Roughness model, Debye-Waller¶

Scope of this page¶

This page documents the Debye-Waller factor as the multiplicative roughness correction appropriate to the long-correlation-length regime of interface roughness, where the diffuse-scattering contribution sits angularly close to the specular reflection and cannot be cleanly separated by a finite-aperture detector. The treatment follows Esashi [1, Sec. 3, Eq. (2)] with explicit reference to the original Debye-Waller and Sinha-Sirota-Garoff-Stanley derivations [2, 3]. The Debye-Waller factor sits as the long-correlation companion to the Nevot-Croce factor of Nevot-Croce, and the choice between them is governed by the criterion in the Selection guide.

Closed-form correction factors¶

For a Gaussian distribution of interface heights with rms amplitude \(\sigma_{i,i+1}\), the Debye-Waller corrections to the reflection and transmission Fresnel amplitudes at the interface between layers \(i\) and \(i+1\) are [1, Eq. (2)]

The reflection exponent carries \(k_{i,z}^{2}\) rather than the \(k_{i,z}k_{i+1,z}\) product that appears in the Nevot-Croce factor, and the transmission exponent has the opposite overall sign relative to the Nevot-Croce form. The two structural differences trace to the underlying scattering geometry [3]. The Nevot-Croce factor follows from the distorted-wave Born approximation in the regime where diffuse scattering escapes the specular cone, so the reflection amplitude is only the coherent contribution and the difference exponent in the transmission factor reflects the energy conservation between specular and diffuse channels. The Debye-Waller factor follows from the same DWBA but in the opposite limit, where diffuse scattering remains inside the specular cone and is folded into the measured signal. The correction therefore symmetrizes around \(k_{i,z}^{2}\) rather than \(k_{i,z}k_{i+1,z}\), and the transmission exponent flips sign because the diffuse channel removes rather than adds to the forward amplitude.

The two factors agree in the limit \(k_{i,z} \to k_{i+1,z}\), where the index contrast vanishes, and they agree numerically to sub-percent precision for \(\sigma < 1\) nm in typical X-ray geometries [1, Figs. 15-18]. They diverge most strongly in the total external reflection regime near the critical angle, where \(k_{i+1,z}\) becomes nearly imaginary and the product \(k_{i,z}k_{i+1,z}\) acquires a strong phase that the Debye-Waller form does not carry.

Application to the eight Passler-Paarmann amplitudes¶

The application logic mirrors the Nevot-Croce treatment. For co-polarized channels the projection \(k_{i,z}^{(p)}\) uses the \(p\)-mode eigenvalue \(q_{i1}\) and similarly for the \(s\) channel. For the cross-polarization channels we adopt the same geometric-mean prescription as in Nevot-Croce under the substitution \(k_{i,z}k_{i+1,z} \to k_{i,z}^{2}\),

with the analogous expression for \(r_{sp}^{\text{rough}}\). As with Nevot-Croce, the cross-polarization extension lacks a published derivation in the birefringent setting, and the regression scaffold is the audit instrument for the choice. The kernel emits a KernelError::UnsupportedConversion diagnostic if the comparison against the graded model fails by more than \(10^{-4}\) in the small-roughness regime.

The transmission corrections take the same form,

with the explicit minus sign that distinguishes the Debye-Waller transmission factor from the Nevot-Croce form.

When to choose Debye-Waller over Nevot-Croce¶

The selection between the two multiplicative factors is controlled by the correlation-length regime of the roughness. Define the angularly-resolved diffuse-scattering bandwidth \(\xi_{\text{diff}} = \lambda/(1 - \cos\theta)\), where \(\lambda\) is the vacuum wavelength and \(\theta\) is the grazing angle of incidence [1, text following Eq. (2)]. If the in-plane correlation length \(\xi\) of the roughness satisfies \(\xi \ll \xi_{\text{diff}}\), the diffuse scattering is angularly broad and can be separated from the specular reflection by a finite-aperture detector. Nevot-Croce applies in this regime. If instead \(\xi \gg \xi_{\text{diff}}\), the diffuse scattering crowds against the specular peak and the detector integrates over both. Debye-Waller applies in this regime.

For X-ray and EUV reflectivity the binding case is short correlation length. Polished silicon wafers have \(\xi \sim 1\) \(\mu\)m, and metal-coated fused quartz has \(\xi\) in the \(100\text{-}200\) nm range [1, references 42 and 43]. At \(\lambda = 0.154\) nm and \(\theta = 5\,\text{deg}\), \(\xi_{\text{diff}} \approx 40\) nm, which sits below the typical roughness correlation length and thereby places most XRR geometries in the Nevot-Croce regime. At \(\lambda = 13.5\) nm and \(\theta = 30\,\text{deg}\), \(\xi_{\text{diff}} \approx 100\) nm, which puts EUV reflectometry on metal-coated optics close to the regime boundary, where neither factor is strictly correct and the graded-interface approach is the safer choice.

For visible-wavelength magneto-optic applications the picture inverts. At \(\lambda = 500\) nm and \(\theta = 45\,\text{deg}\), \(\xi_{\text{diff}} \approx 1.7\) \(\mu\)m, which exceeds typical roughness correlation lengths on polished substrates. The Debye-Waller factor becomes the relevant multiplicative model, not Nevot-Croce. This inversion is the reason the kernel does not silently default to Nevot-Croce across all wavelengths but requires the user to either pass an explicit choice or rely on the auto-dispatcher in the Selection guide.

Validity regime¶

The same perturbative bound \(\sigma k_{i,z} < 1\) governs Debye-Waller as it does Nevot-Croce, and the same small-index-contrast caveat applies. The third condition specific to Debye-Waller is the long-correlation requirement \(\xi \gg \xi_{\text{diff}}\). The kernel rejects a Debye-Waller choice that violates either bound at validation time. When the correlation length sits inside the crossover region \(\xi \sim \xi_{\text{diff}}\), the kernel issues a warning and the user is advised to compare against a graded-interface calculation rather than rely on either multiplicative model.

The numerical disagreement between Nevot-Croce and Debye-Waller in their respective domains of strict validity remains modest in the small-\(\sigma\) regime [1, Figs. 15-18], but the disagreement grows rapidly through the total external reflection region near the critical angle. Users fitting XRR data near the critical angle should be especially cautious about model selection, and the regression scaffold tracks this sensitivity through a dedicated test that sweeps \(\theta\) across the critical angle for both factors and reports the pointwise discrepancy.

Where the code lives¶

In refloxide, the Debye-Waller model is implemented in core::roughness::debye_waller. The module exposes DebyeWaller { sigma_nm: f64, correlation_length_nm: f64 } implementing RoughnessModel. The correct_interface method applies the eight exponential corrections analogously to Nevot-Croce, and the validate method enforces the three conditions \(\sigma k_z < 1\), small index contrast, and \(\xi \gg \xi_{\text{diff}}\).

The Python wrapper exposes the model through Roughness.debye_waller(sigma_nm=..., correlation_length_nm=...). The correlation length is mandatory because it gates the validity check, and the kernel rejects construction without it.

Validation¶

Two regression tests gate the Debye-Waller implementation in addition to the sharp-interface recovery shared with the other two models.

The first is critical-angle agreement with the graded model, where the Debye-Waller amplitudes must agree with a fully discretized graded calculation across the total external reflection region to within \(10^{-5}\). This test catches the most common Debye-Waller failure mode, sign errors in the transmission exponent that produce subtle phase shifts in the critical-angle dip. It lives in tests/regression/test_roughness_debye_critical_angle.py.

The second is co-polarized agreement with Nevot-Croce in the small-roughness, low-contrast regime, where the two multiplicative factors must agree to \(10^{-4}\) when \(\sigma k_z < 0.1\) and the index contrast is below \(10^{-3}\). This is the regime where both factors should reduce to the same DWBA leading-order correction, and disagreement between them indicates a transcription error in one of the two implementations. It lives in tests/regression/test_roughness_debye_vs_nevot.py.

Limitations and known failure modes¶

The Debye-Waller failure modes parallel those of Nevot-Croce with one addition. The shared failures are large \(\sigma k_z\), high index contrast, and non-Gaussian profiles. The Debye-Waller-specific failure is the short-correlation regime, where the diffuse scattering escapes the specular cone and the Debye-Waller assumption that all scattered power remains detected becomes incorrect. The kernel rejects this regime at validation time, so the failure mode is loud rather than silent.

A second Debye-Waller-specific concern is the absence of a published derivation for cross-polarization channels in birefringent media, identical to the Nevot-Croce situation. The geometric-mean prescription used in the kernel is a working ansatz, and the regression scaffold catches its breakdown through the cross-polarization comparison test against the graded model.

References¶

1. Y. Esashi, M. Tanksalvala, Z. Zhang, N. W. Jenkins, H. C. Kapteyn, and M. M. Murnane, "Influence of surface and interface roughness on X-ray and extreme ultraviolet reflectance: A comparative numerical study," OSA Continuum 4, 1497 (2021). DOI.
2. P. Debye, "Interferenz von Roentgenstrahlen und Waermebewegung," Ann. Phys. 348, 49 (1913). DOI.
3. S. K. Sinha, E. B. Sirota, S. Garoff, and H. B. Stanley, "X-ray and neutron scattering from rough surfaces," Phys. Rev. B 38, 2297 (1988). DOI.
4. V. Holy, J. Kubena, I. Ohlidal, K. Lischka, and W. Plotz, "X-ray reflection from rough layered systems," Phys. Rev. B 47, 15896 (1993). DOI.