Evidence for the chiral anomaly in the Dirac semimetal Na3Bi

See allHide authors and affiliations

Science  23 Oct 2015:
Vol. 350, Issue 6259, pp. 413-416
DOI: 10.1126/science.aac6089

Breaking chiral symmetry in a solid

Dirac semimetals have graphene-like electronic structure, albeit in three rather than two dimensions. In a magnetic field, their Dirac cones split into two halves, one supporting left-handed and the other right-handed fermions. If an electric field is applied parallel to the magnetic field, this “chiral” symmetry may break: a phenomenon called the chiral anomaly. Xiong et al. observed this anomaly in the Dirac semimetal Na3Bi (see the Perspective by Burkov). Transport measurements lead to the detection of the predicted large negative magnetoresistance, which appeared only when the two fields were nearly parallel to each other.

Science, this issue p. 413, see also p. 378


In a Dirac semimetal, each Dirac node is resolved into two Weyl nodes with opposite “handedness” or chirality. The two chiral populations do not mix. However, in parallel electric and magnetic fields (E||B), charge is predicted to flow between the Weyl nodes, leading to negative magnetoresistance. This “axial” current is the chiral (Adler-Bell-Jackiw) anomaly investigated in quantum field theory. We report the observation of a large, negative longitudinal magnetoresistance in the Dirac semimetal Na3Bi. The negative magnetoresistance is acutely sensitive to deviations of the direction of B from E and is incompatible with conventional transport. By rotating E (as well as B), we show that it is consistent with the prediction of the chiral anomaly.

The notion of handedness, or chirality, is ubiquitous in the sciences. A fundamental example occurs in quantum field theory. Massless fermions segregate into left- or right-handed groups (they spin clockwise or anticlockwise, respectively, if viewed head on). Because the two groups never mix, we say that chirality is conserved. However, mixing occurs once electromagnetic fields are switched on. This induced breaking of chiral symmetry, known as the chiral anomaly (1), was first studied in pion physics, where it causes neutral pions to decay faster than charged pions by a factor of 3 × 108 (13). In 1983, it was proposed that the anomaly may be observed in a crystal (4). This goal now seems attainable (511) in the nascent field of Dirac/Weyl semimetals (1215).

In Na3Bi (14), strong spin-orbit coupling inverts the bands derived from the Na-3s and Bi-6p orbitals, forcing them to cross at the wave vectors K± = (0, 0, ±kD), with kD ~ 0.1 Å–1 (16, 17). Because symmetry constraints forbid hybridization (13, 15), we have topologically protected Dirac states with energy E(k) = ħv|k|, where the wave vector k is measured from K± and v is the Fermi velocity (Fig. 1A). Furthermore, symmetry dictates that each Dirac node resolves into two massless Weyl nodes with chiralities χ = ±1 that preclude mixing [we calculate χ in (18)]. As discussed in (4), parallel electric and magnetic fields E||B should cause charge pumping between the Weyl nodes, observable as a negative longitudinal magnetoresistance (LMR) (611).

Fig. 1 Weyl nodes and negative longitudinal magnetoresistance in Na3Bi.

(A) Sketch of a Dirac cone centered at K+ represented as two massless Weyl nodes (slightly displaced) with distinct chiralities χ = –1 (gray cone) and +1 (yellow). (B) An intense B field widens the node separation due to the spin Zeeman energy (separation exaggerated for clarity). The Weyl states are quantized into LLs. The N = 0 LL has a linear dispersion with slopes determined by χ. The yellow and green balls represent χ. An E-field ||B generates an axial current observed as a large, negative LMR. (C) The T dependence of the resistivity ρ in B = 0, as inferred from R14,23 (I applied to contacts 1 and 4, voltage measured across contacts 2 and 3) and the Hall coefficient RH (inferred from R14,35). Inset shows the hexagonal crystal J4 (1 mm on each side and 0.5 mm thick), contact labels, and the x and y axes. RH is measured in B <2 T applied ||c. At 3 K, RH corresponds to a density n = 1.04 × 1017 cm–3. The excitation of holes in the valence band leads to a sign change in RH near 70 K and a steep decrease in ρ. (D) Longitudinal magnetoresistance ρxx(B, T) at selected T from 4.5 to 300 K measured with Embedded Image and I applied to contacts 1 and 4. The steep decrease in ρxx(B, T) with increasing B at 4.5 K reflects the onset of the axial current in the lowest LL. As T increases, occupation of higher LLs in conduction and valence bands overwhelms the axial current.

Inspired by these ideas, experimental groups have recently reported negative LMR in Bi1–xSbx (19), Cd3As2 (20, 21), ZrTe5 (22), and TaAs (23). However, because negative LMR also exists in semimetals that do not have a Dirac dispersion [e.g., CdxHg1–xTe (24) and PdCoO2 (25)], it is desirable to go beyond this observation. Here, we found that in Na3Bi the enhanced current is locked to the B vector and hence can be steered by rotating B, even for weak fields.

Crystals of Na3Bi grow as millimeter-sized, deep purple, hexagonal plates with the largest face parallel to the a-b plane (26). We annealed the crystals for 10 weeks before opening the growth tube. Crystals were contacted using silver epoxy in an argon glovebox to avoid oxidation, and were then immersed in paratone in a capsule before rapid cooling. Initial experiments in our lab (27) on samples with a large Fermi energy EF (400 mV) showed only a positive magnetoresistance (MR) with the anomalous B-linear profile reported in Cd3As2 (20).

Progress in lowering EF in Na3Bi has resulted in samples that display a nonmetallic resistivity ρ versus T profile, a low Hall density nH ~ 1 × 1017 cm–3 (Fig. 1C), and a notably large, negative LMR (Fig. 1D). We explain why the negative LMR is not from localization in (18). We estimate the Fermi wave vector kF = 0.013 Å–1 (smaller than kD by a factor of 8). Below ~10 K, the conductivity is dominated by conduction band carriers with mobility μ ~ 2600 cm2 V–1 s–1. Because the energy gap is zero, holes in the valence band are copiously excited even at low T. Above 10 K, the increased hole population leads to a steep decrease in ρ and an inversion of the sign of RH at 62 K. From the maximum in RH at 105 K, we estimate that EF ~ 3kBT ~ 30 mV. These numbers are confirmed by Shubnikov–de Haas (SdH) oscillations observed in the resistivity matrix element ρxx when B is tilted toward c (Fig. 2A). The index plot of 1/Bn (Fig. 2B), where Bn locates the SdH extrema, yields a Fermi surface (FS) cross section SF = 4.8 ± 0.3 T, which gives EF = 29 ± 2 mV, in good agreement with RH. The density inferred from SF (ne = 1.4 × 1017 cm–3) is slightly higher than nH. The deviation from the straight line in Fig. 2B is consistent with a (spin gyromagnetic) g factor of ~20, whereas g ≈ 40 has been estimated for Cd3As2 (28). The SdH oscillations imply that EF enters the N = 0 Landau level at B = 6 to 8 T.

Fig. 2 Shubnikov–de Haas oscillations in Na3Bi.

(A) SdH oscillations resolved in ρxx for several tilt angles θ (relative to Embedded Image) after subtraction of the positive B-linear MR background (vertical scale as shown). The subtraction is explained in (18). (B) Landau level index N versus 1/Bn, where Bn locates the extrema of the oscillations at selected θ (θ is defined in the inset). Uncertainties in 1/Bn are less than 10% (18). The slope at small B yields SF = 4.8 ± 0.3 T, kF = 0.013 Å–1, and EF = 29 ± 2 mV. The deviation at large B is consistent with a g* value of ~20 (18). (C) Field profiles of ρxx (inferred from R14,23) in samples J1 and J4, with B||I. The N = 0 LL is entered at B = 6 to 8 T. The slight increase for B > 5 T reflects the narrow width of the axial current. A slight misalignment of B (the uncertainty here is ±1°) allows the B-linear positive MR component to appear as a background at large B.

The Landau levels (LLs) of the Weyl states in a strong B are sketched in Fig. 1B. In addition to the LLs, the spin Zeeman energy shifts the nodes away from K+ by δkN = χgBB/(ħν), where ħ is the Planck constant divided by 2π and μB is the Bohr magneton (14, 18). For clarity, we show the shifts exaggerated. A distinguishing feature of Weyl states is that the lowest LL (N= 0) disperses linearly to the right or left depending on χ. Application of E||B leads to a charge pumping rate between the two branchesEmbedded Image (1)This is the chiral anomaly (411). The longitudinal (axial) current relaxes at a rate 1/τa ~ |M|2eB/ħv, where M is the matrix element for impurity scattering and eB/ħv is the LL degeneracy (8). Hence, the chiral conductivity σχ ~ Wτa is independent of B in the quantum limit. Equation 1 and the expression for 1/τa apply in the quantum limit at high fields (when only the lowest LL is occupied). However, we emphasize that even in weak fields when many LLs are occupied, the axial current remains observable. In the weak-B limit, Son and Spivak (9) showed thatEmbedded Image (2)with 1/τa now independent of B. As B increases, σχ grows as B2 [see also (29)] but saturates to a B-independent value in the quantum limit.

As shown in Fig. 1D, the resistivity ρxx displays a large negative LMR (Embedded Image, the current; the notch at B = 0 is discussed below). The resistance measured is R14,23 (see Fig. 1C, inset). Raising T above ~100 K suppresses the peak. In Fig. 2C, ρxx (in samples J1 and J4) falls rapidly to saturate to an almost B-independent value above 5 T (the slight upturn is a hint that the axial current is sensitive to misalignment of B at the level ±1°). A large negative LMR is anomalous in a conventional conductor, even with band anisotropy (18).

The axial current is predicted to be large when B is aligned with E. A crucial test then is the demonstration that, if E is rotated by 90°, the negative MR pattern rotates accordingly; that is, the axial current maximum is locked to B and E rather than being pinned to the crystal axes, even for weak B.

To carry out this test, we rotated B in the x-y plane while still monitoring the resistance R14,23. Figure 3A shows the curves of the resistivity ρxx versus B measured at 4.5 K at selected values of ϕ (the angle between B and Embedded Image). The MR is positive for ϕ = 90° (Embedded Image), displaying the nominal B-linear form observed in Cd3As2 (20) and Na3Bi (27) with B||c. As B is rotated toward Embedded Image (ϕ decreased), the MR curves are pulled toward negative values. At alignment (ϕ = 0), the longitudinal MR is very large and fully negative [see (18) for the unsymmetrized curves and results from sample J1].

Fig. 3 Evidence for axial current in Na3Bi.

Transport measurements were taken on sample J4 in an in-plane field B. (A) Resistivity ρxx versus B at selected field-tilt angles ϕ to the x axis (inferred from resistance R14,23; see inset). For ϕ = 90°, ρxx displays a B-linear positive MR. However, as ϕ → 0° (Embedded Image), ρxx is strongly suppressed. (B) R35,26 with E rotated by 90° relative to (A) (B makes an angle ϕ′ relative to Embedded Image; see inset). The resistance R35,26 changes from a positive MR to negative as ϕ′ → 0°. In both configurations, the negative MR appears only when B is aligned with E. (C) Conductance G ≡ 1/R35,26. In weak B, it has the B2 form predicted in Eq. 2 (9). A fit to the parabolic form gives τa0 = 40 to 60. (D) ρxx [as in (A)] extended to 35 T. Above 23 T, a knee-like kink appears at Hk. Above Hk, ρxx increases very steeply (for ϕ > 35°).

We then repeated the experiment in situ with I applied to contacts 3 and 5, so that E is rotated by 90° (the measured resistance is R35,26). Remarkably, the observed MR pattern is also rotated by 90°, even when B < 1 T. Defining the angle of B relative to Embedded Image as ϕ′, we now find that the MR is fully negative when ϕ′ = 0. The curves in Fig. 3, A and B, are nominally similar, except that ϕ = 0 and ϕ′ = 0 refer to Embedded Image and Embedded Image, respectively. As we discuss below, we identify the locking of the negative MR direction to the common direction of E||B as a signature of the chiral anomaly.

The acute sensitivity of the axial current to misalignment at large B, as hinted in Fig. 2C, is surprising. To determine the angular variation, we performed measurements in which R14,23 is measured continuously in fixed field versus tilt angle (with B either in the x-y or the x-z plane). Figure 4 displays the curves of Δσxx(B, ϕ) = σxx(B, ϕ) – σxx(B, 90°) versus ϕ (B in the x-y plane at angle ϕ to Embedded Image), with B fixed at values 0.5 →2 T (Fig. 4A) and 3 →7 T (Fig. 4B). Shown in Fig. 4, C and D, are the same measurements but now with B in the x-z plane at an angle θ to Embedded Image. In both cases, the low-field curves (B2 T) are reasonably described with cosp ϕ (or cosp θ) with p = 4. However, for B > 2 T, the angular widths narrow considerably. Hence, at large B, the axial current is observed as a strongly collimated beam in the direction selected by B and E as ϕ or θ is varied. The strong collimation has not been predicted.

Fig. 4 Angular dependence of the axial current.

The dependence is inferred from measurements of R14,23 in tilted B(θ,ϕ) in sample J4 at 4.5 K. (A and B) B lies in the x-y plane at an angle ϕ to Embedded Image (sketch in insets). The conductance enhancement Δσxx at fixed B is plotted against ϕ for values of B ≤ 2 T (A) and for 3 ≤ B ≤ 7 T (B). Fits to cos4 ϕ (dashed curves), although reasonable below 2 T, become very poor as B exceeds 2 T. The insets show the polar representation of Δσxx versus ϕ. (C and D) B is tilted in the x-z plane. As sketched in the insets, θ is the angle between B and Embedded Image. Curves of Δσxx versus θ for B = 1, 1.5, and 2 T are shown in (C); shown in (D) are curves for 3 ≤ B ≤ 7 T. The axial current is peaked when ϕ → 0 (or θ → 0) with an angular width that narrows as B increases.

The large negative MR in Fig. 3 implies a long relaxation time τa for the axial current. By fitting Eq. 2 to the parabolic profile of G = 1/R35,26 shown in Fig. 3C, we find that τa = 40 to 60 × τ0, the Drude lifetime. Despite its importance, the matrix element M in 1/τa is not well studied. There is debate on whether a large node separation 2δkN is needed to obtain a long τa [using the estimate (18) of g* ~20, we find that δkN > kF when B > 12 T]. Recently, however, it was shown (29) that the ratio τa0 can be very large (in a superlattice model) even for negligible δkN because Berry curvature effects hinder axial current relaxation and chiral symmetry is only weakly violated. This issue should be resolvable by LMR experiments.

A notable feature in the LMR profile (Fig. 1D) is the notch at B = 0, which persists to 120 K. Above 140 K, the notch expands to a V-shaped positive LMR profile. The insensitivity of this feature to the tilt angle of B implies that it is associated with the Zeeman energy. A similar feature is seen in Cd3As2 (20).

We extended measurements of R14,23 to B = 35 T. From the curves of ρxx versus B (Fig. 3D), we find a new feature at the kink field Hk ~ 23 T when Embedded Image. As B is tilted away from Embedded Image (ϕ → 55°), the feature at Hk becomes better resolved as a kink. The steep increase in ρxx above Hk suggests an electronic transition that opens a gap. However, as we decrease ϕ below 45°, Hk(ϕ) moves rapidly to above 35 T. The negative MR curve at ϕ = 0 remains unaffected by the instability up to 35 T (the small rising background is from a weak Bz due to a slight misalignment).

Within standard MR theory, the feature that is most surprising is the locking of the MR pattern to the B vector in Fig. 3, A and B. If one postulates that the narrow plume in Fig. 4 arises from anisotropies in the FS properties (v and τ0 versus k), the direction of maximum conductivity should be anchored to the crystal axes. We should not be able to rotate the resistivity tensor by orienting the weak E and B fields (this violates linear response). However, it agrees with the prediction of the chiral anomaly; the axial current peaks when E aligns with B, even for weak fields.

We believe that this locking pattern in weak B is the quintessential signature of the axial current. The experiment confirms the B2 behavior in weak B and provides a measurement of τa. The narrow angular width of the axial current may provide further insight into its properties.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S6

References (30, 31)

References and Notes

  1. See supplementary materials on Science Online.
  2. Acknowledgments: We thank B.A. Bernevig and Z. Wang for valuable discussions. Supported by Army Research Office grant ARO W911NF-11-1-0379, a MURI award for topological insulators (ARO W911NF-12-1-0461), and the Gordon and Betty Moore Foundation’s EPiQS Initiative through grant GBMF4539 (N.P.O.). The growth and characterization of crystals were performed by S.K.K., J.W.K., and R.J.C. with support from NSF grant DMR 1420541. Some experiments were performed at the National High Magnetic Field Laboratory (NHMFL), which is supported by NSF Cooperative Agreement no. DMR-1157490, the State of Florida, and the U.S. Department of Energy; we thank E. S. Choi for assistance at NHMFL.

Stay Connected to Science

Navigate This Article