# Pitfalls, Limitations, and Extensions *Every theorem comes with a user manual, and the Kramers–Kronig relations are no exception. This chapter lists the mistakes that account for most wrongly-reconstructed dispersions, and sketches the extensions that rescue K-K in settings where the textbook form fails.* The pitfalls below apply in either the physics (`e^{−iωt}`) or engineering (`e^{+jωt}`) convention, with signs adjusted accordingly (Chapter 16). ## Pitfall 1: Wrong sign convention By far the most common mistake. Reading a physics-convention K-K formula and substituting engineering-convention data — or vice versa — produces a reconstruction that is the correct shape but with the imaginary part negated, or the sign of a Lorentzian dispersion wiggle flipped around its resonance. The diagnostic: the reconstructed χ″ for a known passive medium comes out negative at frequencies where absorption is clearly positive. The fix: declare the convention at the top of every derivation and every data-processing pipeline, and check the sign of χ″ at one reference frequency against a known passive-medium datum (e.g. water in the visible, where ε″ > 0 in physics convention for all ω > 0). ## Pitfall 2: Applying K-K to non-passive / active systems K-K holds for any **linear causal** system — passivity is not required for the dispersion relations themselves. What passivity buys is the additional constraint Im χ(ω) ≥ 0 for ω > 0 (physics convention). Active systems — laser gain media, negative-resistance oscillators, parametric amplifiers — are linear and causal and obey K-K, but Im χ < 0 over the gain band. Attempting to interpret a reconstruction by "absorption must be non-negative" will flag that region as an error when it is, in fact, correct physics. The fix: separate causality (which K-K enforces) from passivity (which it does not). Use K-K for consistency checks; use positivity tests only when you already know the medium is passive. ## Pitfall 3: Unsubtracted form when χ(∞) ≠ 0 The textbook integral ``` χ′(ω) = (1/π) P ∫_{−∞}^{∞} χ″(ω′) / (ω′ − ω) dω′ ``` assumes χ(ω) → 0 as |ω| → ∞. Real systems often have χ(∞) = χ_∞ finite (dielectric constant at optical frequencies, background susceptibility in NMR). Using the unsubtracted form then gives a divergent or grossly-biased reconstruction. The fix: use the singly-subtracted form (Chapter 15), which absorbs the constant into an explicit χ_∞ term and leaves a convergent integrand. ## Pitfall 4: Neglecting branch cuts of log r in reflectance K-K The reflectance K-K (Chapter 12) expresses the phase of a Fresnel reflection coefficient r(ω) as a Hilbert transform of log|r(ω)|. But log r(ω) has branch points wherever r(ω) = 0 — and zeros of r occur naturally at the Brewster angle, at interference minima of thin films, and more generally wherever the amplitude dips to zero. Each such zero contributes an extra 2π phase jump that must be tracked explicitly; naive numerical K-K of log|r| misses these contributions and gives the wrong phase by integer multiples of 2π in regions near zeros. The fix: count zeros of r inside the UHP (equivalently, count sign changes in r(ω) as ω crosses minima), and add the corresponding Blaschke-product phase to the K-K-reconstructed phase. ## Pitfall 5: DC conductivity pole A conductive medium has a DC conductivity σ_dc contributing a singular imaginary part ``` ε″(ω) = σ_dc / (ε₀ ω) as ω → 0 ``` This 1/ω pole makes the unsubtracted K-K integral divergent at ω = 0. The fix: subtract the pole analytically before numerical K-K. Decompose ε(ω) = ε_bound(ω) + i σ_dc/(ε₀ ω), apply K-K to ε_bound, and note that the 1/ω pole is its own K-K partner up to a constant (the Hilbert transform of 1/ω is a delta at ω = 0, handled via the distributional definition). Electrochemical impedance spectroscopy has standard tools (KK-Test, Boukamp's algorithm) that do this correctly. ## Pitfall 6: Finite measurement bandwidth The K-K integral needs data from 0 to ∞; measurement always gives a finite band. Truncating the integral biases the reconstruction, especially within a few linewidths of the band edges. The bias is systematic, not random — so averaging many measurements does not help. The fix: extrapolate the measured data with analytic tails matched to the known asymptotic behaviour (Drude at DC for conductors, sum-rule- consistent Lorentzian at high ω for optical response) — see Chapter 29. Never simply zero-extend. ## Pitfall 7: Non-linear systems K-K in its standard form applies to **linear** time-invariant systems. Non-linear systems do not admit a simple frequency-domain transfer function, and the straightforward K-K argument collapses. That said, extensions exist: - **Nonlinear K-K for χ^(n)** (Scandolo, Chernyak, Peiponen): causality of the n-th order nonlinear response kernel χ^(n)(ω_1, …, ω_n) produces dispersion relations in each frequency argument separately, with subtractions and multi-dimensional Hilbert transforms. Substantially more subtle than the linear case. - **Z-scan and pump-probe data** are routinely K-K-analysed using the nonlinear form to cross-check χ^(3) and two-photon-absorption coefficients. The reference is V. Lucarini et al., *Kramers–Kronig Relations in Optical Materials Research*, Springer, 2005, Ch. 10, which works through the Scandolo nonlinear dispersion relations in detail. ## Pitfall 8: Non-stationary (time-varying) systems Causality still holds for linear time-varying systems, but the simple one-dimensional K-K is replaced by a **bi-frequency** dispersion relation on the two-dimensional generalised transfer function H(ω_in, ω_out). Each frequency variable satisfies a K-K-type relation at fixed slices of the other, and the full constraint is a 2D Hilbert transform on the appropriate slice. These forms appear in radar with fast-moving targets, parametric amplifiers with periodic pumping, and time-varying biological signal processing. ## Pitfall 9: Random / stochastic cases For stationary random processes, the causality of the cross- correlation function translates into a K-K-type relation between the real and imaginary parts of the **cross-power spectral density** (CPSD). Specifically, if x(t) and y(t) are jointly stationary and the impulse response relating them is causal, then Re S_xy(ω) and Im S_xy(ω) are a Hilbert pair. This is the formal basis for fluctuation–dissipation theorems: the imaginary part of a linear response function (the dissipative part) is tied to the auto-spectrum of equilibrium fluctuations, and the real part follows by K-K. The Johnson–Nyquist noise of a resistor and the noise temperature of a passive microwave component are both direct consequences. ## Extensions: K-K holography and K-K receivers Two modern applications deserve mention. ### K-K holography (coherent optics) In coherent imaging, the complex field at the detector carries both magnitude and phase information, but an intensity detector measures only |E|². Phase retrieval via K-K is possible when the object's complex transmittance is the boundary value of a function analytic in an appropriate half-plane (engineered by placing a strong reference beam carrier frequency on one side of the object spectrum). The logarithm of the measured intensity can then be Hilbert-transformed to give the phase. Widely used in electron microscopy and digital holographic microscopy. ### K-K coherent receivers in fibre optics Mecozzi, Antonelli, and Shtaif (Optica, 2016) showed that a **single- polarisation, intensity-only** photodiode followed by a K-K phase reconstructor can recover the full complex optical field if a continuous-wave (CW) tone is co-transmitted with the signal at sufficient power and on one side of the signal spectrum. The tone ensures the total field is the boundary value of an UHP-analytic function, whose log-magnitude determines phase via K-K. This removed the need for a local oscillator in a class of coherent receivers and is now deployed in commercial 400G/800G coherent transceivers. Both are examples of **engineered causality**: the experimental setup is arranged so that the measured signal has the right analyticity domain, and K-K then recovers what would otherwise be missing phase. ## See also - [15_subtracted_dispersion.md](15_subtracted_dispersion.md) - [16_sign_conventions.md](16_sign_conventions.md) - [25_vna_measurements.md](25_vna_measurements.md) - [29_numerical_implementation.md](29_numerical_implementation.md) ## References - V. Lucarini, J. J. Saarinen, K.-E. Peiponen, E. M. Vartiainen, *Kramers–Kronig Relations in Optical Materials Research*, Springer, 2005. - S. Scandolo, F. Bassani, "Nonlinear sum rules: The three-level and the anharmonic-oscillator models", *Phys. Rev. B* 51, 6925, 1995. - A. Mecozzi, C. Antonelli, M. Shtaif, "Kramers–Kronig coherent receiver", *Optica*, vol. 3, no. 11, pp. 1220–1227, 2016. - B. A. Boukamp, "A linear Kronig–Kramers transform test for immittance data validation", *J. Electrochem. Soc.* 142, 1885, 1995.