betaFeedback

WorksPhysics and Astronomy

Dirac branch-cut modes with relativistic transport

Abstract

Emergent Dirac fields, exhibiting effective relativistic physics, are most commonly associated with bulk and surface states in materials such as graphene and topological insulators. Here we show that phase dislocations of a complex Dirac mass support a distinct family of propagating Dirac defect states that behave as massless or massive one-dimensional relativistic modes. These traveling-wave defect states, termed Dirac branch-cut (DBC) modes, are governed by a one-dimensional Dirac Hamiltonian whose effective mass is controlled by the phase difference across the branch cut. Using acoustic metamaterials, we experimentally demonstrate a range of relativistic phenomena exhibited by DBC modes, including relativistic dispersion, energy-independent confinement, Klein tunnelling, and transport along freeform (e.g., spiral) trajectories. Our results establish branch-cut defects as a distinct platform for realizing propagating Dirac defect states, extending relativistic Dirac physics from bulk and surface states to propagating modes confined to defect boundaries.

Dirac fields are of profound importance in many areas of physics [1], including particle physics [2] and condensed matter physics [3], because they provide a natural framework for effective relativistic wave phenomena. The discovery of graphene showed that such physics is not restricted to elementary particles, but can emerge as low-energy excitations in crystalline materials [4, 5]. Since then, phenomena such as Klein tunneling [6] have been explored primarily through extended bulk or surface states associated with Dirac cones [7, 8, 9, 10]. In this setting, relativistic behaviour is tied to dispersive Dirac bands and the propagation of waves through the bulk or along material surfaces.

A separate and equally influential direction concerns defect states in Dirac fields. The best-known examples are the Jackiw–Rebbi state [11], localized at a domain-wall defect where a real Dirac mass changes sign, and the Jackiw-Rossi state [18], bound to a vortex of a complex Dirac mass. These states have shaped our understanding of charge fractionalization [12], topological phases of matter [13, 14], and Majorana physics [20, 21, 22], and have been realized across condensed-matter and metamaterial platforms [15, 16, 17, 23, 24, 25, 26, 27, 28, 29, 30, 31]. However, while Dirac cones provide propagating relativistic states, the associated defects have been viewed mainly as mechanisms for binding topological states. Whether a lower-dimensional defect can itself host relativistic Dirac modes, and the resulting consequences, remain largely unexplored.

Here we show that a phase discontinuity in a complex Dirac mass field—a branch-cut defect—supports such states. A branch cut arises when the complex mass field is assigned to a multi-valued complex function, such as a root or logarithm, producing a phase discontinuity along a curve [32, 33]. We find that this phase discontinuity binds a distinct family of propagating relativistic defect states, which we term Dirac branch-cut (DBC) modes. Unlike Jackiw-Rebbi and Jackiw-Rossi states, DBC modes satisfy a one-dimensional (1D) Dirac equation, retaining the relativistic features of the parent Dirac field, including the same Dirac velocity. Their effective mass is determined by the phase discontinuity across the cut, and may be zero (linear dispersion) or nonzero (hyperbolic dispersion); in either case, the DBC modes exhibit energy-independent transverse confinement, a signature property arising from the fact that both the bulk and defect are relativistic. We experimentally realize DBC modes using acoustic metamaterials composed of a Kekulé modulated triangular lattice of rigid pillars [23, 24, 26], with the modulation amplitude controlling the mass magnitude and the modulation phase controlling the phase. The resulting DBC modes exhibit relativistic dispersion, constant transverse confinement, and Klein tunneling along the defect. We also demonstrate DBC modes in a variety of configurations, including line cavities formed by finite cuts and freeform transport along spiral trajectories. These properties establish branch cuts as a distinct platform for realizing Dirac defect states alongside domain walls and vortices, while revealing a new regime in which relativistic Dirac transport occurs along defect boundaries rather than in bulk or surface states.

To derive the existence of DBC modes, we take the Jackiw–Rossi model, which comprises a minimal four-component Dirac equation in 2D space with the Hamiltonian [18, 28].

H=τ3(−iσ1∂∂x−iσ2∂∂y)+m1(r)τ1+m2(r)τ2,(1)H=\tau_{3}\left(-i\sigma_{1}\frac{\partial}{\partial x}-i\sigma_{2}\frac{\partial}{\partial y}\right)+m_{1}(\boldsymbol{r})\tau_{1}+m_{2}(\boldsymbol{r})\tau_{2}, \tag*{(1)}

where the Dirac velocity is normalized to 1, the 2D position vector is r≡(x,y)\boldsymbol{r}\equiv(x,y), and the internal Hilbert space is factorized as C2⊗C2\mathbb{C}^{2}\otimes\mathbb{C}^{2}, with σ1,2,3\sigma_{1,2,3} and τ1,2,3\tau_{1,2,3} denoting Pauli matrices acting on each factor space. There are two Dirac mass fields m1m_{1} and m2m_{2}, which can be consolidated into a complex mass field m(r)≡m1+im2=∣m∣eiθm(\boldsymbol{r})\equiv m_{1}+im_{2}=|m|e^{i\theta}; the model excludes other mass terms that break chiral symmetry [28]. Note that HH is nonetheless Hermitian. Jackiw and Rossi considered a specific m(r)m(\boldsymbol{r}) with θ\theta winding by 2π2\pi around the origin, and showed that it hosts a mid-gap bound state [18]; this was later generalized to an index theorem stating that any m(r)m(\boldsymbol{r}) with a 2π2\pi phase winding implies the existence of such a bound state [19].

Now, instead of phase singularities, we study phase dislocations. Consider the piecewise-constant mass field in Figure 1a, with m=m0eiθ+m=m_{0}e^{i\theta_{+}} for y>0y>0 and m=m0eiθ−m=m_{0}e^{i\theta_{-}} for y<0y<0, where m0∈R+m_{0}\in\mathbb{R}^{+} is a constant. In each domain, there is a bulk gap [−m0,m0][-m_{0},m_{0}], while at the domain wall y=0y=0 there is a phase discontinuity Δθ=θ+−θ−\Delta\theta=\theta_{+}-\theta_{-}. For any Δθ≠0\Delta\theta\ne0, there exist DBC modes along the domain wall, with the Jackiw–Rebbi-like form

Figure 1. Properties of Dirac branch-cut (DBC) modes. a, Upper panel: complex plane for the Jackiw–Rossi mass parameter mm, indicating two points with the same ∣m∣\lvert m\rvert and phase difference Δθ\Delta\theta. Lower panel: a schematic of a piecewise-constant m(r)m(r) distribution in 2D space, constructed from those two mm values with phase dislocation along y=0y=0. b, Phase dislocation mode (DBC mode) energies versus wavenumber kxk_x for Δθ=π\Delta\theta=\pi (left panel) and Δθ=2π/3\Delta\theta=2\pi/3 (middle panel), and versus Δθ\Delta\theta at kx=0k_x=0 (right panel). Cyan dashes show the results obtained analytically using Eqs. (4)–(5). Color maps show the minimum eigenvalue magnitude of the scattering matrix, computed directly from the Hamiltonian (1); a zero value corresponds to a boundary state (see Supplementary Information). The bulk bands are shaded in gray. c, Phase profile of a mass distribution made from a branch function m=(z−3)1/2(z+3)1/2m=(z-3)^{1/2}(z+3)^{1/2}, where z=x+iyz=x+iy. d, Numerically-obtained DBC mode energies versus mode order with the phase of mass distribution in c, and the amplitude extracted from the branch function ∣m∣\lvert m\rvert. e, Spatial profiles for two of the modes in d: the Jackiw–Rossi state at E=0E=0 (left panel), and the order-4 mode (right panel). f, Phase profile of a “phase-only” mass distribution ∣m∣=m0\lvert m\rvert=m_0 and θ=arg⁡[(z−3)1/2(z+3)−1/2]\theta=\arg[(z-3)^{1/2}(z+3)^{-1/2}], with m0=5m_0=5. g, DBC mode energies for the mass distribution in f (blue markers), and for another mass distribution with Δθ=2π/3\Delta\theta=2\pi/3 using the same cut (yellow markers). The theoretically-predicted minigap for the latter is indicated by horizontal dashes. h Representative spatial profiles of two of the DBC modes for Δθ=π\Delta\theta=\pi. For the DBC mode profiles for Δθ=2π/3\Delta\theta=2\pi/3, see Supplementary Information Figure S1,Figure S1. All the DBC modes have very similar confinement lengths, up to high orders approaching the bulk bands.

ψ(y)={e+κye−iτ3θ−/2∣χ−⟩,y<0,e−κye−iτ3θ+/2∣χ+⟩,y>0,(2)\psi(y)= \begin{cases} e^{+\kappa y}e^{-i\tau_{3}\theta_{-}/2}|\chi_{-}\rangle, & y<0,\\ e^{-\kappa y}e^{-i\tau_{3}\theta_{+}/2}|\chi_{+}\rangle, & y>0, \end{cases} \tag*{(2)}

where κ=(m02+kx2−E2)1/2\kappa=(m_{0}^{2}+k_{x}^{2}-E^{2})^{1/2} is the inverse decay depth, kxk_{x} is the wavenumber along xx, and the energy EE lies in the bulk gap (∣E∣<m0|E|<m_{0}), and ∣χ±⟩|\chi_{\pm}\rangle are spinors satisfying

[τ3(kxσ1±iκσ2)+m0τ1]∣χ±⟩=E∣χ±⟩.(3)\left[\tau_{3}\left(k_{x}\sigma_{1}\pm i\kappa\sigma_{2}\right)+m_{0}\tau_{1}\right]|\chi_{\pm}\rangle=E|\chi_{\pm}\rangle. \tag*{(3)}

It can be shown that the continuity of ψ(y)\psi(y) along y=0y=0 constrains the solutions to Eq. (3) to a Δθ\Delta\theta-dependent subspace (see Methods). The DBC mode eigenproblem then maps to a two-component 1D Dirac equation, with

E2=kx2+mb2,mb=m0∣cos⁡(Δθ/2)∣,(4)E^{2}=k_{x}^{2}+m_{b}^{2},\qquad m_{b}=m_{0}\left|\cos(\Delta\theta/2)\right|, \tag*{(4)}
κ=m0∣sin⁡(Δθ/2)∣.(5)\kappa=m_{0}\left|\sin(\Delta\theta/2)\right|. \tag*{(5)}

In Eq. (4), the effective mass mbm_{b} depends on m0m_{0} and the phase dislocation Δθ\Delta\theta. For Δθ=π\Delta\theta=\pi, which corresponds to a Jackiw–Rebbi-like sign flip of mm across the domain wall, we have mb=0m_{b}=0; other values of Δθ\Delta\theta give 0<mb<m00<m_{b}<m_{0}, generating a tunable family of massive 1D Dirac modes. The addition of a phase gradient vθyv_{\theta}y retains this result, but with a correction to mbm_{b} (see Methods). Importantly, κ\kappa is independent of kxk_{x}, so all the DBC modes have the same transverse confinement length regardless of where they lie in the bulk gap.

they lie in the bulk gap. This is due to the 1D dispersion relation (4) and the bulk dispersion relation for guided modes, E2=kx2−κ2+m02E^{2}=k_{x}^{2}-\kappa^{2}+m_{0}^{2}, from Eq. (1); these involve the same Dirac velocity, and combining them leads to Eq. (5). We will return to this robust confinement property later.

In Figure 1b, we compare the analytic result Eq. (4) to numerical results obtained directly from Eq. (1) (see Supplementary Information, Sec. S1). We obtain exact agreement for the dispersion relations (left and middle panels) and their Δθ\Delta\theta-dependence (right panel).

From these findings, we hypothesize that the DBC modes arise chiefly from the presence of a discontinuity in θ(r)\theta(\boldsymbol{r}), and will also exist for phase dislocations along arbitrary 1D curves. A convenient way to generate phase dislocations is to use a branch-cut—a 1D curve connecting different Riemann sheets of a complex multi-valued function, like a complex root or logarithm [32]. A branch function is specified by choosing one Riemann sheet, and is differentiable for all rr except at the cut, where there is a constant phase discontinuity given by the monodromy (i.e., the multiplicity of the Riemann sheets). As an example, take f(z)=(z−z1)α(z−z2)βf(z)=(z-z_{1})^{\alpha}(z-z_{2})^{\beta}, where z=x+iyz=x+iy, and α,β∈Q\alpha,\beta\in\mathbb{Q} with α+β∈Z\alpha+\beta\in\mathbb{Z}. The branch points are z1z_{1} and z2z_{2}, with no branch point at infinity; we can place the cut freely so long as it does not cross itself and terminates at the branch points. All along the cut, there is a phase discontinuity Δθ=2πα≡−2πβ(mod2π)\Delta\theta=2\pi\alpha\equiv-2\pi\beta\pmod{2\pi}.

The cut can be continuously deformed between the branch points without changing its monodromy-induced phase discontinuity Δθ\Delta\theta. This phase discontinuity supports the extended DBC modes, whereas a nonzero net phase winding of the overall mass texture can additionally enforce a Jackiw–Rossi zero mode.

In Figure 1c, we show the distribution of θ(r)=arg⁡[m(r)]\theta(\boldsymbol{r})=\arg[m(\boldsymbol{r})] for m(r)=(z−3)1/2(z+3)1/2m(\boldsymbol{r})=(z-3)^{1/2}(z+3)^{1/2}. The cut is chosen to be a sinusoidal curve connecting the branch points at z=±3z=\pm3, and the phase discontinuity is Δθ=π\Delta\theta=\pi. Since ∣m∣→∞|m|\to\infty as ∣r∣→∞|\boldsymbol{r}|\to\infty, there are no free solutions, only bound states. We solve for these numerically using the finite element method (FEM) (see Methods and Supplementary Information, Sec. S2). The spectrum consists of discrete energies extending to E→±∞E\to\pm\infty, as shown in Figure 1d. As the branch function has a phase winding of 2π2\pi upon encircling both branch points, the index theorem for the Jackiw–Rossi model predicts a mode at E=0E=0 [18, 19], consistent with our obtained spectrum. Thus, in this particular configuration, the Jackiw–Rossi zero mode coincides with the zero-energy member of the DBC spectrum. However, all the DBC modes are strongly bound to the cut. Two of them, the zero mode and the order-4 mode, are plotted in Figure 1e.

We can also devise mass distributions with finite bulk gaps, which are more suited to experimental realization (see below). An expedient way to do this is to take only the phase of a complex branch function, keeping the magnitude constant. As an example, we take ∣m∣=5|m|=5 with θ=arg⁡{[(z−3)/(z+3)]1/2}\theta=\arg\{[(z-3)/(z+3)]^{1/2}\}, using an arc-like cut, as shown in Figure 1f; though this m(r)m(\boldsymbol{r}) is non-holomorphic, that does not invalidate our preceding discussion. There is now a bulk gap E∈[−5,5]E\in[-5,5], with free states outside and DBC modes inside the gap (Figure 1g, blue circles). The net phase winding is now zero, so the DBC modes do not include a Jackiw–Rossi mode. For comparison, we also plot the spectrum for ∣m∣=5\lvert m\rvert=5 and θ=arg⁡{[(z−3)/(z+3)]1/3}\theta=\arg\{[(z-3)/(z+3)]^{1/3}\}, which has phase discontinuity Δθ=2π/3\Delta\theta=2\pi/3 (Figure 1g, yellow circles). These DBC modes exhibit a minigap closely matching the one derived from Eq. (4), as indicated by dashes in the figure. The DBC modes are again strongly confined to the cut, as shown in Figure 1h. Remarkably, even high-order DBC modes very close to the bulk band edge (e.g., mode # 10 in Figure 1g,Figure 1h) have the same tight confinement, consistent with the prediction from Eq. (5), that does not invalidate our preceding discussion.

In the Supplementary Information, Sec. S3, we investigate the mode profiles in greater detail. All DBC modes have profiles matching those in Figure 1e,Figure 1h. Their transverse confinement lengths are all extremely similar and thus energy-independent, consistent with the previous straight-cut case; the same holds for mass distributions with nonzero minigaps. In Sec. S4, we compare the DBC to the interface states of the Bernevig-Hughes-Zhang (BHZ) model [42], which is the continuum limit of the Wu-Hu model [42, 52]. We show that these two families of 1D states are not equivalent, except for the limiting case of gapless (Δθ=π\Delta\theta=\pi) DBCs, which map to a subclass of BHZ interface states. BHZ interface states do not support massive relativistic dynamics, nor do they generally exhibit the DBC modes’ energy-independent transverse confinement.

The Jackiw–Rossi model has previously been realized in several classical-wave experiments [23, 24, 25, 26, 27, 28, 29], and although those works focused on phase singularities and the associated Jackiw–Rossi modes (also called “Dirac vortex states” in this context), we can leverage on such ideas to realize DBC modes. We choose an acoustic crystal design based on the work of Torrent and Sánchez-Dehesa [43]. Starting from a triangular lattice of pillars of equal radius r0r_0 and next-nearest-neighbor distance aa, we apply the modulation shown in Figure 2a. Each pillar is kept at a fixed center position r\mathbf{r}, with radius

Figure 2. a, Schematic of the acoustic structure, consisting of solid pillars (pink circles) surrounded by air. The pillar positions are fixed, and their radii are modulated according to Eq. (6); in this plot, we take θ=−0.35π\theta=-0.35\pi. Dashed and solid circles indicate the radii before and after modulation, respectively. b, A phase map θ(r)\theta(\mathbf{r}) with a π\pi phase dislocation along y=0y=0 (left panel), and the corresponding acoustic structure (right panel). The upper half is assigned θ=−0.35π\theta=-0.35\pi and modulation parameters R0=0.2aR_0=0.2a and ΔR=0.13d\Delta R=0.13d; the lower half has θ=0.65π\theta=0.65\pi, R0=0.19aR_0=0.19a, and ΔR=0.092d\Delta R=0.092d. c, DBC mode dispersion for Δθ=π\Delta\theta=\pi (left panel) and Δθ=2π/3\Delta\theta=2\pi/3 (right panel). Color maps show the Fourier transformed excitation spectrum obtained from experiments, while cyan dashes show the dispersion relations from FEM simulations of the acoustic structure (see Methods). The bulk bands are shaded in gray. Markers in the left and right panels indicate four frequencies between 10.7 kHz–11.3 kHz (with 0.2 kHz spacing), and three frequencies between 11.1 kHz–11.3 kHz (with 0.1 kHz spacing) used in subsequent subplots. d, Experimental setup for measuring spatially-resolved excitation spectra. The location of the acoustic source is indicated by the loudspeaker symbol. e, Measured profiles for DBC modes at 10.7 kHz (upper panel) and 11.1 kHz (lower panel). f, Semilogarithmic plot of the transverse acoustic pressure profiles, measured along a line of constant xx with the frequencies at Δθ=π\Delta\theta=\pi (left panel) and Δθ=2π/3\Delta\theta=2\pi/3 (right panel) marked in c. These experimental results are normalized to the value at y=0y=0, and agree with the frequency-independent theoretical prediction derived from the Jackiw–Rossi model (solid green line).

R(r)=R0+ΔRcos⁡[K⋅r+θ(r)],(6)R(\mathbf{r})=R_0+\Delta R\cos[\mathbf{K}\cdot\mathbf{r}+\theta(\mathbf{r})], \tag*{(6)}

where K=K+−K−\mathbf{K}=\mathbf{K}_+-\mathbf{K}_-, with K±=[±4π/3a,0]\mathbf{K}_{\pm}=[\pm4\pi/\sqrt{3}a,0] denoting the corners of the original lattice’s Brillouin zone [24]. The modulation couples the Dirac cones, and in the long wavelength limit the four mode amplitudes (two

Dirac cones, with two amplitudes each) are governed by Eq. (1) with ΔR↔∣m∣\Delta R\leftrightarrow|m| and θ(r)↔arg⁡(m)\theta(\mathbf{r})\leftrightarrow\arg(m) [26]. In our experiments, we take a=29a=29 mm. Unless stated otherwise, we set R0=0.2aR_0=0.2a (which yields a Dirac frequency of 10.4 kHz for the unmodulated lattice) and ΔR=0.13d\Delta R=0.13d (see Supplementary Information, Sec. S5).

We now implement a sample with a piecewise-constant phase distribution with Δθ=π\Delta\theta=\pi, analogous to Figure 1a. Figure 2b shows the phase distribution (left panel) and the modulated acoustic lattice (right panel). We experimentally determine the excitation spectrum by applying an acoustic source and measuring the acoustic pressure at different points along the cut (Figure 2c,Figure 2d), followed by a Fourier transform (see Methods). In Figure 2c, the color maps show the resulting excitation spectrum for this acoustic sample, as well as another similar sample with Δθ=2π/3\Delta\theta=2\pi/3. Both sets of data agree quantitatively with the dispersion curves calculated directly from the acoustic equations via FEM (see Methods), plotted as cyan dashes. These dispersion curves are approximately hyperbolic near the center of the bulk gap, in accordance with the theory, but deviate from the hyperbolic form as they approach the bulk bands. For Δθ=π\Delta\theta=\pi, the DBC minigap vanishes and agrees with the prediction in Figure 1b. For Δθ=2π/3\Delta\theta=2\pi/3, the minigap approximately matches the theoretical value of 2mb2m_b calculated from Eq. (4) using the observed bulk gap.

We then measure the spatial profiles of the DBC modes in the frequency regime where the dispersion relation is hyperbolic. The results are plotted in Figure 2e,Figure 2f (left panel), and reveal exponential decays with distance yy from the cut. In Figure 2f (left panel), we include the transverse profile predicted by the theoretical model (with m0m_0 again fitted using the observed bulk gap, and the Dirac velocity fitted to the slope of the dispersion curve in Figure 2c), which has decay length κ−1≈30\kappa^{-1}\approx30 mm. Over the measured frequency range, which spans a large portion of the bulk gap, the experimental profiles agree well with the theoretical line. We also plot the exponential decays corresponding to the case with Δθ=2π/3\Delta\theta=2\pi/3 in Figure 2f (right panel). These gapped DBC modes likewise retain nearly energy-independent transverse confinement over the investigated frequency range, and agree with the theoretical prediction.

To illustrate the flexibility of the DBC modes, we now study samples with different cut configurations. In Figure 3a, the upper panel shows a phase distribution constructed from a bounded cut (I) adjacent to an unbounded cut (II), forming a waveguide-cavity system like those found in integrated photonics [39, 40, 47]. The DBC modes on cut I turn into a set of discrete modes for a line cavity (i.e., a Fabry-Pérot-like resonator), while the DBC modes on cut II are waveguide modes that couple to the cavity. A photograph of the experimental sample is shown in the lower panel of Figure 3a. We excite cut II from the left, and monitor the acoustic power transmitted to its other end. Within the bulk gap, the measured transmittance spectrum (Figure 3b, upper panel) exhibits a dip near mid-gap that may be attributed to local symmetry-breaking along the domain wall [46] (as well as the reduction in group velocities above and below the minigap). We also observe a sequence of narrow dips, which are absorption resonances caused by critical coupling to the cavity [47]. These results are in agreement with the more ideal transmittance spectrum produced by FEM simulations (Figure 3b, middle panel), including quantitative matches for the resonance frequencies (vertical dashes). Furthermore, we numerically calculate the eigenmodes for the line cavity in the absence of cut II, and find that the eigenfrequencies closely match the absorption dips (Figure 3b, bottom panel). These simulation results also show that the eigenmodes’ quality (Q) factors (based on radiative losses at the edges of the computational cell) remain approximately constant, or increase slightly, as the eigenfrequencies approach the upper bulk band. In Figure 3c, we present experimental measurements (upper panel) and simulations (lower panel) for the field intensity profile at one of the absorption resonances. The excited line cavity mode is strongly confined to cut I, as expected.

Figure 3. Coupled waveguide-cavity system based on DBC modes. a, Upper panel: diagram θ(r)\theta(\mathbf{r}) created from branch function f(z)=(z−150)1/2(z+150)−1/2f(z)=(z-150)^{1/2}(z+150)^{-1/2}, with z=x+i4yz=x+i4y. Lower panel: the acoustic crystal modulated with the diagram θ(r)\theta(\mathbf{r}). Blue dashed lines indicate the positions of cuts I and II, which are separated by a distance of 70 mm. Scale bar: 100 mm. b, Upper panel: acoustic power transmittance to the right side of cut II. Middle panel: transmittance spectrum from FEM simulations. Lower panel: quality (Q) factors of the DBC modes on cut I. The bulk bands are shaded in gray. c, The profiles of DBC modes from experiment (upper panel) and simulations (lower panel) at the frequency marked by the star in b.

DBC modes have several other major advantages. One is their inherent flexibility: they can transmit along arbitrary paths that can be easily generated by assigning branch-cuts. Another advantage is that DBC modes support a form of Klein tunnelling [7, 9, 10] due to their effective Dirac description; in the Supplementary Information, Sec. S6, we show that the DBC mode along a straight waveguide can propagate through a potential barrier with insignificant reflection.

To demonstrate the utility of these two features, we design a spiral-shaped cut with Δθ=π\Delta\theta= \pi, as shown in Figure 4a. Keeping ∣m∣|m| constant, we use the same scheme as before to generate the acoustic structure. We locally introduce a potential barrier to the spiral path (see the shaded region in Figure 4a). The potential barrier spans the frequency range from 10.5 kHz to 11.1 kHz and is realized by adding an offset of −0.03a-0.03a to the modulation parameter R0R_0. In the Supplementary Information, Sec. S7, we show that the spiral waveguide exhibits high transmittance in a frequency range corresponding to the upper branch of DBC modes (see Figure 2c, left panel). The fabricated sample is shown in Figure 4b. Placing an acoustic source at the center of the spiral, we probe the acoustic pressure distribution at regularly-spaced points across the sample. Evidently, the acoustic waves emitted by the source propagate outward along the spiral and bypass the potential barrier (Figure 4c), with no sign of Anderson localization in the longitudinal direction. Figure 4d shows the field profiles along a line perpendicular to the spiral; the results measured at four representative frequencies exhibit confinement lengths consistent with the continuum theory (solid green line). These results show that the acoustic lattice reproduces the key features of the underlying relativistic continuum model, even along a strongly-deformed freeform cut.

Figure 4. DBC modes on a spiral. a, A modulation phase profile θ(r)\theta(r) with a spiral-shaped cut. b, Photograph of the acoustic lattice generated from this phase profile, with the cut indicated by the cyan curve. The location of the acoustic source in the experiment is indicated by the loudspeaker symbol. c, The acoustic pressure distribution measured at 11 kHz, a frequency within the potential barrier. d, Spatial profiles measured along a line transverse to the spiral (dashes in c) at four different frequencies (blue dots and red circles). In a–c, the shaded regions denote the regions with a potential barrier between 10.5 kHz and 11.1 kHz, assigned a modulation parameter of R0=0.17aR_0 = 0.17a. The solid green line is the theoretical prediction.

In the Supplementary Information, Sec. S7, we compare these results to the case of freeform waveguiding along a spiral path using conventional defect modes. The DBC modes provide significantly better mean transmittance and transmission bandwidth. Unlike a conventional interface between two globally distinct domains, a cut followed by a DBC can terminate at a branch point. This makes possible the present spiral, for which the regions on each side of the cut are globally connected. Comparing to freeform topological waveguides generated using amorphous lattices [39, 40], DBC waveguides are at least as performant, but are easier to design since only a θ(r)\theta(r) profile is required (i.e., there is no need to generate an amorphous lattice by computational close-packing).

In Supplementary Information, Sec. S8, we numerically demonstrate the extension of the DBC design to photonic crystals and highlight its potential for higher-frequency operation through geometric scaling.

In summary, we have identified a previously unrecognized family of propagating Dirac defect states confined to branch cuts in a 2D complex Dirac mass field. These Dirac branch-cut (DBC) modes are distinct from the Jackiw–Rebbi [11] and Jackiw–Rossi [18] states, which are respectively bound to domain wall and vortex defects, and from the corner and hinge states associated with higher-order topological insulators [48, 49, 50, 51]. The DBC modes propagate relativistically along the 1D cut, and can be massless or massive. The key properties predicted by the continuum theory are replicated in a discrete acoustic lattice, including relativistic (linear or hyperbolic) dispersion, energy-independent transverse confinement, Klein tunnelling, and transport along strongly deformed freeform trajectories. We have also demonstrated that the DBC modes can form line-segment cavities, and can propagate along complicated spiral trajectories. In future, we expect that the lattice modulation scheme can be further refined to bring the actual DBC modes closer to their ideal counterparts, including making the dispersion relations hyperbolic over a larger portion of the bulk gap. Thus, our results establish branch-cut defects as a platform for investigating relativistic Dirac physics on lower-dimensional defects and for developing versatile wave-based device applications.

METHODS

Solution for a straight phase dislocation

Starting from the Hamiltonian Eq. (1), we let the mass distribution be invariant along xx, i.e., m1+im2=m(y)m_1 + im_2 = m(y). We aim to derive the boundary states for a phase dislocation along the line y=0y = 0. Introducing the wavenumber kxk_x, the Schrödinger equation reduces to

[τ3(σ1kx−iσ2∂y)+m0(y)Uθ†τ1Uθ]ψ=Eψ,(7)\left[\tau_3\left(\sigma_1 k_x-i\sigma_2\partial_y\right)+m_0(y)U_\theta^\dagger\tau_1U_\theta\right]\psi=E\psi, \tag*{(7)}

where Uθ≡eiθτ3/2U_\theta\equiv e^{i\theta\tau_3/2}, and ψ(y)\psi(y) is a four-component wavefunction. We now define

ψ(y)=Uθ(y)†φ(y).(8)\psi(y)=U_{\theta(y)}^\dagger\varphi(y). \tag*{(8)}

When this is plugged into Eq. (7), the ∂y\partial_y gives rise to an additional gauge term because θ\theta depends implicitly on yy:

∂y[Uθ†φ(y)]=Uθ†[∂y−iθ′2τ3]φ(y),(9)\partial_y\left[U_\theta^\dagger\varphi(y)\right]=U_\theta^\dagger\left[\partial_y-\frac{i\theta'}{2}\tau_3\right]\varphi(y), \tag*{(9)}

where θ′≡dθ/dy\theta' \equiv d\theta/dy. Hence, Eq. (7) simplifies to

[τ3σ1kx−iτ3σ2∂y−12σ2θ′+m0(y)τ1]φ=Eφ.(10)\left[\tau_3\sigma_1k_x-i\tau_3\sigma_2\partial_y-\frac{1}{2}\sigma_2\theta'+m_0(y)\tau_1\right]\varphi=E\varphi. \tag*{(10)}

Let us now choose the mass function

m(y)=m0eivθy×{eiθ+,y>0eiθ−,y<0,(11)m(y)=m_0e^{iv_\theta y}\times \begin{cases} e^{i\theta_+}, & y>0\\ e^{i\theta_-}, & y<0, \end{cases} \tag*{(11)}

where m0∈R+m_0\in\mathbb{R}^+ and vθ∈Rv_\theta\in\mathbb{R}. This has a phase discontinuity of Δθ=θ+−θ−\Delta\theta=\theta_+-\theta_- at y=0y=0. The parameter vθv_\theta gives a uniform phase gradient, with vθ=0v_\theta=0 corresponding to the piecewise-constant mass function in Figure 1a. Hence,

θ′=vθ+Δθ δ(y).(12)\theta'=v_\theta+\Delta\theta\,\delta(y). \tag*{(12)}

We now take the ansatz

φ(y)={∣χ+⟩e−κy,y>0∣χ−⟩eκy,y<0,(13)\varphi(y)= \begin{cases} |\chi_+\rangle e^{-\kappa y}, & y>0\\ |\chi_-\rangle e^{\kappa y}, & y<0, \end{cases} \tag*{(13)}

where κ∈R+\kappa\in\mathbb{R}^+ and ∣χ±⟩|\chi_\pm\rangle are four-component spinors. Plugging this into Eq. (10), for y≠0y\ne0, gives

A±∣χ±⟩=E∣χ±⟩,(14)A_\pm|\chi_\pm\rangle=E|\chi_\pm\rangle, \tag*{(14)}
A±≡kxτ3σ1±iτ3σ2κ−vθ2σ2+m0τ1.(15)A_\pm\equiv k_x\tau_3\sigma_1\pm i\tau_3\sigma_2\kappa-\frac{v_\theta}{2}\sigma_2+m_0\tau_1. \tag*{(15)}

To handle the boundary conditions at y=0y=0, we require the original wavefunction ψ(y)\psi(y) to be continuous at y=0y=0, so the rotated wavefunction φ(y)\varphi(y) is discontinuous. Using Eq. (8), the continuity of ψ(y)\psi(y) requires Uθ+†φ(0+)=Uθ−†φ(0−)U_{\theta_+}^\dagger\varphi(0+)=U_{\theta_-}^\dagger\varphi(0-), which gives

φ(0+)=UΔθφ(0−).(16)\varphi(0+)=U_{\Delta\theta}\varphi(0-). \tag*{(16)}

Referring to Eq. (13), this is satisfied by the boundary condition ∣χ+⟩=UΔθ∣χ−⟩|\chi_+\rangle=U_{\Delta\theta}|\chi_-\rangle. We can plug this into Eq. (14) to eliminate (say) χ+\chi_+, obtaining

(UΔθ†A+UΔθ−A−)∣χ−⟩=0.(17)\left(U_{\Delta\theta}^\dagger A_+U_{\Delta\theta}-A_-\right)|\chi_-\rangle=0. \tag*{(17)}

Using Eq. (15), the terms in parentheses can be simplified to

2iκτ3σ2+m0[τ1(cos⁡Δθ−1)+τ2sin⁡Δθ].(18)2i\kappa\tau_3\sigma_2+m_0\left[\tau_1(\cos\Delta\theta-1)+\tau_2\sin\Delta\theta\right]. \tag*{(18)}

What is the spectrum of this operator? Pick an eigenstate of σ2\sigma_2, i.e., ∣χ−⟩=∣u⟩⊗∣v⟩|\chi_-\rangle=|u\rangle\otimes|v\rangle where σ2∣v⟩=±∣v⟩\sigma_2|v\rangle=\pm|v\rangle. Then Eq. (17) reduces to

((±2iκm0(e−iΔθ−1)m0(eiΔθ−1)∓2iκ))∣u⟩=0.(19)\left( \begin{pmatrix} \pm2i\kappa& m_0(e^{-i\Delta\theta}-1)\\ m_0(e^{i\Delta\theta}-1) & \mp2i\kappa \end{pmatrix} \right)|u\rangle=0. \tag*{(19)}

For either choice of ±\pm, the determinant vanishes when

4κ2−m02∣cos⁡Δθ−1−isin⁡Δθ∣2=0.(20)4\kappa^2-m_0^2\left|\cos\Delta\theta-1-i\sin\Delta\theta\right|^2=0. \tag*{(20)}

Hence, we arrive at Eq. (5). Note that the result for κ\kappa solves Eq. (19) for both choices of ±\pm, i.e., both eigenstates of σ2\sigma_2. It follows that ∣χ−⟩|\chi_-\rangle is a superposition of these two choices. To be specific, let us take the eigenstates of σ2\sigma_2 to be ∣v+⟩=2−1/2[1,i]T|v_+\rangle=2^{-1/2}[1,i]^T and ∣v−⟩=σ1∣v+⟩|v_-\rangle=\sigma_1|v_+\rangle. In each case, the non-Hermitian matrix in Eq. (19) has equal and opposite eigenvalues that are either purely real or imaginary. The zero eigenvalue condition is an exceptional point where the matrix has only one eigenvector, which we denote by ∣u+⟩|u_+\rangle or ∣u−⟩|u_-\rangle. Evidently, ∣u+⟩|u_+\rangle and ∣u−⟩|u_-\rangle depend only on Δθ\Delta\theta (and are independent of m0m_0 and kxk_x), and can be defined such that

∣u+⟩=12(sgn⁡[sin⁡(Δθ/2)]eiΔθ/2),∣u−⟩=τ3∣u+⟩.(21)|u_+\rangle=\frac{1}{\sqrt{2}} \begin{pmatrix} \operatorname{sgn}[\sin(\Delta\theta/2)]\\ e^{i\Delta\theta/2} \end{pmatrix}, \qquad |u_-\rangle=\tau_3|u_+\rangle. \tag*{(21)}

Note that ⟨u+∣u−⟩=0\langle u_+|u_-\rangle=0. With these, the eigenproblem for ∣χ−⟩|\chi_-\rangle is converted into a restricted eigenproblem (i.e., one where the solution is constrained to a subspace):

[kxτ3σ1−iτ3σ2κ−vθ2σ2+m0τ1]∣χ−⟩=E∣χ−⟩,(22)\left[k_x\tau_3\sigma_1-i\tau_3\sigma_2\kappa-\frac{v_\theta}{2}\sigma_2+m_0\tau_1\right]|\chi_-\rangle=E|\chi_-\rangle, \tag*{(22)}

where κ\kappa is given by Eq. (5) and

∣χ−⟩=α∣u+⟩∣v+⟩+β∣u−⟩∣v−⟩(23)|\chi_-\rangle=\alpha|u_+\rangle|v_+\rangle+\beta|u_-\rangle|v_-\rangle \tag*{(23)}

for some α,β∈C\alpha,\beta\in\mathbb{C}. Applying either ⟨u+∣⟨v+∣\langle u_+|\langle v_+| or ⟨u−∣⟨v−∣\langle u_-|\langle v_-| to the left side, we obtain a pair of scalar equations for α\alpha and $\beta that can be consolidated into

(mbσ3+kxσ1)(αβ)=E(αβ),(24)\left(m_b\sigma_3+k_x\sigma_1\right) \begin{pmatrix} \alpha\\ \beta \end{pmatrix} = E \begin{pmatrix} \alpha\\ \beta \end{pmatrix}, \tag*{(24)}
mb=m0sgn⁡[sin⁡Δθ]∣cos⁡(Δθ/2)∣−vθ2.(25)m_b=m_0\operatorname{sgn}[\sin\Delta\theta]\left|\cos(\Delta\theta/2)\right|-\frac{v_\theta}{2}. \tag*{(25)}

This 1D Dirac equation, with effective Dirac mass mbm_b, gives the dispersion relation Eq. (4). The spectral theorem now implies that the restricted eigenproblem Eq. (22) has two solutions for each kxk_x, which is significant as restricted eigenproblems are not guaranteed to possess solutions. This clarifies why the 4×44 \times4 Dirac equation exhibits two, not four, boundary states for each kxk_x, and retroactively justifies the ansatz Eq. (13).

Numerical Solutions

The results presented in Figure 1c–Figure 1h are obtained by numerically solving the Jackiw–Rossi equations using the FEM software package COMSOL Multiphysics with custom-defined partial differential equations expressed in weak form (see Supplementary Information). In these calculations, the cut is located along the curve y=Asin⁡[Nπ(x−z1)/(z2−z1)]y = A \sin[N \pi(x-z_1)/(z_2-z_1)], where z1=−3z_1=-3, z2=3z_2=3, and A=(z2−z1)/6A=(z_2-z_1)/6. We take N=2N=2 in Figure 1c–Figure 1e, and N=1N=1 in Figure 1f–Figure 1h.

The numerical simulations of the acoustic crystal, such as those shown in Figure 2c and Figure 3b–Figure 3c, are also done using COMSOL Multiphysics. The simulations solve the acoustic equations outside the pillars, using air density 1.18 kg m−31.18\ \mathrm{kg\,m^{-3}} and speed of sound 343 m s−1343\ \mathrm{m\,s^{-1}}; the pillar surfaces are treated as hard boundaries, which is valid due to the large impedance difference between air and the resin material. Due to the limited acoustic pressure in our experiment, nonlinearities are neglected [44].

When calculating the dispersion relations for the DBC modes (Figure 2c), we use a supercell with periodic and radiation conditions on the boundaries parallel to yy and xx, respectively. When calculating the transmitted power and mode profiles in Figure 3 and Figure 4, radiation conditions are applied on all edges of the computational cell. We also use the numerical model to calculate the band structure of the acoustic structure, and validate that the Dirac mode amplitudes are governed by the Jackiw–Rossi model as described in Ref. [43]; for details, see the Supplementary Information.

Sample Fabrication and Characterization

Each acoustic sample is composed of an upper plate, lower plate, and sealing plug. The upper plate is made of transparent acrylic with modulus 3000 MPa and density 1190 kg/m31190\ \mathrm{kg/m^3}, with holes at designated positions for inserting acoustic probes. The lower plate and sealing plug are manufactured via stereolithography, using photosensitive resin with modulus 2880 MPa and density 1100 kg/m31100\ \mathrm{kg/m^3}. The pillars are fabricated directly on the lower plate.

For the dispersion measurements of Figure 2c, we set up an acoustic source (Brüel & Kjær Type 2735) to emit a monochromatic signal through a narrow tube (radius ≈1.5\approx1.5 mm and length 100 mm) to a point on the sample’s left edge. The signal is picked up by two probe microphones (Brüel & Kjær Type 4182). The first microphone scans over 48 small holes (diameter 2 mm, step size 12.56 mm) placed along the line of the phase dislocation (parallel to the xx direction). During each measurement, the other holes are filled with sealing plugs to prevent acoustic leakage. The second microphone is fixed towards the source to pick up a reference signal. The acoustic pressure at different positions is recorded with a data collector (Brüel&Kjær Type 3560-C), and a Fourier transform is performed to obtain the kxk_x-dependence. This process is repeated at different frequencies to obtain the results in Figure 2c.

For the intensity profile measurements of Figure 3 and Figure 4, we use a monochromatic acoustic source and two probe microphones in a similar manner. The acoustic signal is picked up along a grid of holes with spacing 16.74 mm in both the xx and yy directions.

Acknowledgements

This work was supported by the Singapore National Research Foundation (NRF) under Competitive Research Program (CRP) Nos. NRF-CRP23-2019-0005, NRF-CRP23-2019-0007, and NRF-CRP29-2022-0003, and the NRF Investigatorship NRF-NRFI08-2022-0001; the Singapore Ministry of Education Singapore (MOE-T2EP50122-0019); the Singapore A*STAR Grant No. A2090b0144; the China Scholarship Council: 202406280469.

B.Z. and Y.D.C. conceived the idea. B.Z. performed the theoretical derivations with the help from Y.D.C. B.Z. performed the numerical calculations and designed the experiments. C.M. conducted the measurements with the help from X.H.Z.. B.Z. analysed the experimental data with the help from C.M. B.Z., Y.D.C. and B.L.Z. wrote the manuscript. Q.W., G.G.L. and Z.C. discussed and commented on the manuscript. Q.J.W., B.L.Z. and Y.D.C. supervised the project.

Supplementary Information for “Dirac branch-cut modes with relativistic transport”

Bofeng Zhu; Chengzhi Ma; Qiang Wang; Gui-Geng Liu; Xiuhai Zhang; Zheyu Cheng; Qi Jie Wang; Baile Zhang; Y. D. Chong

Scattering matrix calculations for straight phase dislocations

In this section, we describe a direct numerical solution for the Jackiw-Rossi model for a straight semi-infinite phase dislocation, which is helpful for checking the analytical derivations (see Figure 1b in the main text). Starting from the 4×44 \times4 Dirac equation in 2D, we replace −i∂/∂x→kx-i\partial/\partial x \to k_x and write the Hamiltonian as

H=[0kx−∂ym(y)0kx+∂y00m(y)m∗(y)00−kx+∂y0m∗(y)−kx−∂y0](26)H = \begin{bmatrix} 0 & k_x-\partial_y & m(y) & 0 \\ k_x+\partial_y & 0 & 0 & m(y) \\ m^*(y) & 0 & 0 & -k_x+\partial_y \\ 0 & m^*(y) & -k_x-\partial_y & 0 \end{bmatrix} \tag*{(26)}

For simplicity, we assume the piecewise-constant mm function shown in Figure 1a of the main text (i.e., vθ=0v_\theta=0 and θ−=0\theta_-=0). Then (S1) can be written as

H=H0(m,kx)+Λ∂∂y,(27)H=H_0(m,k_x)+\Lambda\frac{\partial}{\partial y}, \tag*{(27)}
H0(m,kx)=[0kxm0kx00mm∗00−kx0m∗−kx0],Λ=[0−100+1000000+100−10]=−Λ−1.(28)H_0(m,k_x)= \begin{bmatrix} 0 & k_x & m & 0 \\ k_x & 0 & 0 & m \\ m^* & 0 & 0 & -k_x \\ 0 & m^* & -k_x & 0 \end{bmatrix}, \qquad \Lambda= \begin{bmatrix} 0 & -1 & 0 & 0 \\ +1 & 0 & 0 & 0 \\ 0 & 0 & 0 & +1 \\ 0 & 0 & -1 & 0 \end{bmatrix} =-\Lambda^{-1}. \tag*{(28)}

First, consider y≤0y\leq0. Taking θ−=0\theta_-=0 for simplicity (i.e., m=m0m=m_0), we seek solutions for fixed EE of the form

ψ(y)=eγy∣μ⟩,Hψ(y)=Eψ(y).(29)\psi(y)=e^{\gamma y}|\mu\rangle,\qquad H\psi(y)=E\psi(y). \tag*{(29)}

Plugging this into (S2)–(S3), and doing some rearrangement, yields

Λ[H0(m0,kx)−E]∣μ⟩=γ∣μ⟩.(30)\Lambda\left[H_0(m_0,k_x)-E\right]|\mu\rangle=\gamma|\mu\rangle. \tag*{(30)}

It can be shown that this eigenproblem’s eigenvalues come in pairs {±γ1,±γ2}\{\pm\gamma_1,\pm\gamma_2\}, which can be either real (evanescent waves) or imaginary (propagating waves). We use the ++ label for evanescent waves that grow to the right (γ∈R+\gamma\in\mathbb{R}^+) as well as propagating waves with positive phase velocity (γ=ik\gamma=ik where k∈R+k\in\mathbb{R}^+); and conversely for the −- label. The corresponding eigenvectors are denoted by {∣μ1+⟩,∣μ2+⟩,∣μ1−⟩,∣μ2−⟩}\{|\mu_1^+\rangle,|\mu_2^+\rangle,|\mu_1^-\rangle,|\mu_2^-\rangle\}. Hence, the wavefunction has the form

ψ(y)=a1eγ1y∣μ1+⟩+a2eγ2y∣μ2+⟩+b1e−γ1y∣μ1−⟩+b2e−γ2y∣μ2−⟩,(y≤0),(31)\psi(y)=a_1e^{\gamma_1y}|\mu_1^+\rangle+a_2e^{\gamma_2y}|\mu_2^+\rangle+b_1e^{-\gamma_1y}|\mu_1^-\rangle+b_2e^{-\gamma_2y}|\mu_2^-\rangle,\qquad(y\leq0), \tag*{(31)}

for some coefficients a1(2),b1(2)∈Ca_{1(2)},b_{1(2)}\in\mathbb{C}.

The procedure for y≥0y\geq0 is similar, but with m=m0eiΔθm=m_0e^{i\Delta\theta}. After constructing the eigenproblem analogous to (S5), we obtain a second set of eigenvalues {±η1,±η2}\{\pm\eta_1,\pm\eta_2\}, with eigenvectors {∣ν1+⟩,∣ν2+⟩,∣ν1−⟩,∣ν2−⟩}\{|\nu_1^+\rangle,|\nu_2^+\rangle,|\nu_1^-\rangle,|\nu_2^-\rangle\}. The wavefunction has the form

ψ(y)=c1eη1y∣ν1+⟩+c2eη2y∣ν2+⟩+d1e−η1y∣ν1−⟩+d2e−η2y∣ν2−⟩,(y≥0),(32)\psi(y)=c_1e^{\eta_1y}|\nu_1^+\rangle+c_2e^{\eta_2y}|\nu_2^+\rangle+d_1e^{-\eta_1y}|\nu_1^-\rangle+d_2e^{-\eta_2y}|\nu_2^-\rangle,\qquad(y\geq0), \tag*{(32)}

for c1,2,d1,2∈Cc_{1,2},d_{1,2}\in\mathbb{C}. We now impose continuity of the four-component wavefunction, by equating (S6) to (S7) at y=0y=0:

a1∣μ1+⟩+a2∣μ2+⟩+b1∣μ1−⟩+b2∣μ2−⟩=c1∣ν1+⟩+c2∣ν2+⟩+d1∣ν1−⟩+d2∣ν2−⟩(33)a_1|\mu_1^+\rangle+a_2|\mu_2^+\rangle+b_1|\mu_1^-\rangle+b_2|\mu_2^-\rangle=c_1|\nu_1^+\rangle+c_2|\nu_2^+\rangle+d_1|\nu_1^-\rangle+d_2|\nu_2^-\rangle \tag*{(33)}
⇒a1∣μ1+⟩+a2∣μ2+⟩−d1∣ν1−⟩−d2∣ν2−⟩=−b1∣μ1−⟩−b2∣μ2−⟩+c1∣ν1+⟩+c2∣ν2+⟩(34)\Rightarrow a_1|\mu_1^+\rangle+a_2|\mu_2^+\rangle-d_1|\nu_1^-\rangle-d_2|\nu_2^-\rangle=-b_1|\mu_1^-\rangle-b_2|\mu_2^-\rangle+c_1|\nu_1^+\rangle+c_2|\nu_2^+\rangle \tag*{(34)}
⇒[∣μ1+⟩∣μ2+⟩−∣ν1−⟩−∣ν2−⟩][a1a2d1d2]=[−∣μ1−⟩−∣μ2−⟩∣ν1+⟩∣ν2+⟩][b1b2c1c2].(35)\Rightarrow \begin{bmatrix} |\mu_1^+\rangle& |\mu_2^+\rangle& -|\nu_1^-\rangle& -|\nu_2^-\rangle \end{bmatrix} \begin{bmatrix} a_1\\ a_2\\ d_1\\ d_2 \end{bmatrix} = \begin{bmatrix} -|\mu_1^-\rangle& -|\mu_2^-\rangle& |\nu_1^+\rangle& |\nu_2^+\rangle \end{bmatrix} \begin{bmatrix} b_1\\ b_2\\ c_1\\ c_2 \end{bmatrix}. \tag*{(35)}

On the left, we group together the amplitudes for the modes growing/propagating toward the domain wall, and on the right we have the modes growing/propagating away from the domain wall. Inverting the matrix on the right, we arrive at a scattering matrix relation of the form

S(kx,E,m0,Δθ)[a1a2d1d2]=[b1b2c1c2].(36)S(k_x,E,m_0,\Delta\theta) \begin{bmatrix} a_1\\ a_2\\ d_1\\ d_2 \end{bmatrix} = \begin{bmatrix} b_1\\ b_2\\ c_1\\ c_2 \end{bmatrix}. \tag*{(36)}

Now we can vary EE and/or other parameters, and monitor the eigenvalues of SS. A zero eigenvalue corresponds to the existence of an eigenstate of HH that is exponentially localized to y=0y=0. (Note that propagating waves cannot yield zero eigenvalues due to energy conservation). In Figure 1b of the main text, we plot the minimum squared eigenvalue, min⁡[eig⁡(S†S)]\min[\operatorname{eig}(S^\dagger S)], versus EE and kxk_x and/or Δθ\Delta\theta, and use the zeros of this quantity to locate the Jackiw–Rebbi-like bound states, which agree well with the previous analytical results.

Numerical solutions of bound states at 2D phase dislocation

To solve the Jackiw–Rossi equations using the finite element method (FEM), we express the equations in weak form. We start with the time-independent Schrödinger equation

\begin{bmatrix} 0 & -i\partial_x-\partial_y & m_1-im_2 & 0\\ -i\partial_x+\partial_y & 0 & 0 & m_1-im_2\\ m_1+im_2 & 0 & 0 & i\partial_x+\partial_y\\ 0 & m_1+im_2 & i\partial_x-\partial_y & 0 \end{bmatrix} \begin{bmatrix} \psi_1\\ \psi_2\\ \psi_3\\ \psi_4 \end{bmatrix} = E \begin{bmatrix} \psi_1\\ \psi_2\\ \psi_3\\ \psi_4 \end{bmatrix}, \tag*{(37)}\end{bmatrix}

where ψ1,2,3,4\psi_{1,2,3,4} are the wavefunction components and EE is the energy eigenvalue. Next, we multiply each component of this equation with a set of test functions ϕ1,2,3,4\phi_{1,2,3,4}, and integrate over some area Ω\Omega:

∫Ωϕ1⋅[−i(∂xψ2−i∂yψ2)+(m1−im2)ψ3] d2r=E∫Ωϕ1⋅ψ1 d2r(38)\int_{\Omega}\phi_1\cdot\left[-i(\partial_x\psi_2-i\partial_y\psi_2)+(m_1-im_2)\psi_3\right]\,d^2r = E\int_{\Omega}\phi_1\cdot\psi_1\,d^2r \tag*{(38)}
∫Ωϕ2⋅[−i(∂xψ1+i∂yψ1)+(m1−im2)ψ4] d2r=E∫Ωϕ2⋅ψ2 d2r(39)\int_{\Omega}\phi_2\cdot\left[-i(\partial_x\psi_1+i\partial_y\psi_1)+(m_1-im_2)\psi_4\right]\,d^2r = E\int_{\Omega}\phi_2\cdot\psi_2\,d^2r \tag*{(39)}
∫Ωϕ3⋅[(m1+im2)ψ1+i(∂xψ4−i∂yψ4)] d2r=E∫Ωϕ3⋅ψ3 d2r(40)\int_{\Omega}\phi_3\cdot\left[(m_1+im_2)\psi_1+i(\partial_x\psi_4-i\partial_y\psi_4)\right]\,d^2r = E\int_{\Omega}\phi_3\cdot\psi_3\,d^2r \tag*{(40)}
∫Ωϕ4⋅[(m1+im2)ψ2+i(∂xψ3+i∂yψ3)] d2r=E∫Ωϕ4⋅ψ4 d2r.(41)\int_{\Omega}\phi_4\cdot\left[(m_1+im_2)\psi_2+i(\partial_x\psi_3+i\partial_y\psi_3)\right]\,d^2r = E\int_{\Omega}\phi_4\cdot\psi_4\,d^2r. \tag*{(41)}

We then apply integration by parts to the derivative terms on the left side of each equation. For example, for the first term in the left hand side of (S13), we have

∫Ωϕ1⋅(−i∂xψ2) d2r=i∫Ω(∂xϕ1)⋅ψ2 d2r−i∫∂Ω(ϕ1⋅ψ2)nx dℓ,(42)\int_{\Omega}\phi_1\cdot(-i\partial_x\psi_2)\,d^2r = i\int_{\Omega}(\partial_x\phi_1)\cdot\psi_2\,d^2r -i\int_{\partial\Omega}(\phi_1\cdot\psi_2)n_x\,d\ell, \tag*{(42)}

where Ω\Omega and ∂Ω\partial\Omega are the domain and its boundary, nxn_x is the xx-component of outward normal vector on ∂Ω\partial\Omega, and dℓd\ell is the boundary element. The first term on the right of (S17) enters into our desired weak form of the PDE. The second term is a boundary term that is eliminated in the usual way (e.g., by imposing Dirichlet conditions on distant boundaries). Carrying out this procedure for all the terms in (S13)–(S16), we obtain

∫Ω[+i(∂xϕ1)ψ2+(∂yϕ1)ψ2+(m1−im2)ϕ1ψ3−Eϕ1ψ1] d2r=0(43)\int_{\Omega}\left[+i(\partial_x\phi_1)\psi_2+(\partial_y\phi_1)\psi_2+(m_1-im_2)\phi_1\psi_3-E\phi_1\psi_1\right]\,d^2r=0 \tag*{(43)}
∫Ω[+i(∂xϕ2)ψ1−(∂yϕ2)ψ1+(m1−im2)ϕ2ψ4−Eϕ2ψ2] d2r=0(44)\int_{\Omega}\left[+i(\partial_x\phi_2)\psi_1-(\partial_y\phi_2)\psi_1+(m_1-im_2)\phi_2\psi_4-E\phi_2\psi_2\right]\,d^2r=0 \tag*{(44)}
∫Ω[−i(∂xϕ3)ψ4−(∂yϕ3)ψ4+(m1+im2)ϕ3ψ1−Eϕ3ψ3] d2r=0(45)\int_{\Omega}\left[-i(\partial_x\phi_3)\psi_4-(\partial_y\phi_3)\psi_4+(m_1+im_2)\phi_3\psi_1-E\phi_3\psi_3\right]\,d^2r=0 \tag*{(45)}
∫Ω[−i(∂xϕ4)ψ3+(∂yϕ4)ψ3+(m1+im2)ϕ4ψ2−Eϕ4ψ4] d2r=0(46)\int_{\Omega}\left[-i(\partial_x\phi_4)\psi_3+(\partial_y\phi_4)\psi_3+(m_1+im_2)\phi_4\psi_2-E\phi_4\psi_4\right]\,d^2r=0 \tag*{(46)}

The integrands are the weak expressions entered into the FEM solver. The phase and amplitude of the complex mass are input as m1=∣m(r)∣cos⁡(θ(r))m_1=|m(\mathbf r)|\cos(\theta(\mathbf r)) and m2=∣m(r)∣sin⁡(θ(r))m_2=|m(\mathbf r)|\sin(\theta(\mathbf r)). The numerical solutions are obtained using Dirichlet boundary conditions, ψ1,2,3,4=0\psi_{1,2,3,4}=0, at the edges of the computational cell, far from the plotted region.

Higher-order DBC modes

In Figure 1h of the main text, we plotted the spatial profiles for several Dirac branch-cut (DBC) modes on an exemplary bounded cut. Here, we present the spatial profiles for several other modes. In Figure S1, we plot the profiles for DBC modes # 4 and # 7 at Δθ=π\Delta\theta= \pi (all other parameters are the same as in Figure 1h of the main text). It can be seen that the mode profiles are extremely similar to those of modes # 1 and # 10 from in the main text. In Figure S1, we plot all four modes’ profiles along the center line x=0x = 0. This reveals that the confinement lengths are all highly similar, consistent with the analytic prediction derived from straight cuts.

Figure S1. DBC modes on a bounded cut. a, Spatial profiles of DBC modes on a curved cut with Δθ=π\Delta\theta= \pi. All the other parameters are the same as in Figure 1h of the main text. b, Profiles along the line x=0x = 0 for the two DBC modes in a, as well as the two modes plotted in Figure 1h of the main text. All four modes have very similar confinement lengths. c–d, Mode profiles for Δθ=2π/3\Delta\theta= 2\pi/3.

In Figure S1, we plot the spatial profiles of two representative modes for the Δθ=2π/3\Delta\theta= 2\pi/3 case. The corresponding profiles along x=0x = 0 are plotted in Figure S1. Once again, the modes at very different energies have the same confinement lengths, averaging to κ≈0.87\kappa\approx0.87, while the analytic value [Eq. (5)] of the main text is κ=sin⁡(π/3)/sin⁡(π/2)≈0.86603\kappa= \sin(\pi/3)/\sin(\pi/2) \approx0.86603.

In Figure 1c–Figure 1e of the main text, we presented a case where the Jackiw–Rossi state coincides with one of the DBC modes, and has properties similar to all the other DBC modes. It is also possible to design m(r)m(\mathbf{r}) functions hosting Jackiw–Rossi modes distinct from DBC modes. For example, Figure S2 shows the phase distribution for the mass field m(r)=(z−3)1/3(z+3)2/3m(\mathbf{r}) = (z - 3)^{1/3}(z + 3)^{2/3}. This has a net winding of 2π2\pi at large distances, and therefore hosts a Jackiw–Rossi state at E=0E = 0. At the same time, the phase discontinuity at the cut is 2π/32\pi/3, so the DBC modes are gapped—i.e., their energies cannot overlap with that of the Jackiw–Rossi state, as shown in Figure S2. The spatial profiles for several modes are plotted in Figure S2–Figure S2. In this case, we find that the Jackiw–Rossi mode does not spread out along the cut, as the DBC modes do, but is localized near one of the ends.

Figure S2. Separation of Jackiw–Rossi states and DBC modes. a, Phase map for the mass field m(r)=(z−3)1/3(z+3)2/3m(\mathbf{r})=(z-3)^{1/3}(z+3)^{2/3}, using a sinusoidal half-period (with amplitude 1) for the cut. b, Eigenenergy spectrum, showing the Jackiw–Rossi state at E=0E=0 and a sequence of DBC modes with a finite gap. c–d, Spatial profiles for the Jackiw–Rossi mode, and three representative DBC modes indicated by the corresponding markers in b.

Comparison to the Bernevig-Hughes-Zhang (BHZ) model

In this section, we compare the DBC modes to the topological interface states of Bernevig-Hughes-Zhang (BHZ) type Hamiltonians [42], which constitute the continuum limit for Wu-Hu lattices [52]. We begin with a block diagonal k⋅pk \cdot p type Hamiltonian in the (p+,d+,p−,d−)(p_{+}, d_{+}, p_{-}, d_{-}) basis [52],

H′=[H+00H−],H±=[m0+Bk2A∗k±Ak∓−m0−Bk2],(47)H' = \begin{bmatrix} H_{+} & 0 \\ 0 & H_{-} \end{bmatrix}, \quad H_{\pm} = \begin{bmatrix} m_{0} + Bk^{2} & A^{*}k_{\pm} \\ Ak_{\mp} & -m_{0} - Bk^{2} \end{bmatrix}, \tag*{(47)}

where k±=kx±ikyk_{\pm}=k_x\pm ik_y and k2=kx2+ky2k^2=k_x^2+k_y^2, m0=(ϵp0−ϵd0)/2m_0=(\epsilon_p^0-\epsilon_d^0)/2 is the real Dirac mass, BB and A=viA=vi are positive real and purely imaginary numbers, respectively, and vv is the velocity. The bulk dispersion is

E±=±(m0+Bk2)2+∣A∣2k2.(48)E_{\pm}=\pm\sqrt{(m_0+Bk^2)^2+|A|^2k^2}. \tag*{(48)}

Let us take a straight interface parallel to xx so that kxk_x is conserved. In one uniform half-space, the mode bounded at y=0y=0 must have an evanescent solution Φ(x,y)∝eikxx−κy\Phi(x,y)\propto e^{ik_xx-\kappa y}, κ>0\kappa>0. By inserting replacements ky→iκk_y\to i\kappa and k2→kx2−κ2k^2\to k_x^2-\kappa^2 into the bulk dispersion Eq. (48), we have

E2=(m0+B(kx2−κ2))2+∣A∣2(kx2−κ2).(49)E^2=(m_0+B(k_x^2-\kappa^2))^2+|A|^2(k_x^2-\kappa^2). \tag*{(49)}

After solving this quadratic equation, we find that the decay constant satisfies

κ2=kx2−q±(E)(50)\kappa^2=k_x^2-q_{\pm}(E) \tag*{(50)}

where

q±(E)=−(2m0B+∣A∣2)±∣A∣4+4m0B∣A∣2+4B2E22B2.(51)q_{\pm}(E)=\frac{-(2m_0B+|A|^2)\pm\sqrt{|A|^4+4m_0B|A|^2+4B^2E^2}}{2B^2}. \tag*{(51)}

In the B=0B=0 limit, this Hamiltonian reduces to a first-order Dirac theory whose straight-interface state is equivalent to the Δθ=π\Delta\theta=\pi DBC discussed in the main text. For B≠0B\ne0, Eqs. (50)–(51) therefore give an explicit dependence of the transverse decay constant on energy and longitudinal momentum, even if the dispersion is purely linear, as illustrated in Figure S3–Figure S3.

Figure S3. Comparison of the energy dependence of transverse confinement in the Wu–Hu/BHZ model and DBC modes. a, Wu–Hu/BHZ edge state dispersion. b, Transverse profile of the field strength ∣ψ∣|\psi| for BHZ edge states into the positive-mass half-space, obtained at different energies EE and Bloch wavevector kk marked in a. c, DBC mode dispersion under a phase difference Δθ=π\Delta\theta=\pi. d, Transverse profile of the field strength ∣ψ∣|\psi| for DBC modes marked in c. e, DBC mode dispersion under a phase difference Δθ=2π/3\Delta\theta=2\pi/3. f, Transverse profile of the field strength ∣ψ∣|\psi| for DBC modes marked in e. The Dirac mass is m0=3m_0=3. Other parameters in a–b are B=1B=1 and A=2iA=2i.

Compare this to the DBC family. Moving away from the shared point (Δθ=π\Delta\theta=\pi, see Figure S3–Figure S3) opens a minigap, creating 1D massive relativistic modes (see Figure S3–Figure S3). For a straight-line phase discontinuity, the Hamiltonian can be written as

HDBC=vk∥α∥−ivα⊥∂⊥+Msgn⁡(y)β1+μβ2,(52)H_{\mathrm{DBC}}=vk_{\parallel}\alpha_{\parallel}-iv\alpha_{\perp}\partial_{\perp}+M\operatorname{sgn}(y)\beta_1+\mu\beta_2, \tag*{(52)}

with

M=m0sin⁡(Δθ/2),μ=m0cos⁡(Δθ/2),(53)M=m_0\sin(\Delta\theta/2),\qquad\mu=m_0\cos(\Delta\theta/2), \tag*{(53)}

where β1\beta_1 and β2\beta_2 anticommute with each other and with the kinetic Dirac matrices. Importantly, introducing the second mass does not generate higher-order momentum terms or destroy the Dirac algebra. Projection onto the confined subspace gives

hDBC=vk∥ηz+μηx,(54)h_{\mathrm{DBC}}=v k_{\parallel}\eta_z+\mu\eta_x, \tag*{(54)}

and hence

E2=v2k∥2+μ2,κ=∣M∣/v.(55)E^2=v^2 k_{\parallel}^2+\mu^2,\qquad\kappa=|M|/v. \tag*{(55)}

In all cases, the transverse decay constant κ=∣M∣/v\kappa=|M|/v is independent of longitudinal momentum and energy.

Acoustic structure design

The acoustic structure that we use to realize the Jackiw–Rossi model, and the consequent DBC modes, is similar to the design in Ref. 43. We begin with a triangular lattice of solid pillars surrounded by air, as shown in Figure 2a of the main text. If all pillars have equal radii, the band structure has two Dirac points at the corners of the hexagonal Brillouin zone. If the lattice is viewed in terms of an expanded unit cell with six pillars each, both Dirac points are folded to the center, Γ\Gamma, of the shrunken Brillouin zone [52]. In Figure S4, the gray lines show the acoustic band structure calculated by FEM. Near Γ\Gamma, therefore, we have four bands (two from each valley) governed by an effective Jackiw–Rossi model with m1=m2=0m_1=m_2=0.

Figure S4. Design of the acoustic structure. a, Band structure for the acoustic crystal with Kekulé modulation (red lines, modulation phase θ=−0.35π\theta=-0.35\pi) and without it (gray lines). All other parameters are the same as in Figure 2 of the main text. b Bulk band gap versus modulation phase θ\theta, with bulk bands shaded in gray. c, A phase map θ(r)\theta(\boldsymbol{r}) with a phase singularity. d, Eigenmode spectrum for the acoustic structure generated from c. The state in the bulk gap is a Jackiw–Rossi or “Dirac vortex” state. Inset: spatial profile of the in-gap state, which is localized at the center of the phase singularity and is thus interpretable as the Jackiw–Rossi mode [18]. All results shown here are obtained from FEM simulations.

We then apply a Kekulé-type modulation by varying the pillar radii according to Eq. (6) of the main text, which involves a modulation amplitude ΔR≠0\Delta R\ne0 and phase θ\theta. The modulation lifts the Dirac point degeneracy, as seen in

Fig. S4a (the red lines show the bands for θ=−0.35π\theta=-0.35\pi). Since the modulation phase parameter θ\theta is an angle variable by construction, we infer that it maps to the phase parameter arg⁡[m]\arg[m] for the complex mass field [26, 28]. In fact, this parameter mapping is not exact: the modulation phase also induces a variation in ∣m∣|m|, as seen in the variation in the bulk gap in Figure S4. In the present work, we neglect the ∣m∣|m| variation when designing our DBC waveguide structures; in the future, further optimizing the modulation scheme may enable creating more performant structures. Thus, to realize a Jackiw–Rossi model with spatially-varying arg⁡[m(r)]\arg[m(\boldsymbol{r})], we apply the corresponding spatially-varying modulation phase parameter θ(r)\theta(\boldsymbol{r}) [26, 28]. For instance, it is possible to generate a Jackiw–Rossi (or “Dirac vortex”) state with a θ(r)\theta(\boldsymbol{r}) distribution like the one shown in Figure S4, containing a phase singularity. Using this θ(r)\theta(\boldsymbol{r}), we generate the acoustic structure and calculate its eigenmodes; the results are shown in Figure S4, and indeed feature a localized eigenmode in the gap. In the main text, we proceed to study θ(r)\theta(\boldsymbol{r}) distributions containing branch-cuts, which host DBC modes.

Klein tunnelling of DBC modes

In this section, we demonstrate Klein tunnelling for the Δθ=π\Delta\theta=\pi DBC mode used in the freeform-waveguiding experiment. Klein tunnelling follows from the effective Dirac dynamics and allows transmission through a scalar potential barrier with strongly suppressed reflection. This phenomenon has been previously demonstrated for massless Dirac particles in graphene and photonic crystal [7, 9] and massive Dirac particles in time-domain synthetic lattice [10].

We first verify Klein tunnelling in the continuum model by numerically solving a set of weak form PDEs modified from the Eq. S18 in Sec. S2.

∫Ω[+i(∂xϕ1)ψ2+(∂yϕ1)ψ2+(m1−im2)ϕ1ψ3+(U−iη−E)ϕ1ψ1−s1ϕ1]d2r=0(56)\int_{\Omega}\left[+i(\partial_x\phi_1)\psi_2+(\partial_y\phi_1)\psi_2+(m_1-im_2)\phi_1\psi_3+(U-i\eta-E)\phi_1\psi_1-s_1\phi_1\right]\mathrm{d}^2r=0 \tag*{(56)}
∫Ω[+i(∂xϕ2)ψ1−(∂yϕ2)ψ1+(m1−im2)ϕ2ψ4+(U−iη−E)ϕ2ψ2−s2ϕ2]d2r=0(57)\int_{\Omega}\left[+i(\partial_x\phi_2)\psi_1-(\partial_y\phi_2)\psi_1+(m_1-im_2)\phi_2\psi_4+(U-i\eta-E)\phi_2\psi_2-s_2\phi_2\right]\mathrm{d}^2r=0 \tag*{(57)}
∫Ω[−i(∂xϕ3)ψ4−(∂yϕ3)ψ4+(m1+im2)ϕ3ψ1+(U−iη−E)ϕ3ψ3−s3ϕ3]d2r=0(58)\int_{\Omega}\left[-i(\partial_x\phi_3)\psi_4-(\partial_y\phi_3)\psi_4+(m_1+im_2)\phi_3\psi_1+(U-i\eta-E)\phi_3\psi_3-s_3\phi_3\right]\mathrm{d}^2r=0 \tag*{(58)}
∫Ω[−i(∂xϕ4)ψ3+(∂yϕ4)ψ3+(m1+im2)ϕ4ψ2+(U−iη−E)ϕ4ψ4−s4ϕ4]d2r=0(59)\int_{\Omega}\left[-i(\partial_x\phi_4)\psi_3+(\partial_y\phi_4)\psi_3+(m_1+im_2)\phi_4\psi_2+(U-i\eta-E)\phi_4\psi_4-s_4\phi_4\right]\mathrm{d}^2r=0 \tag*{(59)}

where the scalar potential U(x)U(x) is a Heaviside step function used to create the potential barrier for tunnelling (see Figure S5), η(x)=4tanh⁡((∣x∣/9)5)\eta(x)=4\tanh((|x|/9)^5) is a smooth absorption potential near the boundaries and {s1,s2,s3,s4}\{s_1,s_2,s_3,s_4\} are the excitation terms defined by

Figure S5. The Klein tunnelling of DBC mode. a, The DBC mode spectrum corresponding to regions with scalar potentials U(x)=0U(x)=0 and U(x)=1U(x)=1. b, The DBC mode transport profile at E=0.5E=0.5. The red and shaded regions denote source and absorption potential. c, The transmittance spectrum of DBC modes across the potential barrier. d, The spectrum of DBC waveguides with modulation parameters of R0=0.2aR_0=0.2a and R0=0.17aR_0=0.17a. All the other parameters remain the same. e, The DBC waveguide with potential barrier at x>0x>0. f, The tunnelling transmittance. Inset is the wave profile obtained at 11 kHz and indicated by the arrow. The shaded regions denote the frequency regions of potential barrier.

sj(x,y)=Φj(y)eikx(x−xs)e−((x−xs)/ws)2,j=1,2,3,4(60)s_j(x,y)=\Phi_j(y)e^{ik_x(x-x_s)}e^{-((x-x_s)/w_s)^2},\quad j=1,2,3,4 \tag*{(60)}

where Φ(y)=[Φ1,Φ2,Φ3,Φ4]T\Phi(y)=\begin{bmatrix}\Phi_1,\Phi_2,\Phi_3,\Phi_4\end{bmatrix}^{T} is the right-moving DBC mode profile obtained by numerically solving the weak form PDEs (see Eq. (43)) in one dimension. The source position and width are chosen as xs=−5x_s=-5 and ws=2w_s=2. The phases of regions y>0y>0 and y<0y<0 are the same as those in Figure 2b.

As shown in Figure S5, the scalar term U(x)U(x) lifts the Dirac point position by ΔE=1\Delta E=1, creating a potential barrier between E∈[0,1]E\in[0,1]. However, this potential barrier does not block the transmission of DBC mode, as shown in the waveguide profile in Figure S5. This tunnelling can also be visualized through the transmittance spectrum in Figure S5. The transmittance within the barrier-frequency region remains the same as that outside the barrier.

In the acoustic waveguide structure, we can also create a potential barrier by adjusting the modulation parameter R0R_0 in Eq. (6). For example, by introducing an offset of −0.03a-0.03a to the R0R_0, one can create a potential barrier between 10.5 kHz and 11.08 kHz, as shown in Figure S5. We further construct an acoustic waveguide in Figure S5, analogous to that in Figure S5. The modulation parameter R0R_0 for x>0x>0 is assumed as R0=0.17aR_0=0.17a, but remains the same as in the main text for x<0x<0; all other design parameters are identical to the main text. In the corresponding transmittance spectrum in Figure S5, we observe vanishing reflectance from the potential barrier. This agrees with the experimental results obtained for the spiral waveguide in Figure 4c.

Freeform waveguiding comparison with acoustic crystal defect waveguide

In this section, we compare the freeform waveguiding of DBC modes and acoustic crystal (AC) defect modes. The DBC spiral waveguide is created with the same parameters as those in Figure 4a. The conventional acoustic-crystal (AC) defect consists of a square lattice of solid pillars with lattice constant p=18.4p=18.4 mm and pillar radius r=7r=7 mm; the defect channel is formed by omitting one line of pillars and enlarging the pillars immediately adjacent to the defect to 1.2r1.2r. The spiral path is identical to that in Figure 4. As shown in Figure S6, the DBC mode can achieve a better waveguiding performance, regarding both the transmittance amplitude and bandwidths. In contrast, the AC defect waveguide can only realize waveguiding at several individual frequencies. The DBC mode has an overall uniform distribution along the waveguide, which has been shown in simulations (Figure S6) and demonstrated experimentally (Figure S6). In contrast, as shown in Figure S6, the AC defect mode profile is non-uniform along the path, consistent with stronger backscattering in this implementation.

Figure S6. The freeform waveguiding of DBC and acoustic crystal defect modes. a, Transmittance (between the yellow dashed lines in b and c) of DBC and acoustic crystal (AC) defect modes along a spiral path. The parameters for DBC waveguide is same as that in Figure 4 of the main text. The acoustic-crystal defect waveguide is created from a square lattice of solid pillars with lattice constant p=18.4p = 18.4 mm and pillar radius r=7r = 7 mm; the defect channel is formed by omitting one line of pillars and enlarging the pillars immediately adjacent to the defect to 1.2r1.2r. The brown and blue curves denote the transmittance of DBC and AC defect modes. b, The mode profile of guided AC defect mode at 11.02 kHz. c–d, Simulated and experimental mode profiles of the DBC mode at 11.30 kHz.

Photonic analogue

In this section, we present FEM simulation results showing that the present scheme for realizing DBC modes in acoustics can be transferred to photonics.

In the photonic design, the acoustic structure’s pillars are converted into air holes, while the air regions are converted into GaAs (see Figure S7). With a lattice constant of a=28a = 28 μm and GaAs refractive index of 3.74, the PhC supports a transverse magnetic (TM) band gap near 3.8 THz, as shown in Figure S7. For a straight DBC waveguide, analogous to the acoustic design of Figure 2 in the main text, this yields the dispersion relation shown in Figure S7. Similar to the acoustic case, we can use the DBC modes to form various configurations such as line cavities, as shown in Figure S7–Figure S7. Photonic realizations at higher frequency regimes can also be achieved by scaling down the dimensions of the structure.

Figure S7. Photonic DBC modes. a, A Terahertz-scale photonic structure, similar to the acoustic design but with air holes in a GaAs medium. The lattice constant is a=28a=28 μm, and the refractive index of GaAs is 3.74. The domain wall is marked by dashes. b, Dispersion relations for the transverse magnetic (TM) photonic DBC modes, calculated from FEM simulations. c, Q-factors for the eigenmode of a DBC line cavity, constructed similarly to Figure 3 of the main text, with a cavity length 360 μm. d, Field amplitude (∣Ez∣|E_z|) distributions for two of the eigenmodes in c.

References

  1. [1]B. Thaller, The Dirac equation. (Springer Science and Business Media, 2013).
  2. [2]K. Gottfried and V. F. Weisskopf, Concepts of particle physics. (Oxford University Press, 1986).
  3. [3]T. O. Wehling, A. M. Black-Schaffer and A. V. Balatsky, Dirac materials, Adv. Phys. , 63, 1 (2014).DOI
  4. [4]K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos and A. A. Firsov, Two-dimensional gas of massless Dirac fermions in graphene, Nature 438, 197 (2005).DOI
  5. [5]A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov and A. K. Geim, The electronic properties of graphene, Rev. Mod. Phys. 81, 109 (2009).DOI
  6. [6]O. Klein, Die reflexion von elektronen an einem potentialsprung nach der relativistischen dynamik von dirac, Z. Physik 53, 157 (1929).DOI
  7. [7]M. I. Katsnelson, K. S. Novoselov and A. K. Geim, Chiral tunnelling and the Klein paradox in graphene, Nat. Phys. 2, 620 (2006).DOI
  8. [8]X. Ni, D. Purtseladze, D. Smirnova, A. Slobozhanyuk, A. Alù and A. B. Khanikaev, Spin- and valley-polarized one-way Klein tunneling in photonic topological insulators, Sci. Adv. 4, eaap8802 (2018).DOI
  9. [9]X. Jiang, C. Shi, Zhenglu Li, S. Wang, Y. Wang, S. Yang, S. G. Louie and X. Zhang, Direct observation of Klein tunneling in phononic crystals, Science 370, 1447 (2020).DOI
  10. [10]L. Yu, H. Xue, R. Guo, E. A. Chan, Y. Y. Terh, C. Soci, B. Zhang and Y. D. Chong, Dirac mass induced by optical gain and loss, Nature 632, 63 (2024).
  11. [11]R. Jackiw and C. Rebbi, Solitons with fermion number 1/2, Phys. Rev. D 13, 3398 (1976).DOI
  12. [12]W. P. Su, J. R. Schrieffer, and A. J. Heeger, Solitons in Polyacetylene, Phys. Rev. Lett. 42, 1698 (1979).DOI
  13. [13]M. Z. Hasan, C. L. Kane, Colloquium: topological insulators, Rev. Mod. Phys. 82, 3045 (2010).
  14. [14]X. L. Qi, S.-C. Zhang, Topological insulators and superconductors, Rev. Mod. Phys. 83, 1057 (2011).DOI
  15. [15]T. Ozawa, H. Price, A. Amo, N. Goldman, M. Hafezi, L. Lu, M. Rechtsman, D. Schuster, J. Simon, O. Zil-berger and I. Carusotto, Topological photonics, Rev. Mod. Phys. 91, 015006 (2019).
  16. [16]G. Ma, M. Xiao, and C. T. Chan, Topological phases in acoustic and mechanical systems, Nat. Rev. Phys. 1, 281 (2019).DOI
  17. [17]Z. Chen, T. Zhang, X. Wang, J. Li, Z. Lin, F. Gao, L. Wang, Y. Liu, Q. Wang, X. Zhang, G. Ma, X. Chen, M. Lu, Y. Chen and J. Jiang, Topological phononics, arXiv:2605.20900v1, (2026).arxiv.org/abs/2605.20900
  18. [18]R. Jackiw and P. Rossi, Zero modes of the vortex-fermion system, Nucl. Phys. B 190, 681 (1981).
  19. [19]E. J. Weinberg, Index calculations for the fermion-vortex system, Phys. Rev. D 24, 2669 (1981).
  20. [20]G. E. Volovik, Fermion zero modes on vortices in chiral superconductors, Pisma Zh. Eksp. Teor. Fiz. 70, 601 (1999).DOI
  21. [21]N. Read and D. Green, Paired states of fermions in two dimensions with breaking of parity and time-reversal symmetries and the fractional quantum Hall effect, Phys. Rev. B 61, 10267 (2000).DOI
  22. [22]L. Fu and C. L. Kane, Superconducting Proximity Effect and Majorana Fermions at the Surface of a Topological Insulator, Phys. Rev. Lett. 100, 096407 (2008).
  23. [23]A. Kekulé, Untersuchungen über aromatische Verbindungen Ueber die Constitution der aromatischen Verbindungen., Ann. Chem. Pharm. 137, 129 (1866).
  24. [24]C. Hou, C. Chamon, and C. Mudry, Electron Fractionalization in Two-Dimensional Graphenelike Structures, Phys. Rev. Lett. 98, 186809 (2007).pubmed.ncbi.nlm.nih.gov/17501599
  25. [25]C. Gutiérrez, C.-J. Kim, L. Brown, T. Schiros, D. Nordlund, E. B. Lochocki, K. M. Shen, J. Park, and Abhay N. Pasupathy, Imaging chiral symmetry breaking from Kekulé bond order in graphene, Nat. Phys. 12, 950 (2016).DOI
  26. [26]P. Gao, D. Torrent, F. Cervera, P. San-Jose, J. Dehesa, and J. Christensen, Majorana-like Zero Modes in Kekulé Distorted Sonic Lattices, Phys. Rev. Lett. 123, 196601 (2019).DOI
  27. [27]J. Noh, T. Schuster, T. Iadecola, S. Huang, M. Wang, K. P. Chen, C. Chamon, and M. C. Rechtsman, Braiding photonic topological zero modes, Nat. Phys. 16, 989 (2020).DOI
  28. [28]X. Gao, L. Yang, L. Hao, L. Zhang, J. Li, F. Bo, Z. Wang and L. Lu, Dirac-vortex topological cavities, Nat. Nanotechn. 15, 1012 (2020).DOI
  29. [29]A. J. Menssen, J. Guan, D. Felce, M. J. Booth and I. A. Walmsley, Photonic Topological Mode Bound to a Vortex, Phys. Rev. Lett. 125, 117401 (2020).
  30. [30]L. Yang , G. Li , X. Gao and L. Lu, Topological-cavity surface-emitting laser, Nat. Photon. 16, 279 (2022).DOI
  31. [31]S. Han, Y. Chua, Y. Zeng, B. Zhu, C. Wang, B. Qiang, Y. Jin, Q. Wang, L. Li, A. G. Davies, E. H. Linfield, Y. Chong, B. Zhang and Q. J. Wang, Photonic Majorana quantum cascade laser with polarization-winding emission, Nat. Commun. 14, 707 (2023).DOI
  32. [32]D. Martin, and L. Ahlfors, Complex analysis, New York: McGraw-Hill, (1966).
  33. [33]J. W. Brown and R. V. Churchill Complex variables and applications, 8th ed. (McGraw-Hill 2009).
  34. [39]Q. Wang, Y. Ge, H. Sun, H. Xue, D. Jia, Y. Guan, S. Yuan, B. Zhang, and Y. D. Chong, Vortex states in an acoustic Weyl crystal with a topological lattice defect, Nat. Comm. 12, 3654 (2021).DOI
  35. [40]R. Banerjee, A. Kumar, T. C. Tan, M. Gupta, R. Jia, P. Szriftgiser, G. Ducournau, Y. Chong, and R. Singh, On-chip amorphous terahertz topological photonic interconnects, Sci. Adv. 11, eadu2526 (2025).DOI
  36. [42]B. A. Bernevig, T. L. Hughes, and S-C. Zhang, Quantum Spin Hall Effect and Topological Phase Transition in HgTe Quantum Wells, Science 314, 1757 (2006).
  37. [43]D. Torrent and J. Sánchez-Dehesa, Acoustic Analogue of Graphene: Observation of Dirac Cones in Acoustic Surface Waves, Phys. Rev. Lett. 108, 174301 (2012).pubmed.ncbi.nlm.nih.gov/22680870
  38. [44]C. He, X. Ni, H. Ge, X. Sun, Y. Chen, M. Lu, X. Liu, and Y. Chen, Acoustic topological insulator and robust one-way sound transport, Nat. Phys. 12, 1124 (2016).DOI
  39. [46]S. Xu, Y. Wang and R. Agarwal, Absence of topological protection of the interface states in Z2 photonic crystals, Phys. Rev. Lett. 131, 053802 (2023).DOI
  40. [47]K. J. Vahala, Optical microcavities, Nature 424, 839 (2003).
  41. [48]Z. Song, Z. Fang and C. Fang, (d − 2)-Dimensional Edge States of Rotation Symmetry Protected Topological States, Phys. Rev. Lett. 119, 246402 (2017).
  42. [49]W. A. Benalcazar, B. A. Bernevig, and T. L. Hughes, Electric multipole moments, topological multipole moment pumping, and chiral hinge states in crystalline insulators, Phys. Rev. B 96, 245115 (2017).DOI
  43. [50]C. W. Peterson, W. A. Benalcazar, T. L. Hughes and G. Bahl, A quantized microwave quadrupole insulator with topologically protected corner states, Nature 555, 346 (2018).DOI
  44. [51]J. Noh, W. A. Benalcazar, S. Huang, M. J. Collins, K. P. Chen, T. L. Hughes and M. C. Rechtsman, Topological protection of photonic mid-gap defect modes, Nat. Photon. 12, 408 (2018).
  45. [52]L. H. Wu and X. Hu, Scheme for Achieving a Topo- logical Photonic Crystal by Using Dielectric Material, Phys. Rev. Lett. 114, 223901 (2015).

Paper details

vlnc:W_h6r9jy73
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

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