Abstract
Fifty years ago Walter Kohn speculated that a zerogap semiconductor might be unstable against the spontaneous generation of excitons–electron–hole pairs bound together by Coulomb attraction. The reconstructed ground state would then open a gap breaking the symmetry of the underlying lattice, a genuine consequence of electronic correlations. Here we show that this excitonic insulator is realized in zerogap carbon nanotubes by performing firstprinciples calculations through manybody perturbation theory as well as quantum Monte Carlo. The excitonic order modulates the charge between the two carbon sublattices opening an experimentally observable gap, which scales as the inverse of the tube radius and weakly depends on the axial magnetic field. Our findings call into question the Luttinger liquid paradigm for nanotubes and provide tests to experimentally discriminate between excitonic and Mott insulators.
Introduction
Long ago Walter Kohn speculated that gray tin—a zerogap semiconductor—could be unstable against the tendency of mutually attracting electrons and holes to form bound pairs, the excitons^{1}. Being neutral bosoniclike particles, the excitons would spontaneously occupy the same macroscopic wave function, resulting in a reconstructed insulating ground state with a broken symmetry inherited from the exciton character^{2,3,4,5}. This excitonic insulator (EI) would share intriguing similarities with the Bardeen–Cooper–Schrieffer (BCS) superconductor ground state^{4,6,7,8,9,10,11}, the excitons—akin to Cooper pairs—forming only below a critical temperature and collectively enforcing a quasiparticle gap. The EI was intensively sought after in systems as diverse as mixedvalence semiconductors and semimetals^{12,13}, transition metal chalcogenides^{14,15}, photoexcited semiconductors at quasi equilibrium^{16,17}, unconventional ferroelectrics^{18}, and, noticeably, semiconductor bilayers in the presence of a strong magnetic field that quenches the kinetic energy of electrons^{19,20}. Other candidates include electron–hole bilayers^{21,22}, graphene^{23,24,25,26}, and related twodimensional structures^{27,28,29,30,31,32,33}, where the underscreened Coulomb interactions might reach the critical coupling strength stabilizing the EI. Overall, the observation of the EI remains elusive.
Carbon nanotubes, which are rolled cylinders of graphene whose lowenergy electrons are massless particles^{34,35}, exhibit strong excitonic effects, due to ineffective dielectric screening and enhanced interactions resulting from one dimensionality^{36,37,38,39}. As single tubes can be suspended to suppress the effects of disorder and screening by the nearby substrate or gates^{40,41,42}, the field lines of Coulomb attraction between electron and hole mainly lie unscreened in the vacuum (Fig. 1a). Consequently, the interaction is truly long ranged and in principle—even for zero gap—able of binding electron–hole pairs close to the Dirac point in momentum space (Fig. 1b). If the binding energy is finite, then the ground state is unstable against the spontaneous generation of excitons having negative excitation energy, ε _{ u } < 0. This is the analog of the Cooper instability that heralds the transition to the superconducting state—the excitons replacing the Cooper pairs.
Here we focus on the armchair family of zerogap carbon nanotubes, because symmetry prevents their gap from opening as an effect of curvature or bending^{43}. In this paper we show that armchair tubes are predicted to be EIs by firstprinciples calculations. The problem is challenging, because the key quantities controlling this phenomenon—energy band differences and exciton binding energies—involve manybody corrections beyond density functional theory (DFT) that are of the order of a few meV, which is close to the limits of currently available methods. In turn, such weak exciton binding reflects in the extreme spatial extension of the exciton wave function, hence its localization in reciprocal space requires very high sampling accuracy. To address these problems, we perform stateoftheart manybody perturbation theory calculations within the GW and Bethe–Salpeter schemes^{44}. We find that bound excitons exist in the (3,3) tube with finite negative excitation energies. We then perform unbiased quantum Monte Carlo simulations^{45} to prove that the reconstructed ground state is the EI, its signature being the broken symmetry between inequivalent carbon sublattices—reminescent of the exciton polarization. Finally, to investigate the trend with the size of the system, which is not yet in reach of firstprinciples calculations, we introduce an effectivemass model, which shows that both EI gap and critical temperature fall in the meV range and scale with the inverse of the tube radius. Our findings are in contrast with the widespread belief that electrons in undoped armchair tubes form a Mott insulator—a strongly correlated Luttinger liquid^{46,47,48,49,50,51,52}. We discuss the physical origin of this conclusion and propose independent experimental tests to discriminate between excitonic and Mott insulator.
Results
Exciton binding and instability
For the sake of computational convenience we focus on the smallest (3,3) armchair tube, which was investigated several times from first principles^{53,54,55,56,57,58,59,60}. We first check whether the structural optimization of the tube might lead to deviations from the ideal cylindrical shape, affecting the electronic states. Full geometry relaxation (Methods) yields an equilibrium structure with negligible corrugation. Thus, contrary to a previous claim^{60}, corrugation cannot be responsible of gap opening. We find that the average length of C–C bonds along the tube axis, 1.431 Å, is shorter than around the circumference, 1.438 Å, in agreement with the literature^{53}.
We use DFT to compute the band structure (solid lines in Fig. 2a), which provides the expected^{43} zero gap at the Dirac point K. In addition, we adopt the G0W0 approximation for the selfenergy operator^{44} to evaluate manybody corrections to Kohn–Sham eigenvalues. The highest valence and lowest conduction bands are shown as dashed lines. The zoom near K (Fig. 2b) shows that electrons remain massless, with their bands stretched by ~28% with respect to DFT (farther from K the stretching is ~13%, as found previously^{56}). Since electrons and holes in these bands have linear dispersion, they cannot form a conventional Wannier exciton, whose binding energy is proportional to the effective mass. However, the screened e–h Coulomb interaction V(z) along the tube axis, projected onto the same bands, has long range (Fig. 2c)—a remarkable effect of the topology of the tube holding even for vanishing gap. Consequently, V(q) exhibits a singularity in reciprocal space at q = 0 (smoothed by numerical discretization in the inset of Fig. 2c), which eventually binds the exciton.
We solve the Bethe–Salpeter equation (BSE) over an ultradense grid of 1800 kpoints, which is computationally very demanding but essential for convergence. We find several excitons with negative excitation energies ε _{ u }, in the range of 1–10 meV (Table 1).
The exciton spectral weight is concentrated in a tiny neighborhood of K and K′ points in reciprocal space (Fig. 3b), hence the excitons are extremely shallow, spread over microns along the axis (Fig. 3c). Only e–h pairs with negative k in valley K and positive k in valley K′ contribute to the exciton wave function, which is overall symmetric under time reversal but not under axis reflection within one valley, k → −k, as shown in Fig. 3b (the axis origin is at Dirac point). On the contrary, the wave functions of excitons reported so far in nanotubes^{36,37,56} are symmetric in kspace. The reason of this unusual behavior originates from the vanishing energy gap, since then e–h pairs cannot be backscattered by Coulomb interaction due to the orthogonality of initial and final states^{61}. In addition, pair energies are not degenerate for k → −k, as Dirac cones are slightly asymmetric (Supplementary Discussion and Supplementary Fig. 10).
The exciton with the lowest negative ε _{ u } makes the system unstable against the EI. The transition density, \({\varrho _{{\rm{tr}}}}({\bf{r}}) = \left\langle {{\rm{u}}\left {\hat \varrho ({\bf{r}})} \right0} \right\rangle \), hints at the broken symmetry of the reconstructed ground state, as it connects the noninteracting ground state, \(\left 0 \right\rangle \), to the exciton state, \(\left {\rm{u}} \right\rangle \), through the charge fluctuation operator \(\hat \varrho \) (Fig. 3d). Here we focus on the simpler charge order (spin singlet excitons) and neglect magnetic phenomena (spin triplet), as the only relevant effect of spin–orbit coupling in real tubes^{62,63} is to effectively mix both symmetries. Figure 3d may be regarded as a snapshot of the polarization charge oscillation induced by the exciton, breaking the inversion symmetry between carbon sublattices A and B. Note that this originates from the opposite symmetries of \(\left 0 \right\rangle \) and \(\left {\rm{u}} \right\rangle \) under A ↔ B inversion and not from the vanishing gap. This charge displacement between sublattices is the generic signature of the EI, as its ground state may be regarded as a BCSlike condensate of excitons \(\left {\rm{u}} \right\rangle \) (see the formal demonstration in Supplementary Note 5).
Broken symmetry of the EI
We use quantum Monte Carlo to verify the excitonic nature of the manybody ground state, by defining an order parameter characteristic of the EI, \(\varrho\) _{AB}. In addition, we introduce an alternative order parameter, \(\varrho \) _{Transl}, peculiar to a dimerized charge density wave (CDW) similar to the Peierls CDW predicted by some authors^{57,58,59} for the smallest armchair tubes. The EI order parameter measures the uniform charge displacement between A and B sublattices, \({\varrho _{{\rm{AB}}}} = \left( {\mathop {\sum}\nolimits_{i \in {\rm{A}}} {n_i}  \mathop {\sum}\nolimits_{i \in {\rm{B}}} {n_i}} \right){\rm{/}}{N_{{\rm{atom}}}}\), whereas \(\varrho \) _{Transl} detects any deviation from the periodicity of the undistorted structure by evaluating the charge displacement between adjacent cells, \({\varrho _{{\rm{Transl}}}} = \mathop {\sum}\nolimits_i {n_i}{(  1)^{{i_z}}}{\rm{/}}{N_{{\rm{atom}}}}\) (Fig. 4b–e). Here the undistorted structure is made of a unit cell of 12 C atoms repeated along the z direction with a period of 2.445 Å and labeled by the integer i _{ z }, n _{ i } is the operator counting the electrons within a sphere of radius 1.3 a.u. around the ith atom, and N _{atom} is the total number of atoms in the cluster. Both order parameters \(\varrho \) _{AB} and \(\varrho \) _{Transl} vanish in the symmetric ground state of the undistorted structure, which is invariant under sublatticeswapping inversion and translation symmetries.
We then perform variational Monte Carlo (VMC), using a correlated Jastrow–Slater ansatz that has proved^{64} to work well in 1D correlated systems (Methods), as well as it is able to recover the excitonic correlations present in the meanfield EI wave function^{2,3,4,5} (Supplementary Discussion). We plot VMC order parameters in Fig. 4a. Spontaneously broken symmetry occurs in the thermodynamic limit if the square order parameter, either \(\varrho _{{\rm{AB}}}^2\) or \(\varrho _{{\rm{Transl}}}^2\), scales as 1/N _{atom} and has a non vanishing limit value for N _{atom} → ∞. This occurs for \(\varrho _{{\rm{AB}}}^2\) (black circles in Fig. 4a), confirming the prediction of the EI, whereas \(\varrho _{{\rm{Transl}}}^2\) vanishes (red squares), ruling out the CDW instability (see Supplementary Discussion as well as the theoretical literature^{52,57,58,59} for the Peierls CDW case). We attribute the simultaneous breaking of sublattice symmetry and protection of pristine translation symmetry to the effect of longrange interaction.
The vanishing of \(\varrho \) _{Transl} validates the ability of our finitesize scaling analysis to discriminate between kinds of order in the bulk. Though the value of \(\varrho \) _{AB} after extrapolation is small, \(\varrho \) _{AB}= 0.0165 ± 0.0007, it is non zero within more than twenty standard deviations. Besides, the quality of the fit of Fig. 4a appears good, because the data for the five largest clusters are compatible with the linear extrapolations of both \(\varrho _{{\rm{AB}}}^2\) and \(\varrho _{{\rm{Transl}}}^2\) within an acceptable statistical error. The more accurate diffusion Monte Carlo (LRDMC) values (obtained with the lattice regularization), shown in Fig. 4a as blue circles, confirm the accuracy of the variational calculation. However, as their cost is on the verge of present supercomputing capabilities, we were unable to treat clusters larger that N _{atom}=48, hence the statistical errors are too large to support a meaningful nonzero value in the thermodynamic limit. Nevertheless, we obtain a non zero LRDMC value smaller than the one estimated by VMC but compatible with it within a few standard deviations.
Trends
As the extension of our analysis to systems larger than the (3,3) tube is beyond reach, we design an effectivemass theory to draw conclusions about trends in the armchair tube family, in agreement with firstprinciples findings. We solve the minimal BSE for the massless energy bands ε(k)=±γ k (Fig. 2b and Supplementary Note 1) and the longrange Coulomb interaction V(q), the latter diverging logarithmically in one dimension for small momentum transfer q, V(q)=(2e ^{2}/Aκ _{ r })ln(qR) (inset of Fig. 2c and Supplementary Note 2). Here γ is graphene tightbinding parameter including GW selfenergy corrections, k is the wave vector along the axis, A is the tube length, R is the radius, and κ _{ r } accounts for screening beyond the effectivemass approximation. By fitting the parameters γ=0.5449 eV nm and κ _{ r }=10 to our firstprinciples data, we obtain a numerical solution of BSE recovering ∼60% of the lowest exciton energy ε _{ u } reported in Table 1 (Supplementary Note 3). Moreover, the wave function agrees with the one obtained from first principles (Fig. 3b, c). Importantly, ε _{ u } smoothly converges in an energy range that—for screened interaction—is significantly smaller than the extension of the Dirac cone, with no need of ultraviolet cutoff (Supplementary Fig. 9). Therefore, the exciton has an intrinsic length (binding energy), which scales like R (1/R).
We adopt a meanfield theory of the EI as we expect the longrange character of excitonic correlations to mitigate the effects of quantum fluctuations. The EI wave function can be described as
Here \(\left 0 \right\rangle \) is the zerogap ground state with all valence states filled and conduction states empty, the operator \(\hat c_{k,\sigma }^{\tau + }\) \(\left( {\hat v_{k,\sigma }^{\tau + }} \right)\) creates an electron in the conduction (valence) band with wave vector k, spin σ, valley τ = K or K′, η is an arbitrary phase, and the 2 × 2 matrix χ _{ σσ′} discriminates between singlet and triplet spin symmetries of the e–h pair \(\hat c_{k,\sigma }^{\tau + }\hat v_{k,\sigma '}^\tau \left 0 \right\rangle \) (Fig. 1b). The positive variational quantities u _{ τk } and v _{ τk } are the population amplitudes of valence and conduction levels, respectively, with \(u_{\tau k}^2 + v_{\tau k}^2 = 1\). Whereas in the zerogap state u _{ τk } = 1 and v _{ τk } = 0, in the EI state both u _{ τk } and v _{ τk } are finite and ruled by the EI order parameter Δ(τk), according to \({u_{\tau k}}{v_{\tau k}} = \left {\Delta (\tau k)} \right{\rm{/}}2E(\tau k)\), with \(E(\tau k) = {\left[ {\varepsilon {{(\tau k)}^2} + {{\left {\Delta (\tau k)} \right}^2}} \right]^{1/2}}\). The parameter Δ(τk) obeys the selfconsistent equation
which is solved numerically by recursive iteration (here V includes both long and shortrange interactions as well as form factors, see Supplementary Note 4). As shown in Fig. 5a, in each valley \(\left {\Delta (\tau k)} \right\) is asymmetric around the Dirac point, a consequence of the peculiar character of the exciton wave function of Fig. 3b. The electrons or holes added to the neutral ground state are gapped quasiparticle excitations of the EI, whose energy bands ± E(τk) are shown in Fig. 5b. The order parameter at the Dirac point, \(\left {\Delta (\tau ,k = 0)} \right\), is half the manybody gap. This gap is reminescent of the exciton binding energy, since in the ground state all electrons and holes are bound, so one needs to ionize an excitonlike collective state to create a free electron–hole pair. The gap strongly depends on temperature, with a lowtemperature plateau, a steep descent approaching the critical temperature, and a milder tail (Fig. 5c). The gap approximately scales as 1/R for different tubes (circles in Fig. 5d): whereas at large R such scaling is exact (cf. dashed curve), at small R the gap is enhanced by shortrange intervalley interaction (the decay of Δ will be mitigated if κ _{ r } is sensitive to R).
In experiments, manybody gaps are observed in undoped, ultraclean suspended tubes^{65}, whereas Luttinger liquid signatures emerge in doped tubes^{35,43}. Though it is difficult to compare with the measured manybody gaps^{65}, as the chiralities of the tubes are unknown and the radii estimated indirectly, the measured range of 10–100 meV is at least one order of magnitude larger than our predictions. By doping the tube, we expect that the enhanced screening suppresses the EI order, quickly turning the system into a Luttinger liquid. We are confident that advances in electron spectroscopies will allow to test our theory.
The broken symmetry associated with the EI ground state depends on the exciton spin^{5}. For spin singlet (χ _{ σσ′} = δ _{ σσ′}) and order parameter real (η = 0, π), \(\left {{\Psi _{{\rm{EI}}}}} \right\rangle \) breaks the charge symmetry between A and B carbon sublattices. The charge displacement per electron, Δe/e, at each sublattice site is
where the positive (negative) sign refers to the A (B) sublattice (Supplementary Note 6). For the (3,3) tube this amounts to \(\varrho \) _{AB}=0.0068, which compares well with Monte Carlo estimates of 0.0067 and 0.0165 from LRDMC and VMC, respectively. Note that assessing the energy difference between EI and zerogap ground states is beyond the current capability of quantum Monte Carlo: the meanfield estimate of the difference is below 10^{−6} Hartree per atom, which is less than the noise threshold of the method (10^{−5} Hartree per atom).
Effect of magnetic field
The EI is sensitive to the opening of a noninteracting gap, E _{ g,0}, tuned by the magnetic field parallel to the tube axis, B. The ratio of the flux piercing the cross section, ϕ = πR ^{2} B, to the flux quantum, ϕ _{0} = ch/e, amounts to an Aharonov–Bohm phase displacing the position of the Dirac point along the transverse direction^{66}, k _{⊥} = (ϕ/ϕ _{0})R ^{−1}. Consequently, E _{ g,0} = 2γk _{⊥} is linear with ϕ/ϕ _{0} (red line in Fig. 6a, c). Figure 6a shows the evolution of lowlying singlet (blue lines) and triplet (black lines) excitons of the (3, 3) tube. In addition, we have implemented a full firstprinciples description of B building on a previous method^{67}. Firstprinciples (circles) and model (solid lines) calculations show a fair agreement, which validates the effectivemass theory since all free parameters have been fixed at zero field. Here we rescale energies by R/γ since we expect the plot to be universal, except for small corrections due to shortrange interactions. Excitation energies obtained within the effectivemass model crossover from a lowfield region, where ε _{ u } is almost constant, to a highfield region, where ε _{ u } increases linearly with ϕ/ϕ _{0}. Exciton wave functions are effectively squeezed by the field in real space (Fig. 6b), whereas in reciprocal space they loose their asymmetric character: the amplitudes become evenly distributed around the Dirac points (Supplementary Discussion and Fig. 11) and similar to those reported in literature^{36,37,56}. At a critical flux ϕ _{ c }/ϕ _{0} ≈ 0.035 the excitation energy ε _{ u } becomes positive, hence the tube exits the EI phase and Δ vanishes in a BCSlike fashion. We point out that the critical field intensity, B _{ c } ≈ 460 T ⋅ (R [Å])^{−2}, is out of reach for the (3, 3) tube but feasible for larger tubes. The total transport gap, \({E_g} = {\left( {E_{g,0}^2 + 4{{\left \Delta \right}^2}} \right)^{1/2}}\), first scales with ϕ/ϕ _{0} as E _{ g,0}, then its slope decreases up to the critical threshold ϕ _{ c }/ϕ _{0}, where the linear dependence on ϕ/ϕ _{0} is restored (Fig. 6c). This behavior is qualitatively similar to that observed by Coulomb blockade spectroscopy in narrowgap tubes close to the “Dirac” value of B, which counteracts the effect of E _{ g,0} on the transport gap, fully suppressing the noninteracting contribution^{65}.
Discussion
The observed^{65} manybody gap of armchair tubes was attributed to the Mott insulating state. The system was modeled as a strongly interacting Luttinger liquid with a gap enforced by shortrange interactions^{46,49}, whereas the long tail of the interaction was cut off at an extrinsic, setupdependent length^{47,48,50,51,52}. This model thus neglects the crucial effect of longrange interaction, which was highlighted in Fig. 1: were any cutoff length smaller than the intrinsic exciton length, which is micrometric and scales with R, excitons could not bind.
Whereas armchair carbon nanotubes are regarded as quintessential realizations of the Luttinger liquid, since their lowenergy properties are mapped into those of twoleg ladders^{46}, we emphasize that this mapping is exact for shortrange interactions only. Among e–h pair collective modes with total momentum q = 0, Luttinger liquid theory routinely describes plasmons^{68} but not excitons. Contrary to conventional wisdom, armchair tubes are EIs.
The excitonic and Mott insulators are qualitatively different. The EI exhibits longrange charge order, which does not affect the translational symmetry of the zerogap tube. In the Mott insulator, charge and spin correlations may or may not decay, but always add a 2π/(2k _{ F }) [or 2π/(4k _{ F })] periodicity to the pristine system, k _{ F } being the Fermi wave vector^{50,51}. The EI gap scales like 1/R (Fig. 5d), the Mott gap like 1/R ^{1/(1−g)}, with predicted^{47,50,51,52} values of g pointing to a faster decay, g < 1. The EI order parameter is suppressed at high temperature (Fig. 5c) and strong magnetic field (Fig. 6c); the Mott gap is likely independent of both fields (the Aharonov–Bohm phase does not affect Hubbardlike Coulomb integrals). Importantly, the EI gap is very sensitive to the dielectric environment^{69}, whereas the Mott gap is not. This could explain the dramatic variation of narrow transport gaps of suspended tubes submerged in different liquid dielectrics^{42}.
We anticipate that armchair tubes exhibit an optical absorption spectrum in the THz range dominated by excitons, which provides an independent test of the EI phase. Furthermore, we predict they behave as “chiral electronic ferroelectrics”, displaying a permanent electric polarization P of purely electronic origin^{7}, whereas conventional ferroelectricity originates from ionic displacements. In fact, the volume average of P is zero but its circulation along the tube circumference is finite. Therefore, a suitable timedependent field excites the ferroelectric resonance^{7} associated with the oscillation of P. The special symmetry of armchair tubes^{61} is expected to protect this collective (Goldstone) mode of oscillating electric dipoles from phaselocking mechanisms. The resulting soft mode—a displacement current along the tube circumference—is a manifestation of the longdebated^{6,7,8,9,10,11,70,71} exciton superfluidity.
In conclusion, our calculations demonstrated that an isolated armchair carbon nanotube at charge neutrality is an EI, owing to the strong e–h binding in quasi1D, and the almost unscreened longrange interactions. The emergence of this exotic state of matter, predicted fifty years ago, does not fit the common picture of carbon nanotubes as Luttinger liquids. Our firstprinciples calculations provide tests to discriminate between the EI and the Luttinger liquid at strong coupling, the Mott insulator state. We expect a wide family of narrowgap carbon nanotubes to be EIs. Carbon nanotubes are thus invaluable systems for the experimental investigation of this phase of matter.
Methods
Manybody perturbation theory from first principles
The groundstate calculations for the (3, 3) carbon nanotube were performed by using a DFT approach, as implemented in the Quantum ESPRESSO package^{72}. The generalized gradient approximation (GGA) PW91 parametrization^{73} was adopted together with plane wave basis set and normconserving pseudopotentials to model the electron–ion interaction. The kinetic energy cutoff for the wave functions was set to 70 Ry. The Brillouin zone was sampled by using a 200 × 1 × 1 kpoint grid. The supercell side perpendicular to the tube was set to 38 Bohr and checked to be large enough to avoid spurious interactions with its replica.
Manybody perturbation theory^{44} calculations were performed using the Yambo code^{74}. Manybody corrections to the Kohn–Sham eigenvalues were calculated within the G0W0 approximation to the selfenergy operator, where the dynamic dielectric function was obtained within the plasmonpole approximation. The spectrum of excited states was then computed by solving the BSE. The static screening in the direct term was calculated within the randomphase approximation with inclusion of local field effects; the Tamm–Dancoff approximation for the BSE Hamiltonian was employed after having verified that the correction introduced by coupling the resonant and antiresonant part was negligible. Converged excitation energies, ε _{ u }, were obtained considering respectively three valence and four conduction bands in the BSE matrix. For the calculations of the GW band structure and the Bethe–Salpeter matrix the Brillouin zone was sampled with a 1793 × 1 × 1 kpoint grid. A kinetic energy cutoff of 55 Ry was used for the evaluation of the exchange part of the self energy and 4 Ry for the screening matrix size. Eighty unoccupied bands were used in the integration of the selfenergy.
The effect of the magnetic field parallel to the axis on the electronic structure of the nanotube ground state (eigenvalues and eigenfunctions) was investigated following the method by Sangalli and Marini^{67}. For each value of the field, the eigenvalues and eigenfunctions were considered to build the screening matrix and the corresponding excitonic Hamiltonian.
To obtain the equilibrium structure, we first considered possible corrugation effects. We computed the total energy for a set of structures obtained by varying the relative positions of A and B carbon atoms belonging to different sublattices, so that they were displaced one from the other along the radial direction by the corrugation length Δ and formed two cylinders, as in Fig. 1b of Lu et al.^{60}. Then, we fitted the total energy per carbon atom with an elliptic paraboloid in the twodimensional parameter space spanned by Δ and the carbon bond length. In agreement with Lu et al.^{60}, we find a corrugated structure with a bond length of 1.431 Å and a corrugation parameter Δ 0.018 Å. Eventually, starting from this structure, we performed a full geometry relaxation of the whole system allowing all carbon positions to change until the forces acting on all atoms became less than 5 × 10^{−3} eV Å^{−1}. After relaxation, the final structure presents a negligible corrugation (Δ < 10^{−5} Å) and an average length of C–C bonds along the tube axis, 1.431 Å, slightly shorter than the C–C bonds around the tube circumference, 1.438 Å. The average radius and translation vector of the tube are respectively 2.101 and 2.462 Å, in perfect agreement with the literature^{53}. The obtained equilibrium coordinates of C atoms in the unitary cell are shown in Supplementary Table 1.
Quantum Monte Carlo method
We have applied the quantum Monte Carlo method to carbon nanotubes by using standard pseudopotentials for the 1s core electrons of the carbon atom^{75}. We minimize the total energy expectation value of the firstprinciples Hamiltonian, within the Born–Oppheneimer approximation, by means of a correlated wave function, \(J\left {{\rm{SD}}} \right\rangle \). This is made of a Slater determinant, \(\left {{\rm{SD}}} \right\rangle \), defined in a localized GTO VDZ basis^{75} (5s5p1d) contracted into six hybrid orbitals per carbon atom^{76}, multiplied by a Jastrow term, J. The latter, J = J _{1} J _{2}, is the product of two factors: a oneelectron term, \({J_1} = \mathop {\prod}\nolimits_i {\rm{exp}}\left[ {{u_{{\rm{1body}}}}\left( {{{\bf{r}}_i}} \right)} \right]\), and a twoelectron correlation factor, \({J_2} = \mathop {\prod}\nolimits_{i < j} {\rm{exp}}\left[ {u\left( {{{\bf{r}}_i},{{\bf{r}}_j}} \right)} \right]\). The twobody Jastrow factor J _{2} depends explicitly on the N _{ e } electronic positions, {r _{ i }}, and, parametrically, on the N _{C} carbon positions, R _{ I }, I=1, … N _{C}. The pseudopotential functions, u and u _{1body}, are written as:
where u _{ee}=2^{−1} r/(1 + b _{ee} r) is a simple function, depending on the single variational parameter b _{ee}, which allows to satisfy the electron–electron cusp condition, and u _{ μν } is a symmetric matrix of finite dimension. For nonnull indices, μ, ν > 0, the matrix u describes the variational freedom of J _{2} in a certain finite atomic basis, χ _{ μ }(r), which is localized around the atomic centers R _{ I(μ)} and is made of 3s2p GTO orbitals per atom. Note that the onebody Jastrow term J _{1} is expanded over the same atomic basis and its variational freedom is determined by the first column of the matrix, u _{ μ0}.
We use an orthorombic unit cell L _{ x } × L _{ y } × L _{ z } containing twelve atoms with L _{ x } = L _{ y } = 36 Å and L _{ z } =2.445 Å. This cell is repeated along the z direction for n =1, 2, 3, 4, 5, 6 times, up to 72 carbon atoms in the supercell. Periodic images in the x and y directions are far enough that their mutual interaction can be safely neglected. Conversely, in the z direction we apply twisted periodic boundary conditions and we integrate over that with a number n _{ θ } of twists, n _{ θ } = 80, 40, 30, 20, 20, 20 for n = 1, 2, 3, 4, 5, 6, respectively, large enough to have converged results for each supercell.
The initial Slater determinant was taken by performing a standard LDA calculation. The molecular orbitals, namely their expansion coefficients in the GTO localized basis set, as well as the matrix u determining the Jastrow factor, were simultaneously optimized with well established methods developed in recent years^{77,78}, which allows us to consider up to 3000 independent variational parameters in a very stable and efficient way. Note that the twobody Jastrow term J _{2} can be chosen to explicitly recover the EI meanfield wave function (1), as shown in Supplementary Discussion. After the stochastic optimization the correlation functions/order parameters can be computed in a simple way within VMC.
We also employ latticeregularized diffusion Monte Carlo (LRDMC) within the fixednode approximation, using a lattice mesh of a _{mesh} = 0.2 and a _{mesh} = 0.4 a.u., respectively, in order to check the convergence for a _{mesh} → 0. The fixednode approximation is necessary for fermions for obtaining statistically meaningful groundstate properties. In this case the correlation functions/order parameters, depending only on local (i.e., diagonal in the basis) operators, such as the ones presented in this work, are computed with the forward walking technique^{79}, which allows the computation of pure expectation values on the fixednode ground state.
Code availability
Manybody perturbation theory calculations were performed by means of the codes Yambo (http://www.yambocode.org/) and Quantum ESPRESSO (http://www.quantumespresso.org), which are both open source software. Quantum Monte Carlo calculations were based on TurboRVB code (http://trac.sissa.it/svn/TurboRVB), which is available from S.S. upon reasonable request.
Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Additional information
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
 1.
Sherrington, D. & Kohn, W. Speculations about gray tin. Rev. Mod. Phys. 40, 767–769 (1968).
 2.
Keldysh, L. V. & Kopaev, Y. V. Possible instability of the semimetallic state against Coulomb interaction. Fiz. Tverd. Tela 6, 2791–2798 (1964) [Sov. Phys. Solid State 6, 2219–2224 (1965)].
 3.
des Cloizeaux, J. Excitonic instability and crystallographic anomalies in semiconductors. J. Phys. Chem. Solids 26, 259–266 (1965).
 4.
Jèrome, D., Rice, T. M. & Kohn, W. Excitonic insulator. Phys. Rev. 158, 462–475 (1967).
 5.
Halperin, B. I. & Rice, T. M. The excitonic state at the semiconductorsemimetal transition. Solid State Phys. 21, 115–192 (1968).
 6.
Lozovik, Y. E. & Yudson, V. I. A new mechanism for superconductivity: pairing between spatially separated electrons and holes. Zh. Eksp. i Teor. Fiz. 71, 738–753 (1976) [Sov. Phys.–JETP 44, 389–397 (1976)].
 7.
Portengen, T., Östreich, T. & Sham, L. J. Theory of electronic ferroelectricity. Phys. Rev. B 54, 17452–17463 (1996).
 8.
Balatsky, A. V., Joglekar, Y. N. & Littlewood, P. B. Dipolar superfluidity in electronhole bilayer systems. Phys. Rev. Lett. 93, 266801 (2004).
 9.
Rontani, M. & Sham, L. J. Coherent transport in a homojunction between an excitonic insulator and semimetal. Phys. Rev. Lett. 94, 186404 (2005).
 10.
Su, J. & MacDonald, A. H. How to make a bilayer exciton condensate flow. Nat. Phys. 4, 799–802 (2008).
 11.
Littlewood, P. B. in Problems of Condensed Matter Physics, vol. 139 of International Series of Monographs on Physics (eds Ivanov, A. L. & Tikhodeev, S. G.) Ch. 11 163–181 (Oxford University Press, Oxford, 2008).
 12.
Bucher, B., Steiner, P. & Wachter, P. Excitonic insulator phase in TmSe_{0.45}Te_{0.55}. Phys. Rev. Lett. 67, 2717–2720 (1991).
 13.
Rontani, M. & Sham, L. J. in Novel Superfluids Volume 2, vol. 157 of International Series of Monographs on Physics (eds Bennemann, K. H. & Ketterson, J. B.) Ch. 19 423–474 (Oxford University Press, Oxford, 2014).
 14.
Salvo, F. J. D., Moncton, D. E. & Waszczak, J. V. Electronic properties and superlattice formation in the semimetal TiSe_{2}. Phys. Rev. B 14, 4321–4328 (1976).
 15.
Rossnagel, K. On the origin of chargedensity waves in select layered transitionmetal dichalcogenides. J. Phys. Condens. Matter 23, 213001 (2011).
 16.
Rice, T. M. The electronhole liquid in semiconductors: theoretical aspects. Solid State Phys. 32, 1–86 (1977).
 17.
Keldysh, L. V. in BoseEinstein Condensation (eds Griffin, A., Snoke, D. W. & Stringari, S.) Ch. 12 246–280 (Cambridge University Press, Cambridge, 1995).
 18.
Ikeda, N. et al. Ferroelectricity from iron valence ordering in the chargefrustrated system LuFe_{2}O_{4}. Nature 436, 1136–1138 (2005).
 19.
Spielman, I. B., Eisenstein, J. P., Pfeiffer, L. N. & West, K. W. Resonantly enhanced tunneling in a double layer quantum Hall ferromagnet. Phys. Rev. Lett. 84, 5808–5811 (2000).
 20.
Nandi, A., Finck, A. D. K., Eisenstein, J. P., Pfeiffer, L. N. & West, K. W. Exciton condensation and perfect Coulomb drag. Nature 488, 481–484 (2012).
 21.
Palo, S. D., Rapisarda, F. & Senatore, G. Exciton condensation in a symmetric electronhole bilayer. Phys. Rev. Lett. 88, 206401 (2002).
 22.
Kuneš, J. Excitonic condensation in systems of strongly correlated electrons. J. Phys. Condens. Matter 27, 333201 (2015).
 23.
Khveshchenko, D. V. Ghost excitonic insulator transition in layered graphite. Phys. Rev. Lett. 87, 246802 (2001).
 24.
Vafek, O. & Case, M. J. Renormalization group approach to twodimensional Coulomb interacting Dirac fermions with random gauge potential. Phys. Rev. B 77, 033410 (2008).
 25.
Drut, J. E. & Lände, T. A. Is graphene in vacuum an insulator? Phys. Rev. Lett. 102, 026802 (2009).
 26.
Gamayun, O. V., Gorbar, E. V. & Gusynin, V. P. Supercritical Coulomb center and excitonic instability in graphene. Phys. Rev. B 80, 165429 (2009).
 27.
Lozovik, Y. E. & Sokolik, A. A. Electronhole pair condensation in a graphene bilayer. JETP Lett. 87, 55–59 (2008).
 28.
Dillenschneider, R. & Han, J. H. Exciton formation in graphene bilayer. Phys. Rev. B 78, 045401 (2008).
 29.
Min, H., Bistritzer, R., Su, J. & MacDonald, A. H. Roomtemperature superfluidity in graphene bilayers. Phys. Rev. B 78, 121401(R) (2008).
 30.
Zhang, C. & Joglekar, Y. N. Excitonic condensation of massless fermions in graphene bilayers. Phys. Rev. B 77, 233405 (2008).
 31.
Rodin, A. S. & Castro Neto, A. H. Excitonic collapse in semiconducting transitionmetal dichalcogenides. Phys. Rev. B 88, 195437 (2013).
 32.
Fogler, M. M., Butov, L. V. & Novoselov, K. S. Hightemperature superfluidity with indirect excitons in van der Waals heterostructures. Nat. Commun. 5, 4555 (2014).
 33.
Stroucken, T. & Koch, S. W. Optically bright pexcitons indicating strong Coulomb coupling in transitionmetal dichalcogenides. J. Phys.: Condens. Matter 27, 345003 (2015).
 34.
Saito, R., Dresselhaus, G. & Dresselhaus, M. S. Physical Properties of Carbon Nanotubes (Imperial College Press, London, 1998).
 35.
Ilani, S. & McEuen, P. L. Electron transport in carbon nanotubes. Annu. Rev. Condens. Matter Phys. 1, 1–25 (2010).
 36.
Ando, T. Excitons in carbon nanotubes. J. Phys. Soc. Jpn 66, 1066–1073 (1997).
 37.
Maultzsch, J. et al. Exciton binding energies in carbon nanotubes from twophoton photoluminescence. Phys. Rev. B 72, 241402(R) (2005).
 38.
Wang, F., Dukovic, G., Brus, L. E. & Heinz, T. The optical resonances in carbon nanotubes arise from excitons. Science 308, 838–841 (2005).
 39.
Wang, F. et al. Observation of excitons in onedimensional metallic singlewalled carbon nanotubes. Phys. Rev. Lett. 99, 227401 (2007).
 40.
Waissman, J. et al. Realization of pristine and locally tunable onedimensional electron systems in carbon nanotubes. Nat. Nanotechnol. 8, 569–574 (2013).
 41.
Laird, E. A. et al. Quantum transport in carbon nanotubes. Rev. Mod. Phys. 87, 703–764 (2015).
 42.
Aspitarte, L. et al. Giant modulation of the electronic band gap of carbon nanotubes by dielectric screening. Sci. Rep. 7, 8828 (2017).
 43.
Charlier, J., Blase, X. & Roche, S. Electronic and transport properties of nanotubes. Rev. Mod. Phys. 79, 677–732 (2007).
 44.
Onida, G., Reining, L. & Rubio, A. Electronic excitations: densityfunctional versus manybody Green’s function approaches. Rev. Mod. Phys. 74, 601–659 (2002).
 45.
Foulkes, W. M. C., Mitas, L., Needs, R. J. & Rajagopal, G. Quantum Monte Carlo simulations of solids. Rev. Mod. Phys. 73, 33–83 (2001).
 46.
Balents, L. & Fisher, M. P. A. Correlation effects in carbon nanotubes. Phys. Rev. B 55, R11973–R11976 (1997).
 47.
Kane, C. L., Balents, L. & Fisher, M. Coulomb interaction and mesoscopic effects in carbon nanotubes. Phys. Rev. Lett. 79, 5086–5089 (1997).
 48.
Egger, R. & Gogolin, A. O. Effective lowenergy theory for correlated carbon nanotubes. Phys. Rev. Lett. 79, 5082–5085 (1997).
 49.
Krotov, Y. A., Lee, D. & Louie, S. G. Low energy properties of (n, n) carbon nanotubes. Phys. Rev. Lett. 78, 4245–4248 (1997).
 50.
Yoshioka, H. & Odintsov, A. A. Electronic properties of armchair carbon nanotubes: Bosonization approach. Phys. Rev. Lett. 82, 374–377 (1999).
 51.
Nersesyan, A. A. & Tsvelik, A. M. Coulomb blockade regime of a singlewall carbon nanotube. Phys. Rev. B 68, 235419 (2003).
 52.
Chen, W., Andreev, A. V., Tsvelik, A. M. & Orgad, D. Twist instability in strongly correlated carbon nanotubes. Phys. Rev. Lett. 101, 246802 (2008).
 53.
Liu, H. J. & Chan, C. T. Properties of 4 Å carbon nanotubes from firstprinciples calculations. Phys. Rev. B 66, 115416 (2002).
 54.
Machón, M., Reich, S., Thomsen, C., SánchezPortal, D. & Ordejón, P. Ab initio calculations of the optical properties of 4 Ådiameter singlewalled carbon nanotubes. Phys. Rev. B 66, 155410 (2002).
 55.
Cabria, I., Mintmire, J. W. & White, C. T. Metallic and semiconducting narrow carbon nanotubes. Phys. Rev. B 67, 121406(R) (2003).
 56.
Spataru, C. D., IsmailBeigi, S., Benedict, L. X. & Louie, S. G. Excitonic effects and optical spectra of singlewalled carbon nanotubes. Phys. Rev. Lett. 92, 077402 (2004).
 57.
Bohnen, K., Heid, R., Liu, H. J. & Chan, C. T. Lattice dynamics and electronphonon interaction in (3,3) carbon nanotubes. Phys. Rev. Lett. 93, 245501 (2004).
 58.
Connétable, D., Rignanese, G., Charlier, J. & Blase, X. Room temperature Peierls distortion in small diameter nanotubes. Phys. Rev. Lett. 94, 015503 (2005).
 59.
Dumont, G., Boulanger, P., Côté, M. & Ernzerhof, M. Peierls instability in carbon nanotubes: a firstprinciples study. Phys. Rev. B 82, 035419 (2010).
 60.
Lu, H., Wu, J. & Zhang, W. Corrugationinduced metalsemiconductor transition in singlewall carbon nanotubes with a small radius. Phys. Rev. B 88, 035423 (2013).
 61.
Ando, T., Nakanishi, T. & Saito, R. Berry’s phase and absence of backscattering in carbon nanotubes. J. Phys. Soc. Jpn. 67, 2857–2862 (1998).
 62.
Kuemmeth, F., Ilani, S., Ralph, D. C. & McEuen, P. L. Coupling of spin and orbital motion of electrons in carbon nanotubes. Nature 452, 448–452 (2008).
 63.
Steele, G. A. et al. Large spinorbit coupling in carbon nanotubes. Nat. Commun. 4, 1573 (2013).
 64.
Capello, M., Becca, F., Fabrizio, M., Sorella, S. & Tosatti, E. Variational description of Mott insulators. Phys. Rev. Lett. 94, 026406 (2005).
 65.
Deshpande, V. V. et al. Mott insulating state in ultraclean carbon nanotubes. Science 323, 106–110 (2009).
 66.
Ajiki, H. & Ando, T. Electronic states of carbon nanotubes. J. Phys. Soc. Jpn. 62, 1255–1266 (1993).
 67.
Sangalli, D. & Marini, A. Anomalous AharonovBohm gap oscillations in carbon nanotubes. Nano Lett. 11, 4052–4057 (2011).
 68.
Schulz, H. J. Wigner crystal in one dimension. Phys. Rev. Lett. 71, 1864–1867 (1993).
 69.
Zittartz, J. Transport properties of the “excitonic insulator”: electrical conductivity. Phys. Rev. 165, 605–611 (1968).
 70.
Kohn, W. & Sherrington, D. Two kinds of bosons and Bose condensates. Rev. Mod. Phys. 42, 1–11 (1970).
 71.
Guseinov, R. R. & Keldysh, L. V. Nature of the phase transition under the condition of an “excitonic” instability in the electronic spectrum of a crystal. Zh. Eksp. i Teor. Fiz. 63, 2255–2263 (1972) [Sov. Phys.–JETP 36, 1193–1197 (1973)].
 72.
Giannozzi, P. et al. Quantum ESPRESSO: a modular and opensource software project for quantum simulations of materials. J. Phys. Condens. Matter 21, 395502 (2009).
 73.
Perdew, J. P. et al. Atoms, molecules, solid, and surfaces: applications of the generalized gradient approximation for exchange and correlation. Phys. Rev. B 46, 6671–6687 (1992).
 74.
Marini, A., Hogan, C., Grüning, M. & Varsano, D. Yambo: an ab initio tool for excited state calculations. Comput. Phys. Commun. 180, 1392–1403 (2009).
 75.
Burkatzki, M., Filippi, C. & Dolg, M. Energyconsistent pseudopotentials for quantum Monte Carlo calculations. J. Chem. Phys. 126, 234105 (2007).
 76.
Sorella, S., Devaux, N., Dagrada, M., Mazzola, G. & Casula, M. Geminal embedding scheme for optimal atomic basis set construction in correlated calculations. J. Chem. Phys. 143, 244112 (2015).
 77.
Dagrada, M., Karakuzu, S., Vildosola, V. L., Casula, M. & Sorella, S. Exact special twist method for quantum Monte Carlo simulations. Phys. Rev. B 94, 245108 (2016).
 78.
Umrigar, C. J., Toulouse, J., Filippi, C., Sorella, S. & Rhenning, H. Alleviation of the fermionsign problem by optimization of manybody wave functions. Phys. Rev. Lett. 98, 110201 (2007).
 79.
Calandra Buonaura, M. & Sorella, S. Numerical study of the twodimensional Heisenberg model using a Green function Monte Carlo technique with a fixed number of walkers. Phys. Rev. B 57, 11446–11456 (1998).
 80.
Rontani, M. Anomalous magnetization of a carbon nanotube as an excitonic insulator. Phys. Rev. B 90, 195415 (2014).
Acknowledgements
This work was supported in part by European Union H2020EINFRA20151 program under grant agreement No. 676598 project “MaX–Materials Design at the Exascale”. S.S. acknowledges computational resouces provided through the HPCI System Research Project No. hp160126 on the K computer at RIKEN Advanced Institute for Computational Science. D.V., E.M. & M.R. acknowledge PRACE for awarding them access to the Marconi system based in Italy at CINECA (Grant No. Pra14_3622).
Author information
Author notes
Affiliations
Contributions
M.R. and E.M. initiated this project, D.V., E.M. and M.R. designed a comprehensive strategy to tackle the instability problem by means of different methods, D.V. developed the manybody perturbation theory calculations and analysis, D.V. and D.S. optimized the Yambo code for the calculation in the presence of the magnetic field, S.S. and M.B. developed the quantum Monte Carlo calculations and analysis, M.R. developed the effectivemass theory and wrote the paper, all authors contributed to the analysis of data and critically discussed the paper.
Competing interests
The authors declare no competing financial interests.
Corresponding author
Correspondence to Massimo Rontani.
Electronic supplementary material
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Received
Accepted
Published
DOI
Further reading

Nanomechanical pump–probe measurements of insulating electronic states in a carbon nanotube
Nature Nanotechnology (2019)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.