Plasmon-driven exciton formation in a non-equilibrium Fermi liquid Preprint: AAPM/123-QED Rishi Acharya, Eli Gerber, Nina Bielinski, Hannah E. Aguirre, Younsik Kim, Camille Bernal-Choban, Gaurav Tenkila, Suhas Sheikh, Pranav Mahaadev, Faren Hoveyda-Marashi, Subhajit Roychowdhury, Chandra Shekhar, Claudia Felser, Peter Abbamonte, Benjamin J. Wieder, Fahad Mahmood Abstract Figure 1: Excitations of a Fermi liquid at and away from equilibrium: (a) Schematic depiction of the excitations of a Fermi liquid at equilibrium, represented as scattering processes along the occupied and unoccupied band dispersions ( E vs k ) (left) and as a spectrum showing the excitation energy \omega as a function of the excitation wave vector q (right). In (a), E_{F} and k_{F} indicate the Fermi level and Fermi momentum, respectively. E_{CBM} indicates the energy minimum of the lower conduction band, measured relative to E_{F} , and \omega_{p} denotes the energy of the plasmon. (b) Schematic depiction of the excitations of a Fermi liquid with a secondary chemical potential ( \mu_{NE} ) photo-doped into a conduction band. In (b), \mu_{NE} is measured relative to the bottom of the subset of photo-doped bands. Figure 2: Unoccupied band structure of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} : (a) (001) -surface spectral function of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} , obtained from ab initio density-functional theory (DFT) calculations in the paramagnetic (PM) state (Supplementary Information III.1 ). The spectrum in (a), plotted in the blue shaded (001) -surface Brillouin zone (BZ) in the right schematic, consists of a nearly doubly degenerate pair of surface bands that coexist with the surface projections of the bulk conduction manifold; the surface band minimum lies at an energy that we denote as \widetilde{E} . (b) Schematic of the time- and angle-resolved photoemission spectroscopy (Tr-ARPES) experiment. The UV probe and IR pump are incident along the (x,z) - and (y,z) -planes of the lab frame, both at [formula] to the sample normal (the (001) -surface) and are p -polarized. In (b), the sample is aligned with \overline{\Gamma}-\overline{\textrm{M}} along k_{x} . This measurement geometry probes both the (001) -surface and the bulk electronic structure at k_{z}\approx 0 . (c) Time-integrated ( t-t_{0}\in[$0\text{\,}\mathrm{p}\mathrm{s}$,$2\text{\,}\mathrm{p}\mathrm{s}$] ) ARPES spectrum ( E vs k_{y} ) showing the occupied and unoccupied band structure of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} , with light gray lines showing the bulk band structure at k_{z}=0 along \Gamma-\mathrm{K} from the PM-state DFT calculation, and with dark gray horizontal rules indicating where different scans were truncated to render the composite figure. Above the Fermi level E_{F} in (c), we observe a set of surface states whose dispersion closely matches the DFT-obtained surface states in (a) and whose minimum approximately lies at E_{0}\approx E_{F}+$0.8\text{\,}\mathrm{e}\mathrm{V}$ (Supplementary Information III.2 ). Just below E_{0} lies a bulk conduction band that is nondispersing at \Gamma to leading order (yellow triangle in (c)) in which exciton holes form. (d) A constant-energy cut of circular dichroism (CD-) ARPES measurements of the (001) -surface states taken along the dashed line in (c) ( E_{0}+0.23 eV, \pm 30\text{\,}\mathrm{m}\mathrm{e}\mathrm{V} integration window) at t_{0} , with a hue given by the difference between the ARPES intensities for opposite circular probe photon polarizations and a saturation set by the polarization-integrated photoemission intensity. (e) A constant-energy cut of the local atomic orbital angular momentum (OAM) of the (001) -surface states in the PM-phase DFT calculation in (a), taken at \widetilde{E}+0.23 eV (the dashed line in (a)) and projected along the incident axis of the probe ( \langle L_{xz}\rangle , angled [formula] between the bulk line \Gamma-\textrm{A} and the surface and bulk line \overline{\Gamma}-\overline{\textrm{M}} in the BZ in (a), see Supplementary Information III.3 ). The Tr-ARPES measurements in this figure were taken at 20\text{\,}\mathrm{K} , below T_{C}\approx$25\text{\,}\mathrm{K}$ for the ferromagnetic phase, using 1.2\text{\,}\mathrm{e}\mathrm{V} excitation of low fluence ( 15\text{\,}\mathrm{\SIUnitSymbolMicro} ). Importantly, to facilitate comparison with the DFT calculations, we also performed the same measurements in the PM phase at 42\text{\,}\mathrm{K} , and observed qualitatively identical (CD-) ARPES spectra (Supplementary Information II ). Figure 3: Population dynamics of the unoccupied surface states at representative excitation densities: (a) ARPES spectra of the unoccupied band structure of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} at selected temporal delays after low-fluence (top row) and high-fluence (bottom row) excitation. (b) Energy distribution curves at various temporal delays after photo-excitation for low ( 15\text{\,}\mathrm{\SIUnitSymbolMicro} , left) and high ( 55\text{\,}\mathrm{\SIUnitSymbolMicro} , right) fluence photo-excitation, where the shaded region at E_{X} in the right panel is a guide to the eye. For the experimental data in this figure, the Tr-ARPES measurements were taken at 9.5\text{\,}\mathrm{K} using 1.5\text{\,}\mathrm{e}\mathrm{V} excitation. Collective modes in Fermi liquids are usually regarded as dissipation channels that relax electronic excitations through Landau damping. Whether such modes can instead mediate the formation of correlated electronic states under non-equilibrium conditions remains an open question. Here we show that, under optical photo-doping, a bulk plasmon can drive correlated inter-band transfer within a transient electronic continuum. Using time- and angle-resolved photoemission spectroscopy (Tr-ARPES) on \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} supported by electronic structure calculations, we observe that at high excitation density, plasmons transfer energy from a weakly dispersing bulk band into unoccupied surface states. This bulk-to-surface redistribution stabilizes a long-lived, energy-localized spectral feature consistent with a Mahan exciton. Our results uncover a non-equilibrium regime of Fermi-liquid physics in which collective modes do not merely dissipate energy, but also stabilize correlated bound states. Collective excitations in interacting Fermi systems are traditionally regarded as decay channels. In a conventional Fermi liquid, the plasmon branch lies outside the particle–hole continuum at long wavelengths, but enters it at finite momentum, where Landau damping transfers plasmon energy into electron–hole pairs (Fig. 1 a) 61 ; 99 . This paradigm underlies the standard description of relaxation in metals, where collective modes redistribute electronic energy into particle–hole continua. Such relaxation also extends to non-equilibrium settings: in optically excited metals and semiconductors, bosonic-mode-assisted inter- and intra-band scattering channels have been directly observed to transfer high-energy carriers into lower-energy electronic continua and bound states 58 ; 78 ; 49 . In these cases, the collective mode functions as an efficient pathway toward equilibration. Yet this established framework leaves open a fundamental question: can collective modes, rather than merely dissipating energy, actively drive the formation of correlated electronic states under non-equilibrium conditions? In general, inter-band optical excitation generates a non-equilibrium carrier population that rapidly thermalizes to a secondary chemical potential within transiently filled bands 31 . The resulting inter- and intra-band excitation continua can extend to \omega=0 , overlapping with the plasmon branch (Fig. 1 b). In this regime, collective modes need not act solely as sinks of energy; they may instead mediate interband transfer within the transient continuum. Observing such an interaction channel in a bulk metal, however, is challenging. In three-dimensional systems, strong screening and the large phase space tend to smear interaction-enhanced features. Plasmons are heavily damped once they overlap with bulk continua, and competing scattering channels redistribute spectral weight broadly in momentum and energy. A more favorable setting would involve surface electronic bands interacting with a bulk plasmon. The reduced dimensionality of the surface suppresses phase space, weakens screening, and thus enhances interaction effects. Moreover, the near-field interaction between the dipole moment of a bulk collective mode in a low-carrier-density material and surface-localized conduction electrons is exponentially enhanced at zero-momentum transfer 45 . Thus, in this configuration, a bulk plasmon could transfer energy to the more confined surface electron bands, giving rise to correlated features that can be spectrally resolved. We here find that \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} precisely provides the latter setting. As we demonstrate below, the electronic structure of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} hosts a bulk plasmon at an energy scale comparable to the separation between an unoccupied, weakly dispersing 3D bulk band and an unoccupied 2D surface state. Density-functional theory (DFT) calculations, presented below, establish the existence of these unoccupied surface states, motivating the experimental investigation of plasmon-mediated transfer between the bulk and surface in the non-equilibrium continuum. Using time- and angle-resolved photoemission spectroscopy (Tr-ARPES), we not only uncover this plasmon-mediated transfer, but also find that the reduced screening and constrained phase space of the surface stabilizes a long-lived, energy-localized spectral feature consistent with a Mahan exciton 51 ; 52 . Our results reveal a non-equilibrium regime of Fermi-liquid physics in which collective modes do not merely dissipate energy, but actively reshape electronic distributions by mediating correlated bulk-to-surface transfer. \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} has been previously characterized as semiconducting 75 ; 60 ; 80 or semimetallic throughout its magnetic phase diagram (Supplementary Information I and Ref. 48 ; 72 ), with an antiferromagnetic phase reported below T_{N}=$9\text{\,}\mathrm{K}$ . In the samples used in this work, ferromagnetic (FM) order has been stabilized below T_{C}=$25\text{\,}\mathrm{K}$ through Eu vacancies 72 . Above T_{C} , \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} lies in a paramagnetic (PM) state with bulk inversion ( \mathcal{I} ) and time-reversal ( \mathcal{T} ) symmetries. To analyze the bulk and surface electronic structure of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} , we performed ab initio (DFT) and Wannier-based tight-binding calculations of the electronic structure (Supplementary Information III.1 ). For simplicity, and because our ARPES measurements produced qualitatively identical spectra both above and below T_{C} over the energy and momentum range probed in this study (Supplementary Information II ), we performed our DFT calculations in the PM state. In our DFT-based calculations, we observe unoccupied (001) -surface bands that coexist with the projected continuum of unoccupied bulk states (the projected bulk conduction manifold) and exhibit an approximately hyperbolic dispersion with an energy minimum that we denote as \widetilde{E} (Fig. 2 a, see Supplementary Information III.2 ). Through calculations of the (001) -surface spin texture in Supplementary Information III.3 , we find that the (001) -surface states in fact form a 2D Dirac cone with weak spin-orbit-coupling- (SOC-) driven splitting away from {\bf k}={\bf 0} . To understand the origin of the hyperbolic (001) -surface Dirac cone in Fig. 2 a, we applied the group-theoretic methodology in Ref. 94 to compute the band topology at and away from E_{F} (Supplementary Information III.4 ). Through this analysis, we identified a large range of electronic fillings – spanning from 5 electrons below intrinsic filling to 7 electrons above – at which \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in the PM state hosts 3D topological band-insulating gaps. We may hence associate the unoccupied (001) -surface states in Fig. 2 a to a 3D topological band insulator phase at an electronic filling above E_{F} , which can be accessed by photo-doping experiments. In this sense, \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} is reminiscent of the repeat-topological (RTopo) materials introduced in Ref. 94 , which host topological gaps at and just below E_{F} . For example, in the RTopo material \text{Bi}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{Mg}{\vphantom{\text{X}}}_{\smash[t]{\text{3}}} , the topological band gap below E_{F} gives rise to topological surface resonance bands that coexist with the surface projections of the bulk valence manifold and are detectable in ARPES experiments 94 ; 15 ; 110 . The unoccupied (001) -surface states in \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} and the associated topological band gaps above E_{F} can hence be viewed as a generalization of the concept of repeat-topology – originally introduced to predict topological boundary states accessible in equilibrium ARPES experiments at intrinsic filling – to topological surface conduction bands that are accessible by doping and Tr-ARPES experiments like those in the present study. Having established the presence and energies of unoccupied surface states in \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} , we now use Tr-ARPES to dynamically populate and resolve the surface bands. The geometry of the Tr-ARPES experiment is shown in Fig. 2 b. In our measurements, p -polarized IR pulses at either 1.2\text{\,}\mathrm{e}\mathrm{V} or at 1.5\text{\,}\mathrm{e}\mathrm{V} excite inter-band transitions, after which we collect ARPES spectra using 6\text{\,}\mathrm{e}\mathrm{V} photons with a variable delay from photo-excitation. In momentum space, the photon scattering plane coincides with a crystallographic mirror plane whose normal vector lies along k_{y} and is related to the BZ line \overline{\Gamma}-\overline{\mathrm{K}} in Fig. 2 a by crystal symmetry. In Fig. 2 c, we show an ARPES spectrum ( E vs k_{y} ) of the occupied and unoccupied bulk and surface band structure of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} . Below E_{F} at its equilibrium value ( i.e. , at energies E-E_{0}<$-0.8\text{\,}\mathrm{e}\mathrm{V}$ , where E_{0} is assigned to the bottom of a bright electron-like unoccupied band), a broad valence band is visible, consistent with equilibrium synchrotron ARPES measurements 72 and bulk DFT calculations performed in the present study (gray lines, see Supplementary Information III.1 ). At 0.8\text{\,}\mathrm{e}\mathrm{V} above E_{F} in Fig. 2 c ( \emph{i.e.} , above E_{0} ), we observe a sharp, nearly-linear (hyperbolic) electron-like feature, in strong agreement with the calculated surface spectral function in Fig. 2 a. Slightly below E_{0} in Fig. 2 c, we observe a faint incoherent spectral weight, which we attribute to the bulk conduction band 0.75\text{\,}\mathrm{e}\mathrm{V} above E_{F} , consistent with optical spectroscopy 75 ; 104 and high-energy electron energy loss spectroscopy (EELS) measurements (Supplementary Information IV ). In Fig. 2 d, we show circular dichroism ARPES (CD-ARPES) measurements of the (001) -surface states, which provide further details of their wavefunction textures, and exhibit excellent agreement with theoretical calculations of the surface-projected local (in real space) atomic-orbital contribution to the orbital angular momentum (Fig. 2 e, see Supplementary Information III.3 ). We now investigate the dynamics of the photo-doped surface and bulk conduction bands under different photo-excitation fluences. In Fig. 3 a, we plot ARPES spectra at several different time delays after excitation for two representative fluences. Immediately after excitation, a broad energy range is populated at both low and high fluence (left panels of the top and bottom rows of Fig. 3 a, respectively). The photo-doped population above E_{F} (here, in the vicinity of the bottom of the surface band E_{0} , see Fig. 2 c) specifically corresponds to the non-equilibrium phase schematically depicted in Fig. 1 b. At low fluence, only states at very low energies ( E\nu_{F} (Eq. ( 3 )), which can be accessed by photo-doping experiments like those in the present study. We next turn our attention to the orbital texture of the (001) -surface states, which is closely linked to our CD-ARPES experiments. Specifically, of the different {\bf k} -resolved contributions to the orbital magnetization, CD-ARPES most directly couples to the local (in real space) contribution to the OAM from the underlying atomic orbitals 64 ; 63 ; 97 ; 107 . To probe the atomic-orbital contribution to the OAM, we construct a surface-projected, OAM-dependent spectral function vector \langle{\bf L}(\mathbf{k},E)\rangle , analogous to the surface-projected, spin-dependent spectral function vector \langle{\bf S}(\mathbf{k},E)\rangle in Eq. ( 17 ), whose components are given by 82 ; 77 : \langle L_{\alpha}(\mathbf{k},E)\rangle=-\frac{1}{\pi}\,\mathrm{Im}\,\mathrm{Tr}\bigl[\hat{L}_{\alpha}\,P_{t}G^{R}(\mathbf{k},E)P_{t}\bigr], (19) where G^{R}(\mathbf{k},E) and P_{t} are respectively defined in Eqs. ( 6 ) and ( 7 ), and where \hat{L}_{\alpha} is the \alpha -th component of the OAM operator vector \hat{\bf L}=(\hat{L}_{x},\hat{L}_{y},\hat{L}_{z}) . Like \langle{\bf S}(\mathbf{k},E)\rangle in Eq. ( 17 ), \langle{\bf L}(\mathbf{k},E)\rangle transforms as a \mathcal{T} -odd axial vector, and must hence vanish at each {\bf k} in bulk systems with \mathcal{I} and \mathcal{T} symmetry. However, on the 2D surfaces of 3D systems, \langle{\bf L}(\mathbf{k},E)\rangle can similarly be nonvanishing due to the absence of 3D \mathcal{I} (or \mathcal{I}\times\mathcal{T} ) symmetry. Unlike the spin operator components \hat{S}_{\alpha} , the OAM operator components \hat{L}_{\alpha} have more complicated representations in the Wannier basis, and are also more sensitive to the numerical Wannierization procedure. Specifically, in order to interpret \langle L_{\alpha}(\mathbf{k},E)\rangle as a physical observable, it is crucial that the value of \langle L_{\alpha}(\mathbf{k},E)\rangle at each {\bf k} does not depend on the arbitrary piece of the gauge of the wavefunctions |n\mathbf{k}\rangle (Eq. ( 9 )) of the slab Bloch Hamiltonian H_{\mathrm{slab}}(\mathbf{k}) ( i.e. the free gauge factors that do not affect the real-space model embedding 108 ; 8 ; 47 ). This can most straightforwardly be accomplished by requiring that, like S_{\alpha} in Eq. ( 18 ), each OAM operator component \hat{L}_{\alpha} is represented as a {\bf k} -independent matrix L_{\alpha} in the tight-binding (periodic) gauge of H_{\mathrm{slab}}(\mathbf{k}) . In turn, in order for the matrix representatives L_{\alpha} of \hat{L}_{\alpha} to be {\bf k} -independent, each of the Wannier orbitals associated to the same atomic shell of each atom – specifically the Wannier orbitals for each atom that are assigned the same principal ( n ) and azimuthal ( \ell , i.e. OAM) quantum numbers – must occupy the same positions in real space. This ensures that each OAM operator component \hat{L}_{\alpha} acts locally on the real-space tight-binding basis states, such that the basis-state positions remain invariant under the action of \hat{L}_{\alpha} for all \alpha . In the Wannier tight-binding model that we constructed for the PM state of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in Sec. III.1 , the basis-state positions (Table S1 ) not only satisfy the above requirement, but are even more tightly grouped. Specifically, as shown in Table S1 , each of the Wannier orbitals in our model coincides with the position of its associated atom to within 10^{-4} in the units of the lattice vectors \mathbf{a}_{1,2,3} in Eq. ( 1 ). Our Wannier model is hence suitably constructed for computing the surface-state OAM texture \langle{\bf L}(\mathbf{k},E)\rangle in Eq. ( 19 ). Having established that the surface OAM texture \langle{\bf L}(\mathbf{k},E)\rangle can be computed by applying Eq. ( 19 ) to our Wannier tight-binding model, we next turn our attention to constructing the OAM matrix representatives L_{\alpha} in the Wannier basis. In reduced units in which \hbar=1 , the OAM operator components \hat{L}_{\alpha} are specifically represented in the slab Wannier model constructed in Sec. III.2 as N_{a}N_{L}N_{s}\times N_{a}N_{L}N_{s}=N\times N matrices L_{\alpha} that each take a block form: L_{\alpha}=\left\{\bigoplus_{\xi,\ell_{\xi}\neq 0}L^{\xi,\ell_{\xi}}_{\alpha}\right\}\oplus\left\{\bigoplus_{\xi,\ell_{\xi}=0}\mathbb{0}_{N_{s}N_{L}\times N_{s}N_{L}}\right\}, (20) where each L^{\xi,\ell_{\xi}}_{\alpha} is proportional to the identity in the N_{s}N_{L}\times N_{s}N_{L}=2N_{L}\times 2N_{L} spin and layer subspace (Eq. ( 5 )), and where \mathbb{0}_{N_{s}N_{L}\times N_{s}N_{L}} is the N_{s}N_{L}\times N_{s}N_{L} matrix of zeroes. In Eq. ( 20 ), \xi and \ell_{\xi} respectively run in Table S1 over the “Site” column and the OAM quantum numbers \ell_{\xi} per site in the “Wannier Orbitals” column, such that for example the \ell=2 d and \ell=3 f orbitals of the 1a Eu atom are included in the first direct sum ( \ell_{\xi}\neq 0 ) and excluded in the second ( \ell_{\xi}=0 ), whereas for example the \ell=0 s orbitals of the 2d Cd atoms are excluded from the first direct sum and included in the second. Each L^{\xi,\ell_{\xi}}_{\alpha} in Eq. ( 20 ) is hence an N_{L}N_{s}(2\ell_{\xi}+1)\times N_{L}N_{s}(2\ell_{\xi}+1) matrix 32 . To obtain the matrices L^{\xi,\ell_{\xi}}_{\alpha} in Eq. ( 20 ) in the Wannier tight-binding basis, we first recognize that each atomic orbital in our model is expressed in the real cubic–harmonic basis 57 : \bigl\{s,p_{x},p_{y},p_{z},d_{z^{2}},d_{xz},d_{yz},d_{x^{2}-y^{2}},d_{xy},\ldots\bigr\}, (21) rather than in the complex spherical–harmonic basis \{\,Y_{\ell}^{m}\,\} in which L^{\xi,\ell_{\xi}}_{z} is a diagonal matrix whose only nonzero elements take values given by the orbital magnetic quantum numbers m 32 ; 54 . Our strategy for constructing each L^{\xi,\ell_{\xi}}_{x,y,z} matrix therefore consists of two steps: (i) we first construct L^{\xi,\ell_{\xi}}_{x,y,z} in the familiar basis of spherical harmonics, and (ii) we then apply a unitary transformation to rotate L^{\xi,\ell_{\xi}}_{x,y,z} into the cubic-harmonic Wannier basis. We next recognize that given an OAM matrix representative (L^{\xi,\ell_{\xi}}_{z})^{(\ell)} in the spherical-harmonic basis (denoted by the superscript (\ell) here and below) with the eigenstates |\ell,m\rangle such that: (L^{\xi,\ell_{\xi}}_{z})^{(\ell)}|\ell,m\rangle=m|\ell,m\rangle, (22) there exist ladder operators \hat{L}_{\pm}^{\xi,\ell_{\xi}} that change m by one unit within the \ell_{\xi} manifold 74 : \hat{L}^{\xi,\ell_{\xi}}_{\pm}\,|\ell,m\rangle=\sqrt{\ell_{\xi}(\ell_{\xi}+1)-m(m\pm 1)}\;|\ell,m\pm 1\rangle. (23) Using Eq. ( 23 ), we then construct the matrix representatives (L^{\xi,\ell_{\xi}}_{\pm})^{(\ell)} of \hat{L}^{\xi,\ell_{\xi}}_{\pm} , whose matrix elements are given by: \left[(L^{\xi,\ell_{\xi}}_{\pm})^{(\ell)}\right]_{m,m^{\prime}}=\sqrt{\ell_{\xi}(\ell_{\xi}+1)-m^{\prime}(m^{\prime}\pm 1)}\;\delta_{m,m^{\prime}\pm 1}. (24) The \alpha=x,y,z OAM matrix representatives in the spherical-harmonic basis are then by definition given by 74 ; 32 : (L^{\xi,\ell_{\xi}}_{x})^{(\ell)}=\frac{1}{2}\left\{(L^{\xi,\ell_{\xi}}_{+})^{(\ell)}+(L^{\xi,\ell_{\xi}}_{-})^{(\ell)}\right\},\qquad(L^{\xi,\ell_{\xi}}_{y})^{(\ell)}=\frac{1}{2i}\left\{(L^{\xi,\ell_{\xi}}_{+})^{(\ell)}-(L^{\xi,\ell_{\xi}}_{-})^{(\ell)}\right\}. (25) Lastly, each cubic-harmonic (Wannier-basis) eigenstate |\ell,\mu\rangle_{\rm cub} represents a linear combination of the spherical-harmonic eigenstates |\ell,m\rangle 57 : |\ell,\mu\rangle_{\rm cub}=\sum_{m=-\ell}^{+\ell}\left[U^{\ell}\right]_{\mu m}\,|\ell,m\rangle. (26) Using Eq. ( 26 ), we hence obtain the unitary matrix U^{\ell} that transforms the OAM matrix representatives (L^{\xi,\ell_{\xi}}_{\alpha})^{(\ell)} in the spherical-harmonic basis into the OAM matrix representatives L^{\xi,\ell_{\xi}}_{\alpha} in the cubic-harmonic basis of our Wannier tight-binding model (Eq. ( 20 )): L^{\xi,\ell_{\xi}}_{\alpha}=U^{\ell}\,(L^{\xi,\ell_{\xi}}_{\alpha})^{(\ell)}\,\big(U^{\ell}\big)^{\dagger},\qquad\alpha\in\{x,y,z\}. (27) As shown in Table S1 , our Wannier model contains s ( \ell=0 ), p ( \ell=1 ), d ( \ell=2 ), and f ( \ell=3 ) orbitals. Using the procedure detailed in Eqs. ( 20 ) through ( 27 ), we may now construct the OAM matrix representatives L_{\alpha} required to compute \langle{\bf L}(\mathbf{k},E)\rangle in Eq. ( 19 ). For each of the orbital shells with \ell\neq 0 enumerated below, we will specifically provide explicit matrix representatives in the form of (2\ell+1)\times(2\ell+1) square matrices, each of which is taken to be shorthand notation for a larger N_{L}N_{s}(2\ell+1)\times N_{L}N_{s}(2\ell+1) matrix that is proportional to the identity in the N_{s}N_{L}\times N_{s}N_{L} spin and layer subspace (Eq. ( 5 )). First, for the p shell of the 2d As atoms (denoted with \xi,\ell_{\xi}\equiv(p) ), the real cubic harmonics are expressed in terms of the spherical harmonics Y_{1}^{m} as: p_{z}=Y_{1}^{0},\qquad p_{x}=\frac{1}{\sqrt{2}}\bigl(Y_{1}^{-1}-Y_{1}^{1}\bigr),\qquad p_{y}=\frac{i}{\sqrt{2}}\bigl(Y_{1}^{-1}+Y_{1}^{1}\bigr). (28) In the cubic-harmonic ordering \{p_{z},p_{x},p_{y}\} , the ladder operators L^{(p)}_{\pm} in the Wannier basis are then represented by: L^{(p)}_{+}=\begin{pmatrix}0&1&-i\\ -1&0&0\\ i&0&0\end{pmatrix},\qquad L^{(p)}_{-}=\begin{pmatrix}0&-1&-i\\ 1&0&0\\ i&0&0\end{pmatrix}, (29) and the block 3\times 3 OAM representatives L_{\alpha}^{(p)} , \alpha\in\{x,y,z\} in the Wannier basis are given by: L_{x}^{(p)}=\begin{pmatrix}0&0&-i\\ 0&0&0\\ i&0&0\end{pmatrix},\qquad L_{y}^{(p)}=\begin{pmatrix}0&-i&0\\ i&0&0\\ 0&0&0\end{pmatrix},\qquad L_{z}^{(p)}=\begin{pmatrix}0&0&0\\ 0&0&i\\ 0&-i&0\end{pmatrix}. (30) Next, for the d shell of the 1a Eu atom (denoted with \xi,\ell_{\xi}\equiv(d) ), the real cubic harmonics are expressed in terms of the spherical harmonics Y_{2}^{m} as: \displaystyle d_{z^{2}} \displaystyle=Y_{2}^{0}, (31) \displaystyle d_{xz} \displaystyle=\frac{1}{\sqrt{2}}\!\left(Y_{2}^{-1}-Y_{2}^{1}\right), \displaystyle d_{yz} \displaystyle=\frac{i}{\sqrt{2}}\!\left(Y_{2}^{-1}+Y_{2}^{1}\right), \displaystyle d_{x^{2}-y^{2}} \displaystyle=\frac{1}{\sqrt{2}}\!\left(Y_{2}^{2}+Y_{2}^{-2}\right), \displaystyle d_{xy} \displaystyle=\frac{i}{\sqrt{2}}\!\left(Y_{2}^{2}-Y_{2}^{-2}\right). In the cubic-harmonic ordering \{d_{z^{2}},d_{xz},d_{yz},d_{x^{2}-y^{2}},d_{xy}\} , the ladder operators L^{(d)}_{\pm} in the Wannier basis are then represented by: L^{(d)}_{+}=\begin{pmatrix}0&0&-1&-i&0\\ 0&0&i&-1&0\\ 1&-i&0&0&-\sqrt{3}\\ i&1&0&0&\sqrt{3}i\\ 0&0&\sqrt{3}&-\sqrt{3}i&0\end{pmatrix},\qquad L^{(d)}_{-}=\begin{pmatrix}0&0&1&-i&0\\ 0&0&i&1&0\\ -1&-i&0&0&\sqrt{3}\\ i&-1&0&0&\sqrt{3}i\\ 0&0&-\sqrt{3}&-\sqrt{3}i&0\end{pmatrix}, (32) and the block 5\times 5 OAM representatives L_{\alpha}^{(d)} , \alpha\in\{x,y,z\} in the Wannier basis are given by: L_{x}^{(d)}=\begin{pmatrix}0&0&0&-i&0\\ 0&0&i&0&0\\ 0&-i&0&0&0\\ i&0&0&0&\sqrt{3}\,i\\ 0&0&0&-\sqrt{3}\,i&0\end{pmatrix},\qquad L_{y}^{(d)}=\begin{pmatrix}0&0&i&0&0\\ 0&0&0&i&0\\ -i&0&0&0&\sqrt{3}\,i\\ 0&-i&0&0&0\\ 0&0&-\sqrt{3}\,i&0&0\end{pmatrix},\qquad L_{z}^{(d)}=\begin{pmatrix}0&2i&0&0&0\\ -2i&0&0&0&0\\ 0&0&0&i&0\\ 0&0&-i&0&0\\ 0&0&0&0&0\end{pmatrix}. (33) Lastly, for the f shell of the 1a Eu atom (denoted with \xi,\ell_{\xi}\equiv(f) ), the real cubic harmonics are expressed in terms of the spherical harmonics Y_{3}^{m} as: \displaystyle f_{z^{3}} \displaystyle=Y_{3}^{0}, (34) \displaystyle f_{xz^{2}} \displaystyle=\frac{1}{\sqrt{2}}\bigl(Y_{3}^{-1}-Y_{3}^{1}\bigr), \displaystyle f_{yz^{2}} \displaystyle=\frac{i}{\sqrt{2}}\bigl(Y_{3}^{-1}+Y_{3}^{1}\bigr), \displaystyle f_{z(x^{2}-y^{2})} \displaystyle=\frac{1}{\sqrt{2}}\bigl(Y_{3}^{2}+Y_{3}^{-2}\bigr), \displaystyle f_{xyz} \displaystyle=\frac{i}{\sqrt{2}}\bigl(Y_{3}^{2}-Y_{3}^{-2}\bigr), \displaystyle f_{x(x^{2}-3y^{2})} \displaystyle=\frac{1}{\sqrt{2}}\bigl(Y_{3}^{3}-Y_{3}^{-3}\bigr), \displaystyle f_{y(3x^{2}-y^{2})} \displaystyle=\frac{i}{\sqrt{2}}\bigl(Y_{3}^{3}+Y_{3}^{-3}\bigr). In the cubic-harmonic ordering \{f_{z^{3}},f_{xz^{2}},f_{yz^{2}},f_{z(x^{2}-y^{2})},f_{xyz},f_{x(x^{2}-3y^{2})},f_{y(3x^{2}-y^{2})}\} , the ladder operators L^{(f)}_{\pm} in the Wannier basis are then represented by: L^{(f)}_{+}=\begin{pmatrix}0&\sqrt{6}&-\sqrt{6}i&0&0&0&0\\ -\sqrt{6}&0&0&\frac{\sqrt{10}}{2}&\frac{\sqrt{10}i}{2}&0&0\\ \sqrt{6}i&0&0&\frac{\sqrt{10}i}{2}&-\frac{\sqrt{10}}{2}&0&0\\ 0&-\frac{\sqrt{10}}{2}&-\frac{\sqrt{10}i}{2}&0&0&\frac{\sqrt{6}}{2}&-\frac{\sqrt{6}i}{2}\\ 0&-\frac{\sqrt{10}i}{2}&\frac{\sqrt{10}}{2}&0&0&-\frac{\sqrt{6}i}{2}&-\frac{\sqrt{6}}{2}\\ 0&0&0&-\frac{\sqrt{6}}{2}&\frac{\sqrt{6}i}{2}&0&0\\ 0&0&0&\frac{\sqrt{6}i}{2}&\frac{\sqrt{6}}{2}&0&0\end{pmatrix}, L^{(f)}_{-}=\begin{pmatrix}0&-\sqrt{6}&-\sqrt{6}i&0&0&0&0\\ \sqrt{6}&0&0&-\frac{\sqrt{10}}{2}&\frac{\sqrt{10}i}{2}&0&0\\ \sqrt{6}i&0&0&\frac{\sqrt{10}i}{2}&\frac{\sqrt{10}}{2}&0&0\\ 0&\frac{\sqrt{10}}{2}&-\frac{\sqrt{10}i}{2}&0&0&-\frac{\sqrt{6}}{2}&-\frac{\sqrt{6}i}{2}\\ 0&-\frac{\sqrt{10}i}{2}&-\frac{\sqrt{10}}{2}&0&0&-\frac{\sqrt{6}i}{2}&\frac{\sqrt{6}}{2}\\ 0&0&0&\frac{\sqrt{6}}{2}&\frac{\sqrt{6}i}{2}&0&0\\ 0&0&0&\frac{\sqrt{6}i}{2}&-\frac{\sqrt{6}}{2}&0&0\end{pmatrix}, (35) and the block 7\times 7 OAM representatives L_{\alpha}^{(f)} , \alpha\in\{x,y,z\} in the Wannier basis are given by: L_{x}^{(f)}=\begin{pmatrix}0&0&-\sqrt{6}\,i&0&0&0&0\\ 0&0&0&0&\frac{\sqrt{10}}{2}i&0&0\\ \sqrt{6}\,i&0&0&\frac{\sqrt{10}}{2}i&0&0&0\\ 0&0&-\frac{\sqrt{10}}{2}i&0&0&0&-\frac{\sqrt{6}}{2}i\\ 0&-\frac{\sqrt{10}}{2}i&0&0&0&-\frac{\sqrt{6}}{2}i&0\\ 0&0&0&0&\frac{\sqrt{6}}{2}i&0&0\\ 0&0&0&\frac{\sqrt{6}}{2}i&0&0&0\end{pmatrix}, L_{y}^{(f)}=\begin{pmatrix}0&-\sqrt{6}\,i&0&0&0&0&0\\ \sqrt{6}\,i&0&0&-\frac{\sqrt{10}}{2}i&0&0&0\\ 0&0&0&0&\frac{\sqrt{10}}{2}i&0&0\\ 0&\frac{\sqrt{10}}{2}i&0&0&0&-\frac{\sqrt{6}}{2}i&0\\ 0&0&-\frac{\sqrt{10}}{2}i&0&0&0&\frac{\sqrt{6}}{2}i\\ 0&0&0&\frac{\sqrt{6}}{2}i&0&0&0\\ 0&0&0&0&-\frac{\sqrt{6}}{2}i&0&0\end{pmatrix},\qquad L_{z}^{(f)}=\begin{pmatrix}0&0&0&0&0&0&0\\ 0&0&i&0&0&0&0\\ 0&-i&0&0&0&0&0\\ 0&0&0&0&-2i&0&0\\ 0&0&0&2i&0&0&0\\ 0&0&0&0&0&0&3i\\ 0&0&0&0&0&-3i&0\end{pmatrix}. (36) We have importantly confirmed by direct computation that the OAM matrix representatives L_{\alpha}^{\xi,\ell_{\xi}} in Eqs. ( 30 ) through ( 36 ) satisfy the canonical angular momentum commutation relations 32 : [L_{\alpha}^{\xi,\ell_{\xi}},L_{\beta}^{\xi,\ell_{\xi}}]=i\,\varepsilon^{\alpha\beta\gamma}L_{\gamma}^{\xi,\ell_{\xi}}, (37) where \varepsilon^{\alpha\beta\gamma} is the Levi–Civita tensor and \alpha,\beta,\gamma\in\{x,y,z\} . We have also numerically confirmed that the L_{\alpha}^{\xi,\ell_{\xi}} matrices for each angular momentum shell satisfy the total angular momentum summation relation: (L^{\xi,\ell_{\xi}})^{2}=\sum_{\alpha=x,y,z}(L_{\alpha}^{\xi,\ell_{\xi}})^{2}=\ell_{\xi}(\ell_{\xi}+1)\,\mathbb{I}_{N_{L}N_{s}(2\ell_{\xi}+1)\times N_{L}N_{s}(2\ell_{\xi}+1)}, (38) where \mathbb{I}_{N_{L}N_{s}(2\ell_{\xi}+1)\times N_{L}N_{s}(2\ell_{\xi}+1)} is the N_{L}N_{s}(2\ell_{\xi}+1)\times N_{L}N_{s}(2\ell_{\xi}+1) identity matrix (see the text surrounding Eq. ( 20 )). Having established our OAM formalism, we may now finally compute the (001) -surface OAM texture \langle{\bf L}(\mathbf{k},E)\rangle (Eq. ( 19 )) in our Wannier-based model of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} . In our CD-ARPES experiments, measurements were performed with the photon scattering plane aligned along the the (x,z) -plane of the sample (see Fig. 2 b,d of the main text and the Methods section). In momentum space, the photon scattering plane coincides with a crystallographic mirror plane whose normal vector lies along k_{y} and is related to the BZ line \overline{\Gamma}-\overline{\mathrm{K}} in Fig. S3 a by crystal symmetry. We specifically employed in our CD-ARPES measurements 6\text{\,}\mathrm{e}\mathrm{V} circularly polarized photons with bulk and surface sensitivity that were incident along an axis aligned [formula] degrees between the x and z axes. We then extracted the CD-ARPES signature from the respective ARPES intensities I_{CD}=I_{RCP}-I_{LCP} . We hence interpret our CD-ARPES measurements as probing the \langle L_{xz}\rangle=\langle\frac{L_{x}+L_{z}}{\sqrt{2}}\rangle component of the (001) -surface OAM texture. In Fig. S6 c, we therefore plot \langle L_{xz}({\bf k},E)\rangle for the Wannier-based model at the same constant energy as the overall surface spectral function A_{\mathrm{surf}}(\mathbf{k},E) in Fig. S6 a, using a signed log scale with a color map in which yellow is positive and dark blue is negative. The \langle L_{xz}({\bf k},E)\rangle OAM texture component in Fig. S6 c shows excellent agreement with our experimental CD-ARPES data measured at the same energy relative to the surface band minimum both above and below the PM to FM transition temperature T_{C}\approx$25\text{\,}\mathrm{K}$ (see Fig. S2 c,d and the text following Eq. ( 16 )). Interestingly, unlike the \langle S_{xz}({\bf k},E)\rangle spin texture component in Fig. S6 b, \langle L_{xz}({\bf k},E)\rangle in Fig. S6 c exhibits the same sign for both concentric ring-like Fermi pockets of the surface Dirac cone. To demonstrate a scenario that qualitatively reproduces the spin and OAM textures in Fig. S6 b,c, we introduce a simple {\bf k}\cdot{\bf p} model \mathcal{H}_{kp}(k) that represents a 1D line cut along k through a \mathcal{T} -invariant point k=0 in a 2D surface BZ: \mathcal{H}_{kp}(k)=k\tau_{z}+M\tau_{x}+v\tau_{z}\sigma_{z}. (39) In Eq. ( 39 ), there are four (surface-localized) basis states that arise from spin-degenerate p_{x}\pm ip_{y} orbitals with the orbital magnetic quantum numbers ( L_{z} eigenvalues) m=\pm 1 (see the text surrounding Eq. ( 21 )), and hence overall form a total angular momentum j=3/2 representation 32 . We employ a notation in Eq. ( 39 ) in which: \tau_{i}\sigma_{j}=s_{j}\otimes s_{i}, (40) where s_{i},\ i\in\{x,y,z\} are the 2\times 2 Pauli matrices, s_{0} is the 2\times 2 identity matrix, and where factors of \tau_{0} and \sigma_{0} have been suppressed for notation simplicity. The matrices \tau_{z} and \sigma_{z} respectively index the p_{x}\pm ip_{y} orbital and spin-1/2 system subspaces, such that the matrix representatives of \hat{S}_{z} (Eq. ( 17 )) and \hat{L}_{z} (Eq. ( 19 )) are respectively given by: S_{z}=\sigma_{z},\qquad L_{z}=\tau_{z}, (41) and \mathcal{T} symmetry is represented through the action: \mathcal{T}\mathcal{H}_{kp}(k)\mathcal{T}^{-1}=\tau_{x}\sigma_{y}\mathcal{H}^{*}_{kp}(-k)\tau_{x}\sigma_{y}. (42) In \mathcal{H}_{kp}(k) in Eq. ( 39 ), the velocity term k\tau_{z} hence induces fully OAM-polarized dispersion kL_{z} , M\tau_{x} is a {\bf k} -independent OAM coupling term that breaks L_{z} symmetry, and v\tau_{z}\sigma_{z} is an S_{z} -symmetric SOC splitting term. Figure S7: Low-energy toy model of 2D Dirac surface states with contrasting spin and OAM polarizations: In this figure, we plot the spectral function A_{\mathrm{surf}}(k,E) (Eq. ( 8 ), \langle S_{z}(k,E)\rangle spin texture component (Eq. ( 17 )), and \langle L_{z}(k,E)\rangle OAM texture component (Eq. ( 19 )) of a simplified, \mathcal{T} -invariant {\bf k}\cdot{\bf p} model Hamiltonian \mathcal{H}_{kp}(k) along a 1D line parameterized by k in a 2D surface BZ (Eqs. ( 39 ), ( 41 ), ( 42 ), and ( 43 )). (a, b, c) A_{\mathrm{surf}}(k,E) , \langle S_{z}(k,E)\rangle , and \langle L_{z}(k,E)\rangle , respectively, for \mathcal{H}_{kp}(k) with M=0 and v=-0.05 . The spectrum in (a) consists of four linear bands with (b) narrow spin splitting, where the bands within each spin-split pair exhibit (c) \langle L_{z}\rangle OAM polarizations with the same sign originating from the OAM dispersion term k_{z}\tau_{z}=kL_{z} in Eqs. ( 39 ) and ( 41 ). (d, e, f) A_{\mathrm{surf}}(k,E) , \langle S_{z}(k,E)\rangle , and \langle L_{z}(k,E)\rangle , respectively, for \mathcal{H}_{kp}(k) with vanishing SOC and significant OAM coupling ( M=-0.1 , v=0 ). The spectrum in (d) consists of two pairs of spin-degenerate bands with hyperbolic dispersion that (e) exhibit vanishing \langle S_{z}\rangle at all k , due to the spin degeneracy, and (f) carry the same sign of \langle L_{z}\rangle within each spin-degenerate pair. (g, h, i) A_{\mathrm{surf}}(k,E) , \langle S_{z}(k,E)\rangle , and \langle L_{z}(k,E)\rangle , respectively, for the previous case in (d,e,f) subject to additional weak SOC ( M=-0.1 , v=-0.05 ). The spectrum in (g) consists of two \mathcal{T} -enforced twofold Dirac degeneracies at k=0 , where the bands from each Dirac point exhibit linear dispersion at small k , but otherwise largely follow the hyperbolic energy dispersion in (d). As in our Wannier-based calculations of the unoccupied (001) -surface states of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} (Fig. S6 ), the two singly degenerate branches of each Dirac cone in (g) exhibit (h) oppositely signed \langle S_{z}\rangle spin polarizations, but (i) carry \langle L_{z}\rangle OAM polarizations with the same sign. In Fig. S7 , we respectively plot the total spectral function A_{\mathrm{surf}}(k,E) (Eq. ( 8 ), panels a,d,g in the top row), the \langle S_{z}\rangle component of the spin texture \langle{\bf S}(k,E)\rangle (Eq. ( 17 ), panels b,d,e in the middle row), and the \langle L_{z}\rangle component of the OAM texture \langle{\bf L}(k,E)\rangle (Eq. ( 19 ), panels c,f,i in the bottom row) as functions of E and k for \mathcal{H}_{kp}(k) with varying values of M and v in Eq. ( 39 ). For all panels in Fig. S7 , we have taken the top-surface projector P_{t} (Eq. ( 7 )) to be the 4\times 4 identity matrix: P_{t}=\mathbb{I}_{4\times 4}, (43) because the low-energy Hamiltonian \mathcal{H}_{kp}(k) is already projected into a (sub)space of surface-localized basis states. We first in Fig. S7 a,b,c plot A_{\mathrm{surf}}(k,E) , \langle S_{z}(k,E)\rangle , and \langle L_{z}(k,E)\rangle for \mathcal{H}_{kp}(k) with M=0 and v=-0.05 . The spectral function A_{\mathrm{surf}}(k,E) in Fig. S7 a consists of four linear bands with narrow spin splitting (Fig. S7 b), where the bands within each spin-split pair exhibit \langle L_{z}\rangle OAM polarizations with the same sign (Fig. S7 c). Next, in Fig. S7 d,e,f, we plot A_{\mathrm{surf}}(k,E) , \langle S_{z}(k,E)\rangle , and \langle L_{z}(k,E)\rangle for a case with vanishing SOC and significant OAM coupling in which M=-0.1 and v=0 . In this case, A_{\mathrm{surf}}(k,E) in Fig. S7 d consists of two pairs of spin-degenerate bands with hyperbolic dispersion, for which \langle S_{z}(k,E)\rangle is vanishing at all k due to the spin degeneracy (Fig. S7 e). Importantly, however, the OAM texture \langle L_{z}(k,E)\rangle in Fig. S7 f remains nonvanishing, and the two doubly degenerate bands within each pair continue to exhibit the same sign of \langle L_{z}\rangle at each k. We lastly consider the case of weakly introducing SOC into the previous scenario, and therefore in Fig. S7 g,h,i plot A_{\mathrm{surf}}(k,E) , \langle S_{z}(k,E)\rangle , and \langle L_{z}(k,E)\rangle for \mathcal{H}_{kp}(k) with M=-0.1 and v=-0.05 . The energy spectrum in Fig. S7 g now consists of two \mathcal{T} -enforced Dirac cones whose two branches are only weakly spin-split (Fig. S7 h) and pairwise disperse in the same directions except at very small values of k . Specifically, focusing on the upper Dirac cone in energy in Fig. S7 g, the two Dirac bands together form an essentially hyperbolic dispersion with positive velocities for all k>0 outside of a very small momentum range in the vicinity of k=0 . This closely resembles the dispersion and spin texture distribution of the unoccupied (001) -surface states in our Wannier-based model of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} (Figs. S3 b, S5 a, and S6 a,b). Also like the (001) -surface Dirac cone in our Wannier-based calculations (Fig. S6 c), the two singly degenerate branches of each Dirac cone in Fig. S7 i exhibit \langle L_{z}\rangle OAM polarizations with the same sign, despite exhibiting oppositely signed \langle S_{z}\rangle spin polarizations (Fig. S7 h), owing to the dominant OAM dispersion term k\tau_{z}=kL_{z} in Eqs. ( 39 ) and ( 41 ). To conclude, we have introduced a simple {\bf k}\cdot{\bf p} model that exhibits a 2D surface Dirac fermion with hyperbolic dispersion, weak spin splitting, and an OAM polarization distribution that mirrors our first-principles calculations of the (001) -surface spectrum of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} . It is important to note, however, that the simple model in Eq. ( 39 ) only represents a demonstration that it is feasible to realize a surface Dirac cone with these characteristics, and we leave for future works the more detailed question of the specific physical mechanism that gives rise to the (001) -surface energy dispersion and angular momentum textures in \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} . III.4 Topological classification and band topology above the Fermi level in \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} In our calculations of the (001) -surface spectrum of the PM phase of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in Secs. III.2 and III.3 , we observed (weakly SOC-split) 2D surface Dirac bands around {\bf k}={\bf 0} with an energy minimum \widetilde{E} above the Fermi level E_{F} (Figs. S3 b, S5 a, and S6 ). We will next investigate whether it is possible to associate the unoccupied Dirac surface states in \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} to bulk topological bands. To address this question, we employ the approach of Ref. 94 , in which the methods of Topological Quantum Chemistry 13 ; 21 ; 101 and symmetry-based indicators (SIs) 42 ; 68 ; 87 ; 37 were applied to efficiently screen band structures in nonmagnetic materials (or material phases) at and away from E_{F} for nontrivial topology that is detectable from crystal symmetry eigenvalues. The authors of Ref. 94 specifically first performed DFT calculations using the Vienna Ab-initio Simulation Package ( VASP ) 39 ; 40 , then applied the freely accessible Fortran program VASP2Trace ( https://github.com/zjwang11/irvsp ) 26 to extract the eigenvalues of the unitary crystal symmetries, and lastly uploaded the extracted symmetry eigenvalues to the Check Topological Mat tool on the Bilbao Crystallographic Server (BCS) to determine the band topology (accessible by navigating from http://webbdcrista2.ehu.es/ to https://cryst.ehu.es/cgi-bin/cryst/programs/magnetictopo.pl?tipog=gesp ) 5 ; 4 ; 3 ; 93 . Because our DFT calculations in Sec. III.1 were performed using Quantum Espresso 28 ; 27 ; 29 and not VASP , we began by implementing custom code, detailed below, to extract the crystal symmetry eigenvalues from our band structure calculations in a format compatible with Check Topological Mat . To construct an input file for Check Topological Mat from our DFT band structure calculation in Fig. S3 a, we first analyzed nonmagnetic SG 164 ( P\bar{3}m1 ), the 3D symmetry group G of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in the PM state, to determine the set of maximal {\bf k} points ( i.e. the \mathcal{T} -invariant {\bf k} points, as well as high-symmetry points without \mathcal{T} like K and H in Fig. S3 a, see Refs. 13 ; 21 ). At each maximal {\bf k} point, there specifically exists an infinite subgroup of symmetries G_{\bf k}\subseteq G , termed the little group, for which each symmetry: g\in G_{\bf k}, (44) leaves {\bf k} invariant up to an integer linear combination {\bf K} of the reciprocal lattice vectors \mathbf{b}_{1,2,3} in Eq. ( 2 ): g{\bf k}\text{ mod }{\bf K}={\bf k}. (45) For each grouping of bands that share the same energy eigenvalue E_{n\mathbf{k}} (up to a numerical tolerance factor of 0.002 eV), we next obtained the set of m (approximately) degenerate Bloch eigenstates \{|\psi^{1}_{n\mathbf{k}}\rangle,|\psi^{2}_{n\mathbf{k}}\rangle,\ldots,|\psi^{m}_{n\mathbf{k}}\rangle\} in an orthonormal basis. The m states that share the energy E_{n\mathbf{k}} may either form an enforced degeneracy, or may simply be a set of states for which SOC and orbital splitting is very weak, but in either case transform into themselves or other degenerate states in the set under the action of the little group symmetries g\in G_{\bf k} 12 ; 13 . Using the spglib software library 91 , we then obtained a finite set of coset symmetry representatives with respect to the infinite group of translations for the full SG G , and restricted consideration to those that are also coset representatives of G_{\bf k} using Eq. ( 45 ). Of the coset representatives of G_{\bf k} obtained from spglib , a smaller subset consists only of unitary symmetries O , which hence carry eigenvalues that can be numerically obtained. Specifically, each unitary coset representative O\in G_{\bf k} admits a decomposition of the form 79 ; 12 : O=\{R|t\}, (46) where R is a point group operation and t is a (possibly fractional) translation. For the m (approximately) degenerate Bloch states with energy E_{n\mathbf{k}} , the action of the unitary little group symmetry O is represented by an m\times m matrix O^{\mathbf{k}} : O|\psi^{i}_{n\mathbf{k}}\rangle=\sum_{j=1}^{m}\left[O^{\mathbf{k}}\right]_{ij}\,|\psi^{j}_{n\mathbf{k}}\rangle, (47) where the matrix elements of O^{\mathbf{k}} are hence given by: \left[O^{\mathbf{k}}\right]_{ij}=\langle\psi^{j}_{n\mathbf{k}}|\,O\,|\psi^{i}_{n\mathbf{k}}\rangle. (48) In the plane-wave spinor basis of our DFT calculations, O in Eq. ( 46 ) acts by transforming \mathbf{k} vectors under R , multiplying by the phase factor e^{-i(R\mathbf{k})\cdot t} from the combined action of R and t , and by applying the SU(2) spin rotation that corresponds to the proper rotation piece of R . For each O and (approximate) band multiplet at E_{n\mathbf{k}} , we lastly obtain a character (trace) \chi_{n\mathbf{k}}(O) by computing: \chi_{n\mathbf{k}}(O)=\sum_{i=1}^{m}\langle\psi^{i}_{n\mathbf{k}}|O|\psi^{i}_{n\mathbf{k}}\rangle=\mathrm{Tr}\ O^{\mathbf{k}}. (49) The characters \chi_{n\mathbf{k}}(O) are equal to the sum of the symmetry eigenvalues of O for the m degenerate states at E_{n\mathbf{k}} , and across all maximal {\bf k} points and band degeneracies form a set of symmetry data vectors that can be used to determine band connectivity and topology 13 ; 21 . In Table S2 , we reproduce the output from Check Topological Mat for the symmetry character trace file generated by our code (the tabulation of each \chi_{n\mathbf{k}}(O) in Eq. ( 49 )) applied to the DFT calculation of the PM-state electronic structure of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in Sec. III.1 . The rows of Table S2 indicate groupings of connected bands above the core electronic shell, listed in the order realized by artificially, independently filling the Bloch states at each {\bf k} , rather than using a fixed, horizontal Fermi energy. This allows energetically isolated bands and gaps to be classified as “topological band insulators” that, while metallic in the bulk, still exhibit topological boundary states 100 ; 101 . In order, the columns in Table S2 list the irreducible small corepresentations associated to the characters \chi_{n\mathbf{k}}(O) (Eq. ( 49 )) at each maximal momentum {\bf k} in Fig. S3 a, the total number of connected bands in the grouping including spin (dim), the classification of the bands in the language of Topological Quantum Chemistry (type, e.g. SEBR, see Ref. 94 ), the symmetry indicators (SIs) (Z_{2,3},Z_{4}) in SG 164 ( P\bar{3}m1 ) of the energetically isolated bands (top. ind., respectively the k_{z}=\pi weak \mathbb{Z}_{2} invariant and the strong \mathbb{Z}_{4} invariant), the cumulative band dimension realized by individually occupying each Bloch state up to an electronic filling \nu above the core shell (filling \nu ), and the cumulative topology and SIs (Z_{2,3},Z_{4}) of the symmetry-allowed gap at \nu (all type / top. ind.). Most relevant to the present study of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} , Z_{4}\text{ mod }2 in Table S2 corresponds to the Fu-Kane \mathbb{Z}_{2} invariant for strong 3D topological insulating (TI) states 25 ; 24 ; 33 , such that the rows with odd values of Z_{4} in “all type / top. ind.” indicate electronic fillings at which topological surface Dirac cones may appear, modulo the effects of surface chemical potentials 94 . \Gamma A H K L M dim type top. ind. filling \nu all type / top. ind. (Z_{2,3},Z_{4}) (Z_{2,3},Z_{4}) \bar{\Gamma}_{8}(2) \bar{A}_{8}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{3}\bar{M}_{4}(2) 2 TRIVIAL (0,0) 2 TRIVIAL / (0,0) \bar{\Gamma}_{9}(2) \bar{A}_{9}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 TRIVIAL (0,0) 4 TRIVIAL / (0,0) \bar{\Gamma}_{9}(2) \bar{A}_{9}(2) \bar{H}_{4}\bar{H}_{5}(2) \bar{K}_{4}\bar{K}_{5}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 FRAGILE (0,0) 6 TRIVIAL / (0,0) \bar{\Gamma}_{6}\bar{\Gamma}_{7}(2) \bar{A}_{6}\bar{A}_{7}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 FRAGILE (0,0) 8 TRIVIAL / (0,0) \bar{\Gamma}_{8}(2) \bar{A}_{8}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{3}\bar{M}_{4}(2) 2 TRIVIAL (0,0) 10 TRIVIAL / (0,0) \bar{\Gamma}_{9}(2) \bar{A}_{9}(2) \bar{H}_{4}\bar{H}_{5}(2) \bar{K}_{4}\bar{K}_{5}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 FRAGILE (0,0) 12 TRIVIAL / (0,0) \bar{\Gamma}_{4}\bar{\Gamma}_{5}(2) \bar{A}_{4}\bar{A}_{5}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{3}\bar{M}_{4}(2) 2 SEBR (1,3) 14 SEBR / (1,3) \bar{\Gamma}_{6}\bar{\Gamma}_{7}(2) \bar{A}_{6}\bar{A}_{7}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 SEBR (1,1) 16 TRIVIAL / (0,0) \bar{\Gamma}_{8}(2) \bar{A}_{9}(2) \bar{H}_{4}\bar{H}_{5}(2) \bar{K}_{4}\bar{K}_{5}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 SEBR (1,0) 18 SEBR / (1,0) \bar{\Gamma}_{9}(2) \bar{A}_{8}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{3}\bar{M}_{4}(2) 2 SEBR (1,0) 20 TRIVIAL / (0,0) \bar{\Gamma}_{8}(2) \bar{A}_{8}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{3}\bar{M}_{4}(2) 2 SEBR (1,3) 22 SEBR / (1,3) \bar{\Gamma}_{9}(2) \bar{A}_{9}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 SEBR (1,1) 24 TRIVIAL / (0,0) \bar{\Gamma}_{4}\bar{\Gamma}_{5}(2) \bar{A}_{6}\bar{A}_{7}(2) \bar{H}_{4}\bar{H}_{5}(2) \bar{K}_{4}\bar{K}_{5}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 SEBR (1,0) 26 SEBR / (1,0) \bar{\Gamma}_{8}(2) \bar{A}_{9}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{3}\bar{M}_{4}(2) 2 TRIVIAL (0,0) 28 SEBR / (1,0) \bar{\Gamma}_{6}\bar{\Gamma}_{7}(2) \bar{A}_{4}\bar{A}_{5}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{3}\bar{M}_{4}(2) 2 SEBR (1,0) 30 TRIVIAL / (0,0) \bar{\Gamma}_{9}(2) \bar{A}_{8}(2) \bar{H}_{4}\bar{H}_{5}(2) \bar{K}_{4}\bar{K}_{5}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 FRAGILE (0,0) 32 TRIVIAL / (0,0) \bar{\Gamma}_{8}(2) \bar{A}_{9}(2) \bar{H}_{4}\bar{H}_{5}(2) \bar{K}_{4}\bar{K}_{5}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{3}\bar{M}_{4}(2) 2 SEBR (1,1) 34 SEBR / (1,1) \bar{\Gamma}_{8}(2) \bar{A}_{8}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 SEBR (1,2) 36 SEBR / (0,3) \bar{\Gamma}_{4}\bar{\Gamma}_{5}(2) \bar{\Gamma}_{9}(2) \bar{A}_{8}(2) \bar{A}_{4}\bar{A}_{5}(2) \bar{H}_{6}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{K}_{6}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{5}\bar{M}_{6}(2) \bar{M}_{3}\bar{M}_{4}(2) 4 FRAGILE (0,0) 40 SEBR / (0,3) \bar{\Gamma}_{9}(2) \bar{A}_{9}(2) \bar{H}_{4}\bar{H}_{5}(2) \bar{K}_{4}\bar{K}_{5}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 FRAGILE (0,0) 42 SEBR / (0,3) \bar{\Gamma}_{6}\bar{\Gamma}_{7}(2) \bar{A}_{6}\bar{A}_{7}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{3}\bar{M}_{4}(2) 2 SEBR (0,1) 44 TRIVIAL / (0,0) \bar{\Gamma}_{8}(2) \bar{A}_{9}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 SEBR (0,3) 46 SEBR / (0,3) \bar{\Gamma}_{9}(2) \bar{A}_{9}(2) \bar{H}_{4}\bar{H}_{5}(2) \bar{K}_{4}\bar{K}_{5}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 FRAGILE (0,0) 48 SEBR / (0,3) \bar{\Gamma}_{6}\bar{\Gamma}_{7}(2) \bar{\Gamma}_{9}(2) \bar{A}_{8}(2) \bar{A}_{6}\bar{A}_{7}(2) \bar{H}_{6}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) \bar{M}_{3}\bar{M}_{4}(2) 4 SEBR (1,0) 52 SEBR / (1,3) \bar{\Gamma}_{9}(2) \bar{A}_{9}(2) \bar{H}_{4}\bar{H}_{5}(2) \bar{K}_{4}\bar{K}_{5}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 FRAGILE (0,0) 54 SEBR / (1,3) \bar{\Gamma}_{9}(2) \bar{A}_{9}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 TRIVIAL (0,0) 56 SEBR / (1,3) \bar{\Gamma}_{6}\bar{\Gamma}_{7}(2) \bar{A}_{6}\bar{A}_{7}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{5}\bar{L}_{6}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 FRAGILE (0,0) 58 SEBR / (1,3) \bar{\Gamma}_{9}(2) \bar{A}_{9}(2) \bar{H}_{6}(2) \bar{K}_{6}(2) \bar{L}_{3}\bar{L}_{4}(2) \bar{M}_{5}\bar{M}_{6}(2) 2 SEBR (1,1) 60 TRIVIAL / (0,0) Table S2: Symmetry data and topological classification from Check Topological Mat for the PM state of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} . The output obtained from the Check Topological Mat tool on the Bilbao Crystallographic Server 5 ; 4 ; 3 ; 93 ; 94 for the symmetry characters \chi_{n\mathbf{k}}(O) (trace file, see Eq. ( 49 ) and the surrounding text) from our DFT calculation of the electronic structure of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in the PM state (SG 164 ( P\bar{3}m1 ), see Sec. III.1 ). The rows indicate groupings of connected bands above the core shell, listed in the order realized by independently filling the Bloch states at each {\bf k} . In order, the columns list the irreducible small corepresentations associated to the little group symmetry characters \chi_{n\mathbf{k}}(O) (Eq. ( 49 )) at each maximal momentum {\bf k} in Fig. S3 a, the total number of connected bands in the grouping including spin (dim), the classification of the bands in the language of Topological Quantum Chemistry (type, e.g. SEBR, see Ref. 94 ), the symmetry-based indicators (SIs) of stable band topology (Z_{2,3},Z_{4}) in SG 164 ( P\bar{3}m1 ) for the energetically isolated bands (top. ind., respectively the k_{z}=\pi weak \mathbb{Z}_{2} invariant and the strong \mathbb{Z}_{4} invariant), the cumulative band dimension realized by individually occupying each Bloch state up to an electronic filling \nu above the core shell (filling \nu ), and the cumulative topology and SIs (Z_{2,3},Z_{4}) of the symmetry-allowed gap at \nu (all type / top. ind.). The columns “filling \nu ” and “all type / top. ind.” are hence additive, and are obtained by respectively summing the data in the “dim” and “top. ind.” columns (including modulo operations for the cumulative SIs). Counting above the core shell, \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} has \nu_{F}=51 valence electrons (Eq. ( 50 )), and is hence a \mathcal{T} -enforced metal in the PM state 98 ; 102 (ESFD in the nomenclature of Refs. 93 ; 94 ). The first band gap allowed by symmetry above \nu_{F} occurs at \nu=\nu_{F}+1=52 (dashed horizontal line). The gap at \nu=52 , as well as all of the symmetry-allowed gaps from \nu=\nu_{F}-5=46 to \nu=\nu_{F}+7=58 , exhibit the stable topology of a 3D topological insulator (TI), as indicated by their nontrivial Fu-Kane \mathbb{Z}_{2} invariants Z_{4}\text{ mod }2=1 25 ; 24 ; 33 . We may hence associate the unoccupied (001) -surface Dirac fermion in our DFT calculations of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in Figs. S3 b, S5 a, and S6 to 3D TI gaps away from E_{F} , with the most likely candidates being the unoccupied topological band-insulating gaps at \nu_{F}<\nu\leq 58 (see the text following Eq. ( 50 )). In the PM state, \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} carries an intrinsic (valence) electronic filling above the core shell \nu_{F} of: \nu_{F}=51. (50) Owing to the requirement from \mathcal{T} symmetry that noninteracting, nonmagnetic insulators have even numbers of occupied bands 98 ; 102 , \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in the PM state is therefore, in the language of Ref. 94 , an enforced semimetal with a Fermi degeneracy (ESFD) at \nu=\nu_{F} , and not a band insulator. The first gap in the PM state of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} allowed by symmetry instead occurs at a filling of \nu=\nu_{F}+1=52 electrons, as indicated by the dashed horizontal line in Table S2 . Remarkably, not only does the gap at \nu=52 in \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} carry the cumulative topology of a 3D TI state ( Z_{4}=3 in “all type / top. ind.”), but in fact all of the symmetry-allowed band gaps realize 3D TI states over a range of \Delta\nu=12 electrons spanning from \nu=\nu_{F}-5=46 to \nu=\nu_{F}+7=58 . Hence, though surface chemical potentials may still shift topological surface Dirac cones away from the average energy of nontrivial band-insulating gaps (see the text following Eq. ( 14 )), the huge filling range \Delta\nu=12 of nontrivial 3D TI gaps surrounding \nu_{F} in Table S2 is much larger than the scale of chemical potential effects, implying that the unoccupied (001) -surface Dirac cone in our DFT calculations (Figs. S3 b, S5 a, and S6 ) can be associated to a band-insulating 3D TI phase with \nu\neq\nu_{F} . Even further restricting to fillings above the Fermi energy \nu>\nu_{F} , it is also unlikely that surface chemical potential effects can shift a Dirac cone by energies equivalent to changing \nu by 7 electrons, and we may hence more specifically reasonably associate the (001) -surface Dirac cone in Figs. S3 b, S5 a, and S6 to the smaller set of unoccupied 3D TI gaps with \nu_{F}<\nu\leq 58 in Table S2 . In this sense, \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} is reminiscent of the repeat-topological (RTopo) materials introduced in Ref. 94 , especially the example of Bi 2 Mg 3 , which respects the same symmetry group SG 164 ( P\bar{3}m1 ) as \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in the PM state. RTopo materials were specifically defined in Ref. 94 as materials that are stable topological band insulators at intrinsic filling \nu_{F} , but also host cumulative stable topological band-insulating states at the first band gap below \nu_{F} allowed by symmetry. In Bi 2 Mg 3 , the RTopo classification gives rise to two sets of (001) -surface topological Dirac cones, which at intrinsic filling respectively manifest as surface states in the projected gap just above E_{F} , and as surface resonance bands below E_{F} that coexist with the surface projections of the bulk valence manifold and are associated to the topological band gap at \nu<\nu_{F} 94 ; 15 ; 110 . If \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} in the PM phase were doped to \nu=\nu_{F}+1=52 (the dashed line in Table S2 ), it would also similarly be classified as RTopo. The unoccupied surface Dirac cone in \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} (Figs. S3 b, S5 a, and S6 ) and the associated topological band gaps at \nu>\nu_{F} in Table S2 can hence be viewed as a generalization of the concept of repeat-topology – originally introduced to predict topological boundary states accessible in equilibrium ARPES experiments at intrinsic filling – to topological surface conduction bands that are accessible by doping and pump-probe ARPES experiments like those in the present study. IV Estimating the position of the bulk conduction band with electron energy loss spectroscopy (EELS) Figure S8: High-energy electron energy loss spectroscopy: Momentum-integrated spectra obtained from electron energy loss spectroscopy (EELS), taken in the paramagnetic state of \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} at 293\text{\,}\mathrm{K} . The local minimum near E_{CBM}=E-E_{F}=$0.75\text{\,}\mathrm{e}\mathrm{V}$ that gives way to a broad continuum corresponds to the inter-band transition edge. The plasmon feature (Fig. 4 b of the main text) lies above the vertical range of this plot. Electron energy loss spectroscopy (EELS) measurements at low energies were presented in Fig. 4 b of the main text, and showed a strong peak at approximately 0.12\text{\,}\mathrm{e}\mathrm{V} , which we attribute to a plasmon. The EELS data at higher energies (Fig. S8 ) of the \text{EuCd}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}}\text{As}{\vphantom{\text{X}}}_{\smash[t]{\text{2}}} samples used in this study show a broad continuum of energy loss above E_{CBM}=E-E_{F}=$0.75\text{\,}\mathrm{e}\mathrm{V}$ , indicative of either a bulk or surface band minimum at that energy. In our first-principles electronic band structure calculations in Sec. III , we observe a bulk conduction band at approximately the same energy E_{CBM} (Fig. S3 a). The value of E_{CBM}=$0.75\text{\,}\mathrm{e}\mathrm{V}$ obtained from our EELS measurements is also in agreement with previous optical spectroscopy measurements on samples characterized as semimetallic 104 , which place a conduction band minimum between 0.7\text{\,}\mathrm{e}\mathrm{V} and 0.8\text{\,}\mathrm{e}\mathrm{V} relative to the maximum of the hole-like valence band. Optical spectroscopy and Tr-ARPES studies of semiconducting samples 75 similarly place a local conduction band minimum at 0.8\text{\,}\mathrm{e}\mathrm{V} . V Extracting and fitting transient electronic temperatures and dispersions from Tr-ARPES data Figure S9: Waterfall plots for extracting time-dependent chemical potential and temperature from ARPES EDCs: (a-c) Waterfall plots showing EDCs for the unoccupied surface states and corresponding fitted Fermi-Dirac distributions at time delays between 0\text{\,}\mathrm{p}\mathrm{s} and 3\text{\,}\mathrm{p}\mathrm{s} and energies measured relative to the bottom of the unoccupied surface state at E_{0}\approx E_{F}+$0.8\text{\,}\mathrm{e}\mathrm{V}$ . (d, e) Temperature ( T ) and chemical potential ( \mu ) parameters extracted from the data in (a-c) as functions of time. Error bars for T and \mu (the shaded regions in (a-c)) represent a 68% confidence interval for the fits. We observe that under low-fluence excitation (below 46\text{\,}\mathrm{\SIUnitSymbolMicro} ), the momentum-integrated intensity of ARPES spectra are well-approximated at all times by a linear density of states multiplied by a Fermi-Dirac distribution, convolved with a fixed Gaussian energy resolution: I(E,t)=A_{0}(t)(E-\alpha)\frac{1}{\mathrm{e}^{(E-\mu(t))/k_{B}T(t)}+1}*\text{Gauss}(E,\sigma=$10\text{\,}\mathrm{m}\mathrm{e}\mathrm{V}$). (51) In Eq. ( 51 ), the parameter \alpha is chosen to permit the presence of a nonzero density of states at E_{0} , which we attribute to a rigid background due to either (or both) scattering from impurities and final state effects. These artifacts are independent of pump fluence and time after photo-excitation, and we therefore hold \alpha constant at 0.05\text{\,}\mathrm{e}\mathrm{V} . In Fig. S9 , we show the results of fitting EDCs at different times to the functional form of the ARPES intensity in Eq. ( 51 ). At high excitation fluences, we no longer employ Eq. ( 51 ). As described in the main text, under stronger excitation, we observe that the electronic distribution sharpens 100\text{\,}\mathrm{m}\mathrm{e}\mathrm{V} above the bottom of the unoccupied surface state E_{0}\approx E_{F}+$0.8\text{\,}\mathrm{e}\mathrm{V}$ at increasing delay time after photo-excitation, which we attribute to the formation of excitons. The model of Eq. 51 is then no longer a reasonable physical description, since the excitons (specifically the electrons resulting from breaking excitons with the UV probe) would need to be ascribed their own statistical distribution. Lastly, we note that in Eq. ( 51 ), we have allowed the prefactor A_{0}(t) to carry an explicit time dependence, because A_{0}(t) scales with the total particle number in the unoccupied states, whose conservation is broken by significant radiative and impurity-mediated decay of electrons towards the equilibrium Fermi level. Figure S10: Waterfall plots for extracting dispersion from Lorentzian fits: (a) Time-integrated ARPES spectrum for t\in[$0\text{\,}\mathrm{p}\mathrm{s}$,$2\text{\,}\mathrm{p}\mathrm{s}$] for low fluence excitation ( T=14.5\,$\mathrm{K}$ ). The black line in (a) tracks the centers of Lorentzian fits of the momentum distribution curves (MDCs). (b) MDCs at several energy windows, with Lorentzian fits superimposed. (c) Example imaginary self energy \text{Im }\Sigma(E) extracted from the half width at half maximum (HWHM) of the Lorentzian fits in (b). In a one-band quasi-equilibrium system, the ARPES intensity can be written as 11 ; 18 : I(k,E)=A(k,E)|M(k,E)|^{2}f(E), (52) where f(E) is the Fermi-Dirac distribution, M(k,E) is a matrix element for the excitation of a bound electron to a plane-wave state, and A(k,E) is the single-particle spectral function, given by: A(k,E)=-\frac{1}{\pi}\frac{\Sigma^{\prime\prime}(k,E)}{\left[E-\epsilon(k)-\Sigma^{\prime}(k,E)\right]^{2}+\left[\Sigma^{\prime\prime}(k,E)\right]^{2}}. (53) In Eq. ( 53 ), \epsilon(k) is a single-particle energy eigenvalue, and \Sigma^{\prime}(k,E)=\text{Re }\Sigma and \Sigma^{\prime\prime}(k,E)=\text{Im }\Sigma are respectively the real and imaginary parts of the self energy, which are connected through the Kramers-Kronig relations. When \Sigma^{\prime\prime} is taken to be independent of k (that is, scattering mechanisms encoded by \Sigma^{\prime\prime} are assumed to vary slowly with momentum), cuts of A(k,E) taken at constant E have a Lorentzian profile in k centered at \epsilon(k)+\Sigma^{\prime}(E) with a half width at half maximum (HWHM) proportional to \Sigma^{\prime\prime} . We therefore track the dispersion and imaginary self energy of the unoccupied surface states using Lorentzian fits of the MDCs. Our ARPES spectra were prepared by integrating over a short window starting at t=t_{0} , to account for rapid intra-band scattering and to allow significant spectral weight at all energy levels. As stated in the main text, there is no observed band renormalization; hence, the results of this fitting process are independent of our choice of the integration duration. Representative examples of an ARPES spectrum and a waterfall plot are displayed in Fig. S10 , and show good agreement between individual cuts and the fits. VI Absence of time-dependent photovoltage shifts Figure S11: Time-dependence of photovoltage effects: Plot of the imaginary self energy \text{Im }\Sigma(E) at various delay times after 55\text{\,}\mathrm{\SIUnitSymbolMicro} excitation by 1.5\text{\,}\mathrm{e}\mathrm{V} photons. Rigid vertical offsets have been added for clarity. Across all delay times, we observe no significant shift of the local minimum in \text{Im }\Sigma(E) (green shaded region) relative to the bottom of the surface spectral feature at E_{0}\approx E_{F}+$0.8\text{\,}\mathrm{e}\mathrm{V}$ . Details about the fitting process used to extract \text{Im }\Sigma(E) from ARPES cuts are provided in Supplementary Information V . In wide-gap semiconductors, it is well established that photovoltage effects can lead to shifts in the effective Fermi energy 78 . If present, this could shift the bottom of the surface spectral feature at E_{0}\approx E_{F}+$0.8\text{\,}\mathrm{e}\mathrm{V}$ upwards to E_{X} , leading to an apparent long-lived population at the middle of the dispersion extracted at t\approx$0\text{\,}\mathrm{p}\mathrm{s}$ . Such a rigid shift could additionally lead to time-dependent renormalization, similar to that observed in this work at high fluences in Fig. 3 of the main text. To investigate photovoltage effects in our measurements, we track in Fig. S11 the position of the local minimum in the imaginary part of the self energy \text{Im }\Sigma(E) of the unoccupied states, which necessarily appears at the bottom of the surface states, E_{0} , as a function of delay time. Details about the fitting process used to extract \text{Im }\Sigma(E) from ARPES cuts are provided in Supplementary Information V . This allows to us to address the possibility that the persistent population interpreted in this work as an exciton is instead simply an accumulation of electrons at the bottom of an unoccupied band. In Fig. S11 , we importantly observe no significant time-dependent shift in the local minimum of \text{Im }\Sigma(E) , allowing us to exclude the trivial (non-excitonic) interpretation of our data. VII Rate equation modeling of plasmon-mediated exciton formation Figure S12: Theoretical modeling of plasmon-mediated exciton formation: (a) Calculated and (b) observed difference between time traces for high and low fluence photo-excitation. The maximum population at each energy level was normalized to unity for each photo-excitation fluence. The calculation in (a) was obtained by using the parameters in Table S3 as input for I_{h}(E,t)-I_{\ell}(E,t) , where I_{\ell}(E,t) and I_{h}(E,t) are respectively defined in Eqs. ( 55 ) and ( 57 ). The experimental time trace data in (b) was acquired with fluences of 15\text{\,}\mathrm{a} nd 55\text{\,}\mathrm{\SIUnitSymbolMicro} . Parameter Value Screened electron-plasmon coupling strengths at low fluence ( \alpha_{\ell} ) 1.7\text{\,}\mathrm{p}\mathrm{s}^{-1}\mathrm{e}\mathrm{V}^{-1} Screened electron-plasmon coupling strength at high fluence ( \alpha_{h} ) 0.85\text{\,}\mathrm{p}\mathrm{s}^{-1}\mathrm{e}\mathrm{V}^{-1} Energy at which electronic lifetimes diverge ( E_{CBM} ) -0.1\text{\,}\mathrm{e}\mathrm{V} Impurity scattering time ( \tau_{i} ) 25\text{\,}\mathrm{p}\mathrm{s} Exciton electron width ( \varsigma_{X} ) 0.15\text{\,}\mathrm{e}\mathrm{V} Exciton hole width ( \varsigma_{H} ) 0.075\text{\,}\mathrm{e}\mathrm{V} Exciton electron position ( E_{X} ) 0.1\text{\,}\mathrm{e}\mathrm{V} Exciton hole position ( E_{H} ) -0.06\text{\,}\mathrm{e}\mathrm{V} Excitonic formation time ( \tau_{Xf} ) 2\text{\,}\mathrm{p}\mathrm{s} Excitonic lifetime ( \tau_{Xl} ) 6\text{\,}\mathrm{p}\mathrm{s} Excitonic spectral weight ( A_{X} ) 0.1 Table S3: Physical meaning and selected numerical values for the input parameters for our model of the electronic population dynamics in the unoccupied band structure (Fig. S12 (a) and Eq. ( 57 )). E_{X} , E_{H} , and E_{CBM} are measured relative to the bottom of the surface state, E_{0}\approx E_{F}+$0.8\text{\,}\mathrm{e}\mathrm{V}$ . Figure S13: Decay rate as a function of energy ( E-E_{0} ) for low fluence excitation ( 15\text{\,}\mathrm{\SIUnitSymbolMicro} ), obtained from single exponential fits of Tr-ARPES data, which we use to estimate \alpha_{\ell} in Table S3 and Eq. ( 55 ). We additionally set E_{CBM}=$-0.1\text{\,}\mathrm{e}\mathrm{V}$ , approximately where the decay rate reaches its minimum value. The shaded region represents a 68% confidence interval. To evaluate whether the population dynamics presented in Fig. 5 of the main text are consistent with the proposed mechanism of plasmon-mediated exciton formation, we numerically model the population evolution using a rate equation 111 . In Table S3 , we list the physical meaning and selected values of the minimal parameters needed for this model. The onset of the intensity at the instant of photo-excitation is approximately probe-limited ( 200\text{\,}\mathrm{f}\mathrm{s} , see the Methods section of the main text). We hence treat the onset of the intensity at photo-excitation as a step function, given that the dynamics described above are significantly slower, and set observed intensities for both low and high excitation fluences to zero: I_{\ell}(E,t<0)=0 and I_{h}(E,t<0)=0 . Below, we consider only t\geq 0 . We describe the electronic population I^{e}_{F}(E,t) at a fluence F (with F=\ell,h for low and high fluences respectively), energy E , and time t using \frac{\mathrm{d}I^{e}_{F}(E,t)}{\mathrm{d}t}=-\bar{\alpha}_{F}(E)\ I^{e}_{F}(E,t), (54) where \bar{\alpha}_{F}(E) is the effective energy-dependent scattering rate. For plasmon-mediated decay, we approximate \bar{\alpha}_{F}(E)=\alpha_{F}(E-E_{CBM})+\tau_{i}^{-1} . The parameters \alpha_{\ell} and \alpha_{h} in Table S3 respectively represent the screened electron-plasmon coupling strengths for the low fluence and high fluence cases. We treat the two cases as distinct due to screening effects and differences in the initial temperatures of the photo-doped population. We set \alpha_{\ell} to the slope of the graph of the decay rate over energy near E_{0} for 15\text{\,}\mathrm{\SIUnitSymbolMicro} photo-excitation (Fig. S13 ). \tau_{i} is an energy-independent impurity-mediated scattering time. E_{CBM} is measured relative to the bottom of the surface state, E_{0}\approx E_{F}+$0.8\text{\,}\mathrm{e}\mathrm{V}$ , and lies at the bottom of the broad energy range over which spectral weight from the bulk conduction band is observed (indicated by the yellow triangle in Fig. 2 c of the main text). It coincides with the energy of a bulk conduction band in our PM-state DFT calculation that is nondispersing at \Gamma to leading order (Fig. S3 a). In Fig. S13 , we confirm that the decay rate indeed reaches its minimum value at approximately E_{CBM} . To eliminate effects associated with the density of states (which decreases observed spectral weight at lower energies) and the choice of excitation photon energy (which places an upper bound on the range of electronic energies that can be occupied), we normalize the maximum intensity at each energy level to unity. For low fluence excitation, when the re-absorption of plasmons by cooled electrons is negligible, the observed intensity I_{\ell}(E,t) originates entirely from electrons, with no excitonic contribution, giving I_{\ell}(E,t)=I_{\ell}^{e}(E,t) . Thus, the energy-normalized spectral intensity can be written using Eq. ( 54 ) as: I_{\ell}(E,t\geq 0)=\exp\left[-t(\alpha_{\ell}(E-E_{CBM})+\tau^{-1}_{i})\right]. (55) For high fluence excitation, we add further terms to Eq. ( 55 ) to account for the arrival (along the timescale of exciton formation \tau_{Xf} ) and decay (along the timescale of the exciton lifetime \tau_{Xl} ) of electrons at E_{X} 55 , as well as a corresponding term of opposite sign at E_{H} to account for the conservation of particle number (that is, hole formation). By this approach, we express the observed intensity as the sum of electronic and excitonic contributions: I_{h}(E,t)=I_{h}^{e}(E,t)+I_{h}^{X}(E,t), (56) where I_{h}^{X}(E,t) is positive (negative) at energies where the exciton’s electron (hole) manifests in spectra. We assume the exciton electron and hole to have Gaussian spectral signatures with the respective standard deviations \sigma_{X} and \sigma_{H} , which are related to the full width half max (FWHM) \varsigma_{E,H} of the corresponding features in the EDCs by \sigma_{E,H}=\varsigma_{E,H}/(2\sqrt{2\log{2}}) . E_{X} and E_{H} are measured relative to the bottom of the surface state, E_{0}\approx E_{F}+$0.8\text{\,}\mathrm{e}\mathrm{V}$ , and are identified from normalized differences in time traces (see Fig. 5 b of the main text, displayed without annotation in Fig. S12 b). E_{H} is allowed to lie slightly above E_{CBM} , to match experimental data and avoid numerical divergences. Like E_{CBM} , E_{H} lies within the broad spectral weight associated with the bulk conduction band (Fig. 2 c of the main text). This results in a refinement of Eq. ( 55 ) given by: \displaystyle I_{h}(E,t\geq 0)= \displaystyle\exp\left[-t(\alpha_{h}(E-E_{CBM})+\tau^{-1}_{i})\right] \displaystyle+\frac{A_{X}}{\sigma_{X}\sqrt{2\pi}}\exp\left(-\frac{(E-E_{X})^{2}}{2\sigma_{X}^{2}}\right) \displaystyle\quad\qquad\qquad\left(1-\mathrm{e}^{-t/\tau_{Xf}}\right)\mathrm{e}^{-t(\tau^{-1}_{Xl}+\tau^{-1}_{i})} \displaystyle-\frac{A_{X}}{\sigma_{H}\sqrt{2\pi}}\exp\left(-\frac{(E-E_{H})^{2}}{2\sigma_{H}^{2}}\right) \displaystyle\quad\qquad\qquad\left(1-\mathrm{e}^{-t/\tau_{Xl}}\right)\mathrm{e}^{-t(\tau^{-1}_{Xf}+\tau^{-1}_{i})}. (57) The positive (negative) Gaussian term adds (removes) electronic spectral weight at E_{X} ( E_{H} ) due to breaking of the exciton during photoemission contributing an electron (hole) to the measured intensity. Finally, in Fig. S12 a, we plot I_{h}(E,t)-I_{\ell}(E,t) using the parameter values in Table S3 . The theoretical calculation in Fig. S12 a shows good qualitative agreement with the corresponding experimental results in Fig. S12 b.