{"id":860312,"date":"2026-06-11T17:19:34","date_gmt":"2026-06-11T17:19:34","guid":{"rendered":"https:\/\/www.europesays.com\/us\/860312\/"},"modified":"2026-06-11T17:19:34","modified_gmt":"2026-06-11T17:19:34","slug":"electron-ptychography-reveals-correlated-lattice-vibrations-at-atomic-resolution","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/us\/860312\/","title":{"rendered":"Electron ptychography reveals correlated lattice vibrations at atomic resolution"},"content":{"rendered":"

The first step of CAVIAR is a gradient-based ptychographic reconstruction5<\/a>,10<\/a>,20<\/a>,25<\/a>,26<\/a>,27<\/a>,28<\/a> which fits a transmission function of the specimen by determining the most likely29<\/a> parameters of a numerical model for a given set of experimentally measured intensities. Various aspects of the numerical model are depicted in Fig.\u00a01<\/a> and a detailed mathematical description can be found in the Supplementary Eqs.\u00a01<\/a>\u20137<\/a>. Upon convergence of the fit, one obtains a discrete complex transmission function of the specimen, which in our case has four dimensions: two lateral coordinates (x,y) spanning a plane perpendicular to the beam direction, an out-of-plane coordinate (z), parallel to the beam; and one dimension accommodating incoherent object modes (n). The phase of the transmission function O(x,\u00a0y,\u00a0z,\u00a0n) is directly proportional to the projected electrostatic potential in slice z10<\/a> of the n\u2019th configuration of the object. The second step of CAVIAR is an extraction of the interatomic correlations from an unordered sequence of the retrieved object states n. To do so, we adopt a framework developed for the calculation of phonon dispersion curves from MD simulations18<\/a>. The initial determination of atomic positions can be achieved either manually or, in more complex scenarios, using an object detection network (e.g., ref. 30<\/a>). Subsequently, refining the atomic coordinates can be accomplished by calculating the center of mass within a small circular region around the initially determined atomic positions, using the retrieved phase values as weighting factors. In this way, for each unit cell l, basis atom \u03ba, direction of displacement \u03b1 (x,y,z) and reconstructed object state n one can get real-space atomic positions Rl\u03ba\u03b1n and displacements from the mean position18<\/a>: <\/p>\n

$${u}_{l\\kappa \\alpha n}={R}_{l\\kappa \\alpha n}-\\frac{1}{N}{\\sum }_{n=1}^{N}{R}_{l\\kappa \\alpha n}.$$<\/p>\n

\n (1)\n <\/p>\n

Afterwards, the lattice Green\u2019s function coefficient \\({G}_{l\\kappa \\alpha {l}^{{\\prime} }{\\kappa }^{{\\prime} }{\\alpha }^{{\\prime} }}\\) describing the correlations can be computed as a second moment of the displacements18<\/a>,31<\/a>: <\/p>\n

$${G}_{l\\kappa \\alpha {l}^{{\\prime} }{\\kappa }^{{\\prime} }{\\alpha }^{{\\prime} }}=\\frac{1}{N}{\\sum }_{n=1}^{N}{u}_{l\\kappa \\alpha n}{u}_{{l}^{{\\prime} }{\\kappa }^{{\\prime} }{\\alpha }^{{\\prime} }n}.$$<\/p>\n

\n (2)\n <\/p>\n

Simulated data<\/p>\n

To investigate the feasibility of CAVIAR, we started with simulated data to have well-controlled conditions and the ability to compare with the ground truth, before considering experimental data. As a test sample, we chose a symmetric \u03a39 grain boundary in silicon, MD simulations of phonon spectra of which32<\/a> agree perfectly well with experimental STEM-EELS data24<\/a>. Moreover, prior work has shown that atoms at the grain boundary exhibit distinct vibrational behavior compared to those in bulk regions due to differences in bonding environments24<\/a>. CAVIAR enables this phenomenon to be examined at the level of individual atoms.<\/p>\n

To generate a 4D-STEM dataset33<\/a> we used 30 time-snapshots (object configurations) of a 0.7\u2009nm thick MD super-cell containing four atomic layers. The accelerating voltage, convergence semi-angle, and scan-step were 200\u2009kV, 30\u2009mrad, and 50\u2009pm, respectively. Averaging neighboring diffraction patterns with a Gaussian weighting was used to simulate the effect of partial spatial coherence with an effective source size (full width at half maximum of the Gaussian) of 47\u2009pm. In total we generated four 4D-STEM datasets mimicking different degrees of incoherent scattering: (1) a perfectly coherent dataset based on a single MD snapshot, i.e., without TDS and without additional partial spatial coherence; (2) a dataset based on the Einstein model of thermal vibration17<\/a>,34<\/a>,35<\/a>, i.e., uncorrelated displacements sampled from a normal distribution with standard deviation matched to the average one from the MD data and without additional partial spatial coherence; (3) a dataset based on MD data including correlated atomic movement without partial spatial coherence; (4) a dataset based on MD data with additional partial spatial coherence. Allowing for partial spatial coherence in the 4D-STEM data is important due to the fact that averaging over multiple slightly displaced probes is conceptually equivalent to averaging over multiple slightly displaced objects. Thus, when the single probe mode is used in a reconstruction, partial spatial coherence, absorbed by the object states, introduces a positive offset to the x-x and y-y correlations of the atomic coordinates. The full Green\u2019s tensors in Supplementary Fig.\u00a02<\/a> reflect this effect, showing a uniform offset in the x-x and y-y correlations in the single probe mode reconstructions from the datasets corrupted by partial spatial coherence. Mitigation of this effect can be achieved in two ways: (1) the positive offset can be corrected in the recovered correlations, for example, in the same manner as background subtraction is performed in EELS spectra24<\/a>; (2) A ptychographic fit with single object mode and multiple probe modes10<\/a>,14<\/a> can be performed prior to mixed-object and mixed-probe reconstruction in order to fit multiple probe states that will absorb the offset. In the reconstructions presented further we chose the second approach. All ptychographic fits from the Si grain boundary data were performed using 20 object states and 4 slices with a spacing of \u0394z\u00a0=\u00a00.19\u2009nm. In Fig.\u00a02<\/a> we show both the reconstructed projected phase as well as the retrieved interatomic correlations. The reconstructions from data without additional partial spatial coherence were done with a single probe mode and the reconstruction in Fig.\u00a02<\/a>e was done with 9 probe modes.<\/p>\n

Fig. 2: Ptychographic reconstructions from realistically simulated data for various kinds of atomic motion and partial spatial coherence (PSC).<\/b>\"Fig.The alternative text for this image may have been generated using AI.<\/p>\n

a<\/b> Ground truth phase image from an MD simulation of a \u03a39 grain boundary in silicon with arrows indicating various components of the Green\u2019s tensor defined in eq. (2<\/a>). A randomly picked reference atom is marked by a red circle in (a<\/b>\u2013e<\/b>), its selected neighbors and the grain boundary are shown by a blue and magenta dashed lines in (b<\/b>), respectively. Colors of arrows show different correlation types: yellow\/purple (self-correlations), pink\/light blue (correlation with reference atom), two 2\u00a0\u00d7\u00a02 blocks of the full Green\u2019s tensor are shown schematically in the middle of the top row using the same colors. b<\/b>\u2013e<\/b> Reconstructed phases and correlation matrices obtained from data simulated using different levels of approximation: b<\/b> Perfectly coherent dataset, no PSC\u2014no correlations appear to have been reconstructed; c<\/b> Uncorrelated vibrations (Einstein model), no PSC\u2014correlations between neighbors largely vanish; d<\/b> Correlated vibrations, no PSC\u2014principal directions of correlations match the ground truth; e<\/b> Correlated vibrations, with PSC\u2014the recovered correlations match the ground truth and the offset created by PSC is efficiently mitigated by multiple probe modes; In the bottom of images b<\/b>\u2013e<\/b> we report values of length-similarity (LSM) and cosine-similarity (CSM) metrics (cf. eqs. (3<\/a>) and (4<\/a>)) between MD ground truth and retrieved correlation vectors. All reconstructions used 4 object slices, 20 object states. Reconstructions in (b<\/b>\u2013d)<\/b> were done with a single probe mode. In the reconstruction (e<\/b>) 9 probe modes were used.<\/p>\n

Although there is no fundamental bound prohibiting a full 3D reconstruction, the z-resolution currently achievable with STEM-ptychography20<\/a> is not sufficient to quantify the out-of-plane vibrations. Therefore, we limit our analysis to the in-plane components of vibrations. For a system of N atoms and two spatial coordinates, the retrieved Green\u2019s tensor depicting all correlations is a 2N\u00a0\u00d7\u00a02N matrix. In order to visualize it in a convenient form, we first randomly picked a reference atom, shown in Fig.\u00a02<\/a>b by the red circle, and some adjacent atoms circumscribed by the blue circles in the same image (for clarity of the figure only a few atoms were chosen). For a given pair of atoms we depict correlations as two vectors: the first one (coral) representing correlations of the reference atom\u2019s x coordinate with the neighbor\u2019s x and y coordinates and the second one (cyan) representing correlations between the reference atom\u2019s y coordinate with its neighbor\u2019s x and y coordinates. The yellow and purple arrows represent the atom\u2019s auto-correlation, which corresponds to the two contributions \\(\\langle {u}_{x}^{2}\\rangle\\) and \\(\\langle {u}_{y}^{2}\\rangle\\) of the equivalent isotropic displacement factor36<\/a>.<\/p>\n

Figure\u00a02<\/a>a shows the ground truth of the displacement correlations from the MD snapshots. One can see that the correlations retrieved via CAVIAR (Fig.\u00a02<\/a>d, e) are weaker than they should be, but the angles between the depicted arrows are nearly identical to the ground truth. Further, with our approach we can distinguish between various kinds of vibrations. In case of a coherent simulation (Fig.\u00a02<\/a>b) we cannot detect any displacements at all and for the simulation with uncorrelated atom displacements (Einstein model, Fig.\u00a02<\/a>c) we do not recover any correlations between neighbors.<\/p>\n

To quantitatively compare the correlations from the initial MD data and those extracted from the ptychographic reconstructions, we adopt two orthogonal metrics judging agreement in strength and direction: length similarity metric (LSM) and cosine-similarity metric (CSM)37<\/a>. For each atomic pair i, whose correlations are represented in Fig.\u00a02<\/a> by a vector Vi with two components (denoted j), we evaluate the similarity between the reconstructed (Rec) and ground truth (GT) correlations as follows: <\/p>\n

$$\\,{\\mbox{LSM}}\\,={\\sum }_{ij}\\sqrt{\\frac{| {V}_{ij}^{Rec}{| }^{2}}{| {V}_{ij}^{GT}{| }^{2}}}$$<\/p>\n

\n (3)\n <\/p>\n

$$\\,{\\mbox{CSM}}\\,=\\frac{1}{{N}_{vectors}}{\\sum }_{i=1}^{{N}_{vectors}}\\frac{{\\sum }_{j}{V}_{ij}^{GT}\\cdot {V}_{ij}^{Rec}}{\\sqrt{{\\sum }_{j}{\\left({V}_{ij}^{GT}\\right)}^{2}}\\cdot \\sqrt{{\\sum }_{j}{\\left({V}_{ij}^{Rec}\\right)}^{2}}}.$$<\/p>\n

\n (4)\n <\/p>\n

The CSM metric is bounded between \u22121 and 1, where 1 means perfect similarity. The LSM metric is bounded from below by zero, meaning the absence of vibrations in a reconstruction. LSM\u00a0=\u00a01 also means a perfect similarity. For the reconstruction from the dataset without partial spatial coherence (Fig.\u00a02<\/a>d) the CSM value is the highest (0.98), showing an almost perfect match. When adding partial spatial coherence (Fig.\u00a02<\/a>e), the mentioned positive offset in correlations reduces the CSM value only slightly to 0.96, indicating that multiple probe modes efficiently eliminate the ambiguity. For the coherent dataset and the one with uncorrelated TDS presented in Fig.\u00a02<\/a>b, c, the CSM values are substantially lower. The LSM metric allows to quantify the magnitude of retrieved vibrations, for a perfectly coherent dataset in Fig.\u00a02<\/a>b the value is close to zero, indicating the absence of atomic vibrations. Despite a good match in directions of correlations retrieved from correlated datasets, their magnitudes are approximately two times lower than the ground truth. Overall, we conclude that we may underestimate the strength of correlations due to the resolution in these reconstructions being limited, as well as due to a limited number of object states, but we can explore the directionality and relative differences between neighboring bonds.<\/p>\n

Experimental data<\/p>\n

After validating CAVIAR on simulated data we moved to an experimental 4D-STEM dataset of an approximately 15\u2009nm thick hexagonal boron nitride (hBN) sample in [0001] direction, whose upper and lower halves were twisted by approximately 11\u2218 relative to each other around said zone axis28<\/a>. The 4D-STEM dataset was acquired using a Nion HERMES microscope, operated at an accelerating voltage of 60\u2009kV, with a convergence semi-angle of 40\u2009mrad and a scan step of 30\u2009pm. The reconstructed transmission function included 10 object states and 30 slices with a spacing of 0.48\u2009nm. We further used 5 incoherent probe modes to account for partial spatial coherence of the beam and potentially occurring sample drift.<\/p>\n

In Supplementary Fig.\u00a01<\/a>, we demonstrate that even for a thin \u03a39 grain boundary in silicon, multiple scattering dominates over incoherence arising from sample vibrations. Consequently, successful extraction of interatomic correlations via CAVIAR requires both sufficient slice sampling and adequate depth resolution. To validate that these conditions were met in our experiments, we begin by analyzing the achieved spatial resolution. Figure\u00a03<\/a>a shows the projected phase, summed over all 30 z-slices. In Fig.\u00a03<\/a>b we show the power spectrum incoherently averaged over object slices and states, which indicates an information transfer up to d\u00a0=\u00a047.6 pm. Dividing the achieved resolution by the illumination wavelength \u03bb\u00a0=\u00a04.87 pm gives us a value of d\u00a0=\u00a09.78\u03bb. A ratio d\/\u03bb\u00a0<\u00a010 for a three dimensional potential reconstruction has so far been difficult to achieve10<\/a>. In order to estimate the depth resolution we azimuthally averaged the 3D Fourier transform of the phase over the in-plane spatial frequencies. The corresponding power spectrum is shown in Fig.\u00a03<\/a>c and clearly indicates a missing wedge of three dimensional information10<\/a>. Albeit the in-plane frequencies show an information transfer reaching beyond the double aperture limit, the depth spatial frequencies are still confined in a full convergence angle, in agreement with previously published multislice ptychographic reconstructions20<\/a>. The first two rings of hBN reflections correspond to a depth resolution of approximately 2\u2009nm, but at high scattering angles, the first-order Laue zone diffraction ring is reconstructed, indicating a vertical resolution better than 6.7\u2009\u00c5 (see Supplementary Fig.\u00a09<\/a>). In order to illustrate the separation between upper and lower lattices of the sample we computed average phase profiles as a function of distance r to the center of the atomic columns in each of the reconstructed potential slices. In Fig.\u00a03<\/a>d, e these r\u00a0\u2212\u00a0z phase maps for the upper and lower lattices are shown. In Fig.\u00a03<\/a>f the phase value at the center of the atomic columns is shown as a function of z (line profiles along the left hand sides of Fig.\u00a03<\/a>d, e). The boundaries of each sub-lattice as well as an overlap region caused by the limited resolution along the z-axis can clearly be identified.<\/p>\n

Fig. 3: Experimental ptychographic reconstruction from a 4D-STEM dataset of an approximately 15\u2009nm thick hBN bicrystal.<\/b>\"Fig.The alternative text for this image may have been generated using AI.<\/p>\n

The specimens upper- and lower-halves (the beam propagates vertically) are twisted with respect to each other by approximately 11\u2218. The reconstruction was carried out using 30 slices, 10 object states and 5 probe modes to account for partial spatial coherence and sample drift. a<\/b> Reconstructed phase projected over all slices and averaged over all states. b<\/b> Sum of moduli squared of the Fast Fourier-transforms of all slices and object states indicating an information transfer up to 47.8\u2009pm, way above the double-aperture radius limit (61\u2009pm for an accelerating voltage of 60\u2009kV and convergence semi-angle of 40\u2009mrad). c<\/b> The azimuthal average of the 3D-Fourier moduli squared makes it possible to estimate the achieved resolution along the z-axis\u2014approximately 2\u2009nm for the first two rings of Bragg peaks. d<\/b>, e<\/b> Phase profiles averaged over all reconstructed atoms obtained by azimuthally averaging phase profiles around the centers of atomic columns within the upper and lower slices of the reconstruction, respectively. f<\/b> Phase profiles along z through the centers of atomic columns clearly validating that two crystals have identical thicknesses within the precision limited by the depth resolution.<\/p>\n

The depth resolution analysis above confirms that multiple scattering effects are accounted for in the reconstruction, enabling further analysis of interatomic correlations. For each of the 30 slices and 10 states we extracted atom positions and corresponding displacements. In Fig.\u00a04<\/a>a, b projected phases and interatomic correlations averaged in z-regions from 1.9\u2009nm up to 5.7\u2009nm and from 8.6\u2009nm up to 12.5\u2009nm, respectively, are shown. For each of the two lattices we evaluated correlations in coordinate systems aligned with the respective local lattice frame. In the magenta dashed boxes we show correlations corresponding to a particular atom column (and its neighbors) and in the turquoise boxes we show correlations averaged over all identical unit-cell pairs. The correlation vectors, overlaid at atomic sites, reveal similar spatial structures and orientations in both layers, despite their relative rotation. This confirms that the recovered correlations are physically meaningful and lattice-intrinsic. We further compute a Debye-Waller B-factor by summing self correlations (x-x and y-y) and multiplying the sum by 8\u03c02. For the upper and lower lattices we get 3\u00a0\u00d7\u00a0103\u2009pm2 and 3.5\u00a0\u00d7\u00a0103\u2009pm2 respectively, which are approximately 6 times lower than the reported value of 2\u00a0\u00d7\u00a0104\u2009pm2 obtained from powder X-ray diffraction38<\/a>. Such a discrepancy cannot simply be explained by a lack of z-vibrations in our reconstruction, as the z-resolution of 2\u2009nm in this ptychographic reconstruction is clearly not sufficient to detect fluctuations along this axis. Although for a 2D material, such as hBN, where the layers are bound only by the Van der Waals forces to one another, the z-vibrations might significantly increase the isotropic squared displacements, two additional limiting factors are responsible for reducing the amplitude of reconstructed correlations and the remaining broadening of the images of projected atom positions within the layers: (1) The z-distance between atomic layers in hBN is 0.33\u2009nm, which means that per 0.48\u2009nm thin potential slice there are on average 1.4 layers of atoms, the positions of which are averaged over, and (2) the limited number of only 10 object states may not be sufficient to represent all of the TDS and Debye-Waller factor by an incoherent sum of diffraction patterns from statically displaced atom configurations.<\/p>\n

Fig. 4: Interatomic correlations recovered from experimental mixed-object ptychographic reconstruction of approximately 15\u2009nm thick bulk hBN crystal.<\/b>\"Fig.The alternative text for this image may have been generated using AI.<\/p>\n

a<\/b>, b<\/b> Projected phases of the reconstructed transmission function summed over all object states and 8 out of 30 slices selected from the z-ranges z\u00a0=\u00a01.9\u2009nm \u20265.7\u2009nm and z\u00a0=\u00a08.6\u2009nm \u202612.5\u2009nm, respectively. For each of the potential slices within those ranges the interatomic correlations were computed separately; their averages were computed afterwards. The depicted arrows represent various components of retrieved Green\u2019s tensors. Randomly picked reference atoms are shown by white outline, yellow and purple arrows show correlations of atoms with their own displacement along x- and y-axes. Pink and light blue show correlations of atoms with displacements of a reference atom along x or y direction. Note that for the two stacked crystals the coordinate system was rotated to align the x-direction with one of the three nearest neighbors. The y-axis was then defined to be orthogonal to it. Directions of each coordinate system are shown with white arrows in the lower left corners of each panel as well as the unit cells. It is important to mention that individual atoms may vibrate quite differently, but on average they are expected to behave similarly. In the magenta colored dashed boxes we show correlations corresponding to a particular atomic column (and its neighbors) and in turquoise boxes we show correlations averaged over all identical unit-cell pairs (i.e., with the same differences \\(l-{l}^{{\\prime} }\\) and \\(\\kappa -{\\kappa }^{{\\prime} }\\)). Note that principal directions of averaged correlations are identical in (a<\/b>, b<\/b>).<\/p>\n

Still, despite an underestimated correlation strength, their directions provide direct access to the dynamical matrix of the crystal. Following the procedure described in ref. 18<\/a>, we compute Fourier-transformed displacements by summing over all L available unit cells l: <\/p>\n

$${\\widetilde{u}}_{\\kappa \\alpha }(q)=\\frac{1}{\\sqrt{L}}{\\sum }_{l=1}^{L}{u}_{l\\kappa \\alpha }\\times {e}^{-iq{r}_{l}}.$$<\/p>\n

\n (5)\n <\/p>\n

By assuming boron\u2019s and nitrogen\u2019s harmonically averaged atomic mass of 12.2 a.m.u. and a temperature of 300 K we obtain the dynamical matrix18<\/a>,31<\/a> via: <\/p>\n

$${D}_{\\kappa \\alpha {\\kappa }^{{\\prime} }\\beta }(q)=\\frac{{k}_{B}T}{m}{\\left[\\langle {\\widetilde{u}}_{\\kappa \\alpha }(q){\\widetilde{u}}_{{\\kappa }^{{\\prime} }\\beta }^{*}(q)\\rangle \\right]}_{\\kappa \\alpha {\\kappa }^{{\\prime} }\\beta }^{-1},$$<\/p>\n

\n (6)\n <\/p>\n

where \u03ba, \\({\\kappa }^{{\\prime} }\\) indicate one of two basis atoms and \u03b1, \u03b2 indicate the displacement directions (x and y). Computing the square root of eigenvalues of this matrix directly results in the phonon-dispersion curves. Although the 59 unit cells available in our scanned field of view are not sufficient to examine the curves in detail, we compute an average frequency along the \u0393\u00a0\u2212\u00a0K\u00a0\u2212\u00a0M\u00a0\u2212\u00a0\u0393 path. We obtain 10.7 THz, 14.1 THz, 17.7 THz and 24.0 THz for transverse acoustic, longitudinal acoustic, transverse optical, longitudinal optical branches in the upper lattice. For the bottom lattice we get 10.8\u2009THz, 13.0\u2009THz, 18.2\u2009THz, and 27.0\u2009THz for the same four branches. The reported theoretical values are approximately 1.5 times higher: 14.3\u2009THz, 22.9\u2009THz, 39.5\u2009THz and 39.6\u2009THz39<\/a>. The scaling of energies calculated via CAVIAR is affected by both assumed temperature as well as extracted amplitude of atomic vibrations. Although the electron beam heats the sample during data collection, this effect is negligible, as the temperature raises only by up to a few tens of Kelvin, which is not enough to observe a noticeable change in energy. The vibration amplitude is a far more important factor, and our simulations show that CAVIAR can underestimate it. Still, CAVIAR makes it possible to place frequencies of two acoustic and two optical branches in the correct range for a complementary Vibrational STEM-EELS measurements23<\/a>,40<\/a>. Further, unlike EELS, mixed-object ptychography yields directions of correlated vibrations at the level of a single atom.<\/p>\n","protected":false},"excerpt":{"rendered":"The first step of CAVIAR is a gradient-based ptychographic reconstruction5,10,20,25,26,27,28 which fits a transmission function of the specimen…\n","protected":false},"author":3,"featured_media":860313,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_share_on_mastodon":"0"},"categories":[25],"tags":[27212,10046,51052,10047,492,159,107090,39201,67,132,68],"class_list":["post-860312","post","type-post","status-publish","format-standard","has-post-thumbnail","category-physics","tag-computational-methods","tag-humanities-and-social-sciences","tag-imaging-techniques","tag-multidisciplinary","tag-physics","tag-science","tag-transmission-electron-microscopy","tag-two-dimensional-materials","tag-united-states","tag-unitedstates","tag-us"],"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@us\/116732624084939294","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/posts\/860312","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/comments?post=860312"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/posts\/860312\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/media\/860313"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/media?parent=860312"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/categories?post=860312"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/us\/wp-json\/wp\/v2\/tags?post=860312"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}