betaFeedback

WorksPhysics and Astronomy

Correlation-driven electronic nematicity in the Dirac semimetal BaNiS2

Abstract

In BaNiS2 a Dirac nodal-line band structure exists within a two-dimensional Ni square lattice system, in which significant electronic correlation effects are anticipated. Using scanning tunneling microscopy, we discover signs of correlated-electron behavior, namely electronic nematicity appearing as a pair of C2-symmetry striped patterns in the local density-of-states at ~60 meV above the Fermi energy. In observations of quasiparticle interference, as well as identifying scattering between Dirac cones, we find that the striped patterns in real space stem from a lifting of degeneracy among electron pockets at the Brillouin zone boundary. We infer a momentum-dependent energy shift with d-form factor, which we model numerically within a density wave equation framework that considers spin-fluctuation-driven nematicity. This suggests an unusual mechanism driving the nematic instability, stemming from only a small perturbation to the Fermi surface, in a system with very low density of states at the Fermi energy. The Dirac points lie at nodes of the d-form factor, and are almost unaffected by it. These results highlight BaNiS2 as a unique material in which Dirac electrons and symmetry-breaking electronic correlations coexist.

Materials in which Dirac fermions and significant electronic correlations coexist represent an intersection of two active frontiers of condensed matter research and might host a range of useful or interesting new phenomena [1, 2, 3]. Correlations between electrons can result in instability towards a number of spontaneous symmetry-breaking configurations including magnetic order, charge density waves and unconventional superconductivity. They can also lead to nematic order, which plays an important role in the cuprate and Fe-based superconductors [4, 5, 6, 7]. Basic questions include whether topological band structures can survive in the presence of significant electronic correlations, and whether, or in what way, the topological classification of a system can be altered by correlation-driven symmetry-breaking effects such as nematicity, in which translational symmetries are preserved while rotational symmetry is broken [8–10]. However, the impact of such symmetry breaking effects on Dirac fermions remains to be fully elucidated.

Correlations can be introduced through the inclusion of transition metals in the material composition, while otherwise ensuring the requisite crystal symmetries for an interesting topological phase [11]. Such a situation is realized in the BaCo1−xNixS2\mathrm{BaCo}_{1-x}\mathrm{Ni}_{x}\mathrm{S}_{2} series of compounds. It has recently been shown that the end member of the series, BaNiS2\mathrm{BaNiS}_{2}, hosts four simple Dirac nodal-lines that are well-isolated from other bands and traverse the Brillouin zone (BZ) with fairly weak dispersion along the layer-perpendicular axis [12, 13]. These properties make it one of the systems closest to the ideal of a two-dimensional (2-D) Dirac fermion system. The bands emanating from these nodal-lines have been shown to be susceptible to renormalization under ultrafast light pulses [14], and to a controlled shifting of the Dirac node energy and wavevector upon Co substitution [13].

The dd orbitals of the transition metal square net in BaCo1−xNixS2\mathrm{BaCo}_{1-x}\mathrm{Ni}_{x}\mathrm{S}_{2} introduce correlated-electron behavior, most obviously magnetism, but also further interesting properties. While BaNiS2\mathrm{BaNiS}_{2} is a paramagnetic metal, BaCoS2\mathrm{BaCoS}_{2} is an antiferromagnetic insulator [15]. In the BaCo1−xNixS2\mathrm{BaCo}_{1-x}\mathrm{Ni}_{x}\mathrm{S}_{2} solid solution, at x=0.22x = 0.22 a metal-insulator transition occurs, for which both purely electronic [16–18] and structural mechanisms [19] have been proposed. Although superconductivity is absent, the temperature-composition phase diagram otherwise resembles that of the layered cuprates, and the above properties make BaCo1−xNixS2\mathrm{BaCo}_{1-x}\mathrm{Ni}_{x}\mathrm{S}_{2} an attractive system for the exploration of possible composition-tuned Mott-Hubbard physics on a square lattice, and the search for correlation-driven symmetry breaking phenomena. Even the metallic state of pure BaNiS2\mathrm{BaNiS}_{2}, furthest from the metal-insulator transition, has been shown to be somewhat anomalous, with TT-linear resistivity below T=2T = 2 K possibly hinting at the presence in the phase diagram of a quantum critical point [20].

In this work we use scanning tunneling microscopy (STM) to explore the electronic structure at the cleaved surface of BaNiS2\mathrm{BaNiS}_{2}, observing both nearly ideal topological Dirac fermionic behavior and clear symmetry breaking in the form of bond-order nematicity within the surface Ni square net. In a narrow energy range approximately 60 meV above EFE_{\mathrm{F}}, we observe a lowering of sym-

Properties of the BaNiS2_2(001) surface

BaNiS2_2 is a layered quasi-2-D material with a structure belonging to the nonsymmorphic space group P4/nmmP4/nmm, with lattice parameters reported as a=b=4.43a=b=4.43 Å and c=8.89c=8.89 Å [21]. Its structure is composed of NiS5_5 pyramids whose orientation alternates about the basal plane, and whose basal edges are shared with their neighbors, as depicted in Figure 1(a). Sandwiching these layers are Ba atoms located between, and almost in the plane of, the apical S of the NiS5_5 pyramids. As indicated in Figure 1(b), cleavage can occur between adjacent BaS layers revealing (001) facets. A top-down view of the surface lattice is shown in Figure 1(c). The cutaway on the right-hand side of the image shows the NiA_\mathrm{A} and NiB_\mathrm{B} lattices.

Image not extracted

Figure 1. Overview of the crystal and low-energy electronic structures of BaNiS2_2. (a) A depiction of the primitive cell [21] and (b) the quasi-two-dimensional crystal structure, showing the location of the cleavage plane. (c) A top-down view of the surface lattice. The cutaway view shows the NiA_{\mathrm{A}} and NiB_{\mathrm{B}} square nets. The primitive cell is shown with a black square. Unit-cell and lattice images were created using VESTA [22]. (d) Calculated surface band structure shown along high-symmetry lines. The orbital composition for four of the Ni dd orbitals is indicated according to the color key, with the radial axis (lightness) signifying the spectral weight summing over all orbitals. The dxyd_{xy} orbital is not shown as its spectral weight is negligible in this energy range. (e) A constant-energy cross-section through A(k,E)A(\mathbf{k}, E) at E=−80E=-80 meV, and (f) a cross-section at E=50E=50 meV. In these calculations the P4/nmmP4/nmm space group is assumed, resulting in C4C_4 symmetry of the constant-energy cross-sections, but this assumption will be overturned by the observations of nematicity shown below. In anticipation of this, we draw a distinction between the points X‾a\overline{X}_{a} and X‾b\overline{X}_{b} at the zone boundary. (g) A typical constant-current STM topograph (setpoints V=0.1V=0.1 V, I=100I=100 pA). The observed atomic corrugations likely correspond to the uppermost NiA_{\mathrm{A}} square net. (h) A dIdV(E)\frac{dI}{dV}(E) conductance curve after averaging over the field of view marked as a white dashed square in (g). (i) A dIdV(E)\frac{dI}{dV}(E) curve averaged over the same area, focusing on the region nearer to the Fermi energy EFE_{\mathrm{F}}.

Figure 1(d) shows the (001) surface spectral function A(k,E)A(\mathbf{k}, E) along chosen high-symmetry axes of the BZ. The orbital composition of the bands are shown using hue as indicated in the inset. Approximately halfway along the ΓM‾\overline{\Gamma M} line, there exists a Dirac cone. This feature represents the surface projection of a 3-D cone emanating from one of the four Dirac nodal-lines running between the kz=πck_z = \frac{\pi}{c} and −πc-\frac{\pi}{c} planes [13]. The electron-like pocket surrounding the Γ‾\overline{\Gamma}-point results from a feature previously described as a pinched electron-like tube connecting the kz=±πck_z = \pm\frac{\pi}{c} faces of the BZ [23]. A third noteworthy feature is the electron-like band centered around the X‾\overline{X} point. This results from the collapse into two dimensions of a band surrounding each of the R points at the edges of the bulk BZ, and the previously described Rashba splitting within these bands is also observed here [24, 25].

In Figure 1(e) we show a constant-energy cross-section taken from A(k,E)A(\mathbf{k}, E) at E=−80E = -80 meV. Here the electron-like pocket around the bulk Γ‾\overline{\Gamma} point, and hole-pockets stemming from the Dirac nodal-lines, are seen to collapse into a circle and a set of ellipses, respectively, in the surface BZ. The Dirac cones are composed of a hybridization of the Ni dx2−y2d_{x^2-y^2} and dz2d_{z^2} orbitals. Below the Dirac points, these orbitals contribute the inner (Γ‾\overline{\Gamma}-facing) and outer (M‾\overline{M}-facing) arcs, respectively [13], and this is reversed above the Dirac points as seen at the left-hand-side of Figure 1(d). Figure 1(f) shows a constant-energy cross-section at E=50E = 50 meV, an energy which coincides closely with both the Dirac nodes and the bottom of the electron bands around the X‾\overline{X} point. Like the Dirac cones, these bands are composed of a hybridization of dx2−y2d_{x^2-y^2} and dz2d_{z^2} orbitals. We point out that these numerical calculations do not accurately reflect the energies of these features, which are to be properly established through observations presented below (and to some extent in the Supplemental Information).

A typical constant-current STM topography image acquired at a cleaved surface is shown in Figure 1(g). The observed atomic corrugations almost certainly represent the Ni square net, and specifically NiA_\mathrm{A}. (The periodicity of the corrugations corresponds to one observed atom per surface unit cell – see Supplemental Information.)

Figures 1(h) and 1(i) show typical tunneling conductance curves, denoted by dIdV(E)\frac{dI}{dV}(E), where by convention E=eVE = eV (ee being the electron charge and VV the sample bias). These are averaged over the 6×66 \times6 nm2^2 field of view marked as a white dashed square in Figure 1(g). The curves indicate a density of states close to an ideal semimetal, i.e. becoming small near the Fermi energy. In Figure 1(i), the ‘V’-shaped minimum at E≈27E \approx27 meV can be taken to indicate the approximate energy of the Dirac crossings. The shallow and roughly linear onset toward lower energy likely corresponds to Dirac cones emerging from these crossings. Additional spectroscopy measurements acquired under magnetic fields are shown in the Supplementary Information. These indicate the emergence of a field-independent Landau level, indicating a Berry’s phase of π\pi associated with each Dirac point, as well as giving a more precise estimate of the Dirac points’ energy. We note that EFE_\mathrm{F} is located near the Dirac point energy, where the density-of-states is minimal. This situation would generally be unfavorable for an electronic instability. However, as we will show below, BaNiS2_2 exhibits a likely correlation-driven electronic nematicity.

Observation of Nematic Order

Figure 2 shows high-resolution spectroscopic imaging that reveals an unexpected symmetry breaking within the local density-of-states. Figure 2(a) shows STM topography over the same 6×66 \times6 nm2^2 field of view as marked by a white dashed square in Figure 1(g). A gray dot in the upper left-hand corner marks one of the NiA_\mathrm{A} sites for a point of reference. Differential conductance dIdV(r,E)\frac{dI}{dV}(\mathbf{r}, E) was acquired by measuring a dIdV(E)\frac{dI}{dV}(E) curve at each pixel. Figures 2(b) and 2(c) show images extracted from the normalized differential conductance defined as L(r,E)≡dIdV(r,E)/I(r,E)VL(\mathbf{r}, E) \equiv\frac{dI}{dV}(\mathbf{r}, E)/\frac{I(\mathbf{r}, E)}{V}, at E=55E = 55 meV and 68 meV, respectively. The normalization mitigates the so-called ‘set-point effect’ [26], an unwanted influence of variations in tip-sample spacing on dIdV(r)\frac{dI}{dV}(\mathbf{r}) images. The gray dots in Figures 2(b) and 2(c) are carried over from the one in Figure 2(a). The Fourier transform images of the topographic and L(r)L(\mathbf{r}) images are shown in Figs. 2(d–f). In Figure 2(d) the peaks corresponding to the reciprocal lattice vectors Ga\mathbf{G}_a and Gb\mathbf{G}_b are seen, and show the expected C4C_4 symmetry. In Figs. 2(b) and 2(c), we see the two C2C_2 symmetry stripe patterns, each commensurate with, and in phase with, one of the surface lattice corrugations [i.e. for each pattern qa,b=Ga,b\mathbf{q}_{a,b} = \mathbf{G}_{a,b} as seen in Figs 2(e) and (f)].

Image not extracted

Figure 2. High-resolution spectroscopic imaging. (a) Constant-current topography in the 6×6 nm26 \times6\ \mathrm{nm}^{2} field of view marked by the dashed square in Figure 1(g) (setpoints V=0.1 VV = 0.1\ \mathrm{V}, I=500 pAI = 500\ \mathrm{pA}). (b,c) LL images at selected energies (Vmod=1 mVV_{\mathrm{mod}} = 1\ \mathrm{mV}). The grey dots in (a–c) mark a single NiA_{\mathrm{A}} site, and show that the lines of high intensity in the LL images coincide exactly with the NiA_{\mathrm{A}} lattice. (d–f) Fourier transformations of the images in (a–c). (g) A depiction of the NiA_{\mathrm{A}} square net and its two bond-centered sublattices, marked by arrays of blue and red dots. (h) Sampling of dIdV(E)\frac{dI}{dV}(E) data in the same field-of-view as in (a) (setpoints V=0.15 VV = 0.15\ \mathrm{V}, I=300 pAI = 300\ \mathrm{pA}, and Vmod=2.5 mVV_{\mathrm{mod}} = 2.5\ \mathrm{mV}), on a 8×88 \times8 grid on each bond-centered sublattice. (i) dIdV(E)\frac{dI}{dV}(E) curves after averaging over each of the sets of sampling points. Each arrow marks a subtle kink in the respective curve. (j) Derivatives d2IdV2(E)\frac{d^{2}I}{dV^{2}}(E) enhancing the small variations of the curves shown in (i). The splitting between the peaks is ≈12 meV\approx12\ \mathrm{meV}.

The peaks in intensity of each striped pattern correspond directly with the rows or columns of the NiA_\mathrm{A} lattice, and each stripe lies along a NiA_\mathrm{A}–NiA_\mathrm{A} bond chain (or more strictly, a NiA_\mathrm{A}–Si–NiA_\mathrm{A} bond chain). Figure 2(g) shows a cartoon of this situation, in which the blue (red) stripes indicate the pattern at the lower (upper) energy. As well as the in-plane projections of the NiA_\mathrm{A} and NiB_\mathrm{B} sites, the sites belonging to the two inequivalent bond-centered lattices are also shown, with red and blue dots. As shown in Figure 2(h) we perform a sampling of the raw conductance data in the same field of view as Figs. 2(a–c), on a 8×88 \times8 grid on each of the sublattices. Figure 2(i) shows the resulting dIdV(E)\frac{dI}{dV}(E) curves (averaged over the 64 sites of each grid). In each dIdV(E)\frac{dI}{dV}(E) curve, a subtle shoulder in the steep onset of states above the conductance minimum is marked with a black arrow. A derivative with respect to energy, shown in Figure 2(j) helps to clarify the difference between the two curves. A peak in the second-derivative of an I−VI - V curve is generally known to correspond to the availability of an additional tunneling channel, such as a band edge that becomes available upon reaching the necessary bias. The energy difference between the peaks in the two curves is ∼12\sim12 meV.

A question that reasonably follows from the above observation is whether a domain structure of the nematic order exists. Throughout this work, examining patterns of nematic order in two samples and searching over fields-of-view up to ∼500 nm\sim500\ \mathrm{nm}, no domain boundaries were observed, and the order was seen to be preserved across a step-terrace morphology including three atomic terraces (see Supplemental Information).

Manifestation of nematicity in QPI

The mechanism underlying the observed nematic order might be elucidated by examining the momentum-space electronic structure at the BaNiS2_2 surface, and indirect insights into this can be obtained through interpretations of QPI phenomena. Such measurements can complement previous angle-resolved photo-emission spectroscopy reports [13, 14, 24], but can additionally probe unoccupied bands.

Figure 3(a) shows a selected image from L(r,E)L(\mathbf{r}, E) data acquired in a larger field-of-view, at E=−80 meVE = -80\ \mathrm{meV}, in which modulations resulting from interference caused by scattering of quasiparticles from point defects are seen. Figure 3(b) shows the Fourier transform F[L(r,E=−80 meV)]\mathcal{F}[L(\mathbf{r}, E = -80\ \mathrm{meV})]. We hereafter refer to such Fourier transformed conductance data loosely as Lq(q,E)L_{q}(\mathbf{q}, E). The reciprocal lattice vectors are again denoted by Ga\mathbf{G}_{a} and Gb\mathbf{G}_{b}. At this energy we identify three prominent scattering vectors, labelled q1,2,3\mathbf{q}_{1,2,3}. Figure 3(c) shows linecuts through Lq(q,E)L_{q}(\mathbf{q}, E), running from the origin to Ga\mathbf{G}_{a}, and along a line of the same length running halfway between Ga\mathbf{G}_{a} and Gb\mathbf{G}_{b}. The most interesting features are seen in the linecut running from the origin to Ga\mathbf{G}_{a}. The branches q2\mathbf{q}_{2} and q3\mathbf{q}_{3} form an inverted ‘v’ below EFE_{\mathrm{F}}, and it can be inferred that they represent the shortest and longest possible scattering vectors, respectively, connecting two Dirac cones in adjacent quadrants of the BZ. At positive energy, a branch with apparently very weak dispersion, q4\mathbf{q}_4, extends to meet Ga\mathbf{G}_a. This branch will be discussed in greater detail below.

Image not extracted

Figure 3. Overview of QPI observations. (a) A normalized conductance image L(r)L(r) acquired at E=−80E=-80 meV in a 80 × 80 nm2^2 field of view, showing QPI modulations around point defects (448 × 448 pixels, setpoints V=0.25V=0.25 V, I=1I=1 nA, and Vmod=10V_{\mathrm{mod}}=10 mV). (b) A Fourier transform of the LL map, labeling the reciprocal lattice vectors Ga\mathbf{G}_a and Gb\mathbf{G}_b. The wavevectors q1,2,3\mathbf{q}_{1,2,3} of QPI modulations discussed in the text, are marked purple, red and orange arrows. The point q45∘\mathbf{q}_{45^\circ} marks the endpoint of a vector formed by rotating Ga\mathbf{G}_a by 45°. (c) Linecuts through the Lq(q,E)L_q(q,E) data, from the origin to Ga\mathbf{G}_a and also to q45∘\mathbf{q}_{45^\circ}. The intensity is averaged over the line width of 10 pixels. Here we additionally highlight the wavevector q4\mathbf{q}_4, which will be discussed below. The dashed lines for q1\mathbf{q}_1 and q4\mathbf{q}_4 are guides to the eye. The solid lines for q2\mathbf{q}_2 and q3\mathbf{q}_3 are obtained by a fitting procedure described in detail in the Supplementary Information. (d) A schematic of the surface BZ and calculated spectral function at E=−80E=-80 meV, showing the likely scattering transitions giving rise to q1,2,3\mathbf{q}_{1,2,3}. (e) Calculated QPI pattern Lqcalc.(q)L_q^{\mathrm{calc.}}(q) at E=−80E=-80 meV, based on the spectral function and TT-matrix formalism (see Methods). Each white ‘×’ symbol marks the position of a reciprocal lattice point. (f) Profiles through Lqcalc.(q,E)L_q^{\mathrm{calc.}}(q,E) highlighting the predicted signals q1,2,3,4calc.\mathbf{q}_{1,2,3,4}^{\mathrm{calc.}}. The solid red lines mark the ridges in Lqcalc.(q,E)L_q^{\mathrm{calc.}}(q,E), corresponding to q2\mathbf{q}_2 and q3\mathbf{q}_3, for comparison with the solid lines in (c).

Figure 3(d) shows the likely origin of each of the scattering signals found at E=−80E=-80 meV. We attribute q1\mathbf{q}_1 to scattering between Dirac cones on opposite sides of the BZ. We attribute q2\mathbf{q}_2 and q3\mathbf{q}_3 to scattering between cones in adjacent quadrants of the BZ, and specifically between the least- and most-distance ends of the respective ellipsoids as illustrated in Figure 3(d). Figure 3(e) and 3(f), show the corresponding calculated QPI pattern and linecuts. All of the labeled QPI branches are qualitatively reproduced.

Comparing the data shown in Figure 3(c) and (Figure 3f), we find that the observed dispersions for q2\mathbf{q}_2 and q3\mathbf{q}_3 are only a factor of 0.58 as high as the calculated ones. With reference to recent related work, we speculate that the reason could be a failure to fully account for electronic correlations that might renormalize the band dispersion at low energies [27]. Therefore the inferred ‘squeezing’ of the band structure near EFE_F may reflect significant electronic correlation effects in the real material.

A numerical fitting to the q2\mathbf{q}_2 and q3\mathbf{q}_3 branches, shown as red and orange solid lines in Figure 3(c), yields apparent velocities vq2=8.15×104v_{\mathbf{q}_2}=8.15\times10^4 m s−1^{-1} and vq3=−6.91×104v_{\mathbf{q}_3}=-6.91\times10^4 m s−1^{-1}, respectively. However, these values are related only indirectly to any actual band velocities, since a Fourier transform of QPI modulations visualized the momentum-transfer (qq) space rather than the momentum (kk) space. By making a comparison between the observed and calculated features in qq-space, and also by making a kk-space-to-qq-space comparison of the calculation results, we can eventually estimate some key properties of the Dirac cones. (See the Supplemental Information for full arguments.) In this way, the band velocities of the inner (T-facing) and outer (M-facing) sides of each Dirac cone are estimated to be vk,inner=3.09×105v_{k,\mathrm{inner}}=3.09\times10^5 m s−1^{-1} and vk,outer=−2.98×105v_{k,\mathrm{outer}}=-2.98\times10^5 m s−1^{-1}, respectively. This is reasonably consistent with a rough estimate drawn from recently reported angle-resolved photoemission spectroscopy results, of vk,inner∼vk,outer≈2×105v_{k,\mathrm{inner}}\sim v_{k,\mathrm{outer}}\approx2\times10^5 m s−1^{-1} [13].

Having identified the qq-space signatures of the Dirac cones, we turn our attention to the manifestation in QPI phenomena of the nematic behavior described above. Examining the Lq(q,E)L_q(\mathbf{q}, E) data further, we find that in a small energy range a few tens of meV above EFE_F there appears a lowering of symmetry similar to that shown in Figure 2. Two Lq(q)L_q(\mathbf{q}) images selected from the same data as shown in Figure 3, at E=60meVE = 60\mathrm{meV} and E=80meVE = 80\mathrm{meV}, are shown in Figure 4(a) and Figure 4(b) respectively. Each image shows an asymmetry in intensity between the pairs of reciprocal lattice peaks Ga\mathbf{G}_a and Gb\mathbf{G}_b [recall Figure 2(e) and Figure 2(f)].

Image not extracted

Figure 4. Symmetry breaking in QPI patterns. (a) The image Lq(q,E=60meV)L_q(q, E = 60\mathrm{meV}), from the same measurement described for Figure 3 above. The pairs of reciprocal lattice peaks Ga\mathbf{G}_a and Gb\mathbf{G}_b show a large difference in intensity. (b) The corresponding image at E=80meVE = 80\mathrm{meV}. Here the alternate set of reciprocal lattice peaks become more intense. (c) Comparison of fitting to Lq(q,E)L_q(q, E) along the Ga\mathbf{G}_a and Gb\mathbf{G}_b vectors [i.e., along the red and blue dashed lines marked in (a)]. In the region in which qq approaches ∣Ga,b∣\lvert\mathbf{G}_{a,b}\rvert the branches of points diverge, towards a splitting of ∼12meV\sim12\mathrm{meV} at the endpoint. (d) A schematic of the surface BZ and calculated spectral function at E=50meVE = 50\mathrm{meV}, with the yellow arrow illustrating the scattering vector q4\mathbf{q}_4. (e) The corresponding calculated QPI pattern Lqcalc.(q)L_q^{\mathrm{calc.}}(\mathbf{q}) at E=50meVE = 50\mathrm{meV}. See also Figure 3(f).

Figure 4(c) shows the result of fitting a sum of multiple Lorentzian functions to the Lq(E)∣qL_q(E)|_q curve at each qq-point along the lines from q=0q = 0 to each of the reciprocal points Ga\mathbf{G}_a and Gb\mathbf{G}_b. [This is the precursor step to obtaining the linear fits shown in Figure 3(c), and the full details are described in the Supplemental Information]. The results from each line profile are superimposed along the same axis for comparison. We see that nearly everywhere the peak energies show only small differences, but that a noticeable difference appears as we approach the endpoints q=∣Ga,b∣q = \lvert\mathbf{G}_{a,b}\rvert. We therefore distinguish between the branches along the two axes by relabeling them as q4,aq_{4,a} and q4,bq_{4,b}. As qq approaches ∣Ga,b∣\lvert\mathbf{G}_{a,b}\rvert, the energy splitting between the q4,aq_{4,a} and q4,bq_{4,b} branches approaches ∼12meV\sim12\mathrm{meV}. This is consistent with the splitting observed in intra-unit-cell sampling of dIdV(r,E)\frac{dI}{dV}(\mathbf{r}, E) data shown in Figure 2. The ∼10meV\sim10\mathrm{meV} difference in the absolute energies of the two branches as compared to those of the two peaks in Figure 2(j) is likely due to the fact that the measurements were acquired on two different sample surfaces.

Scattering vectors equal to Ga\mathbf{G}_a (Gb\mathbf{G}_b) in qq-space serve to connect X‾a\overline{X}_a (X‾b\overline{X}_b) and its counterpart at the opposite edge of the BZ. This suggests that the q4q_4 branch captured in Figure 3 and Figure 4 results from scattering between the electron pockets around the X‾a\overline{X}_a.Figure 4(d) and Figure 4(e) depict this scenario with reference to the calculated spectral function, and show the appearance of q4q_4 in the calculated QPI pattern.

Origin of nematicity via the density wave equation

The results presented above describe symmetry-breaking both in the intra-unit cell density-of-states and also in quasiparticle scattering between the pairs of electron-like pockets around opposing X‾a\overline{X}_a and X‾b\overline{X}_b points. We interpret these results as different manifestations of the same underlying phenomenon. The latter manifestation strongly suggests a kk-dependent shift of the band energies, that has an opposite sign near X‾a\overline{X}_a as compared to X‾b\overline{X}_b, and this would also give an intuitive explanation for the former manifestation: The electron bands flatten out exactly at the X‾a,b\overline{X}_{a,b} points so that the last remaining scattering vectors connecting the pairs of bands at the band bottoms are exactly qa,b=Ga,bq_{a,b} = G_{a,b}. The qaq_a and qbq_b striped patterns each appear at the energy that the respective band bottoms out. The simplest kk-dependent energy correction that explains the above results is one with dd-form factor.

A structural distortion is unlikely as the origin of the observed lifting of degeneracy because BaNi2_2As2_2 is tetragonal at room temperature, and no phase transition is reported down to T=2KT = 2\mathrm{K} [20]. It may be more reasonable to assume an electronic origin. We exclude a simple orbital ordering however, because there is apparently no degeneracy of the relevant orbitals that make up the pocket around the X points (as among dxzd_{xz} and dyzd_{yz}, for example). Moreover, the small density-of-states near EFE_{\mathrm{F}} likely excludes mechanisms stemming from a Fermi-surface-instability.

For these reasons we adopt a framework based on the density wave (DW) equation, which has previously been used to discuss dd-form factor energy corrections emerging from many-body correlations in the context of the unconventional superconductors [28–33, 35] as well as exotic density-wave states in other systems [34]. In this scenario, broken symmetry in the charge sector may arise as a many-body correction generated by the interference of spin fluctuations.

Figure 5 shows results of numerical calculations of dd-form factor energy corrections fl(k)f_l(\mathbf{k}). The full details of the calculations are described in the Methods section. Figure 5(a) and Figure 5(b) show plots of the form factors calculated for the two dominant orbitals of the electron-like band near X, namely dz2d_{z^2} and dx2−y2d_{x^2-y^2} [recall Figure 2(a) and (Figure 2c)]. The strength of each fl(k)f_l(\mathbf{k}) is plotted with a common color scale. Although fdz2(k)f_{d_{z^2}}(\mathbf{k}) and fdx2−y2(k)f_{d_{x^2-y^2}}(\mathbf{k}) are related by a sign change and therefore partially cancel each other, the former is a factor of ∼2\sim2 stronger than the latter and a significant dd-form factor correction remains. Note that an inverse Fourier transform of the dd-form factor results in a B2gB_{2g} bond-order pattern in real-space, similar to that illustrated in Figure 2(g). Figure 5(c) shows the resulting band structure after applying the correction to the band energies, sampled along two paths in the BZ whose ΓX\Gamma X segments run perpendicular to each other, as shown in the inset. The blue and red curves show negligible difference except in the region near the X-point. This difference serves to explain the observations of rotational symmetry breaking in QPI and in high-resolution spectroscopic imaging described in the sections above. Figure 5(d) shows the constant-energy contours of the corrected band structure at the energy of 50 meV, roughly the energy at which nematicity was observed. In Figure 5(e) we plot one quadrant of the BZ showing the Fermi contour of a Dirac cone, and that of another Dirac cone rotated by 90° from an adjacent quadrant. This visualizes the effect of the dd-form factor correction on the Dirac cones, which is seen to be fairly small. This is because the ΓM\Gamma M lines on which the Dirac cones are centered are the nodes of the form factor, as shown in Figure 5(a) and Figure 5(b).

Image not extracted

Figure 5. Symmetry breaking corrections via the density wave equation. (a) The calculated form factor fdz2(k)f_{d_{z^2}}(\mathbf{k}). (b) The corresponding form factor fdx2−y2(k)f_{d_{x^2-y^2}}(\mathbf{k}). (c) Band structures after corrections, plotted along two paths through the BZ that capture the rotational symmetry breaking (see inset). The corrected curves are shown in red and blue, corresponding to the red and blue paths depicted in the inset. They exhibit a significant difference in energy near the X-point. (d) A constant-energy band contour plot at EF+50E_{\mathrm{F}} + 50 meV, above the bottom of one set of electron-like pockets, but below the bottom of the other. (e) The effect of dd-form factor corrections on the Dirac cone. The constant energy contour in the positive quadrant of the BZ, at EFE_{\mathrm{F}}, showing the contour of a Dirac cone (gray). A contour from another Dirac cone, rotated by 90° into the positive quadrant, is also shown (magenta).

A discussion of the energy savings originating from the emergence of B2gB_{2g} bond order is given in the Supplemental Information.

Conclusion

To summarize the results of this work, we have characterized the cleaved (001) surface of BaNi2_2As2_2 using STM, where we can access the electronic properties of the uppermost Ni square net. Through tunneling spectroscopy

(including Landau level spectroscopy), we determine the approximate energy of the Dirac nodal-line to be ED≈27E_{\mathrm{D}} \approx27 meV. We estimate the Fermi velocities on the inner and outer arcs of each Dirac cone along the ΓM\Gamma M line, as vk,inner=3.09×105v_{k,\mathrm{inner}} = 3.09 \times10^5 m s−1^{-1} and vk,outer=−2.98×105v_{k,\mathrm{outer}} = -2.98 \times10^5 m s−1^{-1}, respectively. This is in fairly good agreement with previous reports [13].

High-resolution conductance imaging and QPI observations both reveal a lowering of symmetry, from the C4C_4 symmetry expected for a material belonging to the P4/nmmP4/nmm space group, to C2C_2. The C2C_2 configuration consists of a pair of striped patterns appearing at around 60 meV above EFE_{\mathrm{F}}, related to each other by a 90∘90^{\circ} rotation and ∼12\sim12 meV energy splitting, and oriented along the Ni bond chains. This symmetry-breaking phenomenon appears to respect the translational symmetries of the surface lattice but break its rotational symmetry. Hence it constitutes an example of nematic order.

A plausible explanation for the observed nematicity is the dd-form factor modification, fl(k)f_l(\mathbf{k}), to the hopping integrals on the Ni square net, characterized within the DW equation framework as resulting from interference between spin fluctuations. Because the Dirac points lie on the nodes of fl(k)f_l(\mathbf{k}), they are almost unaffected by the nematicity. An additional unusual feature of nematicity in BaNiS2_2 is that it is present even in a semimetal, having very low density-of-states at EFE_{\mathrm{F}}. This is in sharp contrast with mechanisms such as the Peierls-type instability or the band Jahn-Teller effect, which are usually promoted by a large density-of-states at EFE_{\mathrm{F}} and a correspondingly large available energy saving upon reconfiguration of the Fermi surface.

From the above findings we conclude that, in BaNiS2_2, topological Dirac nodal-lines and symmetry-breaking electronic correlations coexist without the former being significantly perturbed by the latter. However, looking ahead, the investigation of the BaCo1−x_{1-x}Nix_xS2_2 solid solution will be of interest because substitution with Co should be expected to increase the influence of electronic correlations on the electronic behavior. Ultimately, this may allow us to observe the fate of topologically protected Dirac cones as the system is driven toward a Mott transition.

Overall, as well as adding to the understanding of the nearly ideal Dirac nodal-line electronic structure of BaNiS2_2, the observations presented here also advance the general understanding of electronic nematic states in correlated transition metal square lattice systems.

Materials and Methods

STM measurements

Single crystals of BaNiS2_2 were synthesized as described previously [36, 37]. Samples were prepared for measurement by cleaving in an ultra-high vacuum chamber (P∼10−10P \sim10^{-10} Torr) on a cleaving stage held at about 77 K. They were then quickly inserted into a modified Unisoku 1300 low-temperature STM system held at 1.5 K [38], equipped with a home-made head similar to that described previously [39]. Scanning tips were formed by electrochemically etching tungsten wire, and characterized and conditioned using a field ion microscope and mild indentation at a clean Cu(111) surface. To measure differential conductance dIdV\frac{dI}{dV}, a lock-in technique was used, with modulation frequency fmod=617.3f_{\mathrm{mod}} = 617.3 Hz. The modulation amplitude VmodV_{\mathrm{mod}} used for each measurement, where relevant, is specified in the respective figure caption.

STM data processing

To prepare L(r)L(\mathbf{r}) image data for fast Fourier transformation, one of two methods was used. For all data shown in Figure 2, a Hann window was applied before transformation. For all imaging of QPI patterns in Figure 3 and Figure 5, Moisan’s periodic plus smooth image decomposition method was used [40]. This avoids the introduction by the Hann window of periodic envelope artifacts in the qq domain [visible in Figure S1 of the Supplemental Information].

Image not extracted

Figure S1. Additional details from STM topographic images. (a) A typical constant-current STM topograph (setpoints V=0.1V = 0.1 V, I=100I = 100 pA), and (b) its Fourier transform. The peak circled in red corresponds to a reciprocal lattice vector with a norm of 0.2210.221 Å−1^{-1} and corresponding periodicity of 4.524.52 Å, close to the expected lattice parameters aa and bb. (c) A constant-current topograph of a step between atomic terraces (V=0.1V = 0.1 V, I=10I = 10 pA), and (d) the corresponding height profile along the red line in (c). The observed height of 99 Å closely matches the lattice parameter cc.

Density functional theory and tight binding model calculations

To model QPI patterns, we first performed density functional theory (DFT) calculations for bulk BaNiS2_2 using the Perdew-Burke-Ernzerhof exchange-correlation functional [41], as implemented in the WIEN2k package [42]. We also performed additional calculations using a modified Becke-Johnson exchange potential [43] to account for the many-body correlation effects not treatable within DFT. These calculations are referred to as correlation-corrected DFT, or in short CC-DFT. For both sets of calculations, we sampled the bulk BZ using a 20×20×1020 \times20 \times10 kk-mesh and muffin-tin RMTR_{\mathrm{MT}} radius for all atoms such that its product with the maximum modulus of reciprocal space Kmax⁡K_{\max} becomes RMTKmax⁡=7.0R_{\mathrm{MT}}K_{\max} = 7.0. The bulk DFT and CC-DFT calculations were then down-folded into a 20-band tight-binding model using maximally localised Wannier functions [44, 45] incorporating Ni-3d3d orbitals as the projection centers. The tight-binding Hamiltonian was then used to construct a 100-layer supercell stacked along the crystalline cc-axis, while spanning the surface BZ over a fine 256×256256 \times256 kk-mesh.

As the CC-DFT results more successfully reproduced observed electronic structures, most notably the electron-like pocket around the Γ‾\overline{\Gamma}-point in photoemission measurements [13], these were adopted for the comparisons discussed in this work.

Calculations of QPI patterns

With the projected wave functions of the topmost two Ni layers, calculations of QPI were carried out using the TT-matrix formalism, in a manner similar to that described previously [46]. The lifetime broadening was chosen to be 1 meV and a localized scalar scatterer with a scattering potential of 0.1 eV in the unitary limit was employed. We performed a basis transformation from the lattice model to the continuum model using the Wannier function [47] constructed from the tight-binding wave-functions between ±1\pm1 eV at a height of 1 nm above the NiA plane. Lock-in broadening and normalization according to the integrated density-of-states were included for direct comparison with the experimental results. Simulations were carried out for a scatterer at both NiA and NiB sites, and summed to simulated a random distribution of impurities.

Calculations of dd-form factor correction via DW equation.

Based on the 20-band tight binding model derived from the CC-DFT results described above, a 2-D five band tight-binding model was constructed by ignoring inter-layer hopping, ignoring spin, and moving from the ‘folded’ to the ‘unfolded’ BZ. In this way the NiA and NiB sites are treated identically. This model is equivalent to one using the folded BZ if the spin-orbit interaction and inter-layer hopping are both ignored. (The results are displayed in Figure 5 after returning to the folded BZ.)

With this model as a starting point, and following previous works, the linearized DW equation [28] can then be used to characterize rotational symmetry breaking in the hopping integrals. As in previous treatments, the DW interaction includes the Maki-Thomson, Hartree, and Aslamazov-Larkin terms, the latter of which captures the effects of interference between spin fluctuations, and can contribute a dd-form factor correction [30]. The interaction is quantified by the momentum-dependent form factor fq,l(k)f_{q,l}(\mathbf{k}) where ll denotes a particular orbital. With q=0q = 0, fl(k)f_l(\mathbf{k}) can describe a uniform bond order as suggested by observation.

We set T=40T = 40 meV and J/U=0.1J/U = 0.1, where JJ is the Hund’s coupling and UU is the on-site Coulomb interaction.

Acknowledgements

We are grateful to T. Machida and A. Gauzzi for helpful discussions. This work was supported by JST CREST Grant No. JPMJCR16F2, and by a Grant-in-Aid for Scientific Research on Innovative Areas ‘Quantum Liquid Crystals’ (KAKENHI Grant No. JP19H05824 and No. JP19H05825) from JSPS of Japan. C.J.B. acknowledges support from RIKEN’s Programs for Junior Scientists.

Author Contributions

S. Shamoto synthesized the BaNiS2 crystals. C.J.B. performed all STM measurements and related data processing. M. S. B. performed the DFT and tight-binding calculations and Y. K. performed the subsequent TT-matrix calculations of QPI patterns. Y. Y., S. O. and H. K. proposed and calculated the dd-form factor band correction. T. H. supervised the project. All authors were involved in interpretation of the results. C. J. B. prepared the manuscript with input from all authors.

Data Availability

The data that support the findings presented here are available from the corresponding authors upon reasonable request.

Correlation-driven electronic nematicity in the Dirac semimetal BaNiS2_2

C. J. Butler; Y. Kohsaka; Y. Yamakawa; M. S. Bahramy; S. Onari; H. Kontani; T. Hanaguri; S. Shamoto

Further Results from STM

Topography

As indicated in Figure 1(b) of the main manuscript, cleavage can occur between adjacent BaS layers revealing (001) facets. The STM topography shown in Figure 1(g) is repeated in Figure S1 below. Its Fourier transform F[T(r)]F[T(\mathbf{r})] is shown in Figure S1 and the primary set of reciprocal lattice peaks represent a periodicity of 4.52 Å, close to the reported value for the surface lattice vectors a=b=4.43a=b=4.43 Å [1]. Although the Ba and apical S are uppermost at the cleaved surface, because the density-of-states at low energies is dominated by Ni orbitals, the atomic corrugation likely represents the Ni square net. Moreover, because the observed reciprocal lattice vector indicates only one Ni atom per unit cell is observed, the corrugation therefore represents specifically the uppermost NiA_\mathrm{A} square net.

An additional topography image that includes a step between two atomic terraces is shown in Figure S1, and a height profile, shown in Figure S1, shows an inter-layer separation of 9 Å, close to the lattice parameter of c=8.89c=8.89 Å [1].

Landau Level Spectroscopy

Under high magnetic fields, in suitably two- or quasi-two-dimensional metals, Landau quantization can be distinguished in conductance spectra measured using STM, in which Landau levels appear as conductance peaks [2, 3, 4]. Figure S2 shows dIdV(E)\frac{dI}{dV}(E) spectra acquired under various magnetic fields. The magnetic field was applied approximately parallel to the cc-axis. (Surface tilt was measured to be less than 1∘1^\circ.) Each curve represents the average over the same field of view as for Figure 1(h) and Figure 1(i) in the main text. The magnetic field was increased in increments of 1 T, up to 6 T, and then in increments of 0.5 T, up to the maximum field of H=12H=12 T.

Image not extracted

Figure S2. Landau level spectroscopy. (a) dI/dV(E)\mathrm{d}I/\mathrm{d}V(E) in the region of EFE_F and EDE_D, under magnetic field varying between 0 and 12 T. The curves are averaged over the field of view marked as a dashed square in Figure 1(g). Each curve is offset vertically by an amount proportional to the applied magnetic field. (b) The field-dependence as represented by subtracting the zero-field curve [dashed curve in (a)] from all others. Each curve is offset vertically as in (a). Anticipating the following result, we label the peak near to EDE_D and the first peak below it as LL0_0 and LL−1_{-1}, respectively. (c) Results of fitting the two labeled peaks using Eqn. 1, and using field-dependence curves from 6 to 12 T in increments of 0.5 Tesla.

A subtle change in the spectra appears in the region between E=0E=0 and 50 meV, and this is clarified upon taking the zero-field conductance curve as a baseline and subtracting it from all other curves, as shown in Figure S2. Two peaks are seen to become prominent approaching H=12H=12 T. The peak at lower energy shifts significantly with magnetic field, while the peak at higher energy appears to have only a small shift. Knowing that under ideal circumstances, the energy of the zeroth Landau level of a topological system should have no field-dependence, we anticipate that the upper of the two peaks can be assigned LL0_0, and the lower of the two can then be assigned as LL−1_{-1}.

Assuming a 2-D Dirac cone, the energy of the nthn^\mathrm{th} Landau level can be expressed as

En=ED+sgn⁡(n)vˉ2eℏ(n+γ)μ0H+gLLμBμ0H(1)E_n=E_D+\operatorname{sgn}(n)\bar{v}\sqrt{2e\hbar(n+\gamma)\mu_0H}+g_{\mathrm{LL}}\mu_{\mathrm{B}}\mu_0H \tag*{(1)}

where vˉ\bar{v} is the average velocity on the contour of the band from which the Landau level forms, ee is the electronic charge, ℏ\hbar is the reduced Planck constant, μ0\mu_0 is the vacuum permeability, gLLg_{\mathrm{LL}} is the effective gg-factor, which we assume to be common to all Landau levels. The phase factor γ\gamma is given by

γ=12−ΦB2π.(2)\gamma=\frac{1}{2}-\frac{\Phi_{\mathrm{B}}}{2\pi}. \tag*{(2)}

Here ΦB\Phi_{\mathrm{B}} is the Berry phase accumulated around the Landau orbit. This signifies the topological character of the system and takes the value π\pi if the orbit encloses a non-trivial Dirac node. In the case that γ=0\gamma=0, and if the Zeeman term is ignored, the energy of the zeroth Landau level (LL0_0) remains at EDE_D independent of magnetic field. With the Zeeman term considered, LL0_0 can shift away from EDE_D with increasing magnetic field given a finite gg-factor gLLg_{\mathrm{LL}}. Using this framework, and adopting the assumption that γ=0\gamma=0, we estimate EDE_D, vˉ\bar{v} and the two gg-factors by fitting to the data as shown in Figure S2. For this procedure, after a two-point (0.5 meV) Gaussian filter is applied to smooth each curve, the Landau level peak energies are initially guessed using Epeak,H=argmin⁡E[d2dE2(dIdV∣H(E)−dIdV∣H=0T(E))]E_{\mathrm{peak},H}=\underset{E}{\operatorname{argmin}}\left[\frac{d^2}{dE^2}\left(\frac{dI}{dV}\bigg|_H(E)-\frac{dI}{dV}\bigg|_{H=0^T}(E)\right)\right]. The fitting estimation yields ED≈26.8E_D \approx26.8 meV, vˉ≈1.07×105\bar{v} \approx1.07 \times10^5 m s−1^{-1}, and gLL≈1.32g_{\mathrm{LL}} \approx1.32.

To summarize, the observation of a field-independent Landau level implies the topologically non-trivial nature of the Dirac cones in BaNiS2_2.

Nematicity in a step-terrace morphology

Figure S3 shows measurements across a step-terrace morphology found at a BaNiS2_2 surface. Two steps were found, separating three adjacent terraces [the lower terrace, seen at the bottom of Figure S3 is the same one as at the upper terrace seen at the bottom of Figure S3]. Images of L(r)L(r) and their Fourier transforms Lq(q)L_q(q), acquired with bias separation of 10 mV, capture the presence of the nematicity described in the main text. The L(r)L(r) and Lq(q)L_q(q) images acquired on the upper and lower terraces, in the regions marked by the dashed squares in Figure S3, are shown in Figure S3 and Figure S3, respectively. The corresponding data acquired above and below the second step, in the regions marked by dashed squares in Figure S3, are shown in Figure S3 and S5(f),

Image not extracted

Figure S3. Nematicity in three adjacent terraces. (a) Topography showing a single unit-cell high step (setpoints V=0.1V = 0.1 V, I=100I = 100 pA). (b) Images of L(r)L(\mathbf{r}) and their Fourier transforms Lq(q)L_q(\mathbf{q}) at E=58E = 58 mV and 68 mV. (c) The corresponding images acquired on the lower terrace. (d) Topography showing another single unit-cell high step. The higher terrace (at the bottom of the image) is the same as that shown in the lower right of (a), as indicated by the asterisk symbols. (e) and (f) The corresponding L(r)L(\mathbf{r}) and Lq(q)L_q(\mathbf{q}) images at E=58E = 58 mV and 68 mV. Here nematicity manifests as a differing intensity between the Lq(q)L_q(\mathbf{q}) peaks corresponding to the reciprocal lattice vectors Ga\mathbf{G}_a and Gb\mathbf{G}_b. The orientation of the striped patterns is the same for all three terraces.

respectively.

Figure S3 and Figure S3 represent data collected on the same terrace [as marked by the asterisk in Figure S3 and Figure S3], but have a different appearance due to the STM tip becoming contaminated by during the series of measurements.

Each set of images exhibits nematicity with the same orientation. This indicates either that the relation between nematicity in adjacent steps has ‘ferro’-type order, or more likely, that the nematicity in adjacent layers is not directly correlated, but rather is coupled to a global condition such as strain.

Fitting of quasiparticle interference signals

Figure S4 shows a comparison of linecuts through measured Lq(q,E)L_q(q,E) data, and full results of fitting that is summarized by the solid red and orange lines shown in Figure 3(c) of the main text. A model consisting of the sum of multiple Lorentzian lineshapes and a constant offset, cqc_q, is fitted at each point along the qq path. This is written as

Lqfit(E)=∑n=1NIn(Γn/2)(E−En)2+(Γn/2)2+cq(3)L_q^{\mathrm{fit}}(E) = \sum_{n=1}^{N} \frac{I_n(\Gamma_n/2)}{(E-E_n)^2+(\Gamma_n/2)^2}+c_q \tag*{(3)}

Here NN is the number of Lorentzian lineshapes included in the model, which is automatically determined for each qq point using a peak finding algorithm [5] after some smoothing.

Quantitative Description of Dirac Cones Derived from QPI Results

As described in the main text, the apparent dispersion of a QPI signal has only an indirect relation to the actual dispersion of the related bands in kk-space, and for this reason interpretation of QPI observations is generally not straightforward. In the present case the relation is complicated by the fact that the Dirac cones are not centered at q=0q = 0, and that only scattering between Dirac cones is observed, and not scattering within a single one. Nevertheless below we draw a comparison between the observed and calculated features in qq-space, and also a kk-space-to-qq-space comparison of the calculation results,

specifically for the QPI signals q2q_2 and q3q_3 shown in Figure 3 of the main text. From this we discuss whether the Dirac point energy can be inferred from QPI observations, and also derive estimates of the Dirac cone velocities parallel to the ΓM\Gamma M line.

First, if the observed scattering branches q2q_2 and q3q_3 result from scattering between adjacent Dirac cones as depicted in Figure 3(d) of the main text, they intersect at the energy where each of the Dirac cones vanishes to a point, so that the shortest and longest possible vectors connecting their contours (namely q2q_2 and q3q_3) become identical. The energy at which they intersect (E=101E = 101 meV) would then seem to be the Dirac point energy. However, we find that this is not true. The calculated Dirac point energy is in fact about 60 meV [see Figure 1(d) in the main text], meaning the extrapolated point of intersection gives an over-estimate, perhaps due to a finite curvature as the band nears the Dirac point. This indicates that an extrapolation from the branches q2q_2 and q3q_3 in the observation (which would yield ED≈47E_D \approx47 meV) would similarly over-estimate the actual Dirac point energy. For this purpose we fall back on the minimum in the dIdV(E)\frac{dI}{dV}(E) curve [Figure 1(i) of the main text], and also Landau level spectroscopy above, to determine the Dirac point energy.

As described above, the point of intersection of q2\mathbf{q}_{2} and q3\mathbf{q}_{3} represents the qq-vector connecting two adjacent Dirac points, which we may call qD−D\mathbf{q}_{\mathrm{D-D}}. The assumption that each Dirac point lies on a ΓM\Gamma M line in kk-space leads to ∣qD−D∣2=2∣kD∣2|\mathbf{q}_{\mathrm{D-D}}|^{2}=2|\mathbf{k}_{\mathrm{D}}|^{2} where kD\mathbf{k}_{\mathrm{D}}, the wavevector of the Dirac point, is directed from the origin along the ΓM\Gamma M line. This results in ∣kD∣≈0.884 nm−1|\mathbf{k}_{\mathrm{D}}|\approx0.884\ \mathrm{nm}^{-1}.

An approximate value for the Dirac cones’ Fermi velocity vF=dEFd∣k∣v_{\mathrm{F}}=\frac{\mathrm{d}E_{\mathrm{F}}}{\mathrm{d}|\mathbf{k}|} along a particular axis can be inferred from the apparent velocity vq,2,3v_{q,2,3} of the scattering branches q2\mathbf{q}_{2} and q3\mathbf{q}_{3}. We start from the assumption that vqv_q for a generic scattering branch always relates to the velocity vkv_k of a chosen point on the band by a proportionality constant that captures geometric details of the scattering vectors allowed within the band’s structure, i.e. vq=cvkv_q=cv_k. (As an example, vq=12vkv_q=\frac{1}{2}v_k in the extremely simple case of scattering across the diameter of an isotropic Dirac cone).

Next, we compare the calculated and observed velocities vqobs.v_q^{\mathrm{obs.}} and vqcalc.v_q^{\mathrm{calc.}} for each branch [see Figure 3(c) and Figure 3(f) of the main text]. These velocities are related by a factor we call the renormalization factor RR, such that vqobs.=Rvqcalc.v_q^{\mathrm{obs.}}=Rv_q^{\mathrm{calc.}}. Since in both calculation and in reality, vq=cvkv_q=cv_k with the same constant cc (determined only by geometry), we also posit that the real band velocity is vk=Rvkcalc.v_k=Rv_k^{\mathrm{calc.}}. In this way a renormalization of the kk-space band structure can reasonably be inferred from the apparent renormalization of qq-space structures.

The resulting calculated apparent velocities [see Figure 3(f) of the main text] are vq2calc=1.41×105 m s−1v^{\mathrm{calc}}_{q2}=1.41\times10^{5}\ \mathrm{m\,s^{-1}} and vq3calc=−1.18×105 m s−1v^{\mathrm{calc}}_{q3}=-1.18\times10^{5}\ \mathrm{m\,s^{-1}}, respectively. Comparing these with the experimental values leads to R=0.58R=0.58 for both inner and outer branches q2q_{2} and q3q_{3}.

Using the same method of fitting as shown in Figure S4, but for the surface spectral function shown in Figure 1(d) of the main text, gives velocities for the inner and outer arcs of the Dirac cone, along the ΓM‾\overline{\Gamma M} line, of vk,innercalc=5.32×105 m s−1v^{\mathrm{calc}}_{k,\mathrm{inner}}=5.32\times10^{5}\ \mathrm{m\,s^{-1}}, and vk,outercalc=−5.14×105 m s−1v^{\mathrm{calc}}_{k,\mathrm{outer}}=-5.14\times10^{5}\ \mathrm{m\,s^{-1}}, respectively. Finally, using R=0.58R=0.58 allows us to infer that the actual band velocities in the sample are vk,inner=3.09×105 m s−1v_{k,\mathrm{inner}}=3.09\times10^{5}\ \mathrm{m\,s^{-1}} and vk,outer=−2.98×105 m s−1v_{k,\mathrm{outer}}=-2.98\times10^{5}\ \mathrm{m\,s^{-1}}. As a comparison, from recently reported angle-resolved photoemission spectroscopy results, we estimate corresponding absolute velocities of ∣vk,inner∣∼∣vk,outer∣≈2×105 m s−1\lvert v_{k,\mathrm{inner}}\rvert\sim\lvert v_{k,\mathrm{outer}}\rvert\approx2\times10^{5}\ \mathrm{m\,s^{-1}}[6].

Image not extracted

Figure S4. Fitting to Lq(q,E)L_q(q,E) data. (a) and (b) Linecuts through measured Lq(q,E)L_q(q,E) data, along the Ga\mathbf{G}_{a} and Gb\mathbf{G}_{b} reciprocal lattice vectors. (c) and (d) Corresponding images composed of the fitting results at each qq point, with a fitting model described by Eq. (3). Where the fitting procedure failed to converge, a vertical black line is shown. (e) and (f) The same fitting results displayed as points and ‘error bars’ (which in this case denote the Lorentzian broadening Γ\Gamma). These plots are superimposed together in Figure 4(c) of the main text.

The band velocities along the ΓM‾\overline{\Gamma M} line are higher than the average Fermi velocity as determined through Landau level spectroscopy above. This is partly because the Dirac cones are elliptical, having the highest velocity along ΓM‾\overline{\Gamma M} and the lowest velocity perpendicular to this line. It may also be because the Fermi contour generally has a lower velocity as the Dirac cone has some curvature, becoming lower in velocity near EDE_{\mathrm{D}}. The estimate of velocities based on QPI dispersions relies on energies between −50-50 and −100 meV-100\ \mathrm{meV}, where the cone has steeper and more linear dispersion.

Origin of Energy Savings Due to Nematic Order

Here, we discuss why an energy gain results from the present finite-energy nematic order. Although the Dirac nodes are almost unchanged by the B2gB_{2g} form factor, the band dispersion along the antinodal direction is significantly modified. First, we discuss the change in the free-particle energy due to the form factor f(k)=f×fˉl(k)f(\mathbf{k})=f\times\bar{f}_{l}(\mathbf{k}), where fˉl(k)\bar{f}_{l}(\mathbf{k}) is the normalized form factor (max⁡l,k∣fˉl(k)∣=1)(\max_{l,\mathbf{k}}\lvert\bar{f}_{l}(\mathbf{k})\rvert=1). In the hole representation, it is given by

ΔEfreef≡Efreef−Efreef=0≈af2+bf4,(4)\Delta E_{\mathrm{free}}^{f}\equiv E_{\mathrm{free}}^{f}-E_{\mathrm{free}}^{f=0}\approx af^{2}+bf^{4}, \tag*{(4)}
Efreef=1N∑n,k(1−f(ϵn,kf))(−ϵn,kf),(5)E_{\mathrm{free}}^{f}=\frac{1}{N}\sum_{n,\mathbf{k}}\left(1-f(\epsilon_{n,\mathbf{k}}^{f})\right)(-\epsilon_{n,\mathbf{k}}^{f}), \tag*{(5)}

where f(ϵ)f(\epsilon) is the Fermi distribution function, ϵn,kf\epsilon_{n,\mathbf{k}}^{f} is the nn-th band dispersion with the form factor. We perform the numerical study at T∼0.01T\sim0.01, where Eq. (4) is very close to the free energy because TSTS is small. In the present model, the relations a<0a<0 and b>0b>0 hold, and therefore ΔEfreef<0\Delta E_{\mathrm{free}}^{f}<0 at a finite ff. This energy gain mainly originates from the electron-like dispersion around the X,Y point (Figure 5). In fact, the bottom of this dispersion (E0≈0.05 eV)(E_{0}\approx0.05\ \mathrm{eV}) is given as EX(Y)≈E0+(−)cf+df2E_{\mathrm{X(Y)}}\approx E_{0}+(-)cf+df^{2}, where c≈1c\approx1 and d>0d>0. In fact, d≈(E0−E0′)−1d\approx(E_{0}-E_{0}^{\prime})^{-1} due to the second-order perturbation theory, and E0′≈−0.5 eVE_{0}^{\prime}\approx-0.5\ \mathrm{eV} is the energy of the nearest valence band at X,Y points. Thus, the negative aa in Eq. (4) is robustly obtained in the present band-structure.

Next, we discuss the total energy with the electron correlation, ΔEf\Delta E^{f}. According to Ref. [7], it is simply given as

ΔEf≈a′f2+bf4,(6)\Delta E^{f}\approx a^{\prime}f^{2}+bf^{4}, \tag*{(6)}

where a′=a(1−1/λ)a^{\prime}=a(1-1/\lambda), and λ\lambda is the eigenvalue of the DW equation. Here, λ>1(<1)\lambda>1(<1) below (above) TnemT_{\mathrm{nem}}. Because a<0a<0 in BaNi2_2As2_2, a′a^{\prime} in Eq. (6) is negative below TnemT_{\mathrm{nem}}, and the order parameter is given as ∣f∣=−a′/2b\lvert f\rvert=\sqrt{-a^{\prime}/2b}. According to Ref. [7], the specific heat jump at TnemT_{\mathrm{nem}} is ΔC/T∼αN(0)\Delta C/T\sim\alpha N(0) with α=(−λ˙T)nem\alpha=(-\dot{\lambda}T)_{\mathrm{nem}}, where λ˙≡dλ/dT\dot{\lambda}\equiv d\lambda/dT. In the BCS superconductor, α=1\alpha=1. In contrast, α≪1\alpha\ll1 is expected in the case of the nematic transition at T=TnemT=T_{\mathrm{nem}}[7].

Temperature Dependence

A natural next step following on from these results is to experimentally characterize the temperature dependence of the nematic order. The energy splitting of ∼12 meV\sim12\ \mathrm{meV} corresponds to a temperature of ∼140 K\sim140\ \mathrm{K}, naively the minimum possible transition temperature for the nematic phase. Unfortunately, tunneling spectroscopy observations suffer from an energy broadening that rises faster than kBTk_{\mathrm{B}}T (by a factor of 3.5), meaning that observations similar to those presented in this work probably become impractical before reaching the temperature required to observe the potential melting of the nematic phase. This means that further investigations using other tools may be necessary to characterize its temperature dependent behavior.

References

  1. [1]I. E. Grey and H. Steinfink. Crystal structure and properties of barium nickel sulfide, a square-pyramidal nickel(II) compound. J. Am. Chem. Soc. 92, 17, 5093–5095 (1970). https://doi.org/10.1021/ja00720a015
  2. [2]P. Cheng, C. Song, T. Zhang, Y. Zhang, Y. Wang, J.-F. Jia, J. Wang, Y. Wang, B.-F. Zhu, X. Chen, X. Ma, K. He, L. Wang, X. Dai, Z. Fang, X. Xie, X.-L. Qi, C.-X. Liu, S.-C. Zhang, and Q.-K. Xue, Landau Quantization of Topological Surface States in Bi₂Se₃. Phys. Rev. Lett. 105, 076801 (2010). https://doi.org/10.1103/PhysRevLett.105.076801
  3. [3]T. Hanaguri, K. Igarashi, M. Kawamura, H. Takagi, and T. Sasagawa, Momentum-resolved Landau-level spectroscopy of Dirac surface state in Bi₂Se₃. Phys. Rev. B 82, 081305(R) (2010). https://doi.org/10.1103/PhysRevB.82.081305
  4. [4]Y.-S. Fu, T. Hanaguri, K. Igarashi, M. Kawamura, M. S. Bahramy, and T. Sasagawa, Observation of Zeeman effect in topological surface state with distinct material dependence. Nat. Commun. 7, 10829 (2016). https://doi.org/10.1038/ncomms10829
  5. [5]https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.find_peaks.html
  6. [6]N. Nilforoushan, M. Casula, A. Amaricci, M. Caputo, J. Caillaux, L. Khalil, E. Papalazarou, P. Simon, L. Perfetti, I. Vobornik, P. K. Das, J. Fujii, A. Barinov, D. Santos-Cottin, Y. Klein, M. Fabrizio, A. Gauzzi, and M. Marsi, Moving Dirac nodes by chemical substitution. Proc. Natl. Acad. Sci. 118, e2108617118 (2021). https://doi.org/10.1073/pnas.2108617118
  7. [7]R. Tazai, S. Matsubara, Y. Yamakawa, S. Onari, H. Kontani, A Rigorous Formalism of Unconventional Symmetry Breaking in Fermi Liquid Theory and Its Application to Nematicity in FeSe. arXiv:2205.02280 (2022). https://doi.org/10.48550/arXiv.2205.02280

Paper details

vlnc:W_pk4k4jhb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents