REVIEW 2 major objections 5 minor 4 references
X-ray Thermal diffuse scattering from real-space displacement correlations
T0 review · 2 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read One Fourier transform computes all-order thermal diffuse scattering within the harmonic approximation, and silicon data match it to 4.8 percent.
desk verdict Useful all-order harmonic TDS method with a clean silicon validation; the finite-grid folding caveat is real but likely fixable, so send it to review. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the three-dimensional difference pair distribution function (3D-ΔPDF), assembled from the phonon-derived reciprocal-space kernel $\gamma_{\kappa\alpha,\kappa'\beta}(\mathbf{q})=(M_\kappa M_{\kappa'})^{-1/2}\sum_\nu A_\nu(\mathbf{q})e^*_{\kappa\alpha,\nu}(\mathbf{q})e_{\kappa'\beta,\nu}(\mathbf{q})$ with thermal amplitude $A_\nu=(2\langle n_\nu\rangle+1)/\omega_\nu$. An inverse DFT on the same $N\times N\times N$ q-grid as the intensity sampling converts this kernel to the real-space covariance $C_{\kappa\kappa'}(R)$, and for every atom pair the assembly lays down two centred Gaussians on the interatomic vector $R+x_\kappa-x_{\kappa'}$: one with the correlated relative-displacement covariance $\Sigma^R_{\kappa\kappa'}=U_\kappa+U_{\kappa'}-2C_{\kappa\kappa'}(R)$, one with $U_\kappa+U_{\kappa'}$. Their difference is that pair's ΔPDF contribution, and a single forward FFT of the summed volume, with the form-factor product applied on transform, yields the all-order diffuse map. The reason the method is exact in the harmonic approximation is that the exponential of the covariance is kept whole, so the assembly never chooses a phonon order; the FFT pair shares one grid, so no interpolation is needed.
What would settle it
Compute the same silicon diffuse map by directly summing the real-space covariance contributions over lattice vectors out to a large cutoff and compare with the single-FFT map on, say, a 30×30×30 grid; if the two disagree near the Bragg peaks by an amount comparable to the 4.8% residual, the folding assumption fails.
Extended reading notes
Core claim
The central claim is that all multiphonon contributions to X-ray thermal diffuse scattering are encoded in the real-space displacement covariance $C_{\kappa\kappa'}(R)=\langle u_{0\kappa}u_{R\kappa'}^T\rangle$, so the diffuse intensity is exactly (inside the harmonic-Gaussian description) the Fourier transform of a 3D-ΔPDF: $I_{\mathrm{diff}}(\mathbf{h})=N\mathcal{F}[\Delta\mathrm{PDF}(\mathbf{r})](\mathbf{h})$. Each atom pair contributes two Gaussians on its interatomic vector, one with the correlated covariance $\Sigma^R_{\kappa\kappa'}=U_\kappa+U_{\kappa'}-2C_{\kappa\kappa'}(R)$ and one with the independent-motion covariance $U_\kappa+U_{\kappa'}$; their difference is the diffuse signal. The intensity formula keeps the exponential $\exp(4\pi^2\mathbf{h}^TC_{\kappa\kappa'}(R)\mathbf{h})$ unexpanded, so no phonon order is dropped and no supercell averaging noise enters. Tests against measured silicon data support the claim: with first-principles phonons used unadjusted and only scale and background refined, the room-temperature diffuse intensity is reproduced to $R^2\approx4.8\%$, and a four-temperature series reproduces the thermal growth with one scale to $R^2\approx9\%$.
Load-bearing premise
The load-bearing premise is that periodically wrapping the slowly decaying acoustic displacement correlations into the finite real-space box, instead of summing them directly, leaves an error below the reported residual; the paper argues this but does not quantify it against a direct sum.
Editorial extensions
If this is right
- The standard one-phonon and two-phonon truncation of TDS is avoided altogether, since every multiphonon order is carried by the unexpanded covariance exponential.
- A full diffuse map over large reciprocal-space volumes comes from one fast Fourier transform, so the cost is governed by the FFT grid; the silicon benchmark finishes in under ten minutes.
- Because thermal and static correlated disorder are expressed in the same pair-correlation representation, crystals that show both can be modeled and ultimately refined together.
- Every step in the construction is differentiable, which opens the way to gradient-based refinement of phonon or disorder parameters against measured diffuse volumes.
- The output is a real-space 3D-ΔPDF, so the calculation yields interpretable per-pair displacement correlations, not just reciprocal-space intensities.
Reading between the lines
- If the periodic-folding error is as small as argued, the pipeline should transfer to anisotropic and lower-symmetry crystals with cost set mainly by the output grid; testing such a crystal would confirm the approach is not silicon-specific.
- The four-temperature series with one scale factor suggests that measured TDS at several temperatures could serve as a sensitive benchmark for phonon frequencies and anharmonic shifts, since the model alone must reproduce the thermal growth; the paper does not explore this use.
- Because the calculation is differentiable and noise-free, it could serve as a forward model in reverse Monte Carlo or machine-learning fits of disorder, where expensive noisy supercell averages are currently the bottleneck.
- A useful benchmark would be comparing this all-order result against one- and two-phonon approximations to map where truncation errors appear in scattering vector and temperature; the paper does not include such a map.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper introduces a method for computing X-ray thermal diffuse scattering (TDS) from real-space displacement correlations using the three-dimensional difference pair distribution function (3D-ΔPDF). Within the harmonic approximation, the derivation expresses the full all-phonon-order diffuse intensity as a Fourier transform of a real-space ΔPDF volume built from atom-pair covariances. The implementation obtains displacement covariances from first-principles phonons via an inverse FFT and then assembles the ΔPDF from Gaussian atom-pair contributions; a single forward FFT yields the diffuse map. The method is validated against room-temperature silicon data measured at ID28 (R2≈4.8%, with only a scale and smooth background refined) and against a four-temperature silicon series from SNBL (R2≈9%). The paper further argues that the shared pair-correlation language with the Yell program allows joint treatment of thermal and static disorder.
Significance. If the claims hold, this is a genuinely useful advance: it gives an all-phonon-order TDS calculation at the cost of one FFT, avoids the truncation and sampling-noise problems of Laval-Born-James and special-displacement approaches, and is differentiable, opening the door to gradient-based refinement. The derivation is clearly presented, the code is available, and the validation uses an external experimental benchmark with no fitted physical parameters—only a scale and a background. The main reservation is that the numerical implementation's finite-grid folding approximation is acknowledged but not quantitatively assessed, and the reported residual is computed after masking the region where that approximation is expected to be largest.
major comments (2)
- [4, Figure 3] The main text reports R²≈4.8% but defers the definition of R² and the punch-and-fill procedure to the Supplementary Information. Since this residual is the central quantitative validation, the main text should state exactly which voxels are included, how the Bragg region is filled, and how the R² is computed, so the reader can assess whether the near-Bragg masking also removes the region where the model error is largest. This is particularly important because the folding-error concern of Section 2.6 is dominated by that same region.
- [Abstract and Section 2.6] The abstract states that 'within the harmonic approximation the method is exact', which is true for the analytic equations derived in Section 2.3 but not for the implemented finite-grid pipeline, whose folding error is acknowledged but not shown to be negligible. I recommend either adding the quantitative convergence evidence requested above or qualifying the 'exact' claim to the infinite-lattice expressions, with the finite-grid implementation presented as a controlled approximation.
minor comments (5)
- [Eq. (16) and Section 3.1] The q=0 acoustic modes make Eq. (16) singular; the text states they are regularised to zero but does not spell out the limiting argument in the main body. A parenthetical explaining the rigid-translation cancellation, or a reference to the Supplementary Information at that point, would help.
- [Section 3.2] '6013 grid' should read '601³ grid'.
- [Figure 3 caption] The bottom-right residual panel uses a diverging blue/white/red map, but the caption does not explain the colour mapping beyond the range; a one-sentence explanation would improve readability.
- [Section 3.2] The tool name 'anaddb' should be typeset as 'Anaddb' or 'ABINIT's anaddb' for consistency with standard nomenclature.
- [Data availability] The statement that the ETH repository data will be available 'after short review' is vague; a persistent identifier or a release date would be more useful to readers and reviewers.
Circularity Check
No significant circularity: derivation is self-contained, phonon covariances come from independent DFPT, and the silicon benchmark is an external experimental validation.
full rationale
The derivation chain is self-contained. Equations (8)-(15) follow from the standard harmonic Gaussian average, and Eq. (16) brings in phonon covariances computed independently by DFPT with Abinit. The 3D-dPDF identity, Eqs. (20)-(21), is a definitional rearrangement of Eq. (15), not a fitted input: the covariance enters the real-space Gaussians from first principles, and no physical parameter of the model is refined. The silicon validation uses external experimental data measured at ID28/ESRF and SNBL/BM01; the only fitted quantities are an overall scale and a smooth polynomial background, which do not enter the physical derivation. The self-citations to Yell, Meerkat, and the 3D-dPDF literature are implementation context and prior published software, not load-bearing support for the intensity formula. Section 2.6 explicitly flags the periodic-folding approximation of acoustic correlations and admits that folding is 'not free of error'; this is an acknowledged numerical approximation and a validation concern, not a circular step, and it does not reduce the central claim to its inputs. No circularity score above 0 is warranted.
Assumptions & free parameters
free parameters (3)
- global intensity scale s =
not stated numerically
- smooth isotropic background =
not stated numerically
- Bragg punch radius r_p =
r_p = 3 (ID28), r_p = 4 (SNBL)
assumptions (4)
- standard math Atomic displacements are harmonic and Gaussian-distributed, so the thermal average of the phase factor reduces to exp(-2π² ⟨(h·Δu)²⟩).
- domain assumption Displacement covariances C_kappa,kappa'(R,T) are obtained from DFPT phonons via Eq. (16), taken from Xu (2010), and the GGA-PBE force constants are accurate enough without adjustment.
- ad hoc to paper Acoustic displacement correlations decay slowly, but folding them periodically into the N×N×N supercell introduces errors that are negligible relative to the target residual.
- standard math Translational invariance allows the double cell sum to be collapsed to a single sum over lattice vectors R.
Cite this review
Pith. "Pith review of X-ray Thermal diffuse scattering from real-space displacement correlations." pith.science (2026). https://pith.science/paper/H6MB7HHU
@misc{pith2026260805051,
author = {Pith},
title = {Pith review of: X-ray Thermal diffuse scattering from real-space displacement correlations},
year = {2026},
howpublished = {\url{https://pith.science/paper/H6MB7HHU}},
note = {Machine review of arXiv:2608.05051}
}
read the original abstract
We introduce a method for calculating X-ray thermal diffuse scattering based on the three-dimensional difference pair distribution function (3D-dPDF). Within the harmonic approximation the method is exact, as it includes all orders of multi-phonon scattering. A single Fourier transform delivers the diffuse intensity over large volumes of reciprocal space. We tested the method against experimental diffuse scattering measured on a silicon single crystal. With nothing refined beyond a scale and a background, the calculation reproduces the measured intensity to an R2 residual below 5%. Because the method uses the same pair-correlation language as the established Yell program, thermal and static correlated disorder enter the analysis on equal footing.
Figures
Figures from the paper (1 more)
Reference graph
Works this paper leans on
-
[1]
Agarwal, R. C. (1978).Acta Crystallogr. A,34(5), 791–
work page 1978
-
[492]
Virtanen, P., Gommers, R., Oliphant, T. E.et al.(2020). Nat Methods,17(3), 261–272. Wall, M. E., Van Benschoten, A. H., Sauter, N. K., Adams, P. D., Fraser, J. S. & Terwilliger, T. C. (2014).Proceedings of the National Academy of Sci- ences,111(50), 17887–17892. Warren, B. E. (1969).X-ray Diffraction. Reading, MA: Addison-Wesley. Weber, T. & Simonov, A. (...
work page 2020
-
[809]
(1942).Reports on Progress in Physics,9(1), 294–333
Born, M. (1942).Reports on Progress in Physics,9(1), 294–333. Bosak, A. & Chernyshov, D. (2008).Acta Crystallogr A Found Crystallogr,64(5), 598–600. Bulled, J., Fahl, B. & Simonov, A. (2026). Domain for- mation in BaTiX 3 from single crystal diffuse scatter- ing. Dataset; experiment session A01-2-1379, beam- line BM01 (SNBL). https://doi.org/10.15151/ESRF...
-
[1271]
Mirone, A. & Wehinger, B. (2013). ab2tds, version 1.1. https://gitlab.esrf.fr/mirone/ab2tds Pan, Y., Hildebrandt, P.-N., Zahn, D., Zacharias, M., Windsor, Y. W., Ernstorfer, R., Caruso, F. & Seiler, H. (2025).ACS nano,19(11), 11381–11389. Sangiorgio, B., Bozin, E. S., Malliakas, C. D., Fechner, M., Simonov, A., Kanatzidis, M. G., Billinge, S. J., Spaldin,...
work page 2013
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.