betaFeedback

WorksPhysics and Astronomy

Symmetry-Controlled Quantum State Transfer in Poly-bi-[8]-annulenylene: From Heisenberg to Hubbard Dynamics

Abstract

We investigate quantum state transfer in the poly-bi-[8]-annulenylene lattice using isotropic Heisenberg and Hubbard models. For a 16-site lattice, reflection symmetry separates the single-excitation Hamiltonian into independent triplet and singlet sectors, allowing analytical determination of the energy spectra. Although perfect state transfer is forbidden for positive couplings, appropriately chosen coupling ratios enable pretty-good transfer with arbitrarily high fidelity. Remarkably, the triplet transfer probability is independent of the interchain rung coupling, while singlet transfer can be selectively enhanced or suppressed, enabling symmetry-selective quantum transport without time-dependent control. Extending the analysis to the Hubbard model, we show that the Heisenberg transfer dynamics emerge in the strong-interaction limit, while finite interactions delay all high-fidelity transfer peaks of a symmetry sector by the same fraction of their transfer time, set by the spectrum, and reduce the peak fidelity, even at the corrected time, by an amount that grows with the transfer time. These findings are relevant to future quantum-network architectures, where symmetry-controlled state transfer could provide a building block for connecting quantum memories and emitter-based interfaces within quantum nodes. Overall, our results demonstrate how molecular network geometry and symmetry can be exploited to engineer and control quantum-state transport in interacting spin systems, providing a theoretical framework for quantum information transfer in molecular-inspired quantum architectures.

Introduction

Transferring quantum states between distant parts of a quantum device is a basic requirement for scalable quantum information processing. Bose proposed doing this with a chain of permanently coupled spins [1]: the state propagates under the natural, time-independent dynamics of the chain, and no control is needed during the transfer. This idea has grown into a broad field that studies how the geometry and couplings of a spin network determine its quality as a quantum channel [2], and state transfer has been demonstrated experimentally, for example in quantum-dot spin chains coupled by Heisenberg exchange [3].

The quality of a channel is measured by the transfer probability P(τ)=∣⟨b∣e−iHτ∣a⟩∣2P(\tau) = |\langle b|e^{-iH\tau}|a\rangle|^2 from a source state ∣a⟩|a\rangle to a target state ∣b⟩|b\rangle at time τ\tau. Perfect transfer (PT), P(τ)=1P(\tau) = 1 at a finite time, can be achieved in chains with engineered couplings [4]. In pretty good transfer (PGT), P(τ)P(\tau) comes arbitrarily close to one at suitable times [5, 6]. Most of this theory concerns XX interactions, for which the single-excitation Hamiltonian is the adjacency matrix of the coupling graph. In the isotropic Heisenberg (XXX) exchange, natural for electron spins in quantum dots and molecules [3, 7], each site acquires an on-site energy set by the total strength of its bonds.

In uniform Heisenberg chains PGT between the ends requires the number of sites to be a power of two [6], unlike the XX case. PGT also says nothing about when high fidelities are reached, so PGT alone is of limited practical use. It is more relevant to ask whether the error can be kept below a tolerance at a prescribed, experimentally accessible time by tuning a few couplings [8]; we call this almost perfect transfer (APT).

Networks with several paths, such as parallel chains [9] and spin ladders [10], offer further possibilities. When the network has a reflection symmetry, the dynamics splits into symmetric and antisymmetric sectors, a reduction used to analyze PGT between mirror-image sites [11]. Chains with alternating couplings, as in the Su–Schrieffer–Heeger (SSH) model [12], host boundary zero modes in their topological phase, which have been used for robust transfer protocols [13, 14], usually at the cost of long transfer times or time-dependent control. These edge states are not required for high-fidelity transfer, however, and SSH-type XXZ chains show none near the isotropic Heisenberg point [15].

Poly-bi-[8]-annulenylene, a polymer of cyclooctatraene units, combines these features: its two arms have SSH-like alternating couplings and are joined by rungs, giving it a reflection symmetry and two propagation paths. In a tight-binding description it behaves, for weak coupling between the arms, as two coupled SSH chains with boundary zero modes [16]. Motivated by the growing role of molecular systems in quantum information science [7], we ask whether this connectivity can serve as an efficient channel when the sites interact through isotropic Heisenberg exchange. Because the spin model adds the on-site terms described above, tight-binding results, including those on edge states, do not carry over automatically.

We treat in detail the structure with N=2N = 2 unit cells hosting 16 sites (Figure 1). Our main results are:

Figure 1. Poly-bi-[8]-annulenylene with N=2N = 2, with coupling parameters (v,w,r)(v,w,r). The horizontal dashed line is the reflection axis that exchanges the two arms, j↔17−jj \leftrightarrow17-j, and defines the triplet and singlet sectors (Sec. II C). The vertical dashed line is the reflection axis that exchanges the ends; each subgraph to either side of it can be considered its own poly-bi-[8]-annulenylene model with N=1N = 1.

  1. Reflection symmetry splits the single-excitation Hamiltonian into two 8×88 \times8 blocks, so transfer reduces to moving a triplet or singlet state from the end pair (1,16)(1,16) to the opposite pair (8,9)(8,9). All eigenenergies follow in closed form. For any two-leg ladder, the rung coupling rr only shifts the triplet sector by a constant, so triplet transfer is that of two uncoupled staggered chains [17].

  1. No positive couplings give PT in either sector: in the triplet sector the transition frequencies are incommensurate, and in the singlet sector the PT conditions are incompatible modulo 4.

  1. Couplings related to Pythagorean triples leave only two rationally independent irrational frequencies and give PGT of both states. The arrival times, bounded with the Lenstra–Lenstra–Lovász algorithm [18], follow the scaling law of Ref. [19]. PGT also holds between mirror-image bulk sites [6].

  1. Because the triplet sector is independent of rr, the rung coupling can tune the singlet transfer probability at a fixed time close to one or zero without changing the triplet transfer. Both states can then be transferred with high fidelity, or an unwanted singlet component can be suppressed.

  1. The rungs leave the clean triplet transfer unchanged but make it more robust to static coupling disorder than a single SSH-type chain.

  1. An out-of-time-order correlator (OTOC) [20] built from projectors, as in fidelity OTOCs [21], equals P(τ)2P(\tau)^2 and carries no information beyond the transfer probability.

To test how charge fluctuations modify these predictions, we extend the analysis to the half-filled Hubbard model. The Heisenberg transfer probabilities are recovered at strong on-site repulsion UU. At large but finite UU, the high-fidelity peaks acquire a relative delay proportional to (tv/U)2(t_v/U)^2, where tvt_v is the hopping amplitude on the vv bonds. We show that, at fixed couplings and initial state, this prefactor is the same for all high-fidelity peaks within a sector: it is a spectral average of the finite-UU shifts of the transition frequencies and does not depend on the peak. The loss that remains after correcting the readout time grows with the transfer time, so the early peaks retain high transfer probabilities at moderate U/tvU/t_v, whereas the late PGT peaks require much stronger repulsion.

The paper is organized as follows: Sec. II introduces the model, Sec. III studies PT, PGT, and APT and their robustness to coupling disorder, § 2 examines transfer in the Hubbard model and its approach to the Heisenberg limit, and § 2.1 concludes. Appendix A shows that the triplet transfer is independent of the rungs for any two-leg ladder, Appendix B gives the explicit transfer probability, Appendix C discusses OTOCs, and Appendix D derives the delay of the transfer peaks at finite UU.

Heisenberg hamiltonian, initial states, eigenenergies

The Poly-bi-[8]-annulenylene lattice

A structure of NN unit cells has 8N8N sites and three exchange couplings: vv and ww alternate along each of the two arms, as in the SSH chain [12], and rr couples the two arms. For weak rr and v<wv < w, the corresponding tight-binding model hosts boundary zero modes under open boundary conditions [16].

We number the sites j=1,…,8Nj = 1,\ldots,8N by following one arm from end to end and then returning along the other, so that j=1,…,4Nj = 1,\ldots,4N lie on one arm and j=4N+1,…,8Nj = 4N+1,\ldots,8N on the other. The bonds are then

v:(j,j+1),j odd,w:(j,j+1),j even,r:(j,8N+1−j),j≤4N, j=2,3(mod4).(1)\begin{aligned} v &: (j,j+1), && j\ \text{odd},\\ w &: (j,j+1), && j\ \text{even},\\ r &: (j,8N+1-j), && j \le4N,\ j = 2,3\pmod4. \tag*{(1)} \end{aligned}

with jj and j+1j+1 on the same arm in the first two lines, giving 4N4N bonds of strength vv, 4N−24N-2 of strength ww and 2N2N rungs. We treat N=2N = 2 (16 sites, Figure 1) in detail; the symmetry decomposition and the rung independence of the triplet sector hold for any NN.

Heisenberg Hamiltonian

We look at the spin-12\frac{1}{2} isotropic Heisenberg (XXX) ferromagnetic model in the single-excitation sector.

For a general NN, the Hamiltonian is

HN=−v∑j=1j odd8N−1σj⋅σj+1−w∑j=2j even; j≠4N8N−2σj⋅σj+1−r∑j=1j≡2,3 (mod 4)4Nσj⋅σ8N+1−j(2)H_N=-v\sum_{\substack{j=1\\j\ \mathrm{odd}}}^{8N-1}\boldsymbol{\sigma}_j\cdot\boldsymbol{\sigma}_{j+1} -w\sum_{\substack{j=2\\j\ \mathrm{even};\ j\ne4N}}^{8N-2}\boldsymbol{\sigma}_j\cdot\boldsymbol{\sigma}_{j+1} -r\sum_{\substack{j=1\\j\equiv2,3\ (\mathrm{mod}\ 4)}}^{4N}\boldsymbol{\sigma}_j\cdot\boldsymbol{\sigma}_{8N+1-j} \tag*{(2)}

where σi=(σix,σiy,σiz)\boldsymbol{\sigma}_i=(\sigma_i^x,\sigma_i^y,\sigma_i^z). We set ℏ=1\hbar=1 and measure the couplings vv, ww, rr in units of an arbitrary energy scale J0J_0, so that the time τ\tau is measured in units of ℏ/J0\hbar/J_0. We use τ\tau for time throughout, since tt denotes the hopping amplitudes of the Hubbard model in Sec. IV. Rescaling all couplings by a factor λ\lambda rescales the dynamics as τ→τ/λ\tau\to\tau/\lambda, so the transfer probabilities depend only on the ratios of the couplings and on the products of couplings and time.

The Hamiltonian in Eq. (2) commutes with the total magnetization along zz,

[HN,∑iσiz]=0,(3)\left[H_N,\sum_i\sigma_i^z\right]=0, \tag*{(3)}

therefore, the Hamiltonian block-diagonalizes in subspaces of fixed number of spins up. For our initial state, we consider a normalized single-excitation state for the spin model. The 8N8N states with a single spin up are denoted as follows

∣1⟩=∣100…0⟩, ∣2⟩=∣010…0⟩,…, ∣8N⟩=∣000…1⟩.(4)|1\rangle=|100\ldots0\rangle,\ |2\rangle=|010\ldots0\rangle,\ldots,\ |8N\rangle=|000\ldots1\rangle. \tag*{(4)}

Singlet and triplet basis, and the single-excitation sector

The poly-bi-[8]-annulenylene molecule is symmetric under reflection about both the horizontal and the vertical axes. Since we are interested in information transfer from left to right, we use the horizontal reflection symmetry to block-diagonalize the Hamiltonian matrix into two 4N×4N4N\times4N blocks. This symmetry also has a clear physical interpretation: in this new basis, we study not the transfer of an individual spin up from site 1 to site 8 or 9, but rather the transfer of a triplet or singlet state from the pair of sites (1,16) to the pair of sites (8,9). The basis states with triplet and singlet symmetry are

∣j±⟩=12(∣j⟩±∣8N+1−j⟩),j=1,…,4N(5)|j_{\pm}\rangle=\frac{1}{\sqrt{2}}\left(|j\rangle\pm|8N+1-j\rangle\right),\qquad j=1,\ldots,4N \tag*{(5)}

In the single-excitation subspace, and for N=2N=2, the Hamiltonian can be written as two blocks of 8×88\times8 matrices, one for the triplet sector and one for the singlet, i.e.

H2=(h(+)00h(−))=h(+)⊕h(−).(6)H_2= \begin{pmatrix} h^{(+)} & 0\\ 0 & h^{(-)} \end{pmatrix} =h^{(+)}\oplus h^{(-)}. \tag*{(6)}

With the shorthand a≡3v+3w+2ra\equiv3v+3w+2r, b≡3v+2w+2rb\equiv3v+2w+2r and c≡3v+2wc\equiv3v+2w, the triplet block reads

h(+)=−2(av000000vbw000000wbv000000vbw000000wbv000000vbw000000wbv000000va),(7)h^{(+)}=-2 \begin{pmatrix} a&v&0&0&0&0&0&0\\ v&b&w&0&0&0&0&0\\ 0&w&b&v&0&0&0&0\\ 0&0&v&b&w&0&0&0\\ 0&0&0&w&b&v&0&0\\ 0&0&0&0&v&b&w&0\\ 0&0&0&0&0&w&b&v\\ 0&0&0&0&0&0&v&a \end{pmatrix}, \tag*{(7)}

and the singlet block is

h(−)=−2(av000000vcw000000wcv000000vbw000000wbv000000vcw000000wcv000000va).(8)h^{(-)}=-2 \begin{pmatrix} a&v&0&0&0&0&0&0\\ v&c&w&0&0&0&0&0\\ 0&w&c&v&0&0&0&0\\ 0&0&v&b&w&0&0&0\\ 0&0&0&w&b&v&0&0\\ 0&0&0&0&v&c&w&0\\ 0&0&0&0&0&w&c&v\\ 0&0&0&0&0&0&v&a \end{pmatrix}. \tag*{(8)}

Eigenenergies and spectral weights of eigenstates

The eigenenergies corresponding to the triplet-symmetry eigenvectors are

E1(+)=−2(2v+3w+2r)E2(+)=−2(4v+3w+2r)E3(+)=−2(3v+2w+2r−v2+w2)E4(+)=−2(3v+2w+2r+v2+w2)E5(+)=−2(3v+2w+2r−v2+w2−2vw)E6(+)=−2(3v+2w+2r+v2+w2−2vw)E7(+)=−2(3v+2w+2r−v2+w2+2vw)E8(+)=−2(3v+2w+2r+v2+w2+2vw)(9)\begin{aligned} E_1^{(+)}&=-2(2v+3w+2r)\\ E_2^{(+)}&=-2(4v+3w+2r)\\ E_3^{(+)}&=-2\left(3v+2w+2r-\sqrt{v^2+w^2}\right)\\ E_4^{(+)}&=-2\left(3v+2w+2r+\sqrt{v^2+w^2}\right)\\ E_5^{(+)}&=-2\left(3v+2w+2r-\sqrt{v^2+w^2-\sqrt{2vw}}\right)\\ E_6^{(+)}&=-2\left(3v+2w+2r+\sqrt{v^2+w^2-\sqrt{2vw}}\right)\\ E_7^{(+)}&=-2\left(3v+2w+2r-\sqrt{v^2+w^2+\sqrt{2vw}}\right)\\ E_8^{(+)}&=-2\left(3v+2w+2r+\sqrt{v^2+w^2+\sqrt{2vw}}\right) \tag*{(9)} \end{aligned}

And the eigenenergies associated with the singletsymmetry eigenvectors are

E1(−)=−2(3v+3w+r−v2+r2)E2(−)=−2(3v+3w+r+v2+r2)E3(−)=−2(3v+2w+r−v2+(w+r)2)E4(−)=−2(3v+2w+r+v2+(w+r)2)E5(−)=−2(3v+2w+r−v2+w2+r2−w2v2+2r2)E6(−)=−2(3v+2w+r+v2+w2+r2−w2v2+2r2)E7(−)=−2(3v+2w+r−v2+w2+r2+w2v2+2r2)E8(−)=−2(3v+2w+r+v2+w2+r2+w2v2+2r2)(10)\begin{aligned} E_1^{(-)} &= -2\left(3v+3w+r-\sqrt{v^2+r^2}\right) \\ E_2^{(-)} &= -2\left(3v+3w+r+\sqrt{v^2+r^2}\right) \\ E_3^{(-)} &= -2\left(3v+2w+r-\sqrt{v^2+(w+r)^2}\right) \\ E_4^{(-)} &= -2\left(3v+2w+r+\sqrt{v^2+(w+r)^2}\right) \\ E_5^{(-)} &= -2\left(3v+2w+r-\sqrt{v^2+w^2+r^2-w\sqrt{2v^2+2r^2}}\right) \\ E_6^{(-)} &= -2\left(3v+2w+r+\sqrt{v^2+w^2+r^2-w\sqrt{2v^2+2r^2}}\right) \\ E_7^{(-)} &= -2\left(3v+2w+r-\sqrt{v^2+w^2+r^2+w\sqrt{2v^2+2r^2}}\right) \\ E_8^{(-)} &= -2\left(3v+2w+r+\sqrt{v^2+w^2+r^2+w\sqrt{2v^2+2r^2}}\right) \tag*{(10)} \end{aligned}

In the triplet energies of Eq. (9), rr appears only through the common shift −4r-4r of every eigenvalue, whereas in the singlet energies of Eq. (10) it enters the square roots. Appendix A shows that the first property holds for any number of unit cells and even for site-dependent rung couplings: the rungs add a multiple of the identity to the triplet block, so its eigenvectors, and with them the triplet transfer probability, do not depend on rr.

The energies for the particular case w=3w=3 and r=1r=1 are shown in Figure 2. We denote by ∣ϕk⟩|\phi_k\rangle the eigenvector with eigenvalue EkE_k of Eq. (9) or Eq. (10), and by pk=∣⟨1±∣ϕk⟩∣2p_k=\lvert\langle1_{\pm}|\phi_k\rangle\rvert^2 its spectral weight on the initial state. The eigenvectors, and with them the weights, follow analytically from Eqs. (7) and (8) but are lengthy, and we do not list them. Figure 2 shows the weights; by mirror symmetry, those on ∣8±⟩|8_{\pm}\rangle are the same.

Energy spectrum and spectral weights for triplet and singlet eigenstates

Figure 2. (a) Energy spectrum for w=3w=3 and r=1r=1. Teal (rose) lines correspond to eigenstates with triplet (singlet) symmetry. Panels (b) and (c) show the spectral weights pk=∣⟨1±∣ϕk⟩∣2p_k=\lvert\langle1_{\pm}|\phi_k\rangle\rvert^2 of the triplet (b) and singlet (c) eigenvectors, for w=3w=3 and r=1r=1; kk labels the eigenvalues as in Eqs. (9) and (10). By mirror symmetry, the weights on ∣8±⟩|8_{\pm}\rangle are the same. The xx axes run over v∈[0,1]v\in[0,1], then over 1/v1/v from 1 to 0. The vertical dashed line marks v=4v=4, the value used for the triplet example of Sec. III.

Perfect, pretty good, and almost perfect transfer in the Heisenberg model

The Hamiltonian is block diagonal in the basis of Eq. (5), so a state prepared in one sector stays there. We therefore treat two independent problems: the transfer of the triplet state ∣1+⟩\lvert1_{+}\rangle to ∣8+⟩\lvert8_{+}\rangle under h(+)h^{(+)} [Eq. (7)], and of the singlet state ∣1−⟩\lvert1_{-}\rangle to ∣8−⟩\lvert8_{-}\rangle under h(−)h^{(-)} [Eq. (8)]. Their transfer probabilities are

P(±)(τ)=∣⟨8±∣e−ihτ∣1±⟩∣2(11)P^{(\pm)}(\tau)=\left\lvert\langle8_{\pm}\lvert e^{-ih\tau}\rvert1_{\pm}\rangle\right\rvert^{2} \tag*{(11)}

and we drop the superscript for statements that hold in both sectors. For transfer between other pairs of Eq. (5) we write Pj→k(±)(τ)=∣⟨k±∣e−ihτ∣j±⟩∣2P_{j\to k}^{(\pm)}(\tau)=\left\lvert\langle k_{\pm}\lvert e^{-ih\tau}\rvert j_{\pm}\rangle\right\rvert^{2}, so that P(±)(τ)=P1→8(±)(τ)P^{(\pm)}(\tau)=P_{1\to8}^{(\pm)}(\tau).

The network exhibits PT if P(τPT)=1P(\tau_{\mathrm{PT}})=1 at some finite time τPT\tau_{\mathrm{PT}} [2, 4], and PGT if the transfer probability gets arbitrarily close to one, even if it never reaches it [5, 6]:

∀ϵ>0 ∃ τϵ:P(τϵ)≥1−ϵ.(12)\forall\epsilon>0\ \exists\ \tau_{\epsilon}:P(\tau_{\epsilon})\geq1-\epsilon. \tag*{(12)}

Following Ref. [19], we call the earliest time at which P(τ)≥1−ϵP(\tau)\geq1-\epsilon the arrival time τϵ\tau_{\epsilon}. APT, the finite-time criterion introduced in § 1, is made precise in Sec. III C.

Consider either of the two 8×88\times8 blocks, denoted by hh. Each is a real symmetric tridiagonal matrix with nonzero nearest-neighbor off-diagonal elements. Its eigenvalues EkE_k are therefore nondegenerate, and every eigenvector ∣ϕk⟩\lvert\phi_k\rangle has nonzero amplitudes at both ends [22]. Each block is also mirror symmetric, hjk=h9−j,9−kh_{jk}=h_{9-j,9-k}. This reflection exchanges the ends within each block and is distinct from the reflection that defines the triplet and singlet sectors. Each eigenvector consequently has a definite parity sk=±1s_k=\pm1 under j→9−jj\to9-j, with successive eigenvectors in energy order having opposite parities [2, 23].

The endpoint amplitudes therefore satisfy ⟨8±∣ϕk⟩=sk⟨1±∣ϕk⟩\langle8_{\pm}\lvert\phi_k\rangle=s_k\langle1_{\pm}\lvert\phi_k\rangle, so the transfer amplitude within the chosen sector is

⟨8±∣e−ihτ∣1±⟩=∑k=18skpke−iEkτ,(13)\langle8_{\pm}\lvert e^{-ih\tau}\rvert1_{\pm}\rangle=\sum_{k=1}^{8}s_kp_ke^{-iE_k\tau}, \tag*{(13)}

where pk=∣⟨1±∣ϕk⟩∣2>0p_k=\left\lvert\langle1_{\pm}\lvert\phi_k\rangle\right\rvert^2>0 and ∑kpk=1\sum_kp_k=1. This is a weighted sum of unit complex numbers ske−iEkτs_ke^{-iE_k\tau}. Since the weights are positive and sum to one, the amplitude has unit magnitude if and only if all these numbers have the same phase. Thus P(τ)=1P(\tau)=1 if and only if

ske−iEkτ=eiθfor all k(14)s_ke^{-iE_k\tau}=e^{i\theta}\qquad\text{for all }k \tag*{(14)}

for some global phase θ\theta. Within the chosen sector, the mirror-even eigenvectors must therefore acquire a common dynamical phase, and the mirror-odd eigenvectors must acquire the opposite phase [2]. PT requires this condition to hold exactly at some finite time, whereas PGT requires it to be approached arbitrarily closely at suitable times. The transfer probability can also be written as a sum of cosines whose frequencies are the eigenvalue differences Ek−ElE_k-E_l.

As we show in Sec. III A, neither sector admits PT, so the relevant question is how quickly Eq. (14) can be approached. Rational frequencies can all be matched at periodic times, but each rationally independent irrational number adds a phase that must reach its target together with the others. As a result, the arrival time grows as

τϵ∼ϵ−KI/2,(15)\tau_{\epsilon}\sim\epsilon^{-K_{\mathrm{I}}/2}, \tag*{(15)}

where KIK_{\mathrm{I}} is the number of rationally independent irrational numbers among the frequencies Ek−ElE_k-E_l [19]. Couplings that make PGT practical are therefore those that minimize KIK_{\mathrm{I}}.

The eigenvalues of Eqs. (9) and (10) contain square roots, and KIK_{\mathrm{I}} is reduced by making as many of them rational as possible, i.e. by choosing couplings that are legs of Pythagorean triples. The conditions differ between the sectors because the rungs act differently.

In the triplet sector rr only shifts the spectrum (Sec. III D), so P(+)(τ)P^{(+)}(\tau) is that of two uncoupled eight-site chains [17], and the only removable root is v2+w2\sqrt{v^2+w^2}, which requires (v,w)(v,w) to be legs of a Pythagorean triple; the nested roots X±=v2+w2±2vwX_{\pm}=\sqrt{v^2+w^2}\pm\sqrt{2vw} are irrational for any rational couplings. In the singlet sector the rungs act as an on-site potential 4r4r on sites 22, 33, 66 and 77, so rr enters the roots v2+r2\sqrt{v^2+r^2} and v2+(w+r)2\sqrt{v^2+(w+r)^2}, and the conditions are on (v,r)(v,r) and (v,w+r)(v,w+r). Satisfying all conditions at once requires large couplings, the smallest being (v,w,r)=(60,11,80)(v,w,r)=(60,11,80). Since P(+)P^{(+)} does not depend on rr, we instead choose the couplings separately for each sector.

For the triplet we take (v,w)=(4,3)(v,w)=(4,3) and r=1r=1; other values of rr shift every eigenvalue by −4(r−1)-4(r-1). In increasing order the eigenvalues are

E2(+)=−54,E8(+)=−2(20+α+),E4(+)=−50,E6(+)=−2(20+α−),E1(+)=−38,E5(+)=−2(20−α−),E3(+)=−30,E7(+)=−2(20−α+),(16)\begin{aligned} E_{2}^{(+)}&=-54, & E_{8}^{(+)}&=-2(20+\alpha_{+}),\\ E_{4}^{(+)}&=-50, & E_{6}^{(+)}&=-2(20+\alpha_{-}),\\ E_{1}^{(+)}&=-38, & E_{5}^{(+)}&=-2(20-\alpha_{-}),\\ E_{3}^{(+)}&=-30, & E_{7}^{(+)}&=-2(20-\alpha_{+}), \tag*{(16)} \end{aligned}

with α±=25±122\alpha_{\pm}=\sqrt{25\pm12\sqrt{2}}. The integer eigenvalues belong to mirror-even eigenvectors and the irrational ones to mirror-odd eigenvectors, alternating as stated above. Every frequency Ek−ElE_k-E_l is an integer combination of 11, α+\alpha_{+} and α−\alpha_{-}, so KI=2K_{\mathrm{I}}=2; the explicit form of P(+)(τ)P^{(+)}(\tau) is given in Appendix B.

For the singlet we take (v,w,r)=(8,9,6)(v,w,r)=(8,9,6), for which v2+r2=10\sqrt{v^2+r^2}=10, v2+(w+r)2=17\sqrt{v^2+(w+r)^2}=17 and 2(v2+2r2)=417\sqrt{2(v^2+2r^2)}=4\sqrt{17}, giving

E2(−)=−134,E8(−)=−2(48+γ+),E4(−)=−130,E6(−)=−2(48+γ−),E1(−)=−94,E5(−)=−2(48−γ−),E3(−)=−62,E7(−)=−2(48−γ+),(17)\begin{aligned} E_{2}^{(-)}&=-134, & E_{8}^{(-)}&=-2(48+\gamma_{+}),\\ E_{4}^{(-)}&=-130, & E_{6}^{(-)}&=-2(48+\gamma_{-}),\\ E_{1}^{(-)}&=-94, & E_{5}^{(-)}&=-2(48-\gamma_{-}),\\ E_{3}^{(-)}&=-62, & E_{7}^{(-)}&=-2(48-\gamma_{+}), \tag*{(17)} \end{aligned}

with γ±=181±3617\gamma_{\pm} = \sqrt{181 \pm36\sqrt{17}}. Again the integer eigenvalues are mirror-even, the irrational ones mirror-odd, and KI=2K_I = 2.

Perfect transfer

Neither sector exhibits PT for any positive couplings. By Eq. (14), PT at a time τPT\tau_{\mathrm{PT}} requires that eigenvalues of equal parity differ by integer multiples of 2π/τPT2\pi/\tau_{\mathrm{PT}}, and that eigenvalues of opposite parity differ by odd multiples of π/τPT\pi/\tau_{\mathrm{PT}}. In the triplet sector, E1(+)E_1^{(+)},..., E4(+)E_4^{(+)} are mirror-even and E5(+)E_5^{(+)},..., E8(+)E_8^{(+)} are mirror-odd. Applying the first condition within each parity class requires vv, ww, v2+w2\sqrt{v^2+w^2} and X±X_{\pm} all to be integer multiples of π/(2τPT)\pi/(2\tau_{\mathrm{PT}}). Since v,w>0v,w>0, the identity X+2−X−2=22vwX_+^2-X_-^2=2\sqrt{2}vw then implies 2∈Q\sqrt{2}\in\mathbb{Q}, a contradiction; the triplet sector therefore admits no PT.

In the singlet sector, with A≡v2+r2A\equiv\sqrt{v^2+r^2}, B≡v2+(w+r)2B\equiv\sqrt{v^2+(w+r)^2} and Y±Y_{\pm} the nested roots of Eq. (10), the same conditions make AA, BB, ww and Y±Y_{\pm} integer multiples of π/(2τPT)\pi/(2\tau_{\mathrm{PT}}), and the identity Y+2+Y−2=2(A2+w2)Y_+^2+Y_-^2=2(A^2+w^2) is then incompatible, modulo 4, with the parity conditions, so this sector admits no PT either. We therefore turn to PGT.

Pretty good transfer

In both examples KI=2K_I=2, so Eq. (15) predicts τϵ∼ϵ−1\tau_{\epsilon}\sim\epsilon^{-1}. Figure 3 tests this for the triplet transfer ∣1+⟩→∣8+⟩|1_+\rangle\to|8_+\rangle with (v,w)=(4,3)(v,w)=(4,3) and the singlet transfer ∣1−⟩→∣8−⟩|1_-\rangle\to|8_-\rangle with (v,w,r)=(8,9,6)(v,w,r)=(8,9,6). It shows, on a log-log scale, times τϵ\tau_{\epsilon} at which P(τϵ)=1−ϵP(\tau_{\epsilon})=1-\epsilon, obtained as described below; each is an upper bound on the arrival time for

Times $\tau_{\epsilon}/\pi$ at which $P(\tau_{\epsilon})=1-\epsilon$ for triplet and singlet transfer

Figure 3. Times τϵ/π\tau_{\epsilon}/\pi at which P(τϵ)=1−ϵP(\tau_{\epsilon}) = 1-\epsilon, for the triplet (teal, (v,w)=(4,3)(v,w)=(4,3)) and the singlet (rose, (v,w,r)=(8,9,6)(v,w,r)=(8,9,6)). Dashed lines are linear fits in log-log scale, with fitted exponents 1.008 (teal) and 1.005 (rose).

jjqjq_jP(+)(τj=qjπ/2)P^{(+)}(\tau_j=q_j\pi/2)
160.97
25350.9990
3148160.99993
412162790.99999992
5361445840.999999998
6104400454720.99999999997
7604118669630.999999999990
82789869825860.9999999999998
989589035791680.99999999999993
104464316125728290.999999999999998
jjqjq_jP(+)(τj=qjπ/2)P^{(+)}(\tau_j=q_j\pi/2)
160.97
25350.9990
3148160.99993
412162790.99999992
5361445840.999999998
6104400454720.99999999997
7604118669630.999999999990
82789869825860.9999999999998
989589035791680.99999999999993
104464316125728290.999999999999998

Table 1. Values of qjq_j and P(+)(τj)P^{(+)}(\tau_j) for v=4v=4 and w=3w=3 using rational approximations for the irrational numbers appearing in the eigenvalues Eq. (16): 25±122\sqrt{25\pm12\sqrt{2}}.

jjqjq_jP(−)(τj=qjπ/2)P^{(-)}(\tau_j=q_j\pi/2)
1270.990
24060.9998
3138430.99998
43607570.9999996
51085436530.999999990
625600567230.9999999990
711777641299580.999999999997
838428002939270.99999999999993
97024867913989120.999999999999996
10339378499059391000.999999999999999990

Table 2. Values of qjq_j and P(−)(τj)P^{(-)}(\tau_j) for v=8v=8, w=9w=9 and r=6r=6 using rational approximations for the irrational numbers appearing in the eigenvalues Eq. (17): 181±3617\sqrt{181\pm36\sqrt{17}}.

its ϵ\epsilon. Linear fits give exponents 1.008 (triplet) and 1.005 (singlet), in agreement with KI/2=1K_I/2=1.

Figure 3 was obtained as follows. At times τ=qπ/2\tau=q\pi/2 with integer qq, the integer eigenvalues in Eqs. (16) and (17) are all twice an odd number, so the mirror-even eigenvectors acquire the common phase (−1)q(-1)^q (other values of rr only add a global phase). The mirror-odd eigenvectors acquire e±iπqα±e^{\pm i\pi q\alpha_{\pm}} [or e±iπqγ±e^{\pm i\pi q\gamma_{\pm}}]. By Eq. (14), P(qπ/2)P(q\pi/2) is close to one when e±iπqα±≈(−1)q+1e^{\pm i\pi q\alpha_{\pm}}\approx(-1)^{q+1}, i.e. when qα+q\alpha_+ and qα−q\alpha_- are both close to integers with the parity of q+1q+1. Continued fractions give the best rational approximations of a single number, but there is no equally simple algorithm for approximating two numbers with a common denominator qq, so we find such qq with the Lenstra–Lenstra–Lovász (LLL) lattice basis reduction algorithm [24, 25]. We denote the successive values, ordered by increasing accuracy, by qjq_j, and the corresponding times by τj=qjπ/2\tau_j=q_j\pi/2. They are listed in Tables I and II. Figure 4 shows P(τ)P(\tau) around τ2\tau_2: the high-fidelity peak is isolated and sharp.

Transfer probabilities near the second characteristic time

Figure 4. P(+)(τ)P^{(+)}(\tau) for v=4v=4, w=3w=3, and r=1r=1 in the neighborhood of τ2\tau_2 of Table I, and P(−)(τ)P^{(-)}(\tau) for v=8v=8, w=9w=9 and r=6r=6 in the neighborhood of τ2\tau_2 of Table II.

We now ask whether PGT also holds between bulk sites. The condition for high transfer at τj\tau_j, Eq. (14), does not depend on the site. The same times τj\tau_j therefore give high transfer between every mirror pair of the triplet chain, ∣j+⟩→∣(9−j)+⟩|j_+\rangle\to|(9-j)_+\rangle. Applying the transfer twice returns the state to its origin, so the retrieval probability P1→1(+)P^{(+)}_{1\to1} is also close to one at 2τj=qjπ2\tau_j=q_j\pi. Table III confirms this for (v,w)=(4,3)(v,w)=(4,3): all these probabilities approach one as jj increases.

jjqjq_jP2→7(+)(τj=qjπ/2)P^{(+)}_{2\to7}(\tau_j=q_j\pi/2)P4→5(+)(τj=qjπ/2)P^{(+)}_{4\to5}(\tau_j=q_j\pi/2)P1→8(+)(τj=qjπ)P^{(+)}_{1\to8}(\tau_j=q_j\pi)
160.9620.9680.88
25350.9900.99900.996
3148160.999950.999930.9997
412162790.999999940.999999920.9999997
5361445840.9999999980.9999999980.999999990
6104400454720.999999999980.999999999970.99999999990
7604118669630.9999999999900.9999999999900.99999999996
82789869825860.99999999999980.99999999999970.9999999999991
989589035791680.999999999999950.999999999999930.9999999999997
104464316125728290.9999999999999980.9999999999999970.999999999999991

Table III. Bulk transfer and retrieval in the triplet sector for (v,w)=(4,3)(v,w)=(4,3), at the times τj=qjπ/2\tau_j=q_j\pi/2 of Table I (retrieval at 2τj=qjπ2\tau_j=q_j\pi). At these times P3→6(+)=P2→7(+)P^{(+)}_{3\to6}=P^{(+)}_{2\to7}, and P4→5(+)P^{(+)}_{4\to5} coincides with P1→8(+)P^{(+)}_{1\to8} in Table I.

At the times τ=qπ/2\tau=q\pi/2 we also find P3→6(+)=P2→7(+)P^{(+)}_{3\to6}=P^{(+)}_{2\to7} and P4→5(+)=P1→8(+)P^{(+)}_{4\to5}=P^{(+)}_{1\to8}, although these probabilities differ at generic times. This follows from a symmetry of the mirror-odd eigenvectors that relates sites ii and 5−i5-i. Table III therefore lists only P2→7(+)P^{(+)}_{2\to7} and P4→5(+)P^{(+)}_{4\to5}.

Almost perfect transfer

The scaling law of Eq. (15) makes PGT alone impractical, but a network that exhibits PGT is a natural candidate for APT at experimentally relevant times. Since rescaling all couplings by λ\lambda rescales time by 1/λ1/\lambda (Sec. II B), any target time could be met by speeding up the dynamics, so APT must be defined at a fixed energy scale. We keep ww fixed and say that the network exhibits

APT for a tolerance Δ\Delta at a time τΔ\tau_\Delta if

∃ (vΔ,rΔ):P(τΔ)>1−Δ.(18)\exists\,(v_\Delta,r_\Delta):\quad P(\tau_\Delta)>1-\Delta. \tag*{(18)}

Here Δ\Delta is the maximum tolerable transmission error and τΔ\tau_\Delta an experimentally realistic time. Reference [23] uses the name almost perfect state transfer for the property called PGT here (see also Ref. [6]); we reserve the name APT for the finite-time criterion of Eq. (18).

As an example we keep w=3w=3 and fix τ=535π/2\tau=535\pi/2, the time τ2\tau_2 of Table I at which (v,w)=(4,3)(v,w)=(4,3) already gives P(+)=0.9990P^{(+)}=0.9990, and look for high values of the transfer probability as a function of vv, sampled in steps of 10−410^{-4} for the triplet and 2×10−42\times10^{-4} for the singlet. The probabilities P(+)(τ=535π/2)P^{(+)}(\tau=535\pi/2) and P(−)(τ=535π/2)P^{(-)}(\tau=535\pi/2), the latter with r=1r=1, are shown in Figure 5.

Transfer probabilities as a function of $v$

Figure 5. (a) shows P(+)(τ=535π/2)P^{(+)}(\tau=535\pi/2) as a function of vv for w=3w=3 in the triplet state. The dark gray circles show the 33 peaks where P(+)>0.9P^{(+)}>0.9, corresponding to Δ<0.1\Delta<0.1. (b) shows P(−)(τ=535π/2)P^{(-)}(\tau=535\pi/2) as a function of vv for w=3w=3, r=1r=1 in the singlet state. The dark gray circles show the 34 peaks where P(−)>0.9P^{(-)}>0.9, corresponding to Δ<0.1\Delta<0.1.

Tuning in P(−)P^{(-)}

The probabilities are highly oscillating functions, and for a given time at which P(+)P^{(+)} is almost zero or one, P(−)P^{(-)} could take any value between zero and one. The fact that P(+)P^{(+)} does not depend on rr may be used to tune P(−)P^{(-)}, either close to one, or close to zero. Examples of the four possible situations are given in Table IV.

Then, suppose we have to transmit a triplet state, but the preparation leaves a small singlet component,

∣ψ(0)⟩=ω+∣1+⟩+ω−∣1−⟩,(19)|\psi(0)\rangle=\omega_{+}|1+\rangle+\omega_{-}|1-\rangle, \tag*{(19)}

where ∣ω+∣2+∣ω−∣2=1|\omega_{+}|^{2}+|\omega_{-}|^{2}=1 and ∣ω+∣2≫∣ω−∣2|\omega_{+}|^{2}\gg|\omega_{-}|^{2}.

Since HNH_N does not couple the two sectors, the probabilities of finding the triplet and the singlet at the target are

∣⟨8±∣e−iHNτ∣ψ(0)⟩∣2=∣ω±∣2P(±)(τ).(20)\left|\langle8\pm|e^{-iH_N\tau}|\psi(0)\rangle\right|^{2}=|\omega_{\pm}|^{2}P^{(\pm)}(\tau). \tag*{(20)}

To suppress the singlet component for (v,w)=(4,3)(v,w)=(4,3), we can set r=4.828r=4.828. At τ=535π/2\tau=535\pi/2, Table IV then gives a singlet probability ∣ω−∣2P(−)=3×10−5∣ω−∣2|\omega_{-}|^{2}P^{(-)}=3\times10^{-5}|\omega_{-}|^{2}, the product of two small quantities, while the triplet is still transferred with P(+)=0.9990P^{(+)}=0.9990.

τ\taurrP1→8(+)(τ)P^{(+)}_{1\to8}(\tau)P1→8(−)(τ)P^{(-)}_{1\to8}(\tau)
535π/2535\pi/29.88890.99900.997
535π/2535\pi/24.8280.99900.00003
535π535\pi6.1110.00070.993
535π535\pi5.9290.00070.000007

Table IV. Values of P1→8(+)(τ)P^{(+)}_{1\to8}(\tau) and P1→8(−)(τ)P^{(-)}_{1\to8}(\tau) close to one or close to zero for v=4v=4, w=3w=3 and selected values of τ\tau and rr.

Robustness to coupling disorder

In an experiment the couplings will not take exactly their design values, and static disorder is known to degrade state transfer in spin chains [26, 27, 28]. We model it by replacing each coupling Jb∈{v,w,r}J_b\in\{v,w,r\} of Eq. (2) by Jb(1+ub)J_b(1+u_b), with ubu_b drawn uniformly from [−η,η][-\eta,\eta] independently for every bond, and average over 400 realizations.

We compare the triplet transfer ∣1+⟩→∣8+⟩|1+\rangle\to|8+\rangle with the end-to-end transfer of an eight-site Heisenberg SSH chain with the same couplings vv and ww. Without disorder the two are identical, since the triplet block is the single-arm block shifted by a constant Appendix A and [17], so any difference comes from how the two structures respond to disorder. Because disorder shifts the transfer peaks by a different amount in each realization, we measure the best transfer within τ0±0.5\tau_0\pm0.5, where τ0\tau_0 is the time of a transfer peak of the clean system. Unless stated otherwise, η=0.05\eta=0.05, (v,w)=(4,3)(v,w)=(4,3) and τ0=3π\tau_0=3\pi, the time τ1\tau_1 of Table I, where the best clean transfer in this window is 0.976.

Figure 6(a) shows the result as a function of rr. The SSH chain reaches 0.72. Without rungs the lattice does worse, 0.57: the two components of ∣1+⟩|1+\rangle travel along two arms with different disorder and lose their relative phase. For r≳0.2r\gtrsim0.2 the lattice overtakes the SSH chain, and it reaches 0.77–0.79 for 1≲r≲201\lesssim r\lesssim20. The rungs lock the two arms together, so the triplet sees the average of the disorder on the two arms, which fluctuates less than the disorder on either arm. Consistently, the advantage disappears when the disorder is the same on mirror-image bonds (0.72 at r=10r=10), and disorder on the rungs themselves does not affect the triplet transfer, because the rungs only shift the triplet block by a constant Eq. (34). Very strong rungs help less (0.73 at r=50r=50 and 100), so the rung coupling has an optimal range.

Triplet transfer under coupling disorder

Figure 6. Triplet transfer under coupling disorder of strength η=0.05\eta= 0.05: best transfer within τ0±0.5\tau_0 \pm0.5 of the clean peak time τ0\tau_0, averaged over 400 realizations. (a) As a function of rr for (v,w)=(4,3)(v,w) = (4,3) and τ0=3π\tau_0 = 3\pi; r=0r = 0 is plotted at the left edge. Error bars give the standard error of the mean. The horizontal lines give the clean value and the eight-site Heisenberg SSH chain. (b) As a function of v/wv/w for w=3w = 3 and r=10r = 10, with τ0\tau_0 the highest clean peak in 0<τ<300 < \tau< 30; the dashed line marks v/w=4/3v/w = 4/3.

Figure 6(b) repeats the comparison for other coupling ratios at r=10r=10. The lattice is never significantly below the SSH chain. The gain depends on the coupling ratio: it is negligible at v/w=0.2v/w=0.2 and 1.1, where disorder affects the peak only weakly, and largest, 0.07, at (v,w)=(4,3)(v,w)=(4,3).

The late peak at τ=535π/2\tau=535\pi/2 used in Secs. III B and III C is much more fragile, because the errors in the phases grow with time. At η=10−3\eta=10^{-3} the best triplet transfer near this time falls from 0.9994 to 0.69, and the singlet transfer, suppressed to 3×10−53\times10^{-5} with r=4.828r=4.828 in Table IV, rises to 0.09; only η≈10−4\eta\approx10^{-4} preserves both (0.992 and 0.001). For realistic disorder the early peaks are therefore the useful ones, as we also find for the Hubbard model below.

Transfer in the Hubbard Model

The Heisenberg couplings of Eq. (2) arise naturally as the superexchange of electrons, or of fermionic atoms in an optical lattice, in a half-filled Hubbard model. Unlike the spin model, the Hubbard model retains charge fluctuations: an electron can virtually or really occupy a site together with a second electron. State transfer has also been studied for particles hopping on engineered bosonic and fermionic networks [29]; the Hubbard model adds the on-site interaction that generates the exchange couplings. Here we ask which features of the transfer found above survive in this setting. The analysis is exact within the relevant sector of the Hubbard model and involves no truncation.

Model and strong-coupling limit

On the network of Figure 1 we consider

HHub=−∑b=⟨ij⟩tb∑σ=↑,↓(ciσ†cjσ+cjσ†ciσ)+U∑ini↑ni↓,(21)H_{\mathrm{Hub}}=-\sum_{b=\langle ij\rangle}t_b\sum_{\sigma=\uparrow,\downarrow}\left(c_{i\sigma}^{\dagger}c_{j\sigma}+c_{j\sigma}^{\dagger}c_{i\sigma}\right)+U\sum_i n_{i\uparrow}n_{i\downarrow}, \tag*{(21)}

with hopping tb∈{tv,tw,tr}t_b \in\{t_v,t_w,t_r\} on the bonds labeled vv, ww, rr, and on-site repulsion U>0U > 0. At half filling and U≫∣tb∣U \gg|t_b|, degenerate perturbation theory (equivalently a Schrieffer–Wolff transformation [30, 31]) reduces Eq. (21) to the antiferromagnetic Heisenberg model [32, 33, 34]

Heff=∑b(4tb2/U)(Si⋅Sj−14)+O(tb4/U3).(22)H_{\mathrm{eff}}=\sum_b\left(4t_b^2/U\right)\left(\mathbf{S}_i\cdot\mathbf{S}_j-\frac{1}{4}\right)+O(t_b^4/U^3). \tag*{(22)}

With Si=σi/2\mathbf{S}_i=\boldsymbol{\sigma}_i/2, in the Pauli normalization of Eq. (2) each bond contributes (tb2/U)(σi⋅σj−1)(t_b^2/U)(\boldsymbol{\sigma}_i\cdot\boldsymbol{\sigma}_j-1), so the couplings are

v=tv2U,w=tw2U,r=tr2U.(23)v=\frac{t_v^2}{U},\qquad w=\frac{t_w^2}{U},\qquad r=\frac{t_r^2}{U}. \tag*{(23)}

We take the strong-coupling limit U/tb→∞U/t_b\to\infty at fixed (v,w,r)(v,w,r), i.e. with tb=JbUt_b=\sqrt{J_bU} for Jb∈{v,w,r}J_b\in\{v,w,r\}. The exchange couplings, and with them every time scale of the spin dynamics, then stay fixed, so the Hubbard and Heisenberg dynamics can be compared at the same time τ\tau (ℏ=1\hbar=1, energies in the units of Eq. (2)). We quantify the interaction strength by U/tvU/t_v.

The couplings obtained this way are antiferromagnetic, whereas Eq. (2) is ferromagnetic. Writing HFM≡−∑bJb σi⋅σjH_{\mathrm{FM}}\equiv-\sum_bJ_b\,\boldsymbol{\sigma}_i\cdot\boldsymbol{\sigma}_j for Eq. (2), the leading term of Eq. (22) is

HAFM≡∑bJb(σi⋅σj−1)=−HFM−Jtot,Jtot≡∑bJb=8v+6w+4r.(24)H_{\mathrm{AFM}}\equiv\sum_bJ_b(\boldsymbol{\sigma}_i\cdot\boldsymbol{\sigma}_j-1)=-H_{\mathrm{FM}}-J_{\mathrm{tot}}, \qquad J_{\mathrm{tot}}\equiv\sum_bJ_b=8v+6w+4r. \tag*{(24)}

where JtotJ_{\mathrm{tot}} sums the couplings over the 8 vv, 6 ww and 4 rr bonds of the network. Every eigenstate ∣ϕk⟩|\phi_k\rangle of HFMH_{\mathrm{FM}} with energy −Ek−Jtot-E_k-J_{\mathrm{tot}}: the spectrum is reflected and rigidly shifted. The shift fixes the zero of energy. The polarized state ∣F↑⟩|F_\uparrow\rangle, with HFM∣F↑⟩=−Jtot∣F↑⟩H_{\mathrm{FM}}|F_\uparrow\rangle=-J_{\mathrm{tot}}|F_\uparrow\rangle, has HAFMH_{\mathrm{AFM}} energy 0, which is also its exact Hubbard energy, since Pauli blocking forbids every hop in the fully polarized half-filled state. It is therefore the highest state of HAFMH_{\mathrm{AFM}}, not the ground state; the transfer protocol requires only that it be an exact eigenstate.

The sign does not affect the transfer. By Eq. (13), P(τ)=∑k,lskslpkplcos⁡[(Ek−El)τ]P(\tau)=\sum_{k,l}s_ks_lp_kp_l\cos[(E_k-E_l)\tau], with the parities sks_k and weights pkp_k defined there, where kk and ll both run over the eigenvectors ∣ϕk⟩|\phi_k\rangle of HFMH_{\mathrm{FM}} in the chosen sector. By Eq. (24), Ek−El→−(Ek−El)E_k-E_l\to-(E_k-E_l): the constant JtotJ_{\mathrm{tot}} cancels from every difference, and the sign reversal leaves each cosine unchanged. A ferromagnetic realization is also available with two-component bosons [35]. Their superexchange has been observed directly [36] and used to follow a single spin impurity moving on a polarized background [37].

The transfer problem in the Hubbard model

Sector. We take the single excitation to be one down spin on the polarized state ∣F↑⟩|F_{\uparrow}\rangle, i.e. N↑=15N_{\uparrow}=15 and N↓=1N_{\downarrow}=1, which Eq. (21) conserves separately. Flipping all spins, which commutes with Eqs. (2) and (21), maps this onto the single up spin of Sec. II B, so the transfer probabilities are the same in both conventions. The sector is spanned by the 256 states

∣d,h⟩=cd↓†ch↑∣F↑⟩,∣F↑⟩=c1↑†c1↓†⋯c16↑†∣0⟩(25)|d,h\rangle=c^\dagger_{d\downarrow}c_{h\uparrow}|F_{\uparrow}\rangle,\qquad|F_{\uparrow}\rangle=c^\dagger_{1\uparrow}c^\dagger_{1\downarrow}\cdots c^\dagger_{16\uparrow}|0\rangle \tag*{(25)}

labeled by the position dd of the down electron and the site hh from which an up electron is missing. For d=hd=h this is the single spin flip ∣n,n⟩=Sn−∣F↑⟩|n,n\rangle=S_n^-|F_{\uparrow}\rangle; for d≠hd\ne h site dd is doubly occupied and site hh empty. Equation (21) acts as

HHub∣d,h⟩=U(1−δdh)∣d,h⟩−∑d′tdd′∣d′,h⟩+∑h′thh′∣d,h′⟩(26)H_{\mathrm{Hub}}|d,h\rangle=U(1-\delta_{dh})|d,h\rangle-\sum_{d'}t_{dd'}|d',h\rangle+\sum_{h'}t_{hh'}|d,h'\rangle \tag*{(26)}

with tijt_{ij} the hopping on bond (i,j)(i,j). The sector is therefore exactly a two-body problem: a down particle hopping with −tb-t_b and an up hole hopping with +tb+t_b, with an on-site attraction −U-U relative to the continuum. The single spin flips (magnons) are its bound states. For U≫tbU\gg t_b there are 16 bound states near E=0E=0 and 240 doublon–holon states near E=UE=U. The mirror symmetries of Figure 1 are even permutations of the sites, so they leave ∣F↑⟩|F_{\uparrow}\rangle invariant and act on ∣d,h⟩|d,h\rangle without fermionic signs. The two sectors of opposite chain-exchange parity (128 states each) therefore remain exactly decoupled at any UU.

Initial and target states. The pair states of Eq. (5) become

∣1±⟩=12(S1−±S16−)∣F↑⟩=12(∣1,1⟩±∣16,16⟩)|1_{\pm}\rangle=\frac{1}{\sqrt{2}}(S_1^-\pm S_{16}^-)|F_{\uparrow}\rangle=\frac{1}{\sqrt{2}}(|1,1\rangle\pm|16,16\rangle)
∣8±⟩=12(∣8,8⟩±∣9,9⟩)(27)|8_{\pm}\rangle=\frac{1}{\sqrt{2}}(|8,8\rangle\pm|9,9\rangle) \tag*{(27)}

Because Sn−=cn↓†cn↑S_n^-=c^\dagger_{n\downarrow}c_{n\uparrow} is a bilinear, these carry no fermionic sign ambiguity: ∣1+⟩|1_+\rangle and ∣1−⟩|1_-\rangle are the triplet and singlet states of the pair (1,16)(1,16), as in the spin model.

Transfer amplitude. The triplet and singlet transfer probabilities are defined as in Eq. (11), P(±)(τ)=∣⟨8±∣e−iHHubτ∣1±⟩∣2P^{(\pm)}(\tau)=|\langle8_{\pm}|e^{-iH_{\mathrm{Hub}}\tau}|1_{\pm}\rangle|^2, and the amplitude has the form of

⟨8±∣e−iHHubτ∣1±⟩=∑k=1128skpke−iEkτ(28)\langle8_{\pm}|e^{-iH_{\mathrm{Hub}}\tau}|1_{\pm}\rangle=\sum_{k=1}^{128}s_kp_ke^{-iE_k\tau} \tag*{(28)}

where ∣ϕk⟩|\phi_k\rangle and EkE_k are now the eigenvectors and energies of HHubH_{\mathrm{Hub}} in the chosen sector, taken with definite parity sks_k under the reflection that exchanges the ends (see Sec. III), and pk=∣⟨1±∣ϕk⟩∣2p_k=|\langle1_{\pm}|\phi_k\rangle|^2. Keeping only the eight bound states in the sum gives the bound-band amplitude of Figure 8(e).

Figure 8. Transfer in the Hubbard model. (a)–(c) P(+)(τ)P^{(+)}(\tau) for (v,w,r)=(4,3,1)(v,w,r)=(4,3,1) at U/tv=5U/t_v=5, 15.815.8 and 5050 (solid), compared with the Heisenberg model (dashed). (d) Infidelity, 1−Pbest1-P_{\mathrm{best}}, in the windows 3π±0.53\pi\pm0.5 and 535π/2±3535\pi/2\pm3 [triplet (+)(+)] and 27π/2±0.527\pi/2\pm0.5 and 203π±3203\pi\pm3 [singlet (−)(-), (8,9,6)(8,9,6)]; dashed lines give the Heisenberg values in the same windows. Below U/tv≈100U/t_v\approx100 (535π/2535\pi/2) and 200200 (203π203\pi) a neighboring peak in the window is higher than the delayed one; these points are omitted. (e) Relative delay of the transfer peak (peak of the bound-band amplitude) in units of (tv/U)2(t_v/U)^2. Dotted lines give its limit κ\kappa of Eq. (29); colors as in (d). (f) Transfer at the Heisenberg peak time τHeis\tau_{\mathrm{Heis}} (×) and at the corrected time τHeis[1+κ(tv/U)2]\tau_{\mathrm{Heis}}[1+\kappa(t_v/U)^2] (•); colors as in (d).

Observable. In this sector nn↓(1−nn↑)=∣n,n⟩⟨n,n∣n_{n\downarrow}(1-n_{n\uparrow})=|n,n\rangle\langle n,n|. Hence, exactly at any UU,

P(±)(τ)=∣⟨8±∣e−iHHubτ∣1±⟩∣2=2∣⟨8,8∣e−iHHubτ∣1±⟩∣2=2⟨n8↓(1−n8↑)⟩τ\begin{aligned} P^{(\pm)}(\tau)=|\langle8_{\pm}|e^{-iH_{\mathrm{Hub}}\tau}|1_{\pm}\rangle|^2 \\ &=2|\langle8,8|e^{-iH_{\mathrm{Hub}}\tau}|1_{\pm}\rangle|^2=2\langle n_{8\downarrow}(1-n_{8\uparrow})\rangle_{\tau} \end{aligned}

where ⟨⋅⟩τ\langle\cdot\rangle_{\tau} is the expectation value in e−iHHubτ∣1±⟩e^{-iH_{\mathrm{Hub}}\tau}|1_{\pm}\rangle. The second line uses Eq. (27) and chain-exchange symmetry, ⟨9,9∣e−iHHubτ∣1±⟩=±⟨8,8∣e−iHHubτ∣1±⟩\langle9,9|e^{-iH_{\mathrm{Hub}}\tau}|1_{\pm}\rangle=\pm\langle8,8|e^{-iH_{\mathrm{Hub}}\tau}|1_{\pm}\rangle. The transfer probability is thus a single-site, spin- and density-resolved occupation, as measured in quantum-gas microscopes [38, 39].

All results below are obtained by exact diagonalization of the 256×256256\times256 matrix of Eq. (26). We use the triplet example (v,w,r)=(4,3,1)(v,w,r)=(4,3,1) of Secs. III B and III C and the singlet example (8,9,6)(8,9,6) of Sec. III B.

Spectrum

Figure 7 shows the structure implied by Eq. (26). For U/tv≳5U/t_v \gtrsim5 a band of eight bound states in each parity sector is separated from the doublon–holon continuum, and the bound states carry most of their weight on d=hd=h. The bound energies converge to the Heisenberg spectrum reflected and shifted as in Eq. (24), E=−Ek(+)−JtotE=-E_k^{(+)}-J_{\mathrm{tot}}, with Ek(+)E_k^{(+)} the triplet energies of Eq. (9) [inset of Figure 7]. The level ordering is inverted: the uniform magnon, the ferromagnetic ground state with E2(+)=−JtotE_2^{(+)}=-J_{\mathrm{tot}}, becomes the top of the band. It stays at E=0E=0 exactly for every UU, since ∑n∣n,n⟩\sum_n \lvert n,n\rangle is an exact eigenstate: applying Eq. (26), the UU term vanishes because d=hd=h, and the two hopping terms give −∑n,n′tnn′∣n′,n⟩-\sum_{n,n'}t_{nn'}\lvert n',n\rangle and +∑n,n′tnn′∣n,n′⟩+\sum_{n,n'}t_{nn'}\lvert n,n'\rangle, which cancel because tnn′=tn′nt_{nn'}=t_{n'n}.

Figure 7. Spectrum of the Hubbard model in the triplet (+) part of the single-excitation sector (N↑,N↓)=(15,1)(N_{\uparrow},N_{\downarrow})=(15,1), for (v,w,r)=(4,3,1)(v,w,r)=(4,3,1) with tb=JbUt_b=\sqrt{J_bU}. Main panel: all 128 levels in units of UU, with eight bound, magnon-like states near zero (teal) and 120 doublon–holon states near UU (rose). Inset: the bound band (solid) for 5≤U/tv≤305\le U/t_v\le30 and the Heisenberg prediction E=−Ek(+)−JtotE=-E_k^{(+)}-J_{\mathrm{tot}} (dashed), with Ek(+)E_k^{(+)} from Eq. (9) and JtotJ_{\mathrm{tot}} from Eq. (24).

Approach to the Heisenberg limit and the transfer time

Figure 8(a)–(c) compares the transfer probability with its Heisenberg counterpart. At U/tv=5U/t_v=5 the charge fluctuations reshape the signal. At U/tv=15.8U/t_v=15.8 the peak structure is recognizable but visibly delayed, and at U/tv=50U/t_v=50 the two curves are almost indistinguishable. Figure 8(d) quantifies this for the peaks studied in Secs. III B and III C. Because finite UU shifts the peaks, we quote the best transfer within the windows of Figure 8(d) in both models; for the Heisenberg model these values slightly exceed those at exactly τj\tau_j in Tables I and II:

  • The early triplet peak near τ=3π\tau=3\pi reaches 0.43, 0.76 and 0.94 at U/tv=5U/t_v=5, 10 and 15.8, and its Heisenberg value 0.976 from U/tv≈30U/t_v\approx30 on.

  • The pretty-good-transfer peak at τ=535π/2\tau=535\pi/2 requires much stronger interactions: 0.961 at U/tv=158U/t_v=158 and 0.9989 at U/tv=500U/t_v=500, against 0.9994 in the Heisenberg model.

  • The singlet peaks follow the same pattern. The early peak near 27π/227\pi/2 gives 0.60, 0.81, 0.987 and 0.989 at U/tv=11.2U/t_v=11.2, 35.4, 112 and 354 (Heisenberg: 0.989), and the pretty-good-transfer peak at τ=203π\tau=203\pi gives 0.990 and 0.997 at U/tv=354U/t_v=354 and 500 (Heisenberg: 0.9998).

The exact rr independence of the triplet transfer in the Heisenberg model is lost at O(tb4/U3)O(t_b^4/U^3), but the effect is small: at U/tv=50U/t_v=50 the best 3π3\pi peak varies only between 0.9763 and 0.9769 for 0.5≤r≤30.5\le r\le3 at fixed (v,w)=(4,3)(v,w)=(4,3). The main finite-UU effect is a systematic delay of the high-fidelity transfer peaks,

τHub≃τHeis[1+κ(tvU)2](29)\tau_{\mathrm{Hub}}\simeq\tau_{\mathrm{Heis}}\left[1+\kappa\left(\frac{t_v}{U}\right)^2\right] \tag*{(29)}

with κ=3.153\kappa=3.153 and 3.150 for the triplet peaks at τ1=3π\tau_1=3\pi and τ2=535π/2\tau_2=535\pi/2 of Table I and κ=2.455\kappa=2.455 and 2.458 for the singlet peaks at τ1=27π/2\tau_1=27\pi/2 and τ2=203π\tau_2=203\pi of Table II. Here τHeis\tau_{\mathrm{Heis}} is the time of the singlet peaks at τ1=27π/2\tau_1=27\pi/2 and τ2=203π\tau_2=203\pi of Table II. Here τHeis\tau_{\mathrm{Heis}} is the time of the Heisenberg maximum within the window of Figure 8(d), which differs slightly from the nominal τj\tau_j of the tables (for the first triplet peak, τHeis=9.4366\tau_{\mathrm{Heis}}=9.4366 rather than 3π=9.42483\pi=9.4248), τHub\tau_{\mathrm{Hub}} is the corresponding peak of the finite-UU bound-band amplitude, and κ\kappa is the U→∞U\to\infty limit of the scaled delay illustrated in Figure 8(e). At fixed couplings and initial state, κ\kappa is the same for every high-fidelity peak of a sector, up to small corrections from imperfect phase alignment, with κ≃3.15\kappa\simeq3.15 for the triplet and κ≃2.46\kappa\simeq2.46 for the singlet example.

The value of κ\kappa and its near independence of the peak follow from the spectral form of P(τ)P(\tau) (Appendix D). If finite UU shifts the transition frequencies Δkl=Ek(±)−El(±)\Delta_{kl}=E_k^{(\pm)}-E_l^{(\pm)} of Eq. (13) by δΔkl\delta\Delta_{kl}, the maximum of PP near a peak with aligned phases moves to

τHubτHeis−1≃−∑k,lpkplΔklδΔkl∑k,lpkplΔkl2(30)\frac{\tau_{\mathrm{Hub}}}{\tau_{\mathrm{Heis}}}-1\simeq-\frac{\sum_{k,l}p_kp_l\Delta_{kl}\delta\Delta_{kl}}{\sum_{k,l}p_kp_l\Delta_{kl}^{2}} \tag*{(30)}

a weighted average of the relative frequency shifts that does not depend on the peak. Equation (30) gives κ=3.150\kappa=3.150 for the triplet and 2.459 for the singlet, in agreement with the late peaks; the early peaks, whose phases are not perfectly aligned, differ by less than 0.004. Re-timing removes only this average shift. The non-uniform part of δΔkl\delta\Delta_{kl} leaves an infidelity that grows as [(τ(tv/U)2)]2[(\tau(t_v/U)^2)]^2 (0.047 for the 535π/2535\pi/2 peak at U/tv=150U/t_v=150), so a fixed error requires U/tv∝τU/t_v\propto\sqrt{\tau}, which is why the late peaks need much stronger interactions.

The O(tb4/U3)O(t_b^4/U^3) terms of Eq. (22) reduce the effective exchange (on a chain, Jb=tb2/U−4tb4/U3J_b=t_b^2/U-4t_b^4/U^3 in the normalization of Eq. (23)) [40, 41], so the magnon moves more slowly. Figure 8(f) compares readout at the Heisenberg time with readout at the delayed time:

  • Measured at the Heisenberg time, the triplet 3π3\pi peak falls to 0.48 at U/tv≈17U/t_v\approx17, the triplet 535π/2535\pi/2 peak to 0.40 at U/tv=150U/t_v=150, and the singlet peaks at 27π/227\pi/2 and 203π203\pi to 0.77 at U/tv≈59U/t_v\approx59 and 0.69 at U/tv=200U/t_v=200.

  • At the corrected time the same peaks give 0.94, 0.95, 0.95 and 0.92, essentially the best values of Figure 8(d).

Equation (29) is therefore the time map needed to carry the Heisenberg protocol over to the Hubbard model; because κ\kappa does not depend on the peak, the same map applies to all high-fidelity peaks of a sector. For pretty good transfer, finite UU has a practical cost: the 535π/2535\pi/2 and 203π203\pi peaks recover their Heisenberg values only for U/tvU/t_v of several hundred, where the exchange v=tv2/Uv=t_v^2/U is hundreds of times slower than the hopping. In a Hubbard realization the early peaks, which reach 0.94 at U/tv≈17U/t_v\approx17 after re-timing, are the relevant ones.

Conclusion and Outlook

We have studied quantum-state transfer in the 16-site poly-bi-[8]-annulenylene network using Heisenberg and Hubbard Hamiltonians. In the Heisenberg model, the reflection symmetry splits the dynamics into triplet and singlet sectors with closed-form spectra. Neither sector admits perfect transfer. Couplings chosen from Pythagorean triples, however, give pretty good transfer of both states, with arrival times scaling as τϵ∼ϵ−1\tau_\epsilon\sim\epsilon^{-1}.

Because the rungs only shift the triplet block by a constant, the triplet transfer is independent of rr, while the singlet transfer is not. The rung coupling can therefore enhance or suppress singlet transfer without affecting the triplet and without time-dependent control. The rungs also make the triplet transfer more robust to coupling disorder than a single SSH-type chain.

In the Hubbard model, the Heisenberg dynamics is recovered in the strong-interaction limit, while finite interactions affect the transfer in two ways. First, the high-fidelity transfer peaks are delayed, τHub≃τHeis[1+κ(tv/U)2]\tau_{\mathrm{Hub}}\simeq\tau_{\mathrm{Heis}}[1+\kappa(t_v/U)^2], where κ\kappa is a spectral average of the finite-UU frequency shifts and is therefore the same for every high-fidelity peak of a sector, with κ≈3.15\kappa\approx3.15 for the triplet and 2.462.46 for the singlet example. Read out at the Heisenberg time, the transfer drops sharply, so this re-timing is essential. Second, even at the corrected time, the peak fidelity remains below its Heisenberg value until the interaction is strong, since the remaining loss grows as [τHeis(tv/U)2]2[\tau_{\mathrm{Heis}}(t_v/U)^2]^2. The early triplet peak reaches its Heisenberg value only from U/tv≈30U/t_v\approx30 on, and the long-time pretty-good-transfer peaks only for U/tvU/t_v of several hundred, where they are also the most sensitive to disorder. In both models, the early peaks are therefore the practically relevant ones.

These results provide a starting point for symmetry-controlled state transfer in molecular-inspired quantum networks. Such networks could serve as internal state-transfer components in emitter-based quantum nodes, connecting quantum memories, processing elements and spin–photon interfaces. Establishing this will require further study of entanglement generation and preservation, decoherence, coupling to optical emitters, and the fidelity of spin–photon conversion.

Acknowledgments

This work was supported by the U.S. Department of Energy (DOE), Office of Basic Energy Sciences, under Grant No. DE-SC0026309; and the US Department of Energy, the Office of Science through the Quantum Science Center (QSC), a National Quantum Information Science Research Center. We also acknowledge support from the Goodnight Distinguished Chair in Quantum Computing at North Carolina State University.

Rung couplings and the triplet sector of two-leg ladders

We show that, in the single-excitation sector of any two-leg ladder, the rungs shift the triplet block by a multiple of the identity, whatever their number, positions and strengths. Number the 2M2M sites j=1,…,Mj=1,\ldots,M along one leg and jˉ≡2M+1−j\bar{j}\equiv2M+1-j along the other, and let Π\Pi be the reflection j↔jˉj\leftrightarrow\bar{j}. We write

H=H0+Hrung,Hrung=−∑j=1Mrj σj⋅σjˉ,(31)H=H_0+H_{\mathrm{rung}},\qquad H_{\mathrm{rung}}=-\sum_{j=1}^{M}r_j\,\boldsymbol{\sigma}_j\cdot\boldsymbol{\sigma}_{\bar{j}}, \tag*{(31)}

with rj=0r_j=0 where there is no rung, and require only that H0H_0 conserve ∑iσiz\sum_i\sigma_i^z and satisfy ΠH0Π=H0\Pi H_0\Pi=H_0; the leg couplings are otherwise arbitrary. Poly-bi-[8]-annulenylene has M=4NM=4N, H0H_0 given by the vv and ww bonds of Eq. (2), and rj=rr_j=r for j≡2,3(mod4)j\equiv2,3\pmod{4}, rj=0r_j=0 otherwise.

In the single-excitation sector, σj⋅σjˉ\boldsymbol{\sigma}_j\cdot\boldsymbol{\sigma}_{\bar{j}} has diagonal element −1-1 on ∣j⟩\lvert j\rangle and ∣jˉ⟩\lvert\bar{j}\rangle and +1+1 on every other state, and couples ∣j⟩\lvert j\rangle to ∣jˉ⟩\lvert\bar{j}\rangle with matrix element 22. Hence, for k=1,…,Mk=1,\ldots,M,

⟨k∣Hrung∣k⟩=⟨kˉ∣Hrung∣kˉ⟩=rk−∑j≠krj,\langle k|H_{\mathrm{rung}}|k\rangle=\langle\bar{k}|H_{\mathrm{rung}}|\bar{k}\rangle=r_k-\sum_{j\ne k}r_j,
⟨k∣Hrung∣kˉ⟩=−2rk,(32)\langle k|H_{\mathrm{rung}}|\bar{k}\rangle=-2r_k, \tag*{(32)}

and all other matrix elements vanish. In the basis ∣k±⟩=(∣k⟩±∣kˉ⟩)/2\lvert k_\pm\rangle=(\lvert k\rangle\pm\lvert\bar{k}\rangle)/\sqrt{2}, which is Eq. (5) for M=4NM=4N, HrungH_{\mathrm{rung}} is therefore diagonal, with

⟨k±∣Hrung∣k±⟩=rk−∑j≠krj∓2rk={−R(triplet),4rk−R(singlet),(33)\langle k_\pm|H_{\mathrm{rung}}|k_\pm\rangle=r_k-\sum_{j\ne k}r_j\mp2r_k= \begin{cases} -R & \text{(triplet)},\\ 4r_k-R & \text{(singlet)}, \end{cases} \tag*{(33)}

where R=∑jrjR=\sum_jr_j. Physically, σj⋅σjˉ=+1\boldsymbol{\sigma}_j\cdot\boldsymbol{\sigma}_{\bar{j}}=+1 whenever the pair (j,jˉ)(j,\bar{j}) is polarized or in its triplet state, and in the triplet sector this holds for every pair. Since Π\Pi commutes with H0H_0 and with every rung term, the two sectors decouple (Sec. II C), and the triplet block is

h(+)({rj})=h0(+)−RIM,(34)h^{(+)}(\{r_j\})=h_0^{(+)}-R I_M, \tag*{(34)}

with h0(+)h_0^{(+)} the triplet block of H0H_0. The rungs thus change triplet amplitudes only by the global phase eiRτe^{iR\tau}, so every triplet transfer probability [Eq. (13)], between the ends or between bulk pairs, is independent of the rung couplings. In the singlet sector the rungs instead add the site-dependent potential 4rk4r_k, which changes both the eigenvalues and the eigenvectors [Eq. (10)].

Explicit transfer probability

For (v,w)=(4,3)(v,w)=(4,3), inserting the eigenvalues of Eq. (16) and the corresponding eigenvectors of Eq. (7) into Eq. (13) gives P(+)(τ)P^{(+)}(\tau). It is the same for every rr, since rr only shifts all triplet eigenvalues by a constant:

P(+)(τ)=1269600{337[40cos⁡(4τ)+10cos⁡(8τ)+40cos⁡(12τ)+25cos⁡(16τ)+16cos⁡(20τ)+10cos⁡(24τ)]+200α+2cos⁡(4α−τ)+200α−2cos⁡(4α+τ)+50(337+337)cos⁡(250−2337 τ)+50(337−337)cos⁡(250+2337 τ)+∑σ=±[−16850cos⁡(2αστ)(cos⁡(2τ)+2cos⁡(10τ)+cos⁡(14τ))+10βσsin⁡(2αστ)(5sin⁡(2τ)−6sin⁡(10τ)−5sin⁡(14τ))+43583]}.(35)\begin{aligned} P^{(+)}(\tau)=\frac{1}{269600}\Bigl\{& 337\bigl[40\cos(4\tau)+10\cos(8\tau)+40\cos(12\tau)+25\cos(16\tau)+16\cos(20\tau)+10\cos(24\tau)\bigr]\\ &+200\alpha_+^2\cos(4\alpha_-\tau)+200\alpha_-^2\cos(4\alpha_+\tau)\\ &+50(337+\sqrt{337})\cos\bigl(2\sqrt{50-2\sqrt{337}}\,\tau\bigr)\\ &+50(337-\sqrt{337})\cos\bigl(2\sqrt{50+2\sqrt{337}}\,\tau\bigr)\\ &+\sum_{\sigma=\pm}\Bigl[-16850\cos(2\alpha_\sigma\tau)\bigl(\cos(2\tau)+2\cos(10\tau)+\cos(14\tau)\bigr)\\ &\qquad+10\beta_\sigma\sin(2\alpha_\sigma\tau)\bigl(5\sin(2\tau)-6\sin(10\tau)-5\sin(14\tau)\bigr)+43583\Bigr]\Bigr\}. \tag*{(35)} \end{aligned}

where α±=25±122\alpha_\pm=\sqrt{25\pm12\sqrt{2}} and β±=337(137±962)\beta_\pm=\sqrt{337(137\pm96\sqrt{2})}. The frequencies are the eigenvalue differences of Eq. (16): the integers 4,8,…,244,8,\ldots,24, the combinations 2α±±22\alpha_\pm\pm2, 2α±±102\alpha_\pm\pm10, 2α±±142\alpha_\pm\pm14, 4α±4\alpha_\pm, and 2(α+∓α−)=250∓23372(\alpha_+\mp\alpha_-)=2\sqrt{50\mp2\sqrt{337}}. All of them are integer combinations of 11, α+\alpha_{+} and α−\alpha_{-}, so KI=2K_{I}=2.

OTOC and transfer

OTOCs are commonly defined as four-point correlation functions

FWV(τ)=⟨W†(τ)V†W(τ)V⟩,W(τ)=eiHτWe−iHτ,(36)F_{WV}(\tau)=\langle W^{\dagger}(\tau)V^{\dagger}W(\tau)V\rangle,\qquad W(\tau)=e^{iH\tau}We^{-iH\tau}, \tag*{(36)}

and quantify the spreading of initially local operators under many-body dynamics [20]. Fidelity OTOCs [21] take VV to be the projector on the initial state. We make the same choice, V=∣a⟩⟨a∣V=\lvert a\rangle\langle a\rvert, take W=∣b⟩⟨b∣W=\lvert b\rangle\langle b\rvert to be the projector on the target state, and evaluate the OTOC in ∣a⟩\lvert a\rangle. Since V∣a⟩=∣a⟩V\lvert a\rangle=\lvert a\rangle and VV has rank one,

F(τ)=⟨a∣W(τ)VW(τ)V∣a⟩=⟨a∣W(τ)∣a⟩2=∣⟨b∣e−iHτ∣a⟩∣4=Pa→b(τ)2.(37)\begin{aligned} F(\tau)&=\langle a\lvert W(\tau)VW(\tau)V\rvert a\rangle\\ &=\langle a\lvert W(\tau)\rvert a\rangle^{2}\\ &=\left\lvert\langle b\lvert e^{-iH\tau}\rvert a\rangle\right\rvert^{4}\\ &=P_{a\to b}(\tau)^{2}. \tag*{(37)} \end{aligned}

For ∣a⟩=∣1⟩\lvert a\rangle=\lvert1\rangle and ∣b⟩=∣9⟩\lvert b\rangle=\lvert9\rangle this is the density OTOC ⟨1∣n9(τ)n1n9(τ)n1∣1⟩\langle1\lvert n_{9}(\tau)n_{1}n_{9}(\tau)n_{1}\rvert1\rangle, since in the single-excitation sector ni=∣i⟩⟨i∣n_{i}=\lvert i\rangle\langle i\rvert. The squared commutator likewise reduces to

⟨a∣[W(τ),V]†[W(τ),V]∣a⟩=Pa→b(τ)[1−Pa→b(τ)].\langle a\lvert[W(\tau),V]^{\dagger}[W(\tau),V]\rvert a\rangle=P_{a\to b}(\tau)\left[1-P_{a\to b}(\tau)\right].

Both identities hold for any Hamiltonian, including the tight-binding and Hubbard models. For this choice of operators the OTOC is therefore not an independent transport observable: it encodes the same amplitude as the transfer probability. This has a practical side: OTOCs are usually measured with time-reversal or interferometric protocols [42, 43], whereas Pa→b(τ)P_{a\to b}(\tau) is a direct occupation measurement at the target, so for state transfer the OTOC offers no experimental advantage. It also means that the time dependence of such projector OTOCs reflects only the propagation of the excitation, not scrambling.

Delay of the transfer peaks at finite UU

This appendix derives Eq. (30) and the infidelity that remains after re-timing. As in the definition of κ\kappa [Figure 8(e)], we keep the eight bound states in Eq. (28). Let ϵk\epsilon_k be their energies, labeled so that −ϵk−Jtot→Ek-\epsilon_k-J_{\mathrm{tot}}\to E_k for U→∞U\to\infty, where Ek=EkHE_k=E_k^{\mathrm{H}} are the Heisenberg energies of Eq. (9) or Eq. (10). We write

−ϵk−Jtot=Ek+δEk,δΔkl=δEk−δEl,(38)-\epsilon_k-J_{\mathrm{tot}}=E_k+\delta E_k,\qquad\delta\Delta_{kl}=\delta E_k-\delta E_l, \tag*{(38)}

with δEk∝(tv/U)2\delta E_k\propto(t_v/U)^2 at large UU. The reflection and shift do not change the transfer probability (§ 2 A), and we take the parities sks_k and the weights pkp_k from the Heisenberg model (see the end of this appendix). With Δkl=Ek−El\Delta_{kl}=E_k-E_l and ∑kpk=1\sum_k p_k=1,

1−P(τ)=∑k,lpkpl{1−skslcos⁡[(Δkl+δΔkl)τ]}.(39)1-P(\tau)=\sum_{k,l}p_kp_l\left\{1-s_ks_l\cos\left[(\Delta_{kl}+\delta\Delta_{kl})\tau\right]\right\}. \tag*{(39)}

We first assume that the phases are perfectly aligned at the Heisenberg peak τHeis\tau_{\mathrm{Heis}}, i.e. that Eq. (14) holds there. No peak satisfies this exactly (Sec. III A); the corrections are given at the end of this appendix. Then skslcos⁡(ΔklτHeis)=1s_ks_l\cos(\Delta_{kl}\tau_{\mathrm{Heis}})=1 and sin⁡(ΔklτHeis)=0\sin(\Delta_{kl}\tau_{\mathrm{Heis}})=0 for all k,lk,l. Near this peak, at τ=τHeis+δτ\tau=\tau_{\mathrm{Heis}}+\delta\tau, the argument of each cosine is ΔklτHeis+xkl\Delta_{kl}\tau_{\mathrm{Heis}}+x_{kl}, with

xkl=Δklδτ+δΔklτHeis(40)x_{kl}=\Delta_{kl}\delta\tau+\delta\Delta_{kl}\tau_{\mathrm{Heis}} \tag*{(40)}

up to the higher-order term δΔklδτ\delta\Delta_{kl}\delta\tau. Each term of Eq. (39) then becomes 1−cos⁡xkl1-\cos x_{kl} and to second order

1−P≃12∑k,lpkplxkl2=12(S2δτ2+2S1τHeisδτ+S0τHeis2),(41)\begin{aligned} 1-P\simeq\frac{1}{2}\sum_{k,l}p_kp_lx_{kl}^{2} \\ &=\frac{1}{2}\left(S_{2}\delta\tau^{2}+2S_{1}\tau_{\mathrm{Heis}}\delta\tau+S_{0}\tau_{\mathrm{Heis}}^{2}\right), \tag*{(41)} \end{aligned}

where

S2=∑k,lpkplΔkl2,S0=∑k,lpkplδΔkl2,S_{2}=\sum_{k,l}p_kp_l\Delta_{kl}^{2},\qquad S_{0}=\sum_{k,l}p_kp_l\delta\Delta_{kl}^{2},
S1=∑k,lpkplΔklδΔkl.(42)S_{1}=\sum_{k,l}p_kp_l\Delta_{kl}\delta\Delta_{kl}. \tag*{(42)}

There is no term linear in xklx_{kl}, because sin⁡(ΔklτHeis)=0\sin(\Delta_{kl}\tau_{\mathrm{Heis}})=0. The expansion requires only that the xklx_{kl} be small near the new peak. The phase errors δΔklτHeis\delta\Delta_{kl}\tau_{\mathrm{Heis}} at the old peak need not be: for the late peaks the error can be of order one where the transfer at τHeis\tau_{\mathrm{Heis}} has collapsed [Figure 8(f)].

Minimizing Eq. (41) over δτ\delta\tau gives the peak time τHub=τHeis+δτ\tau_{\mathrm{Hub}}=\tau_{\mathrm{Heis}}+\delta\tau with δτ/τHeis=−S1/S2\delta\tau/\tau_{\mathrm{Heis}}=-S_{1}/S_{2}, which is Eq. (30). The peak time τHeis\tau_{\mathrm{Heis}} has dropped out, so all aligned peaks of a sector are delayed by the same fraction. The ratio −S1/S2-S_{1}/S_{2} is the average of −δΔkl/Δkl-\delta\Delta_{kl}/\Delta_{kl} with weights pkplΔkl2p_kp_l\Delta_{kl}^{2}; for a uniform rescaling, δΔkl=−λΔkl\delta\Delta_{kl}=-\lambda\Delta_{kl}, it gives δτ/τHeis=λ\delta\tau/\tau_{\mathrm{Heis}}=\lambda to leading order. Since δΔkl∝(tv/U)2\delta\Delta_{kl}\propto(t_v/U)^2, Eq. (29) follows with κ=−lim⁡U→∞(U/tv)2S1/S2\kappa=-\lim_{U\to\infty}(U/t_v)^2S_{1}/S_{2}.

Inserting the optimal δτ\delta\tau into Eq. (41) gives the infidelity at τHub\tau_{\mathrm{Hub}},

1−P(τHub)≃τHeis22(S0−S12S2).(43)1-P(\tau_{\mathrm{Hub}})\simeq\frac{\tau_{\mathrm{Heis}}^{2}}{2}\left(S_{0}-\frac{S_{1}^{2}}{S_{2}}\right). \tag*{(43)}

By the Cauchy–Schwarz inequality, S12≤S0S2S_{1}^{2}\le S_{0}S_{2}, with equality only for a uniform rescaling of the frequencies. Equation (D6) is therefore the part of the frequency shifts that re-timing cannot remove. Since S0−S12/S2∝(tv/U)4S_{0}-S_{1}^{2}/S_{2}\propto(t_v/U)^4, it grows as [τHeis(tv/U)2]2[\tau_{\mathrm{Heis}}(t_v/U)^2]^2. For the 535π/2535\pi/2 peak at U/tv=150U/t_v=150 it gives 0.0473; the exact loss relative to the Heisenberg peak, at the corrected time of Figure 8(f), is also 0.0473. For the 203π203\pi peak at U/tv=200U/t_v = 200 the two values are 0.087 and 0.082.

Real peaks are not perfectly aligned, and the derivation neglects three effects. (i) At the Heisenberg maximum the term linear in δτ\delta\tau still vanishes, since its coefficient is proportional to dP/dτ\mathrm{d}P/\mathrm{d}\tau. However, the quadratic terms acquire the factor skslcos⁡(ΔklτHeis)s_k s_l \cos(\Delta_{kl}\tau_{\mathrm{Heis}}), and the change of the weights pkp_k at finite UU, of relative order (tv/U)2(t_v/U)^2, now shifts the peak at that order. Together these change κ\kappa by less than 0.004 for the early peaks and 0.001 for the late ones. (ii) Misalignment also adds a loss linear in δΔkl\delta\Delta_{kl}. It decreases only as τHeis(tv/U)2\tau_{\mathrm{Heis}}(t_v/U)^2 and accounts for most of the difference between Eq. (43) and the exact loss below about 1%. (iii) The initial state has a weight of order (tv/U)2(t_v/U)^2 on doublon–holon states, which lowers PP by a similar amount, negligible for the late peaks; its fast oscillations are excluded from κ\kappa, which is defined by the peak of the bound-band amplitude.

References

References

  1. [1]S. Bose, Quantum Communication through an Unmodulated Spin Chain, Physical Review Letters 91, 207901 (2003).pubmed.ncbi.nlm.nih.gov/14683398
  2. [2]A. Kay, Perfect, efficient, state transfer and its application as a constructive tool, International Journal of Quantum Information 08, 641 (2010).DOI
  3. [3]Y. P. Kandel, H. Qiao, S. Fallahi, G. C. Gardner, M. J. Manfra, and J. M. Nichol, Coherent spin-state transfer via Heisenberg exchange, Nature 573, 553 (2019).DOI
  4. [4]M. Christandl, N. Datta, A. Ekert, and A. J. Landahl, Perfect State Transfer in Quantum Spin Networks, Physical Review Letters 92, 187902 (2004).pubmed.ncbi.nlm.nih.gov/15169534
  5. [5]C. Godsil, S. Kirkland, S. Severini, and J. Smith, Number-Theoretic Nature of Communication in Quantum Spin Systems, Physical Review Letters 109, 050502 (2012).DOI
  6. [6]L. Banchi, G. Coutinho, C. Godsil, and S. Severini, Pretty good state transfer in qubit chains—The Heisenberg Hamiltonian, Journal of Mathematical Physics 58, 032202 (2017).
  7. [7]A. Gaita-Ariño, F. Luis, S. Hill, and E. Coronado, Molecular spins for quantum computation, Nature Chemistry 11, 301 (2019).DOI
  8. [8]P. Serra, A. Ferrón, and O. Osenda, Almost perfect quantum state transfer on isotropic and anisotropic Heisenberg spin chains with tailored site dependent exchange couplings, Physics Letters A 449, 128362 (2022), 2101.03194.DOI
  9. [9]D. Burgarth and S. Bose, Conclusive and arbitrarily perfect quantum-state transfer using parallel spin-chain channels, Physical Review A 71, 052315 (2005).DOI
  10. [10]C. B. Pushpan, K. Harikrishnan, and A. K. Pal, Quantum state transfer on 2D lattices: A low-energy effective Hamiltonian approach, Physics Letters A 511, 129543 (2024).DOI
  11. [11]M. Kempton, G. Lippner, and S.-T. Yau, Pretty good quantum state transfer in symmetric spin networks via magnetic field, Quantum Information Processing 16, 210 (2017).DOI
  12. [12]W. P. Su, J. R. Schrieffer, and A. J. Heeger, Soliton excitations in polyacetylene, Physical Review B 22, 2099 (1980).
  13. [13]M. P. Estarellas, I. D’Amico, and T. P. Spiller, Topologically protected localised states in spin chains, Scientific Reports 7, 42904 (2017).DOI
  14. [14]F. Mei, G. Chen, L. Tian, S.-L. Zhu, and S. Jia, Robust quantum state transfer via topological edge states in superconducting qubit chains, Physical Review A 98, 012331 (2018).DOI
  15. [15]L. Moragues, D. S. Acosta Coden, O. Osenda, and A. Ferrón, Anisotropic Heisenberg Su-Schrieffer-Heeger spin chain as a quantum channel, Physics Letters A 571, 131273 (2026).
  16. [16]V. Muruganandam, M. Sajjan, and S. Kais, Foray into the topology of poly-bi-[8]-annulenylene, Natural Sciences 3, e20230015 (2023).arxiv.org/abs/2305.06264
  17. [17]P. Serra, A. Ferrón, and O. Osenda, Exact solution of a family of staggered Heisenberg chains with conclusive pretty good quantum state transfer, Journal of Physics A: Mathematical and Theoretical 55, 405302 (2022).DOI
  18. [18]A. K. Lenstra, H. W. Lenstra, and L. Lovász, Factoring polynomials with rational coefficients, Mathematische Annalen 261, 515 (1982).
  19. [19]P. Serra, A. Ferrón, and O. Osenda, The scaling law of the arrival time of spin systems that present pretty good transmission, Journal of Physics A: Mathematical and Theoretical 57, 015304 (2024).arxiv.org/abs/2309.02207
  20. [20]S. Xu and B. Swingle, Scrambling Dynamics and Out-of-Time-Ordered Correlators in Quantum Many-Body Systems, PRX Quantum 5, 010201 (2024).DOI
  21. [21]R. J. Lewis-Swan, A. Safavi-Naini, J. J. Bollinger, and A. M. Rey, Unifying scrambling, thermalization and entanglement through measurement of fidelity out-of-time-order correlators in the Dicke model, Nature Communications 10, 1581 (2019).
  22. [22]B. N. Parlett, The Symmetric Eigenvalue Problem, Classics in Applied Mathematics No. 20 (Society for Industrial and Applied Mathematics, Philadelphia, 1998).
  23. [23]L. Vinet and A. Zhedanov, Almost perfect state transfer in quantum spin chains, Physical Review A 86, 052319 (2012).DOI
  24. [24]J. C. Lagarias, The Computational Complexity of Simultaneous Diophantine Approximation Problems, SIAM Journal on Computing 14, 196 (1985).
  25. [25]W. Bosma and I. Smeets, Finding simultaneous Diophantine approximations with prescribed quality, The Open Book Series 1, 167 (2013).arxiv.org/abs/1001.4455
  26. [26]G. De Chiara, D. Rossini, S. Montangero, and R. Fazio, From perfect to fractal transmission in spin chains, Physical Review A 72, 012323 (2005).DOI
  27. [27]D. Petrosyan, G. M. Nikolopoulos, and P. Lambropoulos, State transfer in static and dynamic spin chains with disorder, Physical Review A 81, 042307 (2010).DOI
  28. [28]A. Zwick, G. A. Álvarez, J. Stolze, and O. Osenda, Robustness of spin-coupling distributions for perfect quantum state transfer, Physical Review A 84, 022311 (2011).DOI
  29. [29]M.-H. Yung and S. Bose, Perfect state transfer, effective gates, and entanglement generation in engineered bosonic and fermionic networks, Physical Review A 71, 023310 (2005).DOI
  30. [30]J. R. Schrieffer and P. A. Wolff, Relation between the Anderson and Kondo Hamiltonians, Phys. Rev. 149, 491 (1966).
  31. [31]S. Bravyi, D. P. DiVincenzo, and D. Loss, Schrieffer–Wolff transformation for quantum many-body systems, Ann. Phys. 326, 2793 (2011).DOI
  32. [32]K. A. Chao, J. Spalek, and A. M. Olés, Kinetic exchange interaction in a narrow S-band, J. Phys. C: Solid State Phys. 10, L271 (1977).DOI
  33. [33]A. Auerbach, Interacting Electrons and Quantum Magnetism, Graduate Texts in Contemporary Physics (Springer, New York, 1994).DOI
  34. [34]P. Fazekas, Lecture Notes on Electron Correlation and Magnetism, Series in Modern Condensed Matter Physics, Vol. 5 (World Scientific, Singapore, 1999).
  35. [35]L.-M. Duan, E. Demler, and M. D. Lukin, Controlling spin exchange interactions in ultracold atoms in optical lattices, Phys. Rev. Lett. 91, 090402 (2003).
  36. [36]S. Trotzky, P. Cheinet, S. Fölling, M. Feld, U. Schnorrberger, A. M. Rey, A. Polkownikov, E. A. Demler, M. D. Lukin, and I. Bloch, Time-resolved observation and control of superexchange interactions with ultracold atoms in optical lattices, Science 319, 295 (2008).
  37. [37]T. Fukuhara, A. Kantian, M. Endres, M. Cheneau, P. Schauß, S. Hild, D. Bellem, U. Schollwöck, T. Giamarchi, C. Gross, I. Bloch, and S. Kuhr, Quantum dynamics of a mobile spin impurity, Nat. Phys. 9, 235 (2013).DOI
  38. [38]M. Boll, T. A. Hilker, G. Salomon, A. Omran, J. Nespolo, L. Pollet, I. Bloch, and C. Gross, Spin- and density-resolved microscopy of antiferromagnetic correlations in Fermi-Hubbard chains, Science 353, 1257 (2016).DOI
  39. [39]M. F. Parsons, A. Mazurenko, C. S. Chiu, G. Ji, D. Greif, and M. Greiner, Site-resolved measurement of the spin-correlation function in the Fermi-Hubbard model, Science 353, 1253 (2016).
  40. [40]M. Takahashi, Half-filled Hubbard model at low temperature, J. Phys. C: Solid State Phys. 10, 1289 (1977).
  41. [41]A. H. MacDonald, S. M. Girvin, and D. Yoshioka, t/u expansion for the Hubbard model, Phys. Rev. B 37, 9753 (1988).pubmed.ncbi.nlm.nih.gov/9944372
  42. [42]B. Swingle, G. Bentsen, M. Schleier-Smith, and P. Hayden, Measuring the scrambling of quantum information, Physical Review A 94, 040302 (2016).DOI
  43. [43]M. Gärttner, J. G. Bohnet, A. Safavi-Naini, M. L. Wall, J. J. Bollinger, and A. M. Rey, Measuring out-of-time-order correlations and multiple quantum spectra in a trapped-ion quantum magnet, Nature Physics 13, 781 (2017).

Paper details

vlnc:W_89dkma3w
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
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