WorksPhysics and Astronomy
Origin of Flat Bands and Role of Electron Correlation in Lutetium Hydrides
Abstract
Lutetium hydrides (LuH, ) form a diverse series of phases, several of which superconduct under pressure. Characterizing their electronic properties has remained challenging owing to a high propensity for hydrogen defect formation, and recent angle-resolved photoemission (ARPES) measurements reveal puzzling flat-band regions that position these materials as candidates where superconductivity and flat-band physics may intersect. Here, by combining density functional theory, dynamical mean-field theory, and the constrained random-phase approximation, we uncover the microscopic origin and correlation nature of these flat bands. Across all compositions, the screened on-site Coulomb interaction is larger for H-s states than for Lu- states due to compact hydrogen orbitals. Nevertheless, these systems remain weakly correlated metals: the nearly filled H- shell admits little charge fluctuation, so its large interaction acts as a static level shift rather than a source of correlation. Although hydrogen primarily occupies tetrahedral sites at , we discover that anti-site defects–where hydrogens occupy slightly unfavorable octahedral sites–generate both the ARPES flat-band features and the low-energy optical absorption peak, attesting to the usual defective nature of such materials in experimental samples. We further find that correlation strength is governed primarily by hydrogen orbital filling at these sites rather than the interaction magnitude itself. Consequently, we identify hydrogen orbital filling as the fundamental organizing principle dictating correlation and low-energy flat-band physics in lutetium hydrides.
Introduction
Many metal hydrides are well described within conventional band theory, as the metal-derived electrons are mostly itinerant and they effectively screen the hydrogen states. The coupled electronic, structural and phonon properties of stoichiometric and non-stoichiometric lanthanide hydrides have been a puzzling topic of study for over half century. Motivated by the argument of Matthias [1], Merriam and Schreiber [2] searched for superconductivity in lanthanum hydrides to below 1 K. Two decades later, a sub-kelvin for was estimated based on Bardeen–Cooper–Schrieffer (BCS) theory [3]. This superconductivity was ultimately confirmed [4] in slightly supersoichiometric . A growing body of later work suggested that certain rare-earth hydrides exhibit behavior beyond the standard band-theory description, including electron–electron correlation effects and correlation-driven metal–insulator transitions.[5, 6, 7, 8] This beyond-band-theory behavior originates in hydrogen lattice itself: a periodic array of such correlation-stabilized ions is a genuine many-body problem exhibiting metallic, intermediate, and Mott-insulating regimes [9], and correlation-based treatments have accordingly cast as correlation-driven insulators [10, 11], with hydrogen ordering proposed as an additional contributing factor [12, 13].
Lutetium hydride is a particularly clean system in which to test these ideas. Its filled shell removes the -electron correlation that complicates the lighter rare earths, isolating the hydrogen and - states as the only candidates for correlated behavior; and its accessible dihydride–trihydride range spans hydrogen fillings from the tetrahedral-only dihydride to the octahedrally-filled trihydride, allowing the hydrogen occupation to be tuned systematically. Weaver et al. [14] measured optical reflectance spectra that established the dominance of Drude absorption and yielded the carrier plasma frequency, while synchrotron photoemission resolved the Lu- conduction states near the Fermi level and a H- band several eV below it, providing evidence for charge transfer from the lutetium conduction () manifold to the hydrogen site [15]. The same measurements locate the filled Lu levels as sharp features near eV; being a closed, chemically inert shell, the states do not participate in the bonding with hydrogen but distinguish lutetium from lanthanum, whose empty shell underlies the larger lanthanum lattice constant. The electronic structure of the lutetium dihydride has been characterized experimentally: angle-resolved photoemission on Lu–H–N have been interpreted as indicating a flat band eV below the Fermi level [?], which has been suggested to reflect strong correlation; whether the near- flat band is genuinely correlation-driven, however, remains an open question.
Further, theoretical studies of Sufyan and Larson [?] report that pristine , in its trigonal structure, hosts nontrivial topological features and van Hove singularities near the Fermi level, and Denchfield et al. [?] show that specific substitutions of nitrogen in the hydrogen sublattice generates flat bands and sharply peaked densities of states.
We show that electronic correlation in the rare-earth hydrides is governed by hydrogen occupation rather than interaction strength — the central result of this work, and one that inverts the usual expectation. Throughout, we use correlation strength to mean the magnitude of many-body renormalization of the one-particle properties, quantified by the mass enhancement and by the local spin susceptibility and instantaneous moment ; we reserve interaction strength for the magnitude of the screened on-site and obtained from cRPA. The two need not track one another, and in these materials they do not: hydrogen carries large cRPA interaction yet remains the weakly renormalized site. Moreover, two hydrogen sublattices are electronically inequivalent, with octahedral hydrogen the more strongly hybridizing of the two, and we find that specific arrangements of tetrahedral vacancies and octahedral occupancies give rise to narrow bands and van Hove singularities near .
The remainder of this paper is organized as follows. Section II presents the results: the electronic inequivalence of the tetrahedral and octahedral hydrogen sublattices, the mass enhancements and local moments obtained from DMFT, the DFT and DMFT densities of states, the effects of octahedral occupancy and hydrogen vacancies, and comparison with experiment. Section III discusses the origin of the hydrogen filling and its relation to the observed renormalization, and compares our results with spectroscopic measurements on lutetium and other lanthanide hydrides. The paper concludes with a summary of how hydrogen occupation controls the strength of correlation effects in rare-earth hydrides.
Results
Inequivalent Tetrahedral and Octahedral Hydrogen Sublattices
All LnH structures considered here (Ln = Lu, La) are built on the cubic fluorite (CaF-type) framework, space group . The metal atoms form a face-centered cubic (FCC) sublattice on the sites, and hydrogen occupies the tetrahedral sites, each coordinated by four metal atoms. This accounts for two hydrogen per formula unit, giving the dihydride LnH. The remaining octahedral interstitial, the site and coordinated by six metal atoms, is vacant in the ideal fluorite structure; progressive filling of this site carries the composition from toward the trihydride LnH, in which both hydrogen sublattices are fully occupied. We begin by asking whether the two hydrogen sublattices in LnH are electronically equivalent. Figure 1 shows the model structure for LuH, with tetrahedral hydrogen (H) in pink and octahedral

Figure 1. Idealized fluorite structure of LuH. Tetrahedral hydrogens (H) are represented by pink atoms, octahedral hydrogens (H) are represented by blue atoms and Lutetium atoms are represented by turquoise atoms.
hydrogen (H) in blue.
To characterize the two hydrogen sites electronically, we extracted their on-site energies, for LnH, from the diagonal elements of the Wannier Hamiltonian (Table 1). The tetrahedral and octahedral hydrogen sublattices are found to be electronically inequivalent, differing substantially in their on-site energies (Table 1). For LuH, the tetrahedral H- level lies 0.97 eV below the octahedral H- level, with both situated well below the Fermi level. The Lu- on-site energies, by contrast, lie above , split by the crystal field into two sets at and eV, well separated from the hydrogen manifold. The tetrahedral hydrogen, fully coordinated and filled already in the dihydride, is the more strongly bound and more deeply lying species; the octahedral hydrogen—the “extra” hydrogen added on going from the di- to the trihydride—is more weakly bound and sits closer to .
The ordering is reversed in LaH, where the octahedral level lies 0.38 eV below the tetrahedral one. The two compounds differ substantially in lattice constant (5.01 versus 5.65 Å), and the octahedral interstitial, being the larger and less tightly confined site, is expected to be the more sensitive of the two to this difference; the reversal is consistent with the site-resolved occupations, which show the octahedral hydrogen to be the more filled species in LaH and the less filled one in LuH. We do not attempt to separate the geometric effect from possible chemical differences between the two rare earths.
Cubic trihydrides: interactions do not open the gap
LaH and YH are semiconductors [5], whereas standard band theory predict these trihydrides to be metallic; hydrogen ordering [12, 13] and electronic correlation [10, 11] have both been proposed as the missing ingredient to open the band gap. Figure 2 tests the correlation scenario directly.

Figure 2. Hubbard interactions alone do not gap the cubic trihydrides. Orbital-resolved DOS of LuH (left) and LaH (right) at the Wannier level (, top), from DMFT at the cRPA couplings (middle), and from DMFT with eV at fixed hydrogen interactions (bottom).
At the Wannier level (, top row) LuH and LaH are low-density-of-states metals: lies in a shallow minimum with and states/eV respectively, produced by residual overlap between the H-derived valence weight and the RE- conduction manifold, and part of the H- weight remains unoccupied.
Applying correlation at the cRPA couplings (middle row) rearranges the spectrum substantially. The static, orbital-dependent shift of drives the occupied H-derived weight below eV and removes the unoccupied H- admixture, leaving a wide region of negligible spectral weight beneath . Both compounds nonetheless remain metallic: the is just above the conduction manifold with the total density of states and states/eV and a finite occupation per rare earth atom. Raising to 8 eV, nearly twice the cRPA value (bottom row), leaves this unchanged, only shifting the states below by a little amount due to the remaining unoccupied H orbitals.
The occupations identify this orbital dependent correlation effect. Hydrogen is a nearly filled shell (–) and the manifold is nearly empty (– of ten spin-orbitals); the insulating state these compounds are expected to reach is therefore the ionic RE(H) band insulator, requiring and . The two sets are themselves well separated: in LuH the doublet lies at eV and the triplet at eV, and in LaH at and eV respectively. The set therefore lies below in both compounds—the inverse of the familiar octahedral ordering, and the expected result for the eight-fold cubic coordination of the metal by tetrahedral hydrogen in the fluorite structure.
Hubbard interactions push the system toward the insulator limit but cannot complete the charge transfer, because the residual occupation is sustained by the strong – hybridization. Doubling accordingly changes nothing. It however is possible to open the gap through distorting the hydrogen sublattice, consistent with DFT+ and quantum Monte Carlo results in which hydrogen distortions open a small gap that interactions subsequently enlarge [?]—which motivates the vacancy-
and occupancy-resolved analysis of the following sections.
Correlation and Vacancy Effects
We now consider the electronic structure of LuH at five compositions, , 2, 2.25, 2.875, and 3. The compositions studied here therefore span partial vacancy of the tetrahedral sublattice below and partial filling of the octahedral sublattice above it. Figure 3 shows the DFT-relaxed structures for the compositions considered, together with their band structures and densities of states; the sub-stoichiometric compounds, in which tetrahedral sites are partially vacant, develop flat bands immediately below the Fermi level. The DFT-optimized lattice parameters and the cRPA interaction parameters for all compositions are given in Sections S2.1 and S2.2, respectively.

Figure 3. Lu-H structures with different hydrogen occupation of and studied and their associated DFT calculated electronic properties. For 4 Lu atoms: a) LuH has 7 . b) LuH has 8 . c) LuH has 8 and 1 . d) LuH has 8 and 3 . e) LuH has 8 and 4 .
Across the LuH series the hydrogen shell is close to being fully filled at both sites (–), consistent with an H-like anion, but it is not a closed shell: the deficit from is recovered almost exactly as excess occupation (Table 18, ), establishing that the two sublattices are hybridized rather than ionically decoupled. The hydrogen sublattices are themselves re-
| Channel | Quantity | undoped | ||
| Lu- | 1.158 | 1.417 | 1.670 | |
| Lu- | 0.1513 | 0.2108 | 0.2346 | |
| Lu- | 0.0375 | 0.0389 | 0.0391 | |
| Lu- | 4.04 | 5.42 | 6.01 | |
| H | 1.849 | 1.848 | 1.847 | |
| H | 0.0045 | 0.0045 | 0.0046 | |
| H | 0.0040 | 0.0040 | 0.0041 | |
| H | 1.12 | 1.13 | 1.13 | |
| H | 1.725 | 1.720 | 1.720 | |
| H | 0.0125 | 0.0129 | 0.0130 | |
| H | 0.0096 | 0.0098 | 0.0099 | |
| H | 1.30 | 1.31 | 1.31 |
Table 18. Response of the correlated subspace to rigid-band doping in LuH (7 tet + 1 oct). One electron is added to (+1e) or removed from (−1e) the total electron count in the DMFT step at fixed structure, corresponding to doping. Occupations and local spin and charge susceptibilities , from the CTQMC solver at . is a total-shell quantity and is therefore not directly comparable between the five-orbital manifold and the single-orbital hydrogen sites.
solvably inequivalent. At every composition containing octahedral hydrogen, – and –, which follows directly from the on-site energies of Sec. II A: the shallower octahedral level is less completely occupied than the deeper tetrahedral one.
Despite the predominantly closed-shell character, residual charge fluctuations are substantial: –, which corresponds to non-closed-shell configuration 15–27% of the time. The associated instantaneous moment reaches for the octahedral sites, versus 0.145–0.19 on tetrahedral ones. For a single orbital these are not independent diagnostics: identically, and with throughout, the moment is a direct readout of the charge deviation from the ionic limit. Local charge and spin fluctuations therefore measure the hybridization strength site by site, and are largest where the on-site level lies closest to .
The hydrogen mass enhancement increases systematically with octahedral occupation. At the tetrahedral site it rises monotonically from in LuH through 1.22 in LuH and 1.29 in LuH to 1.34 in LuH, and the octahedral site reaches 1.36 in LuH. These values of mass enhancement are comparable to the kagome metal ScVSn () [?] and well below the value for moderately correlated oxides such as SrVO () [?].
The Lu- manifold is similarly weakly correlated but follows a different trend. Its occupation decreases monotonically with hydrogen content, from in LuH to 0.68 in LuH, as hydrogen withdraws charge from the band. The mass enhancement, however, is non-monotonic: it peaks near the dihydride (– for LuH) and decreases toward the trihydride (1.16–1.18 in LuH). The two crystal-field-split -sets track each other closely at all compositions.
The local susceptibilities distinguish the two channels more sharply than the static moments alone. The instantaneous moment measures the magnitude of the local moment at a given instant; the local spin susceptibility additionally weights its persistence in imaginary time, and their ratio therefore measures how long-lived the moment is relative to its size. For hydrogen this ratio is at tetrahedral sites but 0.046–0.050 at octahedral sites: the octahedral moment is not merely larger but also longer-lived per unit moment. As we show below, this distinction is not incidental: octahedral and tetrahedral H sites occupy geometrically distinct positions relative to the Lu orbitals, and this governs which channel each couples to.
The spin and charge susceptibilities are moreover nearly equal on hydrogen (– at every site and composition), the signature of a site whose fluctuations are hybridization-driven rather than moment-forming, since local-moment behavior requires charge fluctuations to be suppressed relative to spin fluctuations. The shell behaves differently: falls from 4.6–5.4 near the dihydride to 1.7–1.9 at the trihydride, and itself collapses by a factor of four across the same range, tracking the emptying of the band. The manifold is thus moment-like — behaving as a localized, correlation-susceptible orbital — where it is close to half-filled, and becomes progressively itinerant or band-like as hydrogen withdraws its charge and the shell empties. Hydrogen, by contrast, remains itinerant and does not develop local-moment character at any composition, consistent with its H- band being governed by filling rather than by interaction strength. We note that is reported here at a single temperature ( eV); distinguishing Curie from Pauli behavior would require its temperature dependence.
In the ideal fluorite structure the Lu site has symmetry, and each hydrogen sublattice hybridizes with only one channel: tetrahedral H lies along , where the – two-center integrals vanish, and octahedral H along , where the – integrals vanish [?]. Since the H- level lies well below , the coupled set is pushed upward by antibonding repulsion while the uncoupled set is left unshifted. In the tetrahedral-only dihydride this raises and leaves essentially nonbonding, so charge accumulates preferentially in : against . (At intermediate compositions the octahedral sublattice is partially occupied and the site symmetry is reduced, so the separation is only approximate; it is exact in stoichiometric LuH and LuH.) Because the tetrahedral sublattice is occupied at every composition, the environment is fixed across the series; only the octahedral sublattice varies, and it couples to alone. Adding additional hydrogen to the octahedral site therefore acts almost exclusively on the channel, which acquires a hybridization partner of its own and is pushed up in turn. The occupations bear this out: from LuH to LuH, falls by a factor of four (0.49 0.12; 0.11 in LaH) while is nearly unchanged (0.12 0.11), closing the imbalance from to . Both sets remain far from the half-filled configuration throughout. The composition dependence of the -shell occupations thus follows from hybridization geometry and filling rather than from any change in interaction strength.
The local spin susceptibility tracks the occupation across the series: the compositions with have –, while those with have –. The -shell spin fluctuations therefore follow the filling of the channel that hydrogen depopulates, independently of any change in the interaction parameters. The trends above follow from changing the hydrogen content, which alters the filling and the hybridization geometry together. To separate the two we consider rigid-band doping at fixed structure, which changes the carrier count without touching either hydrogen sublattice (Table S10).
The correlated-subspace occupations show where the doped charge is accommodated (Table S10). Across the full range the Lu- occupation shifts by 0.51 electrons, while both hydrogen sites change by less than 0.005—below the resolution at which we quote them. The local susceptibilities behave the same way: on the shell rises by 55% from hole- to electron-doping and its ratio to the charge susceptibility climbs from 4.0 to 6.0, whereas the hydrogen susceptibilities and their ratios are unchanged to within 3% at both sites. Doping therefore acts exclusively on the manifold; the hydrogen sublattices, including the more strongly hybridized octahedral site, are spectators. The two perturbations agree: whether the filling is changed chemically, by octahedral hydrogen, or rigidly, by doping, on the shell follows the filling while the hydrogen sites do not respond.
The picture that emerges—a nearly filled but hybridized hydrogen shell, weak renormalization in both channels, and an octahedral environment (both the H site and the orbitals it couples to by symmetry) that is systematically the more strongly hybridized of the two coordination geometries—is one in which every site-resolved distinction traces back to occupation and orbital geometry rather than to interaction strength. This is the central claim of the present work in microscopic form, and it makes a direct prediction for the spectral function: features appearing near should be one-body in origin and controlled by octahedral occupancy rather than by interaction strength. We test this against photoemission in the following section.
Electronic Structure, Correlated Spectral Function, and Optical Response
To trace the band-structure origin of the configurational dependence observed in the photoemission and optical response (Figure 5, 6,), we computed the electronic band structures of four representative LuH configurations: (a) pure LuH with full tetrahedral occupancy, (b) LuH with one tetrahedral H removed, (c) LuH with one additional octahedral H, and (d) nominally LuH with one tetrahedral vacancy compensated by one octahedral occupant (see Figure 4).

Figure 4. Comparision of band structures of various LuH structures showing that flat-bands and van Hove singularity can be tuned near . The band structures are for the following structures: (a) LuH with only tetrahedral-H occupancy (b) LuH with one tetrahedral-H vacancy (c) LuH with one octahedral-H occupancy (d) LuH with one tetrahedral-H vacancy and one octahedral-H occupancy.

Figure 5. Correlated electronic structure of LuH₂ with Hₒct: theory–experiment comparison. Top: Wannier bands and DMFT spectral function along ––, undoped and at carrier doping. Relative to the Wannier bands, DMFT produces a narrow band pinned near with a saddle at , and damped flat features near eV; doping shifts both rigidly in energy (electron doping downward, hole doping upward) without displacing them in momentum. Bottom: ARPES cuts adapted from [?], licensed under CC by 4.0, showing the near- flat band with its emerging light band (left), compared with the undoped DMFT spectral function along – (middle) and the model structure of LuH₂ with Hₒct – teal atoms represent Lu, pink atoms represent Hₜₑₜ and blue atom represents Hₒct (right).
The pure LuH band structure (panel a) is the canonical fluorite-dihydride result: dispersive Lu-5d bands crossing along , with H-derived weight confined well below the Fermi level. The two H-sublattice modifications produce distinct changes near . Removing a tetrahedral H (panel d, LuH) introduces a substantially narrow band approximately 0.2 eV below along . Adding an octahedral H without removing a tetrahedral one (panel c, LuH) shifts the band crossing at downward onto , bringing a saddle-point-like feature into coincidence with the Fermi level. The combined configuration (panel b) exhibits both effects simultaneously: a weakly dispersive band near inherited from the tetrahedral vacancy, and a van Hove-like feature at shifted onto by the octahedral occupant.
The two effects therefore appear to act through different mechanisms. Tetrahedral vacancies introduce a localised defect-derived state whose narrow bandwidth reflects the loss of metal-hydrogen hybridization in the immediate vacancy environment. Octahedral occupants shift the Fermi level through electron donation while simultaneously perturbing the Lu-5d states they hybridize with, moving pre-existing dispersion features into coincidence with . The combination yields enhanced density of states at the Fermi level relative to the pure dihydride.
The states introduced by tetrahedral vacancies and octahedral occupants are narrow, and it is for narrow bands that correlation can least safely be assumed away, being largest for precisely the weakly dispersing features these modifications place near — and the ARPES flat band has itself been interpreted as a correlation feature [?]. Whether such a band is correlated in the many-body sense cannot be decided from the band structure alone, nor from a static Hubbard correction, which shifts levels but cannot distinguish a band reconstructed by a static self-energy from one narrowed by a small quasiparticle residue. We therefore turn to DMFT, which treats the frequency dependence of the self-energy explicitly.
Correlation neither removes nor shifts these features. The flat band at and the feature eV below it survive the inclusion of dynamical correlations essentially unchanged, so that both remain in agreement with the ARPES measurements, complementing the agreement already obtained for the optical conductivity. The self-energies underlying this conclusion, and their decomposition into static shift, mass renormalization and damping, are given in Section S4.
Figure 5 compares the Wannier bands of LuH containing octahedral hydrogen with the DMFT spectral function along . Because both are built from the same Wannier Hamiltonian, every difference between them is attributable to the self-energy. Three effects are visible, and we keep them separate throughout.
The first and largest is static. The orbital-dependent repositions the hydrogen-derived states relative to the Lu- manifold, and where the resulting weakly dispersing level intersects bands of predominantly Lu- character, hybridization quenches their dispersion. The clearest consequence is along : a band that disperses appreciably in the Wannier spectrum becomes a narrow, nearly flat feature pinned close to the Fermi level. At this narrow band forms a saddle, hole-like along one in-plane direction and electron-like along the other, lying within a few tens of meV of .
The second effect, dynamical mass renormalization, is comparatively weak. The Matsubara self-energies give for either of or orbitals, bounding the correlation-induced bandwidth reduction to at most . The observed flattening of the near- band substantially exceeds this bound, identifying it as a reconstruction of the band structure by the static, orbital-dependent part of the self-energy rather than as a mass-enhancement effect.
The third effect is damping. A second set of flat features appears near eV, but with markedly lower intensity and larger energy width than the near- band. These features are flat in the dispersive sense; what distinguishes them is lifetime, since grows away from the Fermi level and broadens them into low-intensity ridges. The narrow band near remains sharp and intense by contrast, protected by the vanishing of at low frequency.
The doped spectra differ from the stoichiometric one by an essentially rigid displacement in energy. With each spectrum referenced to its own Fermi level, electron doping raises the chemical potential and the narrow band and its saddle appear at lower energy, moving further below ; hole doping shifts them upward toward and through . No displacement in momentum accompanies these shifts, as required for a momentum-independent self-energy, and the dispersions, intensities and linewidths are unchanged within our resolution. Linear extrapolation of the saddle energy places it at the Fermi level for a hole doping of .
For the LuH with H structure we also calculate the in-plane and out of plane energy distribution curves (EDC) and compare with EDC curves obtained from ARPES experiment (Figure S3). [?] The construction of such curves, which introduces no adjustable parameters beyond the experimental temperature, energy resolution, and escape-depth momentum broadening, is described in Section S4.3.

Figure S3. Correlated electronic structure of LuH with octahedral hydrogen: theory–experiment comparison. In-plane and out-of-plane energy-distribution curves, adapted from [6], licensed under CC by 4.0, and calculated. Both show a sharp, momentum-independent peak near ; deeper structure is dispersive in-plane and comparatively flat out-of-plane. Momenta are not registered between calculation and experiment throughout; the comparison is made on energies and band shapes only (see text).
In plane, a sharp maximum sits immediately below at fixed energy across the cut, with deeper structure that shifts markedly from curve to curve—the distinction drawn in the measured curves between a flat band and the light band emerging from it, with the calculated near- maximum at meV against a measured flat-band binding energy of – eV. Out of plane, both show a stationary maximum just below , persisting across photon energies spanning more than one Brillouin zone along ; in the calculation this peak is stationary to within meV over the full range. The calculation additionally resolves a second out-of-plane maximum near eV, dispersing by eV, which appears in the measurement not as a distinct peak but within the broad enhancement between and eV; we attribute this to momentum integration over the escape-depth window and to the ordered octahedral sublattice of our supercell, which produces coherent minibands where a disordered sample would show an incoherent resonance at the same energy (Sec. S4.3). The comparison is therefore one of feature count, approximate energy, and dispersive character; peak intensities are not compared, since the measured curves carry photoemission matrix elements and an inelastic background absent from . Likewise, the calculated DMFT DOS for LuH with H exhibits a pronounced peak at the Fermi level and a broad structure centred near eV, separated by a shallow minimum (Figure S4). The Fermi-level peak is the density-of-states signature of the narrow band identified above: a band of width concentrates its spectral weight over that width, so a band of eV width at necessarily produces a peak of this kind. Its orbital decomposition is dominated by Lu- states, with the set carrying the larger share and hydrogen weight negligible in this window, the latter having been displaced to higher binding energy by the static shift.
!Photon energy dependent DOS adapted from [6] and calculated DMFT DOS bbox=0.17,0.09,0.84,0.42
Figure S4. Photon energy dependent DOS adapted from [6] licensed under CC by 4.0 (left), and calculated DMFT DOS (right)
Taken together, these results show that the correlated orbital creates a flat band it does not itself occupy. Correlation on the hydrogen states is substantial but almost entirely static: displaces the H-derived weight to higher binding energy, while leaves the quasiparticle residue close to unity. The narrow band remaining at is therefore Lu- in character, flattened where the displaced H- level crosses it—a hybridization effect driven by a correlation-induced shift, not a heavy quasiparticle. The same self-energy accounts for both photoemission features, the sharp Lu- peak at protected by the vanishing of and the broad, damped H- structure near eV. Agreement with the measured flat-band binding energy improves under electron doping. Because the mechanism is a level alignment rather than a mass enhancement, its strength is set by where the hydrogen sits and how much of it is present—which is what the comparisons that follow address.
These single-particle differences appear directly in the optical response. Figure 6 shows the absorptivity , computed from the independent-particle dielectric function (Figure S4), for the tetrahedral-only series [panel (b)] and for the octahedrally compensated LuH₂ [panel (a)], together with the measured spectra of Weaver et al. [14]. Configurations (a) and (d) share the same stoichiometry, so any difference between their spectra is structural rather than compositional.

Figure 6. Calculated optical absorption spectra of LuHₓ structures compared with experiment. (a) Calculated absorptivity for a LuH₂ structure in which one hydrogen occupies an octahedral site, compared with experimental data for LuH₁.₉₈, which also shows weak octahedral occupation. Arrows mark the low-energy feature associated with transitions involving octahedral hydrogen: orange for the calculated LuH₂ curve and green for the experimental LuH₁.₉₈ curve. This feature is absent in the tetrahedral-only structures of panel (b). (b) Calculated absorptivity for LuH₂, LuH₁.₈₇₅, and LuH₁.₇₅, which contain only tetrahedral hydrogen, compared with experimental data for LuH₁.₈₃, which is likewise tetrahedral-only.
In pure LuH, falls below unity between 1.0 and 2.2 eV (Figure S4): the tetrahedral-only band structure supports no direct transitions in this window, hydrogen weight lying several eV below . Within the window crosses zero at 1.05 eV with of order unity, so the loss function reaches order unity and the absorptivity shows the corresponding sharp plasma edge—a single maximum of 0.96 at 1.05 eV over an essentially flat sub-edge background, the canonical Drude-plus-interband response of a clean CaF-structure dihydride. Tetrahedral vacancies (LuH, LuH) progressively broaden this edge, raise the low-energy background, and shift the maximum toward 1.7–2.0 eV, but introduce no new structure below it.
Octahedral occupancy does. The compensated structure develops an occupied H-derived peak near eV in the projected density of states (Figure S2), with no counterpart in any tetrahedral-only configuration. Transitions from this state into the unoccupied Lu-5d manifold set in near 0.65 eV and produce an oscillator peaking at 1.05 eV, consistent with the unoccupied DOS maximum at eV. The zero of moves down to 0.57 eV, where now and the loss function is reduced to ; the single sharp maximum is replaced by two damped ones, at 0.7 and 1.75 eV, separated by a minimum near 1.15 eV, and the integrated spectral weight below 1 eV substantially exceeds that of any tetrahedral-only spectrum. Both maxima fall at minima of rather than at its peaks: for the reflectivity tracks , so is largest where is smallest, and neither maximum marks a transition energy.
Discussion
We now examine the physical origin of the trends established above. We first consider how the hydrogen on-site energies fix the sublattice occupations, and how the resulting filling — together with the screening environment set by the Lu-d manifold — determines the correlation strength on hydrogen. We then compare our results with optical and photoemission measurements on lutetium and related lanthanide hydrides, and assess what they imply for the electronic character of the flat bands near the Fermi level.
Origin of Filling and Correlation
The on-site energies provide a direct, one-body explanation for the hydrogen filling that underlies our central result. An orbital’s occupation is set by its energy relative to the Fermi level, and in LuH both H- sublattices lie 3–4 eV below : the hydrogen shells are therefore filled and anionic (H-like), consistent with the near-full DMFT occupation () and with photoemission evidence for charge transfer to hydrogen [15, ?].
The Wannier analysis further reveals the electronic inequivalence of the two hydrogen sublattices: the octahedral H- level lies eV above the tetrahedral one, identifying the octahedral hydrogen as the more weakly bound, shallower species. This is significant because the octahedral hydrogen is the common actor across the phenomenology of these materials: it controls the carrier density probed optically [14], it is the Kondo-like impurity of the breathing-hydrogen picture [10, 11], and it is the sublattice whose distortion opens the gap in cubic LuH [?]. Our calculation shows directly that, for lutetium hydrides, it is the hydrogen level closest to —the least deeply bound and hence the most likely to be electronically and structurally active—while the tetrahedral hydrogen, filled already in the dihydride, lies deeper and is correspondingly inert.
The susceptibilities sharpen this picture and provide the most robust site-resolved discriminant in our data. In every Lu composition the octahedral hydrogen carries 1.8–2.8 times the local spin susceptibility of the tetrahedral hydrogen, without exception—including LuH, where the mass enhancements of the two sites are inverted relative to the other compositions. Because is a static thermodynamic average rather than a low-frequency derivative of a stochastic quantity, it is the less noise-sensitive diagnostic, and we take the susceptibility ordering as the more reliable statement of site inequivalence. Its ordering follows the on-site energies directly: the shallower octahedral level supports larger and longer-lived local fluctuations than the deeper tetrahedral one. The reversal in LaH, where the octahedral level lies below the tetrahedral one, is reproduced in the susceptibilities as well—octahedral falls below tetrahedral—so that the level ordering, the occupations, the local moments, and the susceptibilities all identify the same sublattice as the electronically active one in each compound. That four independent diagnostics agree indicates the site hierarchy is a robust feature of the electronic structure rather than an artifact of any single observable.
The level alignment directly rationalizes the weak hydrogen correlation. A shell buried well below is filled and far from the half-filled configuration where the interaction could localize it; the large H- interaction therefore acts only as a static level shift rather than a mass renormalization, exactly as found in our DMFT self-energies. Level position, filling, and correlation thus form a single causal chain: the host sets the hydrogen level, the level sets the filling, and the filling—not the magnitude of —controls the correlation. This is the filling-control principle familiar from Hund’s metals and orbital-selective correlation in -electron systems [?, ?], operating here in its simplest setting: a single orbital with negligible Hund’s coupling. The same principle couples the two manifolds—the hydrogen acts as the electron reservoir that sets the -band filling and thereby its (weak) correlation—as detailed in Section 3C.
The rigid-band doping calculations make the reservoir behavior explicit. Adding or removing one electron per formula unit shifts the Lu- occupation by the full doped charge while the hydrogen occupations, moments, and local susceptibilities are unchanged—at the octahedral site as well as the tetrahedral one, even though the octahedral hydrogen is the more electronically active of the two. The hydrogen shell is thus closed not only in its static occupation but in its response: charge introduced at the Fermi level is absorbed entirely by the manifold. This is the dynamical counterpart of the deep, filled anionic level identified in the on-site energies, and it is the condition under which the rigid-band comparison with photoemission (Secion II D) is justified—the hydrogen-derived states near shift with the chemical potential but do not themselves reconstruct. The susceptibility, by contrast, grows steadily with filling and becomes increasingly spin-dominated ( rising from 4.0 to 6.0), the same trend seen across the stoichiometric series as hydrogen content is reduced. The two routes—varying hydrogen content at fixed electron count, and varying electron count at fixed hydrogen content—thus give a consistent account of which manifold is compressible.
The -channel behavior follows from the crystal-field structure of the fluorite lattice. The metal is coordinated by eight tetrahedral hydrogen at the cube corners rather than by six ligands octahedrally, and the cubic field inverts the familiar ordering: , whose lobes point between the ligands, lies below . The set consequently fills first and sits closer to half filling, carrying the larger per-orbital moment. Adding octahedral hydrogen introduces a competing field of the opposite sign—the 4b site is octahedrally disposed with respect to the metal—which partially cancels the cubic splitting; this accounts for the narrowing of the occupation contrast from the dihydride to the trihydride.
Because both sets lie below half filling throughout, the spin susceptibility is on the rising side of its filling dependence, and the shell moves away from the moment-forming regime as hydrogen content increases. The rigid-band doping calculations confirm this directly: adding electrons raises and monotonically with no turnover (Table 18), as expected when approaching half filling from below. The stoichiometric series and the doping series therefore describe the same dependence traversed in opposite directions, and identify the occupation—set by hydrogen content—as the variable controlling the -shell spin response.
Comparision with Experiment
Angle-resolved photoemission on (nominally nitrogen-doped) lutetium hydride reports a flat band 0–200 meV below and a nearby van Hove singularity, interpreted as a signature of strong correlation [?]. Our results bear directly on the origin of this feature. The DMFT Lu-5 mass enhancement is modest (–1.3), whereas a correlation-driven flat band would require a heavy quasiparticle mass, . We therefore find that the near- flat band is unlikely to be primarily a correlation-renormalization effect, pointing instead toward a disorder-related (hydrogen-vacancy) or filling/Van Hove origin, consistent with the absence of such a band in pure-LuH DFT [?]. We note that the precise hydrogen and nitrogen content of the ARPES sample is not firmly established; if the sample is closer to LuH than to a nitrogen-doped composition, this conclusion is unaffected, since the argument rests on the magnitude of the computed renormalization rather than on the dopant.
Figure 6 compares the calculated optical absorptivity of LuH in four configurations against the experimental spectra of Weaver et al. [14] at and . The dielectric functions (Fig. S[E]) show why octahedral occupation, and only octahedral occupation, restructures the spectrum. The oscillator introduced by the octahedral H falls inside the transparency window of pure LuH—the same window that hosts its plasma edge. Filling the window has two consequences: the added interband screening shifts the zero of from 1.05 down to 0.57 eV, and at the crossing rises by an order of magnitude, collapsing the loss function. The sharp plasma edge of the dihydride is thereby eliminated, leaving a damped feature at 0.7 eV and a residual maximum at 1.75 eV in the part of the window that survives above the oscillator. Tetrahedral vacancies, by contrast, redistribute oscillator strength within the existing bands; they broaden the edge, but the window—and with it the single-maximum line shape—remains.
The measured spectra display the same dichotomy. LuH shows a sharp edge and no sub-eV structure, consistent with predominantly tetrahedral occupation near the dihydride phase boundary; its maximum at 2.05 eV is bracketed by the calculated LuH (1.05 eV) and LuH (1.85 eV). Near-stoichiometric LuH shows an additional shoulder at 0.3–0.5 eV absent from LuH, in the region where the octahedral calculation peaks, though at somewhat lower energy and with far smaller amplitude: against the calculated , so any octahedral occupancy in the measured sample must be small. Two caveats attend the assignment. A Drude term with a finite scattering rate produces an inflection of similar shape with no interband contribution at all, and a firm identification would require the calculated absorptivity followed as a continuous function of octahedral filling, together with an independent structural determination of the site occupancies. A systematic offset of the calculated edge relative to experiment is also apparent—1.05 eV for the all-tet reference against the measured eV; we discuss its possible origins in Sec. ??, and the interpretation here rests on the relative behavior across configurations, which is reproduced unambiguously.
The configurational comparison in Figure 6 provides a direct band-structure-level confirmation of the inference first made by Weaver et al. [14] that the sub-eV opti<lemma cal feature observed at near-stoichiometric compositions arises from octahedral H occupation rather than from tetrahedral disorder. The crucial control is the comparison between LuH₁.₇₅ and LuH₁.₈₇₅ (both tetrahedral-only) and LuH₂ with a single octahedral occupant. Tetrahedral vacancies smear the plasmon edge but do not introduce a new oscillator anywhere below 1.5 eV; octahedral occupation does. This separates site type from H content as the controlling variable—a separation that the original optical study could only approach indirectly through the quench-versus-slow-cool experiment on YH₁.₈₉ and through the systematic lattice-constant trend across Sc, Lu, Y.
It is worth recalling how the octahedral assignment was established historically, since the present calculation completes a chain of reasoning begun nearly half a century ago. Weaver et al.’s 1979 optical inference [14] was supported on the theoretical side by the companion self-consistent band calculation of Peterman et al. [?], which showed that tetrahedral-only CaF₂-structure YH₂ has no interband transitions in the 0.5–1.5 eV window, establishing that the sub-eV optical features required states outside the tetrahedral picture. Direct structural confirmation came shortly afterwards: proton NMR rigid-lattice second moments found 10–15% octahedral occupation in YH₁.₉₂ and YH₁.₉₈ [?], and neutron diffraction and inelastic scattering subsequently tracked the temperature dependence of the octahedral occupancy in YH₂.₀ and identified its optic-mode signature [?]. For LuHₓ specifically the direct structural evidence is sparser, and the present supercell calculation occupies a role analogous to Peterman’s but in the affirmative: where the tetrahedral-only band structure could only exclude certain transitions, we include an octahedral occupant and show spectroscopically that it produces the observed feature.
The dichotomy between tetrahedral and octahedral hydrogen is rooted in local coordination. In the fluorite structure, tetrahedrally coordinated H sits in a four-fold metal environment with comparatively short metal–hydrogen distances and strong – hybridization, which pulls the tetrahedral bonding band well below . Octahedrally coordinated H has six metal neighbors at longer distance and weaker hybridization, placing its bonding combination closer to . This ordering—tetrahedral deep, octahedral shallow—is the one we obtain for LuH₃, and it is the ordering established in the early rare-earth trihydride band-structure literature [?, ?, ?]. Both sites lie well below the Fermi level, consistent with the universal H⁻ level-alignment framework of Van de Walle and Neugebauer [?].
Coordination is not, however, the only variable. The octahedral interstitial is the larger and less tightly confined of the two sites, and its level is correspondingly more sensitive to the cell volume. In LaH₃, whose lattice constant exceeds that of LuH₃ by more than 10%, this sensitivity is sufficient to reverse the ordering, placing the octahedral level below the tetrahedral one (Table I). The tet/oct hierarchy therefore reflects a competition between coordination-driven hybridization, which favors a deeper tetrahedral level, and the volume dependence of the octahedral site; it is not a fixed property of the fluorite structure.
This level alignment is precisely the hierarchy revealed by our on-site Wannier energies in LuH₃, where the tetrahedral H 1s level sits at eV and the octahedral H 1s level at eV with a 0.9 eV splitting in the expected direction (Sec. II A). The on-site energies set the hierarchy, not the transition energies: hybridization with the Lu-5d manifold carries octahedral-derived weight up to the occupied peak at eV seen in the supercell density of states (Fig. S2), and it is transitions from this hybridized manifold into the unoccupied Lu-5d states that produce the oscillator at 1.05 eV. Tetrahedral-derived weight remains several eV below even after hybridization, so its transitions lie far outside the Weaver window. The sub-eV structure in our supercell is therefore a direct spectroscopic signature of the shallow octahedral manifold, not a generic stoichiometry effect.
The systematic dependence on the host metal across the trivalent dihydrides reinforces this picture. The octahedral hole grows with the lattice constant in the sequence ScH ( Å), LuH (5.03 Å), YH (5.20 Å), and the spectroscopic signature of octahedral occupation follows in lockstep: absent in ScH, weak near 0.4 eV in LuH, and strong with two distinct features at 0.35 and 1.25 eV in YH [14]. This is consistent both with the steric argument (larger octahedral holes are easier to occupy) and with the bonding argument (the octahedral H bonding level rises further toward as the metal-hydrogen octahedral distance lengthens, increasing oscillator strength for transitions into the conduction manifold).
This optical interpretation closes the loop on the picture assembled from our cRPA, on-site-alignment, and mass-enhancement analyses, which together identify the octahedral hydrogen sublattice as the electronically active component of LuH in the near-stoichiometric regime. The screened interaction is systematically lower than in the H-s-only subspace and inverts upon enlargement of the active space to include Lu- screening, identifying the octahedral H as the more strongly coupled to the metallic conduction electrons; the on-site Wannier energy places it within hybridization range of the Fermi level; and the optical response now supplies an independent, experimentally anchored confirmation that the spectroscopic fingerprint of stoichiometry-driven changes in LuH near resides in transitions involving this shallow octahedral manifold.
Conclusions
We investigate electronic correlation in the rare-earth hydrides LuH (–3) together with other LnH compound using dynamical mean-field theory (DMFT) with the Hubbard values obtained from constrained random-phase approximation (cRPA). Our studies reveal that hydrogen-derived orbitals has much larger screened on-site Coulomb interactions than the Lu states. A joint analysis of Wannier spread functions and interaction parameters (see discussion in Section S2.2) demonstrates that these trends originate from the interplay of orbital localization and screening efficiency as evidenced by the pronounced dependence of on the choice of correlated subspace. The screened hydrogen interaction rises with hydrogen content at nearly fixed lattice constant, driven by the progressive anionic conversion of the octahedral hydrogen, which depletes the conduction-electron screening; this is the same carrier-density reduction observed experimentally as the decreasing Drude plasma frequency across the series [14], here manifested in the interaction parameters.
Despite carrying the largest interaction in the system, the hydrogen is weakly correlated. In hydrides LuH (–3) and LaH, DMFT calculations indicates that these materials are weakly correlated metals; within this weakly correlated regime, however, the correlation is strongly filling-controlled. The LuH hydrogen is a nearly filled, closed-shell anion with a deep on-site level consistent with the photoemission evidence for charge transfer to hydrogen [15], whose large interaction is inert because the filled shell admits no charge fluctuations. The Lu- band correlation in LuH decreases as octahedral hydrogen dilutes the filling. The hydrogen content thus emerges as the principal tuning parameter for correlation in these materials—it is the electron reservoir that sets the filling of every manifold.
These results bear directly on the experimental phenomenology. Our calculations indicate that the flat bands reported near by photoemission [?] arise from the presence of H and H vacancies. At the same time, tetrahedral vacancies and octahedral occupancy can be tuned to place narrow bands and van Hove singularities near , so that hydrogen stoichiometry offers a structural route to engineering the low-energy density of states. The 0.9 eV splitting we compute between the tetrahedral and octahedral hydrogen levels, and the correlation contrast we predict between the lutetium and lanthanum trihydrides, are in principle accessible to photoemission and optical spectroscopy.
The broader lesson is a simple organizing principle for correlation in hydrogen-rich materials: the host band structure sets the hydrogen level, the level sets the filling, and the filling—not the magnitude of —controls the correlation. A site-resolved treatment distinguishing the two hydrogen sublattices in LaH, where the hydrogen is correlation-active, is the natural next step and would test the site-selective correlation this principle predicts.
Acknowledgments
This research was supported by the National Science Foundation (NSF) (DMR-2104881, A.L., R.J.H.), the Department of Energy (DOE) National Nuclear Security Administration (NNSA) through the Chicago/DOE Alliance Center (DE-NA0004153; R.J.H.), the DOE Office of Science (SC) (DE-CO0020340, R.J.H.), and NSF SI2-SSE Grant 2513657 (H.P.). P. G. (mentorship, analysis, writing) and H. S. (mentorship, analysis, writing) were supported by the DOE SC, Basic Energy Sciences (BES), Materials Sciences and Engineering Division, as part of the Computational Materials Sciences Program and Center for Predictive Simulation of Functional Materials. The authors gratefully acknowledge the Advanced Cyberinfrastructure for Education and Research (ACER) group at the University of Illinois Chicago for providing computational resources and services needed to deliver the research results presented in this paper. Similarly, this research used resources of the National Energy Research Scientific Computing Center (NERSC), a DOE SC User Facility located at Lawrence Berkeley National Laboratory, operated under Contract No. DE-AC02-05CH11231 using NERSC award BES-ERCAP0023615. This research used resources of the Argonne Leadership Computing Facility, which is a DOE SC User Facility supported under contract DE-AC02-06CH11357. This manuscript has been authored by UT-Battelle, LLC under Contract No. DE-AC05-00OR22725 with the DOE. Part of the DFT-based research was conducted as part of a user project at the Center for Nanophase Materials Sciences (CNMS), a DOE SC User Facility at Oak Ridge National Laboratory. Contributions to the manuscript were also sponsored by UT-Battelle, LLC under the same contract. The U.S. Government (USG) retains, and the publisher, by accepting the article for publication, acknowledges that the USG retains, a non-exclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for USG purposes. The DOE will provide public access to the results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan).
References
Computational details
First-principles calculations were performed within the framework of density functional theory (DFT) using the Vienna Ab initio Simulation Package (VASP) [4]. The projector augmented-wave (PAW) method [2] was employed to describe the interaction between valence electrons and ionic cores, and the exchange–correlation effects were treated within the generalized gradient approximation (GGA) in the Perdew–Burke–Ernzerhof (PBE) parametrization [9].
The plane-wave basis set was truncated at an energy cutoff of eV. Brillouin-zone integrations were performed using a -centered Monkhorst–Pack -point mesh of . Structural relaxations were carried out until the residual forces on each atom were smaller than eV/Å. Electronic self-consistency was achieved with an energy convergence criterion of eV.
Wannier basis construction
To construct the localized basis used for the effective low-energy model, maximally localized Wannier functions (MLWFs) were generated using the Wannier90 code [8]. The Wannier functions were constructed from a set of Bloch states obtained from VASP through the VASP2Wannier interface. Initial projections were chosen according to the relevant atomic orbitals of the correlated subspace (e.g., Lu orbitals and hydrogen orbitals). The disentanglement procedure was performed within an energy window extending from eV to 10 eV relative to the Fermi level. The resulting Wannier functions reproduce the DFT band structure in the selected energy range with high fidelity.
Constrained Random Phase Approximation
The effective screened Coulomb interaction parameters were calculated within the constrained random phase approximation (cRPA) as implemented in VASP. In this framework, the fully screened interaction is obtained from the bare Coulomb interaction and the polarization function within the random phase approximation (RPA):
In order to construct a low-energy Hamiltonian for the chosen correlated subspace, screening processes arising from electronic transitions within that subspace must be excluded. The total polarization is therefore decomposed as
where corresponds to transitions within the correlated subspace and represents the remaining screening channels. The partially screened interaction defining the effective Hubbard interaction is then
The static limit of this interaction () provides the effective Hubbard interaction parameters used in model Hamiltonians:
where are the Wannier orbitals spanning the correlated subspace.
The matrix elements of the screened interaction were evaluated in the Wannier basis constructed from the DFT Bloch states. The calculated interaction matrix was subsequently used to extract the intra-orbital Hubbard interaction , the inter-orbital interaction , and the Hund’s coupling .
Density Functional Theory + Dynamical Mean Field Theory methodology
To capture the effects of strong electronic correlations beyond conventional density functional theory (DFT), we employed dynamical mean-field theory (DMFT) in combination with a Wannier-based low-energy Hamiltonian. The DFT+DMFT calculations were performed using the DMFTwDFT framework [11], which interfaces DFT codes with many-body solvers via Wannier90.
In this approach, a correlated subspace is constructed from a set of localized Wannier orbitals, which provide an effective low-energy Hamiltonian,
where is the Wannier-projected non-interacting Hamiltonian, describes the local Coulomb interactions within the correlated subspace, and is the double-counting correction accounting for interactions already included at the DFT level.
Within DMFT, the lattice problem is mapped onto a self-consistent quantum impurity problem, in which the local Green’s function is given by
where is the local self-energy and is the chemical potential. The impurity problem is solved iteratively to obtain , which is then used to update the lattice Green’s function until convergence is achieved.
The interacting Hamiltonian is expressed in terms of effective Hubbard interactions (, , ), obtained from constrained random phase approximation (cRPA) calculations. The DMFT impurity problem was solved using the continuous-time quantum Monte Carlo (CTQMC) method in the hybridization expansion, as implemented in the DMFTwDFT package.
Charge self-consistency between DFT and DMFT was achieved by computing the DMFT density matrix and feeding it back into the DFT calculation, ensuring a consistent treatment of electronic correlations at both the one-particle and many-body levels [11]. This iterative loop continues until convergence in both the charge density and self-energy is reached.
Finally, the momentum-resolved spectral function was obtained from the analytically continued Green’s function,
Correlation diagnostics: self-energy, quasiparticle weight, and local moments
We characterize the strength of electronic correlation using three complementary quantities extracted from the converged DMFT solution: the quasiparticle mass enhancement obtained from the Matsubara self-energy, the per-atom occupation, and the instantaneous local moment obtained from the CTQMC impurity solver’s atomic-state probabilities.
Quasiparticle weight and mass enhancement. Near the Fermi level, the real part of the self-energy can be linearized as , which defines the quasiparticle weight
For local/momentum-independent self-energy, is exactly the inverse mass enhancement, . We have:
Occupation and local moment. The per-atom occupation of each correlated orbital is obtained from the impurity solver and cross-checked against the trace of the local Green’s function. For a single-orbital impurity we have the probabilities of the empty, singly-occupied (spin-up and spin-down), and doubly-occupied configurations, , , , and , from which the occupation and the instantaneous local moment follow directly,
Double counting. The double-counting correction removes the portion of the local Coulomb interaction already present, in a mean-field sense, in the DFT exchange-correlation functional, and must be subtracted to avoid counting the same interaction twice. We employ the fully localized limit (around-mean-field) form,
evaluated self-consistently using the DMFT-derived occupation of the correlated orbital at each charge-self-consistency iteration, so that the double-counting potential tracks the converged correlated occupation rather than a fixed nominal value.
Supplementary Results
DFT structures and lattice trends
Figure 1 shows the Lu–H structures considered in this work. The stoichiometric end members LuH and LuH adopt the structure. Hydrogen-deficient structures were generated by removing H from octahedral sites for LuH and from tetrahedral sites for LuH, and all structures were relaxed within DFT; the optimized lattice parameters are summarized in tbl:S1 3.
All compositions relax to cubic or near-cubic cells, with and to within 0.02°. The lattice constant varies by less than 0.4% across the entire series and does so non-monotonically, with a shallow minimum near LuH. This near-invariance of the cell geometry is important for interpreting the interaction parameters that follow: because the lattice constant changes so little—and non-monotonically—across the series, any systematic, monotonic trend in the screened interactions with hydrogen content (Sec. 2.2.2) cannot be attributed to cell geometry and must instead be electronic in origin. We return to this point in the cRPA analysis.
Interaction Parameters
Dependence of interaction parameters on orbital character
We performed constrained random-phase approximation (cRPA) calculations of the effective Coulomb interaction parameters for three correlated subspaces: (i) RE- only, (ii) H- only, and (iii) the combined RE- + H- manifold. From the interaction matrices we extracted the intra-orbital Hubbard , inter-orbital , and Hund’s coupling (Table 5,Table 13 and following).
| in LuH | In Lu-d subspace | In H-s subspace | In H-s + Lu-d subspace | |||||||
| LuH | 4.45 | – | – | 8.89 | – | – | 9.1 | – | – | |
| LuH | 2.875 | 4.6462 | 1.4191 | 0.0518 | 7.7576 | 1.2005 | 0.0005 | 7.7057 | 1.4204 | 0.0236 |
| LuH | 2.25 | 3.4047 | 0.6362 | 0.0915 | 5.7296 | 0.1233 | 0.0036 | 5.7109 | 0.6628 | 0.0541 |
| LuH | 2 | 4.3008 | 0.8210 | 0.0952 | 6.5219 | 0.3624 | 0.0014 | 5.1666 | 0.5072 | 0.0564 |
| LuH | 1.75 | 3.7432 | 0.6625 | 0.0951 | 6.2499 | 0.1254 | 0.0013 | 4.5974 | 0.4535 | 0.0600 |
Table 5. cRPA parameters in Lu-d, H-s, and combined subspaces for LuH.
| System | Subspace | Site | |||
| LuH | H-s only | tet (all) | 6.25 | 0.13 | 0.001 |
| RE- only | – | 3.74 | – | 0.09 | |
| LuH | H-s only | tet (all) | 6.52 | 0.36 | 0.001 |
| RE- only | – | 4.30 | 0.82 | 0.095 | |
| combined | – | 5.17 | 0.51 | 0.056 | |
| LuH | H-s only | tet | 7.95 | – | – |
| oct | 7.25 | – | – | ||
| (avg) | 7.76 | 1.20 | 0.001 | ||
| RE- only | – | 4.65 | – | – | |
| LuH | H-s only | tet | 9.21 | – | – |
| oct | 8.25 | – | – | ||
| (avg) | 8.89 | 2.23 | 0.001 | ||
| RE- only | – | 4.46 | 1.67 | 0.102 | |
| combined | tet | 12.20 | – | – | |
| oct | 12.5 | – | – | ||
| (avg) | 9.19 | 2.76 | 0.045 | ||
| LaH | H-s only | tet | 8.76 | – | – |
| oct | 8.01 | – | – | ||
| (avg) | 8.51 | 2.11 | 0.001 | ||
| RE- only | – | 4.69 | – | – | |
| combined | – | 7.45 | – | – |
Table 13. Summary of cRPA screened interaction parameters across LuH stoichiometries and LaH, for the three correlated subspaces (RE- only, H-s only, and combined RE-+H-s). , , and are the averaged intra-orbital Hubbard, inter-orbital, and Hund’s coupling parameters. For the H-s channel in the trihydrides, site-resolved values are given for tetrahedral (tet) and octahedral (oct) hydrogen where both are occupied. All values in eV.
Across all stoichiometries the screened intra-orbital interaction on the H- orbitals is substantially larger than on the RE- orbitals. This is the opposite of the naive expectation that the more spatially compact orbital, or the rare-earth manifold often presumed more strongly correlated, should carry the larger . We find that two distinct factors must be separated to explain the hierarchy: orbital localization and screening.
The bare local Coulomb matrix element scales inversely with orbital extent, , so a more localized Wannier function yields a larger bare interaction. The H- Wannier functions are indeed more compact (spread Å) than the RE- functions (– Å), which is consistent with at the level of the bare interaction. However, localization alone does not control the screened interaction. The decisive test is the change of correlated subspace: enlarging the subspace from H- only to the combined RE- + H- manifold increases the H- Wannier spreads (from to Å), which on the geometric argument alone would reduce . The cRPA result is the opposite: increases (from – to – eV). This demonstrates that the screened H- interaction is governed primarily by the available screening channels rather than by orbital extent. In cRPA the screened interaction is , where the polarization excludes processes within the correlated subspace; enlarging the subspace to include RE- transfers the RE- polarization out of , reducing the screening of and raising it. Because the H- spread increases while also increases, the effect cannot be geometric and must be the loss of RE- screening.
The RE- sector behaves oppositely, and this asymmetry is itself informative. Enlarging the subspace removes the H- screening of RE-, yet barely changes (and slightly decreases, from – to eV). The H- manifold—filled, deep below , and weakly polarizable—is a poor screener of RE-, so removing it has little effect; the small residual decrease tracks a slight increase in the RE- Wannier spreads upon rehybridization. Thus RE- screens H- strongly, but H- screens RE- weakly: the screening is markedly asymmetric, consistent with RE- being the itinerant, polarizable manifold near and H- the localized, filled one well below it. Notably, in the combined subspace the RE- onsite becomes nearly uniform ( eV) even though the Wannier spreads retain their () structure—a further indication that is set by screening rather than orbital extent.
Taken together, the H- states are compact and weakly screened, giving large local interactions, while the RE- states are more extended and more effectively screened despite their multi-orbital character. The hierarchy is therefore driven primarily by differences in screening rather than by localization alone.
Table 5 compares the interaction parameters for the Lu and La hydrides. In the dihydrides, LuH shows larger , , and than LaH in both the RE- and H- sectors. Both crystallize in but with very different lattice constants (5.011 Å for , 5.655 Å for ); the smaller cell yields more compact Wannier functions and correspondingly larger interactions, so the dihydride comparison is governed by the lattice constant. This geometric ordering does not, however, carry over cleanly to the trihydrides: while remains slightly larger for than (8.89 vs 8.40 eV), the RE- interaction reverses, with now larger (4.69 vs 4.45 eV) despite its much larger cell. The RE- channel in the trihydride is therefore no longer controlled by geometry alone, reflecting the change in -band filling and screening octahedral occupation (Sec. S2.2).
Evolution of the screened interaction with stoichiometry
Table 1 collects the cRPA H-s interaction, the DMFT hydrogen occupation, and the lattice constant across the series. The screened H-s Hubbard rises substantially with hydrogen content, from 6.25 eV in to 8.89 eV in —an increase of roughly 42%. Over the same range the lattice constant is nearly invariant, varying by less than 0.4% (5.027 to 5.014 Å) and non-monotonically. The enhancement of the hydrogen interaction is therefore not a geometric (lattice-volume) effect: at essentially fixed cell size, the screened interaction nonetheless grows by several eV.
| Orbital | (eV) | |
| LuH | LaH | |
| H-s | ||
| H-s |
Table 1. On-site ( diagonal) Wannier energies of the tetrahedral and octahedral hydrogen states relative to the Fermi level, from the MLWF Hamiltonian, for LuH ( eV) and LaH ( eV).
| System | (Å) | (eV) | (eV) | |
| 5.027 | 6.25 | 3.74 | 1.91 | |
| 5.011 | 6.52 | 4.30 | 1.91 | |
| 5.008 | 7.76 | 4.65 | 1.88 | |
| 5.014 | 8.89 | 4.45 | 1.75 |
Table 1. Evolution across the series: lattice constant , averaged cRPA H-s and RE- Hubbard , and DMFT per-atom H-s occupation. Octahedral occupation begins above . The lattice constant is nearly invariant while rises by .
The rise sets in specifically as octahedral sites become occupied. In and , where only tetrahedral sites are filled, the H-s interaction is comparatively small (6.25 and 6.52 eV) and the matrix is uniform across sites. Once octahedral hydrogen appears (, ), the average jumps to 7.76 and 8.89 eV, and the H-s interactions split by site type (Sec. 2.2.3). The RE- interaction, by contrast, varies only weakly and non-monotonically across the series (–4.65 eV) and shows no comparable systematic trend, indicating that the strong stoichiometry dependence is specific to the hydrogen sublattice.
The screened H-s Hubbard rises monotonically and substantially with hydrogen content, from 6.25 eV in to 8.89 eV in (Table 1), an increase of . The lattice constant, by contrast, varies by less than 0.4% across these compositions and does so non-monotonically, with a minimum near and the trihydride larger than (Table 3). The interaction and the lattice constant thus differ in both magnitude and trend: grows steadily while wanders within a half-percent window. The enhancement of the hydrogen interaction is therefore decoupled from the cell geometry and is electronic in origin.
| Composition | (Å) | (°) | Volume (ų) |
| 5.0272 | 90.000 | 127.055 | |
| 5.0279 | 89.987 | 127.104 | |
| 5.0273 | 90.015 | 127.059 | |
| 5.0288 | 90.018 | 127.179 | |
| 5.0244 | 90.000 | 126.838 | |
| 5.0112 | 90.000 | 125.844 | |
| 5.0165 | 90.000 | 126.238 | |
| 5.0077 | 90.000 | 125.576 | |
| 5.0201 | 90.000 | 126.511 |
Table 3. Lattice parameters of the relaxed structures.
Site-resolved hydrogen interactions: tetrahedral versus octahedral
In the trihydride, where both interstitial sublattices are occupied, the screened H-s interaction resolves into two site-distinct values. Identifying the sites from the Wannier-function centers, the four octahedral H-s orbitals carry eV and the eight tetrahedral eV in LuH, a difference of eV (). The same ordering holds in LaH (, eV) and LuH ( eV), and is absent in LuH and LuH, where only tetrahedral sites are occupied and the H-s interactions are uniform.
This difference is not a consequence of orbital extent. The tetrahedral and octahedral H-s Wannier functions have nearly equal spreads, Å and Å, differing by only (Table 2). For a localized -like Wannier function the bare onsite Coulomb integral scales inversely with the orbital radius, , so the spread difference predicts a bare-interaction ratio of only
| Site | H-s only (Å) | H-s only (eV) | combined (Å) | combined (eV) |
| Octahedral | 0.537 | 8.25 | 0.542 | 12.5 |
| Tetrahedral | 0.513 | 9.21 | 0.535 | 12.20 |
Table 2. Site-resolved H-s Wannier spreads and screened onsite in LuH, in the H-s-only and combined RE-+H-s subspaces. The spreads are nearly equal between sublattices, yet differs by eV; the tet/oct ordering inverts on enlarging the subspace.
i.e. a geometric effect. The observed ratio of the screened interactions is
a effect. The geometry therefore accounts for only about one-sixth of the observed difference; the remainder must arise from screening. The two hydrogen sublattices carry essentially the same bare interaction but are screened to differing degrees, the octahedral hydrogen being the more strongly screened.
The screening origin is confirmed directly by enlarging the correlated subspace from H-s only to combined RE-+H-s, which removes the RE- states from the cRPA screening. Both H-s interactions rise by – eV, but the octahedral interaction rises more steeply ( eV) than the tetrahedral ( eV), inverting the ordering so that the octahedral interaction becomes the larger (Table 2). Throughout this change the Wannier spreads remain nearly constant (, Å), so the several-eV shifts in cannot be geometric. Because the octahedral interaction gains more when the RE- screening channels are removed, the octahedral hydrogen is the sublattice more strongly screened by the rare-earth conduction electrons.
Hydrogen filling
To probe the origin of the hydrogen filling directly, we examine the on-site (R = 0, diagonal) elements of the Wannier Hamiltonian, , which give the bare one-body energy of each correlated orbital before hybridization. Referencing these to the Fermi level isolates where the hydrogen states sit relative to the conduction manifold, providing a more direct measure of the hydrogen charge state than the occupation alone.
The maximally-localized Wannier functions for are well converged, with small, positive spreads (0.69 Å for octahedral and 0.86 Å for tetrahedral H-s, and Å for Lu-), and centers located precisely on the crystallographic sites. The H-s Wannier functions resolve into two symmetry-distinct groups: eight tetrahedral and four octahedral, as expected for the fluorite trihydride.
Correlation effect calculations
Self-energy
We plot the imaginary part of the Matsubara self-energy, fig:S1b and the analytically continued real-axis self-energy fig:S1a. For every orbital and every stoichiometry, extrapolates toward zero as and rises approximately linearly at low frequency. Also, shows no upturn or divergence. This is a Fermi-liquid behavior of coherent, finite-lifetime quasiparticles. This shows that are correlated metals rather than Mott insulators at the temperature considered here.
The low-frequency slope, which sets the mass enhancement , orders the Lu- correlation strength as , i.e. the Lu- correlations decrease with increasing hydrogen content. The overall magnitude of is moderate (peaking at eV), consistent with moderately—rather than strongly—renormalized quasiparticles. On the real axis, the Lu- incoherent scattering peaks lie just below ; they are broadest, with a weak above- shoulder, for and collapse to a single sharp sub- peak for , mirroring the Matsubara-axis ordering.
The H-s self-energy is roughly an order of magnitude smaller than that of Lu- for all stoichiometries, confirming weak H-s correlation despite the large H-s Hubbard (discussed below). Its frequency dependence is noisier at low , reflecting the small signal, so the H-s extrapolation is the least certain. Notably, the H-s scattering grows with hydrogen content () — the opposite trend to Lu- — and on the real axis its scattering is a narrow peak located essentially at .
DFT and DMFT Densities of States
We compare the orbital-projected density of states (DOS) obtained from DFT (top row) and DMFT (bottom row) for the three stoichiometries in Figure S2. Within DFT, all three compositions are metallic, with substantial Lu- (blue) spectral weight at and around the Fermi level and H-s (orange) bonding states found several eV below . This DFT result is expected from mean-field itinerant-band description, with the Lu- manifold crossing . Now, as we include dynamical correlation through DMFT, the spectral weight is redistributed in all three cases. Occupied Lu- weight is pushed to lower energy and becomes incoherent. Next, the DOS near is diminished, and a broad high-energy feature emerges in the range – eV, reminiscent of an upper Hubbard band. The strength of the near- suppression varies across the series: LuH and LuH show DOS minimum at , LuH shows a strong but metallic (bad-metal-like) suppression. The H-s states are mostly situated near the occupied region and are not show much weight transfer.

Figure S2. Comparision of DFT and DMFT densities of states of Lu-H structures calculated using cRPA calculated values for various stoichiometries. Panels (a,d), (b,e), and (c,f) correspond to LuH, LuH, and LuH, respectively.

Figure S1. Imaginary part of the self-energy as calculated for various stoichiometries. Left side is analytically continued and right side is averaged over last few converged DMFT iterations
In every LuH stoichiometry the H-1s shell is near-full (–, i.e. 88–95% occupied) and shows weaker renormalized (, ), despite carrying the largest cRPA interaction (up to eV). The large H-1s interaction manifests instead as a sizable static shift of the real part of the self-energy, – eV, which grows with across the series and displaces the H-derived spectral weight from eV (DFT) to eV (DMFT).
The substoichiometric compounds exhibit multiple sharp features within a narrow energy window of eV around the Fermi level, in contrast to LuH, which shows a comparatively smooth DOS in this region. The DMFT spectral functions over a wider energy range ( to eV) are shown in Figure 2. The sharp features identified in the DFT DOS can be traced to flat bands located above in the non-interacting description. A closer view of the low-energy electronic structure is provided in Figure 4. LuH exhibits a pronounced quasiparticle peak near eV. In addition, the DMFT spectra clearly resolve the crystal-field splitting between the and manifolds. DFT DOS shows LuH as semi metallic whereas DMFT shows LuH as H-dominated metal.
To isolate the variable controlling the H- correlation, we repeated the LuH and LaH calculations with the H- interaction raised to eV, well above the cRPA values, holding all else fixed. In LuH, where H- is near-full, this 58% increase in changes the H- quasiparticle weight (). H- correlation is controlled by the orbital filling, not by the magnitude of : on-site interaction can renormalize a half-filled H- shell but is inactive on a near-full one, regardless of its strength. Even at eV, however, LuH and LaH remains a correlated metal, not an insulator.
Quasiparticle renormalization and local moments
We quantify the local electronic correlation in each manifold using three complementary diagnostics extracted from the DFT+DMFT solution: the quasiparticle mass enhancement , obtained from the low-frequency slope of the Matsubara self-energy ; the per-atom occupation ; and the instantaneous local moment , obtained directly from the CTQMC atomic-state probabilities. The mass enhancement is a low-energy Fermi-liquid quantity and is the most sensitive to statistical noise; the occupation and local moment are static expectation values and are correspondingly robust. We therefore treat the local moment and occupation as the primary correlation diagnostics and report only where the self-energy slope is resolvable above the statistical noise (verified by a Fermi-liquid check on the low-frequency self-energy and a fit-window sensitivity analysis). Tables 15 and 16 collect the results for the H- and RE- manifolds, respectively.
Extended discussion of cRPA screening trends
Screening versus localization in the cRPA literature. That screening, rather than orbital extent, controls interaction trends is established for the transition-metal oxides. Across the SrMO series the interaction decreases with atomic number, contrary to the atomic localization trend, because increasing screening overcompensates the increasing orbital localization [14]. The decisive separation of the two effects—bare interaction (orbital extent) versus screening—was demonstrated for metal surfaces, where the variation of was shown not to arise from changes in the Wannier-function spread (the bare interaction being essentially unchanged) but from changes in screening, such that can even decrease where a simple coordination argument predicts an increase [10]. Our finding that rises even as the H- Wannier functions become more extended is an instance of the same screening-over-localization principle, and we use the same logic to attribute the H- interaction differences to screening.
Dihydride Lu/La comparison. The LuH/LaH comparison, by contrast, is governed primarily by structure. The much smaller LuH lattice constant yields more localized Wannier functions for both RE- ( vs Å) and H- ( vs Å), accounting for the larger interactions in LuH. This geometric ordering is specific to the dihydride: in the trihydride the RE- interaction is larger for LaH despite its larger cell, indicating that octahedral occupation introduces screening and filling changes that override the geometric trend in the channel. The rare-earth species also differ in -shell occupation (filled for Lu, empty for La), which may further modify the local potential and screening; our calculations do not isolate this from the lattice-constant effect, and we do not assign it a definite role.
Relation to the experimental lattice contraction. The interpretation of the stoichiometry trend aligns with the long-standing anionic picture of hydrogen in the rare-earth hydrides [12, 3] and with the experimental observation that heavy rare-earth -RH contracts with increasing hydrogen content, attributed to increasing ionic character of the octahedral hydrogen [13]. Our fixed-composition calculations are not designed to reproduce this small structural contraction directly—the lattice constants across our series vary by less than half a percent and non-monotonically—and we therefore treat the structural trend as experimental context. What our calculations establish is the electronic side of the same picture: at fixed geometry, the screened hydrogen interaction tracks the anionic conversion of the octahedral sublattice. The site-resolved interactions reinforce this reading: the octahedral hydrogen, which carries the additional anionic charge, is the sublattice whose interaction is most sensitive to conduction-electron screening, as shown by the inversion of the tetrahedral/octahedral ordering when the RE- screening channels are removed.
Filling control of the manifold; scope of the analogy. The hydrogen reservoir sets the -band filling: adding octahedral hydrogen withdraws charge from the band, reducing from in the dihydrides to in LuH, and the correlation tracks this dilution, weakening from to 1.04, with the lowest filling (LuH, ) showing no resolvable enhancement. We stress that the filling-control principle at work here is not an orbital-selective Mott transition—there is only a single hydrogen orbital per site, with no intra-shell bandwidth disparity to drive orbital-selective localization—but a demonstration that the underlying principle extends to, and is most transparently seen in, a single-orbital host.
Construction of energy-distribution curves
Calculated energy-distribution curves are vertical cuts of the converged spectral function at fixed momentum, evaluated on the same Wannier Hamiltonian and self-energy used for Figure S3. Because the self-energy is momentum independent, these require no additional DMFT iterations: the Green’s function is re-evaluated on the desired list at fixed and chemical potential.
Three operations convert into a measured-like intensity, applied in the order
where is the Fermi function at the measurement temperature K, is a Gaussian of 30 meV full width at half maximum representing the analyzer resolution, and denotes after momentum broadening. For out-of-plane cuts the momentum broadening is a Lorentzian of half width 0.15 , corresponding to the momentum uncertainty associated with a photoelectron escape depth ; it is applied by convolution along the cut, with the spectral function mirror-extended about each endpoint so that the kernel is not truncated at the segment edges. In-plane cuts are shown without momentum broadening. All three parameters are fixed by the experimental conditions and none is adjusted to improve agreement. Peak positions quoted in the main text are obtained by parabolic refinement of the maximum of the raw within an energy window bracketing a single band, before the Fermi cutoff and resolution are applied, since both displace the apparent maximum.
Out-of-plane curves are computed along the pure line at fixed in-plane momentum , i.e. the – segment of the simple-cubic zone. The dispersion of the eV feature is quoted over the full segment rather than at a single point: since , the derivative vanishes at both the zone centre and the zone boundary, so curves selected from the immediate vicinity of either endpoint appear flat by symmetry irrespective of the dispersion in between.
Limitations of the comparison of EDC
The measured and calculated curves differ in three respects that bear on the comparison. Energetically, the measurement was performed at 15–30 K with meV analyzer resolution, whereas the impurity problem is solved at eV; the calculated linewidths are therefore upper bounds, and analytic continuation of limits the precision of feature positions near to a few tens of meV. In momentum, the calculation is evaluated at definite , whereas photoemission integrates over an intrinsic uncertainty set by the escape depth and, in these samples, over an additional in-plane spread arising from their multi-domain character, for which high-symmetry directions could not be established. Neither the in-plane origin nor the of the measured curves is therefore known, and no momentum-resolved agreement is claimed. In intensity, the measured curves include photoemission matrix elements, which are orbital-, geometry-, and photon-energy-dependent, together with an inelastic background that grows with binding energy; neither is contained in , which is why the calculated curves return to zero at high binding energy while the measured ones do not.
A further caveat concerns the ordered arrangement of octahedral hydrogen in our supercell. The energy of the octahedral-derived states is robust, but their residual dispersion and fine structure are artifacts of that ordering: in a disordered sample the same states would appear as an incoherent resonance at similar energy rather than as coherent minibands. Any residual excess sharpness of the calculated features relative to experiment is expected on these grounds alone.
cRPA Interactions
The screened on-site interaction on H-s substantially exceeds that on RE- across all compositions, and this hierarchy is a screening effect rather than a localization effect. The decisive evidence is the subspace test: enlarging the correlated subspace from H-s-only to the combined RE-+H-s manifold increases both the H-s Wannier spread and the screened —a combination the bare-interaction (geometric) picture cannot produce, and which can only arise from the removal of RE- screening channels. This is the rare-earth-hydride analogue of the well-known result for SrVO, where including the oxygen states in the correlated subspace nearly doubles the static by excluding the O- screening channel [1]; here the RE- manifold plays the role of the oxygen states. Likewise, cRPA results also show that screening can override orbital localization in setting interaction trends [14, 10]. We have a further discussion in Sec. SX.
The screening is noticibly asymmetric. The itinerant RE- manifold, straddling , screens the H-s interaction strongly, whereas the filled, deep-lying H-s manifold screens RE- only weakly: removing H-s from the screening channels leaves essentially unchanged. In the combined manifold the RE- interaction becomes nearly uniform even though the Wannier spreads retain their crystal-field 2 + 3 structure—a further sign that the interaction is set by screening rather than orbital extent.
The screened hydrogen interaction also rises strongly with stoichiometry at nearly fixed lattice constant, identifying the trend as electronic rather than geometric. The mechanism is the progressive anionic conversion of the octahedral hydrogen above : each octahedral hydrogen that converts toward H withdraws an electron from the conduction band, depleting the carriers that screen the local interactions—the same carrier-density reduction observed optically as a decreasing Drude plasma frequency [15]—so that rises as octahedral sites fill. The effect is specific to hydrogen; the RE- interaction shows no comparable trend. The dihydride Lu/La comparison, the role of the -shell difference, and the relation to the experimental lattice contraction are discussed in Sec. SX.
Details on Absorptivity calculation
We finally remark on the systematic plasmon-position offset between calculation and experiment ( eV). Several conventional sources may contribute: the equilibrium DFT lattice constant of LuH may differ from the experimental composition LuH by a few percent with corresponding effect on the Drude carrier density; convergence of the intraband contribution within epsilon.x requires very dense k-mesh sampling; and the intraband broadening parameter is treated phenomenologically rather than from first principles. We do not attribute the offset to mass renormalization, since our DMFT analysis gives for the LuH end member (Sec. 4.1.1), too close to unity to account for a factor of in the squared plasma frequency. The relative ordering and qualitative shape of the configurational dependence are robust regardless of the precise plasmon position, and these are what carry the interpretive weight of the present comparison.
Flat Bands, van Hove Singularity, and Comparison with Photoemission
We compare the band structures of the following cubic fluorite based structures: LuH (one H vacancy), LuH, LuH (one H occupancy), oct-LuH (one H vacancy and one H occupancy) to identify the ingredients for flat bands and van Hove singularity near . Tetrahedral vacancies provide flat bands below the Fermi level whereas octahedral occupancy pulls the van Hove singularity above M point down to the Fermi level. The structure with both H vacancy and H occupancy has van Hove singularity situated right at the Fermi level and flat bands below the Fermi level Figure 4. The octahedral-H manifold, which the on-site Wannier alignment places at eV in LuH, hybridizes with the Lu-5d states crossing in the dihydride and drives a van Hove-like feature at toward the Fermi level, while tetrahedral vacancies contribute a localized, weakly dispersing defect band. Where both coexist—the minimal model for the observed LuH optical feature— is enhanced relative to the clean dihydride. At the Wannier level these ingredients are present but incomplete: the narrow band is not yet pinned at and the saddle is not yet degenerate.
Interactions complete the construction. Since the Wannier and DMFT panels of Figure 5 share a Hamiltonian, and since the mass enhancements extracted from the Matsubara self-energies are bounded by 1.33, the near- flattening is too large to be quasiparticle mass enhancement. It originates instead in the static, orbital-dependent shift , which relocates the hydrogen level so that its hybridization with the Lu- manifold flattens the composite band and brings the saddle into coincidence with . This is an interaction effect—it is absent at —but it is a band reconstruction rather than a correlation-induced narrowing, and its magnitude is sensitive to the double-counting prescription in the hydrogen channel, the dominant systematic uncertainty here.
Carrier doping then tunes the result. The features shift rigidly in energy at 30 meV per percent doping without displacing in momentum, electron doping moving them below and hole doping toward and through it. This quantifies, within a correlated framework, the proposal of [6] that aligning the flat band and saddle with would enhance a superconducting instability. The compositional series realizes the same tuning self-consistently, since each additional octahedral hydrogen removes – conduction electrons per formula unit while introducing new octahedral levels.
The photoemission comparison supports a two-component reading of both data sets. Experimentally, a sharp flat band lies immediately below with a light band emerging from it; in the calculation, the weakly dispersing hybridized band near is pro tected against lifetime broadening by the smallness of at low frequency, and the deeper structure near eV is dispersive and resolves into individual bands. The energy-distribution curves of Figure S3 reproduce this correspondence: both show a sharp, momentum-independent maximum just below , and deeper structure that disperses in plane and is comparatively flat out of plane. That the calculated deep feature appears as separate peaks rather than the single measured hump follows from momentum integration along over the escape-depth window and from the ordered octahedral sublattice of our supercell, which produces coherent minibands where a disordered sample would show an incoherent resonance at the same energy. Our calculations contain neither nitrogen nor oxygen, both present in the measured samples of Ref. [6]; occupation of the octahedral site by hydrogen alone therefore suffices to generate the observed phenomenology, and the all-tetrahedral calculation, in which neither feature appears, identifies octahedral occupation rather than stoichiometry as the controlling variable. Consistent with this, similar photoemission results have been obtained on nitrogen-free Lu–H crystals [7], indicating that nitrogen is not required for the flat band and saddle to appear.
The comparison is restricted to energies and qualitative band shapes. The measured crystals are multi-domain with no established high-symmetry directions, the flat-band and saddle data were taken at different and undetermined , and our momenta refer to a supercell zone; no momentum-resolved agreement is claimed. On the theory side the impurity problem is solved at eV, so linewidths are upper bounds, and analytic continuation limits feature positions near to a few tens of meV. Within these limits the picture is consistent and, notably, weakly correlated: mass enhancements nowhere exceed [1.33, local moments are fully screened, and the low-frequency self-energies are Fermi-liquid-like. The flat band and saddle are consequences of octahedral-site chemistry sharpened by an interaction-induced level shift—not manifestations of strong correlation, as has been suggested. What this leaves open is whether such features, once tuned to , matter for pairing.
Tables
Below are the tables.
Table 3: Lattice parameters of the relaxed structures. All cells are (near-)cubic with and ; deviations of from are . The lattice constant varies by less than across the full composition range.
| Composition | (eV) | a (Å) | b (Å) | c (Å) | (°) | Volume (Å) |
| LuH | 0.000, 0.000 | 5.0272 | 5.0272 | 5.0272 | 90.000, 90.000, 90.000 | 127.055 |
| LuH | 2.000, 6.000 | 5.1032 | 5.1032 | 5.1032 | 90.000, 90.000, 90.000 | 132.897 |
| LuH | 0.000, 0.000 | 5.0112 | 5.0112 | 5.0112 | 90.000, 90.000, 90.000 | 125.841 |
| LuH | 4.300, 6.522 | 5.1771 | 5.1771 | 5.1771 | 90.000, 90.000, 90.000 | 138.759 |
| LuH [5] | – | 5.033 | 5.033 | 5.033 | 90.000, 90.000, 90.000 | 127.491 |
Table 4. Lattice parameters and volume comparision between lattice parameters optimized using DFT or DFT+. The last row is experimentally measured lattice parameter from [5]
Table 6. Screened Coulomb interaction matrix for the H-s subspace.
| H site | Spread (Å) | (eV) |
| 1 | 0.6278 | 6.0695 |
| 2 | 0.6259 | 6.2429 |
| 3 | 0.6293 | 6.3170 |
| 4 | 0.6259 | 6.2429 |
| 5 | 0.6293 | 6.3170 |
| 6 | 0.6259 | 6.2429 |
| 7 | 0.6293 | 6.3170 |
Table 7. Wannier spreads and onsite Hubbard for the H-s subspace.
Table 8. Screened Coulomb interaction matrix for the Lu-d subspace.
| Orbital | Spread (Å) | (eV) |
| 1 | 1.3801 | 3.8959 |
| 2 | 1.4409 | 3.6419 |
| 3 | 1.4408 | 3.6417 |
| 4 | 1.3800 | 3.8964 |
| 5 | 1.4410 | 3.6416 |
Table 9. Wannier spreads and onsite Hubbard for the Lu- subspace. Lu- sites show the 2 + 3 orbital grouping in spreads and onsite .

Figure S5. Calculated dielectric function of LuH without (left) and with (right) octahedral hydrogen. (a) Re ; (b) Im . In the tetrahedral-only structure, falls below unity between 1.0 and 2.2 eV and crosses zero within this window at 1.05 eV, giving a sharp, weakly damped plasma edge. Octahedral occupancy introduces an interband oscillator peaking at 1.05 eV inside this window, moving the zero crossing of down to 0.57 eV and raising at the crossing from to , which damps the edge and splits the single absorptivity maximum of Figure 6 into the two observed features.
| H site | Spread (Ų) | (eV) |
| 21 | 0.6953 | 6.9423 |
| 22 | 0.7199 | 7.2335 |
| 23 | 0.7319 | 7.6308 |
| 24 | 0.7199 | 7.2335 |
| 25 | 0.7319 | 7.6308 |
| 26 | 0.7198 | 7.2335 |
| 27 | 0.7319 | 7.6309 |
Table 10. Wannier spreads and onsite Hubbard for H-s orbitals in the combined Lu-d + H-s correlated subspace.
| Lu atom | Orbital index | Spread (Ų) | Onsite (eV) |
| 1 | 1 | 1.42437075 | 3.6296 |
| 1 | 2 | 1.54659042 | 3.6299 |
| 1 | 3 | 1.54662202 | 3.6299 |
| 1 | 4 | 1.42443276 | 3.6301 |
| 1 | 5 | 1.54662241 | 3.6292 |
| 2 | 6 | 1.42437067 | 3.6296 |
| 2 | 7 | 1.54655558 | 3.6298 |
| 2 | 8 | 1.54660548 | 3.6298 |
| 2 | 9 | 1.42443295 | 3.6301 |
| 2 | 10 | 1.54659633 | 3.6291 |
| 3 | 11 | 1.42437190 | 3.6296 |
| 3 | 12 | 1.54655256 | 3.6297 |
| 3 | 13 | 1.54653023 | 3.6297 |
| 3 | 14 | 1.42443215 | 3.6301 |
| 3 | 15 | 1.54657519 | 3.6290 |
| 4 | 16 | 1.42437323 | 3.6296 |
| 4 | 17 | 1.54652925 | 3.6297 |
| 4 | 18 | 1.54652519 | 3.6297 |
| 4 | 19 | 1.42443113 | 3.6301 |
| 4 | 20 | 1.54657976 | 3.6290 |
Table 11. Wannier spreads and onsite Hubbard for Lu-d orbitals in the combined Lu-d + H-s correlated subspace.
| RE-d subspace | H-s subspace | RE-d + H-s subspace | |||||||
| System | (eV) | (eV) | (eV) | (eV) | (eV) | (eV) | (eV) | (eV) | (eV) |
| LuH₂ | 4.3008 | 0.8210 | 0.0952 | 6.5219 | 0.3624 | 0.0014 | 5.1666 | 0.5072 | 0.0564 |
| LaH₂ | 3.3265 | 0.8140 | 0.0738 | 5.8031 | 0.2466 | 0.0009 | 4.8154 | 0.7786 | 0.0451 |
| LuH₃ | 4.45 | – | – | 8.89 | – | – | 9.1 | – | – |
| LaH₃ | 4.69 | – | – | 8.40 | – | – | 7.45 | – | – |
Table 12. Screened interaction parameters obtained from cRPA for LuHₓ and LaHₓ for . The table compares intra-orbital Hubbard , inter-orbital interaction , and Hund’s coupling in the RE-d, H-s, and combined RE-d + H-s correlated subspaces.
| Subspace | |||
| H-s only | 9.21 | 8.25 | +0.96 |
| Lu- + H-s | 12.20 | 12.5 | −0.3 |
Table 14. Site-resolved H-s screened (eV) in LuH, showing the inversion of the tetrahedral/octahedral ordering upon removal of RE- screening (H-s-only → combined subspace).
Table 15: H- correlation diagnostics. Occupation , atomic-state probabilities, instantaneous local moment , and quasiparticle mass enhancement ; local spin and charge susceptibilities and from the CTQMC solver at . Tetrahedral and octahedral H sites are treated as inequivalent impurities.
| System | Site | |||||||
| LuH | tet | 1.833 | 0.838 | 0.157 | 1.199 | 0.0052 | 0.0047 | 1.11 |
| LuH | tet | 1.807 | 0.814 | 0.179 | 1.221 | 0.0058 | 0.0052 | 1.12 |
| LuH (7tet+1oct) | tet | 1.848 | 0.852 | 0.145 | 1.278 | 0.0046 | 0.0040 | 1.13 |
| oct | 1.720 | 0.732 | 0.257 | 1.318 | 0.0129 | 0.0098 | 1.31 | |
| LuH | tet | 1.789 | – | – | – | 0.0065 | 0.0058 | 1.13 |
| oct | 1.645 | – | – | – | 0.0169 | 0.0132 | 1.28 | |
| LuH | tet | 1.799 | 0.806 | 0.187 | 1.292 | 0.0063 | 0.0054 | 1.16 |
| oct | 1.731 | 0.742 | 0.246 | 1.255 | 0.0121 | 0.0095 | 1.26 | |
| LuH | tet | 1.812 | 0.818 | 0.176 | 1.335 | 0.0057 | 0.0049 | 1.16 |
| oct | 1.758 | 0.767 | 0.224 | 1.361 | 0.0104 | 0.0083 | 1.25 | |
| LaH | tet | 1.830 | 0.835 | 0.161 | 1.301 | 0.0061 | 0.0053 | 1.15 |
| oct | 1.887 | 0.889 | 0.110 | 1.197 | 0.0055 | 0.0049 | 1.12 |
Table 15.
Table 16: RE- correlation diagnostics. Total -shell occupation and per-orbital occupations , from the converged density matrix; instantaneous local moment (group total) and mass enhancement resolved by crystal-field set; local spin and charge susceptibilities , at . In the cubic field of the fluorite structure the set lies below and is correspondingly the more occupied of the two at every composition.
| System | ||||||||||
| LuH | 1.588 | 0.188 | 0.462 | 0.642 | 0.946 | 1.211 | 1.247 | 0.199 | 0.051 | 3.90 |
| LuH | 1.428 | 0.123 | 0.489 | 0.427 | 1.001 | 1.278 | 1.295 | 0.217 | 0.047 | 4.63 |
| LuH (7tet+1oct) | 1.417 | 0.204 | 0.360 | 0.665 | 0.753 | 1.283 | 1.320 | 0.211 | 0.039 | 5.42 |
| LuH | 1.296 | 0.137 | 0.405 | – | – | – | – | 0.161 | 0.053 | 3.04 |
| LuH | 0.830 | 0.122 | 0.186 | 0.487 | 0.343 | 1.204 | 1.199 | 0.060 | 0.031 | 1.93 |
| LuH | 0.679 | 0.111 | 0.122 | 0.400 | 0.279 | 1.178 | 1.161 | 0.049 | 0.030 | 1.67 |
| LaH | 0.531 | 0.070 | 0.105 | 0.355 | 0.175 | 1.213 | 1.203 | 0.075 | 0.041 | 1.84 |
Table 16.
| System | ||||||
| LuH | 1.750 | 3.208 | 0.292 | 1.588 | 0.338 | 0.046 |
| LuH | 2.000 | 3.614 | 0.386 | 1.428 | 0.428 | 0.042 |
| LuH (7tet+1oct) | 2.000 | 3.664 | 0.336 | 1.417 | 0.417 | 0.081 |
| LuH | 2.875 | 5.113 | 0.637 | 0.830 | 0.705 | 0.068 |
| LuH | 3.000 | 5.382 | 0.618 | 0.679 | 0.679 | 0.061 |
| LaH | 3.000 | 5.547 | 0.453 | 0.531 | 0.531 | 0.078 |
Table 17. Charge sharing between the hydrogen and - sublattices. For the closed-shell ionic reference has every hydrogen as , giving and per formula unit. The hydrogen deficit and the excess are obtained from Tables 15 and 16. The two track one another across the series with a correlation coefficient of 0.99; the residual is constant to within its spread and equals the deviation of the total correlated-subspace occupation from the nominal valence count , i.e. the weight lying outside the Wannier subspace. Charge missing from the hydrogen shell is thus recovered on the shell, establishing that the two sublattices are hybridized rather than ionically decoupled.
References
- [1]F. Aryasetiawan, T. Miyake, and R. Sakuma. The constrained RPA method for calculating the Hubbard U from first-principles. In Eva Pavarini, Erik Koch, Dieter Vollhardt, and Alexander Lichtenstein, editors, The LDA+DMFT approach to strongly correlated materials, volume 1 of Modeling and Simulation, chapter 7. Forschungszentrum Jülich, Jülich, 2011. ISBN 978-3-89336-734-4. URL https://www.cond-mat.de/events/correll11/manuscript/Aryasetiawan.pdf. Lecture Notes of the Autumn School 2011.
- [2]P. E. Blöchl. Projector augmented-wave method. Physical Review B, 50 (24):17953–17979, December 1994. doi: 10.1103/PhysRevB.50.17953. URL https://link.aps.org/doi/10.1103/PhysRevB.50.17953.DOI
- [3]M. Gupta and J. P. Burger. Electronic structure of the rare-earth hydrides LaH₂ and LaH₃. Phys. Rev. B, 22(12):6074–6084, 1980. doi: 10.1103/PhysRevB.22.6074.
- [4]G. Kresse and J. Furthmüller. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Physical Review B, 54 (16):11169–11186, October 1996. doi: 10.1103/PhysRevB.54.11169. URL https://link.aps.org/doi/10.1103/PhysRevB.54.11169.DOI
- [5]Xin Li, Ying Wang, Yuhao Fu, Simon A. T. Redfern, Shuqing Jiang, Pinwen Zhu, and Tian Cui. Stabilization of High-Pressure Phase of Face-Centered Cubic Lutetium Trihydride at Ambient Conditions. Advanced Science, 11 (29):2401642, 2024. ISSN 2198-3844. doi: 10.1002/advs.202401642. URL https://onlinelibrary.wiley.com/doi/abs/10.1002/advs.202401642. eprint: https://advanced.onlinelibrary.wiley.com/doi/pdf/10.1002/advs.202401642.DOI
- [6]Xin Liang, Zihan Lin, Jun Zhang, Jianfa Zhao, Shiyu Feng, Wenlong Lu, Guodong Wang, Luchuan Shi, Ningning Wang, Pengfei Shan, Zhao Zhang, Muntaser Naameh, Runzhe Liu, Bastien Michon, Jinguang Cheng, Changqing Jin, Yang Ren, and Junzhang Ma. Observation of flat band and van hove singularity in non-superconducting nitrogen-doped lutetium hydride, 2023. URL https://arxiv.org/abs/2308.16420.
- [7]J. Ma. private communication.
- [8]Arash A. Mostofi, Jonathan R. Yates, Young-Su Lee, Ivo Souza, David Vanderbilt, and Nicola Marzari. wannier90: A tool for obtaining maximally-localised Wannier functions. Computer Physics Communications, 178(9):685–699, May 2008. ISSN 0010-4655. doi: 10.1016/j.cpc.2007.11.016. URL https://www.sciencedirect.com/science/article/pii/S0010465507004936.DOI
- [9]John P. Perdew, Kieron Burke, and Matthias Ernzerhof. Generalized Gradient Approximation Made Simple. Physical Review Letters, 77(18):3865–3868, October 1996. doi: 10.1103/PhysRevLett.77.3865. URL https://link.aps.org/doi/10.1103/PhysRevLett.77.3865.DOI
- [10]E. Şaşıoğlu, C. Friedrich, and S. Blügel. Strength of the effective Coulomb interaction at metal and insulator surfaces. Phys. Rev. Lett., 109(16):166405, 2012. doi: 10.1103/PhysRevLett.109.166405.
- [11]Vijay Singh, Uthpala Herath, Benny Wah, Xingyu Liao, Aldo H. Romero, and Hyowon Park. DMFTwDFT: An open-source code combining Dynamical Mean Field Theory with various density functional theory packages. Computer Physics Communications, 261:107778, April 2021. ISSN 0010-4655. doi: 10.1016/j.cpc.2020.107778. URL https://www.sciencedirect.com/science/article/pii/S001046552030388X.DOI
- [12]A. C. Switendick. The change in electronic properties on hydrogen alloying and hydride formation. In G. Alefeld and J. Völkl, editors, Hydrogen in Metals I: Basic Properties, volume 28 of Topics in Applied Physics, pages 101–129. Springer, Berlin, 1978.DOI
- [13]P. Vajda. Hydrogen in rare-earth metals, including RH₂+x phases. In K. A. Gschneidner and L. Eyring, editors, Handbook on the Physics and Chemistry of Rare Earths, volume 20, chapter 137, pages 207–291. Elsevier, Amsterdam, 1995. doi: 10.1016/S0168-1273(05)80071-6.
- [14]Loïc Vaugier, Hong Jiang, and Silke Biermann. Hubbard U and Hund’s exchange J in transition metal oxides: Screening versus localization trends from constrained random phase approximation. Phys. Rev. B, 86(16):165105, 2012. doi: 10.1103/PhysRevB.86.165105.DOI
- [15]J. H. Weaver, R. Rosei, and D. T. Peterson. Electronic structure of metal hydrides. I. Optical studies of Sc H₂, Y H₂, and Lu H₂. Phys. Rev. B, 19(10):4855–4866, May 1979. ISSN 0163-1829. doi: 10.1103/PhysRevB.19.4855. URL https://link.aps.org/doi/10.1103/PhysRevB.19.4855.DOI