betaFeedback

WorksPhysics and Astronomy

Low-energy model for doped graphene nanoribbons

Abstract

We analyse in this article the many-body behavior of free-standing doped graphene nanoribbons where the chemical potential lies inside the bulk single-particle bands. We perform an exact mapping from both an extended and an on-site Hubbard model of the ribbons to a Kanamori model, which includes ferromagnetic exchange and pair-hopping interactions. We determine the resulting Coulomb matrix elements analytically, and identify their scaling behavior as a function of ribbon width and length. We propose a low-energy version of the Kanamori Hamiltonian to address the response of the ribbons to external fields, with a view to their use as transport channels in nanoelectronics. We find that the model and the proposed ribbon parameters can produce open-shell, high-spin many-body states that can lead to shell- and spin-blockade responses.

Introduction

Graphene was originally envisioned as the dreamed-of electronic material [1]. The vision included the realization of nano-electronic devices such as transistors or quantum dots, frequently based on graphene nanoribbons (GNRs) [2]. However, the expected revolution was hindered by a variety of factors, among which are the difficulties in producing atomically-precise GNRs [3]. Recently, atomically-precise GNRs have been grown in ultra-high vacuum chambers and transferred to substrates with metal electrodes. Transport through single- or few-nanoribbon devices has recently been measured, showing single-electron transistor and quantum dot behavior [4, 5, 6, 7]. In understanding their behavior, it is important to realize that graphene nanoribbons are different from carbon nanotubes because ±k\pm k wave-vectors are mixed by boundary conditions, so valleys are not good quantum numbers. Spin-splitting is expected to be absent in graphene nanoribbons except when the chemical potential is pinned around the edge-state energy window. Furthermore, both the intrinsic and Rashba spin orbit interactions are expected to be small like in bulk graphene, which exhibits a gap of about 25 μeV25\ \mu\mathrm{eV} [8, 9, 10].

In previous work [11, 12], we diagonalized analytically the tight-binding model of GNRs with arbitrary length and width, and showed that the existence or not, magnetic nature and energy of edge states in N=5N = 5, 7 and 9 GNRs could be described by the use of an effective dimer Hubbard model. The tight-binding solution was extended to the full Hubbard model of a GNR in Ref. 13. The Hubbard term was rewritten in the ribbon eigen-state basis, whereby a huge sum containing operator products c^1σ†c^2σ′†c^3σ′c^4σ\hat{c}_{1\sigma}^{\dagger}\hat{c}_{2\sigma'}^{\dagger}\hat{c}_{3\sigma'}\hat{c}_{4\sigma} was generated. The sum was truncated by keeping only direct 1−41-4, 2−32-3 pairings, and the Coulomb integrals were calculated analytically. Explicit analytical mean-field equations for the paramagnetic, ferromagnetic and antiferromagnetic phases were deduced. The electronic band structure for narrow ribbons agreed with previous mean-field Hubbard model simulations [14, 15]. The band structure for bulk-sized samples with edges did contain both the bulk Dirac cone as well as the edge states band at the same time, something that had not been achieved before.

In this article, we consider free-standing finite-length graphene nanoribbons (GNR). The ribbons are doped and possibly subject to a gate voltage so that the chemical potential lies inside the bands of bulk states. We provide an exact mapping of the extended Hubbard model of GNRs [16, 17, 18] to a Kanamori Hamiltonian [19], which is written in terms of the single-particle bulk eigen-states of the ribbon. We thus show that Hubbard interactions lead not only to direct Coulomb integrals among these single-particle bulk eigen-states, but also to ferromagnetic exchange and pair hopping terms. The pristine extended Hubbard Hamiltonian is written in a real-space site basis; in contrast the Kanamori Hamiltonian is written in the basis of single-particle eigen-states. This new Hamiltonian can be useful for the prediction or the analysis of phenomena that address given energy windows. An example are recent transport experiments of single- or few-ribbon junctions [4, 5, 6, 7, 20], that have uncovered their quantum dot response. In these low-temperature experiments, a gate voltage tunes the chemical potential, and a narrow energy-window of bulk ribbon eigen-states is addressed. Detailed stability diagram featuring Coulomb diamonds and excited-state lines can be drawn by sweeping both the bias and gate voltage in those experiments.

We will discuss in this article how the new Hamiltonian allows us to address cleanly narrow energy-windows of eigen-states, whose position is given by the chemical potential and whose width is determined by and can be adapted to the number of single-particle states that are half-filled. The resulting low-energy Hamiltonian possesses a small enough Fock space that can be analyzed exactly and in detail either analytically or numerically. We find that the Kanamori exchange interactions are ferromagnetic and thus produce high-spin open-shell many-body states in doped GNR. Several theorems preclude the existence of saturated ferromagnetism for the Hubbard model [21, 22], but do predict high-spin (ferrimagnetic) ground states that fulfill specific conditions [22, 23]. We find equivalent conditions for the high-spin ground states described below. Similarly, some studies predict mesoscopic Stoner instabilities in open quantum dots [24] as we do here.

The low-energy Hamiltonian approach is complementary to describing GNRs by Density Functional theory (DFT) simulations. This is because the Coulomb parameters introduced in the initial extended Hubbard model can and must be deduced from DFT [11, 16, 17, 18]. On the other hand, correlated electron phenomena like Coulomb blockade or Kondo physics cannot be described correctly by DFT simulations. The reason is that DFT is a one-body theory so that even the exact functional does not describe the energy position and number of many-body quasi-particle poles [25, 26]. A DFT-based analysis of Coulomb blockade phenomena delivers a wrong number and positioning of Coulomb diamonds, and of excited state lines.

We will make this article self-contained and will therefore make use of some results that have been published previously [11, 13]. The layout of the article is as follows. Section II summarizes the analytical solution of the tight-binding model of graphene ribbons of arbitrary width and length [11]. Section III explains the extended Hubbard model of bulk graphene and introduces it for graphene nanoribbons. This section also shows how the extended Hubbard Hamiltonian can be mapped to a Kanomori model [19], and determines the Coulomb matrix elements. Section IV introduces an onsite Hubbard model and fits the UU parameter using Density Functional Theory (DFT) simulations. The Coulomb matrix elements are vastly simplified here, and found to obey simple relationships among them, and a remarkable size-scaling law. The section also discusses analytical mean field solutions that were found earlier [13]. Section V uses the previous results to propose a low-energy Kanamori Hamiltonian to describe the open-shell and spin behavior of doped GNR. The low-energy Hamiltonian is solved for the case of two and three shells, and an specific example using a N=9N=9 armchair ribbon is discussed. Section VI introduces a transfer Hamiltonian and discusses transition rates in the Sequential Electron Tunneling regime. The section thus show how the model can lead to shell- and spin-blockade in electrical transport experiments using the ribbons as channels connecting two leads. A short conclusion closes the article.

Tight-binding model of graphene ribbons

We consider nanoribbons whose lengths and widths are MxM_x and My=N+12M_y=\frac{N+1}{2}, according to the convention set in Figure 1. Unit cells are rectangles having four atoms. We measure the lengths along the cartesian XX- and YY-axes in units of 3c\sqrt{3}c and cc, respectively, where c=2.46c=2.46 Å is graphene’s lattice constant. Cell coordinates are R=(nx,ny)\mathbf{R}=(n_x,n_y) with nx/y=1,2,…,Mx/yn_{x/y}=1,2,\ldots,M_{x/y}. Furthermore, we address for simplicity ribbons with no edge states contribute to the physics. This condition happens in both armchair and zigzag ribbons whenever the effective doping or the gate voltage are high enough that the chemical potential lies inside the conduction or valence bands.

Graphene nanoribbon unit cells with real and fake atoms

Figure 1. Real graphene atoms are indicated by red circles. Fake atoms, whose wave-function coefficients are set to zero, are indicated by blue circles. Each unit cell is surrounded with a gray dotted box.

The tight-binding Hamiltonian is constructed by placing a Wannierized pzp^z orbital at each site i=R+dA/Bi=\mathbf{R}+\mathbf{d}_{A/B} of the lattice with unit cell coordinates R\mathbf{R} and intra-cell coordinates dA/B\mathbf{d}_{A/B}. The corresponding states are represented by the kets ∣i⟩=∣R+dA/B⟩|i\rangle=|\mathbf{R}+\mathbf{d}_{A/B}\rangle. The second-quantized Hamiltonian is

H^0=−t∑⟨i,j⟩,σr^iσ†r^jσ(1)\hat{H}^{0}=-t\sum_{\langle i,j\rangle,\sigma}\hat{r}^{\dagger}_{i\sigma}\hat{r}_{j\sigma} \tag*{(1)}

where the sum is restricted to nearest neighbor sites. The linear operator r^iσ†\hat{r}^{\dagger}_{i\sigma} creates an electron with spin σ\sigma at the ribbon site i\mathbf{i}, and n^iσ=r^iσ†r^iσ\hat{n}_{i\sigma}=\hat{r}^{\dagger}_{i\sigma}\hat{r}_{i\sigma} is the corresponding number operator, with n^i=∑σn^iσ\hat{n}_i=\sum_{\sigma}\hat{n}_{i\sigma}.

The ribbon boundary conditions define the set of allowed wave-vectors, that we define as kˉ=(kˉx,kˉy)=(3ckx,cky)\bar{\mathbf{k}}=(\bar{k}_x,\bar{k}_y)=(\sqrt{3}ck_x,ck_y). Here, kˉy=km=πMym\bar{k}_y=k_m=\frac{\pi}{M_y}m with m=1,…,Mym=1,\ldots,M_y. The wave-vectors kˉx=kmα\bar{k}_x=k_{m\alpha}, with α=1,…,2Mx\alpha=1,\ldots,2M_x, must be computed numerically and are not spaced evenly. The subset km 2Mxk_{m\,2M_x} with km≥23πk_m\geq\frac{2}{3}\pi corresponds to edge ribbon states [11, 13] and are discarded, while all the rest label bulk ribbon states and are kept. The Hamiltonian H^0\hat{H}^{0} can be diagonalized analytically for any non-chiral finite-length ribbon [11]. The site operators can be decomposed in terms of bulk eigen-state operators as follows

r^iσ†=∑mατr^mατσ†⟨ϕmατ∣i⟩(2)\hat{r}_{i\sigma}^{\dagger}=\sum_{m\alpha\tau}\hat{r}_{m\alpha\tau\sigma}^{\dagger}\langle\phi_{m\alpha\tau}|i\rangle \tag*{(2)}
⟨i∣ϕmατ⟩=∑Rx,Ry=1Mx,y2sin⁡km(Ry−di)(MxMyAmα)1/2×(−τ(−1)αsin⁡(kmα(Rx−di))sin⁡(kmα(Mx+1−(Rx+di))))(3)\langle i|\phi_{m\alpha\tau}\rangle=\sum_{R_x,R_y=1}^{M_{x,y}}\frac{2\sin k_m(R_y-d_i)}{(M_xM_yA_{m\alpha})^{1/2}}\times\left(-\tau(-1)^\alpha\frac{\sin(k_{m\alpha}(R_x-d_i))}{\sin(k_{m\alpha}(M_x+1-(R_x+d_i)))}\right) \tag*{(3)}

where Mx=4Mx+1\mathcal{M}_x=4M_x+1, and τ=±1\tau=\pm1 is a conduction/valence band index, and d1=0d_1=0 and d2=1/2d_2=1/2. {m,α,τ,σ}\{m,\alpha,\tau,\sigma\} is a complete set of quantum numbers in the ribbon. The eigenstates are spin-degenerate, and we call the pair (mατ↑)(m\alpha\tau\uparrow), (mατ↓)(m\alpha\tau\downarrow) a shell from now on. The normalization factor is

Amα=F1(kmα)−δkm,πMxF2(kmα)(4)A_{m\alpha}=F^1(k_{m\alpha})-\frac{\delta_{k_m,\pi}}{\mathcal{M}_x}F^2(k_{m\alpha}) \tag*{(4)}
F1(k)=1−sin⁡(Mxk/2)Mxsin⁡(k/2)(5)F^1(k)=1-\frac{\sin(\mathcal{M}_x k/2)}{\mathcal{M}_x\sin(k/2)} \tag*{(5)}
F2(k)=1−cos⁡(Mxk/2)Mxcos⁡(k/2)(6)F^2(k)=1-\frac{\cos(\mathcal{M}_x k/2)}{\mathcal{M}_x\cos(k/2)} \tag*{(6)}

The diagonalized Hamiltonian and the single-particle eigen-energies are

H^0=∑mατϵmατ0n^mατ(7)\hat{H}^0=\sum_{m\alpha\tau}\epsilon^0_{m\alpha\tau}\hat{n}_{m\alpha\tau} \tag*{(7)}
ϵmατ0=τ1+Δm2+2Δmcos⁡kmα2(8)\epsilon^0_{m\alpha\tau}=\tau\sqrt{1+\Delta_m^2+2\Delta_m\cos\frac{k_{m\alpha}}{2}} \tag*{(8)}

where Δm=2cos⁡km2\Delta_m=2\cos\frac{k_m}{2}.

Extended Hubbard model of graphene

Bulk graphene

The extended Hubbard model is constructed by adding to H^0\hat{H}^0 the following interaction Hamiltonian:

H^I=U0∑in^i↑n^i↓+∑n,i,δnUnn^in^i+δn(9)\hat{H}^I=U^0\sum_i\hat{n}_{i\uparrow}\hat{n}_{i\downarrow}+\sum_{n,i,\delta^n}U^n\hat{n}_i\hat{n}_{i+\delta^n} \tag*{(9)}

The second sum extends to the different atomic neighbor shells nn, while δn\delta^n runs over all the atoms in the nthn^{\mathrm{th}}-neighbor shell of atom ii. The on-site and extended Hubbard parameters U0U^0 and UnU^n correspond to the effective Coulomb interactions among electrons residing in the wannierized pzp^z orbitals. These interactions correspond to the bare Coulomb interaction among pzp^z orbitals, screened by all the remaining higher energy orbitals, and eventually by the presence of a substrate and perhaps an overlayer. U0U^0 and UnU^n have been obtained for free-standing graphene and for bilayer graphene encapsulated or deposited on idealized metallic substrates in references [16, 18]. The reported Hubbard parameters UnU^n for free-standing graphene can be rewritten in terms of the distance rr between two sites and are very well fitted to the law [16, 18]

U(r)=10.31+0.6r eV(10)U(r)=\frac{10.3}{1+0.6r}\ \mathrm{eV} \tag*{(10)}

where rr must be given in Å. In addition, reference 17 has analyzed whether an on-site Hubbard model can describe free-standing graphene. The article finds that this is the case and reports an effective on-site U0≈4.3 eVU^0\approx4.3\ \mathrm{eV}, except for extreme doping cases very far away from half-filling.

Graphene ribbons

We rewrite in this section the interaction Hamiltonian H^I\hat{H}^I in the basis of single-particle eigen-states of the ribbon. Inserting the decomposition 2 into Eq. 9, we find a large sum of terms of the type

r^m1α1τ1σ1†r^m2α2τ2σ2†r^m3α3τ3σ3r^m4α4τ4σ4(11)\hat{r}_{m_1\alpha_1\tau_1\sigma_1}^{\dagger}\hat{r}_{m_2\alpha_2\tau_2\sigma_2}^{\dagger}\hat{r}_{m_3\alpha_3\tau_3\sigma_3}\hat{r}_{m_4\alpha_4\tau_4\sigma_4} \tag*{(11)}

that are multiplied by Coulomb integrals. However, many of these integrals are zero because momentum conservation implies the condition km1α1+km2α2=km3α3+km4α4k_{m_1\alpha_1}+k_{m_2\alpha_2}=k_{m_3\alpha_3}+k_{m_4\alpha_4}. The reason for the cancellation is that the wave-vectors kmk_m are not separated evenly for a ribbon, so the condition is not fulfilled except when k1=k3\mathbf{k}_1=\mathbf{k}_3 and k2=k4\mathbf{k}_2=\mathbf{k}_4, or when k1=k4\mathbf{k}_1=\mathbf{k}_4 and k2=k3\mathbf{k}_2=\mathbf{k}_3. In addition, the two single-particle bulk states having the same kmαk_{m\alpha} and opposite τ\tau are effectively unentangled because they possess very different energies. As a consequence, momentum conservation allows us to rewrite the following effective Hamiltonian for bulk ribbons states, which is of the Kanamori type [19]:

H^=∑aϵa0n^a+∑aVan^a↑n^a↓+∑a≠b(Vabnn^an^b−VabSS^a⋅S^b+VabΔΔ^a†Δ^b+Δ^b†Δ^a2)(12)\hat{H}=\sum_a\epsilon_a^0\hat{n}_a+\sum_aV_a\hat{n}_{a\uparrow}\hat{n}_{a\downarrow}+\sum_{a\ne b}\left(V_{ab}^{n}\hat{n}_a\hat{n}_b-V_{ab}^{S}\hat{\mathbf{S}}_a\cdot\hat{\mathbf{S}}_b+V_{ab}^{\Delta}\frac{\hat{\Delta}_a^\dagger\hat{\Delta}_b+\hat{\Delta}_b^\dagger\hat{\Delta}_a}{2}\right) \tag*{(12)}

where aa and bb are shorthands for the single-particle quantum numbers (mατ)(m\alpha\tau) and (m′α′τ′)(m'\alpha'\tau'). S^a=r^aσ†σσσ′2r^aσ′\hat{\mathbf{S}}_a=\hat{r}_{a\sigma}^{\dagger}\frac{\boldsymbol{\sigma}_{\sigma\sigma'}}{2}\hat{r}_{a\sigma'} are spin operators while Δ^a†=r^a↑†r^a↓†\hat{\Delta}_a^\dagger=\hat{r}_{a\uparrow}^\dagger\hat{r}_{a\downarrow}^\dagger are pair-hopping operators. The Hamiltonian parameters possess the following explicit expressions

εa=τεa0+Vaa(13)\varepsilon_a = \tau\varepsilon_a^0 + V_{aa} \tag*{(13)}
Va=Vaa0+Vaa(14)V_a = V_{aa}^0 + V_{aa} \tag*{(14)}
Vabn=Vab0+Vab4−Jab2(15)V_{ab}^n = \frac{V_{ab}^0 + V_{ab}}{4} - \frac{J_{ab}}{2} \tag*{(15)}
VabS=VabΔ=Vab0+2JabVabΔ(16)V_{ab}^S = V_{ab}^{\Delta} = V_{ab}^0 + 2J_{ab}V_{ab}^{\Delta} \tag*{(16)}

The direct and exchange matrix elements can be written as follows:

Vab=Vab1+Vab2+Vab3+…(17)V_{ab} = V_{ab}^1 + V_{ab}^2 + V_{ab}^3 + \ldots \tag*{(17)}
Jab=Jab1+Jab2+Jab3+…(18)J_{ab} = J_{ab}^1 + J_{ab}^2 + J_{ab}^3 + \ldots \tag*{(18)}

where each VabnV_{ab}^n and JabnJ_{ab}^n derives from the Hubbard terms Unn^in^i+δnU^n\hat{n}_i\hat{n}_{i+\delta_n} that connect sites separated by nn–shells.

The onsite-derived interactions are

Vmα;m′α′0=2CU0∑i=1,2∑nxsin⁡2(kmαnxi)sin⁡2(km′α′nxi)×∑nysin⁡2(kmnyi)sin⁡2(km′nyi)(19)V_{m\alpha;m'\alpha'}^0 = 2CU^0 \sum_{i=1,2}\sum_{n_x}\sin^2(k_{m\alpha}n_x^i)\sin^2(k_{m'\alpha'}n_x^i)\times\sum_{n_y}\sin^2(k_m n_y^i)\sin^2(k_{m'}n_y^i) \tag*{(19)}
C=16Mx2My2AmαAm′α′(20)C = \frac{16}{M_x^2M_y^2A_{m\alpha}A_{m'\alpha'}} \tag*{(20)}

with vectors n1=(nx,ny)\mathbf{n}^1=(n_x,n_y) and n2=(nx−1/2,ny−1/2)\mathbf{n}^2=(n_x-1/2,n_y-1/2). The explicit expressions for the shell-dependent matrix elements depend on whether UnU^n connects same- or opposite-sublattices. When UnU^n connects same-sublattice shells with neighbor vectors δn\boldsymbol{\delta}^n, we find

Vmα;m′α′n=CUn∑i=1,2∑δn∑nxsin⁡2(kmαnxi)sin⁡2(km′α′(nxi+δxn))∑nysin⁡2(kmnyi)sin⁡2(km′(nyi+δyn))(21)V_{m\alpha;m'\alpha'}^n = CU^n\sum_{i=1,2}\sum_{\delta^n}\sum_{n_x}\sin^2(k_{m\alpha}n_x^i)\sin^2(k_{m'\alpha'}(n_x^i+\delta_x^n))\sum_{n_y}\sin^2(k_m n_y^i)\sin^2(k_{m'}(n_y^i+\delta_y^n)) \tag*{(21)}
Jmα;m′α′n=CUn∑i=1,2∑δn∑nxsin⁡(kmαnxi)sin⁡(km′α′nxi)sin⁡(kmα(nxi+δxn))sin⁡(km′α′(nxi+δxn))×∑nysin⁡(kmnyi)sin⁡(km′nyi)sin⁡(km(nyi+δyn))sin⁡(km′(nyi+δyn))(22)\begin{aligned} J_{m\alpha;m'\alpha'}^n ={}& CU^n\sum_{i=1,2}\sum_{\delta^n}\sum_{n_x}\sin(k_{m\alpha}n_x^i)\sin(k_{m'\alpha'}n_x^i)\sin(k_{m\alpha}(n_x^i+\delta_x^n))\sin(k_{m'\alpha'}(n_x^i+\delta_x^n))\\ &\times\sum_{n_y}\sin(k_m n_y^i)\sin(k_{m'}n_y^i)\sin(k_m(n_y^i+\delta_y^n))\sin(k_{m'}(n_y^i+\delta_y^n)) \tag*{(22)} \end{aligned}

In contrast, when UnU^n connects opposite-sublattice shells with neighbor vectors δn\boldsymbol{\delta}^n, we find

Vmα;m′α′n=CUn∑i=1,2∑δn∑nxsin⁡2(kmαnxi)sin⁡2(km′α′(Mx+1−(nxi+δxn)))∑nysin⁡2(kmnyi)sin⁡2(km′(nyi+δyn))(23)V_{m\alpha;m'\alpha'}^n = CU^n\sum_{i=1,2}\sum_{\delta^n}\sum_{n_x}\sin^2(k_{m\alpha}n_x^i)\sin^2(k_{m'\alpha'}(M_x+1-(n_x^i+\delta_x^n)))\sum_{n_y}\sin^2(k_m n_y^i)\sin^2(k_{m'}(n_y^i+\delta_y^n)) \tag*{(23)}
Jmα;m′α′n=CUn∑i=1,2∑δn∑nxsin⁡(kmαnxi)sin⁡(km′α′nxi)sin⁡(kmα(Mx+1−(nxi+δxn)))sin⁡(km′α′(Mx+1−(nxi+δxn)))×∑nysin⁡(kmnyi)sin⁡(km′nyi)sin⁡(km(nyi+δyn))sin⁡(km′(nyi+δyn))(24)\begin{aligned} J_{m\alpha;m'\alpha'}^n ={}& CU^n\sum_{i=1,2}\sum_{\delta^n}\sum_{n_x}\sin(k_{m\alpha}n_x^i)\sin(k_{m'\alpha'}n_x^i)\sin(k_{m\alpha}(M_x+1-(n_x^i+\delta_x^n)))\sin(k_{m'\alpha'}(M_x+1-(n_x^i+\delta_x^n)))\\ &\times\sum_{n_y}\sin(k_m n_y^i)\sin(k_{m'}n_y^i)\sin(k_m(n_y^i+\delta_y^n))\sin(k_{m'}(n_y^i+\delta_y^n)) \tag*{(24)} \end{aligned}

Finally, the Hubbard parameters UnU^n are well-fitted by Eq. (10) for free-standing bulk graphene [16, 18], but are unknown for graphene ribbons. A possible strategy consists of borrowing and using the bulk parameters in Eq. (10).

On-site Hubbard model of graphene ribbons

Determination of the onsite Hubbard parameter U0U^0

We follow in the rest on this article a second strategy, which consists of reducing the parameters of the extended Hubbard model down to an onsite renormalized interaction U0U^0. This strategy was implemented for bulk graphene in Ref. 17, where the authors found a renormalized U0≃4.3U^0 \simeq4.3 eV. We attempt a similar task in this section for graphene ribbons. We make use of the fact that DFT is a mean field theory akin in spirit to performing a Hartree decomposition of the on-site Hubbard model. We perform DFT and Hartree decompositions of the onsite Hubbard model for armchair ribbons with My=4M_y=4, 5, 9 (corresponding to N=7N=7, 9, 17) and lengths ranging from Mx=2M_x=2 up to infinity. We find that an onsite Hubbard U0=4.9U^0=4.9 eV fits the DFT bulk gap and the lowest-lying electron and hole excitations (also called HOMO-1, HOMO, LUMO and LUMO+1) rather closely, as demonstrated in Tables I, II and III. We therefore choose U0=4.9U^0=4.9 eV for the remaining of this article.

Notice that the matrix elements in Eqs. (13)-(18) simplify for the on-site Hubbard model and fulfill 4Vabn=VabS=2VabΔ=Vab04V_{ab}^n=V_{ab}^S=2V_{ab}^{\Delta}=V_{ab}^0. so that the Hamiltonian in Eq. (12) simplifies to

MxM_xDFT H/LMFH H/LDFT H-1/L+1MFH H-1/L+1
20.9430.9583.8093.775
30.5960.6003.1273.199
40.5560.5342.5452.650
50.5460.5042.2102.237
60.5450.4912.0171.975
70.5460.4841.8981.801
80.5470.4821.8171.681
90.5470.4811.7601.595
100.5470.4801.7191.532

Table 1. HOMO-LUMO gaps in eV for My=4M_y = 4, N=7N = 7 armchair GNRs of different lengths MxM_x, as obtained from our Density Functional Theory (DFT) and antiferromagnetic mean-field Hubbard (MFH) calculations with U0=4.9U^0 = 4.9 eV. The infinite DFT gap is 1.559 eV.

MxM_xDFT H/LMFH H/LDFT H-1/L+1MFH H-1/L+1
20.6990.7452.7602.662
30.6630.6782.0752.063
40.6600.6721.7131.749
50.6590.6581.4981.558
60.6580.6581.3571.431
70.6570.6571.2581.341
80.6580.6571.1861.275
90.6570.6571.1331.225
100.6580.6571.0881.186

Table 2. Same as in Table I, with My=5M_y = 5, N=9N = 9. The infinite DFT gap is in this case 0.830 eV.

H^=∑a(ϵa0n^a+Vaa0n^a↑n^a↓)+∑a≠bVab0(n^an^b4−S^a⋅S^b+Δ^a†Δ^b+Δ^b†Δ^a2)(25)\hat{H} = \sum_a \left(\epsilon_a^0 \hat{n}_a + V_{aa}^0 \hat{n}_{a\uparrow}\hat{n}_{a\downarrow}\right) + \sum_{a\ne b} V_{ab}^0 \left(\frac{\hat{n}_a\hat{n}_b}{4} - \hat{\mathbf{S}}_a \cdot\hat{\mathbf{S}}_b + \frac{\hat{\Delta}_a^\dagger\hat{\Delta}_b + \hat{\Delta}_b^\dagger\hat{\Delta}_a}{2}\right) \tag*{(25)}

A further advantage of this model is that the on-site matrix elements in Eq. (19) can be computed analytically [11, 13]. This allows us to determine them even for ribbons with sizes of the order of micrometers. The expressions are

Vmα,m′α′0=Dδmm′+Cmα,m′α′B,++δmm′−Cmα,m′α′B,−/MxΛmαϕΛm′α′ϕ(26)V_{m\alpha,m'\alpha'}^0 = \mathcal{D}\frac{\delta_{mm'}^+ C_{m\alpha,m'\alpha'}^{B,+} + \delta_{mm'}^- C_{m\alpha,m'\alpha'}^{B,-}/M_x}{\Lambda_{m\alpha}^{\phi}\Lambda_{m'\alpha'}^{\phi}} \tag*{(26)}
MxM_xDFT H/LMFH H/LDFT H-1/L+1MFH H-1/L+1
20.7930.7060.8480.931
30.6240.6090.8840.919
40.5920.5790.8920.914
50.5850.5700.8040.807
60.5840.5670.6660.678
70.5660.5660.5830.585
80.4900.5140.5840.566
90.4310.4580.5840.566
100.3840.4140.5840.566

Table 3. Same as in Table I, with My=9M_y = 9, N=17N = 17. The infinite DFT gap is 0.114 eV.

where

D=U02MxMy(27)\mathcal{D} = \frac{U^0}{2M_x M_y} \tag*{(27)}
δmm′+=1+12(δm,m′+δm,Myδm′,My)(28)\delta_{mm'}^+ = 1 + \frac{1}{2}\left(\delta_{m,m'} + \delta_{m,M_y}\delta_{m',M_y}\right) \tag*{(28)}
δmm′−=12δm+m′,My−(δm,My+δm′,My)(29)\delta_{mm'}^- = \frac{1}{2}\delta_{m+m',M_y} - \left(\delta_{m,M_y} + \delta_{m',M_y}\right) \tag*{(29)}
Cmα,m′α′B,+=2F1(kmα)+2F1(km′α′)−F1(k+)−F1(k−)(30)C_{m\alpha,m'\alpha'}^{B,+} = 2F^1(k_{m\alpha}) + 2F^1(k_{m'\alpha'}) - F^1(k_+) - F^1(k_-) \tag*{(30)}
Cmα,m′α′B,−=2F2(kmα)+2F2(km′α′)−F2(k+)−F2(k−)(31)C_{m\alpha,m'\alpha'}^{B,-} = 2F^2(k_{m\alpha}) + 2F^2(k_{m'\alpha'}) - F^2(k_+) - F^2(k_-) \tag*{(31)}

with k±=kmα±km′α′k_{\pm} = k_{m\alpha} \pm k_{m'\alpha'}.

Upon computing numerically the matrix elements Vmα;m′α′0V_{m\alpha;m'\alpha'}^0 for ribbons of different sizes, we come to the conclusion that they collapse into three types that only depend on the ribbon size: same-state integrals Vmα0=Vaa0=U(Mx,My)V_{m\alpha}^0 = V_{aa}^0 = \mathcal{U}(M_x,M_y), same-band integrals Vmα,mα′0=V(Mx,My)V_{m\alpha,m\alpha'}^0 = \mathcal{V}(M_x,M_y) and the rest Vmα,m′α′0=W(Mx,My)V_{m\alpha,m'\alpha'}^0 = \mathcal{W}(M_x,M_y). Consequently, only three Coulomb integrals are needed for each ribbon size. An example of this behaviour for an armchair ribbon of size My=5M_y = 5 (N=9N = 9) and Mx=80M_x = 80 is shown in Figure 2. We have also found that these three Coulomb integrals follow closely the ratios V=2/3U\mathcal{V} = 2/3\mathcal{U} and W=2/3V\mathcal{W} = 2/3\mathcal{V}, up to an accuracy of less than 1 per cent, as can also be inferred from the same figure. As a consequence, it suffices to determine one of the three, and we choose U\mathcal{U}. We have fitted U\mathcal{U} for armchair ribbons with widths and lengths ranging from My=3M_y = 3 to 9 and Mx=3M_x = 3 to 110 respectively, and have found that the size-scaling relationship

Histograms of the Coulomb matrix elements

Figure 2. Histograms of the Coulomb matrix elements of an armchair ribbon of width My=5M_y = 5 (N=9N = 9) and length Mx=80M_x = 80 (L=34.1 nmL = 34.1\ \mathrm{nm}), as a function of energy measured in meV, for the four bands m=1,…,4m = 1,\ldots,4, and all wave-vectors α=1,…,2Mx−1\alpha= 1,\ldots,2M_x-1. VmαV_{m\alpha} is shown in black color; the peak is placed at 6.8 meV. Intra-band Vmα,mα′V_{m\alpha,m\alpha'} and inter-band Vmα,m′≠mα′V_{m\alpha,m'\ne m\alpha'} are shown in red and green colors, respectively. The two peaks are placed at 4.53 and 3.02 meV.

U=3.19(1+1.156Mx)My eV.(32)\mathcal{U} = \frac{3.19}{(1 + 1.156M_x)M_y}\ \mathrm{eV}. \tag*{(32)}

is accurate up to one or two percent.

Mean-field solution for undoped armchair ribbons

We have performed the Kanamori mapping for undoped ribbons in a previous article [13]. There, we kept both edge and bulk states but dropped the exchange and pair-hopping terms. We performed an analytical mean field treatment of the Hamiltonian and found paramagnetic, ferromagnetic and antiferromagnetic solutions. These solutions agreed with previous numerical calculations carried out for infinite length ribbons, both for the bulk and for the edge bands.

bulk and for the edge bands. We also found that the three solutions delivered the same properties for bulk states, and only differed on the nature and energies of the edge states. Specifically, mean-field bulk states kept a one-to-one relationship with their non-interacting counterparts: they retained the quantum numbers (m α τ σ)(m\ \alpha\ \tau\ \sigma) and were spin-degenerate. All in all, bulk states could still be classified by shells labeled by (m α τ)(m\ \alpha\ \tau). We also found that the filled bulk mean field states undergo an energy shift

ϵmασ0⟶ϵmασMF=ϵmα0+Unmα−σ+∑m′α′occVmα;m′α′0nm′α′(33)\epsilon^0_{m\alpha\sigma} \longrightarrow\epsilon^{MF}_{m\alpha\sigma} = \epsilon^0_{m\alpha} + U n_{m\alpha-\sigma} + \sum_{\substack{m'\alpha'\\ \mathrm{occ}}} V^0_{m\alpha;m'\alpha'} n_{m'\alpha'} \tag*{(33)}

where the sum runs over all occupied states. The energy shift is essentially the same for all bulk states, and so it amounts to a rigid shift of all the bulk bands. These mean field bulk energies are illustrated in Figure 3 for a My=5M_y = 5 ribbon of length Mx=80M_x = 80. The mean-field eigen-energies can be used to determine the single-particle addition energies Δϵmα,m′α′=ϵmαMF−ϵm′α′MF\Delta\epsilon_{m\alpha,m'\alpha'} = \epsilon^{MF}_{m\alpha} - \epsilon^{MF}_{m'\alpha'} between consecutive-energy states. Notice that the two single-particle energies can belong to the same band or two different bands. The first case leads to a uniform pattern, while the second follows a random distribution. This behavior is illustrated in Figure 4 for the same ribbon, with My=5M_y = 5 and Mx=80M_x = 80. Panel (a) shows Δϵab\Delta\epsilon_{ab}, where the uniform patterns belonging to same-band pairs can be appreciated; superimposed to them, the random cloud of addition energies correspond to different-band pairs. Panel (b) shows the histogram of addition energies. This histogram resembles a Wigner-Dyson random distribution, although we do not pursue this issue further due to the limited number of counts. Instead, we would like to highlight the large number of counts with Δϵab<U≈6.8\Delta\epsilon_{ab} < U \approx6.8 meV. This means that the electronic correlations of ribbons of these widths and lengths belong to the intermediate regime, and may lead to interesting behavior beyond that expected for weak electron correlations. This intermediate regime behavior will be exploited and analysed in the rest of this article.

Bulk eigen-energies of the antiferromagnetic mean field solution

Figure 3. Bulk eigen-energies ϵmαMF\epsilon^{MF}_{m\alpha} of the antiferromagnetic mean field solution of an armchair ribbon of width My=5M_y = 5 (N=9N = 9) and length Mx=80M_x = 80 (L=34.1 nmL = 34.1\ \mathrm{nm}) [13]. These eigen-energies are discrete; they resemble bands because of the close spacing of the wave-vectors kmαk_{m\alpha} for this ribbon.

Mean-field single-particle energy differences and histogram

Figure 4. (a) Mean-field single-particle energy differences Δϵab=ϵaMF−ϵbMF\Delta\epsilon_{ab} = \epsilon^{MF}_a - \epsilon^{MF}_b between consecutive-energy bulk states belonging to the valence band, all extracted from Figure 3. The horizontal red line is placed at an energy equal to U=6.8U = 6.8 meV, which corresponds to the Coulomb matrix element UU of this ribbon. (b) Histogram of the same energy differences.

Low-energy Hamiltonian

Hamiltonian

We remind that we analyse in this article situations where the chemical potential μ\mu lies well within the bands of bulk single-particle states. With these conditions, bulk shells with energies sufficiently below μ\mu are filled and inert, and we call them closed shells. Similarly, bulk shells with energies sufficiently above μ\mu are empty and inert, and similar words can be said for the ribbon edge shells, provided that they exist. Inert means here that virtual transitions towards/from those states, triggered by either quantum or thermal fluctuations, have negligible probabilities. As a consequence, we expect that only a few shells will be active for a given value of the chemical potential, and we call them open shells. Therefore, when a gate voltage is swept, shells are opened and closed sequentially.

We expect that a mean-field treatment should be sufficient to describe closed shells. However, shells that are close enough to the chemical potential might be open or closed depending of the delicate interplay between kinetic and Coulomb energies. We thus discard from the Kanamori Hamiltonian those closed shells whose mean-field energies are sufficiently far from the chemical potential and keep only a small number OO of shells with mean-field energies around μ\mu. We thus arrive to the following low-energy many-body hamiltonian that addresses cleanly these OO putative open shells

H^=∑a∈OϵaMFn^a+U∑a∈On^a↑n^a↓+∑a,b∈OVab0(n^an^b4−S^a⋅S^b+Δ^a†Δ^b+Δ^b†Δ^a2)(34)\hat{H} = \sum_{a\in O} \epsilon_a^{MF}\hat{n}_a + \mathcal{U}\sum_{a\in O}\hat{n}_{a\uparrow}\hat{n}_{a\downarrow} + \sum_{a,b\in O} V_{ab}^{0}\left(\frac{\hat{n}_a\hat{n}_b}{4}-\hat{\mathbf{S}}_a\cdot\hat{\mathbf{S}}_b+\frac{\hat{\Delta}_a^\dagger\hat{\Delta}_b+\hat{\Delta}_b^\dagger\hat{\Delta}_a}{2}\right) \tag*{(34)}
ϵaMF=ϵa0+∑b∈occVab0nb(35)\epsilon_a^{MF} = \epsilon_a^0 + \sum_{b\in occ} V_{ab}^{0}n_b \tag*{(35)}

where the same-state, opposite-spin interaction of the open shells is kept out of the mean-field treatment. The Coulomb integrals Vab0V_{ab}^{0} are equal to V=2/3 U\mathcal{V}=2/3\,\mathcal{U} or W=(2/3)2U\mathcal{W}=(2/3)^2\mathcal{U} depending on whether aa and bb belong or not to the same band, respectively.

The Fock space is spanned in the single-particle number basis {∣niσ)=∣n1↑,n1↓,…,nO↑,nO↓⟩}\{|n_{i\sigma})=|n_{1\uparrow},n_{1\downarrow},\ldots,n_{O\uparrow},n_{O\downarrow}\rangle\} and has dimension 4O4^O. The many-body Hamiltonian commutes with the total number operator N^\hat{N} and with the total spin operators S^T2\hat{S}_T^2, S^Tz\hat{S}_T^z. As a consequence the Fock space can be classified into 2O+12O+1 sub-spaces labeled by N=0,…,2ON=0,\ldots,2O. Each of these subspaces can be further split into subspaces with given SS. We find that there exists frequently more than one subspace with quantum numbers (N,S)(N,S). Therefore, an additional quantum number RR needs to be added to identify unambiguously each subspace (N,S,R)(N,S,R). Each of these subspaces contains 2S+12S+1 eigenstates. The highest spin subspace available occurs at half-filling N=ON=O and the spin equals ST=O/2S_T=O/2.

The ribbon many-body eigen-states are decomposed in the single-particle basis as

∣Nλ⟩=∑∣niσ⟩(niσ∣Nλ⟩)=∑∣niσ⟩Rniσ,λN(36)|N\lambda\rangle= \sum|n_{i\sigma}\rangle(n_{i\sigma}|N\lambda\rangle)=\sum|n_{i\sigma}\rangle\mathcal{R}^{N}_{n_{i\sigma},\lambda} \tag*{(36)}

where λ\lambda is a shorthand for the many body quantum numbers (R,S,Sz)(R,S,S^z), and where the sum is restricted to the combinations that sum up to NN. The expansion coefficients define the matrices RN\mathcal{R}^{N} that diagonalize the Hamiltonian within each NN subspace. The diagonalized Hamiltonian is

H^=∑N,λEN,λΓ^N,λ†Γ^N,λ(37)\hat{H}=\sum_{N,\lambda} E_{N,\lambda}\hat{\Gamma}_{N,\lambda}^{\dagger}\hat{\Gamma}_{N,\lambda} \tag*{(37)}

where the linear operator Γ^Nλ†\hat{\Gamma}_{N\lambda}^{\dagger} creates the ribbon eigen-state (N,λ)(N,\lambda).

The low-energy many-body Hamiltonian describes adequately the competition between open- and closed-shell behavior, beyond a mean field treatment. This competition is driven by the interplay between Δϵab=ϵbMF−ϵaMF\Delta\epsilon_{ab}=\epsilon_b^{MF}-\epsilon_a^{MF}, U\mathcal{U} and Vab0V_{ab}^{0}. We discuss below the conditions needed for two shells aa, bb to be open. The exchange term in Hamiltonian 34 promotes that STS_T is maximized if the two shells are open, because Vab0V_{ab}^{0} is always positive.

The low-energy Hamiltonian can be diagonalized analytically for O=2O=2 or 3, but becomes analytically unmanageable for larger OO because of the exponential size growth of its Fock space. However, the mechanisms leading to open shell and high-spin behavior are captured already for O=2O=2 and O=3O=3. These mechanisms are discussed in the following two sections. The conditions that will be found below for the existence of open-shell high-spin states are equivalent to similar conditions that have been established for the Hubbard model [22, 23]. Those results are also akin to the mesoscopic Stoner instabilities predicted to exist in open quantum dots [24, 27].

Two shell model

The two-shell model illustrates how more than one shell can be open at the same time. The Hamiltonian is

H^=ϵ1n^1+Un^1↑n^1↓+ϵ2n^2+Un^2↑n^2↓\hat{H}=\epsilon_1\hat{n}_1+\mathcal{U}\hat{n}_{1\uparrow}\hat{n}_{1\downarrow}+\epsilon_2\hat{n}_2+\mathcal{U}\hat{n}_{2\uparrow}\hat{n}_{2\downarrow}
+V12(n^1n^22−2S^1⋅S^2+Δ^1†Δ^2+Δ^2†Δ^1)(38)\quad+V_{12}\left(\frac{\hat{n}_1\hat{n}_2}{2}-2\hat{\mathbf{S}}_1\cdot\hat{\mathbf{S}}_2+\hat{\Delta}_1^\dagger\hat{\Delta}_2+\hat{\Delta}_2^\dagger\hat{\Delta}_1\right) \tag*{(38)}

where we have dropped the super-indices MFMF and 00, and where we assume that ϵ1<ϵ2\epsilon_1<\epsilon_2. In addition, V12V_{12} can be equal to V\mathcal{V} or W\mathcal{W}, depending on whether 1 and 2 belong to the same band or not. The Fock space has five subspaces subspaces labeled by N=0,1,2,3,4N=0,1,2,3,4. The highest spin subspace ST=1S_T=1 is achieved at half-filling N=2N=2. The ground state for N=1N=1 corresponds to the open-shell doublet ∣1,0,0,0⟩|1,0,0,0\rangle, ∣0,1,0,0⟩|0,1,0,0\rangle with energy E1,1/2,1/2,±1/2=E_{1,1/2,1/2,\pm1/2}= ϵ1\epsilon_1. The subspace N=2N = 2 has two closed-shell singlets with energies and wavefunctions

E2,1−2,0,0=ϵ1+ϵ2+U±(Δϵ21)2+V122(39)E_{2,1-2,0,0} = \epsilon_1 + \epsilon_2 + U \pm\sqrt{(\Delta\epsilon_{21})^2 + V_{12}^2} \tag*{(39)}
∣2,1−2,0,0⟩=A+∣1,1,0,0⟩±A−∣0,0,1,1⟩(40)\lvert2,1-2,0,0\rangle= A_+\lvert1,1,0,0\rangle\pm A_-\lvert0,0,1,1\rangle \tag*{(40)}
A±=12(1∓Δϵ21(Δϵ21)2+V122)1/2(41)A_{\pm} = \frac{1}{\sqrt{2}}\left(1 \mp\frac{\Delta\epsilon_{21}}{\sqrt{(\Delta\epsilon_{21})^2 + V_{12}^2}}\right)^{1/2} \tag*{(41)}

with Δϵ21=ϵ2−ϵ1\Delta\epsilon_{21} = \epsilon_2 - \epsilon_1. The singlet ∣2,2,0,0⟩\lvert2,2,0,0\rangle has a higher weight in the single-particle state ∣1,1,0,0⟩\lvert1,1,0,0\rangle and vice-versa. The subspace also contains an open-shell triplet

E2,1,1,1−3=ϵ1+ϵ2(42)E_{2,1,1,1-3} = \epsilon_1 + \epsilon_2 \tag*{(42)}
∣2,1,1,1⟩=∣1,0,1,0⟩(43)\lvert2,1,1,1\rangle= \lvert1,0,1,0\rangle \tag*{(43)}
∣2,1,1,0⟩=∣1,0,0,1⟩+∣0,1,1,0⟩2(44)\lvert2,1,1,0\rangle= \frac{\lvert1,0,0,1\rangle+ \lvert0,1,1,0\rangle}{\sqrt{2}} \tag*{(44)}
∣2,1,1,−1⟩=∣0,1,0,1⟩(45)\lvert2,1,1,-1\rangle= \lvert0,1,0,1\rangle \tag*{(45)}

and an open-shell singlet

E2,3,0,0=ϵ1+ϵ2+2V12(46)E_{2,3,0,0} = \epsilon_1 + \epsilon_2 + 2V_{12} \tag*{(46)}
∣2,3,0,0⟩=∣1,0,0,1⟩−∣0,1,1,0⟩2(47)\lvert2,3,0,0\rangle= \frac{\lvert1,0,0,1\rangle- \lvert0,1,1,0\rangle}{\sqrt{2}} \tag*{(47)}

The open shell singlet-triplet energy difference defines the exchange energy

J=12(Esinglet−Etriplet)=V12(48)J = \frac{1}{2}(E_{\mathrm{singlet}} - E_{\mathrm{triplet}}) = V_{12} \tag*{(48)}

where the convention is that J>0J > 0 means ferromagnetic alignment.

On the other hand, the ground state is either the closed shell singlet ∣2,2,0,0⟩\lvert2,2,0,0\rangle or the open shell triplet ∣2,1,1,1−3⟩\lvert2,1,1,1-3\rangle depending on whether the effective exchange constant

J=12(Esinglet′−Etriplet)=12(U−(Δϵ21)2+V122)(49)\mathcal{J} = \frac{1}{2}(E'_{\mathrm{singlet}} - E_{\mathrm{triplet}}) = \frac{1}{2}\left(U - \sqrt{(\Delta\epsilon_{21})^2 + V_{12}^2}\right) \tag*{(49)}

is smaller or larger than zero, respectively. Notice that if the two shells are open at the same time, then the effective exchange is positive (e.g.: ferromagnetic) and the ground state has high spin S=1S = 1. The full level ordering in the N=2N = 2 subspace is illustrated in Figure 5.

Illustration of the level ordering for the N = 2 subspace of the two shell model

Figure 5. Illustration of the level ordering for the N=2N = 2 subspace of the two shell model.

Three shell model

The three-shell model describes how a higher spin S=3/2S = 3/2 can emerge in an armchair ribbon. The Hamiltonian

is

H^=∑i=1,2,3(ϵin^i+Un^i↑n^i↓)+V12(n^1n^22−2S^1⋅S^2+Δ^1†Δ^2+Δ^2†Δ^1)+V13(n^1n^32−2S^1⋅S^3+Δ^1†Δ^3+Δ^3†Δ^1)+V23(n^2n^32−2S^2⋅S^3+Δ^2†Δ^3+Δ^3†Δ^2)(50)\hat{H} = \sum_{i=1,2,3}\left(\epsilon_i\hat{n}_i + U\hat{n}_{i\uparrow}\hat{n}_{i\downarrow}\right) + V_{12}\left(\frac{\hat{n}_1\hat{n}_2}{2} - 2\hat{\mathbf{S}}_1\cdot\hat{\mathbf{S}}_2 + \hat{\Delta}_1^\dagger\hat{\Delta}_2 + \hat{\Delta}_2^\dagger\hat{\Delta}_1\right) + V_{13}\left(\frac{\hat{n}_1\hat{n}_3}{2} - 2\hat{\mathbf{S}}_1\cdot\hat{\mathbf{S}}_3 + \hat{\Delta}_1^\dagger\hat{\Delta}_3 + \hat{\Delta}_3^\dagger\hat{\Delta}_1\right) + V_{23}\left(\frac{\hat{n}_2\hat{n}_3}{2} - 2\hat{\mathbf{S}}_2\cdot\hat{\mathbf{S}}_3 + \hat{\Delta}_2^\dagger\hat{\Delta}_3 + \hat{\Delta}_3^\dagger\hat{\Delta}_2\right) \tag*{(50)}

with ϵ1<ϵ2<ϵ3\epsilon_1 < \epsilon_2 < \epsilon_3. In addition, VabV_{ab} can be equal to V\mathcal{V} or W\mathcal{W}, depending on whether aa and bb belong to the same band or not. The Hamiltonian is diagonalized in detail in the appendix. The Fock space has seven subspaces labeled by N=0,1,…,6N = 0,1,\ldots,6. The highest spin subspace ST=3/2S_T = 3/2 is achieved at half-filling N=3N = 3.

We discuss now the conditions for shell opening and closing for the N=2N = 2 subspace. The subspace contains three triplet (S=1S = 1) and six singlet eigen-states. We analyse the limit ϵ3≫ϵ1,2\epsilon_3 \gg\epsilon_{1,2} to provide a simpler discussion. The two relevant shells are then i=1,2i = 1,2. We find that the lowest singlet corresponds always to a close shell situation, while the triplets correspond to open shells. The energy difference between the lowest-lying triplet and singlet states then tells whether the two shells are open or closed, and also defines the effective exchange constant

J=12(Esinglet−Etriplet)=U−(Δϵ21)2+V122(51)\mathcal{J} = \frac{1}{2}(E_{\mathrm{singlet}} - E_{\mathrm{triplet}}) = U - \sqrt{(\Delta\epsilon_{21})^2 + V_{12}^2} \tag*{(51)}

In other words, if the two shells are open, then the effective exchange is ferromagnetic J<0\mathcal{J} < 0, and vice versa.

The sub-space N=3N = 3 corresponds to the half-filled model. It contains 20 states that are split into 8 doublets with S=1/2S = 1/2 and a quadruplet with S=3/2S = 3/2. Six doublets contain one fully occupied, one half-occupied and one empty shell. The remaining two have three open shells but their eigen-energies are too large. The quadruplet corresponds to the three shells being open at the same time. The quadruplet is the ground state if the conditions

U+Vab+Vcb2>(Δϵac+Vab−Vcb2)2+(Vac)2(52)\mathcal{U}+\frac{V_{ab}+V_{cb}}{2}>\sqrt{\left(\Delta\epsilon_{ac}+\frac{V_{ab}-V_{cb}}{2}\right)^2+(V_{ac})^2} \tag*{(52)}

for the three possible configurations a≠b≠c=1,2,3a\ne b\ne c=1,2,3 are fulfilled at the same time. The tightest of these conditions defines the energy difference between the lowest quadruplet and the lowest doublet, and the effective exchange constant J\mathcal{J}. So if the three shells are open, then J\mathcal{J} is ferromagnetic; otherwise one shell is closed, another is empty and a third is half-filled. In addition, several of the doublet states contain one full or one empty shell. The model could in these cases be reduced to a simpler two-shell model. Furthermore, this feature illustrates the mechanism by which a model containing a large number of shells folds into a simpler effective model with fewer shells.

The discussion above can be extended to larger OO-shell models to determine the conditions for the existence of spin-O/2O/2 states, as well as the probability of finding them in a given GNR of size (Mx,My)(M_x,M_y). The exponential growth of the Fock space prevents however the analytical determination of these conditions and probabilities. The whole discussion, however sheds strong indications that states with spin as large as 3 or even 4 may be found in GNR.

Example: a N=9 armchair ribbon

To provide a more explicit example, we discuss the armchair GNR with N=9N=9, Mx=80M_x=80 introduced in Figure 3 in view of the three shell model, although this discussion applies to any ribbon with similar widths and lengths, as shown in Table IV. Our estimates indicate that this free-standing ribbon has a Coulomb U≈6.8\mathcal{U}\approx6.8 meV. Shell pairs or triples with energy differences smaller than 6.8 meV are candidates to show open shell high-spin behavior, according to Equs. 51 and 52. The two panels in Figure 4 show that a significant fraction of shell pairs may fulfill these conditions. We have determined that the fraction of shells fulfilling the condition 51 is PS=1≈29P_{S=1}\approx29 per cent. Similarly, we have also found that a fraction PS=3/2∼17P_{S=3/2}\sim17 per cent of the shells fulfill the three conditions in Eq. (52) for the existence of S=3/2S=3/2 states in this ribbon. We have finally computed these two fractions for GNR of different sizes to estimate the likelihood to find high-spin open-shell behavior. The results are summarized in Table IV and demonstrate that GNRs should contain a non-negligible number of open shells. The table indicates that the chances to display double and triple open shell behavior increase, but saturate, with the length and width of the ribbon.

MxM_x10100500
MyM_y
422 (12) / 7 (0)28 (14) / 15 (4)28 (15) / 17 (6)
529 (14) / 13 (4)30 (16) / 19 (6)31 (17) / 19 (7)
934 (20) / 19 (4)36 (17) / 24 (8)36 (21) / 25 (9)
1631 (18) / 19 (3)39 (21) / 27 (8)40 (22) / 29 (10)
2634 (20) / 24 (8)41 (23) / 30 (11)42 (24) / 31 (12)
3638 (19) / 24 (7)41 (23) / 30 (11)42 (24) / 32 (12)

Table IV. Percentage of states fulfilling the conditions for S=1S=1 / S=3/2S=3/2 for the half-filled two- / three-shell models in a doped armchair GNR of length MxM_x and width MyM_y. Numbers inside the parentheses indicate the percentages when an effective dielectric constant ε=2\varepsilon=2 is taken into account.

Graphene ribbons are however experimentally addressed when they are deposited on a substrate. Furthermore, in electrical experiments, the ribbons are contacted at the two sides by metallic electrodes. To provide a phenomenological understanding of these substrate and contact effects, we have included an effective dielectric constant ε\varepsilon in our calculations, so that Ueff=U0/εU^{\mathrm{eff}}=U^0/\varepsilon. Table IV shows in parentheses the percentages found when ε=2\varepsilon=2. Dielectric screening reduces the fraction of open shells because it decreases the size of the effective Coulomb matrix elements.

Electronic transport through doped graphene nanoribbons

Graphene ribbons may have use in different technologies, and we single out here their possible use as channel components in nanoelectronics [5, 6, 7, 20]. Electronic transport through a ribbon connected to two electrodes A=LA=L, RR is governed by the transfer Hamiltonian

H^t=∑A∑k,σ(VA,k,i,σr^iσ†c^A,kσ+h.c.)=∑A∑k,σ,λ,λ′(VA,kσ;Nλ→N+1λ′Γ^N+1,λ′†Γ^Nλc^A,kσ+h.c.)(53)\begin{aligned} \hat{H}_t=\sum_A\sum_{k,\sigma}\left(\mathcal{V}_{A,k,i,\sigma}\hat{r}_{i\sigma}^{\dagger}\hat{c}_{A,k\sigma}+h.c.\right) \\ &=\sum_A\sum_{k,\sigma,\lambda,\lambda'}\left(\mathcal{V}_{A,k\sigma;N\lambda\to N+1\lambda'}\hat{\Gamma}_{N+1,\lambda'}^{\dagger}\hat{\Gamma}_{N\lambda}\hat{c}_{A,k\sigma}+h.c.\right) \tag*{(53)} \end{aligned}

where c^A,kσ†\hat{c}_{A,k\sigma}^{\dagger} creates an electron with spin σ\sigma and wavevector kk in electrode AA and VA,k,i,σ\mathcal{V}_{A,k,i,\sigma} denotes a single-particle transfer matrix element. The matrix elements are written in the many-body eigen-state basis as follows

VA,kσ;Nλ→N+1λ′=∑iVA,k,i,σMλ′λN(i,σ)(54)\mathcal{V}_{A,k\sigma;N\lambda\to N+1\lambda'}=\sum_i\mathcal{V}_{A,k,i,\sigma}M_{\lambda'\lambda}^{N}(i,\sigma) \tag*{(54)}
Mλ′λN(i,σ)=⟨N+1λ′∣r^iσ†∣Nλ⟩=(RN+1)λ′n′⟨n′∣r^iσ†∣n⟩RλnN(55)\begin{aligned} M_{\lambda'\lambda}^{N}(i,\sigma)=\langle N+1\lambda'|\hat{r}_{i\sigma}^{\dagger}|N\lambda\rangle\\ &=\left(R^{N+1}\right)_{\lambda'n'}\langle n'|\hat{r}_{i\sigma}^{\dagger}|n\rangle R_{\lambda n}^{N} \tag*{(55)} \end{aligned}

where we have made use of Eq. (36). The transfer rates in the Single Electron Transfer regime (SET) are

γA,Nλ→N+1λ′=2πℏgA(μA)f(EN+1λ′−ENλ−μA)××∑iσ∣VA,kF,i,σ∣2∣Mλ′λN(i,σ)∣2(56)\begin{aligned} \gamma_{A,N\lambda\to N+1\lambda'}=\frac{2\pi}{\hbar}g_A(\mu_A)f(E_{N+1\lambda'}-E_{N\lambda}-\mu_A)\times\\ &\times\sum_{i\sigma}\left|\mathcal{V}_{A,k_F,i,\sigma}\right|^2\left|M_{\lambda'\lambda}^{N}(i,\sigma)\right|^2 \tag*{(56)} \end{aligned}
γA,Nλ→N−1λ′=2πℏgA(μA)(1−f(ENλ−EN−1λ′−μA))××∑iσ∣VA,kF,i,σ∣2∣Mλ′λN−1(i,σ)∣2(57)\begin{aligned} \gamma_{A,N\lambda\to N-1\lambda'}=\frac{2\pi}{\hbar}g_A(\mu_A)\left(1-f(E_{N\lambda}-E_{N-1\lambda'}-\mu_A)\right)\times\\ &\times\sum_{i\sigma}\left|\mathcal{V}_{A,k_F,i,\sigma}\right|^2\left|M_{\lambda'\lambda}^{N-1}(i,\sigma)\right|^2 \tag*{(57)} \end{aligned}

where gA(μA)g_A(\mu_A) is the density of states at the chemical potential μA\mu_A of the AA electrode.

The above expressions for electron transfer together with the conditions for open shell behavior found in this article lead to a clear mechanism for spin and shell blockade in the transport response of nanoribbons. The mechanism for shell blockade in the SET regime is illustrated with the following example: the transition matrix element (1,1,0,0,0,0∣r^iσ∣0,0,1,0,0,0)(1,1,0,0,0,0|\hat{r}_{i\sigma}|0,0,1,0,0,0) is always zero because the transition needs two electrons to change their single-particle state simultaneously. Similarly, the origin of the mechanism for spin-blockade in the SET regime lies in spin-conservation. For example, the transition matrix element ⟨N=2,R,S=0,Sz∣r^iσ∣N=3,R′,S=3/2,Sz′⟩\langle N=2,R,S=0,S_z|\hat{r}_{i\sigma}|N=3,R^\prime,S=3/2,S_z^\prime\rangle is always zero because a single electron can raise the spin by only 1/21/2.

The explicit expressions for the three-shell model are

γA,Nλ→N+1λ′=γA,1∣Mλ′λN(1)∣2+γA,2∣Mλ′λN(2)∣2+γA,3∣Mλ′λN(3)∣2(58)\gamma_{A,N\lambda\to N+1\lambda^\prime}=\gamma_{A,1}\left|M_{\lambda^\prime\lambda}^{N}(1)\right|^2+\gamma_{A,2}\left|M_{\lambda^\prime\lambda}^{N}(2)\right|^2+\gamma_{A,3}\left|M_{\lambda^\prime\lambda}^{N}(3)\right|^2 \tag*{(58)}

where we have defined γA,i=2πℏgA(μA)∣VA,kF,i∣2\gamma_{A,i}=\frac{2\pi}{\hbar}g_A(\mu_A)|V_{A,k_F,i}|^2 and ∣Mλ′λN(i)∣2=∣Mλ′λN(i,↑)∣2+∣Mλ′λN(i,↓)∣2\left|M_{\lambda^\prime\lambda}^{N}(i)\right|^2=\left|M_{\lambda^\prime\lambda}^{N}(i,\uparrow)\right|^2+\left|M_{\lambda^\prime\lambda}^{N}(i,\downarrow)\right|^2.

Conclusions

In conclusion, we have addressed the electronic behavior of free-standing doped GNR using an extended Hubbard model. We have performed an exact mapping of the model to a Kanamori Hamiltonian. We have reduced the range of the Hubbard parameters, and showed that an onsite Hubbard model with U≈4.9U\approx4.9 eV fits well the low-lying mean field solutions to DFT simulations of ribbons of different widths and lengths. We have determined from the onsite model, analytical equations for the Coulomb matrix elements. We have found that those matrix elements follow a remarkable scaling law, and that they collapse into just three for each ribbon. We have then introduced an effective low-energy Hamiltonian to address the many-body behavior of ribbons where the chemical potential lies inside the bulk bands. The model allows us to unveil a robust mechanism that stabilizes open-shell high-spin states. We have estimated the percentage of cases where open-shell high-spin states are expected to exist. These percentages increase but saturate with the ribbon lengths and widths and decrease when an effective dielectric constant is included in the calculations.

Acknowledgments

JF would like to acknowledge very useful discussions with Malte Rössner and technical assistance from Alvaro Correal. JF and AGF have been supported by MCIN/AEI/10.13039/501100011033/FEDER, by UE and by SEKUENS through the projects PID2022-137078NB-100, TRILMAX 101159646, and UNoANO IDE/2024/000678.

At Delft, the work was supported by a HORIZON-EIC-2022-PATHFINDEROPEN project (ATYPICAL, project number 101099098).

Diagonalization of the three shell model

The Hamiltonian for the three-shell model is

H^=∑i=1,2,3(ϵin^i+Un^i↑n^i↓)+V12(n^1n^22−2S^1⋅S^2+Δ^1†Δ^2+Δ^2†Δ^1)+V13(n^1n^32−2S^1⋅S^3+Δ^1†Δ^3+Δ^3†Δ^1)+V23(n^2n^32−2S^2⋅S^3+Δ^2†Δ^3+Δ^3†Δ^2).(59)\begin{aligned} \hat{H}=\sum_{i=1,2,3}\left(\epsilon_i\hat{n}_i+U\hat{n}_{i\uparrow}\hat{n}_{i\downarrow}\right) \\ &+V_{12}\left(\frac{\hat{n}_1\hat{n}_2}{2}-2\hat{\mathbf{S}}_1\cdot\hat{\mathbf{S}}_2+\hat{\Delta}_1^\dagger\hat{\Delta}_2+\hat{\Delta}_2^\dagger\hat{\Delta}_1\right) \\ &+V_{13}\left(\frac{\hat{n}_1\hat{n}_3}{2}-2\hat{\mathbf{S}}_1\cdot\hat{\mathbf{S}}_3+\hat{\Delta}_1^\dagger\hat{\Delta}_3+\hat{\Delta}_3^\dagger\hat{\Delta}_1\right) \\ &+V_{23}\left(\frac{\hat{n}_2\hat{n}_3}{2}-2\hat{\mathbf{S}}_2\cdot\hat{\mathbf{S}}_3+\hat{\Delta}_2^\dagger\hat{\Delta}_3+\hat{\Delta}_3^\dagger\hat{\Delta}_2\right). \tag*{(59)} \end{aligned}

The states are ordered so that ϵ1<ϵ2<ϵ3\epsilon_1<\epsilon_2<\epsilon_3. The Fock space F\mathcal{F} has dimensions 43=644^3=64 and is spanned by the single-particle basis {∣n1↑,n1↓,n2↑,n2↓,n3↑,n3↓⟩\{|n_{1\uparrow},n_{1\downarrow},n_{2\uparrow},n_{2\downarrow},n_{3\uparrow},n_{3\downarrow}\rangle. The Hamiltonian commutes with the total number operator N^\hat{N}. N^\hat{N} has eigen-values N=0,1,…,6N=0,1,\ldots,6, so that F\mathcal{F} is split into seven subspaces labeled by NN. The Hamiltonian also commutes with total spin operator S^2\hat{\mathbf{S}}^2, so that each of those subspaces is further split into subspaces labeled by SS. However, H^\hat{H}, N^\hat{N} and S^\hat{\mathbf{S}} do not constitute a complete set of commuting observables because there exist for some of the sub-spaces, more than one same-SS multiplet. We label these different subspaces with an additional integer number RR whenever they exist. All in all, we use the following notation to describe the Hamiltonian eigen-states: ∣N,R,S,Sz⟩|N,R,S,S_z\rangle. We describe below the subspaces N=0N=0, 11, 22 and 33, their eigen-states and their energies. Sub-spaces N=4N=4, 55 and 66 are easily deduced from the electron-hole symmetry of the Hamiltonian. The highest spin subspace occurs at half-filling, N=3N=3, and corresponds to a quadruplet with S=3/2S=3/2. We denote Δϵab=ϵa−ϵb\Delta\epsilon_{ab}=\epsilon_a-\epsilon_b.

  • Subspace N=0N=0. This trivial subspace is one-dimensional and contains only the vacuum state

∣0,0,0⟩=∣000000⟩E=0|0,0,0\rangle=|000000\rangle\qquad E=0
  • Subspace N=1N=1. This subspace has dimensions equal to 6 and is made up of the following three doublets

∣1,1,1/2,1/2⟩=∣100000⟩E=ϵ1|1,1,1/2,1/2\rangle=|100000\rangle\qquad E=\epsilon_1
∣1,1,1/2,−1/2⟩=∣010000⟩E=ϵ1|1,1,1/2,-1/2\rangle=|010000\rangle\qquad E=\epsilon_1
∣1,2,1/2,1/2⟩=∣001000⟩E=ϵ2|1,2,1/2,1/2\rangle=|001000\rangle\qquad E=\epsilon_2
∣1,2,1/2,−1/2⟩=∣000100⟩E=ϵ2|1,2,1/2,-1/2\rangle=|000100\rangle\qquad E=\epsilon_2
∣1,3,1/2,1/2⟩=∣000010⟩E=ϵ3|1,3,1/2,1/2\rangle=|000010\rangle\qquad E=\epsilon_3
∣1,3,1/2,−1/2⟩=∣000001⟩E=ϵ3|1,3,1/2,-1/2\rangle=|000001\rangle\qquad E=\epsilon_3
  • Subspace N=2N=2. This subspace has dimensions equal to 15 and is made up of six singlets and three triplets. The Hamiltonian for the basis states ∣1,1,0,0,0,0⟩|1,1,0,0,0,0\rangle, ∣0,0,1,1,0,0⟩|0,0,1,1,0,0\rangle and ∣0,0,0,0,1,1⟩|0,0,0,0,1,1\rangle is

(2ϵ1+UV12V13V122ϵ2+UV23V13V232ϵ3+U)(60)\begin{pmatrix} 2\epsilon_{1}+U & V_{12} & V_{13}\\ V_{12} & 2\epsilon_{2}+U & V_{23}\\ V_{13} & V_{23} & 2\epsilon_{3}+U \tag*{(60)} \end{pmatrix}

Solving the eigen-value equation yields a complicated cubic equation, whose explicit solution delivers three eigen-energies. The corresponding three eigenstates are singlets of the form

∣2,R,0,0⟩=AR1∣1,1,0,0,0,0⟩+AR2∣0,0,1,1,0,0⟩+AR3∣0,0,0,0,1,1⟩|2,R,0,0\rangle=A_{R1}|1,1,0,0,0,0\rangle+A_{R2}|0,0,1,1,0,0\rangle+A_{R3}|0,0,0,0,1,1\rangle

where R=1,2,3R=1,2,3. The remaining three singlet eigen-states and eigen-energies are

∣2,4,0,0⟩=∣100100⟩−∣011000⟩2,E=ϵ1+ϵ2+2V12|2,4,0,0\rangle=\frac{|100100\rangle-|011000\rangle}{\sqrt{2}},\qquad E=\epsilon_{1}+\epsilon_{2}+2V_{12}
∣2,5,0,0⟩=∣100001⟩−∣010010⟩2,E=ϵ1+ϵ3+2V13|2,5,0,0\rangle=\frac{|100001\rangle-|010010\rangle}{\sqrt{2}},\qquad E=\epsilon_{1}+\epsilon_{3}+2V_{13}
∣2,6,0,0⟩=∣001001⟩−∣000110⟩2,E=ϵ2+ϵ3+2V23|2,6,0,0\rangle=\frac{|001001\rangle-|000110\rangle}{\sqrt{2}},\qquad E=\epsilon_{2}+\epsilon_{3}+2V_{23}

The three triplets are

∣2,1,1,1⟩=∣101000⟩,E=ϵ1+ϵ2|2,1,1,1\rangle=|101000\rangle,\qquad E=\epsilon_{1}+\epsilon_{2}
∣2,1,1,0⟩=∣100100⟩+∣011000⟩2,E=ϵ1+ϵ2|2,1,1,0\rangle=\frac{|100100\rangle+|011000\rangle}{\sqrt{2}},\qquad E=\epsilon_{1}+\epsilon_{2}
∣2,1,1,−1⟩=∣010100⟩,E=ϵ1+ϵ2|2,1,1,-1\rangle=|010100\rangle,\qquad E=\epsilon_{1}+\epsilon_{2}
∣2,2,1,1⟩=∣100010⟩,E=ϵ1+ϵ3|2,2,1,1\rangle=|100010\rangle,\qquad E=\epsilon_{1}+\epsilon_{3}
∣2,2,1,0⟩=∣100001⟩+∣010010⟩2,E=ϵ1+ϵ3|2,2,1,0\rangle=\frac{|100001\rangle+|010010\rangle}{\sqrt{2}},\qquad E=\epsilon_{1}+\epsilon_{3}
∣2,2,1,−1⟩=∣010001⟩,E=ϵ1+ϵ3|2,2,1,-1\rangle=|010001\rangle,\qquad E=\epsilon_{1}+\epsilon_{3}
∣2,3,1,1⟩=∣001010⟩,E=ϵ2+ϵ3|2,3,1,1\rangle=|001010\rangle,\qquad E=\epsilon_{2}+\epsilon_{3}
∣2,3,1,0⟩=∣000110⟩+∣001001⟩2,E=ϵ2+ϵ3|2,3,1,0\rangle=\frac{|000110\rangle+|001001\rangle}{\sqrt{2}},\qquad E=\epsilon_{2}+\epsilon_{3}
∣2,3,1,−1⟩=∣000101⟩,E=ϵ2+ϵ3|2,3,1,-1\rangle=|000101\rangle,\qquad E=\epsilon_{2}+\epsilon_{3}

The lowest-lying states are the closed shell singlet ∣2,1,0,0⟩|2,1,0,0\rangle and the open shell triplet ∣2,1,1,Sz⟩|2,1,1,S^{z}\rangle. The condition for the ground state to be a triplet with spin S=1S=1 can be found only numerically.

We analyse here two situations in more detail. The first highlights how effective a ferromagnetic exchange energy arises, and how a third shell may be or may be not enter in the calculations. The second highlight how a shell becomes closed.

  1. Limit ϵ3≫ϵ2\epsilon_{3}\gg\epsilon_{2}.

Here, the third shell is completely empty. The problem thus reduces to a simpler half-filled two-shell model. The two singlet states simplify to

∣2,R,0,0⟩=AR1∣1,1,0,0⟩+AR2∣0,0,1,1⟩|2,R,0,0\rangle=A_{R1}|1,1,0,0\rangle+A_{R2}|0,0,1,1\rangle

where R=1,2R=1,2, with eigen-energies

E=ϵ1+ϵ2+U±(Δϵ21)2+V122E=\epsilon_{1}+\epsilon_{2}+U\pm\sqrt{(\Delta\epsilon_{21})^{2}+V_{12}^{2}}

The energy difference between the singlet and the triplet defines an effective exchange coupling

J=U−(Δϵ21)2+V122(61)\mathcal{J}=U-\sqrt{(\Delta\epsilon_{21})^{2}+V_{12}^{2}} \tag*{(61)}

The triplet state ∣2,1,1,Sz⟩|2,1,1,S^{z}\rangle becomes the ground state if J>0\mathcal{J}>0, e.g.: if the effective exchange coupling becomes ferromagnetic.

  1. Limit ϵ1≪ϵ2\epsilon_{1}\ll\epsilon_{2}.

This limit corresponds to the closing on the lowest shell. Again the equations simplify so that the singlet states are

∣2,R,0,0⟩=AR1∣1,1,0,0⟩+AR2∣0,0,1,1⟩|2,R,0,0\rangle=A_{R1}|1,1,0,0\rangle+A_{R2}|0,0,1,1\rangle

where R=2,3R=2,3, and with eigen-energies

E=ϵ2+ϵ3+U±(Δϵ32)2+V232E=\epsilon_{2}+\epsilon_{3}+U\pm\sqrt{(\Delta\epsilon_{32})^{2}+V_{23}^{2}}
  • Subspace N=3N=3. This subspace has dimensions equal to 20 and is made up of eight doublets and a quadruplet. We list first all the basis states

∣A,Sz=1/2⟩=∣111000⟩,EA=2ϵ1+ϵ2+U+V12|A,S^{z}=1/2\rangle=|111000\rangle,\qquad E_{A}=2\epsilon_{1}+\epsilon_{2}+U+V_{12}
∣B,Sz=−1/2⟩=∣110100⟩,EB=EA|B,S^{z}=-1/2\rangle=|110100\rangle,\qquad E_{B}=E_{A}
∣C,Sz=1/2⟩=∣001011⟩,EC=ϵ2+2ϵ3+U+V23|C,S^{z}=1/2\rangle=|001011\rangle,\qquad E_{C}=\epsilon_{2}+2\epsilon_{3}+U+V_{23}
∣D,Sz=−1/2⟩=∣000111⟩,ED=EC|D,S^{z}=-1/2\rangle=|000111\rangle,\qquad E_{D}=E_{C}
∣E,Sz=1/2⟩=∣110010⟩,EE=2ϵ1+ϵ3+U+V13|E,S^{z}=1/2\rangle=|110010\rangle,\qquad E_{E}=2\epsilon_{1}+\epsilon_{3}+U+V_{13}
∣F,Sz=−1/2⟩=∣110001⟩,EF=EE|F,S^{z}=-1/2\rangle=|110001\rangle,\qquad E_{F}=E_{E}
∣G,Sz=1/2⟩=∣001110⟩,EG=2ϵ2+ϵ3+U+V23|G,S^{z}=1/2\rangle=|001110\rangle,\qquad E_{G}=2\epsilon_{2}+\epsilon_{3}+U+V_{23}
∣H,Sz=−1/2⟩=∣001101⟩,EH=EG|H,S^{z}=-1/2\rangle=|001101\rangle,\qquad E_{H}=E_{G}
∣I,Sz=1/2⟩=∣101100⟩,EI=ϵ1+2ϵ2+U+V12|I,S^{z}=1/2\rangle=|101100\rangle,\qquad E_{I}=\epsilon_{1}+2\epsilon_{2}+U+V_{12}
∣J,Sz=−1/2⟩=∣011100⟩,EJ=EI|J,S^{z}=-1/2\rangle=|011100\rangle,\qquad E_{J}=E_{I}
∣K,Sz=1/2⟩=∣100011⟩,EK=ϵ1+2ϵ3+U+V13|K,S^{z}=1/2\rangle=|100011\rangle,\qquad E_{K}=\epsilon_{1}+2\epsilon_{3}+U+V_{13}
∣L,Sz=−1/2⟩=∣010011⟩,EL=EK|L,S^{z}=-1/2\rangle=|010011\rangle,\qquad E_{L}=E_{K}
∣M,Sz=1/2⟩=∣101001⟩,EM=ϵα+V13+V23|M,S^{z}=1/2\rangle=|101001\rangle,\qquad E_{M}=\epsilon_{\alpha}+V_{13}+V_{23}
∣N,Sz=1/2⟩=∣100110⟩,EN=ϵα+V12+V23|N,S^{z}=1/2\rangle=|100110\rangle,\qquad E_{N}=\epsilon_{\alpha}+V_{12}+V_{23}
∣O,Sz=1/2⟩=∣011010⟩,EO=ϵα+V12+V13|O,S^{z}=1/2\rangle=|011010\rangle,\qquad E_{O}=\epsilon_{\alpha}+V_{12}+V_{13}
∣P,Sz=−1/2⟩=∣010110⟩,EP=ϵα+V13+V23|P,S^{z}=-1/2\rangle=|010110\rangle,\qquad E_{P}=\epsilon_{\alpha}+V_{13}+V_{23}
∣Q,Sz=−1/2⟩=∣011001⟩,EQ=ϵα+V12+V23|Q,S^{z}=-1/2\rangle=|011001\rangle,\qquad E_{Q}=\epsilon_{\alpha}+V_{12}+V_{23}
∣R,Sz=−1/2⟩=∣100101⟩,ER=ϵα+V12+V13|R,S^{z}=-1/2\rangle=|100101\rangle,\qquad E_{R}=\epsilon_{\alpha}+V_{12}+V_{13}

where ϵα=ϵ1+ϵ2+ϵ3\epsilon_{\alpha}=\epsilon_{1}+\epsilon_{2}+\epsilon_{3} and ϵb=ϵα+(V12+V13+V23)\epsilon_{b}=\epsilon_{\alpha}+(V_{12}+V_{13}+V_{23}). The first set has six doublets, all of them having one fully occupied, one half-occupied and one empty shell. The first two doublets in this set are made of the states ∣A⟩|A\rangle and ∣C⟩|C\rangle for spin Sz=+1/2S^{z}=+1/2 and ∣B⟩|B\rangle and ∣D⟩|D\rangle for spin Sz=−1/2S^{z}=-1/2. They have inter-state couplings VAC=VBD=V13V_{AC}=V_{BD}=V_{13}. Both Hamiltonians are identical, so we write just the first one:

(EAVABVABEB)(62)\begin{pmatrix} E_{A} & V_{AB}\\ V_{AB} & E_{B} \tag*{(62)} \end{pmatrix}

The doublets’ eigen-states and eigen-energies ∣N,L,S,SZ⟩|N,L,S,S_Z\rangle are

∣3,1,1/2,1/2⟩=u+∣A⟩−u−∣C⟩|3,1,1/2,1/2\rangle= u_+|A\rangle-u_-|C\rangle
∣3,1,1/2,−1/2⟩=u+∣B⟩−u−∣D⟩|3,1,1/2,-1/2\rangle= u_+|B\rangle-u_-|D\rangle
ϵ3,1,1/2,±1/2=EA+EC2−(EC−EA2)2+VAC2=ϵa+U+V12+V232−(Δϵ31+V23−V122)2+V132\begin{aligned} \epsilon_{3,1,1/2,\pm1/2}=\frac{E_A+E_C}{2}-\sqrt{\left(\frac{E_C-E_A}{2}\right)^2+V_{AC}^2} \\ &=\epsilon_a+U+\frac{V_{12}+V_{23}}{2}-\sqrt{\left(\Delta\epsilon_{31}+\frac{V_{23}-V_{12}}{2}\right)^2+V_{13}^2} \end{aligned}
∣3,2,1/2,1/2⟩=u−∣A⟩+u+∣C⟩|3,2,1/2,1/2\rangle= u_-|A\rangle+u_+|C\rangle
∣3,2,1/2,−1/2⟩=u−∣B⟩+u+∣D⟩|3,2,1/2,-1/2\rangle= u_-|B\rangle+u_+|D\rangle
ϵ3,2,1/2,±1/2=EA+EC2+(EC−EA2)2+VAC2=ϵa+U+V12+V232+(Δϵ31+V23−V122)2+V132\begin{aligned} \epsilon_{3,2,1/2,\pm1/2}=\frac{E_A+E_C}{2}+\sqrt{\left(\frac{E_C-E_A}{2}\right)^2+V_{AC}^2} \\ &=\epsilon_a+U+\frac{V_{12}+V_{23}}{2}+\sqrt{\left(\Delta\epsilon_{31}+\frac{V_{23}-V_{12}}{2}\right)^2+V_{13}^2} \end{aligned}

The next two doublets are made of the states ∣E⟩|E\rangle and ∣G⟩|G\rangle for spin SZ=+1/2S^Z=+1/2 and ∣F⟩|F\rangle and ∣H⟩|H\rangle for spin SZ=−1/2S^Z=-1/2. Their inter-state couplings VEG=VFH=V12V_{EG}=V_{FH}=V_{12}. The Hamiltonian, eigen-states and eigen-energies are the same as above upon the replacement A→EA\rightarrow E, B→FB\rightarrow F, C→GC\rightarrow G and D→HD\rightarrow H, and the doublets are labeled now as ∣3,3,1/2,±1/2⟩|3,3,1/2,\pm1/2\rangle and ∣3,4,1/2,1/2⟩|3,4,1/2,1/2\rangle. Their eigen-energies are

E3,3,1/2,±1/2=ϵa+U+V13+V232−(Δϵ21+V23−V132)2+V122E_{3,3,1/2,\pm1/2}=\epsilon_a+U+\frac{V_{13}+V_{23}}{2}-\sqrt{\left(\Delta\epsilon_{21}+\frac{V_{23}-V_{13}}{2}\right)^2+V_{12}^2}
E3,4,1/2,±1/2=ϵa+U+V13+V232+(Δϵ21+V23−V132)2+V122E_{3,4,1/2,\pm1/2}=\epsilon_a+U+\frac{V_{13}+V_{23}}{2}+\sqrt{\left(\Delta\epsilon_{21}+\frac{V_{23}-V_{13}}{2}\right)^2+V_{12}^2}

The last two doublets in this set are made of the states ∣I⟩|I\rangle and ∣K⟩|K\rangle for spin SZ=+1/2S^Z=+1/2 and ∣J⟩|J\rangle and ∣L⟩|L\rangle for spin SZ=−1/2S^Z=-1/2. Their inter-state couplings are VIK=VJL=V23V_{IK}=V_{JL}=V_{23}. The Hamiltonian, eigen-states and eigen-energies are again the same as above upon the replacement A→IA\rightarrow I, B→JB\rightarrow J, C→KC\rightarrow K and D→LD\rightarrow L, and the doublets are labeled now as ∣3,5,1/2,±1/2⟩|3,5,1/2,\pm1/2\rangle and ∣3,6,1/2,1/2⟩|3,6,1/2,1/2\rangle. Their eigen-energies are

E3,5,1/2,±1/2=ϵa+U+V13+V122−(Δϵ32+V13−V122)2+V232E_{3,5,1/2,\pm1/2}=\epsilon_a+U+\frac{V_{13}+V_{12}}{2}-\sqrt{\left(\Delta\epsilon_{32}+\frac{V_{13}-V_{12}}{2}\right)^2+V_{23}^2}
E3,6,1/2,±1/2=ϵa+U+V13+V122+(Δϵ32+V13−V122)2+V232E_{3,6,1/2,\pm1/2}=\epsilon_a+U+\frac{V_{13}+V_{12}}{2}+\sqrt{\left(\Delta\epsilon_{32}+\frac{V_{13}-V_{12}}{2}\right)^2+V_{23}^2}

The last two doublets have three open shells,

∣3,7,1/2,1/2⟩=1D+1/2((V23−V12−R)∣M⟩+(V13−V23+R)∣N⟩+(V12−V13)∣O⟩)|3,7,1/2,1/2\rangle=\frac{1}{D_+^{1/2}}\left((V_{23}-V_{12}-R)|M\rangle+(V_{13}-V_{23}+R)|N\rangle+(V_{12}-V_{13})|O\rangle\right)
∣3,7,1/2,−1/2⟩=1D+1/2((V23−V12−R)∣P⟩+(V13−V23+R)∣Q⟩+(V12−V13)∣R⟩)|3,7,1/2,-1/2\rangle=\frac{1}{D_+^{1/2}}\left((V_{23}-V_{12}-R)|P\rangle+(V_{13}-V_{23}+R)|Q\rangle+(V_{12}-V_{13})|R\rangle\right)
∣3,8,1/2,1/2⟩=1D−1/2((V23−V12+R)∣M⟩+(V13−V23−R)∣N⟩+(V12−V13)∣O⟩)|3,8,1/2,1/2\rangle=\frac{1}{D_-^{1/2}}\left((V_{23}-V_{12}+R)|M\rangle+(V_{13}-V_{23}-R)|N\rangle+(V_{12}-V_{13})|O\rangle\right)
∣3,8,1/2,−1/2⟩=1D−1/2((V23−V12+R)∣P⟩+(V13−V23−R)∣Q⟩+(V12−V13)∣R⟩)|3,8,1/2,-1/2\rangle=\frac{1}{D_-^{1/2}}\left((V_{23}-V_{12}+R)|P\rangle+(V_{13}-V_{23}-R)|Q\rangle+(V_{12}-V_{13})|R\rangle\right)
R=12((V12−V13)2+(V12−V23)2+(V13−V23)2)1/2R=\frac{1}{2}\left((V_{12}-V_{13})^2+(V_{12}-V_{23})^2+(V_{13}-V_{23})^2\right)^{1/2}
D±=2R1/2(R±12(V12+V13)−V23)1/2D_{\pm}=2R^{1/2}\left(R\pm\frac{1}{2}(V_{12}+V_{13})-V_{23}\right)^{1/2}

Their energies are

ϵ3,7,±1/2=ϵb−R\epsilon_{3,7,\pm1/2}=\epsilon_b-R
ϵ3,8,±1/2=ϵb+R\epsilon_{3,8,\pm1/2}=\epsilon_b+R

The quadruplet eigen-states are

∣3,3/2,3/2⟩=∣1,0,1,0,1,0⟩|3,3/2,3/2\rangle=|1,0,1,0,1,0\rangle
∣3,3/2,1/2⟩=13(∣M⟩+∣N⟩+∣O⟩)|3,3/2,1/2\rangle=\frac{1}{\sqrt{3}}(|M\rangle+|N\rangle+|O\rangle)
∣3,3/2,−1/2⟩=13(∣P⟩+∣Q⟩+∣R⟩)|3,3/2,-1/2\rangle=\frac{1}{\sqrt{3}}(|P\rangle+|Q\rangle+|R\rangle)
∣3,3/2,−3/2⟩=∣0,1,0,1,0,1⟩|3,3/2,-3/2\rangle=|0,1,0,1,0,1\rangle

and their eigen-energy is E=ϵaE = \epsilon_a.

Inspecting the above expressions for the eigen-energies, we come to the conclusion that the two doublets having all open shells have too large eigen-energies. The quadruplet thus competes with the first set of six doublets to be the ground state. The ground state will be the quadruplet state provided that the following three conditions are met

together:

U+V12+V232−(Δϵ31+V23−V122)2+V132>0(63)\mathcal{U} + \frac{V_{12}+V_{23}}{2} - \sqrt{\left(\Delta\epsilon_{31}+\frac{V_{23}-V_{12}}{2}\right)^2+V_{13}^2}>0 \tag*{(63)}
U+V13+V232−(Δϵ21+V23−V132)2+V122>0(64)\mathcal{U} + \frac{V_{13}+V_{23}}{2} - \sqrt{\left(\Delta\epsilon_{21}+\frac{V_{23}-V_{13}}{2}\right)^2+V_{12}^2}>0 \tag*{(64)}
U+V13+V122−(Δϵ32+V13−V122)2+V232>0(65)\mathcal{U} + \frac{V_{13}+V_{12}}{2} - \sqrt{\left(\Delta\epsilon_{32}+\frac{V_{13}-V_{12}}{2}\right)^2+V_{23}^2}>0 \tag*{(65)}

References

References

  1. [1]A. K. Geim, Science 320, 1530 (2009).
  2. [2]R. M. Westervelt, Science 320, 324 (2009).
  3. [3]J. Cai, P. Ruffieux, R. Jaafar, M. Bieri, T. Braun, S. Blankenburg, M. Muoth, A. P. Seitsonen, M. Saleh, X. Feng, K. Mullen, and R. Fasel, Nature 466, 7305 (2010).
  4. [4]J. Lawrence, A. Berdonces-Layunta, S. Edalatmanesh, J. Castro-Este­ban, T. Wang, A. Jimenez-Martin, B. de la Torre, R. Castrillo-Bodero, P. Angulo-Portugal, M. Mohammed, A. Matej, M. Vilas-Varela, F. Schiller, M. Corso, P. Jelinek, D. Peña, and D. G. de Oteyza, Nature Chemistry 14, 1451 (2022).
  5. [5]J. Zhang, L. Qian, G. B. Barin, A. Daaoub, P. Chen, K. Mullen, S. Sangtarash, P. Ruffieux, R. Fasel, H. Sadeghi, J. Zhang, M. Calame, and M. L. Perrin, Nature Electronics 6, 572 (2023).
  6. [6]J. Zhang, G. B. Barin, R. Furrer, C.-Z. Du, X.-Y. Wang, K. Mullen, P. Ruffieux, R. Fasel, M. Calame, and M. L. Perrin, Nano Lett. 23, 8474 (2023).
  7. [7]W. Niu, S. Sopp, A. Lodi, A. Gee, F. Kong, T. Pei, P. Gehring, J. Nagele, C.-S. Lau, J. Ma, J. Liu, A. Narita, J. Mol, M. Burghard, K. Mullen, Y. Mai, X. Feng, and L. Bogani, Nature Materials 22, 180 (2023).
  8. [8]E. A. Laird, F. Kuen­meth, G. Steele, K. Grove-Rasmussen, J. Nygard, H. Flensberg, and L. P. Kouwenhoven, Rev. Mod. Phys. 87, 703 (2015).
  9. [9]S. Konschuh, M. Gmitra, and J. Fabian, Phys. Rev. B 82, 245412 (2010).
  10. [10]K. Zollner, M. Kurpas, M. Gmitra, and J. Fabian, Nat. Rev. Physics 7, 255 (2025).
  11. [11]A. Garcia-Fuente, D. Carrascal, G. Ross, and J. Ferrer, Phys. Rev. B 107, 115403 (2023).
  12. [12]A. Talkachov and E. Babaev, Physical Review B 107, 045419 (2023).
  13. [13]J. Ferrer and A. Garcia-Fuente, Phys. Rev. B 112, 165101 (2025).
  14. [14]J. Fernandez-Rossier, Physical Review B 77, 075430 (2008).
  15. [15]J. Jung and A. H. MacDonald, Physical Review B 79, 235433 (2009).
  16. [16]T. O. Wehling, E. Sasioglu, C. Friedrich, A. I. Lichtenstein, M. I. Katsnelson, and S. Blugel, Phys. Rev. Lett. 106, 236805 (2011).
  17. [17]M. Schuller, M. Rossner, T. O. Wehling, A. I. Lichtenstein, and M. I. Katsnelson, Phys. Rev. Lett. 111, 036601 (2013).
  18. [18]M. Rossner, E. Sasioglu, C. Friedrich, S. Blugel, and Wehling, Phys. Rev. B 92, 085102 (2015).
  19. [19]J. Kanamori, Progress of theoretical Physics 30, 275 (1963).
  20. [20]J. Zhang, B. Ghawri, D. Dutta, R. Fasel, M. Calame, G. B. Barin, and M. L. Perrin, Nature Reviews Materials 11, 194 (2026).
  21. [21]E. Lieb and D. Mattis, Physical Review 125, 164 (1962).
  22. [22]H. Tasaki, J. Phys.: Condens. Matt. 10, 4353 (1998).
  23. [23]P. Pieri, S. Daul, D. Baeriswyl, M. Dzierzawa, and P. Fazekas, Physical Review B 45, 9250 (1996).
  24. [24]I. S. Burmistrov, D. S. Gefen, Y. Shapiro, and A. Shnirman, Phys. Rev. Lett. 124, 196801 (2020).
  25. [25]D. J. Carrascal and J. Ferrer, Phys. Rev. B 85, 045110 (2012).
  26. [26]D. J. Carrascal, J. Ferrer, J. C. Smith, and K. Burke, J. Phys: Condens. Matter 27, 393001 (2015).
  27. [27]I. L. Kurland, I. L. Aleiner, and B. L. Altshuler, Phys. Rev. B 62, 14886 (2000).

Paper details

vlnc:W_qajd4mtf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents