betaFeedback

WorksPhysics and Astronomy

An Absorption Correction for Reliable Pair-Distribution Functions from Low Energy X-ray Sources

Abstract

This paper explores the development and testing of a simple absorption correction model for processing x-ray powder diffraction data from Debye-Scherrer geometry laboratory x-ray experiments. This may be used as a pre-processing step before using PDFgetX3 to obtain reliable pair distribution functions (PDFs). The correction was found to depend only on muD, the product of the x-ray attenuation coefficient and capillary diameter. Various experimental and theoretical methods for estimating muD were explored, and the most appropriate muD values for correction were identified for different capillary diameters and x-ray beam sizes. We identify operational ranges of muD where reasonable signal to noise is possible after correction. A user-friendly software package, diffpy.labpdfproc, is presented that can help estimate muD and perform absorption corrections, with a rapid calculation for efficient processing.

package, DIFFPY.LABPDFPROC, is presented that can help estimate μD\mu D and perform absorption corrections, with a rapid calculation for efficient processing.

Introduction

Historically, the study of local atomic structures using x-ray pair distribution function (PDF) analysis has predominantly been done on data from synchrotron facilities using rapid acquisition PDF (RAPDF) techniques.[1] However, using laboratory-based x-ray sources to get good PDFs is of great interest. This has been demonstrated from laboratory diffractometers equipped with Mo and Ag KαK_{\alpha} sources in various examples including crystalline materials[2] and nanoparticles.[3]

The current authors recently tested protocols for optimizing experimental setups for such measurements.[4] That work discussed that some factors affecting the data reduction to obtain quantitatively accurate PDFs that are often neglected in RAPDF experiments become more relevant for lab-based experiments, especially for those done with Mo KαK_{\alpha} radiation. In particular, multiplicative corrections to the measured intensities from sample absorption are likely to have a non-negligible QQ-dependence for Mo KαK_{\alpha} data but not for RAPDF data.[4, 5] This is for two reasons. First, the lower x-ray energies result in overall higher sample absorption effects. Second, in lab experiments it is necessary to measure over a large range of 2θ2\theta, typically up to 140° or higher with Mo or Ag x-rays, making the intensities more susceptible to any angle-dependent multiplicative corrections.

The influence of sample absorption has been discussed for powder diffraction in general. Absorption leads to differences in calculated and measured powder diffraction patterns and has to be accounted for in Rietveld refinements.[6] In Bragg-Brentano geometries finite sample thickness effects must be considered when measuring low absorbing specimens.[5, 7] Absorption effects are also crucial in successful quantitative XRD analysis by multiphase refinements.[8, 9, 10] Here, we present a detailed investigation of the influence of sample absorption in bench-top PDF analysis, i.e., for lab diffractometers with capillary (Debye-Scherrer) geometry. We present a software package for making sample-absorption corrections to measured data for this geometry.

Among the numerous available software packages to obtain G(r)G(r) from raw data, [11, 12, 13, 14] PDFgetX3[15] is widely used by the community as it follows a simple ad hoc approach to data reduction.[16] The ad hoc algorithm used in PDFgetX3 does a good job of correcting for parasitic scattering (unwanted additive contributions to the signal) but makes no correction for multiplicative effects. This works well for RAPDF data where they are small and their angle dependence is even smaller. However as discussed above, when applied to data from a laboratory diffractometer this simplifying assumption may not be valid in general, which motivated this study.

This work also allows us to suggest best practices for sample preparation for PDF experiments on laboratory x-ray diffractometers to mitigate the worst effects of sample absorption. These strategies are not new, but here we validate and quantify them somewhat to give more precise guidance.

Absorption correction

Here, we address the multiplicative correction due to sample self-absorption effects for the most common geometry used for PDF analysis, the Debye-Scherrer geometry, in which a beam is incident on a cylindrical sample in a capillary with a detector rotating around the sample. The absorption correction for this geometry was developed in detail many years ago[17, 18, 19, 20] and is routinely applied in neutron scattering measurements. Various codes implement these corrections as part of the total scattering data reduction workflow.[13, 21, 22] They require rather detailed knowledge of the sample composition and packing fraction, and for the greatest accuracy, also the sample and sample-container geometry.

In this work, we are more concerned with obtaining sufficient accuracy in these corrections rections in combination with ad hoc data reduction approaches used as PDFGETX3, and understanding the effects of approximate corrections on resulting PDFs. We explore this for a range of possible sample compositions with Mo KαK_{\alpha} and Ag KαK_{\alpha} radiation. We first briefly describe a simplified derivation of the angle dependence of the absorption in this geometry and then explore the nature of the resulting curves in different situations. The full derivation is in the Supplementary Information, but is summarized here as it can help to build intuition about absorption effects in a laboratory x-ray setting.

Assuming a homogeneous ideal powder in the absence of sample absorption the measured x-ray intensity at some point in QQ, Im(Q)I_m(Q), would be proportional to the illuminated sample volume, VV. Here QQ is the magnitude of the scattering vector, Q=∣ki−ks∣=4πsin⁡θλQ = \lvert\mathbf{k}_i - \mathbf{k}_s \rvert= \frac{4\pi\sin\theta}{\lambda} where ki\mathbf{k}_i and ks\mathbf{k}_s are the incident and scattered wave vectors, respectively, θ\theta is the Bragg angle which is half the scattering angle, 2θ2\theta, which is the angle between the incident and scattered x-ray beams (see Figure 1) and λ\lambda is the x-ray wavelength. A normalized coherent scattering intensity per unit volume of sample, ici_c, can therefore be obtained by dividing ImI_m by VV.

Incoming and outgoing x-ray beam paths for x-rays undergoing scattering in two representative pixels

Figure 1. Incoming and outgoing x-ray beam paths for x-rays undergoing scattering in two representative pixels at an angle (a) 2θ=40∘2\theta= 40^\circ and (b) 2θ=135∘2\theta= 135^\circ. In each case, the large black circle indicates the edge of a cross-section of the cylindrical sample capillary. The x-rays arrive horizontally from the left. A scattered ray is then shown at the given 2θ2\theta angle. The small blue-grey dots indicate positions of a uniform grid of voxels in the circle and the red and yellow dots indicate the point on the surface of the sample where the x-rays enter and exit, respectively. The path-length for the x-ray scattered in the vvth voxel at angle 2θ2\theta is then ℓv(2θ)=ℓvi+ℓvo(2θ)\ell_v(2\theta) = \ell_v^i + \ell_v^o(2\theta), the sum of the lengths of the red line and the yellow lines.

In the presence of sample absorption, the scattered intensity from any small volume element (voxel) in the sample will be reduced by absorption of the x-ray as it travels along the incoming, ℓvi\ell_v^i, and outgoing, ℓvo\ell_v^o, path through the sample. This is shown in Figure 1 for the case of a cylindrical capillary mounted perpendicular to the beam. The derivation of the path lengths is reproduced in detail in Section 1 in the Supplementary Information. When there is significant sample absorption, to get ici_c we would divide ImI_m not by the full illuminated volume but by an absorption corrected effective volume, VeV_e, so that

ic(2θ)=Im(2θ)Ve(2θ),(1)i_c(2\theta) = \frac{I_m(2\theta)}{V_e(2\theta)}, \tag*{(1)}

where the effective volume is given by,1

Ve(2θ)=∑ve−μsℓv(2θ)ΔVv,(2)V_e(2\theta) = \sum_v e^{-\mu_s \ell_v(2\theta)} \Delta V_v, \tag*{(2)}

where ΔVv\Delta V_v is the volume of the vvth voxel and μsλ\mu_s^\lambda is the linear absorption coefficient of the sample for x-rays of wavelength λ\lambda. ℓv(2θ)\ell_v(2\theta) is the total path-length, i.e., the sum of the incident and outgoing path lengths (see Figure 1),

ℓv(2θ)=ℓvi+ℓvo(2θ),(3)\ell_v(2\theta) = \ell_v^i + \ell_v^o(2\theta), \tag*{(3)}

for the vvth voxel and scattering angle 2θ2\theta. The normalized intensity from Eq. 1 is then

ic=Im∑ve−μsλℓvΔVv.(4)i_c = \frac{I_m}{\sum_v e^{-\mu_s^\lambda\ell_v}\Delta V_v}. \tag*{(4)}

For the case where all the voxels have the same volume ΔVv=V/Nv\Delta V_v = V/N_v and we get

ic=NvImV∑ve−μsλℓv(2θ).(5)i_c = \frac{N_v I_m}{V\sum_v e^{-\mu_s^\lambda\ell_v(2\theta)}}. \tag*{(5)}

We can then define an absorption correction cvec_{ve} as

cve(2θ)=Nν∑νe−μsλℓν(2θ).(6)c_{ve}(2\theta)=\frac{N_\nu}{\sum_\nu e^{-\mu_s^\lambda\ell_\nu(2\theta)}}. \tag*{(6)}

Both the strength of the x-ray attenuation, and its angle dependence, are dependent on the material and wavelength specific linear absorption coefficient, μsλ\mu_s^\lambda. It also depends on a sum over all the path lengths an x-ray takes through the sample as it arrives from the source and exits after scattering.

For simpler flat geometries it is known that the curve shape depends only on the product μt\mu t, where tt is the thickness of the sample, and not independently on μ\mu and tt. We show here that this is also true for the capillary geometry: the curve depends only on the product μD\mu D where DD is the capillary diameter. This greatly simplifies the analysis by reducing the dimensionality of the space of possibilities we need to consider. Details of the proof are provided in Section 2 in the Supplementary Information. There we show that μD\mu D can be factored out of the sum, ∑νe−μsλℓν(2θ)\sum_\nu e^{-\mu_s^\lambda\ell_\nu(2\theta)} in Eq. 7 which can be rewritten as

cve(2θ)=Nν∑νe−μsλD⋅dℓν(2θ).(7)c_{ve}(2\theta)=\frac{N_\nu}{\sum_\nu e^{-\mu_s^\lambda D\cdot d\ell_\nu(2\theta)}}. \tag*{(7)}

where dℓν(2θ)d\ell_\nu(2\theta) is the path length that the x-ray would traverse for a capillary of unit diameter. Thus, the absorption correction depends not on μ\mu and DD independently but on the product μD\mu D (where for compactness, we drop hereafter the superscript and subscript that explicitly indicate that it is μD\mu D for the sample at a particular wavelength).

We explore below the effect on the PDF of applying this correction to data from a variety of samples of different absorptions, discuss different ways of estimating μD\mu D for a given sample, and describe a software package for rapidly obtaining and applying the correction.

Validation Experiments

To explore the effects of the correction on real data, powder x-ray diffraction patterns were collected for four samples with varying absorption cross-sections. To cover a large range of μD\mu D, ZrO2_2, CeO2_2, and HfO2_2 were packed in Kapton (polyimide tubes) with varying inner diameters (ID’s). To get a sense of how the μD\mu D’s are distributed, the theoretical values were computed based on each sample composition, mass density, and capillary diameter using the XrayDB database. [23] The mass density was determined by measuring the mass of the packed powder and the length of the powder bed. The sample list, along with the corresponding information, are presented in Table 1, ranging from μD=0.8\mu D = 0.8 to almost 12. For reference, a subset of the samples (ID=1 mm) were also measured using synchrotron x-rays.

sampleID (mm)density (g/cm3g/cm^3)μD\mu D
ZrO2_26351.0090.795
8130.8560.864
10241.1221.426
CeO2_26351.7064.229
8131.4354.554
10241.4575.824
HfO2_26351.7418.168
8131.96311.79
Wire2––

Table 1. List of samples that were measured on the Bruker lab diffractometers. The samples were packed in Kapton (polyimide) tubes with different inner diameters (ID). The density was calculated based on the amount of powder packed in a given segment of the cylindrical Kapton tubes. μD\mu D’s were calculated using the database from XrayDB.

The laboratory PDF measurements were performed on a Bruker D8 Discovery diffractometer equipped with a Mo KαK_{\alpha} source (Kα1α2K_{\alpha1\alpha2} double emission, average wavelength λ\lambda=0.71073 Å) using a capillary geometry to ensure a constant sample illumination. The configuration included a focusing Goebel mirror, a divergence slit of 1.0 mm for IDs 0.635 mm and 0.813 mm, and 1.2 mm for IDs 1.024 mm, a 2.5° axial Soller slit, a scattering guard after the source, and an additional 2.5° axial Soller slit in the diffraction beam before the detector. X-ray generator settings of 50 kV and 50 mA were employed. The acquisitions were conducted using the staircase-counting-time (SCT) measurement strategy described in our previous work that ensures increased counting statistics in the high-QQ region.[4]

The SCT acquisition protocol consisted of 5 scans with a constant step size of 0.025°, decreasing 2θ2\theta-range and increasing counting time per step as shown in Table 2.

2θ\theta start2θ\theta stopcounting time (sec)
scan 121401.8
scan 2751403.6
scan 31071407.2
scan 412414014.4
scan 513214028.8

Table 2. Staircase-counting-time (SCT) protocol for the lab PDF measurements on the Bruker D8 Discover diffractometer.

Synchrotron total scattering measurements were conducted at ID31 at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France. The sample powders were loaded into cylindrical slots (1 mm thickness) held between Kapton windows in a high-throughput sample holder. Each sample was measured in transmission with an incident X-ray energy of 75.00 keV (λ=0.1653\lambda= 0.1653 Å). A Pilatus CdTe 2M detector (1679 ×\times 1475 pixels, 172 ×\times 172 μ\mum each) was positioned with the incident beam in the corner of the detector. The sample-to-detector distance was approximately 0.3 m. Background measurements for the empty windows were measured and subtracted. NIST SRM 660b (LaB6) was used for geometry calibration performed with the software pyFAI[24] followed by image integration, including a flat-field, geometry, solid-angle, and polarization corrections. Data processing to obtain the PDF after any pre-processing to correct for absorption was done using PDFgetX3.[15] The Fourier transform from F(Q)F(Q) to G(r)G(r) was done with the parameters Qmax⁡=16.6Q_{\max} = 16.6 Å, Qmin⁡=1.0Q_{\min}=1.0 Å, and rpolyr_{\mathrm{poly}} in the range 1.2–1.5. A Qmax⁡Q_{\max} of 30 Å was used for the synchrotron data. The value of rpolyr_{\mathrm{poly}} was set according to normal protocols to ensure that the shape of the F(Q)F(Q) function had a concave baseline whilst keeping rpolyr_{\mathrm{poly}} less than the nearest neighbor bond-length in the material.

Models were fit to the PDF data using DIFFPY-CMI.[25] As we did in our previous work,[4] we fit models to uncorrected, corrected, and synchrotron data to evaluate the effect of the absorption correction by comparison to the fitted synchrotron data. Structural models for m-ZrO2_2, c-CeO2_2, and m-HfO2_2 were taken from the Inorganic Crystal Structure Database (ICSD) (entry IDs 80047, 184584, and 60902, respectively).[26, 27, 28] The fit was performed on the Nyquist-Shannon (NS) grid to facilitate propagation of valid estimated uncertainties.[29] The refined parameters include the scaling variable s1s_1, the damping factor QdampQ_{\mathrm{damp}}, the broadening parameter QbroadQ_{\mathrm{broad}}, the correlated motion parameter δ2\delta_2, the lattice parameters, and the atomic displacement parameters (ADPs). The ADP constraints and fitted rr range vary slightly for different datasets and will be presented separately for each dataset.

Results

Assessment of the absorption correction for different μD\mu D’s

In Figure 2, by way of example, we show curves of cvec_{\mathrm{ve}} for various choices of μD\mu D. In Figure 2(a) μD\mu D varies from 6.14 (lower dark blue curve) to 71.51.

Figure 2. Absorption correction cvec_{\mathrm{ve}} calculated using the brute-force method for various values of μD\mu D, over the entire QQ-range for Mo Kα_\alpha radiation. From bottom to top in (a) cvec_{\mathrm{ve}} for μD=6.14\mu D = 6.14 (dark blue), 7.05, 17.15, 28.16, 71.51 (light blue), and (b) cvec_{\mathrm{ve}} for μD=0.4\mu D = 0.4 (dark green), 1, 2, 3, and 4 (light green). The insets show the cvec_{\mathrm{ve}} curves for lower μD\mu D’s on expanded scales. In (a), the cvec_{\mathrm{ve}} value for the largest QQ at the highest μD\mu D is labeled in the same light blue color as the corresponding curve.

The μD\mu D values are representative of samples of TiO2_2, ZrO2_2, SnO2_2, CeO2_2, and HfO2_2, respectively, measured with Mo KαK_{\alpha} radiation in a 1 mm tube, but are only chosen as representative of real materials from different rows in the period table for comparison.

All the curves fall from a high value at low-QQ to a smaller value at high-QQ. For the heavier elements the low-QQ signal would be multiplied by up to 200,000 times, falling to around 140× at the highest-QQ of 16.5 Å−1^{-1}. Experiments are obviously not tenable with a 10510^5 attenuation in signal and for heavier elements at modest x-ray energies thinner samples

are needed, as is widely known. We explore this in more quantitative detail below. In the smallest inset of Figure 2(a), we show the curves for more experimentally reasonable, albeit somewhat high, μD\mu D values of around 6. In Figure 2(b) we show curves for μD\mu D less than 6 (0.4<μD<40.4 < \mu D < 4). Even for these experimentally more realistic cases, the value of cvec_{ve} has a strong QQ-dependence. For the higher range of μD\mu D, the value of cvec_{ve} can change by ≈50%\approx50\% from the low-QQ to the high-QQ end. A μD\mu D in the vicinity of unity is generally considered optimal (shown in the inset to Figure 2(b)). Below we explore data quality for higher μD\mu D values for the cases where it is difficult to make samples for PDF that are sufficiently thin.

Exploration of different μD\mu D cases

To understand how the cvec_{ve} curves vary in shape with μD\mu D we compute them and plot them scaled to go from zero to one, in Figure 3. They are normalized to go from zero to one as we are interested in the shape of the curve and not its absolute magnitude in this analysis. The figure shows curves in the range from 0.1<μD<900.1 < \mu D < 90. All the curves fall off more slowly, followed by a rapid fall-off with increasing QQ and finally a long tail in the high-QQ region. What is characteristic of increasing μD\mu D is that the crossover happens at lower angle, and therefore lower-QQ. For the lower μD\mu D values the curves are relatively flat over a much wider angular range. The flatness of these curves for small μD\mu D allows the absorption correction to be neglected in the ad hoc PDFgetX3 algorithm[15] for the case of high-energy synchrotron x-ray measurements in the RAPDF[1] geometry. In that geometry, the maximum 2θ2\theta angle is around 40-50°, much less than the 2θmax⁡=140∘2\theta_{\max} = 140^\circ in typical laboratory PDF measurements, which places the absorption correction even more in a flat region of the cvec_{ve} curves. For lower-energy x-rays, which are more absorbing and require wider measurement angles, whether measured on laboratory instruments or at the synchrotron, this absorption correction should be considered in most cases.

Comparison of the shape of absorption correction $c_{ve}$ curves for values of $\mu D$ from 0.1 to 90

Figure 3. Comparison of the shape of absorption correction cvec_{ve} curves for the range of μD\mu D values from 0.1 to 90. The cvec_{ve} curves are normalized from 0 to 1 to emphasize the shape of the curves vs 2θ2\theta for different values of μD\mu D. The values of cmax⁡c_{\max} and cmin⁡c_{\min} reported in the panels are the maximal and minimal values, respectively, of cvec_{ve} before normalization for the angular range shown (0<2θ<140∘0 < 2\theta< 140^\circ).

The maximal and minimal absolute values of cvec_{ve} before the normalization are also reproduced in each panel of the figure to give an idea about how large of an angle dependence

is required on an absolute scale. For μD≤2\mu D \le2, even for the wide angular range of the data considered here, the angle dependence of the correction is quite small, but it grows rapidly for larger μD\mu D values. By a μD=6\mu D = 6 the low-QQ correction is 7×7 \times larger than the high-QQ correction and there is a significant angular dependence, though the data are probably still usable after applying the correction we lay out here.

To summarize, scattering properties are optimal and absorption corrections are minimal for μD\mu D values of around 1. However, samples with μD\mu D up to around 6 result in reasonable absorption corrections, but should have a correction applied for data collection over a wide angle. For samples with larger μD\mu D values the corrections become large and the data are not likely to be good.

Estimating μD\mu D for a sample

In this Section we consider a number of different ways for estimating μD\mu D for a sample. At first sight this is straightforward since this quantity can, in principle, be calculated from mostly known quantities. However, in practice, some quantities such as the sample density or its chemical composition, are often not well known. Also, the model we used to compute the correction makes some assumptions that are not necessarily true in practice. For example, that of a parallel beam of the same width as the sample is not necessarily true in practice. We therefore seek to determine an “effective” μD\mu D, μDe\mu D_e, for our sample which is the μD\mu D that gives the most appropriate cvec_{ve} curve given our experimental conditions. We compare a number of different approaches for determining μDe\mu D_e by evaluating their effect on the refined structural parameters. We would like to understand empirically which approaches for estimating μD\mu D are preferred, as well as, in general, how big of an effect the cvec_{ve} correction has on refined parameters.

We first consider a number of different ways for estimating μD\mu D for a sample. Later we compare them by modeling. The μD\mu D of a sample can be measured directly for a given diffractometer setup by measuring the x-ray attenuation of the capillary specimens placed in the incident beam as shown in Figure 4. In this measurement, the sample stage is moved vertically (defined here as the zz-axis) traversing the incident beam. In our case, for this, the Bruker Lynxeye detector is set to 0D-mode, which means that the whole detected area is integrated.

Figure 4. Scheme of the diffractometer setup for direct measurements of the sample μD\mu D, called here zz-scans. The sample stage is moved along zz so the sample traverses the incident beam whilst recording the x-ray intensity as a function of zz. The capillary sample is displayed as an orange circle, with its principle axis perpendicular to the viewing plane. This results in a U-shaped profile (orange curve on the right). The capillary diameter, marked by the black bar on the right, is slightly shorter than the full opening of the orange U-shaped curve because of the finite width of the beam.

The resulting absorption profile can be fit with a function that is a convolution of the width of the incident beam and the absorption characteristics of the sample, assuming it is cylindrical, as we describe below.

We tested absorption scans where the incident beam height was made narrower and more tightly collimated than was the case for the actual measurement of the PDF data. We also tested making the absorption scan with wider incident beam heights, including the height used in the actual PDF measurements. To accomplish the experiment with the narrower beam, either the active channels on the detector were reduced, or a horizontal slit was placed in the incident beam before the sample to narrow the beam. The former limits the number of pixels that are active, limiting the active area on the detector to a horizontal stripe. The latter masks the beam and limits beam divergence. We found that both have a similar effect on the resulting U-shaped zz-scan. We then compared three methods to determine μD\mu D from the zz-scans.

Method 1: An approximate μD\mu D is determined using

μDm1=ln⁡(Imax⁡/Imin⁡),(8)\mu D_{m1} = \ln(I_{\max}/I_{\min}), \tag*{(8)}

where Imax⁡≈I0I_{\max} \approx I_0 and Imin⁡≈I0⋅e−μDI_{\min} \approx I_0 \cdot e^{-\mu D}. The rationale is that the minimum attenuation (Imin⁡I_{\min}) is reached when the center of the capillary aligns with the center of the x-ray beam. If the vertical beam height is small enough relative to the capillary diameter, then at this point the sample thickness doesn’t change much across the beam diameter and the measurement resembles a standard μt\mu t measurement of a sample of uniform thickness tt. Knowing μt\mu t at the position of the diameter, and the diameter, we can obtain μ\mu, and therefore μD\mu D for the capillary. This reasoning breaks down when the beam height becomes comparable to the sample height.

Method 2: We carry out a fit of the zz-scan curve to a model that assumes a circular cross-section capillary of uniform density and a parallel x-ray beam of height hh, which estimates μD\mu D. The mathematical details are included in Supplementary § 4. The fit yields μDm2\mu D_{m2} by fitting parameters μ\mu, DD, hh, I0I_0, z0z_0, the height of the center of the capillary, and mm, a linear coefficient for I0I_0 that we found depends on zz. This model ignores effects of the sample container which we expect to have a negligible effect for thin polyimide or quartz tubes at Mo or Ag KαK_{\alpha} energies. It will be a less good approximation if the sample container absorption is significant.

Method 3: Method 3 is the same as Method 2 except that for μDm3\mu D_{m3} the capillary diameter, DD, is fixed to the known value from the manufacturer and not allowed to vary as it is in Method 2. In an ideal world, Method 2 would return a fit diameter that is very close to the known physical diameter, but we found that this was not always the case due to inadequacies in our model and we wanted to understand the effect this has on the results.

It is also possible to estimate μD\mu D theoretically using the known x-ray wavelength, sample composition, and densities and/or powder packing fractions. and so we define also Method 4: This makes use of tabulated attenuation coefficients from online resources such as XrayDB,[23] APS,[30] and the NIST Standard Reference Database 126.[31] Here we calculate μDth\mu D_{\mathrm{th}} values using a measured mass density for our loaded samples and the lookup tables that makes use of the python XrayDB database.[23] The computing details are described in Supplementary Section 5.

For convenience we have developed μD\mu D calculators in the Python software package DIFFPY.UTILS[32] for Methods 2 and 4, which are free to use. For Methods 2 and 3, we also explored various experimental settings for the zz-scan, including varying x-ray beam heights hh and detector channel conditions. The full set of μD\mu D values and fitted parameters that we tried is presented in Tables S9 and S10 in Supplementary Information. Here we confine ourselves to zz-scan data for CeO2\mathrm{CeO_2} with ID=0.635 mm\mathrm{ID} = 0.635\ \mathrm{mm} which was chosen since it had a high, but not excessive, μD\mu D (μDth=4.229\mu D_{\mathrm{th}} = 4.229), which, according to the μD\mu D curves in Figure 3, suggests that an inappropriate estimation of this will result in a significant effect on the measured intensities. Examples of measured zz-scans are shown in Figure 5 as blue circles, computed using Method 2, with fits shown as red lines. The theoretical curve for the sample transmission before convolution with the beam height is shown as the brown curves.

Figure 5. Examples of zz-scans computed using Method 2 (blue circles) with model fits (red lines) for CeO2_2 with ID=0.635I_D = 0.635 mm. The green lines offset below show the residuals. In each case the brown U-shaped curve is the unconvoluted intensity which gives an indication of the edges of the capillary. The panels are arranged so side-by-side panels are measured with closed (left) and open (right) detector channel settings, and measured with increasing height of beam slit going down the columns h=0.05h = 0.05 mm, 0.20.2 mm, 0.60.6 mm, and 11 mm.

When comparing across rows where hh is constant, we observe that the maximal intensity Imax⁡I_{\max} is around twice as large for open channels, while the minimal intensity Imin⁡I_{\min} is about 4 times larger. This results in lower μDm2\mu D_{m2}’s for open detector channels. In addition, as hh increases (comparing down columns), we find that the slope of the U-shape curve (blue curve) becomes less steep. In all cases the fits of our model are satisfactory, albeit returning quite different μDm2\mu D_{m2}’s.

The different μDm2\mu D_{m2}’s are summarized in Figure 6 (a) and (b), along with those of μDm1\mu D_{m1} and μDm3\mu D_{m3}, for the same experimental conditions considered in Figure 5.

Figure 6. μD\mu D values for different CeO2\mathrm{CeO_2} datasets determined from zz-scans using varying incident beam heights (hh) and channel conditions. In all panels, the IDs are reported, with μDm1\mu D_{m1}, μDm2\mu D_{m2}, μDm3\mu D_{m3}, and μDth\mu D_{th} shown. μDth\mu D_{th} for panels (a), (b), (c), and (d) are 4.229, 4.229, 4.554, and 5.824, respectively, as indicated by the dashed grey line. In panels (a) and (b), lighter color bars indicate larger hh values, h=0.05h = 0.05 mm, 0.2 mm, 0.6 mm, and 1 mm. Panel (a) shows the reduced channel detector condition, while (b) shows the open channel detector condition. Panels (c) and (d) show data with reduced channels with h=0.1h = 0.1 mm.

Comparing across the methods, we see that for each experimental setting indicated by the same color, Methods 1-3 give very similar values as each other. This means that we could use any of these methods to estimate μD\mu_D and it wouldn’t matter. On the other hand, Method 4, using the theoretically determined μD\mu_D, indicated by the dashed grey line, gives a consistently higher value.

However, we see large variations in the estimate of μD\mu_D depending on the experimental conditions used to measure it. Comparing across color lightness (beam height, hh) and between panels (a) and (b) (closed vs. open detector channels), the resulting μD\mu_D can vary by almost a factor of two.

A question therefore arises, which of these values of μD\mu_D is the most appropriate for correcting the XRD data for PDF analysis, and that is explored in the next Section. Here we can make some observations that help us to understand why we see these variations.

We first note that the beam height, limited by physical slits upstream of the sample and the channel settings of the detector have a similar effect. The former limits the effective area of the sample that is illuminated, and the latter limits the area of the beam on the detector that is accounted for. Comparing panels (a) and (b), we always get slightly lower μD\mu_D values with the open-channel setting, but the difference is small when hh is small (i.e., 0.05 and 0.2 mm).

We note that another important length-scale in the problem which is the diameter of the capillary (0.635 mm in this case). When hh is much smaller than the capillary diameter, we get larger estimates of μD\mu_D that are also closer to the theoretical μD\mu_D’s but much smaller estimates of μD\mu_D when the zz-scan is measured with beam-slit heights comparable to or larger than the sample capillary.

Understanding the effect of different μD\mu D correction magnitudes on fits and refined parameters

We next investigate the effect on the fit-quality and refined parameters of correcting the ceria PDFs using various different μD\mu D’s. We use CeO2_2 data from the 1.024 mm capillary as an example, where the absorption effect is strong. For the purpose of comparison, in addition to the uncorrected data, we consider three different μD\mu D values for the corrected data: the experimental μD=5.328\mu D = 5.328 obtained from the zz-scan with h=0.1h = 0.1 mm and reduced channels using Method 2, a purposefully underestimated μD=3\mu D = 3, and an overestimated μD=10\mu D = 10. We do this by computing structure functions and PDFs from the measured data after each of the three corrections, and fitting the PDFs of the ceria crystal structure model using Diffpy-CMI.

The effect of these corrections on the raw data, F(Q)F(Q), and G(r)G(r) are shown in Figure 7.

Figure 7. Comparison of CeO2\mathrm{CeO}_2 data (ID = 1.024 mm) with and without absorption correction, using small (μD=3\mu D = 3), middle (μD=5.328\mu D = 5.328), and overestimated (μD=10\mu D = 10) values. Panel (a) shows F(Q)F(Q), where the dark blue, light blue, green, and red curves represent no correction and corrections with small, middle, and overestimated μD\mu D, respectively. The curves overlap or change in amplitude across QQ-ranges, marked by the light grey dashed vertical line, with the order of plotting adjusted to reflect changes without disrupting overall trends. Panels (b)–(d) show the original intensity (blue) vs. corrected intensity (red), scaled so that the highest intensity peak equals to one. Plots are shown on a QQ scale, focusing on Q>7Q > 7 Å−1^{-1}. Panels (e)–(g) show the best-fit calculated PDFs (red lines) on top of their respective measured, absorption corrected, PDFs (blue circles). The PDFs were fitted between rmin⁡=1.0r_{\min} = 1.0 and rmax⁡=40.0r_{\max} = 40.0. The corrections in left columns (b) and (e) correspond to the small μD\mu D, the middle columns (c) and (f) to the middle μD\mu D, and the right columns (d) and (g) to the overestimated μD\mu D.

We consider first the effect on the high-QQ region of the raw data (Figure 7(b-Figure 7d)). In these plots the curves are normalized so the strongest peak in each curve is set to unity, allowing us to conveniently compare the behavior at high-QQ. Similar to what was shown in Schertenlieb et al.,[4] albeit in that case for lower absorbing samples, we see that the absorption correction reduces an overall upturn in the data whilst at the same time suppressing the high-QQ signal. In other words, neglecting the correction will result in an amplified signal at high-QQ, which might resemble an under-estimated Debye-Waller factor in subsequent refinements. Unsurprisingly, the effect on the signal becomes larger as the μD\mu D correction is increased, though the slight upturn in the data doesn’t completely go away even for μD=10\mu D = 10 which we believe to be a significant overestimate of the actual sample absorption. As discussed briefly before,[4] there is an additional contribution to the upturn that is not coming from absorption effects and so simply removing the upturn cannot be used to estimate μD\mu D.

The effect on F(Q)F(Q) plotted over the whole range of QQ is shown in Figure 7(a), where we find under-estimation in low-QQ signal and over-estimation in the high-QQ signal when comparing uncorrected to corrected data. The progressive suppression of the high-QQ signal (Q>12Q > 12)

with increasing correction is very clear.

We turn to fits of structural models to the PDF to understand the effect of the corrections on the fit quality and on refined structural parameters.

The G(r)G(r) fits for the corrected data are shown in Figure 7 (e)-(Figure 7g). A larger version is shown in Figure S14 in the Supplementary Information. All the curves result in reasonable model fits, with small misfits, as evident in the green difference curves. The RwR_w value for no correction is approximately 0.72, and it improves to 0.6, 0.44, and 0.36 for small, middle, and overestimated μD\mu D, respectively. The fit quality, as measured by the RwR_w, and visually in the difference curve, gets better with increasing magnitude of the absorption correction. This trend occurs even when we use a correction that is known to be too strong. This means that fit quality alone, as measured by RwR_w, is not a good measure of the best absorption correction to use. Of course, the purpose of the fits is to obtain high quality values for refined structural parameters. We therefore turn to an analysis of the refined parameter values obtained with different degrees of absorption correction.

Figure S14. CeO₂ PDF fits for synchrotron, uncorrected, and corrected data with different μD\mu D’s (reported in the panels), over rmin⁡=1.0r_{\min}=1.0 and rmax⁡=40.0r_{\max}=40.0. The plots show the measured PDFs (blue), the best-fit PDFs (red), and the residuals (green).

The refinement results for the three selected μD\mu D corrections are shown in Table 3. We are most concerned about the refined values of structural parameters that are, in this case, aa, Ce(UisoU_{\mathrm{iso}}) and O(UisoU_{\mathrm{iso}}). The effect of the correction on the lattice parameter is small, varying at the thousandths of an Ångström level, but does show a small decrease in lattice parameter with increasing strength of correction. The Ce(UisoU_{\mathrm{iso}}) and O(UisoU_{\mathrm{iso}}) values vary monotonically with μD\mu D. Ce(UisoU_{\mathrm{iso}}) increases with increasing absorption correction in line with the progressive suppression of high-QQ signal by the correction. Surprisingly, O(UisoU_{\mathrm{iso}}) shows the opposite behavior, decreasing with increasing correction magnitude, but this may reflect the fact that the oxygen contributions to the measured PDFs in CeO2_2 is very small and so these ADPs are not very reliable.

ParametersynchrotronuncorrectedμD=3\mu D = 3μD=5.328\mu D = 5.328μD=10\mu D = 10
s1s_10.37434(21)0.2191(9)0.2972(12)0.3693(12)0.3287(9)
QdampQ_{\mathrm{damp}}0.02386(4)0.03186(25)0.03074(24)0.02737(23)0.01470(32)
QbroadQ_{\mathrm{broad}}0.01814(7)0.0220(5)0.0251(5)0.0320(4)0.0461(4)
δ2\delta_29.07(4)9.92(28)9.21(17)7.87(13)4.99(13)
aa5.414232(7)5.40240(5)5.40160(5)5.40018(5)5.39814(6)
Ce(Uiso)\mathrm{Ce}(U_{\mathrm{iso}})0.003571(5)0.002403(24)0.002671(25)0.003154(29)0.00467(4)
O(Uiso)\mathrm{O}(U_{\mathrm{iso}})0.04120(10)0.0990(24)0.0696(13)0.0472(7)0.0353(4)
QmaxQ_{\mathrm{max}}30.016.616.616.616.6
grid0.104720.1892530.1892530.1892530.189253
RwR_w0.1608230.7189910.5964040.4391310.363539
χred2\chi^2_{\mathrm{red}}961.5289581173.593401753.497185447.645936452.603539

Table 3. Results of the refinement for CeO2_2 data (ID=1.024 mm) for synchrotron, uncorrected, and corrected data with selected μD\mu D’s, over rmin⁡=1.0r_{\min} = 1.0 and rmax⁡=40.0r_{\max} = 40.0. The μD\mu D’s are the values used for each correction.

To decide which μD\mu D values are optimal we compare to fits to the synchrotron data. The refined lattice parameters from the lab data are closer to reported values, [27] indicating that the synchrotron data overestimates the lattice parameter, which may be related to a calibration issue with the synchrotron data. Among the three selected correction strengths, we find that the experimentally determined μD=5.328\mu D = 5.328 provides UisoU_{\mathrm{iso}} values that agree the best with the synchrotron data.

A detailed exploration of μDe\mu D_e for all CeO2_2 data is presented in Supplementary § 6, where we propose a recommended reference value. Additional examples for both low and high sample absorption cases are discussed in Supplementary § 7.

diffpy.labpdfproc software for absorption corrections

The corrections described here have been implemented in an easy to use Python package, diffpy.labpdfproc, as part of the diffpy[33] family of software packages. It can be used to compute absorption corrections and apply them to measured data before the data are fed to, for example, PDFgetX3, to obtain PDFs.

The software can be easily installed from the Python Package Index (PyPI)[34] or from conda-forge[35] and so is straightforward to use alongside other diffpy programs. Full instructions for installation and use can be found at the GitHub readme[36] and the online documentation.[37]

To use this tool, you only need your 1D diffraction pattern data, along with either the μD\mu D value, a zz-scan file, or relevant chemical information. The program is fast and easy to use. By default, we use a fast calculation for cvec_{ve} correction, described in Supplementary Section 3. It can be run on single file or on directories of files. It also implements the more computationally intensive brute-force calculation method that is described in § 2. DIFFPY.LABPDFPROC can also be used to facilitate estimating μD\mu D for a sample of known composition before making the samples as part of the experiment design.

Recommended Protocols for sample preparation and absorption correction for PDF measurements in laboratory settings

Overall, we find that making an absorption correction is better than not, especially for 2<μD<72 < \mu D < 7. For high μD>8\mu D > 8, we are unlikely to get usable data. Here, we provide our recommended protocols for handling sample absorption in detail. For other experimental decisions for PDF experiments from lab sources, please refer to our previous work.[4] The linear absorption coefficient, μ\mu, can be estimated if an approximate composition is known by using our program or using websites mentioned herein. If you do not know the packing density of your sample, use an approximate value of 0.5. Choose a sample diameter to give a μD\mu D of less than 2 ideally but anything up to 6 or even 7 is acceptable. For very absorbing samples, where no tube or capillary is available that is thin enough, you can consider coating the outside of a narrow tube or glass fiber in grease and sprinkling a thin layer of powder on the outside. The results of doing this for our HfO2_2 sample are reproduced in the Supplementary Information Section 7.

To estimate the most appropriate μD\mu D for correction, μDe\mu D_{\mathrm{e}}, we propose the following: (1) If the capillary diameter is close to the x-ray beam height (D≥0.8∗hD \ge0.8 \ast h), either compute μD\mu D theoretically or measure the experimental μD\mu D using a zz-scan with reduced channels and small beam heights. (2) If the capillary diameter is much smaller than the beam height, scale the theoretical μD\mu D with D/hD/h or do a zz-scan with the actual experimental settings.

Conclusion

In contrast to rapid acquisition PDF experiments, $3 PDF experiments carried out on laboratory diffractometers, or at synchrotrons but with low x-ray energy, require absorption corrections. Here we revisit the nature of absorption corrections, and their effect on the resulting structural refinements, for laboratory PDF measurements which are becoming more popular. We report a new software program, diffpy.labpdfproc, that can carry out absorption corrections in a fast and straightforward way before data are then propagated to the popular PDFgetX3 program. We also assessed the effects of different absorption corrections on subsequent PDF refinements, as well as exploring different methods for determining the best \mu D$ values in a particular experimental situation. The goal is to make this step as straightforward as possible so that it can be easily incorporated into laboratory PDF experimental workflows. As a result, we present protocols for sample preparation to improve data quality.

Acknowledgements

Synchrotron total scattering measurements were performed at beamline ID31 at the European Synchrotron Radiation Facility (ESRF). Work in the Billinge Group was supported by U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences (DOE-BES) under contract No. DE-SC0024141. T. Schertenleib was supported by the Swiss National Science Foundation under grant number 200021 188536.

Data availability

Data used for all the plots in the manuscript are available on Zenodo at https://doi.org/10.5281/zenodo.15199572.

Supplementary Information

Derivation of the path length of an x-ray through the sample

We derive here the relations for determining ℓv(2θ)\ell_v(2\theta), the path length through the sample of an x-ray scattered at angle 2θ2\theta from the vvth voxel. As shown in Eq. 3, this is given by the sum of the path lengths of the incoming and outgoing x-ray paths, ℓvi\ell_v^i and ℓvo\ell_v^o, respectively.

In our simplified model we assume that the incoming beam is parallel (non-diverging). The geometry is shown schematically in Figure 1 of the main paper, where the circle in this figure represents the cross-section of the capillary and the incoming beam arrives horizontally from the left. Placing the origin of our coordinate system at the center of the circle we can consider a pixel, vv, located at position with coordinates (xv,yv)(x_v,y_v) (as examples, see the large grey dots in Figure 1). The incident path length will be given by ℓvi=∣xv−xi∣\ell_v^i = |x_v-x_i|, where the coordinates where the beam enters the capillary circle are (xi,yi)(x_i,y_i) (indicated by the red dots in Figure 1). Points on the capillary circle satisfy x2+y2=R2x^2+y^2=R^2, where R=D2R=\frac{D}{2} is the radius of the capillary circle. Since yi=yvy_i=y_v which is known, we take the negative value and get that

xi=−R2−yv2x_i=-\sqrt{R^2-y_v^2}

and

ℓvi=∣xv+R2−yv2∣.(9)\ell_v^i=\left|x_v+\sqrt{R^2-y_v^2}\right|. \tag*{(9)}

To get the path length of the out-going x-ray we notice that, if the coordinates of the point on the circle where the ray leaves the capillary are (xo,yo)(x_o,y_o) (indicated by the red dots in Figure 1), then

ℓvo(2θ)=(xo−xv)2+(yo−yv)2.(10)\ell_v^o(2\theta)=\sqrt{(x_o-x_v)^2+(y_o-y_v)^2}. \tag*{(10)}

Since xvx_v and yvy_v are known, it remains to find xox_o and yoy_o.

When 2θ≠90∘2\theta\ne90^\circ, we find the yellow line equation y=ax+by=ax+b that passes through both the voxel and the exit point, where its slope a=tan⁡(2θ)a=\tan(2\theta) is determined by the scattering angle and its intersection point can be computed from the grid points b=yv−xv⋅tan⁡(2θ)b = y_v - x_v \cdot\tan(2\theta).

Therefore, we have that xo=yo−bax_o = \frac{y_o-b}{a}. Since the exit point is an intersection point that passes through the line and the circle, we solve by establishing simultaneous equations for the circle and the line. We then have:

xo2+yo2=R2(11)x_o^2 + y_o^2 = R^2 \tag*{(11)}
(yo−ba)2+yo2=R2(12)\left(\frac{y_o-b}{a}\right)^2 + y_o^2 = R^2 \tag*{(12)}
(1+a2)yo2−2byo+(b2−a2R2)=0.(13)(1+a^2)y_o^2 - 2by_o + (b^2-a^2R^2) = 0. \tag*{(13)}

Since a2≥0a^2 \ge0, 1+a2≠01+a^2 \ne0, and we can use the quadratic formula to get yoy_o:

yo=2b±4b2−4(1+a2)(b2−a2R2)2(1+a2)(14)y_o = \frac{2b \pm\sqrt{4b^2 - 4(1+a^2)(b^2-a^2R^2)}}{2(1+a^2)} \tag*{(14)}
=b±a(1+a2)R2−b21+a2.(15)= \frac{b \pm a\sqrt{(1+a^2)R^2-b^2}}{1+a^2}. \tag*{(15)}

As we consider only voxels within the cylinder (xv2+yv2≤R2)(x_v^2+y_v^2 \le R^2), the discriminant of this quadratic is positive:

(1+a2)R2−b2=R2+a2R2−(yv−axv)2(16)(1+a^2)R^2-b^2 = R^2+a^2R^2-(y_v-ax_v)^2 \tag*{(16)}
=(R2−yv2)+(a2R2−a2xv2)+2axvyv(17)= (R^2-y_v^2)+(a^2R^2-a^2x_v^2)+2ax_vy_v \tag*{(17)}
≥xv2+a2yv2+2axvyv(18)\ge x_v^2+a^2y_v^2+2ax_vy_v \tag*{(18)}
=(xv+ayv)2(19)= (x_v+ay_v)^2 \tag*{(19)}
≥0.(20)\ge0. \tag*{(20)}

Since we only have to consider paths leaving the circle in direction of the detector, the exit point is above the grid point thus is the upper intersection point. So we take the positive value here. Substituting and rearranging the equations, we get:

xo=(1+tan⁡2(2θ))R2−(yv−xvtan⁡(2θ))2−yvtan⁡(2θ)+xvtan⁡2(2θ)1+tan⁡2(2θ)(21)x_o = \frac{\sqrt{\left(1+\tan^2(2\theta)\right)R^2-\left(y_v-x_v\tan(2\theta)\right)^2}-y_v\tan(2\theta)+x_v\tan^2(2\theta)}{1+\tan^2(2\theta)} \tag*{(21)}
yo=yv−xvtan⁡(2θ)+tan⁡(2θ)(1+tan⁡2(2θ))R2−(yv−xvtan⁡(2θ))21+tan⁡2(2θ)(22)y_o = \frac{y_v-x_v\tan(2\theta)+\tan(2\theta)\sqrt{\left(1+\tan^2(2\theta)\right)R^2-\left(y_v-x_v\tan(2\theta)\right)^2}}{1+\tan^2(2\theta)} \tag*{(22)}

which can then be plugged back into Eq. 2.

This function is not well behaved in the vicinity of 2θ=90∘2\theta=90^\circ. At 2θ=90∘2\theta=90^\circ, xo=xvx_o=x_v and so yo=R2−xv2y_o=\sqrt{R^2-x_v^2}, and

ℓv(2θ=90∘)=∣yv−R2−xv2∣.(23)\ell_v(2\theta=90^\circ)=\left|y_v-\sqrt{R^2-x_v^2}\right|. \tag*{(23)}

Derivation of the result that cvec_{v_{\mathrm{e}}} depends on the product μD\mu D: Express ℓv(2θ)\ell_v(2\theta) in radius RR

From our previous expressions for ℓv(2θ)\ell_v(2\theta), Eq. 1, Eq. 2, Eq. (23), we find that if we can express xvx_v and yvy_v linearly in terms of RR, then we can factor RR out for all ℓv(2θ)\ell_v(2\theta) and therefore use the product μD\mu D as an independent variable of cvec_{v_{\mathrm{e}}} rather than μ\mu and DD separately.

Here, we present a method for expressing the Cartesian coordinates (xV,yV)(x_V,y_V) of any point VV inside a circle in terms of the radius RR. This process is also visualized in Figure S1 below.

Figure S1. Visualization of the same V′V'th pixel for circles with varying diameters. The pixels are denoted by the blue dots, VV and V′V', for the small and large circle, respectively. The blue lines represent the line segments connecting each pixel to the origin, and the dashed grey lines are the perpendicular lines intersecting each circle at two points. Here, we pick the light brown upper intersection point as AA and A′A'. These light brown lines thus represent the radii. We can then find αV\alpha_V and βV\beta_V, which are the same for VV and V′V'.

Consider the line segment OVOV from the origin, OO, at the center of the circle to the point VV inside the circle. Draw a line perpendicular to OVOV that goes through VV. This line intersects the boundary of the circle at two points. Choosing one such point and denoting it as AA, we establish the radius OA=ROA=R. Let αV=∠VOA\alpha_V=\angle VOA, then cos⁡(αV)=OV/OA\cos(\alpha_V)=OV/OA, which gives us OV=Rcos⁡(αV)OV=R\cos(\alpha_V). Transitioning to polar coordinates, we have that xV=OVcos⁡(βV)x_V=OV\cos(\beta_V) and yV=OVsin⁡(βV)y_V=OV\sin(\beta_V), where βV\beta_V represents the angle between the positive xx-axis and OVOV. We therefore have xV=Rcos⁡(αV)cos⁡(βV)x_V=R\cos(\alpha_V)\cos(\beta_V) and yV=Rcos⁡(αV)sin⁡(βV)y_V=R\cos(\alpha_V)\sin(\beta_V).

Denote αV\alpha_V as the angle formed between the line segment connecting the VV’th pixel to the origin and the radius of the circle to the origin, and βV\beta_V as the angle between the positive xx-axis and the line segment connecting the pixel to the origin. Since xV=Rcos⁡(αV)cos⁡(βV)x_V = R \cos(\alpha_V)\cos(\beta_V) and yV=Rcos⁡(αV)sin⁡(βV)y_V = R \cos(\alpha_V)\sin(\beta_V), we have the following:

ℓVi=∣Rcos⁡(αV)cos⁡(βV)+R2−(Rcos⁡(αV)sin⁡(βV))2∣(24)\ell_V^{i} = \left| R \cos(\alpha_V)\cos(\beta_V) + \sqrt{R^2 - \left(R \cos(\alpha_V)\sin(\beta_V)\right)^2} \right| \tag*{(24)}
=∣cos⁡(αV)cos⁡(βV)+1−cos⁡2(αV)sin⁡2(βV)∣R,(25)= \left| \cos(\alpha_V)\cos(\beta_V) + \sqrt{1-\cos^2(\alpha_V)\sin^2(\beta_V)} \right| R, \tag*{(25)}
ℓV(2θ=90∘)=∣Rcos⁡(αV)sin⁡(βV)−R2−(Rcos⁡(αV)cos⁡(βV))2∣(26)\ell_V(2\theta= 90^\circ) = \left| R \cos(\alpha_V)\sin(\beta_V) - \sqrt{R^2 - \left(R \cos(\alpha_V)\cos(\beta_V)\right)^2} \right| \tag*{(26)}
=∣cos⁡(αV)sin⁡(βV)−1−cos⁡2(αV)cos⁡2(βV)∣R,(27)= \left| \cos(\alpha_V)\sin(\beta_V) - \sqrt{1-\cos^2(\alpha_V)\cos^2(\beta_V)} \right| R, \tag*{(27)}

and

ℓV(2θ≠90∘)=(xo−xV)2+(yo−yV)2(28)\ell_V(2\theta\ne90^\circ) = \sqrt{(x_o-x_V)^2 + (y_o-y_V)^2} \tag*{(28)}
=G2+H2R,(29)= \sqrt{G^2 + H^2}R, \tag*{(29)}

where

E=1+tan⁡2(2θ)−cos⁡2(αV)(sin⁡(βV)−cos⁡(βV)tan⁡(2θ))2(30)E = \sqrt{1+\tan^2(2\theta)-\cos^2(\alpha_V)\left(\sin(\beta_V)-\cos(\beta_V)\tan(2\theta)\right)^2} \tag*{(30)}
F=cos⁡(αV)(sin⁡(βV)−cos⁡(βV)tan⁡(2θ))(31)F = \cos(\alpha_V)\left(\sin(\beta_V)-\cos(\beta_V)\tan(2\theta)\right) \tag*{(31)}
G=E−tan⁡(2θ)F1+tan⁡2(2θ)−cos⁡(αV)cos⁡(βV)(32)G = \frac{E-\tan(2\theta)F}{1+\tan^2(2\theta)}-\cos(\alpha_V)\cos(\beta_V) \tag*{(32)}
H=F+tan⁡(2θ)E1+tan⁡2(2θ)−cos⁡(αV)sin⁡(βV)(33)H = \frac{F+\tan(2\theta)E}{1+\tan^2(2\theta)}-\cos(\alpha_V)\sin(\beta_V) \tag*{(33)}
xo=E−tan⁡(2θ)F1+tan⁡2(2θ)R(34)x_o = \frac{E-\tan(2\theta)F}{1+\tan^2(2\theta)}R \tag*{(34)}
yo=F+tan⁡(2θ)E1+tan⁡2(2θ)R.(35)y_o = \frac{F+\tan(2\theta)E}{1+\tan^2(2\theta)}R. \tag*{(35)}

Since αV\alpha_V, βV\beta_V, and 2θ2\theta do not change for different RR, we can factor out RR from ℓV(2θ)\ell_V(2\theta). Denote ℓV(2θ)=D⋅dℓV(2θ)\ell_V(2\theta)=D\cdot d\ell_V(2\theta), using Eq. 7, we have

cve=NV∑Ve−μD⋅dℓV(2θ).(36)c_{ve}=\frac{N_V}{\sum_V e^{-\mu D\cdot d\ell_V(2\theta)}}. \tag*{(36)}

This concludes the proof that the same μD\mu D would give the same cvec_{ve}, and we can consider it as an independent variable.

Fast calculation of the absorption correction using a polynomial approach

Part of the popularity of PDFgetX3 is the simplicity and speed of applying ad hoc corrections to the diffraction data that require little user input and still result in reliable PDFs.4 For longer wavelength x-rays where absorption is important we would like a similarly quick and straightforward absorption correction that can be incorporated into the PDFgetX3 workflow.

Here we describe a fast approximate calculator of cvec_{ve} for a given μD\mu D. Since our recommended protocols for data acquisition are to measure samples with moderate absorption, i.e., 0.5≤μD≤70.5 \le\mu D \le7, we confine the estimates to this range. As a start, we explore estimating cvec_{ve} through an interpolative reconstruction approach using polynomials, given a known value for μD\mu D. Our strategy is to build a relatively sparse database of cvec_{ve} curves computed by brute force, and to find curves for μD\mu D values in between using interpolation of polynomials. The 1/cve1/c_{ve} curves vary less strongly than cvec_{ve}, with all values falling between 0 and 1, and we found the interpolation works better on 1/cve1/c_{ve} than on cvec_{ve} itself.

We started by creating a sparse dataset of 1/cve1/c_{ve} curves for μD=0.5\mu D = 0.5, 1, 2, 3, 4, 5, 6, and 7 over a range 1≤2θ≤180∘1 \le2\theta\le180^\circ on a grid of spacing of Δ2θ=0.1∘\Delta2\theta= 0.1^\circ using the brute-force approach. We then carried out a fit of a polynomial to each of these curves. After experimenting with using different polynomial degrees, we found that a sixth-degree polynomial provided reliable reconstructions. The reconstructed curves are very close to the original ones as can be seen in Figure S2 which shows a comparison on the brute force and polynomial fits for the cvec_{ve} curves chosen above. The fits are very good, as the blue brute-force curves are barely visible. Small errors are seen clustered around low 2θ2\theta, with maximum percentage error of around 0.68% for μD=7\mu D = 7 at 2θ=1∘2\theta= 1^\circ. The maximum percentage errors for μD=4\mu D = 4, 5, and 6 also occurred at 2θ=1∘2\theta= 1^\circ, with values of 0.03%, 0.12%, 0.32%, respectively. For μD≤3\mu D \le3, the errors are almost negligible, with the maximal percentage error much less than ≤10−2\le10^{-2}%.

Figure S2. Comparison of brute-force and fast calculation of cνec_{\nu_e}. (a) Polynomial fits (red) and brute-force-calculated curves (blue, almost invisible) for μD\mu D values between 0.5 and 7 from our database of curves. (b) Polynomial curves found by interpolation from the database curves (brown), and brute-force curves computed for the same μD\mu D values (blue, almost invisible), for various μD\mu D’s randomly chosen between μD\mu D’s in the sparse dataset. In both panels, the insets show curves for smaller μD\mu D’s on an expanded scale. From bottom to top in (a) μD=0.5\mu D = 0.5, 1, 2, 3, 4, 5, 6, and 7, and in (b) μD=0.8637\mu D = 0.8637, 1.5973, 2.47, 3.42, 4.344, 5.2, 5.824, 6.5, and 6.8.

The polynomial coefficients appropriate to each μD\mu D are then stored in a database.

We then need a method to find the cvec_{ve} curves for any μD\mu D that may lie between those in our sparse database. To interpolate between curves of different μD\mu D, we used quadratic interpolation for the polynomial coefficients. We also experimented with linear and cubic interpolations: the former is not accurate enough and the latter tends to over-fit, settling on quadratic interpolation as the best approach. The full cvec_{ve} curve is then reconstructed using the resulting polynomial function. The accuracy of this approach is illustrated in Figure S2 (b). In the figure the blue curves are computed for intermediate μD\mu D values using the full brute force calculation, and the light brown curve is the one estimated using the polynomial interpolation approach. Since the maximum percentage errors are <1.1%< 1.1\% for all μD\mu D’s, we conclude that our model provides highly accurate estimations for cvec_{ve} across various μD\mu D values within our chosen range of 0.5≤μD≤70.5 \le\mu D \le7, while significantly reducing the computation time from hours to seconds.

Estimation of μD\mu D using model fitting of zz-scans

Here we develop a model for estimating μD\mu D experimentally from zz-scan data. We define the xx direction as the direction of travel of the x-ray beam and zz to be the vertical direction perpendicular to the capillary (see Figure 4). We place the origin at the center of the x-ray beam. Let R=D/2R = D/2 be the radius of the capillary cross-section, and hh to be the height of the x-ray beam. Finally, we define z0z_0 to be the height of the center of the capillary. z0z_0 could take any value without loss of generality but for convenience we try and set z0∼0z_0 \sim0. This offset will be refined in the fit.

The path length through the sample of an x-ray ray at the height, zz is then given by

l={2R2−(z−z0)2for ∣z−z0∣≤R0otherwise.(37)l = \begin{cases} 2\sqrt{R^2 - (z-z_0)^2} & \text{for } |z-z_0| \le R \\ 0 & \text{otherwise.} \end{cases} \tag*{(37)}

We define I0I_0 as the intensity of the unattenuated x-ray beam (assumed to have a uniform cross-section). We found that, in general, I0I_0 had a small, approximately linear, dependence on zz, so we also define a coefficient, mm, to account for this to obtain the best fits. In addition, we assume zz is sampled at discrete intervals with a constant step size Δz\Delta z.

We therefore define the intensity function I[z]I[z] as

I[z]={(I0−mz)⋅e−2μR2−(z−z0)2for ∣z−z0∣≤R,I0−mzotherwise.(38)I[z] = \begin{cases} (I_0-mz)\cdot e^{-2\mu\sqrt{R^2-(z-z_0)^2}} & \text{for } |z-z_0|\le R,\\ I_0-mz & \text{otherwise.} \end{cases} \tag*{(38)}

The normalized kernel function K[z]K[z] to account for the finite width beam is given by

K[z]=1C{1for ∣z∣≤h/2,0otherwise,(39)K[z] = \frac{1}{C} \begin{cases} 1 & \text{for } |z|\le h/2,\\ 0 & \text{otherwise,} \end{cases} \tag*{(39)}

where CC is a constant such that ∑zK[z]=1\sum_z K[z]=1. Therefore, CC is the number of xx such that ∣z∣≤h/2|z|\le h/2, and C=2⌊h2Δz⌋+1C=2\left\lfloor\frac{h}{2\Delta z}\right\rfloor+1.

To fit the measured curves we need to convolute I[z]I[z] with K[z]K[z], denoted as (I⊛K)[z](I\circledast K)[z], given by

(I⊛K)[z]=∑tI[t]⋅K[z−t](40)(I \circledast K)[z] = \sum_t I[t] \cdot K[z-t] \tag*{(40)}
=∑tI[t]{1Cfor ∣z−t∣≤h/20otherwise(41)= \sum_t I[t]\begin{cases} \frac{1}{C} & \text{for } |z-t| \le h/2 \\ 0 & \text{otherwise} \end{cases} \tag*{(41)}
=1C∑t:∣z−t∣≤h/2I[t](42)= \frac{1}{C}\sum_{t:|z-t|\le h/2} I[t] \tag*{(42)}
=1C∑t:∣z−t∣≤h/2{(I0−mt)⋅e−2μR2−(t−z0)2for ∣t−z0∣≤RI0−mtotherwise(43)= \frac{1}{C}\sum_{t:|z-t|\le h/2} \begin{cases} (I_0-mt)\cdot e^{-2\mu\sqrt{R^2-(t-z_0)^2}} & \text{for } |t-z_0|\le R \\ I_0-mt & \text{otherwise} \end{cases} \tag*{(43)}
=1C∑k=1C{(I0−mtk)⋅e−2μR2−(tk−z0)2for ∣tk−z0∣≤RI0−mtkotherwise(44)= \frac{1}{C}\sum_{k=1}^{C} \begin{cases} (I_0-mt_k)\cdot e^{-2\mu\sqrt{R^2-(t_k-z_0)^2}} & \text{for } |t_k-z_0|\le R \\ I_0-mt_k & \text{otherwise} \end{cases} \tag*{(44)}

where tk∈[z−h/2,z+h/2]t_k \in[z-h/2,z+h/2] with ti+1=ti+Δtt_{i+1}=t_i+\Delta t for i=1,2,…,C−1i=1,2,\ldots,C-1.

As seen from Eq. (44), the convolution at each zz is essentially an average of I[tk]I[t_k] for each tk∈[z−h/2,z+h/2]t_k \in[z-h/2,z+h/2]. As hh increases, the convolution deviates more significantly from the intensity function in Eq. (38) and as h→0h \to0, C→1C \to1 meaning that the number of tkt_k we are averaging approaches 1, in which case, Eq. (44) is just the unconvoluted intensity, Eq. (38).

We have developed a user-friendly fitting program in diffpy.utils[32] that accurately and efficiently extracts μD\mu D by optimizing the six parameters: μ\mu, DD, hh, I0I_0, z0z_0, and mm. The function only takes in a set of experimental zz and II values and minimizes the sum of squared residuals between the experimental data and the model convolution. The convolution is computed using SciPy signal’s convolve function with same mode, based on the intensity and kernel functions defined in Eq. (38) and Eq. (39). Boundary effects occur because the kernel only partially overlaps with the intensity signal at the edges, resulting in incomplete averaging. To address this, we extend the zz values for both intensity and kernel functions to ensure full coverage for the given zz values. The same mode also introduces a potential shift in convolution as it crops the result to match the size of zz. As a result, we recenter the kernel around z−zˉz-\bar{z}, where zˉ\bar{z} is the average of zz, aligning the intensity with the convolution.

To optimize the search for the global minimum, we use SciPy’s dual annealing method, a stochastic global optimization algorithm. This approach provides a robust solution to find the global minimum. By combining the theoretical results from Eq. (38), Eq. (39), and Eq. (44) with the fitting program, we achieve a more accurate estimation of μD\mu D.

Estimation of μD\mu D using a theoretical database

The process for estimating the theoretical linear attenuation coefficient μ\mu involves using the XrayDB database,[23] which can then be used to compute μD\mu D by multiplying with the capillary diameter DD. The inputs required for this calculation are the sample composition, the x-ray energy, and either a measured mass density, ρs\rho_s, for the sample, or the packing fraction ff, where ρs=ρm⋅f\rho_s=\rho_m\cdot f, where ρm\rho_m is the mass density of fully dense sample that can be estimated from the atomic structure if it is known.

To calculate μ\mu, we first compute the mass attenuation coefficient (μρ)e\left(\frac{\mu}{\rho}\right)_e for each element ee in the sample, in cm2^2/g. A mass-weighted sum approach is then applied to determine the total mass attenuation coefficient for the mixture. Let EE be the set of sample elements in the sample, fef_e be the number of atoms per element ee, and AeA_e be its atomic mass in g/mol. Then the mass contribution of each element ee is given by me=fe⋅Aem_e=f_e\cdot A_e, from which the total mass attenuation coefficient (μρ)tol\left(\frac{\mu}{\rho}\right)_{\mathrm{tol}} and total mass mtolm_{\mathrm{tol}} are computed as

(μρ)tol=∑e∈E(μρ)e⋅me(45)\left(\frac{\mu}{\rho}\right)_{\mathrm{tol}}=\sum_{e\in E}\left(\frac{\mu}{\rho}\right)_e\cdot m_e \tag*{(45)}
mtol=∑e∈Eme.(46)m_{\mathrm{tol}}=\sum_{e\in E}m_e. \tag*{(46)}

The final μ\mu value is then obtained by normalizing (μρ)tol\left(\frac{\mu}{\rho}\right)_{\mathrm{tol}} by mtolm_{\mathrm{tol}} and scaling by the sample mass density ρs\rho_s:

μ=ρs⋅(μρ)tolmtol.(47)\mu= \rho_s \cdot\frac{\left(\frac{\mu}{\rho}\right)_{\mathrm{tol}}}{m_{\mathrm{tol}}}. \tag*{(47)}

Selecting μDe\mu D_e in different situations

In the main paper we assessed the effect of different cvec_{ve} corrections on refined structural parameters from models. Here, for completeness, we present more complete results of this study by taking the optimal μD\mu D corrections but applying them to data from different thickness CeO2_2 samples with different degrees of absorption.

In Table S3 we show the refined parameters for synchrotron, uncorrected and best-corrected data for all CeO2_2 datasets, with additional μD\mu D’s tested listed in Table S11, Table S12, and Table S13. We find that μDe\mu D_e depends on different experimental conditions, specifically the relative x-ray beam height hh and capillary diameter DD. For data collected with D≈0.8∗hD \approx0.8\ast h (IDs 1.024 mm and 0.813 mm here, with h=1.2h = 1.2 mm and 1.0 mm, respectively), the μDe\mu D_e’s, 6.5 and 4, are close to their theoretical values, 5.824 and 4.554, respectively. For data with D≪hD \ll h (ID = 0.635 mm with h=1h = 1 mm), μDe=1.938\mu D_e = 1.938 which is much smaller than μDth=4.229\mu D_{th} = 4.229. Notably, this value for μDe\mu D_e was obtained from the zz-scan absorption measurement that was made with the slit and detector in the experimental condition.

Parameter1.024 mm0.813 mm0.635 mm1.024 mm corrected0.813 mm corrected0.635 mm correctedsynchrotron
s1s_10.2191(9)0.3060(12)0.2923(9)0.3711(11)0.4199(13)0.4331(11)0.37434(21)
QdampQ_{\mathrm{damp}}0.03186(25)0.02971(23)0.02839(19)0.02448(24)0.02650(21)0.02655(17)0.02386(4)
QbroadQ_{\mathrm{broad}}0.0220(5)0.0241(4)0.02405(34)0.0373(4)0.0308(4)0.02841(31)0.01814(7)
δ2\delta_29.92(28)9.70(17)9.68(12)7.14(11)7.18(9)7.18(7)9.07(4)
aa5.40240(5)5.40233(5)5.40398(4)5.39937(6)5.40071(5)5.40291(4)5.414232(7)
Ce(UisoU_{\mathrm{iso}})0.002403(24)0.002751(24)0.003098(21)0.003388(31)0.003474(28)0.003629(23)0.003571(5)
O(UisoU_{\mathrm{iso}})0.0990(24)0.0765(14)0.0670(9)0.0405(5)0.0500(6)0.0542(6)0.04120(10)
Qmax⁡Q_{\max}16.616.616.616.616.616.630.0
grid0.1892530.1892530.1892530.1892530.1892530.1892530.10472
RwR_w0.7189910.5813250.5079610.3909740.3548270.341720.160823
χred2\chi^2_{\mathrm{red}}1173.593401758.117681864.835553390.094422330.294145434.106299961.528958

Table S3. Results of fittings for CeO2_2 synchrotron, uncorrected, and best-corrected data, over rmin⁡=1.0r_{\min}=1.0 and rmax⁡=40.0r_{\max}=40.0. The uncorrected data are listed with their IDs (1.024 mm, 0.813 mm, and 0.635 mm), and the best-corrected data are labeled with the ID followed by “corrected”.

ParametersynchrotronuncorrectedμD=1.912\mu_D = 1.912μD=1.971\mu_D = 1.971μD=2.005\mu_D = 2.005μD=3.054\mu_D = 3.054μD=3.489\mu_D = 3.489μD=3.666\mu_D = 3.666μD=3.741\mu_D = 3.741
s1s_10.37434(21)0.2923(9)0.4288(11)0.4305(11)0.4314(11)0.4534(10)0.4547(10)0.4540(10)0.4539(10)
QdampQ_{\mathrm{damp}}0.02386(4)0.02839(19)0.02627(18)0.02621(18)0.02618(17)0.02525(17)0.02463(16)0.02444(16)0.02435(16)
QbroadQ_{\mathrm{broad}}0.01814(7)0.02404(34)0.02720(32)0.02725(32)0.02728(32)0.02813(29)0.02782(27)0.02769(27)0.02759(27)
δ2\delta_29.07(4)9.67(12)8.26(9)8.22(9)8.20(9)7.15(7)7.14(6)7.14(6)7.13(6)
aa5.414232(7)5.40398(4)5.40301(4)5.40298(4)5.40297(4)5.40255(4)5.40242(4)5.40239(4)5.40237(4)
Ce(Uiso)\mathrm{Ce}(U_{\mathrm{iso}})0.003571(5)0.003099(21)0.003712(24)0.003735(24)0.003749(24)0.004317(26)0.004734(27)0.004878(27)0.004959(27)
O(Uiso)\mathrm{O}(U_{\mathrm{iso}})0.04120(10)0.0670(9)0.0560(6)0.0558(6)0.0556(6)0.0519(5)0.0516(5)0.0517(4)0.0517(4)
Qmax⁡Q_{\max}30.016.616.616.616.616.616.616.616.6
RgridR_{\mathrm{grid}}0.104720.1892530.1892530.1892530.1892530.1892530.1892530.1892530.189253
ww0.1608230.5079610.3443610.3405310.3383420.2700470.2440280.2370340.233655
χred2\chi_{\mathrm{red}}^2961.528958864.835556439.844024432.38352428.156929309.2045271.704349262.39813258.344267

Table S11. Results of the refinement for CeO2\mathrm{CeO}_2 data (ID=0.635 mm) for synchrotron, uncorrected, and corrected data with different μD\mu_D’s, over rmin⁡=1.0r_{\min} = 1.0 and rmax⁡=40.0r_{\max} = 40.0. The μD\mu_D’s are the values used for each correction.

Table S12: Results of the refinement for CeO2_2 data (ID=0.813 mm) for synchrotron, uncorrected, and corrected data with different μD\mu_D’s, over rmin⁡=1.0r_{\min}=1.0 and rmax⁡=40.0r_{\max}=40.0. The μD\mu_D’s are the values used for each correction.

ParametersynchrotronuncorrectedμD=2.15\mu_D=2.15μD=2.5\mu_D=2.5μD=3\mu_D=3μD=3.5\mu_D=3.5μD=4\mu_D=4μD=4.554\mu_D=4.554μD=5\mu_D=5
s1s_10.37434(21)0.3060(12)0.3562(13)0.3708(13)0.3902(13)0.4065(13)0.4173(13)0.4269(12)0.4234(12)
QdampQ_{\mathrm{damp}}0.02386(4)0.02971(23)0.02875(22)0.02842(22)0.02785(22)0.02713(21)0.02628(21)0.02513(21)0.02438(21)
QbroadQ_{\mathrm{broad}}0.01814(7)0.0241(4)0.0261(4)0.0266(4)0.0276(4)0.0287(4)0.0300(4)0.0316(4)0.0322(4)
δ2\delta^29.07(4)9.69(17)9.04(14)8.86(13)8.56(13)8.22(12)7.83(11)7.36(11)7.15(9)
aa5.414232(7)5.40233(5)5.40187(5)5.40168(5)5.40141(5)5.40110(5)5.40079(5)5.40023(5)5.40019(5)
Ce(Uiso)\mathrm{Ce}(U_{\mathrm{iso}})0.003571(5)0.002751(24)0.002970(26)0.003052(26)0.003188(27)0.003351(28)0.003533(29)0.003760(31)0.004033(31)
O(Uiso)\mathrm{O}(U_{\mathrm{iso}})0.04120(10)0.0765(14)0.0650(10)0.0622(10)0.0582(8)0.0544(7)0.0511(7)0.0479(6)0.0464(5)
Qmax⁡Q_{\max}30.016.616.616.616.616.616.616.616.6
RgridR_{\mathrm{grid}}0.104720.1892530.1892530.1892530.1892530.1892530.1892530.1892530.189253
ww0.1608230.5813250.4934630.4678030.4297340.3910460.3558440.3225140.298757
χred2\chi_{\mathrm{red}}^2961.528958758.117673554.991541506.666438441.041522381.046435332.191765293.028937265.892271

Table S12.

ParametersynchrotronuncorrectedμD=3\mu_D = 3μD=5.328\mu_D = 5.328μD=5.824\mu_D = 5.824μD=6\mu_D = 6μD=6.5\mu_D = 6.5
s1s_10.37434(21)0.2191(9)0.2972(12)0.3693(12)0.3785(12)0.3748(12)0.3712(11)
QdampQ_\mathrm{damp}0.02386(4)0.03186(25)0.03074(24)0.02737(23)0.02612(23)0.02593(23)0.02449(24)
QbroadQ_\mathrm{broad}0.01814(7)0.0220(5)0.0251(5)0.0320(4)0.0347(4)0.0353(4)0.0372(4)
δ2\delta_29.07(4)9.92(28)9.21(17)7.87(13)7.42(13)7.15(11)7.14(11)
aa5.414232(7)5.40240(5)5.40160(5)5.40018(5)5.39959(6)5.39967(6)5.39937(6)
Ce(Uiso)\mathrm{Ce}(U_\mathrm{iso})0.003571(5)0.002403(24)0.002671(25)0.003154(29)0.003216(30)0.003265(29)0.003389(31)
O(Uiso)O(U_\mathrm{iso})0.04120(10)0.0990(24)0.0696(13)0.0472(7)0.0435(6)0.0425(6)0.0405(5)
QmaxQ_\mathrm{max}30.016.616.616.616.616.616.6
RgridR_\mathrm{grid}0.104720.1892530.1892530.1892530.1892530.1892530.189253
ww0.1608230.7189910.5964040.4391310.4172040.4079560.390974
χred2\chi_\mathrm{red}^{2}961.5289581173.593401753.497185447.645936419.475451407.497971390.094511

Table S13. Results of the refinement for CeO2_2 data (ID=1.024 mm) for synchrotron, uncorrected, and corrected data with different μD\mu_D’s, over rmin=1.0r_\mathrm{min} = 1.0 and rmax=40.0r_\mathrm{max} = 40.0. The μD\mu_D’s are the values used for each correction.

If it is possible to measure sample absorption using a zz-scan and using the fitting program in our DIFFPY.LABPDFPROC software to extract μD\mu D, it is recommended to do this with the slits for the zz-scan in the same settings as was used to collect the experimental data. However, if it is not possible, μDe\mu D_e can be approximated by scaling the theoretical value μDth\mu D_{th} by D/hD/h to get an approximately correct value for the effective μD\mu D to use in the cvec_{ve} calculation. Future investigations with more data will be necessary to fully prove this equation.

Refinement across extended μD\mu D values

For a low μD<2\mu D < 2, the uncorrected lab data already provides a reasonably good fit (low RwR_w). This is demonstrated in Table S4 for ZrO2_2 packed in a 1.024 mm capillary, where the theoretical μD\mu D is ∼1.5\sim1.5. We have applied two corrections, corresponding to the largest and smallest experimental μD\mu D’s computed from Method 2, using h=0.05h=0.05 mm and 1.2 mm, respectively, with a reduced channel. To avoid overfitting, isotropic ADPs were used for zirconium and oxygen.

ParametersynchrotronuncorrectedμD=1.127\mu D = 1.127μD=1.517\mu D = 1.517
s1s_10.49519(28)0.5314(27)0.5771(28)0.6077(29)
QdampQ_{\mathrm{damp}}0.02282(5)0.0201(5)0.0218(4)0.0215(4)
QbroadQ_{\mathrm{broad}}0.03182(13)0.0329(9)0.0298(8)0.0294(8)
δ2\delta_21.294(17)1.94(10)1.85(9)1.80(9)
aa5.148597(23)5.14610(30)5.14606(29)5.14622(28)
bb5.211724(23)5.20949(34)5.20991(32)5.21009(31)
cc5.317103(24)5.31141(35)5.31192(34)5.31218(33)
β\beta99.2208(4)99.205(6)99.201(6)99.201(5)
Zr(UisoU_{\mathrm{iso}})0.0010703(25)0.00451(8)0.00507(9)0.00532(9)
O(UisoU_{\mathrm{iso}})0.03721(12)0.0397(10)0.0438(10)0.0445(10)
Qmax⁡Q_{\max}30.016.616.616.6
grid0.104720.1892530.1892530.189253
RwR_w0.2187420.2240710.1970610.191054
χred2\chi^2_{\mathrm{red}}1298.81348137.00162432.54691233.036836

Table S4. Results of fittings for ZrO2_2 data (ID=1.024 mm) for synchrotron, uncorrected, and corrected lab data, over rmin⁡=1.0r_{\min} = 1.0 and rmax⁡=40.0r_{\max} = 40.0. The μD\mu D’s are the values applied for correction.

Overall, we find that the uncorrected lab data provides results comparable to the synchrotron data. As with the ceria data described in the main paper, the lattice parameters of the synchrotron data are slightly overestimated. In ceria, the ADP on the metal site was most important. Here the difference in scattering power between oxygen and Zr is smaller and the oxygen ADPs are more significant. We find that the ADPs of both the metal and oxygen increase with increasing absorption correction as expected following the arguments made in the main paper. However, even for the uncorrected ZrO2_2, the ADPs of the lab data are higher than those from the synchrotron data. We are not sure why this is the case. The refinements are quite reproducible between the synchrotron and lab data, considering the significant differences between the resolutions and QQ-ranges of the different measurements, but do not allow a clear differentiation between the different absorption corrections.

The PDFs of the synchrotron and laboratory data are plotted in Figure S5. The most significant difference is that the lab data have slightly broader peaks compared to the synchrotron data due to the lower Qmax⁡Q_{\max}, but all the fits are of high quality.

Figure S5. Results of ZrO2_2 synchrotron, uncorrected, and corrected laboratory data over rmin⁡=1.0r_{\min}=1.0 and rmax⁡=40.0r_{\max}=40.0. The μD\mu D’s reported inside the panels are the values used for correction. The measured PDFs are the blue circles, the best-fit PDFs are the red lines, and the residuals shown are green.

We now turn to the case of very high μD\mu D by considering data from the HfO2_2 sample. As discussed in the main paper, the significant absorption of the sample results in a strong suppression of the diffraction signal, resulting in a very low signal/noise ratio. This is evident by looking at the dark blue curve in Figure S6, which shows the XRD data for a measurement of HfO2_2 using a capillary of 0.635 mm, where the theoretical μD\mu D is just above eight. As shown in the inset, only a few weak signals are observed, and the curve remains relatively flat until the end, where there is a strong upturn. While the absorption correction would adjust signal amplitudes, it amplifies both the signals and the noise by the same amount, leaving the signal/noise ratio unaffected. As a result, the data is functionally useless for PDF analysis.

Figure S6. XRD, PDF, and zz-scan analysis of HfO2\mathrm{HfO_2} data. (a) XRD comparison of HfO2\mathrm{HfO_2} measured with a capillary (dark blue) and wire (light blue), with an inset showing an expanded portion of the capillary data highlighting its only signals. (b) PDF fit for wire data, showing the measured PDF (blue), the best-fit PDF (red), and residuals (green), fitted between rmin⁡=1.0r_{\min}=1.0 and rmax⁡=20.0r_{\max}=20.0. (c) zz-scan fit for wire data, with the original intensity (blue), the fit (red), the unconvoluted intensity (brown), and residuals (green).

It is possible to load the sample into a thinner capillary if one is available. However, a simple workaround is to take a glass or plastic wire and rub some grease and powder on the outside of the wire (see Figure S7). This approach is very effective if the powder scatters well enough, which is typically the case for elements with large x-ray cross-sections.

Photograph of the wire coated with the powder of HfO₂

Figure S7. Photograph of the wire coated with the powder of HfO₂. The wire was coated with some grease, then, the hafnia powder was dispersed over the coated area.

In Figure S6, the XRD data for the wire is shown in light blue. Comparing it to the dark blue curve (representing HfO2_2 in the 0.635 mm capillary), the wire data shows a significant improvement in the signal/noise ratio, with a reduced upturn at the end, indicating much better data quality. To evaluate the effectiveness further, Figure S6 shows the PDF obtained from the uncorrected wire diffraction pattern, fitted between rmin⁡=1.0r_{\min} = 1.0 and rmax⁡=20.0r_{\max} = 20.0 Å. The refined parameters are provided in Table S8. The small residuals of the fit confirm the high data quality, making the wire data the preferred choice.

ParameterUncorrected
s1s_10.667(13)
QdampQ_{\mathrm{damp}}0.0480(22)
QbroadQ_{\mathrm{broad}}0.0(5)
δ2\delta_26.47(24)
aa5.1106(11)
bb5.1780(15)
cc5.2932(10)
β\beta99.544(23)
U11,0U_{11,0}0.0094(4)
U22,0U_{22,0}0.0091(5)
U33,0U_{33,0}0.00808(32)
U12,0U_{12,0}0.0092(4)
U13,0U_{13,0}0.0088(4)
U23,0U_{23,0}0.0087(4)
U11,4U_{11,4}0.0083(25)
U22,4U_{22,4}0.0101(26)
U33,4U_{33,4}0.0113(25)
U12,4U_{12,4}0.0094(25)
U13,4U_{13,4}0.0099(25)
U23,4U_{23,4}0.0106(25)
U11,8U_{11,8}0.0109(28)
U22,8U_{22,8}0.0096(25)
U33,8U_{33,8}0.0100(27)
U12,8U_{12,8}0.0104(26)
U13,8U_{13,8}0.0103(27)
U23,8U_{23,8}0.0104(26)
x0x_00.2747(4)
y0y_00.46602(31)
z0z_00.7046(4)
x4x_40.1041(19)
y4y_40.2126(19)
z4z_40.8875(19)
x8x_80.4889(21)
y8y_80.7553(18)
z8z_80.9863(22)
Qmax⁡Q_{\max}16.6
grid0.189253
RwR_w0.366525
χred2\chi^2_{\mathrm{red}}41.722268

Table S8. Results of the refinement for uncorrected HfO2_2 wire data over rmin⁡=1.0r_{\min}=1.0 and rmax⁡=20.0r_{\max}=20.0.

Finally, we present a method to estimate the theoretical μD\mu D for wire data, which is useful for determining the effective μD\mu D to use to apply for the absorption correction. Direct theoretical estimation is impossible since we are using a wire instead of a capillary, meaning that we do not know the powder density and exact diameter of the wire and powder together. As a result, we can use a zz-scan to obtain experimental values, using, for instance, a small x-ray beam height of 0.1 mm with a reduced channel. This returns an experimental μD\mu D value close to the theoretical one. Figure S6 shows the zz-scan fit using Method 2 with the experimental settings mentioned above. Although the fitted red curve shows some deviation from the original blue intensity data, making it less accurate, we find it still reliable, as verified through both Method 1 (μDm1=0.777\mu D_{m1} = 0.777) and Method 2 (μDm2=0.691\mu D_{m2} = 0.691).

Additional Tables and Figures

SampleID (mm)Density (g/cm3g/cm^3)hh (mm)ChannelMethod 1Method 2Method 3Method 4
ZrO2\mathrm{ZrO}_20.6351.0090.10Closed0.6150.5380.5170.795
0.8130.8560.10Closed0.7110.5920.5920.864
1.0241.1220.05Closed1.5551.5171.5291.426
0.20Closed1.3991.3981.421
0.60Closed1.1201.1201.206
1.20Closed1.1421.1271.200
CeO2\mathrm{CeO}_20.6351.7060.05Closed3.7413.6613.6244.229
0.10Closed3.7143.6663.671
0.20Closed3.5193.4893.519
0.60Closed3.1083.1083.430
1.00Closed3.0553.0543.384
0.05Open3.2803.2453.208
0.20Open3.0013.0113.001
0.60Open1.9712.0052.098
1.00Open1.9121.9382.016
0.8131.4350.10Closed4.5434.4654.5094.554
1.0241.4570.10Closed5.4195.3285.3915.824
HfO2\mathrm{HfO}_20.6351.741–––––8.168
0.8131.9630.10Open5.8945.5755.57011.79
Wire–0.10Closed0.7770.691––

Table S9. List of μD\mu D’s computed using the four different methods described in the texts, labeled as Method 1, Method 2, Method 3, and Method 4.

SampleDgivenD_{\mathrm{given}} (mm)Density (g/cm3g/cm^3)hgivenh_{\mathrm{given}} (mm)Channelμtheoretical\mu_{\mathrm{theoretical}} (mm−1\mathrm{mm}^{-1})μfit\mu_{\mathrm{fit}} (mm−1\mathrm{mm}^{-1})DfitD_{\mathrm{fit}} (mm)hfith_{\mathrm{fit}} (mm)
ZrO2_20.6351.0090.10Closed1.2520.8140.6600.095
0.8130.8560.10Closed1.0620.7280.8130.042
1.0241.1220.05Closed1.3921.4931.0160.067
0.20Closed1.3871.0080.167
0.60Closed1.1840.9480.290
1.20Closed1.1720.9620.295
CeO2_20.6351.7060.05Closed6.6595.7070.6420.073
0.10Closed5.7810.6340.075
0.20Closed5.5420.6300.148
0.60Closed5.4010.5750.276
1.00Closed5.3300.5830.280
0.05Open5.0520.6420.092
0.20Open4.7250.6370.176
0.60Open3.3040.6070.322
1.00Open3.1740.6100.332
0.8131.4350.10Closed5.6015.5460.8050.672
1.0241.4570.10Closed5.6875.2651.0120.073
HfO2_20.6351.741––12.86–––
0.8131.9630.10Open14.506.8510.8140.083
Wire–0.10Closed–2.8040.2470.104

Table S10. List of fitted and theoretical or given physical parameters μ\mu, DD, and hh. μtheoretical\mu_{\mathrm{theoretical}} is the theoretical μ\mu computed from XrayDB, DgivenD_{\mathrm{given}} is the given inner diameter (ID) of each capillary, hgivenh_{\mathrm{given}} is the given x-ray beam height. μfit\mu_{\mathrm{fit}}, DfitD_{\mathrm{fit}}, and hfith_{\mathrm{fit}} are the fitted parameters computed from Method 2 for each zz-scan data.

References

  1. [1]Chupas, P. J.; Grey, C. P.; Hanson, J. C.; Kim, J.-Y.; Rodriguez, J.; Qiu, X.; Billinge, S. J. L.; Lee, P. L. In-Situ Time Resolved Powder Diffraction Studies in Heterogenous Catalysis; Coupling the Study of Long Range and Local Structural Changes. Commission on Powder Diffraction Newsletter, International Union of Crystallography 2003, 29, 24–25.
  2. [2]Confalonieri, G.; Dapiaggi, M.; Sommariva, M.; Gateshki, M.; Fitch, A. N.; Bernasconi, A. Comparison of Total Scattering Data from Various Sources: The Case of a Nanometric Spinel. Powder Diffraction 2015, 30, S65–S69.
  3. [3]Prinz, N.; Schwensow, L.; Wendholt, S.; Jentys, A.; Bauer, M.; Kleist, W.; Zobel, M. Hard X-ray-based Techniques for Structural Investigations of CO₂ Methanation Catalysts Prepared by MOF Decomposition. Nanoscale 2020, 12, 15800–15813.DOI
  4. [4]Schertenleib, T.; Schmückler, D.; Chen, Y.; Jin, G. B.; Queen, W. L.; Billinge, S. J. L. Protocols for Obtaining Reliable PDFs from Laboratory X-Ray Sources Using PDFgetX3. 2024.arxiv.org/abs/2406.18177
  5. [5]Egami, T.; Billinge, S. J. L. Underneath the Bragg Peaks: Structural Analysis of Complex Materials, 2nd ed.; Pergamon Materials Series 16; Elsevier: Amsterdam, 2012.
  6. [6]Pitschke, W.; Mattern, N.; Hermann, H. Incorporation of Microabsorption Corrections into Rietveld Analysis. Powder Diffraction 1993, 8, 223–228.DOI
  7. [7]Dinnebier, R. E.; Leineweber, A.; Evans, J. S. O. Rietveld Refinement: Practical Powder Diffraction Pattern Analysis Using TOPAS; Walter de Gruyter GmbH & Co KG, 2018.
  8. [8]Brindley, G. XLV. The Effect of Grain or Particle Size on x-Ray Reflections from Mixed Powders and Alloys, Considered in Relation to the Quantitative Determination of Crystalline Substances by x-Ray Methods. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science 1945, 36, 347–369.DOI
  9. [9]Taylor, J. C.; Matulis, C. E. Absorption Contrast Effects in the Quantitative XRD Analysis of Powders by Full Multiphase Profile Refinement. Journal of Applied Crystallography 1991, 24, 14–17.DOI
  10. [10]Lipp, J.; Banerjee, R.; Patwary, M. F.; Patra, N.; Dong, A.; Girgsdies, F.; Bare, S. R.; Regalbuto, J. R. Extension of Rietveld Refinement for Benchtop Powder XRD Analysis of Ultrasmall Supported Nanoparticles. Chemistry of Materials 2022, 34, 8091–8111.DOI
  11. [11]Petkov, V. RAD , a Program for Analysis of X-ray Diffraction Data from Amorphous Materials for Personal Computers. Journal of Applied Crystallography 1989, 22, 387–389.
  12. [12]Jeong, I.-K.; Thompson, J.; Turner, A. M. P.; Billinge, S. J. L. PDFgetX: A Program for Determining the Atomic Pair Distribution Function from X-ray Powder Diffraction Data. J. Appl. Crystallogr. 2001, 34, 536.DOI
  13. [13]Qiu, X.; Thompson, J. W.; Billinge, S. J. L. PDFgetX2: A GUI-driven Program to Obtain the Pair Distribution Function from X-ray Powder Diffraction Data. Journal of Applied Crystallography 2004, 37, 678–678.DOI
  14. [14]Soper, A. K. GudrunN and GudrunX: Programs for Correcting Raw Neutron and x-Ray Diffraction Data to Differential Scattering Cross Section; Science & Technology Facilities Council: Swindon, UK, 2011.
  15. [15]Juhas, P.; Davis, T.; Farrow, C. L.; Billinge, S. J. L. PDFgetX3: A Rapid and Highly Automatable Program for Processing Powder Diffraction Data into Total Scattering Pair Distribution Functions. Journal of Applied Crystallography 2013, 46, 560–566.DOI
  16. [16]Billinge, S. J. L.; Farrow, C. L. Towards a Robust Ad Hoc Data Correction Approach That Yields Reliable Atomic Pair Distribution Functions from Powder Diffraction Data. Journal of Physics: Condensed Matter 2013, 25, 454202.DOI
  17. [17]Paalman, H. H.; Pings, C. J. Numerical Evaluation of X-Ray Absorption Factors for Cylindrical Samples and Annular Sample Cells. Journal of Applied Physics 1962, 33, 2635–2639.DOI
  18. [18]Kendig, A. P.; Pings, C. J. X-Ray Absorption Factors for Cylindrical Samples in Annular Sample Cells Exposed to Incident Beams of Limited Width. Journal of Applied Physics 1965, 36, 1692–1698.DOI
  19. [19]Poncett, P. Inelastic Scattering from a Liquid Sample Confined within a Cylindrical Container: Substraction and Attenuation Corrections; Internal Report 77P0139S, 1977.
  20. [20]Soper, A. K.; Egelstaff, P. Multiple Scattering and Attenuation of Neutrons in Concentric Cylinders: I. Isotropic First Scattering. Nuclear Instruments and Methods 1980, 178, 415–425.DOI
  21. [21]Toby, B. H.; Von Dreele, R. B. GSAS-II: The Genesis of a Modern Open-Source All Purpose Crystallography Software Package. Journal of Applied Crystallography 2013, 46, 544–549.DOI
  22. [22]Soper, A. K.; Barney, E. R. Extracting the Pair Distribution Function from White-Beam X-ray Total Scattering Data. Journal of Applied Crystallography 2011, 44, 714–726.DOI
  23. [23]Newville, M.; easyXAFS; Levantino, M.; Schlepuetz, C.; Guenzing, D.; Rak- itin, M.; Kim, S.-W.; kalvdans X-rayDB: X-ray Reference Data in SQLite. https://github.com/xraypy/XrayDB, 2024.
  24. [24]Ashiotis, G.; Deschilibre, A.; Nawaz, Z.; Wright, J. P.; Karkoulis, D.; Picca, F. E.; Kieffer, J. The Fast Azimuthal Integration Python Library: pyFAI. Journal of Applied Crystallography 2015, 48, 510–519.pubmed.ncbi.nlm.nih.gov/25844080
  25. [25]Juhás, P.; Farrow, C. L.; Yang, X.; Knox, K. R.; Billinge, S. J. L. Complex Modeling: A Strategy and Software Program for Combining Multiple Information Sources to Solve Ill Posed Structure and Nanostructure Inverse Problems. Acta Crystallographica Section A Foundations and Advances 2015, 71, 562–568.DOI
  26. [26]Bondars, B.; Heidemanе, G.; Grabis, J.; Laschke, K.; Boysen, H.; Schneider, J.; Frey, F. Powder Diffraction Investigations of Plasma Sprayed Zirconia. Journal of Materials Science 1995, 30, 1621–1625.DOI
  27. [27]Scavini, M.; Coduri, M.; Allieta, M.; Brunelli, M.; Ferrero, C. Probing Complex Disorder in Ce1−xGdxO2−x/2 Using the Pair Distribution Function Analysis. Chemistry of Materials 2012, 24, 1338–1345.DOI
  28. [28]Hann, R. E.; Suitch, P. R.; Pentecost, J. L. Monoclinic Crystal Structures of ZrO2 and HfO2 Refined from X-ray Powder Diffraction Data. Journal of the American Ceramic Society 1985, 68, C–285–C–286.
  29. [29]Toby, B. H.; Billinge, S. J. L. Determination of Standard Uncertainties in Fits to Pair Distribution Functions. Acta Crystallographica Section A Foundations of Crystallography 2004, 60, 315–317.pubmed.ncbi.nlm.nih.gov/15218209
  30. [30]Von Dreele, R. B.; Suchomel, M. R.; Toby, B. H. Compute X-ray Absorption. https://11bm.xray.aps.anl.gov/absorb/absorb.php, 2024.
  31. [31]Hubbell, J.; Seltzer, S. Tables of X-Ray Mass Attenuation Coefficients and Mass Energy-Absorption Coefficients, NIST Standard Reference Database 126. https://dx.doi.org/10.18434/T4D01F, 2004.
  32. [32]Farrow, C. L.; Juhás, P.; Yang, A.; Billinge, S. J. L.; Yang, X.; Knox, K.; Calamari, J.; Frandsen, B.; Yang, L.; Lee, S.; Chen, Y.; Myers, C.; Zhang, T. diffpy.utils. https://github.com/diffpy/diffpy.utils, 2024.
  33. [33]DiffPy - Atomic Structure Analysis in Python. https://github.com/diffpy.
  34. [34]Python Package Index (PyPI). https://pypi.org/.
  35. [35]Conda-Forge. https://conda-forge.org/.
  36. [36]diffpy.labpdfproc. https://github.com/diffpy/diffpy.labpdfproc, 2024.
  37. [37]diffpy.labpdfproc Documentation. https://www.diffpy.org/diffpy.labpdfproc/, 2024.

Paper details

vlnc:W_bvhen3zz
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents