X-ray Structure Function (SF)

• X-ray structure function (SF) is a statistical measure used to characterize temporal variability in X-ray sources and reveal structural correlations in scattering experiments.
• In time-domain astronomy, SF quantifies lag-dependent flux or count-rate differences, providing insights into AGN variability and flare dynamics such as those seen in Sgr A*.
• In total scattering and XFEL imaging, SF operates in reciprocal space to derive pair distribution functions and intensity models, enabling reconstruction of real-space structures and electron densities.

Searching arXiv for recent and relevant papers on X-ray structure function usages across astronomy, total scattering, and XFEL imaging. arxiv_search(query="X-ray structure function SF total scattering variability XFEL", max_results=10, sort_by="relevance") arxiv_search(query="X-ray structure function SF total scattering variability XFEL", max_results=10) In the literature considered here, the term X-ray structure function (SF) is used in distinct ways across subfields. In time-domain X-ray astronomy, it denotes a lag-dependent statistic of flux or count-rate variability. In total scattering and pair distribution function analysis, it denotes the reciprocal-space quantity $S(Q)$, or the reduced function $F(Q)=Q[S(Q)-1]$, from which real-space pair correlations are obtained by Fourier transform. In single-particle XFEL imaging, it denotes the reciprocal-space intensity model $w(i)$ inferred from sparse diffraction frames. In crystallography, the closely related structure factor $F_{hkl}$ or $F(\mathbf{r}^*)$ provides the Fourier coefficients of the electron density (Vagnetti et al., 2016, Kovyakh et al., 2021, Sapnik et al., 30 Apr 2025, Philipp et al., 2012, Bortel et al., 2023, Batke et al., 2015).

1. Core definitions

For X-ray variability studies, the SF is a second-order statistic of temporal differences. One normalized form is

$$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$

where $R(t)$ is the count rate and $\tau$ is the time lag (Li et al., 2015). Ensemble AGN work also uses logarithmic-flux forms such as

$$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$

with rest-frame lags $\tau_{\rm rest}=\tau_{\rm obs}/(1+z)$, and the noise term defined from photometric errors (Vagnetti et al., 2016, Vagnetti et al., 2011). A closely related formulation is

$F(Q)=Q[S(Q)-1]$0

used for ensemble quasar variability over months-to-decades baselines (Prokhorenko et al., 2024).

For total scattering, the X-ray structure function is the reduced reciprocal-space signal

$F(Q)=Q[S(Q)-1]$1

derived from corrected and normalized scattering data. Its Fourier transform gives the pair distribution function

$F(Q)=Q[S(Q)-1]$2

This definition underlies both conventional isotropic PDF analysis and recent spatially resolved or ultrafast XFEL implementations (Kovyakh et al., 2021, Sapnik et al., 30 Apr 2025).

For anisotropic scattering, the deviation structure function $F(Q)=Q[S(Q)-1]$3 may be expanded in spherical harmonics,

$F(Q)=Q[S(Q)-1]$4

with the real-space pair density expanded analogously (Zhang et al., 2022).

In single-particle XFEL imaging, the SF is the model intensity $F(Q)=Q[S(Q)-1]$5 in reciprocal space, with low-flux photon counts treated as Poisson samples from randomly rotated copies of that model (Philipp et al., 2012). In crystallography, the corresponding reciprocal-space coefficients are written as

$F(Q)=Q[S(Q)-1]$6

or

$F(Q)=Q[S(Q)-1]$7

linking diffraction to atomic positions or electron density (Bortel et al., 2023, Batke et al., 2015).

2. Time-domain variability in X-ray astronomy

The SF is widely used for ensemble X-ray variability because it operates in the time domain and remains usable when individual light curves are sparsely sampled. In serendipitous AGN samples from XMM-Newton and Swift, the ensemble SF follows an approximate power law $F(Q)=Q[S(Q)-1]$8, shows a strong anti-correlation with X-ray luminosity, and does not show evidence of a break in high-luminosity AGN (Vagnetti et al., 2011). In the MEXSAS analysis, the updated SF scales as $F(Q)=Q[S(Q)-1]$9, its normalization varies approximately as $w(i)$0, and its frequency dependence follows $w(i)$1, corresponding to a softer when brighter trend (Vagnetti et al., 2016). In a ROSAT–XMM quasar sample extending to $w(i)$2 years rest-frame, the SF increases monotonically with lag, is fitted by $w(i)$3 with $w(i)$4, and shows no evidence for a plateau at the longest sampled lags (Middei et al., 2016). In the SRG/eROSITA quasar study, $w(i)$5 increases with time lag, is anti-correlated with luminosity, grows with decreasing Eddington ratio, and is larger for less massive black holes at fixed Eddington ratio and timescale; most subsamples have slopes between $w(i)$6 and $w(i)$7, with one low-$w(i)$8, low-$w(i)$9 subsample reaching $F_{hkl}$0 (Prokhorenko et al., 2024).

For Sagittarius A$F_{hkl}$1, the SF was fitted jointly with the count-rate distribution using Monte Carlo simulations and MCMC. The $F_{hkl}$2 keV light curve was decomposed into a quiescent component with a constant count rate of $F_{hkl}$3 count s$F_{hkl}$4 and a flare component with a power-law fluence distribution $F_{hkl}$5 with $F_{hkl}$6. The duration–fluence correlation was modeled as $F_{hkl}$7 with $F_{hkl}$8 at $F_{hkl}$9 confidence. The observed SF rises with $F(\mathbf{r}^*)$0 and flattens above $F(\mathbf{r}^*)$1 s, which was interpreted as the largest event duration in the flaring process; these statistics were found consistent with a self-organized criticality system with spatial dimension $F(\mathbf{r}^*)$2 (Li et al., 2015).

In long-term monitoring of the M31 center, source-by-source $F(\mathbf{r}^*)$3 keV SFs were compared with the ensemble AGN SF

$F(\mathbf{r}^*)$4

and with its $F(\mathbf{r}^*)$5 upper limit $F(\mathbf{r}^*)$6. Sources with Rank $F(\mathbf{r}^*)$7 were taken as probable X-ray binaries, corresponding to a $F(\mathbf{r}^*)$8 excess above the AGN SF; Rank $F(\mathbf{r}^*)$9 corresponds to $$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$0. Using this method, 220 X-ray sources with luminosities $$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$1 erg s$$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$2 were found to have SFs with significantly more variability than the ensemble AGN SF, and were likely X-ray binaries (Barnard et al., 2013).

3. Statistical estimation, likelihoods, and corrections

A central feature of SF work is explicit noise subtraction. In AGN ensemble studies, the observed SF contains intrinsic variability plus photometric noise, and the noise term is subtracted under the square root after averaging over all observation pairs in each lag bin (Vagnetti et al., 2011, Vagnetti et al., 2016). In the eROSITA quasar analysis, each pairwise estimator is

$$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$3

with

$$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$4

and bootstrap methods are used to estimate the uncertainty in each $$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$5 bin (Prokhorenko et al., 2024).

The Sgr A$$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$6 study combines SF fitting with the count-rate distribution in a forward-modeling framework. For each parameter set, Monte Carlo simulations generate synthetic light curves; the SF likelihood is written as

$$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$7

where $$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$8 is the covariance matrix derived from simulations. The total likelihood is $$V(\tau) = \frac{\left\langle [R(t+\tau)-R(t)]^2 \right\rangle}{S^2}, \qquad S^2=\left\langle [R(t)-\overline{R(t)}]^2 \right\rangle,$$9, and the Metropolis-Hastings algorithm implemented through CosmoMC is used to explore the posterior (Li et al., 2015).

The AGN variability literature also uses SF results to diagnose biases in other estimators. The MEXSAS analysis states that an improper use of the normalised excess variance may lead to an underestimate of the intrinsic variability. Using the empirical SF scaling $R(t)$0, it proposes a duration correction

$R(t)$1

to compare light curves standardized to a common rest-frame duration, and introduces a V-correction based on the observed dependence $R(t)$2 when converting observer-frame variability to rest-frame bands (Vagnetti et al., 2016). This methodological emphasis reinforces the earlier conclusion that SF is more appropriate than fractional variability and excess variance for sparse ensemble studies affected by cosmological time dilation (Vagnetti et al., 2011).

4. Reciprocal-space SF in total scattering and PDF analysis

In total scattering, the X-ray SF is the intermediate reciprocal-space quantity between corrected intensity and real-space pair correlations. In a scanning nanostructure X-ray microscopy workflow, 2D detector images are azimuthally integrated to obtain $R(t)$3, standard corrections are applied, and the structure function is computed as

$R(t)$4

The subsequent Fourier transform yields the PDF,

$R(t)$5

This workflow was implemented with semi-automated data reduction, normalization, modeling, and metadata-linked spatial slicing, allowing maps of quantities such as lattice parameter, crystallite size, or fit agreement factor across thin-film arrays (Kovyakh et al., 2021).

When the sample response is anisotropic, angular integration is no longer valid. The spherical harmonics method expands both $R(t)$6 and $R(t)$7 in angular basis functions and relates the coefficients through spherical Bessel transforms,

$R(t)$8

For uniaxial fields, symmetry permits truncation to $R(t)$9 and often to $\tau$0, so that $\tau$1 recovers the isotropic PDF while higher terms encode directional structural response. Applied to PbZr$\tau$2Ti$\tau$3O$\tau$4 ceramics under a 5 kV/mm electric field, the method resolved bond-length changes, domain switching, and lattice distortion in direction-dependent PDFs (Zhang et al., 2022).

XFEL-based total scattering extends this reciprocal-space usage to femtosecond timescales. At the HED instrument of the European XFEL, normalised total scattering data were reported for $\tau$5 from crystalline, nanocrystalline, amorphous, liquid, and solution samples, and high-quality data were obtained from a single approximately 30 fs XFEL pulse. The analysis pipeline included dark-image subtraction, detector-response and flat-field correction, azimuthal integration with pyFAI, beam-intensity normalization, detector merging across an overlap region, background subtraction, and downstream analysis by Rietveld refinement, small-box PDF refinement, joint reciprocal-real space refinement, cluster refinement, and Debye scattering analysis (Sapnik et al., 30 Apr 2025).

5. XFEL structure recovery from sparse and single-pulse data

In single-particle imaging at XFELs, the SF is the reciprocal-space intensity model $\tau$6 that would be observed from an aligned, noise-free particle. The measurements consist of sparse diffraction frames from randomly oriented particles, with photon counts modeled as independent Poisson samples,

$\tau$7

In the low-flux limit, the likelihood of frame $\tau$8 at rotation $\tau$9 reduces to

$$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$0

where $$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$1 is the set of photon-hit pixels. The expectation-maximization algorithm alternates an E-step computing the posterior over rotations and an M-step updating the model via

$$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$2

Using this aggregate processing, reconstruction was demonstrated at an average of only 2.5 photons per frame, without any prior knowledge of the object or its orientation. The same work introduced an information rate ratio $$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$3 from mutual information to quantify the impact of background photons on reconstructability, and argued that model-based likelihood assignment outperforms correlation-based methods in the ultra-sparse regime (Philipp et al., 2012).

A different XFEL route to structure uses Kossel lines from crystals. In a single 25 fs XFEL pulse, hundreds of Bragg reflections were measured in parallel from GaAs and GaP through Kossel line patterns, whose profiles encode both the amplitude and phase of the corresponding structure factors. After line indexing, geometry refinement, and profile fitting based on dynamical diffraction theory, the extracted complex structure factors were inserted into the inverse Fourier sum

$$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$4

yielding the 3D electron density without indirect phase retrieval (Bortel et al., 2023).

6. Structure factors, relativistic effects, and nomenclature

In crystallographic charge-density work, the structure factor is the Fourier transform of the electron density, and heavy-atom systems can require explicit relativistic treatment. For the model systems $$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$5 with $$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$6, four-component molecular wave functions were used to compute X-ray structure factors, and the overall effects of relativity on the structure factors were found to average 0.47, 0.80, and 1.27% for Ni, Pd, and Pt, respectively. For individual reflections or reflection series, the effects can be orders of magnitude larger. The DKH2 and ZORA Hamiltonians account for these effects to a large extent, reducing the differences by one order of magnitude (Batke et al., 2015).

The literature therefore supports a strict distinction between structure function and structure factor, even though both are reciprocal-space or statistical descriptors derived from X-ray data. In astronomy, the SF measures temporal variability. In total scattering, it is the normalized reciprocal-space function entering the PDF transform. In crystallography, the primary object is usually the structure factor. A further source of ambiguity is acronym collision: in the detector paper on low-pressure SF$$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$7, the term “structure function (SF)” is not explicitly mentioned as a formal quantity or with a formula; there, “SF” denotes the gas species rather than an X-ray statistical observable (McLean et al., 2023).

Different source classes also exhibit different long-lag SF behavior. Sgr A$$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$8 flares show a plateau at $$S(\tau)\equiv \sqrt{\left\langle [\log f_X(t+\tau)-\log f_X(t)]^2\right\rangle-\sigma_{\rm noise}^2},$$9 s, interpreted as the largest flare duration in that process, whereas the long-term AGN soft-X-ray ensemble study found no evidence for a plateau up to $\tau_{\rm rest}=\tau_{\rm obs}/(1+z)$0 years rest-frame (Li et al., 2015, Middei et al., 2016). This suggests that no single asymptotic SF morphology is universal across X-ray applications.