1/(\omega_0^2-\omega^2-i\gamma\omega)$ with $\omega_0=1$ and $\gamma=0.3$ (units of $\omega_0$ ).

Step 1. Split into parts: $\chi'(\omega)=(\omega_0^2-\omega^2)/D$ and $\chi''(\omega)=\gamma\omega/D$ , with $D=(\omega_0^2-\omega^2)^2+\gamma^2\omega^2$ . At $\omega=\omega_0=1$ : $\chi'(\omega_0)=0$ and $\chi''(\omega_0)=\gamma\omega_0/(\gamma^2\omega_0^2)=1/(\gamma\omega_0)=1/0.3\approx\mathbf{3.33}$ — the absorption peak, and the dispersion crosses zero exactly there.

Step 2. Feed only $\chi''(\omega')$ into the folded KK integral $\chi'_{\rm KK}(\omega)=\tfrac{2}{\pi}\mathcal{P}\!\int_0^\infty\omega'\chi''(\omega')/(\omega'^2-\omega^2)\,\mathrm{d}\omega'$ , evaluated on a fine grid with the principal value taken by symmetric excision around $\omega'=\omega$ .

Step 3. Compare the reconstruction with the analytic $\chi'$ at several frequencies. At $\omega=0.5$ : analytic $\chi'=(1-0.25)/[(0.75)^2+(0.3\cdot0.5)^2]=0.75/0.585=\mathbf{1.282}$ ; the KK integral returns $1.28$ to three figures. Across $0.2\le\omega\le2$ the reconstruction matches the analytic curve to better than 1%.

Step 4. Draw the lesson: the entire real (dispersive) response was rebuilt from the imaginary (absorptive) response alone, using nothing but causality. Dispersion is not independent data — it is the KK transform of absorption, and the sub-1% agreement is the numerical fingerprint of that lock.

Example 2 — The dielectric $f$ -sum rule as an electron counter. This turns an abstract sum rule into a concrete number. Take a single Lorentz oscillator $\varepsilon(\omega)=1+\omega_p^2/(\omega_0^2-\omega^2-i\gamma\omega)$ modelling one bound resonance, with plasma frequency $\omega_p$ .

Step 1. Its absorption is $\varepsilon''(\omega)=\omega_p^2\,\gamma\omega/[(\omega_0^2-\omega^2)^2+\gamma^2\omega^2]$ , a Lorentzian peaked near $\omega_0$ .

Step 2. Compute the first moment $\int_0^\infty\omega\,\varepsilon''(\omega)\,\mathrm{d}\omega$ numerically; the integral returns $(\pi/2)\,\omega_p^2$ to high accuracy, independent of the resonance position $\omega_0$ and the damping $\gamma$ .

Step 3. Interpret with real numbers: for aluminium, $n=1.8\times10^{29}\,\mathrm{m^{-3}}$ conduction electrons give $\omega_p=\sqrt{ne^2/\varepsilon_0 m}=\sqrt{(1.8\times10^{29})(1.602\times10^{-19})^2/[(8.854\times10^{-12})(9.109\times10^{-31})]}\approx\mathbf{2.4\times10^{16}\ rad/s}$ , an ultraviolet plasma frequency ( $\hbar\omega_p\approx15.8$ eV).

Step 4. Conclude: no matter how the bound-electron resonances shuffle absorption around in frequency, the KK-derived sum rule pins $\int_0^\infty\omega\varepsilon''\,\mathrm{d}\omega=(\pi/2)\omega_p^2$ to the electron density. The $f$ -sum rule literally counts electrons, which is why spectroscopists use it to calibrate absolute absorption spectra.

Origins · 1 minralph de laer kronig and how this was discovered

ralph de laer kronig

1904–1995, German-American theoretical physicist · Ralph de Laer Kronig derived the dispersion relation between the real and imaginary parts of the atomic scattering amplitude in 1926, translating causality into the analytic structure of the response; Hendrik Kramers gave the complementary optical form in 1927, and the two together are the Kramers–Kronig relations. Kronig had earlier, in 1925, proposed that the electron carries an intrinsic spin (angular momentum ℏ/2) — the idea Goudsmit and Uhlenbeck published months later — but was dissuaded from publishing it by Pauli's skepticism, one of the most famous near-misses in physics. He went on to contribute the Kronig–Penney model of electrons in a periodic potential (1931), a cornerstone of solid-state band theory, and spent much of his career at Delft. His dispersion relations remain the daily tool by which optical constants are extracted from absorption data.

History

1925

— Kronig proposes intrinsic electron spin but is dissuaded from publishing by Pauli's skepticism

1926

— Kronig derives the dispersion relation for X-ray scattering amplitudes from causality

1927

— Kramers gives the optical dispersion relation linking refractive index and absorption

1931

— Kronig and Penney publish the periodic-potential band model

1937

— Titchmarsh's theorem places the causality-analyticity equivalence on rigorous ground

Going further

References. L. D. Landau & E. M. Lifshitz, Statistical Physics Part 1 (3rd ed.), §123–125 and Electrodynamics of Continuous Media §82–84 · R. Kubo, M. Toda & N. Hashitsume, Statistical Physics II, Ch. 4 · M. Kardar, Statistical Physics of Particles, Ch. 9 · J. D. Jackson, Classical Electrodynamics (3rd ed.), §7.10 · H. B. Callen, Thermodynamics and an Introduction to Thermostatistics, ch. on fluctuation–dissipation.

Provenance. The dispersion relations are standard, from Landau–Lifshitz §123–125 and Electrodynamics of Continuous Media §82–84, Jackson §7.10, and Kubo–Toda–Hashitsume Ch. 4. A verifier recomputes: (1) the contour derivation giving the master relation $\chi(\omega)=(i\pi)^{-1}\mathcal{P}\!\int\chi(\omega')/(\omega'-\omega)\,\mathrm{d}\omega'$ ; (2) the numerical reconstruction of $\chi'$ from $\chi''$ for the damped oscillator ( $\omega_0=1,\gamma=0.3$ ), matching the analytic $\chi'(0.5)=1.282$ to <1%; (3) the folded-form Debye pair $\chi'=\chi_0/(1+\omega^2\tau^2)$ , $\chi''=\chi_0\omega\tau/(1+\omega^2\tau^2)$ ; (4) the $f$ -sum-rule constant $(\pi/2)\omega_p^2$ by numerical integration of $\omega\varepsilon''$ for a Lorentz oscillator, and the aluminium $\omega_p\approx2.4\times10^{16}$ rad/s. Kramers–Kronig is a cross-track tool (optics owns the bare slug at cutover); the optical dispersion form $n(\omega)-1$ is cross-referenced as a label. No nonstandard numbers asserted.

Builds on: The Langevin Equation · The Fluctuation-Dissipation Theorem · The Ornstein-Uhlenbeck Process. Leads to: The Fluctuation-Dissipation Theorem · Kramers' Escape Rate Theory · Generalized Susceptibility and Spectral Functions.

Do Kramers–Kronig relations really give the real part "for free" from the imaginary part?

Mathematically yes: knowing $\chi''(\omega)$ over the entire frequency axis determines $\chi'(\omega)$ everywhere through the principal-value integral, and vice versa. The "for free" is genuine but comes with a catch — the integral is non-local, so it needs the absorption at all frequencies, not just near your frequency of interest. In practice one measures a finite window and must supplement it with modelled or asymptotic tails; missing high-frequency weight biases the reconstruction. The only physical assumption beyond $\chi(t)=0$ for $t<0$ is that $\chi(\omega)\to0$ (or a known constant) at infinity — everything else is complex analysis.

Does KK work with truncated experimental data over a finite frequency window?

Approximately, with care. The KK integral formally needs the full spectrum, so truncating at a maximum measured frequency introduces error, largest near the edges of the window. Practitioners handle this by anchoring the reconstruction with a measured reference point (a "singly subtracted" KK, which cancels much of the tail sensitivity), by appending physically motivated high-frequency tails (free-electron $1/\omega^2$ ), or by using sum rules as consistency checks. The reconstruction is most trustworthy in the interior of a broad window and degrades toward its ends — which is why spectroscopists prize the widest possible spectral coverage.

Is there any physics hiding in "causality" beyond χ(t)=0 for t<0?

Very little — that is the beauty of it. The single input is that the response cannot begin before the perturbation, which is $\chi(t)=0$ for $t<0$ . The one technical addition is a growth condition: $\chi(\omega)$ must be bounded (in fact $\to0$ ) at high frequency, guaranteed physically because no medium can respond to an infinitely fast drive. Given these, Titchmarsh's theorem makes the equivalence rigorous: causality in time $\Leftrightarrow$ upper-half-plane analyticity $\Leftrightarrow$ the Hilbert-transform (KK) relations. There is no additional dynamical assumption, which is exactly why KK holds model-independently for resistors, dielectrics, scattering amplitudes, and Green's functions alike.

Why does an absorption line force a specific dispersion curve around it?

Because $\chi'$ and $\chi''$ are two projections of one analytic function, KK ties them point by point. A localized absorption peak in $\chi''$ contributes to the principal-value integral for $\chi'$ with a kernel $1/(\omega'-\omega)$ that is positive below the peak and negative above it, producing a reactive response that rises just below the resonance, crosses its background value exactly at the peak, and dips just above — the S-shaped "anomalous dispersion." The stronger and narrower the absorption, the sharper the dispersion swing. You cannot draw an absorption line without its dispersion partner; they are the same causal object.

How do Kramers–Kronig relations connect to the fluctuation–dissipation theorem?

They are complementary halves of the response-theory toolkit. The fluctuation–dissipation theorem fixes the dissipative part: classically $\chi''(\omega)=\omega S(\omega)/2k_BT$ , tying absorption to the equilibrium fluctuation spectrum $S(\omega)$ . But that alone gives only the imaginary part. Kramers–Kronig then supplies the reactive part $\chi'(\omega)$ from $\chi''$ , reconstructing the entire complex susceptibility — including the static response $\chi(0)=\tfrac{2}{\pi}\int_0^\infty\chi''(\omega)/\omega\,\mathrm{d}\omega$ — with no further dynamical input. Together they close the loop: fluctuations fix dissipation (FDT), and causality fixes dispersion from dissipation (KK).

Why the Cauchy principal value, and what happens if I forget it?

The KK integrand has a pole at $\omega'=\omega$ sitting right on the integration path, so the ordinary integral diverges. The principal value tames it by excising a symmetric interval around the pole and taking the limit as the interval shrinks — the divergent contributions from the two sides cancel by symmetry, leaving a finite, well-defined number. Physically this excision is the real-axis half-residue that the contour derivation produces. Forgetting the principal value (e.g. naively integrating through the singularity numerically) gives a divergent or grid-dependent garbage answer; correct KK codes always implement the symmetric excision or an equivalent Hilbert-transform algorithm.