Kramers–Kronig Relations

We take the general theory from aq-sm-kramers-kronig-relations: the statistical-mechanics track owns the causal linear response function $\chi(t)$ , the theorem that causality forces analyticity in the upper half of the complex-frequency plane, the Hilbert-transform pair with its principal value, the contour argument and Titchmarsh's theorem behind it, and the bridge to the fluctuation–dissipation theorem. None of that is re-derived here — it is recalled by name and put to work on light. What optics adds is a startling practical statement: measure only how much a material absorbs, at every frequency, and you have determined its refractive index at every frequency too — without ever measuring a phase or bending a ray. The optical dispersion relations make $n(\omega)-1$ and the extinction coefficient $\kappa(\omega)$ a locked pair: each is a principal-value integral of the other. This article owns that pair and its two consequences: the f-sum rule, pinning the frequency-weighted absorption integral to $\tfrac{\pi}{2}\omega_p^2$ — a count of electrons, however chemistry rearranges the lines — and the anomalous-dispersion S-curve every line wears, here a theorem rather than a model output. The chapter's Lorentz oscillator (aq-optics-lorentz-oscillator-model) is the test bench: feed only its absorption $\chi''$ into the Kramers–Kronig integral and out comes $\chi'(589\ \mathrm{nm})=1.286$ against the analytic $1.287$ — dispersion rebuilt from absorption alone, to a tenth of a percent.

Key takeaways

The optical Kramers–Kronig (KK) relations lock $n(\omega)-1$ to $\kappa(\omega)$ (equivalently $\chi'$ to $\chi''$ ): each is a principal-value integral of the other over all positive frequencies.
The sole physical input is causality — polarization cannot precede the field. The causality → analyticity → Hilbert-pair machinery is owned by aq-sm-kramers-kronig-relations, recalled by name, not re-derived.
The optical f-sum rule $\int_0^\infty\omega\,\varepsilon''(\omega)\,\mathrm{d}\omega=\tfrac{\pi}{2}\omega_p^2$ pins the integrated absorption to the electron density alone — $1.96350$ in $\omega_0=1$ units for the model solid, independent of damping.
Concrete check (recall aq-optics-lorentz-oscillator-model): KK rebuilds $\chi'(589\ \mathrm{nm})=1.286$ from $\chi''$ alone vs the analytic $1.287$ ; the S-curve crosses zero exactly on the absorption peak.
Key idea: causality welds dispersion and absorption into one analytic function — absorption everywhere determines the refractive index everywhere, and the total absorption counts the electrons.

The core insight

Causality makes $n(\omega)-1$ and $\kappa(\omega)$ the real and imaginary shadows of one analytic function: the optical dispersion relations compute either from a principal-value integral of the other, and their high-frequency limit fixes the f-sum rule $\int_0^\infty\omega\,\varepsilon''\,\mathrm{d}\omega=\tfrac{\pi}{2}\omega_p^2$ — absorption anywhere fixes dispersion everywhere.

Where this lives today

Kramers–Kronig inversion is the production line of optical-constants metrology. Handbook values of $n$ and $\kappa$ for metals and semiconductors are not measured independently: reflectance is recorded from the far infrared to the vacuum ultraviolet, extended with Drude and free-electron tails, and KK-transformed to recover the phase and hence $\tilde n(\omega)$ — every coating design and ellipsometry fit inherits the theorem. The sum rules ride along as auditors: a spectrum that misses $(\pi/2)\omega_p^2$ has lost spectral weight — in superconductors that bookkeeping (Ferrell–Glover–Tinkham) weighs the condensate; terahertz spectroscopy checks its measured phase against the KK transform of its amplitude to flag artefacts. The theorem also underwrites the rest of this chapter. The Sellmeier equation of aq-optics-sellmeier-cauchy-equations is the KK integral (6) evaluated in a transparent window with the distant UV and IR bands idealized as poles — why its fitted coefficients land on real absorption edges. And the anomalous dispersion of aq-optics-normal-anomalous-dispersion is KK's local signature: slow light in electromagnetically induced transparency and superluminal group velocity in gain media are engineered by sculpting $\chi''$ — causality permits no other currency.

The physical picture

Strike a dielectric with an impulse of field at $t=0$ . The bound electrons ring — but only afterwards: no pre-echo, because the polarization is built from the field's past, never its future. That single statement, $G(\tau)=0$ for $\tau<0$ , is the entire physical content of this article. The rest is complex analysis, and it is not ours: aq-sm-kramers-kronig-relations proves that a one-sided response has a Fourier transform analytic throughout the upper half $\omega$ -plane, whose real and imaginary parts on the axis are Hilbert transforms of each other. Optics reads the theorem in its own variables: the real part is refraction, the imaginary part is absorption — two projections of one analytic object.

Figure

Causality → analyticity → the optical lock: the response cannot precede its cause, so χ(ω) is analytic in the upper half ω-plane, and n − 1 and κ become a Hilbert-transform pair — specify one and the other is forced

The consequences read like magic but are bookkeeping. A material that absorbs nowhere has $\chi''\equiv0$ , and the Hilbert transform of zero is zero: $n\equiv1$ , vacuum. Refraction requires absorption somewhere — glass bends light only because it absorbs in the ultraviolet. Conversely, every absorption line drags a dispersive fingerprint: the kernel is positive below the line and negative above, so $n$ must swing up on the red side, crash through its background exactly at the peak, and dip on the blue side — the S-curve of aq-optics-normal-anomalous-dispersion, forced by causality. The second figure runs the theorem as a computation on the chapter's Lorentz solid ( $\omega_0=1.884\times10^{16}$ rad/s, $\chi(0)=1.25$ , $\gamma=0.02\,\omega_0$ ; recall aq-optics-lorentz-oscillator-model): only $\chi''$ is fed in, and the dispersion that emerges threads the line in the forced S-shape, landing on the analytic $\chi'$ to a tenth of a percent.

Figure

The punchline, computed for the Lorentz solid in ω₀ = 1 units: feeding only the absorption χ″ (red, peak 62.5) into the KK integral rebuilds χ′ (blue S-curve) — KK gives χ′(589 nm) = 1.286 vs analytic 1.287; the ω-weighted χ″ area is the f-sum (π/2)ωₚ² = 1.96350; resonance region arcsinh-compressed

The core equations

The optical response is a causal convolution: the polarization now is assembled from the field's past through a real memory kernel. It is the optical instance of the general response function whose theory aq-sm-kramers-kronig-relations owns; $G$ is the medium's dielectric after-effect function.

\mathbf{P}(t)=\varepsilon_0\int_{0}^{\infty} G(\tau)\,\mathbf{E}(t-\tau)\,\mathrm{d}\tau,\qquad G(\tau)=0\ \text{for}\ \tau<0 \qquad (1)

where: $\mathbf{P}$ — polarization density; $\mathbf{E}$ — applied field; $G(\tau)$ — real dielectric response (memory) function; $\tau$ — time since the field acted. The lower limit $0$ is causality: the medium has no access to the field's future. Fourier-transforming the one-sided kernel gives the complex susceptibility. Because the integral runs only over $\tau>0$ , pushing $\omega$ upward in the complex plane multiplies the integrand by a decaying $e^{-(\mathrm{Im}\,\omega)\tau}$ — the convergence aq-sm-kramers-kronig-relations converts into full analyticity.

\chi(\omega)=\int_0^{\infty} G(\tau)\,e^{i\omega\tau}\,\mathrm{d}\tau=\chi'(\omega)+i\chi''(\omega) \qquad (2)

where: $\chi'$ — dispersive (in-phase) part, which sets $n$ ; $\chi''$ — absorptive (quadrature) part, which sets $\kappa$ and $\alpha$ . Analyticity of $\chi$ in the upper half-plane is the recalled theorem, not an assumption. Because $G(\tau)$ is real, the susceptibility obeys a crossing condition that fixes the parity of its parts — what lets the full-axis Hilbert transform be folded onto the positive frequencies an experiment measures.

\chi(-\omega)=\chi^*(\omega)\ \Rightarrow\ \chi'(-\omega)=\chi'(\omega),\qquad \chi''(-\omega)=-\chi''(\omega) \qquad (3)

where: $\chi^*$ — complex conjugate. $\chi'$ is even, $\chi''$ odd; the verify script checks both for the Lorentz model. Folding the recalled Hilbert-transform pair with the parity (3) produces the working optical form of the Kramers–Kronig relations — the version spectroscopists code.

\chi'(\omega)=\frac{2}{\pi}\,\mathrm{P}\!\int_0^{\infty}\frac{\omega'\,\chi''(\omega')}{\omega'^2-\omega^2}\,\mathrm{d}\omega',\qquad \chi''(\omega)=-\frac{2\omega}{\pi}\,\mathrm{P}\!\int_0^{\infty}\frac{\chi'(\omega')}{\omega'^2-\omega^2}\,\mathrm{d}\omega' \qquad (4)

where: $\mathrm{P}$ — Cauchy principal value (symmetric excision of the pole at $\omega'=\omega$ ); $\omega'$ — integration frequency. The folded kernel changes sign across the pole — which is what carves the S-curve. The same premises hold for $\tilde n(\omega)-1=n-1+i\kappa$ : analytic in the upper half-plane, vanishing at high frequency like $-\omega_p^2/2\omega^2$ (no medium follows an infinitely fast drive). Applying (4) with $\chi'\to n-1$ , $\chi''\to\kappa$ gives the optical dispersion relations — the centrepiece this article owns.

n(\omega)-1=\frac{2}{\pi}\,\mathrm{P}\!\int_0^{\infty}\frac{\omega'\,\kappa(\omega')}{\omega'^2-\omega^2}\,\mathrm{d}\omega',\qquad \kappa(\omega)=-\frac{2\omega}{\pi}\,\mathrm{P}\!\int_0^{\infty}\frac{n(\omega')-1}{\omega'^2-\omega^2}\,\mathrm{d}\omega' \qquad (5)

where: $n$ — real refractive index; $\kappa$ — extinction coefficient (recall aq-optics-complex-refractive-index-extinction for $\tilde n=n+i\kappa$ ). Refraction and extinction are one analytic function's two shadows: tabulate either over all frequencies and the other is computable. Substituting the Beer–Lambert connection $\kappa=c\,\alpha/(2\omega)$ cancels the extra $\omega'$ in the kernel and turns (5) into the spectrometer-facing statement.

n(\omega)-1=\frac{c}{\pi}\,\mathrm{P}\!\int_0^{\infty}\frac{\alpha(\omega')}{\omega'^2-\omega^2}\,\mathrm{d}\omega' \qquad (6)

where: $\alpha(\omega')$ — intensity absorption coefficient, what a transmission measurement delivers; $c$ — vacuum light speed. Measure $\alpha$ broadband, extrapolate the tails, integrate — how the optical-constants handbooks are built. Finally, feeding the high-frequency limit back into (4) yields the optical f-sum rule — a model-independent constraint on the total absorption strength, derived below.

\int_0^{\infty}\omega\,\varepsilon''(\omega)\,\mathrm{d}\omega=\frac{\pi}{2}\,\omega_p^2,\qquad \omega_p^2=\frac{Ne^2}{\varepsilon_0 m} \qquad (7)

where: $\varepsilon''=2n\kappa$ — imaginary part of the relative permittivity; $\omega_p$ — plasma frequency; $N$ — electron number density; $e,m$ — electron charge magnitude and mass. Binding can shuffle absorption between lines but never change the integral: the $\omega$ -weighted absorption spectrum counts electrons.

Derivation AFrom causality to the optical dispersion relations

The goal is to assemble the optical dispersion relations (5)–(6) — the $n\leftrightarrow\kappa$ lock this article owns — from the single physical input that the dielectric response is causal, importing the causality→analyticity→Hilbert-transform theorem by name from aq-sm-kramers-kronig-relations rather than re-deriving its contour machinery. The strategy: state the recalled pair, fold it onto positive frequencies with the reality condition, promote it to the complex index, substitute the absorption coefficient.

Step 1. Establish causality of the optical response: the polarization (1) is a convolution over the field's past only, $G(\tau)=0$ for $\tau<0$ — bound charges cannot respond to a field that has not arrived. Why: this is the only physics that enters — every later line is inherited complex analysis.

Step 2. Recall from aq-sm-kramers-kronig-relations (stated, not re-derived): a causal response has $\chi(\omega)$ analytic in the upper half-plane, and a contour closed there with a principal-value indentation at the real-axis pole yields the Hilbert-transform pair $\chi'(\omega)=\tfrac1\pi\mathrm{P}\!\int_{-\infty}^{\infty}\chi''(\omega')/(\omega'-\omega)\,\mathrm{d}\omega'$ and its sign-flipped partner; Titchmarsh's theorem makes causality ⇔ analyticity ⇔ Hilbert pair rigorous. Why: the statistical-mechanics owner keeps the derivation; optics applies it to light, and the by-name recall marks the seam.

Step 3. Fold onto positive frequencies: split the full-axis integral at $\omega'=0$ , substitute $\omega'\to-\omega'$ in the negative half, and use the parity (3); the integrands combine through $\tfrac{1}{\omega'-\omega}+\tfrac{1}{\omega'+\omega}=\tfrac{2\omega'}{\omega'^2-\omega^2}$ , giving (4). Why: experiments measure only positive ω; the folding identity is checked symbolically.

Step 4. Promote to the index. $\tilde n(\omega)-1$ satisfies the same two premises as $\chi$ : analytic in the upper half-plane (inherited from $\varepsilon=\tilde n^2$ , whose values avoid the negative real axis there, so the square root cuts nothing), and vanishing at high frequency, $\tilde n-1\to-\omega_p^2/2\omega^2$ . The recalled theorem applies verbatim with $\chi'\to n-1$ , $\chi''\to\kappa$ , producing (5). Why: KK is a property of an analytic class, not one function; this promotion, not the contour, is the optics-owned step.

Step 5. Substitute $\kappa(\omega')=c\,\alpha(\omega')/(2\omega')$ into the first relation of (5): the numerator's $\omega'$ cancels, leaving equation (6). Why: α is what a spectrophotometer reads — the form in which the theorem becomes an instrument: absorption in, refractive index out.

Go deeper · Derivation B · 1 minThe optical f-sum rule from the high-frequency tail

Derivation BThe optical f-sum rule from the high-frequency tail

The goal is the model-independent constraint (7), $\int_0^\infty\omega\,\varepsilon''\,\mathrm{d}\omega=\tfrac{\pi}{2}\omega_p^2$ : the frequency-weighted absorption spectrum of any material integrates to a number fixed by its electron density alone. The strategy: expand the folded KK relation for $\varepsilon'-1$ far above every resonance, where the whole absorption spectrum collapses into a single moment, and match it against the exact free-electron limit — two asymptotic expansions of one function, equated term by term.

Step 1. Fix the high-frequency limit physically. Far above every resonance, binding and damping are negligible beside inertia, so each of the $N$ electrons per unit volume responds as a free charge: Newton gives $x=eE/m\omega^2$ (opposite the force), hence $\varepsilon(\omega)\to1-\omega_p^2/\omega^2$ with $\omega_p^2=Ne^2/(\varepsilon_0 m)$ . Why: chemistry sets where electrons absorb, but at high enough frequency they all shake like free particles — which is what makes the rule model-independent; the verify script confirms $(\varepsilon'-1)\,x^2\to-\chi_0$ at $x=50$ , $200$ , $1000$ .

Step 2. Expand the folded KK relation in the same limit. For $\omega\gg\omega'$ across the absorbing range, the kernel becomes $2\omega'/(\omega'^2-\omega^2)\approx-2\omega'/\omega^2$ , so $\varepsilon'(\omega)-1\to-\tfrac{2}{\pi\omega^2}\int_0^\infty\omega'\,\varepsilon''(\omega')\,\mathrm{d}\omega'$ . Why: far above the spectrum the pole is harmless, and the entire $\varepsilon''$ curve enters only through its first moment.

Step 3. Match the two $1/\omega^2$ coefficients: $\tfrac{2}{\pi}\int_0^\infty\omega'\,\varepsilon''(\omega')\,\mathrm{d}\omega'=\omega_p^2$ — the f-sum rule (7). Why: two valid expansions of one analytic function must agree term by term; causality plus one universal limit fixes the total absorption strength with no model of the lines.

Step 4. Evaluate for the model solid. In $\omega_0=1$ units the single Lorentz line has $\chi_0=1.25$ , so the rule predicts $\int_0^\infty x\,\chi''(x)\,\mathrm{d}x=\tfrac{\pi}{2}\times1.25=1.96350$ ; direct integration returns exactly that — and the same value when the damping is reset to $g=0.05$ or $0.2$ . Why: the invariance under γ is the rule's whole content — spectral weight is conserved, merely redistributed, which is why "missing spectral weight" is a meaningful experimental diagnosis.

Limiting cases

Case	Result	Meaning
Static limit $\omega\to0$	$\chi'(0)=\dfrac{2}{\pi}\int_0^{\infty}\dfrac{\chi''(\omega')}{\omega'}\,\mathrm{d}\omega'$	DC susceptibility = weighted integral of all absorption — the model solid returns $\chi_0=1.25$ , $n(0)=1.5$
High frequency	$\varepsilon\to1-\omega_p^2/\omega^2$ ⇒ f-sum rule	every medium is a free-electron plasma to a hard X-ray; matching the tail to KK counts electrons
Single narrow Lorentz line	$\chi''$ Lorentzian ⇒ $\chi'$ S-curve, zero at $\omega_0$ , extrema at $\omega_0\mp\gamma/2$	the anomalous band is exactly the linewidth — line shape fixes the swing, detail for detail
Transparent window	$n-1\approx\dfrac{2}{\pi}\int_{\text{bands}}\dfrac{\omega'\kappa(\omega')}{\omega'^2-\omega^2}\,\mathrm{d}\omega'$	with $\kappa\approx0$ locally, the index is set by distant UV/IR bands — the KK far-tail Sellmeier parametrizes
Background $\varepsilon_\infty\ne1$	apply KK to $\varepsilon(\omega)-\varepsilon_\infty$	KK needs a response vanishing at $\infty$ ; subtract the constant first, or the integrals lie

Common trapAbsorption and refraction are independent properties, and a finite slice of absorption data fixes the index locally

Both halves are wrong, and together they are how KK analysis fails in practice. $n(\omega)$ and $\kappa(\omega)$ are not independent — they are one analytic function's real and imaginary parts, so tabulating them separately either duplicates information or contradicts causality. And the lock is non-local in frequency: $n$ at one $\omega$ needs $\chi''$ at all $\omega'$ , so truncating the integral to a measured window silently biases the result, worst near the window's edges. Practitioners extrapolate with physical tails (Drude at low $\omega$ , free-electron fall-off at high) or anchor a singly subtracted relation where $n$ is known — and never skip the principal value: integrating through the pole returns grid-dependent garbage.

Worked examples

1 · KK rebuilds the Lorentz index from its absorption line alone

Take the chapter's single-resonance solid (recall aq-optics-lorentz-oscillator-model): resonance at $\lambda_0=100$ nm so $\omega_0=1.884\times10^{16}$ rad/s, $\chi_0=1.25$ , $\gamma=0.02\,\omega_0$ . Using only the absorption spectrum $\chi''(\omega)$ , compute $\chi'$ at the sodium d-line, 589.3 nm, and compare with the analytic Lorentz value.

Step 1. Go dimensionless. With $x=\omega/\omega_0$ the Lorentz pair reads $\chi''(x)=\chi_0\,g\,x/[(1-x^2)^2+(gx)^2]$ and $\chi'(x)=\chi_0(1-x^2)/[(1-x^2)^2+(gx)^2]$ , $g=0.02$ . The test point is $x_t=\omega(589.3\ \mathrm{nm})/\omega_0=100/589.3=0.170$ — far below resonance, where glass-like transparency lives.

Step 2. Set up $\chi'(x_t)=\tfrac{2}{\pi}\mathrm{P}\!\int_0^\infty x\,\chi''(x)/(x^2-x_t^2)\,\mathrm{d}x$ and respect its two hazards: the principal-value pole at $x=x_t$ (excise $\pm0.05$ symmetrically) and the sharp resonance at $x=1$ (a quadrature break point there; split the infinite range at $x=5$ so no break point rides an infinite limit).

Step 3. Integrate. The three pieces $[0,x_t-0.05]$ , $[x_t+0.05,5]$ , $[5,\infty)$ sum to $\chi'_{\mathrm{KK}}(x_t)=\mathbf{1.286}$ .

Step 4. Compare with the analytic Lorentz dispersion: $\chi'(x_t)=\chi_0(1-x_t^2)/[(1-x_t^2)^2+(gx_t)^2]=\mathbf{1.287}$ — agreement to $0.1\%$ , and the implied index $n=\sqrt{1+1.286}=\mathbf{1.512}$ reproduces the Lorentz article's direct value. Causality alone converted an absorption line at 100 nm into a refractive index at 589 nm. KK gives χ′(589 nm) = 1.286 from χ″ alone, vs analytic 1.287; n = 1.512.

2 · The f-sum rule as an electron counter

For the same model solid, verify the optical f-sum rule numerically and use it to count the electrons.

Step 1. In $\omega_0=1$ units the rule (7) predicts $\int_0^\infty x\,\chi''(x)\,\mathrm{d}x=\tfrac{\pi}{2}\chi_0=\tfrac{\pi}{2}\times1.25=\mathbf{1.96350}$ , since $\omega_p^2/\omega_0^2=\chi(0)=\chi_0$ for a single Lorentz line.

Step 2. Integrate the absorption spectrum directly (splitting at the resonance and at $x=5$ ): the numerical value is $1.96350$ — and stays $1.96350$ when the damping is reset to $g=0.05$ or $0.2$ . Broadening lowers the peak ( $\chi_0/g$ : $62.5\to6.25$ ) but the $\omega$ -weighted area is untouchable: spectral weight is conserved.

Step 3. Convert to a count. Restoring units, $\omega_p^2=\chi_0\,\omega_0^2$ gives $\omega_p=\sqrt{1.25}\times1.884\times10^{16}=\mathbf{2.106\times10^{16}\ rad/s}$ , and inverting $\omega_p^2=Ne^2/(\varepsilon_0 m)$ gives $N=\omega_p^2\varepsilon_0 m/e^2=\mathbf{1.394\times10^{29}\ m^{-3}}$ — a sensible electron density for a solid. This is why the rule serves as an absolute calibration: integrate a measured $\omega\varepsilon''$ and you must recover the electron density — any shortfall means the experiment missed spectral weight. ∫₀^∞ ω ε″ dω = (π/2)ωₚ² = 1.96350 (ω₀ = 1 units); N = 1.394×10²⁹ m⁻³.

Physics of the portrait

Every equation here descends from one sentence of physics — the response cannot precede the cause — and that sentence, not any person, is the portrait: a metaphysical-sounding principle with a hard quantitative payload.

Origins · 1 mincausality and how this was discovered

causality

no effect before its cause — the single physical input that welds dispersion to absorption · optics of dispersive media

Causality entered optics as an escape from model-dependence. By the early 1920s, measured dispersion near absorption lines was well fitted by resonance models — but the models were classical scaffolding in a quantum world. The way out was to ask what must hold whatever the microscopic machinery. Ralph Kronig, analysing X-ray dispersion in 1926, and Hendrik Kramers, addressing the Como congress in 1927, found the answer: if the medium's response cannot precede the field, the dispersive and absorptive parts are Hilbert transforms of each other — full stop. A decade later Titchmarsh made the logic watertight: causality in time, analyticity of $\chi(\omega)$ in the upper half-plane, and the transform pair are strictly equivalent — three languages for one fact; Bode's network theory and particle physics' dispersion relations reuse the same structure. In optics its daily meaning is concrete: absorption spectra become refractive indices, optical constants are audited for acausal errors, and the f-sum rule counts a material's electrons from its spectrum. The principle costs nothing to state and forbids almost everything: of all conceivable pairs of $n(\omega)$ and $\kappa(\omega)$ , causality admits only the Hilbert-locked ones.

History

1926

— Kronig derives dispersion relations connecting X-ray dispersion to absorption from the requirement of causality

1927

— Kramers presents the optical dispersion relations at the Como congress

1937

— Titchmarsh proves the rigorous equivalence of causality and upper-half-plane analyticity

1945

— Bode's network analysis publishes the engineering twin: gain–phase relations for stable amplifiers

Going further

Texts. Hecht, Optics, 5th ed., §3.5–3.6 (dispersion, causality) · Foot, Atomic Physics, §7 (susceptibility, absorption and dispersion) · Loudon, The Quantum Theory of Light, 3rd ed., Ch. 1–2 (KK relations, sum rules) · Saleh & Teich, Fundamentals of Photonics, 2nd ed., §5.5 (resonant media, KK) · Jackson, Classical Electrodynamics, 3rd ed., §7.10 (causality and KK) · Landau & Lifshitz, Electrodynamics of Continuous Media, §82–84 · Demtröder, Laser Spectroscopy (dispersion and absorption profiles) Key numerical anchors for this article: KK reconstruction gives $\chi'(589\ \mathrm{nm})=1.286$ vs the analytic Lorentz value $1.287$ ; the f-sum $=(\pi/2)\chi_0=1.96350$ , independent of damping; $\chi'(0)=1.25$ ; $\chi''(\omega_0)=62.5$ ; $\omega_p=2.106\times10^{16}$ rad/s; $N=1.394\times10^{29}\ \mathrm{m^{-3}}$ .

Builds on: Dielectric Susceptibility and Permittivity · Electric Polarization and the Displacement Field · The Lorentz Oscillator Model · Complex Refractive Index and the Extinction Coefficient Leads to: Dispersion: Normal and Anomalous · Absorption Coefficient and Beer–Lambert Law · Sellmeier and Cauchy Dispersion Equations · Group Velocity and Group Index

Why can't a material refract light without absorbing somewhere?

Because the dispersion relation (5) computes $n(\omega)-1$ as an integral of $\kappa$ over all frequencies: if $\kappa\equiv0$ everywhere, the integral is zero everywhere and $n\equiv1$ — vacuum. Any index different from 1, at even one frequency, certifies absorption somewhere. For glass that somewhere is the deep ultraviolet and the infrared; the visible window sits between the debts. It also fixes the sign of normal dispersion: the kernel weights the UV absorption more strongly as $\omega$ climbs toward it, so $n$ rises toward the blue — the prism's spread is the shadow of ultraviolet absorption.

My absorption data only cover a finite band — can I still use KK?

Yes, carefully; everyone's data are finite. The integral formally needs $\chi''$ from 0 to $\infty$ , so the honest procedure is: measure wide, extrapolate physically, anchor. Extrapolation uses the tails physics dictates — Drude at low frequency for conductors, free-electron fall-off at high. Anchoring uses a singly subtracted relation: write KK for $n(\omega)-n(\omega_1)$ with $\omega_1$ a frequency where the index is independently known — the subtraction cancels most sensitivity to the unmeasured tails. Reconstructions are most reliable deep inside a broad window and degrade toward its edges.

What does this article own that aq-sm-kramers-kronig-relations doesn't?

The statistical-mechanics article owns the general theory: the causal response function $\chi(t)$ , analyticity in the upper half-plane, the contour derivation of the Hilbert-transform pair with its principal value, Titchmarsh's theorem, and the fluctuation–dissipation connection — for any linear response. This article owns the optical instance: the promotion from $\chi$ to $\tilde n-1$ , the dispersion relations between $n-1$ and $\kappa$ , the absorption-coefficient form (6) spectroscopists integrate, the f-sum rule as an electron counter, and the Lorentz-line reconstruction with numbers. Need the theorem proved? Go there. Need an index from an absorption spectrum? Done here.

Is the f-sum rule classical or quantum?

Both — that is its power. The derivation here is classical: KK plus the free-electron high-frequency limit. Quantum mechanics reproduces the identical statement as the Thomas–Reiche–Kuhn sum rule: the oscillator strengths of all transitions from any state sum to the electron number, $\sum_j f_j=Z$ . The correspondence is exact because the high-frequency limit probes only the commutator $[x,p]=i\hbar$ and the electron count — none of the binding details. That is why the rule survived the quantum revolution while the Lorentz model's literal springs did not: it is a statement about causality and electron number, not any model of the lines.

Is Sellmeier's equation secretly a Kramers–Kronig integral?

Essentially yes. In a transparent window the local $\kappa$ is negligible, so equation (5) says the index there is generated entirely by the distant absorption bands. Idealize each band as a delta function in $\kappa$ at $\omega_j$ and the KK integral evaluates in closed form to a sum of terms $B_j\lambda^2/(\lambda^2-C_j)$ : exactly the Sellmeier form of aq-optics-sellmeier-cauchy-equations. That is why fitted Sellmeier wavelengths land on real electronic (UV) and vibrational (IR) absorption edges, the coefficients are positive, and the formula holds to $10^{-6}$ across a glass catalogue: not an arbitrary fit but the causal far-field of the absorption spectrum.

Do the relations hold for gain media and negative-index metamaterials?

Yes — causality does not negotiate. A pumped gain medium has $\chi''<0$ over the amplifying band (it violates passivity, not causality), so KK still links its dispersion to its gain spectrum; that lock constrains the anomalous dispersion, and hence the superluminal group velocities, achievable in amplifiers. Metamaterials and negative-index media must equally obey KK, with real teeth: the relations force any strongly dispersive engineered response to carry loss, and the standard bounds on broadband cloaking and low-loss negative index descend from them. A response that fails a KK check is acausal and wrong — the theorem doubles as physics' error detector.