betaFeedback

WorksPhysics and Astronomy

From geometric curvature to effective backgrounds: a dictionary for Dirac fermions in one, two and three spatial dimensions

Abstract

Dirac equation on curved backgrounds can be rewritten as a flat-space equation in which geometry appears as a tensorial Fermi velocity together with a set of background fields. We derive this correspondence from a single Clifford-algebra identity and follow it through one, two, and three spatial dimensions. Once the spin connection is written in terms of Ricci rotation coefficients, only two contractions can occur. The trace produces a term with the index structure of a vector potential in every dimensionality; it enters with an imaginary factor i and is precisely the term that keeps the operator Hermitian when the Fermi velocity depends on position. The totally antisymmetric contraction produces an object that depends on the number of indices available to the Levi-Civita symbol: a trace on a line, a pseudoscalar mass on a surface, and an axial vector potential in a volume. Neither contraction is invariant under local rotations of the frame, and we give the transformation laws. On a surface, the curl of the trace term can always be removed by a frame rotation, while its divergence is proportional to the Gaussian curvature. We apply the correspondence to a Gaussian bump, where the local density of states at the center falls below its flat value and obeys a closed form at low energy; to a wire, where a nonuniform lapse binds states inside the gap and a nonuniform velocity cannot; to stationary frames on a surface, where a circulating drag shifts the Dirac mass locally with either sign; and to twisted and rotating frames in a volume, where the axial term takes the form of a chiral chemical potential produced by the frame alone, or separates Weyl nodes in a rotating sample. We separate the geometric contribution from the lattice pseudogauge field that dominates the response of strained samples

Submitted to: Physica Scripta

Introduction

The low energy electronic structure of graphene is governed by a massless Dirac operator, a fact that follows from the honeycomb lattice and was already implicit in the band calculation of Wallace [1] long before the material was isolated [2, 3]. Real samples, however, are never flat. Ripples appear spontaneously at finite temperature, substrates impose corrugations, and deliberate straining bends the sheet in controlled ways. The carriers then propagate on a curved two dimensional (2D) manifold, and the question of how that geometry enters their dynamics stops being academic. Two traditions have grown around the question. One starts from the tight binding description and treats the modulation of the hopping amplitudes as the primary effect, which produces a valley dependent gauge potential whose curl behaves as a magnetic field of large magnitude [4, 5, 6]. Scanning tunneling spectroscopy of strained nanobubbles has confirmed the prediction by resolving pseudo Landau levels [7], and later work mapped the fields directly [8, 9]. The other tradition writes the Dirac equation covariantly on the deformed surface [10, 11, 12, 13] and asks what the vielbein and the spin connection do to the spectrum, an approach that has been pushed far enough to propose graphene as a setting for quantum field theory in curved spacetime [14, 15, 16]. The apparent tension between the two descriptions was resolved by de Juan, Sturla and Vozmediano, who showed that both lead to a space dependent Fermi velocity of the same form [17], a conclusion later supported by spectroscopy on corrugated monolayers [18].

What makes the covariant route attractive is not elegance but arithmetic. A curved manifold forces one to carry Christoffel symbols, a spin connection and a position dependent Clifford algebra through every calculation. If the same physics can be written as a flat space Dirac equation with a tensorial Fermi velocity and a few background fields, then the whole apparatus of flat space quantum field theory becomes available: ordinary perturbation theory, scattering techniques, functional determinants, factorization methods for bound states. Problems that look forbidding in curved coordinates, such as the spectrum of a graphene wormhole [19], become problems of a kind that has been solved many times. The translation itself, though, is usually presented for a particular geometry and in a particular dimension. The leftover piece of the spin connection is then read off case by case, and its nature tends to be asserted rather than derived. That is a gap worth closing, because three dimensional (3D) Dirac and Weyl materials [20] raise the same questions and there is no reason to expect the 2D answer to carry over unchanged.

Parts of the construction are not new, and it is fair to say so at the outset. The splitting of the spin connection into a trace and a totally antisymmetric piece is standard in the study of Dirac fields in Riemann-Cartan spaces, where only the axial part of the torsion couples to the spinor [21, 22]. In graphene, the trace term is the correction that keeps the Hamiltonian Hermitian when the Fermi velocity depends on position [17, 23]. What we have not found in the literature is a treatment that follows the antisymmetric contraction across dimensions, states which background it becomes in each case, and keeps track of how every entry of the correspondence changes with the choice of frame. Those are the aims of this work, together with a set of worked applications in which the correspondence produces definite results.

The present work derives the correspondence from a single identity of the Clifford algebra and follows it through one, two and three spatial dimensions, as illustrated in Figure 1. The outcome is that the trace of the Ricci rotation coefficients always produces a term with the structure of a vector potential, while the totally antisymmetric contraction produces an object whose character is dictated by how many indices the Levi-Civita symbol has room for: nothing on a line, a pseudoscalar mass on a surface, an axial vector potential in a volume. We have organized the rest of the manuscript as follows. Section 2 fixes the geometric setting, section 3 establishes the correspondence and its behavior under changes of frame, section 4 works out membranes, a Gaussian bump, a wire, stationary frames on a surface, and twisted and rotating frames in a volume, and section 5 delimits what the correspondence does and does not describe.

Schematics of the correspondence between the Dirac equation on a curved manifold and a flat plane with background fields

Figure 1. Schematics of the correspondence between the Dirac equation on a curved manifold and a flat plane with background fields.

Dirac fermions on a curved background

Two families of coordinates are needed throughout. The set {xμ}μ=0n\{x^\mu\}_{\mu=0}^{n} labels points on the curved manifold, while hatted indices refer to a local orthonormal frame in which the metric takes the Minkowski form ηα^β^=diag⁡(1,−1,…,−1)\eta^{\hat{\alpha}\hat{\beta}}=\operatorname{diag}(1,-1,\ldots,-1) in n+1n+1 dimensions. Since the manuscript is concerned with Dirac materials, the limiting speed of the analogue metric is the Fermi velocity from the start, and the time coordinate is x0=vFtx^0=v_{\mathrm{F}}t. The vielbein eμα^e^{\hat{\alpha}}_{\mu} carries one index of each kind and relates the two descriptions through

gμν=ηα^β^eμα^eνβ^,eμα^eβ^μ=δβ^α^.(1)g_{\mu\nu}=\eta_{\hat{\alpha}\hat{\beta}}e^{\hat{\alpha}}_{\mu}e^{\hat{\beta}}_{\nu}, \qquad e^{\hat{\alpha}}_{\mu}e^{\mu}_{\hat{\beta}}=\delta^{\hat{\alpha}}_{\hat{\beta}}. \tag*{(1)}

The correspondence derived below holds for any orthonormal frame. The individual backgrounds it produces, however, are not frame invariant, and § 3.4 gives their transformation laws. A convenient choice of frame shortens the algebra considerably in the applications of § 4.

A spinor propagating on such a background obeys

(eα^μγα^vFPμ+mvF2)Ψ=0,(2)\left(e^{\mu}_{\hat{\alpha}}\gamma^{\hat{\alpha}}v_{\mathrm{F}}P_{\mu}+mv_{\mathrm{F}}^{2}\right)\Psi=0, \tag*{(2)}

with the generalized momentum Pμ=−iℏ(∂μ+Ωμ)P_{\mu}=-i\hbar(\partial_{\mu}+\Omega_{\mu}). The piece that distinguishes equation (2) from its flat space counterpart is the affine spinor connection

Ωμ=12ωα^β^μΣα^β^,Σα^β^=14[γα^,γβ^],(3)\Omega_{\mu}=\frac{1}{2}\omega^{\hat{\alpha}\hat{\beta}}{}_{\mu}\Sigma_{\hat{\alpha}\hat{\beta}},\qquad \Sigma^{\hat{\alpha}\hat{\beta}}=\frac{1}{4}\left[\gamma^{\hat{\alpha}},\gamma^{\hat{\beta}}\right], \tag*{(3)}

built from the spin connection

ωα^β^μ=−eνβ^(∂μeα^ν−Γλμνeα^λ),(4)\omega^{\hat{\alpha}}{}_{\hat{\beta}\mu}=-e^{\nu}{}_{\hat{\beta}}\left(\partial_{\mu}e^{\hat{\alpha}}{}_{\nu}-\Gamma^{\lambda}{}_{\mu\nu}e^{\hat{\alpha}}{}_{\lambda}\right), \tag*{(4)}

where Γλμν\Gamma^{\lambda}{}_{\mu\nu} are the Christoffel symbols of the Levi-Civita connection. Metric compatibility forces ωα^β^μ=−ωβ^α^μ\omega_{\hat{\alpha}\hat{\beta}\mu}=-\omega_{\hat{\beta}\hat{\alpha}\mu}, and this antisymmetry in the frame indices is used repeatedly below.

The spin connection is an awkward object precisely because it mixes index types. Two of its labels live in the frame and one lives on the manifold, so it transforms partly as a tensor and partly as a connection. The remedy is to push the last index into the frame as well,

ωα^β^γ^≡eμγ^ωα^β^μ,(5)\omega_{\hat{\alpha}\hat{\beta}\hat{\gamma}}\equiv e^{\mu}{}_{\hat{\gamma}}\omega_{\hat{\alpha}\hat{\beta}\mu}, \tag*{(5)}

which defines the Ricci rotation coefficients [24, 25]. Because the vielbein is invertible by construction, equation (5) loses no information and can be read backwards, ωα^β^μ=eγ^μωα^β^γ^\omega_{\hat{\alpha}\hat{\beta}\mu}=e^{\hat{\gamma}}{}_{\mu}\omega_{\hat{\alpha}\hat{\beta}\hat{\gamma}}. What the change of basis buys is that the resulting coefficients are ordinary functions carrying three frame indices, antisymmetric in the first pair, and they are the only geometric data the Dirac operator responds to. Every statement in § 3 is a statement about how those functions get distributed among the effective backgrounds.

The dictionary

The claim to be established is that equation (2) can be traded for a flat space problem of the form

(vFSμα^γα^(pμ+ieAμ)+mvF2+V)Ψ=0,(6)\left(v_{\mathrm{F}}S^{\mu}{}_{\hat{\alpha}}\gamma^{\hat{\alpha}}\left(p_{\mu}+ieA_{\mu}\right)+mv_{\mathrm{F}}^{2}+\mathcal{V}\right)\Psi=0, \tag*{(6)}

written with the ordinary momentum pμ=−iℏ∂μp_{\mu}=-i\hbar\partial_{\mu}. Here vFv_{\mathrm{F}} is the Fermi velocity, Sμα^S^{\mu}{}_{\hat{\alpha}} plays the role of a strain matrix that makes that velocity tensorial and position dependent, AμA_{\mu} is a term with the index structure of a vector potential and V\mathcal{V} collects whatever is left over. With the analogue metric written in terms of vFv_{\mathrm{F}}, comparison of equations (2) and (6) identifies the strain matrix directly,

Sμα^≡eμα^.(7)S^{\mu}{}_{\hat{\alpha}}\equiv e^{\mu}{}_{\hat{\alpha}}. \tag*{(7)}

An anisotropic, position dependent Fermi velocity is therefore not an extra ingredient of the model but the inverse vielbein itself, which is consistent with what the tight binding treatment of nonuniform strain produces [17, 26] and with the direct spectroscopic evidence for a spatially varying velocity in corrugated samples [18].

All the remaining work sits in the connection term. Contracting equation (3) with the Dirac structure and using (5) to project the curved index gives

Q≡eα^ μγα^Ωμ=12ωβ^γ^α^γα^Σβ^γ^.(8)Q \equiv e_{\hat{\alpha}}^{\ \mu}\gamma^{\hat{\alpha}}\Omega_\mu= \frac{1}{2}\omega_{\hat{\beta}\hat{\gamma}\hat{\alpha}}\gamma^{\hat{\alpha}}\Sigma^{\hat{\beta}\hat{\gamma}}. \tag*{(8)}

Everything now hinges on the product γα^Σβ^γ^\gamma^{\hat{\alpha}}\Sigma^{\hat{\beta}\hat{\gamma}}, and that product is fixed by the Clifford algebra alone. It splits into a piece proportional to a single gamma matrix, with two frame indices contracted by the metric, and a piece proportional to the dual of the index triplet. In 3+13+1 dimensions the statement reads

γα^Σβ^γ^=12(ηα^β^γγ^−ηα^γ^γβ^)−i2ϵα^β^γ^σ^γσ^γ5,(9)\gamma^{\hat{\alpha}}\Sigma^{\hat{\beta}\hat{\gamma}} = \frac{1}{2}\left(\eta^{\hat{\alpha}\hat{\beta}}\gamma^{\hat{\gamma}}-\eta^{\hat{\alpha}\hat{\gamma}}\gamma^{\hat{\beta}}\right)-\frac{i}{2}\epsilon^{\hat{\alpha}\hat{\beta}\hat{\gamma}\hat{\sigma}}\gamma_{\hat{\sigma}}\gamma^{5}, \tag*{(9)}

and appendix Appendix § 8 shows how it follows from the standard product of three gamma matrices. The first term is what survives when two of the three indices are contracted, the second is what survives when all three are distinct. That division carries the whole content of the dictionary, because the spin connection is antisymmetric in its first two labels and therefore admits exactly those two independent contractions with a three index object.

It is convenient to name them once. The trace vector and its dual partner are

aγ^≡ωα^γ^α^=ηα^β^ωβ^γ^α^,bσ^≡ϵσ^α^β^γ^ωβ^γ^α^.(10)a_{\hat{\gamma}} \equiv\omega^{\hat{\alpha}}{}_{\hat{\gamma}\hat{\alpha}}=\eta^{\hat{\alpha}\hat{\beta}}\omega_{\hat{\beta}\hat{\gamma}\hat{\alpha}},\qquad b^{\hat{\sigma}}\equiv\epsilon^{\hat{\sigma}\hat{\alpha}\hat{\beta}\hat{\gamma}}\omega_{\hat{\beta}\hat{\gamma}\hat{\alpha}}. \tag*{(10)}

Neither object is exotic. The first is the contraction on which a Weyl rescaling of the spinor acts, and the second is the totally antisymmetric part of the anholonomy of the frame. With these definitions, inserting (9) into (8) and using the antisymmetry of ω\omega to fold the two metric terms into one leaves

Q=12aγ^γγ^−i4bσ^γσ^γ5,(11)Q=\frac{1}{2}a_{\hat{\gamma}}\gamma^{\hat{\gamma}}-\frac{i}{4}b^{\hat{\sigma}}\gamma_{\hat{\sigma}}\gamma^{5}, \tag*{(11)}

so that the curved space equation (2) becomes

(vFeα^ μγα^pμ−iℏvF2aγ^γγ^−ℏvF4bσ^γσ^γ5+mvF2)Ψ=0.(12)\left(v_{\mathrm{F}}e_{\hat{\alpha}}^{\ \mu}\gamma^{\hat{\alpha}}p_\mu-\frac{i\hbar v_{\mathrm{F}}}{2}a_{\hat{\gamma}}\gamma^{\hat{\gamma}}-\frac{\hbar v_{\mathrm{F}}}{4}b^{\hat{\sigma}}\gamma_{\hat{\sigma}}\gamma^{5}+mv_{\mathrm{F}}^{2}\right)\Psi=0. \tag*{(12)}

Comparison with (6) is now a matter of reading off coefficients. The vector term is the trace of the Ricci rotation coefficients pulled back to the manifold,

Aμ=−ℏ2eeγ^μaγ^,(13)\mathcal{A}_\mu=-\frac{\hbar}{2e}e^{\hat{\gamma}}{}_\mu a_{\hat{\gamma}}, \tag*{(13)}

with ee the electron charge.

Two properties of this term must be kept in view, because they decide how it should be interpreted. It enters equation (6) with a factor of ii, so it is anti-Hermitian with respect to the flat measure and is not a U(1)U(1) gauge field: it cannot be removed by a change of phase, only by a change of normalization, and its curl is not an observable field. Its role is a different one. For any static frame with e^0=vF dt\hat e^0 = v_F\,dt and time independent spatial legs ea^ie_{\hat a}^i (a^,i=1,…,n\hat a,i=1,\ldots,n), the trace reduces to a divergence,

aa^=1g∂i(gea^i),(14)a_{\hat a}=\frac{1}{\sqrt{g}}\partial_i\left(\sqrt{g}e_{\hat a}^i\right), \tag*{(14)}

with gg the determinant of the spatial metric. A nonuniform lapse adds the gradient of its logarithm, as section 4.3 shows. The substitution Ψ=g−1/4χ\Psi=g^{-1/4}\chi, which makes the measure flat, then brings the spatial part of the operator to the symmetrized form

ea^iγa^(∂i+Ωi)Ψ=g−1/4γa^12{ea^i,∂i}χ.(15)e_{\hat a}^i\gamma^{\hat a}(\partial_i+\Omega_i)\Psi=g^{-1/4}\gamma^{\hat a}\frac{1}{2}\{e_{\hat a}^i,\partial_i\}\chi. \tag*{(15)}

The trace term is therefore exactly the correction that renders the Dirac operator Hermitian when the velocity tensor depends on position, the same structure obtained from the lattice by de Juan, Sturla and Vozmediano [17, 23]. We keep the notation Aμ\mathcal{A}_\mu for it, and refer to it as the geometric trace connection, but the reader should not read it as a gauge potential.

The third background is the term carrying bσ^b^{\hat\sigma}, and its character depends on the number of spatial dimensions in a way that deserves to be spelled out case by case rather than asserted in general.

One spatial dimension

On a line the frame indices run over two values, so the totally antisymmetric contraction in (10) has nowhere to live and the dual structure collapses. With the representation γ0=σz\gamma^0=\sigma^z and γ1=iσx\gamma^1=i\sigma^x, the chirality matrix is γ5=γ0γ1=−σy\gamma^5=\gamma^0\gamma^1=-\sigma^y, which squares to the identity. An explicit check reported in the appendix shows that the two candidate forms of γα^Σβ^γ^\gamma^{\hat\alpha}\Sigma^{\hat\beta\hat\gamma} agree term by term,

12(γα^γβ^γγ^−ηα^γ^γβ^)=+12εβ^γ^γα^γ5.(16)\frac{1}{2}\left(\gamma^{\hat\alpha}\gamma^{\hat\beta}\gamma^{\hat\gamma}-\eta^{\hat\alpha\hat\gamma}\gamma^{\hat\beta}\right)=+\frac{1}{2}\varepsilon^{\hat\beta\hat\gamma}\gamma^{\hat\alpha}\gamma^5. \tag*{(16)}

The reason is elementary and worth stating, since it colors the interpretation of the result. Multiplication by γ5\gamma^5 in two spacetime dimensions exchanges the two gamma matrices, as γ0γ5=γ1\gamma^0\gamma^5=\gamma^1 and γ1γ5=γ0\gamma^1\gamma^5=\gamma^0. Vector and axial couplings are thus two names for a single object, a fact closely tied to the chiral anomaly of two dimensional models.

What survives is therefore the trace term and nothing else,

Q=12aγ^γγ^,Aμ=−ℏ2eeμγ^aγ^,V=0.(17)Q=\frac{1}{2}a_{\hat\gamma}\gamma^{\hat\gamma},\qquad\mathcal{A}_\mu=-\frac{\hbar}{2e}e_\mu^{\hat\gamma}a_{\hat\gamma},\qquad\mathcal{V}=0. \tag*{(17)}

One dimensional Dirac systems cannot generate a scalar background out of geometry alone. Whatever the deformation of the line, the fermion sees at most the trace term, and by (16) the freedom to call that term vector or axial is a matter of convention rather than a physical distinction. Section 4.3 shows that a nonuniform velocity produces no connection at all, and that the content of a one dimensional geometry lies in the lapse.

Two spatial dimensions

Surfaces are the case of direct experimental interest and also the one where the dual structure is richest, because in three spacetime dimensions the Levi-Civita symbol has exactly as many indices as the spin connection. Taking γ0=σz\gamma^0=\sigma^z, γ1=iσx\gamma^1=i\sigma^x and γ2=iσy\gamma^2=i\sigma^y, the spinor remains two component and the Clifford algebra closes on the 2D identity,

{γα^,γβ^}=2ηα^β^I2×2,[γα^,γβ^]=−2iεα^β^γ^γγ^.(18)\left\{\gamma^{\hat{\alpha}},\gamma^{\hat{\beta}}\right\}=2\eta^{\hat{\alpha}\hat{\beta}}I_{2\times2},\qquad\left[\gamma^{\hat{\alpha}},\gamma^{\hat{\beta}}\right]=-2i\varepsilon^{\hat{\alpha}\hat{\beta}\hat{\gamma}}\gamma_{\hat{\gamma}}. \tag*{(18)}

The product that governs the dictionary becomes

γα^Σβ^γ^=12(ηβ^γ^γα^−ηα^γ^γβ^)−i2εα^β^γ^I2×2,(19)\gamma^{\hat{\alpha}}\Sigma^{\hat{\beta}\hat{\gamma}}=\frac{1}{2}\left(\eta^{\hat{\beta}\hat{\gamma}}\gamma^{\hat{\alpha}}-\eta^{\hat{\alpha}\hat{\gamma}}\gamma^{\hat{\beta}}\right)-\frac{i}{2}\varepsilon^{\hat{\alpha}\hat{\beta}\hat{\gamma}}I_{2\times2}, \tag*{(19)}

so the dual piece is no longer accompanied by a gamma matrix. It is a multiple of the identity in spinor space. Writing b≡εα^β^γ^ωα^β^γ^b\equiv\varepsilon^{\hat{\alpha}\hat{\beta}\hat{\gamma}}\omega_{\hat{\alpha}\hat{\beta}\hat{\gamma}} for the single independent totally antisymmetric contraction, equation (8) gives

Q=12aγ^γγ^−i4bI2×2,(20)Q=\frac{1}{2}a_{\hat{\gamma}}\gamma^{\hat{\gamma}}-\frac{i}{4}bI_{2\times2}, \tag*{(20)}

and the curved equation reduces to

(vFeα^μγα^pμ−iℏvF2aγ^γγ^−ℏvF4b+mvF2)Ψ=0.(21)\left(v_{\mathrm F}e_{\hat{\alpha}}^{\mu}\gamma^{\hat{\alpha}}p_{\mu}-\frac{i\hbar v_{\mathrm F}}{2}a_{\hat{\gamma}}\gamma^{\hat{\gamma}}-\frac{\hbar v_{\mathrm F}}{4}b+mv_{\mathrm F}^{2}\right)\Psi=0. \tag*{(21)}

Two spatial dimensions is the only case in which the triad of effective backgrounds announced in equation (6) is realized with a spinor scalar in the third slot,

Sα^μ=eα^μ,Aμ=−ℏ2ee~γ^μaγ^,V=−ℏvF4b.(22)S_{\hat{\alpha}}^{\mu}=e_{\hat{\alpha}}^{\mu},\qquad\mathcal{A}_{\mu}=-\frac{\hbar}{2e}\tilde{e}^{\mu}_{\hat{\gamma}}a^{\hat{\gamma}},\qquad\mathcal{V}=-\frac{\hbar v_{\mathrm F}}{4}b. \tag*{(22)}

The position of V\mathcal{V} in equation (21) shows what kind of object it is. It sits next to mvF2mv_{\mathrm F}^{2}, so it shifts the Dirac mass locally, mvF2→mvF2−ℏvFb/4mv_{\mathrm F}^{2}\to mv_{\mathrm F}^{2}-\hbar v_{\mathrm F}b/4, rather than adding a potential energy to the Hamiltonian. With two component spinors in 2+12+1 dimensions the mass term is odd under parity and under time reversal, and bb, which carries an uncontracted Levi-Civita symbol, shares those properties; this is the sense in which we call it a pseudoscalar. Because the three frame indices must be distinct, one of them is the time label, and bb vanishes for every static frame in which e0^=vF dte^{\hat{0}}=v_{\mathrm F}\,dt and the spatial legs do not depend on time. Section 4.4 shows that frames which mix time and space, such as those of a rotating or dragged medium, do produce it.

Three spatial dimensions

In four spacetime dimensions the Weyl representation is the natural choice,

γ^μ^=[0σμ^σˉμ^0],σμ^=(I2×2,σx,σy,σz),σˉμ^=(I2×2,−σx,−σy,−σz),(23)\hat{\gamma}^{\hat{\mu}} = \begin{bmatrix} 0 & \sigma^{\hat{\mu}} \\ \bar{\sigma}^{\hat{\mu}} & 0 \end{bmatrix}, \qquad \sigma^{\hat{\mu}} = (I_{2\times2},\sigma^x,\sigma^y,\sigma^z), \qquad \bar{\sigma}^{\hat{\mu}} = (I_{2\times2},-\sigma^x,-\sigma^y,-\sigma^z), \tag*{(23)}

for which γ5=iγ0γ1γ2γ3=diag⁡(−I2×2,I2×2)\gamma^5=i\gamma^0\gamma^1\gamma^2\gamma^3=\operatorname{diag}(-I_{2\times2},I_{2\times2}). This is the case treated by equations (9) to (12), and the dictionary reads

Sαμ=eαμ,Aμ=−ℏ2eeαμγ^α,V=−ℏvF4bγθ^σ^γγ5.(24)S^\mu_\alpha=e^\mu_\alpha,\qquad \mathcal{A}_\mu=-\frac{\hbar}{2e}e^\mu_\alpha\hat{\gamma}^\alpha,\qquad \mathcal{V}=-\frac{\hbar v_F}{4}b^{\hat{\theta}}_\gamma\hat{\sigma}^\gamma\gamma^5. \tag*{(24)}

The third entry is no longer a scalar. It carries a free frame index and a γ5\gamma^5, which makes it an axial vector potential coupling to the chiral current rather than a mass term. Calling it a scalar background, as the 2D intuition invites one to do, would misrepresent both its transformation properties and its physical effect.

The two kinds of component act differently, and both have a clear meaning once the equation is written in Hamiltonian form. The time component enters the Hamiltonian as a term proportional to (ℏvF/4)b0^γ5(\hbar v_F/4)b^{\hat{0}}\gamma^5, a chiral chemical potential that shifts the nodes of opposite chirality in opposite directions in energy. The spatial components enter as a term proportional to (ℏvF/4)bk^Sk(\hbar v_F/4)b^{\hat{k}}S^k, with Sk≡γ5γ0γk^S^k\equiv\gamma^5\gamma^0\gamma^{\hat{k}} the spin matrix, and shift the nodes of opposite chirality in opposite directions in momentum. In 3D Dirac and Weyl materials [20] these are the two familiar axial backgrounds, an energy separation and a momentum separation of the nodes, and neither is a mass. Whether a given bθ^b^{\hat{\theta}} is physical, however, depends on the frame, as § 3.4 explains and § 4.5 illustrates with a twisted frame in flat space.

The pattern across the three cases is set by how much room the Levi-Civita symbol has. On a line the dual contraction degenerates into the trace, on a surface it saturates the spinor indices and produces a pseudoscalar mass, and in a volume it leaves one index free and produces an axial vector. The trace contraction, by contrast, always yields the term of equation (13). That term is the only entry of the dictionary that can appear in every dimension, but it is not the pseudogauge field that experiments on strained samples detect [7, 9]; that field has a lattice origin, as § 5 discusses.

Behavior under local frame rotations

The orthonormal frame is defined only up to a local Lorentz transformation eμα^→Λ^β^α^(x)eμβ^e^{\hat{\alpha}}_\mu\to\hat{\Lambda}^{\hat{\alpha}}_{\hat{\beta}}(x)e^{\hat{\beta}}_\mu. The Dirac operator is covariant under it, provided the spinor transforms with the corresponding element of the spin group, but the spin connection picks up an inhomogeneous term, ω→ΛωΛ−1+Λ dΛ−1\omega\to\Lambda\omega\Lambda^{-1}+\Lambda\,\mathrm{d}\Lambda^{-1}. That term feeds into both contractions of (10), so aγ5a\gamma_5, bθ^b^{\hat{\theta}}, Aμ\mathcal{A}_\mu and its curl all change with the frame. The correspondence (6) holds in every frame, but its entries are frame dependent, and only combinations that are insensitive to Λ\Lambda can be given a physical meaning.

On a static surface the relevant transformations are rotations of the spatial legs by an angle θ(x,y)\theta(x,y), e^1→cos⁡θ e^1+sin⁡θ e^2\hat{e}^{1} \to\cos\theta\,\hat{e}^{1}+\sin\theta\,\hat{e}^{2} and e^2→−sin⁡θ e^1+cos⁡θ e^2\hat{e}^{2} \to-\sin\theta\,\hat{e}^{1}+\cos\theta\,\hat{e}^{2}. A direct computation gives

δAi=−ℏ2eg εijgjk∂kθ,δFxy=ℏ2eg Δgθ,(25)\delta A_i=-\frac{\hbar}{2e}\sqrt{g}\,\varepsilon_{ij}g^{jk}\partial_k\theta,\qquad\delta\mathcal{F}_{xy}=\frac{\hbar}{2e}\sqrt{g}\,\Delta_g\theta, \tag*{(25)}

with εxy=1\varepsilon_{xy}=1, Fxy=∂xAy−∂yAx\mathcal{F}_{xy}=\partial_x A_y-\partial_y A_x, gijg_{ij} the spatial metric and Δg\Delta_g its Laplace-Beltrami operator. In a flat plane with a rotated frame this is the result Fxy=(ℏ/2e)∇2θ\mathcal{F}_{xy}=(\hbar/2e)\nabla^2\theta, which is not zero although the plane is flat. The divergence, by contrast, does not change, and it carries the curvature,

∇iAi=ℏ2eK,(26)\nabla_i A^i=\frac{\hbar}{2e}K, \tag*{(26)}

where indices are raised with gijg^{ij} and KK is the Gaussian curvature of the surface. Equations (25) and (26) say that the trace connection on a surface is, up to a factor, the Hodge dual of the spin connection. Its curl is pure frame choice, and its divergence is geometry.

In 2+12+1 dimensions the pseudoscalar bb is also affected by rotations that depend on time, and in 3+13+1 dimensions bσ^b^{\hat{\sigma}} is affected by rotations that depend on position. § 4.5 gives an explicit case in which a frame twisted along one axis produces a constant bθ^b^{\hat{\theta}} in flat space.

Applications

Membranes and the choice of frame

Two spatial dimensions is where the dictionary meets the laboratory, so it deserves an explicit deformation rather than a statement about a general metric. A static membrane is described by the frame

e^0=vF dt,e^1=λ1(x,y) dx,e^2=λ2(x,y) dy,(27)\hat{e}^{0}=v_{\mathrm{F}}\,dt,\qquad\hat{e}^{1}=\lambda_1(x,y)\,dx,\qquad\hat{e}^{2}=\lambda_2(x,y)\,dy, \tag*{(27)}

with positive functions λ1\lambda_1 and λ2\lambda_2 that encode how much the sample is stretched along each direction. Building the Christoffel symbols, feeding them into (4), projecting with (5) and contracting as in (10) is a mechanical but lengthy computation whose outcome is compact,

Aμdxμ=−ℏ2e(∂xln⁡λ2 dx+∂yln⁡λ1 dy),b=0.(28)\mathcal{A}_{\mu}dx^{\mu}=-\frac{\hbar}{2e}\left(\partial_x\ln\lambda_2\,dx+\partial_y\ln\lambda_1\,dy\right),\qquad b=0. \tag*{(28)}

The crossing of labels is not a misprint. The component along xx is governed by the stretching along yy and the other way round, which is already a hint that isotropic and anisotropic deformations will behave differently. Consider first a conformal profile, λ1=λ2=eφ\lambda_1=\lambda_2=e^{\varphi}. Equation (28) collapses to Aμ=−(ℏ/2e)∂μφ\mathcal{A}_{\mu}=-(\hbar/2e)\partial_{\mu}\varphi, a pure gradient. Because of the factor of ii that accompanies the minimal coupling in (6), removing a gradient is not a change of phase but a change of normalization. The substitution Ψ=e−φ/2χ\Psi=e^{-\varphi/2}\chi eliminates the connection term, but it leaves the velocity and the measure untouched. Writing H0=−iℏvFγ0γa∂^aH_0=-i\hbar v_F\gamma^0\gamma^a\hat{\partial}_a for the flat massless operator, the spatial part of the equation becomes

−iℏvFe−φγaγ^a(∂a+Ωa)Ψ=e−φ/2(e−φH0)χ,(29)-i\hbar v_F e^{-\varphi}\gamma^a\hat{\gamma}^a(\partial_a+\Omega_a)\Psi=e^{-\varphi/2}\left(e^{-\varphi}H_0\right)\chi, \tag*{(29)}

while the norm becomes ∫g∣Ψ∣2 d2x=∫eφ∣χ∣2 d2x\int\sqrt{g}\lvert\Psi\rvert^2\,\mathrm{d}^2x=\int e^{\varphi}\lvert\chi\rvert^2\,\mathrm{d}^2x. The problem for χ\chi is a Dirac operator with the isotropic, position dependent velocity vFe−φv_Fe^{-\varphi} and a nonuniform measure. Equivalently, in terms of χ′=eφ/2χ\chi'=e^{\varphi/2}\chi the operator takes the symmetrized form (15) with flat measure. The geometric background of a conformally deformed membrane is therefore a modulation of the Fermi velocity and nothing else. This is the same structure exploited in the program that turns negatively curved graphene into a table top realization of quantum field theory in curved space [14, 27, 16].

Whether the connection term can be removed by a rescaling of the spinor in the frame (27) is decided by the curl of (28). Writing Fij=∂iAj−∂jAiF_{ij}=\partial_iA_j-\partial_jA_i and inserting the potential gives the closed form

Fij=−ℏ2eεij∂x∂yln⁡(λ1λ2),(30)F_{ij}=-\frac{\hbar}{2e}\varepsilon_{ij}\partial_x\partial_y\ln\left(\frac{\lambda_1}{\lambda_2}\right), \tag*{(30)}

with εxy=−εyx=1\varepsilon_{xy}=-\varepsilon_{yx}=1. Written in this way the antisymmetry in ijij is explicit. The mixed derivative is symmetric, but the logarithm is odd under the relabeling x↔yx\leftrightarrow y, which also exchanges λ1\lambda_1 and λ2\lambda_2. The curl vanishes if and only if the ratio λ1/λ2\lambda_1/\lambda_2 factorizes into a function of xx times a function of yy. Within the diagonal frame (27) and the laboratory coordinates in which it is written, isotropy is a sufficient condition and anisotropy that varies along both directions obstructs the rescaling.

This statement must not be read as a property of the membrane. By equation (25), a rotation of the frame with g Δgθ=−(2e/ℏ)Fxy\sqrt{g}\,\Delta_g\theta=-(2e/\hbar)F_{xy} removes the curl, and such an angle always exists locally. In the rotated frame the trace connection becomes a gradient, A=−(ℏ/2e) dfA=-(\hbar/2e)\,\mathrm{d}f, and equation (26) fixes the function through

Δgf=−K.(31)\Delta_g f=-K. \tag*{(31)}

The same conclusion follows from the existence of isothermal coordinates, in which every 2D metric takes the conformal form [28]: there λ1=λ2\lambda_1=\lambda_2, the curl vanishes, and f=φf=\varphi solves (31). For a static surface, after a suitable rotation of the frame, the connection term can always be traded for a rescaling of the spinor whose logarithm is sourced by the Gaussian curvature, and the geometry reaches the spectrum only through the velocity tensor and the measure. Figure 2 shows the curl (30) for a smooth Gaussian conformal factor in the diagonal frame with λ1=eφ\lambda_1=e^\varphi, λ2=1\lambda_2=1, and its vanishing for λ1=λ2\lambda_1=\lambda_2.

Figure 2. Conformal factor φ=εexp⁡[−(x2+y2)/(2σ2)]\varphi= \varepsilon\exp[-(x^2 + y^2)/(2\sigma^2)] with ε=0.3\varepsilon= 0.3. Panel (a) shows φ\varphi, which enters the frame (27); it is a property of the frame and not a height profile. Panel (b) shows the curl (30) for the anisotropic choice λ1=eφ\lambda_1 = e^{\varphi}, λ2=1\lambda_2 = 1, Fxy=−(ℏ/2e)∂x∂yφ\mathcal{F}_{xy} = -(\hbar/2e)\partial_x\partial_y\varphi, in units of ℏ/(eσ2)\hbar/(e\sigma^2); its maximum modulus is ε/(2e)\varepsilon/(2e), with ee the base of the natural logarithm. The quadrupolar pattern follows the mixed derivative of φ\varphi. For the isotropic choice λ1=λ2=eφ\lambda_1 = \lambda_2 = e^{\varphi} the curl vanishes at every point. Both statements refer to the diagonal frame; by equation (25) a local rotation of the frame changes panel (b).

Shear behaves in the same spirit but with a different anatomy. Replacing the orthogonal frame by e1=dx+u(x,y) dye^1=\mathrm{d}x+u(x,y)\,\mathrm{d}y and e2=dye^2=\mathrm{d}y leaves a single nonvanishing component, Ay=(ℏ/2e)∂xuA_y=(\hbar/2e)\partial_xu, and a curl Fxy=(ℏ/2e)∂x2uF_{xy}=(\hbar/2e)\partial_x^2u controlled by the curvature of the shear profile along the direction it displaces. A surface of revolution, by contrast, with λ1=1\lambda_1=1 and λ2=f(x)\lambda_2=f(x), gives Ax=−(ℏ/2e)∂xln⁡fA_x=-(\hbar/2e)\partial_x\ln f and no curl, as equation (30) anticipates.

A Gaussian bump

The conformal factor of the previous section is a property of a frame. A deformed sample is described instead by its height h(x,y)h(x,y) above the reference plane, and the metric it induces on the carriers is gij=δij+∂ih ∂jhg_{ij} = \delta_{ij} + \partial_i h\,\partial_j h. We take the Gaussian bump

h(x,y)=h0exp⁡(−x2+y22σ2),(32)h(x,y) = h_0 \exp\left(-\frac{x^2+y^2}{2\sigma^2}\right), \tag*{(32)}

which is the standard idealization of the nanobubbles and corrugations of the experiments cited in the introduction, and whose Gaussian curvature at the apex is K(0)=h02/σ4K(0) = h_0^2/\sigma^4. Panel (a) of Figure 3 shows KK for h0=σh_0 = \sigma: positive on the cap, weakly negative on the surrounding ring where the bump meets the plane.

Figure 3. Gaussian bump (32). (a) Gaussian curvature Kσ2K\sigma^2 for h0=σh_0=\sigma; the black line marks K=0K=0. (b) Curl FxyF_{xy} of the trace connection, in units of ℏ/(eσ2)\hbar/(e\sigma^2), in the symmetric frame for an elliptical bump with widths σ\sigma along xx and 2σ2\sigma along yy and h0=σh_0=\sigma; for the circular bump the same quantity vanishes identically. (c) Local density of states at the center of the circular bump divided by its flat value, from the radial problem (33); the low energy limit is given by equation (34). Only the geometric contribution is included.

Two frames are natural here. The symmetric frame e^i=(g1/2)ai\hat{e}^i = (g^{1/2})_{ai}, built from the matrix square root of gijg_{ij}, stays close to the Cartesian axes. The polar frame e^1=λ(r) dr\hat{e}^1 = \lambda(r)\,dr, e^2=r dϕ\hat{e}^2 = r\,d\phi, with λ=[1+h′(r)2]1/2\lambda= [1+h'(r)^2]^{1/2}, is diagonal in polar coordinates and gives Frϕ=0\mathcal{F}_{r\phi} = 0 by equation (30), since λ1/λ2=λ(r)/r\lambda_1/\lambda_2 = \lambda(r)/r does not depend on ϕ\phi. For the circular bump the symmetric frame also gives Fxy=0\mathcal{F}_{xy} = 0 at every point, because it differs from the polar frame by a rotation through the polar angle, which is harmonic away from the apex. This is in line with the perturbative finding of de Juan, Cortijo and Vozmediano that the geometric gauge term does not contribute at first order for axially symmetric bumps [29]. An elliptical bump with widths σ\sigma and 2σ2\sigma breaks that coincidence; panel (b) shows the resulting curl in the symmetric frame, a quadrupole of maximum modulus 0.039 ℏ/(eσ2)0.039\,\hbar/(e\sigma^2) for h0=σh_0 = \sigma, which a further rotation of the frame would remove.

The frame independent content of the bump is in the spectrum. For the circular bump the problem separates in the polar frame. Writing the proper radial distance as ρ\rho and the radius of the circle of constant ρ\rho as R(ρ)=rR(\rho)=r, the spinor RΨ\sqrt{R}\Psi, with components referred to the polar frame, obeys in each channel of total angular momentum j∈Z+1/2j\in\mathbb{Z}+1/2,

Hj=ℏvF(−iσx∂ρ+σyjR(ρ))(33)H_j=\hbar v_{\mathrm{F}}\left(-i\sigma^x\partial_\rho+\sigma^y\frac{j}{R(\rho)}\right) \tag*{(33)}

which reduces to the flat problem when R=ρR=\rho. Only the channels j=±1/2j=\pm1/2 reach the center. Integrating the regular solution outwards and comparing its asymptotic amplitude with the flat one gives the local density of states at the apex relative to its flat value, without a finite box or an artificial broadening. Panel (c) of Figure 3 shows the result. The density is reduced at every energy, and the reduction disappears once the wavelength is shorter than the width of the bump. At low energy, where the wavelength exceeds the bump, the regular solution is amplified across the cap by a factor that can be computed in closed form, and

ν(0,E→0)νflat(E)=exp⁡[−∫0∞(λ(r)−1)drr](34)\frac{\nu(0,E\to0)}{\nu_{\mathrm{flat}}(E)}=\exp\left[-\int_0^\infty\left(\lambda(r)-1\right)\frac{\mathrm{d}r}{r}\right] \tag*{(34)}

which gives 0.940, 0.790 and 0.605 for h0/σ=0.5h_0/\sigma=0.5, 1 and 1.5 and reproduces the numerical curves at the lowest energies. The sign agrees with the perturbative result of Ref. [29], which found the corrected density always below the flat one. The calculation keeps only the geometric part of the problem; the lattice pseudogauge field of a real bubble, which has a threefold pattern and produces pseudo Landau levels [7, 4], has to be added on top of it.

One spatial dimension: velocity and lapse

A line has no intrinsic curvature, so the interest of the 1D case lies in how the frame is chosen in time and space. Consider first a static frame e^0=v0 dt\hat{e}^{0}=v_0\,\mathrm{d}t, e^1=v0 dx/v(x)\hat{e}^{1}=v_0\,\mathrm{d}x/v(x) with v(x)>0v(x)>0, which describes a wire with a nonuniform Fermi velocity v(x)v(x). The spin connection vanishes identically: in one spatial dimension there is no rotation of a static frame. The Dirac operator is −iℏγ0γ1v(x)∂x-i\hbar\gamma^0\gamma^1v(x)\partial_x, which is Hermitian with respect to the proper measure v0 dx/vv_0\,\mathrm{d}x/v, and the coordinate s=∫dx v0/v(x)s=\int\mathrm{d}x\,v_0/v(x) maps it onto a flat operator with constant mass. The spectrum of H2H^2 is therefore bounded below by the square of the gap Δ=mv02\Delta=mv_0^2, and no modulation of the velocity alone, however strong, can bind a state inside the gap. What remains of the velocity profile is a reparametrization of the wire, whose only global trace is the total time of flight ∫dx/v\int\mathrm{d}x/v.

The lapse behaves differently. For e^0=N(x) v0 dt\hat{e}^{0}=N(x)\,v_0\,\mathrm{d}t and e^1=dx\hat{e}^{1}=\mathrm{d}x the trace connection is the gradient Ax=−(ℏ/2e) ∂xln⁡N\mathcal{A}_x=-(\hbar/2e)\,\partial_x\ln N, removed by Ψ=N−1/2χ\Psi=N^{-1/2}\chi. Multiplying the equation by NN, dividing by v0v_0 and passing to ds=dx/N\mathrm{d}s=\mathrm{d}x/N, with x0=v0tx^0=v_0t,

(−iℏγ0∂0−iℏγ1∂s+mv0N(s))χ=0(35)\left(-i\hbar\gamma^0\partial_0-i\hbar\gamma^1\partial_s+mv_0N(s)\right)\chi=0 \tag*{(35)}

a flat Dirac equation with a position dependent mass N(s)ΔN(s)\Delta. A lapse modulates the velocity and the gap together, keeping their ratio fixed, and it is that joint modulation that carries the geometric content of a 1D background. Physically, it describes a 1D Dirac system in which the overall energy scale of the hopping varies along the wire, for instance a dimerized chain in which the mean hopping and its dimerization are rescaled by the same factor N(x)N(x), so that velocity and gap vary together.

Equation (35) is the setting in which the factorization methods mentioned in the introduction apply directly. With W(s)=N(s) mv0/ℏW(s)=N(s)\,mv_0/\hbar, the square of the Hamiltonian splits into the supersymmetric partners −∂s2+W2∓W′-\partial_s^2+W^2\mp W' [30], and a dip in NN binds states inside the gap. Figure 4 shows the in-gap spectrum for N(x)=1−ηexp⁡[−x2/(2w2)]N(x)=1-\eta\exp[-x^2/(2w^2)] with w=2ℏ/(mv0)w=2\hbar/(mv_0). One bound state pair appears as soon as η>0\eta>0, and more follow as the well deepens; for η=0.6\eta=0.6 the bound energies are ∣E∣/Δ=0.510|E|/\Delta=0.510, 0.7100.710, 0.8720.872 and 0.9850.985. The lowest level approaches (1−η)Δ(1-\eta)\Delta, the minimum of the local gap, as the well grows. Since W>0W>0 everywhere, no zero mode appears and the spectrum is symmetric. For the velocity-only profile with the same η\eta the same computation finds no state inside the gap, as it must.

Stationary frames on a surface

A frame whose time leg has spatial components describes a medium that drags the carriers, the analogue of a gravitomagnetic background. For

e^0=dx0+Wi(x,y) dxi,e^1=dx,e^2=dy,(36)\hat{e}^{0}=dx^{0}+W_i(x,y)\,dx^i,\qquad\hat{e}^{1}=dx,\qquad\hat{e}^{2}=dy, \tag*{(36)}

Figure 4. A massive Dirac fermion on a line. (a) The profile 1−ηexp⁡[−x2/(2w2)]1-\eta\exp[-x^2/(2w^2)] with η=0.6\eta=0.6 and w=2ℏ/(mv0)w=2\hbar/(mv_0), used either as a velocity v(x)/v0v(x)/v_0 or as a lapse N(x)N(x). (b) In-gap energies for the lapse, from equation (35), as a function of η\eta. The shaded bands are the continuum and the dashed lines mark ±(1−η)Δ\pm(1-\eta)\Delta. A velocity profile alone produces no in-gap states.

with x0=vFtx^0=v_{\mathrm{F}}t and dimensionless WiW_i, the trace vanishes and the pseudoscalar is the curl of the drag,

aγ^=0,b=∂xWy−∂yWx.(37)a_{\hat{\gamma}}=0,\qquad b=\partial_x W_y-\partial_y W_x. \tag*{(37)}

By equation (22) the drag shifts the Dirac mass locally by −ℏvFb/4-\hbar v_{\mathrm{F}}b/4. For a localized circulating drag of strength w0w_0 and width σ\sigma, Wx=−w0(y/σ)e−r2/(2σ2)W_x=-w_0(y/\sigma)e^{-r^2/(2\sigma^2)} and Wy=w0(x/σ)e−r2/(2σ2)W_y=w_0(x/\sigma)e^{-r^2/(2\sigma^2)}, the pseudoscalar is

b=w0σe−r2/(2σ2)(2−r2σ2),(38)b=\frac{w_0}{\sigma}e^{-r^2/(2\sigma^2)}\left(2-\frac{r^2}{\sigma^2}\right), \tag*{(38)}

positive in the core and negative on the surrounding ring, with a change of sign at r=2σr=\sqrt{2}\sigma, as Figure 5 shows. The local mass changes sign near the core when mvF2<ℏvFw0/(2σ)mv_{\mathrm{F}}^2<\hbar v_{\mathrm{F}}w_0/(2\sigma). Since the surfaces of constant tt must remain spacelike, which requires ∣W∣<1|W|<1 and hence w0<e1/2w_0<e^{1/2}, a sign change is possible only for σ≲ℏ/(mvF)\sigma\lesssim\hbar/(mv_{\mathrm{F}}). The region of inverted mass is then no larger than the Compton length, and the familiar picture of states bound to a mass domain wall cannot be taken over without a separate calculation.

Figure 5. Pseudoscalar bb for the circulating drag of equation (38). (a) Map of bb in units of ω0/σ\omega_0/\sigma, with arrows indicating WiW_i and the black line marking b=0b=0. (b) Radial profile; the sign changes at r=2 σr=\sqrt{2}\,\sigma. By equation (22) the local Dirac mass is shifted by −ℏvFb/4-\hbar v_F b/4.

A rigidly rotating frame belongs to the same family. In coordinates that rotate with angular velocity Ω\Omega, e^1=dx+(Ω/vF)y dx0\hat{e}^{1}=dx+(\Omega/v_{\mathrm{F}})y\,dx^0 and e^2=dy−(Ω/vF)x dx0\hat{e}^{2}=dy-(\Omega/v_{\mathrm{F}})x\,dx^0, one finds aγ^=0a_{\hat{\gamma}}=0 and the uniform value b=−2Ω/vFb=-2\Omega/v_{\mathrm{F}}. The mass shifts by ℏΩ/2\hbar\Omega/2. This uniform bb is physical only because the material itself rotates with the frame, and together with the orbital term −ΩLz-\Omega L_z the shift completes the coupling −ΩJz-\Omega J_z of a spin one half particle to rotation [31, 32]. In graphene the two spinor components refer to the sublattice pseudospin rather than to the electron spin, so the identification presumes that the pseudospin carries angular momentum, as argued by Mecklenburg and Regan [33]. The size of the effect deserves a plain statement: for a mechanical rotation at $\Omega/2\pi=10\ \mathrm{kHz} the shift is of order 10−11 eV10^{-11}\,\mathrm{eV}. The point of the example is structural. A static bend of a sample produces no pseudoscalar at all, and any genuine bb on a surface signals a frame that mixes time and space or, as discussed in § 5, a connection with torsion.

Twisted and rotating frames in a volume

Two frames in flat 3D space make the frame dependence of the axial term explicit and separate its artificial and physical parts. In the first, the spatial legs rotate about the zz axis as one moves along it,

e^1=cos⁡(τz) dx+sin⁡(τz) dy,e^2=−sin⁡(τz) dx+cos⁡(τz) dy,e^3=dz,(39)\hat{e}^{1}=\cos(\tau z)\,dx+\sin(\tau z)\,dy,\qquad\hat{e}^{2}=-\sin(\tau z)\,dx+\cos(\tau z)\,dy,\qquad\hat{e}^{3}=dz, \tag*{(39)}

with e^0=dx0\hat{e}^{0}=dx^{0}. The space is flat, yet aγ^=0a_{\hat{\gamma}}=0 and bσ^=(2τ,0,0,0)b^{\hat{\sigma}}=(2\tau,0,0,0), a term with the form of a uniform chiral chemical potential of magnitude ℏvFτ/2\hbar v_F\tau/2. It is entirely an artifact of the frame: the spin rotation exp⁡(−iτzS3/2)\exp(-i\tau zS^{3}/2) that undoes the twist removes it. This example answers in the negative the question of whether bσ^b^{\hat{\sigma}} is observable by itself. A twist of the frame is not a twist of the material, and a lattice twist enters the Dirac operator through a different route, as an axial pseudogauge field produced by the modulation of the node separation [34, 35, 36].

In the second frame the coordinates rotate rigidly about zz,

e^1=dx+ΩvFy dx0,e^2=dy−ΩvFx dx0,e^3=dz.\hat{e}^{1}=dx+\frac{\Omega}{v_F}y\,dx^{0},\qquad\hat{e}^{2}=dy-\frac{\Omega}{v_F}x\,dx^{0},\qquad\hat{e}^{3}=dz.

Again aγ^=0a_{\hat{\gamma}}=0, and now bσ^=(0,0,0,2Ω/vF)b^{\hat{\sigma}}=(0,0,0,2\Omega/v_F), a spatial axial vector along the rotation axis. In Hamiltonian form it is the spin part ℏΩS3/2\hbar\Omega S^{3}/2 of the coupling −Ω⋅J-\boldsymbol{\Omega}\cdot\mathbf{J} between angular momentum and rotation [32], which for a Weyl node of chirality χ=±1\chi=\pm1 shifts the node along kzk_z by −χΩ/(2vF)-\chi\Omega/(2v_F), up to an overall sign fixed by conventions. Here the frame is tied to a physical observer, the rotating sample. The orbital part −ΩLz-\Omega L_z is also present, so the statement concerns the spin part of the coupling rather than a conserved momentum. In a Weyl material S3S^3 acts on the band pseudospin, and its identification with angular momentum carries the same caveat as in graphene. It is also the microscopic ingredient behind the chiral vortical effect, the axial current that a rotating chiral system carries along its axis [37, 38]. For Ω/2π=10 kHz\Omega/2\pi= 10\ \mathrm{kHz} and vF=3×105 m/sv_{\mathrm{F}} = 3 \times10^{5}\ \mathrm{m/s} the shift is about 0.1 m−10.1\ \mathrm{m}^{-1}, far below any lattice scale; the value of the example is again that it shows which entry of the dictionary rotation feeds, and that the answer is an axial vector rather than a mass.

Scope of the correspondence

Nothing in sections 2 and 3 involves an expansion. The map from curvature to effective backgrounds is an identity between two ways of writing the same operator, valid for any invertible vielbein, and in that sense it does not degrade as the deformation grows. Approximations enter one step earlier, when a physical displacement field of a real sample is translated into a frame, and one step later, when the flat space problem is actually solved. The second point matters for geometries such as the graphene wormhole, where the deformation near the throat is in no sense small [19]: the dictionary still applies, but treating the resulting backgrounds perturbatively does not.

A more delicate issue concerns what the trace connection of equation (13) actually is. It descends from the spin connection, it is anti-Hermitian, and section 3 shows that it is the term that keeps the operator Hermitian when the velocity depends on position; its curl depends on the frame and is not observable. It is not the pseudogauge field that dominates the pseudomagnetic response measured in strained samples. That one has a lattice origin, arises from the modulation of the hopping amplitudes and enters the continuum description as a valley dependent vector potential that is Hermitian and produces pseudo Landau levels [4, 5, 23]. The two contributions coexist and have been reconciled explicitly within the covariant formalism [17, 6], where the modulation of the hopping amplitudes produces both a correction to the velocity tensor, which can be absorbed into the frame of equation (7), and a valley dependent vector potential, which has no counterpart in Ωμ\Omega_\mu. Reporting the geometric piece as if it accounted for the fields inferred from pseudo Landau levels [7, 8, 9] would therefore misattribute the effect. What the geometry does leave in the spectrum is a modulation of the density of states governed by the velocity tensor and, on a surface, by the Gaussian curvature, as the bump of § 4.2 illustrates.

The silence of the third background in static geometries invites a second remark. Its source, the totally antisymmetric part of the Ricci rotation coefficients, is exactly the quantity that a connection with torsion makes generically nonzero [21, 22], and torsion is the natural language for dislocations and disclinations in a continuum description of defective media [39, 40]. Topological defects in graphene have long been treated along those lines [12, 11], so the appearance of a pseudoscalar mass in a 2D Dirac system should be read as evidence of torsion or of a frame that mixes time and space rather than as evidence of curvature. The present work assumes a Levi-Civita connection throughout, and extending the dictionary to Riemann-Cartan backgrounds is the obvious next step. Chirality deserves a closing caution. In three spatial dimensions the third entry of the dictionary couples through γσγ^5\gamma^\sigma\hat{\gamma}^{5}, so it acts with opposite sign on the two chiral components of the spinor. Any statement transported from the 2D case, where that entry is a single function multiplying the identity and shifting the mass, has to be reexamined before being applied to a Weyl or Dirac semimetal. The same warning applies in reverse to one spatial dimension, where equation (16) makes the vector and axial descriptions indistinguishable and no amount of algebra will separate effects that the 2D intuition keeps apart.

Concluding remarks

Reading the geometry of a Dirac system through the Ricci rotation coefficients turns a question about manifolds into a question about two contractions, and the answer to that question changes with dimension in a way that is easy to overlook when one works exclusively on surfaces. The practical consequence is that results migrating from graphene to 3D Dirac and Weyl materials should be checked term by term. A pseudoscalar shift of the Dirac mass and an axial potential coupling to the chiral current do different things to a spectrum, and only one of them exists in each setting.

The applications show where the correspondence pays off and where it needs care. Its entries depend on the frame, so the statements that survive are about invariant combinations: on a surface, the divergence of the trace connection, which is the Gaussian curvature, and the velocity tensor that governs the density of states of a bump; on a line, the lapse, which modulates gap and velocity together and can bind states that a velocity profile cannot; on a surface or in a volume, the pseudoscalar and axial terms of frames tied to physical observers, which reproduce the coupling between spin and rotation. Twisted frames in flat space show how easily an axial term can be produced by the frame alone.

Several directions open from here. Extending the construction to Riemann-Cartan backgrounds is the most immediate, since torsion is precisely what switches on the contraction that stays silent for static membranes, and dislocations and disclinations in a defective lattice are described exactly in that language. A second direction is computational. Because the correspondence is an identity rather than an expansion, it can be fed to numerical scattering codes written for flat space, with the deformation entering through the frame; geometries with a throat or a neck, where no small parameter exists, are the natural test cases. A third is the systematic use of factorization methods on the resulting one dimensional radial problems, of which equation (35) is the simplest instance.

What would sharpen the picture most is an experiment sensitive to the distinction drawn in § 5. The lattice gauge field and the geometric terms enter the same Dirac operator through different doors. A deformation engineered so that the curvature-driven modulation of the density of states can be measured separately from the pseudo Landau levels of the lattice field would decide how much of the observed response the geometry is entitled to claim.

Acknowledgments

AR acknowledges support from Consejo de la Investigación Científica of the Universidad Michoacana de San Nicolás de Hidalgo under grant 18371.

Gamma matrix conventions and the identities used in the text

Signs are fixed once and for all by ε01=ε012=ε0123=+1\varepsilon^{01}=\varepsilon^{012}=\varepsilon^{0123}=+1 and by the mostly minus metric. In every dimensionality the gamma matrices satisfy {γα^,γβ^}=2ηα^β^I\{\gamma^{\hat{\alpha}},\gamma^{\hat{\beta}}\}=2\eta^{\hat{\alpha}\hat{\beta}}I, where the identity is 2×22\times2 for n=1n=1 and n=2n=2 and 4×44\times4 for n=3n=3.

On a line, with γ0=σz\gamma^0=\sigma^z and γ1=iσx\gamma^1=i\sigma^x, the chirality matrix is γ5=γ0γ1=−σy\gamma^5=\gamma^0\gamma^1=-\sigma^y, which satisfies (γ5)2=I(\gamma^5)^2=I and anticommutes with both gamma matrices. The product iγ0γ1i\gamma^0\gamma^1 squares to −I-I and is not a chirality matrix in this signature. The single independent commutator is [γ0,γ1]=2γ5[\gamma^0,\gamma^1]=2\gamma^5, equivalently 4Σα^β^=2εα^β^γ54\Sigma^{\hat{\alpha}\hat{\beta}}=2\varepsilon^{\hat{\alpha}\hat{\beta}}\gamma^5. Evaluating the components one by one confirms equation (16), and also that in 1+11+1 dimensions the product γα^Σβ^γ^\gamma^{\hat{\alpha}}\Sigma^{\hat{\beta}\hat{\gamma}} equals the metric term alone.

On a surface the third matrix is γ2=iσy\gamma^2=i\sigma^y. The spinor keeps two components, so the anticommutator closes on I2×2I_{2\times2}, and the commutators are those of equation (18). Equation (19) then follows by writing Σβ^γ^=−i2εβ^γ^γλ^\Sigma^{\hat{\beta}\hat{\gamma}}=-\frac{i}{2}\varepsilon^{\hat{\beta}\hat{\gamma}}\gamma_{\hat{\lambda}} and using the Clifford algebra once.

In four spacetime dimensions the required input is the product of three gamma matrices,

γμ^γν^γρ^=ημ^ν^γρ^+ην^ρ^γμ^−ημ^ρ^γν^−iεσ^μ^ν^ρ^γσ^γ5.(40)\gamma^{\hat{\mu}}\gamma^{\hat{\nu}}\gamma^{\hat{\rho}}=\eta^{\hat{\mu}\hat{\nu}}\gamma^{\hat{\rho}}+\eta^{\hat{\nu}\hat{\rho}}\gamma^{\hat{\mu}}-\eta^{\hat{\mu}\hat{\rho}}\gamma^{\hat{\nu}}-i\varepsilon^{\hat{\sigma}\hat{\mu}\hat{\nu}\hat{\rho}}\gamma_{\hat{\sigma}}\gamma^5 . \tag*{(40)}

Antisymmetrizing the last two indices and dividing by four turns (A.1) into the master identity (9), since the metric terms combine in pairs while the Levi-Civita term doubles. Relation (A.1) is equivalent to the self duality of σμ^ν^=i2[γμ^,γν^]\sigma^{\hat{\mu}\hat{\nu}}=\frac{i}{2}[\gamma^{\hat{\mu}},\gamma^{\hat{\nu}}] under multiplication by γ5\gamma^5.

Every identity in this appendix, the reductions (11) and (20), equations (14), (15), (25), (26) and the backgrounds quoted in § 4 were checked symbolically in the matrix representations quoted above, by building the spin connection from the frame and comparing QQ computed directly with the gamma matrices against the closed forms.

References

  1. [1]Wallace P R 1947 Phys. Rev. 71 622
  2. [2]Novoselov K S, Geim A K, Morozov S V, Jiang D, Katsnelson M I, Grigorieva I V, Dubonos S V and Firsov A A 2005 Nature 438 197
  3. [3]Castro Neto A H, Guinea F, Peres N M R, Novoselov K S and Geim A K 2009 Rev. Mod. Phys. 81 109
  4. [4]Vozmediano M A H, Katsnelson M I and Guinea F 2010 Phys. Rep. 496 109–148
  5. [5]Guinea F, Katsnelson M I and Geim A K 2010 Nature Phys. 6 30–33
  6. [6]Amorim B, Cortijo A, de Juan F, Grushin A G, Guinea F, Gutiérrez-Rubio A, Ochoa H, Parente V, Roldán R, San-Jose P, Schiefele J, Sturla M and Vozmediano M A H 2016 Phys. Rep. 617 1–54
  7. [7]Levy N, Burke S A, Meaker K L, Panlasigui M, Zettl A, Guinea F, Castro Neto A H and Crommie M F 2010 Science 329 544–547
  8. [8]Meng L, He W Y, Zheng H, Liu M, Yan H, Yan W, Chu Z D, Bai K, Dou R F, Zhang Y, Liu Z, Nie J C and He L 2013 Phys. Rev. B 87 205405
  9. [9]Jiang Y, Mao J, Duan J, Lai X, Watanabe K, Taniguchi T and Andrei E Y 2017 Nano Lett. 17 2839–2843
  10. [10]Kavalo A R, Kostov I K and Sedraky an A G 1986 Phys. Lett. B 175 331–334
  11. [11]Cortijo A and Vozmediano M A H 2007 Europhys. Lett. 77 47002
  12. [12]Cortijo A and Vozmediano M A H 2007 Nucl. Phys. B 763 293
  13. [13]Mesaros A, Sadri D and Zaanen J 2010 Phys. Rev. B 82 073405
  14. [14]Iorio A 2011 Ann. Phys. 326 1334–1353
  15. [15]Iorio A and Lambiase G 2012 Phys. Lett. B 716 334–337
  16. [16]Gallerati A 2021 J. Phys.: Condens. Matter 33 135501
  17. [17]de Juan F, Sturla M and Vozmediano M A H 2012 Phys. Rev. Lett. 108 227205
  18. [18]Yan H, Chu Z D, Yan W, Liu M, Meng L, Yang M, Fan Y, Wang J, Dou R F, Zhang Y, Liu Z, Nie J C and He L 2013 Phys. Rev. B 87 075405
  19. [19]González J and Herrero J 2010 Nucl. Phys. B 825 426–443
  20. [20]Armitage N P, Mele E J and Vishwanath A 2018 Rev. Mod. Phys. 90 015001
  21. [21]Hehl F W and Datta B K 1971 J. Math. Phys. 12 1334–1339
  22. [22]Shapiro I L 2002 Phys. Rep. 357 113
  23. [23]de Juan F, Mañes J L and Vozmediano M A H 2013 Phys. Rev. B 87 165131
  24. [24]Nakahara M 2003 Geometry, Topology and Physics 2nd ed (Bristol: IOP Publishing)
  25. [25]Birrell N D and Davies P C W 1984 Quantum Fields in Curved Space (Cambridge: Cambridge University Press)
  26. [26]Oliva-Leyva M and Naumis G G 2015 Phys. Lett. A 379 2645–2651
  27. [27]Iorio A and Lambiase G 2014 Phys. Rev. D 90 025006
  28. [28]Chern S S 1955 Proc. Amer. Math. Soc. 6 771–782
  29. [29]de Juan F, Cortijo A and Vozmediano M A H 2007 Phys. Rev. B 76 165409
  30. [30]Cooper F, Khare A and Sukhatme U 1995 Phys. Rep. 251 267–385
  31. [31]Mashhoon B 1988 Phys. Rev. Lett. 61 2639–2642
  32. [32]Hehl F W and Ni W T 1990 Phys. Rev. D 42 2045–2048
  33. [33]Mecklenburg M and Regan B C 2011 Phys. Rev. Lett. 106 116803
  34. [34]Cortijo A, Ferreirós Y, Landsteiner K and Vozmediano M A H 2015 Phys. Rev. Lett. 115 177202
  35. [35]Pikulın D I, Chen A and Franz M 2016 Phys. Rev. X 6 041021
  36. [36]Ilan R, Grushin A G and Pikulin D I 2020 Nat. Rev. Phys. 2 29–41
  37. [37]Vilenkin A 1979 Phys. Rev. D 20 1807–1812
  38. [38]Landsteiner K, Megías E and Pena-Benitez F 2011 Phys. Rev. Lett. 107 021601
  39. [39]Katanaev M O and Volovich I V 1992 Ann. Phys. 216 1–28
  40. [40]Katanaev M O 2005 Phys. Usp. 48 675–701

Paper details

vlnc:W_x9k3mv7t
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

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