Ultrafast disordering of vanadium dimers in photoexcited VO2

See allHide authors and affiliations

Science  02 Nov 2018:
Vol. 362, Issue 6414, pp. 572-576
DOI: 10.1126/science.aau3873

Snapshots of a phase transition

Time-resolved x-ray scattering can be used to investigate the dynamics of materials during the switch from one structural phase to another. So far, methods provide an ensemble average and may miss crucial aspects of the detailed mechanisms at play. Wall et al. used a total-scattering technique to probe the dynamics of the ultrafast insulator-to-metal transition of vanadium dioxide (VO2) (see the Perspective by Cavalleri). Femtosecond x-ray pulses provide access to the time- and momentum-resolved dynamics of the structural transition. Their results show that the photoinduced transition is of the order-disorder type, driven by an ultrafast change in the lattice potential that suddenly unlocks the vanadium atoms and yields large-amplitude uncorrelated motions, rather than occurring through a coherent displacive mechanism.

Science, this issue p. 572; see also p. 525


Many ultrafast solid phase transitions are treated as chemical reactions that transform the structures between two different unit cells along a reaction coordinate, but this neglects the role of disorder. Although ultrafast diffraction provides insights into atomic dynamics during such transformations, diffraction alone probes an averaged unit cell and is less sensitive to randomness in the transition pathway. Using total scattering of femtosecond x-ray pulses, we show that atomic disordering in photoexcited vanadium dioxide (VO2) is central to the transition mechanism and that, after photoexcitation, the system explores a large volume of phase space on a time scale comparable to that of a single phonon oscillation. These results overturn the current understanding of an archetypal ultrafast phase transition and provide new microscopic insights into rapid evolution toward equilibrium in photoexcited matter.

Manipulating the structure and function of materials remains one of the ultimate challenges of modern science and technology (1). Ultrafast pulses can create new phases not accessible in equilibrium by transiently modifying the interatomic potential (24) and the atomic positions (57). However, probing the short-lived transient structures of 1023 atoms is a challenge, and the dynamics are often simplified by assuming they evolve on a potential energy surface described by a few degrees of freedom with a well-defined spatial periodicity (3, 811). Although ultrafast electron (8, 12, 13) and x-ray diffraction (2, 10, 1417) are sensitive to the structure with atomic length scale and femtosecond time resolution, they measure averages over many unit cells, and thus are less sensitive to random displacements. To overcome this limitation, we used femtosecond total x-ray scattering (18) to study the dynamics of the structural transition of bulk VO2 at all length scales. We observed that the structural transition proceeds by uncorrelated disordering of the vanadium ions from their initial dimerized distribution, rather than the previously proposed synchronized motion along an optical phonon mode (8, 12, 14, 19). Supported by ab initio molecular dynamics, our results show that the highly anharmonic, flat potential energy surface allows the quasi-rutile structure in the photoexcited state to develop on femtosecond time scales, by disrupting the vanadium pairs and populating a continuum of modes that enables the system to reach ergodicity within 150 fs. More generally, these results question the common interpretation of many ultrafast measurements in terms of motion along a well-defined reaction coordinate, or order parameter, on a potential energy surface (5, 8, 10, 12, 16, 20) and provide a new level of understanding regarding how complex systems can reach ergodicity on an ultrafast time scale.

Ultrafast structural transitions are usually described in terms of a few phonon modes with well-defined wave vectors, and it is assumed that the system evolves along a potential surface connecting the initial and final structures that also determines the speed of the photoexcited transition (24, 14, 19). However, this assumes a coordinated motion along a well-defined reaction coordinate and ignores the role of disorder and entropy, which are thought to develop on slower time scales, even though entropy may be the driving factor in the thermodynamic transition (21). To date, all attempts to measure the “molecular movie” of the dynamics of a phase transition at the atomic scale have exploited Bragg scattering, which can only measure the average motions of atoms and is insensitive to deviations from the average reaction coordinate. Using femtosecond pulses from a free-electron laser (FEL), we measured total x-ray scattering, containing discrete Bragg peaks as well as the diffuse continuum between them, and show that the lattice disorder in photoexcited VO2, manifested as diffuse intensity across momentum space, reaches close to its equilibrium value on a time scale comparable to that of a single atomic oscillation.

Vanadium dioxide (VO2) can be switched from an insulating to a metallic state with ultrafast pulses. The insulator-to-metal transition (IMT) is accompanied by a structural change from a monoclinic (M1) to a rutile (R) structure (Fig. 1, A and B), and numerous prior efforts have focused on elucidating the evolution of both the electronic and lattice degrees of freedom via ultrafast pump-probe measurements (4, 8, 9, 12, 14, 21). Generally, a solid-solid phase transition where the symmetry of the crystal is raised (from M1 to R in VO2) as the temperature increases can be dominated by either a displacive process or an order-disorder process (Fig. 1, C and D). The distortion at the IMT in VO2 has widely been interpreted as a displacive transition corresponding to a phonon soft mode of the high-symmetry metallic phase, based on symmetry considerations and diffraction measurements (812), as well as lattice dynamics simulations (2225). In a displacive transition, the atoms collaboratively reshuffle their positions under the effect of one or few well-defined, spatially coherent vibrational modes, of amplitude ξ, that link the low- and high-symmetry structures. This is illustrated for the dimerized vanadium pairs of the monoclinic (M1) phase of VO2 in Fig. 1C, with wave vector at the R point in the rutile Brillouin zone, k(R) = (0, ½, ½)R [here, subscripts denote the Bravais basis convention (M1 or R), while the superscript (R) represents the special point in the Brillouin zone]. Below the critical temperature, Tc = 340 K, the system is locked near an energy minimum at a large amplitude of ξ, yielding the structural distortion and additional superstructure diffraction peaks, owing to the doubling of the unit cell volume, represented by the red dots in Fig. 1E. Above Tc, the onset of a displacive mechanism would show characteristic soft phonons at the associated wave-vector (red crosses at k(R) in Fig. 1F, corresponding to additional Brillouin zone centers in Fig. 1E) (2628). Alternatively, in an order-disorder transition, the atoms move from the low- to the high-symmetry structure in a spatially incoherent manner with no characteristic correlation length (26, 28) and without soft modes at any wave vector (Fig. 1D). The two possible types of transformation can be distinguished by probing for diffuse intensity extended across reciprocal space, which is a hallmark of large-amplitude motions due to the uncorrelated disorder (orange shaded areas, Fig. 1G). Notably, in both cases, the average atomic position may follow the same trajectory, and thus an ultrafast diffraction measurement probing the Bragg peaks alone cannot distinguish between displacive and order-disorder mechanisms (Fig. 1, H and I).

Fig. 1 VO2 structural distortion in real and reciprocal space.

(A and B) Correspondence between the M1 and R phases, highlighting the pattern of dimerization of V pairs along the rutile (R) c axis, leading to diagonal layers of tilted dimers (green bands), along the wave vector corresponding to the modulation of the R phase, k(R) = (½, 0, ½) or (0, ½, ½) reciprocal lattice units. Gray boxes represent the unit cell of each structure. (C and D) Alternate possible transition mechanisms between the M1 phase and the R phase upon photoexcitation. The spatially coherent distortion (C) with coordinate ξ is a symmetry-allowed transformation path between R and M1, which matches low-energy phonon excitations of the rutile phase at the R point, localized at k(R) in reciprocal space. The disordering mechanism (D) has no well-defined wave vector and causes extended diffuse scattering in k-space. Diffractograms for the (E) M1 phase and (F and G) rutile-like phases resulting from either displacive distortion (F) or a disordering transition (G). Note the appearance in M1 of superstructure Bragg peaks at k(R) points of the rutile phase (crosses at coordinates with half integers in R phase). The shaded orange regions in (G) represent a diffuse scattering intensity that is more extended than the exact k(R) modulation wave vector. (H and I) Schematic time evolutions of the intensity of Bragg peaks for M1 and R phases and diffuse signal, for a displacive transition (H) or an order-disorder transition (I). Zero time corresponds to photoexcitation.

Optical probes of photoexcited VO2 show that ~6-THz coherent optical phonons, corresponding to the motion of the vanadium ions, are induced for low-intensity excitation (4, 9, 19). At the same time, structural probes (8, 12) have shown that the long-range order of the V dimerization is lost in <300 fs, but these measurements lacked the sub–100 fs time resolution required to probe the transition on the time scale of relevant vibrational periods. Furthermore, these experiments lacked sensitivity to deviations from average lattice periodicity at arbitrary wave vectors, which are a signature of disorder in the transition mechanism. Here, we used femtosecond total x-ray scattering (~50-fs time resolution) with an x-ray FEL to obtain a more complete picture of the dynamics during the photoinduced structural transition in VO2. We detected a rapid disordering of the low-temperature structure that starts as soon as 50 fs after photoexcitation, as observed through a rapid increase in diffuse x-ray intensity and an equally fast decrease in the intensity of the Bragg peaks of the low-symmetry structure. Furthermore, the diffuse intensity distribution in reciprocal space at 150 fs resembles the rutile equilibrium distribution and suggests that the lattice reaches a near-ergodic state in that time scale. This unequivocally shows that the nonequilibrium photoinduced transition is of the order-disorder type, driven by an ultrafast change in the lattice potential that suddenly unlocks the V dimers and yields large-amplitude uncorrelated motions, rather than occurring along a coherent displacive coordinate. Our ab initio molecular dynamics (AIMD) simulations based on density functional theory (DFT) potentials corroborate the observed ultrafast time scale of the disordering. The DFT calculations show that the ionic potential in the photoexcited phase is highly anharmonic and flat, which explains the observed large V displacements, and is also the origin of the high phonon entropy of the equilibrium metallic phase (21).

Total x-ray scattering patterns were collected with 9.5-keV photons at the XPP end station at the Linac Coherent Light Source (LCLS) from high-quality single crystals of VO2 [see methods for details (29)]. The difference in the equilibrium scattering for T = 353 K (R) and T = 295 K (M1) is shown in Fig. 2A [see fig. S2 for M1 and R raw patterns (29)]. The change is dominated by the suppression of three strong Bragg peaks of the M1 phase labeled (−1, 0, −2)M1, (−1, 2, 0)M1, and (−1, 2, 2)M1. In addition, a rectangular-shaped diffuse feature strongly increases in intensity. Our simulations of the change in the total scattering (Fig. 2B), based on the equilibrium force constants obtained from DFT [see methods (29)], reproduce the diffuse signature well and indicate that it originates from soft transverse acoustic phonons of the rutile phase (21). This enables the V atoms to adopt a large distribution of displacements about their average positions in the rutile phase.

Fig. 2 Diffuse scattering from the thermal and photoinduced phase transitions in VO2.

Saturated false-color plot of the difference in the thermal diffuse scattering intensity between the R (353 K) and the M1 phases (295 K) [ΔI(Q) = I(Q, 353 K) − I(Q, 295 K)], (A) measured at LCLS and (B) calculated on the basis of the equilibrium force constants. The positions of several Bragg peaks are indicated. The rings in (A) arise from x-ray scattering at damaged edges of the sample, and the central semicircle-like feature corresponds to laser scatter [see methods and figs. S1 and S2 for details (29)]. (C) Pump-induced change in the scattering signal when excited at 20 mJ/cm2 at a fixed delay of 350 fs [ΔI(Q) = I(Q, 350 fs) − I(Q, −11 ns)], showing changes very similar to those of the thermal response. (D) Pump fluence dependence of the (−1, 2, 0)M1 peak amplitude and mean diffuse scattering as integrated over the boxed area at 350 fs and 500 ps. The gray dots represent the background, taken as the mean intensity at low scattering angles. A linear fit (dashed line) to the high-fluence part of the diffuse scattering gives a threshold of Fc ~ 5 mJ/cm2.

In Fig. 2C, we show the photoinduced changes in the total scattering 350 fs after photoexcitation for an absorbed pump fluence of 20 mJ/cm2, in the saturation regime for the photoinduced phase transition (19). The photoinduced changes in the total scattering show a marked resemblance to the thermal changes in Fig. 2A. The intensities of the Bragg peaks are strongly suppressed as a result of the change in the crystalline symmetry, whereas the extended diffuse scattering strongly increases. This shows that on this very fast time scale, not only has the long-range, average symmetry of the crystal changed, as indicated by the fast suppression of the M1 peaks, but also the V atoms have already adopted a rutile-like displacement distribution.

Before turning to the time scale in more detail, we examine the fluence dependence of the photoinduced phase transition. Figure 2D shows the fluence dependence of the (−1, 2, 0)M1 Bragg peak and the integrated diffuse scattering at ~350 fs and 500 ps after excitation. The Bragg peak shows a linear decrease in intensity with absorbed fluence for fluences below a threshold of Fc ~ 5 mJ/cm2, after which the change is more rapid until saturation and complete suppression occurs. The dynamics of the Bragg peak also show distinct changes below and above threshold, with coherent phonons observed to modulate the Bragg intensity below threshold, whereas the coherent signal disappears above threshold (fig. S11), in agreement with optical measurements (4). The diffuse scattering also shows threshold-like behavior, rising abruptly above Fc ~ 5 mJ/cm2, indicating that the V ions only adopt the broad, rutile-like distribution above a critical fluence.

Figure 3A shows the change in intensity ΔI(Q, t) at delays of 50 fs, 100 fs, and 2 ps for excitation in the saturation regime. It is clear that changes in the diffuse scattering already begin to occur over extended portions of reciprocal space within 50 fs of photoexcitation, and the Q dependence of the scattering shows little subsequent change after reaching its full intensity at about 150 fs. To examine this time scale in more detail, we follow the temporal evolution of the intensity for specific regions of reciprocal space. Figure 3B shows the dynamics of multiple Bragg peaks and of the diffuse scattering intensity in the high-fluence regime. All curves clearly show that the time scale for the structural transition is at least twice as fast as the 300 fs reported from previous diffraction measurements (8, 12). Notably, this time scale is above the ~50-fs experimental time resolution. The time scale for the rise of the integrated diffuse intensity is independent of fluence above the threshold, as shown in fig. S12. However, below threshold and at long delays, the Bragg peaks can be affected by strain, which leads to slow changes in intensity at the few-ps time scale.

Fig. 3 Time dependence of Bragg peaks and diffuse scattering for above-threshold pumping of 22 mJ/cm2.

(A) Snapshots of the increase in the diffuse intensity ΔI(Q, t) at several delays after laser excitation. (B) Temporal evolution of selected Bragg peaks and integrated diffuse scattering signal compared to AIMD simulations (solid lines). With the exception of the diffuse scattering signal, all data are normalized to the pre–time zero amplitude for both experiments and simulations with no further scale factor. The diffuse scattering data are compared to the variance of the vanadium positions obtained from the AIMD simulations and are thus separately normalized. The horizontal bar indicates half of the period of the relevant 3-THz acoustic modes of the rutile phase (see text for details). The additional Bragg peaks that do not appear in (A) were obtained in a different experimental geometry, shown in fig. S3.

The dynamics of both the Bragg peaks and diffuse scattering observed here indicate a direct structural transition between the M1 and R phase, without evidence of passing through intermediate structures previously reported (12). M1 superstructure peaks [e.g., (−1, 0, −2)M1, (−1, 2, 0)M1] are not present in the rutile structure, and all rapidly drop to zero intensity. Peaks labeled R are present in both phases and can initially increase or decrease, depending on their structure factor response to the average distortion, Fhkl(ξ). Notably, the diffuse scattering signal changes on the same time scale as that of the Bragg peaks, showing that the disordering occurs as fast as the loss of M1 long-range order. This indicates that the phase transition is dominated by disordering of the V ions rather than a coherent atomic motion along the ξ coordinate.

To gain greater insight into the phase-change dynamics, we performed AIMD simulations of the photoinduced phase transition. The lattice potential was obtained from DFT, including electronic correlation effects [see (29) for additional information], and the effect of photoexcitation was modeled as a sudden increase in the electronic temperature, Tel [see supplementary materials for details (29)]. The relaxed ground-state structure in the M1 phase shows a clear threshold behavior in Tel above which the V dimerization disappears (fig. S4), leading to a rutile-like phase, similar to the trend observed with increasing hole concentration (30) and in agreement with the experimental observation of a threshold laser fluence. The time-dependent Bragg intensities derived from the ionic positions after electronic excitation are plotted in Fig. 3B, together with the measured diffraction signals. Excellent agreement is obtained both in terms of the time scale for the dynamics and the amplitude of the change. The agreement is particularly noteworthy considering that the only adjustable parameter was the global temporal offset between the data and the simulation.

Figure 4A shows the AIMD evolution of short dimerized bonds (dS) and long V-V bonds (dL) along the rutile c axis, after a sudden increase in Tel at t = 0, starting with a thermal M1 structure at 300 K. Bond-length distributions, averaged over specified time windows, are shown in Fig. 4B. The initial distribution is bimodal (dS = 2.50 Å, dL = 3.16 Å), reflecting the monoclinic dimerization pattern (Fig. 1A). After the prompt transition, it quickly collapses to the unimodal distribution expected for the rutile phase, dS = dL = 2.81 Å = dR. Already within the first 75 fs, the tails of the distributions show some vanadium pairs with the rutile bond length, but the means only cross at about 150 fs and slightly overshoot the ideal rutile bond length at t ~ 200 fs before rebounding, resulting in a damped oscillation converging to the equidistant configuration. This overshoot is responsible for the single-cycle overdamped oscillation in the (−1, 0, −2)M1 and (−1, −1, 3)M1 peaks. Similar dynamics are observed in the evolution of the twist angle, and the frequency of the damped oscillation is about 3 THz (fig. S5), but is doubled (~6 THz) in the diffraction data shown in Fig. 3B because |Fhkl(ξ)|2 ~ ξ2 when dL = dS = dR.

Fig. 4 Simulated VO2 response under photoexcitation, based on first-principles DFT simulations.

(A) Time evolution of dimer bond lengths from ab initio molecular dynamics. Short (dS) and long (dL) bonds are indicated in Fig. 1A. (B) Distribution of dimer bond lengths from (A), integrated over several time intervals before and after photoexcitation. The merging of the distributions after a brief overshoot is evident, indicating the evolution toward the rutile phase with chains of equidistant V atoms. (C) Effect of photoexcitation on the potential for the distortion coordinate ξ. Rutile corresponds to ξ =0 Å and M1 to ξ = ±0.24 Å. The photoexcitation lifts the two M1 minima of the double-well potential to stabilize the rutile phase in a flat, strongly quartic potential. Markers are DFT energies, and curves are polynomial fits. (D) Phonon density of states, and elemental decomposition on V and O atoms, obtained from AIMD trajectories for the equilibrium M1 phase before photoexcitation (bottom) and after photoexcitation (middle). The appearance of the pronounced V vibration peak at 3 THz (dashed line) after photoexcitation is characteristic of the rutile phase equilibrium phonon dynamics, shown in the top curves (T = 380 K).

Our measurements, confirmed by simulations, show that the changes in bond length and angle occur concomitantly and within a few hundred femtoseconds, in contrast to previous scenarios based on analysis of time-resolved electron Bragg scattering alone (8). The vibrational modes at 3 THz are a factor of 2 slower than the 6-THz mode previously suggested to limit the phase transition (19), but agrees well with the frequency of V-dominated flat transverse acoustic modes of the R phase, as seen in our first-principles simulations of the phonon density of states (Fig. 4D) and dispersions (fig. S6d), showing that no single wave vector is responsible for the transition. In addition, the dynamics of the standard deviation of the bond lengths is consistent with the time scale of the diffuse scattering (solid line and symbols in Fig. 3B, top trace), further demonstrating that a broad distribution of modes is required to describe the evolution of the distribution of displacements in Fig. 4, A and B. The AIMD simulations thus indicate that photoexcitation drives an instantaneous change in the lattice potential, leading to rutile-like dynamics of the V ions. The effect of changing the electronic temperature on the potential energy surface along the distortion ξ (M1-to-R distortion in Fig. 1C) is shown in Fig. 4C. As can be seen, the electronic excitation results in the loss of the double-well potential characteristic of the M1 phase and instead stabilizes the flat, strongly anharmonic quartic-like potential of the rutile phase.

These observations allow us to build a radically new description of the photoinduced phase transition in VO2. To date, most theoretical models and interpretations of experimental data have assumed a transition driven coherently by phonon modes of well-defined wave vector (typically kR) and frequency (8, 9, 12, 14, 19), and have ignored the role of fluctuations and disorder. Our combined ultrafast total scattering measurements and first-principles simulations of dynamics show that ultrafast disordering is a key aspect of the photoexcited state of VO2. Photoexcitation instantaneously creates a new lattice potential, which can be described by a hot electron distribution within the DFT framework. The new potential quickly alters the dynamics of V ions, effectively severing the intradimer spring constant, causing them to rapidly assume the new high-symmetry configuration. We note that these V-V interactions in the rutile phase yield a flat, dispersiveless transverse acoustic phonon branch (21). These features of the potential cause the transformation to proceed without spatial coherence and no well-defined wave vector.

The time scale of this disordering process is another particularly unexpected observation. The rapid excitation of phonons in a broad range of wave vectors, observed through the diffuse scattering and occurring within 150 fs, is faster than other disorder-driven phase transitions (13, 17). Whereas the speed of a soft mode–driven transition is governed by the curvature of the potential surface along the ξ coordinate (3), the nonequilibrium order-disorder transition time scale reflects the time for the displacements to become uncorrelated and reach a quasi-equilibrium distribution. The dynamics observed here are fast because the average distance between M1 and rutile is relatively short (ξ0 = 0.24 Å, the position of the minimum of the potential in Fig. 4C). Indeed, the average vanadium velocity at room temperature is 3.8 Å/ps, so the V atoms could traverse the distance ξ0 in ~60 fs. Although this is a simplified model, the flatness of the photoexcited potential in Fig. 4C suggests that the dynamics are partially inertial and that the speed of the transition is mainly limited by the initial thermal distribution of velocities in the M1 phase and the amplitude of the M1-to-R distortion. The change in the potential produces an ultrafast activation of a large number of low-energy V-dominated phonon modes spread out in reciprocal space, as seen in the broadly distributed diffuse scattering signal, and in the phonon dispersions in fig. S6. The rapid evolution toward ergodicity is thus enabled by the large phase space with low-frequency V-dominated transverse acoustic modes, which was also noted in (21) to yield a large phonon entropy gain stabilizing the rutile phase and the strong damping of phonons in the rutile phase. Our current observations thus reveal an interesting parallel to the thermal transition.

Our results show that the structural transformation in VO2 occurs along many concurrent degrees of freedom spanning a large phase space, and thus a complete description must extend beyond that of a single-order parameter or discrete set of displacements. The rapid, large-amplitude disordering observed here is caused by the flatness of the photoexcited potential and is intimately related to the high entropy of the rutile phase (21); but the implications of our findings extend well beyond the VO2 system. The role of disorder has been neglected in many ultrafast solid-solid phase transitions, even though the ultimate control of properties on demand relies on a reversible pathway between the two states. Our findings suggest that disorder may play an important role in some materials and that a description in terms of a single degree of freedom is incomplete. This disorder presents both opportunities as well as challenges to ultrafast material control. Disorder ultimately prevents coherent control of phase transitions and complete control over the phase transition pathway. However, transient disorder can become a new knob by which materials can be controlled, just as disordered solids (glasses) can show marked differences compared with their crystalline counterparts in near-equilibrium conditions. Understanding whether disorder plays a general role in vibrationally excited solids (7) could ultimately provide a new perspective on how to control matter, suggesting a new direction in which light-induced superconductivity may be achieved (5).

Supplementary Materials

Materials and Methods

Figs. S1 to S12

Movies S1 and S2

References (3148)

References and Notes

  1. Materials and methods are available as supplementary materials
Acknowledgments: Preliminary static diffuse scattering patterns were collected at the 11-3 beamline of the Stanford Synchrotron Radiation Light Source (SSRL). Funding: Work at ICFO was supported by Spanish MINECO (Severo Ochoa grant SEV-2015-0522), Ramón y Cajal programme RYC-2013-14838, and Fondo Europeo de Desarrollo Regional FIS2015-67898-P (MINECO/FEDER), as well as by the European Union’s Horizon 2020 research and innovation programme (grant agreement no. 758461) Fundació Privada Cellex, and CERCA Programme –Generalitat de Catalunya. First-principles simulations and analysis (S.Y., O.D.) were supported by the U.S. Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division, under the Early Career Award no. DE-SC0016166. M.K., T.H., M.T., and D.A.R. were supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences through the Division of Materials Sciences and Engineering under contract no. DE-AC02-76SF00515. Use of the Linac Coherent Light Source (LCLS), and the Stanford Synchrotron Radiation Lightsource at SLAC National Accelerator Laboratory, is supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences under contract no. DE-AC02-76SF00515. Additional x-ray measurements were performed at BL3 of SACLA with the approval of the Japan Synchrotron Radiation Research Institute (JASRI) (proposal no. 2016A8008). Ab initio calculations were performed using resources of the National Energy Research Scientific Computing Center, a U.S. DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under contract no. DE-AC02-05CH11231. First principles simulations for this research used resources of the Oak Ridge Leadership Computing Facility at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under contract no. DE-AC05-00OR22725. Author contributions: L.V., M.C., J.M.G., M.K., T.K., T.H., M.J., T.A.M., D.A.R., and O.D. contributed to the data collection led by S.W. and M.T. Data analysis was performed by S.W. and M.T. Sample synthesis was performed by L.A.B. S.Y. performed DFT calculations under the supervision of O.D. S.W., O.D., and M.T. proposed, planned, and coordinated all aspects of the project. Competing interests: The authors declare no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper or the supplementary materials.

Stay Connected to Science

Navigate This Article