betaFeedback

WorksPhysics and Astronomy

Triplon current generation in solids

Abstract

A triplon refers to a fictitious particle that carries angular momentum S=1S = 1 corresponding to the elementary excitation in a broad class of quantum dimerized spin systems. Such systems without magnetic order have long been studied as a testing ground for quantum properties of spins. Although triplons have been found to play a central role in thermal and magnetic properties in dimerized magnets with singlet correlation, a spin angular momentum flow carried by triplons, a triplon current, has not been detected yet. Here we report spin Seebeck effects induced by a triplon current: triplon spin Seebeck effect, using a spin-Peierls system CuGeO3_3. The result shows that the heating-driven triplon transport induces spin current whose sign is positive, opposite to the spin-wave cases in magnets. The triplon spin Seebeck effect persists far below the spin-Peierls transition temperature, being consistent with a theoretical calculation for triplon spin Seebeck effects.

Introduction

Spin Seebeck effects[1] (SSEs) refer to the generation of a spin current, a flow of spin angular momentum of electrons, from a temperature gradient applied to a bilayer system comprising a magnet and heavy metal such as Pt. Spin current in the magnet propagates along the temperature gradient and reaches the heavy metal[2]. When the randomly polarized electrons in the heavy metal scatter at the interface with magnet, the electron spin experience an exchange interaction form the spin current carrier of the magnet. Such interaction aligns spins in the heavy metal along the spin direction of spin current carrier in the magnet and induces spin accumulation at the heavy metal side of interface. The diffusion of such spin accumulation causes a pure spin current in the heavy metal, in which spin current can be detected as a voltage signal via the inverse spin-Hall effect (ISHE)[3, 4]. The SSE has been observed in various insulating systems, in which parasitic thermal effects originating from itinerant electrons can be eliminated. Some types of spin carriers including magnons in magnets[5, 6, 7, 8], paramagnons[9], antiferromagnetic magnons[7, 8], and spinons in a spin liquid[10, 11] have been demonstrated for different mechanisms of the SSE.

Among quantum spin systems without magnetic order, dimerized magnets spin systems occupy an important position, in which two neighboring spins are frozen as S=0S = 0 singlets in the ground state. The elementary spin excitations are S=1S = 1 triplet states, called a triplon. A triplon has been expected to carry spin angular momentum.

In this work, we report the observation of spin current in CuGeO3_3, carried by triplons in terms of longitudinal SSE measurements. A typical spin-dimer system is a spin-Peierls material CuGeO3_3, which contains one-dimensional spin-1/2 chains with antiferromagnetic exchange interaction for nearest-neighbor spins. In the spin chain, with lowering the temperature, neighboring spins dimerizes to form a spin-gapped phase; the transition is called a spin Peierls (SP) transition[12]. Below the SP transition temperature TSPT_{\mathrm{SP}}, the chain distorts so that the distances between neighboring spins change alternately. The bond-alternating exchange interaction causes neighboring spins to dimerize to reduce the total energy, creating a gap in the spin excitation energy spectrum. CuGeO3_3 is the firstly discovered inorganic system that exhibits a SP transition[13]. High-quality single crystals of CuGeO3_3 are easier to obtain than organic SP materials and CuGeO3_3 has typically been used to study spin excitations in SP systems[14].

Results

Sample structure and the concept of the study. Figure 1a shows the crystal structure of CuGeO3_3.

One dimensional Cu2+^{2+} spin (S=1/2S = 1/2) chains are aligned along the cc-axis, as illustrated as dotted lines. The nearest-neighbor exchange interaction along the cc-axis is Jc∼120J_c \sim120 K, much greater than the interchain exchange coupling[15], Jb∼0.1JcJ_b \sim0.1J_c and Ja∼−0.01JcJ_a \sim-0.01J_c along the bb and aa-axis, respectively. CuGeO3_3 is thus considered as a quasi-one-dimensional spin system. As the temperature TT decreases below the SP transition temperature TSP∼14.5T_{\mathrm{SP}} \sim14.5 K, the lattice of CuGeO3_3 is spontaneously distorted to form a bond alternating configuration[16]. The spin configuration in the ground state is shown schematically in Fig. 1b, where all spins are frozen as dimerized S=0S = 0 singlets. The elementary spin excitation from the ground state is an S=1S = 1 triplon[17], also shown in Fig. 1b. The excitation gap of a triplon is estimated to be ∼23\sim23 K from ESR[18] and neutron scattering[15, 19, 20] experiments.

Based on the previous neutron scattering results[19] and theoretical calculations[21], the dispersion relation of triplons along the cc-axis is sketched in Fig. 1c. In the absence of a magnetic field, μ0H=0\mu_0H = 0, the triplon states are threefold degenerated: ∣↑↑⟩\lvert\uparrow\uparrow\rangle, (∣↑↓⟩+∣↓↑⟩)/2(\lvert\uparrow\downarrow\rangle+ \lvert\downarrow\uparrow\rangle)/\sqrt{2}, and ∣↓↓⟩\lvert\downarrow\downarrow\rangle corresponding to states with different spin quantum numbers, Sz=1S_z = 1, 00, and −1-1, respectively. When a magnetic field is applied, the degenerated triplet bands split into three due to the Zeeman effect, where the triplet state with the parallel spin direction to HH, ∣↑↑⟩\lvert\uparrow\uparrow\rangle, has the lowest energy[18, 19, 20]. The ∣↑↑⟩\lvert\uparrow\uparrow\rangle state thus exhibits the highest probability for thermal excitation, while ∣↓↓⟩\lvert\downarrow\downarrow\rangle the lowest. Driven by a thermal gradient, different occupancy among the three states can result in a flow of a net spin angular momentum in an external magnetic field: triplon SSE. One of the key features of the triplon SSE is that it should exhibit the opposite sign of ISHE signal to the magnon SSE in ferri-and ferro- magnets. This is because triplon spin currents are carried by excitation with the parallel spin direction to the external field; while magnons carry antiparallel spins to the external field, as shown in Fig. 1d. Due to the opposite polarization direction of spin current carrier in these two systems, the electron spin in Pt feels reverse direction of exchange interaction, resulting in spin accumulation with reverse polarization direction and different sign in ISHE. Our theoretical calculation based on the tunnel spin current (see Supplementary Note G3 for details) suggests that the opposite sign of triplon SSE to magnon SSE is a universal property for gapped triplet-spin systems.

Material characterization and measurement setup. First, we measured temperature dependence of magnetization, M(T)M(T), for the CuGeO3\mathrm{CuGeO_3} sample. Figure 1e shows the magnetic susceptibility χ(T)=M(T)/H\chi(T) = M(T)/H measured at μ0H=1\mu_0H = 1 T ∥\parallel bb-axis. When TT is decreased from 20 K, χ\chi exhibits a sudden drop at the SP transition temperature TSP∼14.5T_{\mathrm{SP}} \sim14.5 K and rapidly decreases towards zero. This indicates that Cu spins along the spin chain (∥\parallel cc-axis) form dimers and a spin gap develops below TSPT_{\mathrm{SP}}. By measuring χ(T)\chi(T) at different values of HH (see Supplementary Fig. 1 for details), we obtain the HH-TT spin-phase diagram of the CuGeO3\mathrm{CuGeO_3} sample in Fig. 1f. In the phase diagram, for large HH in the SP phase, CuGeO3\mathrm{CuGeO_3} undergoes another phase transition into a magnetic phase at a transition field, HmH_{\mathrm{m}}, where magnetization is recovered by forming a spin soliton lattice in the chain[22]. The transition can be confirmed as a sudden jump of magnetization as a function of fields[23] (also see Supplementary Fig. 1). The obtained phase diagram is in good agreement with a previous study[24].

The experimental setup for the SSE measurement and the sample used in the present study are shown in Figure 2a and Figure 2b. A trilayer structure comprising a Pt wire for ISHE, an insulating SiO2_2 layer, and an Au wire as a heater was fabricated on the top of the (001) plane of a CuGeO3_3 single crystal. The spin Seebeck voltage signal VSSEV_{\mathrm{SSE}} in the Pt wire is measured by a lock-in method, where an a.c. current is applied to the Au layer under an in-plane magnetic field, HH, applied perpendicular to the Pt wire. We normalize VSSEV_{\mathrm{SSE}} by the heater power and the detector resistance as

Figure 2. SSE measurement results for CuGeO3_3/Pt. a, Experimental setup for the SSE measurement. Temperature gradient (∇T\nabla T) is applied by using an Au heater electrically insulated from the Pt layer by a SiO2_2 film. An a.c. current is applied to the Au heater and SSE voltage in the Pt is measured with a lock-in amplifier as the second harmonic voltage V2fV_{2f}. b, An optical micrograph for a on-chip SSE device made on the top of CuGeO3_3. c and d, The HH dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} (the SSE signal normalized to the heating power and the detector resistance) for the CuGeO3_3/Pt sample at 15 K and 2 K, respectively. e and f, V~SSE(H)\widetilde{V}_{\mathrm{SSE}}(H) measured for Y3_3Fe5_5O12_{12}/Pt at 2 K and 15 K.

V~SSE=VSSEPAu⋅RPt.\widetilde{V}_{\mathrm{SSE}} = \frac{V_{\mathrm{SSE}}}{P_{\mathrm{Au}} \cdot R_{\mathrm{Pt}}}.

Observation of the triplon SSE. The magnetic field dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} at some selected temperatures is shown in Figure 2c and Figure 2d. At 15 K, just above TSP∼14.5T_{\mathrm{SP}} \sim14.5 K, no HH-dependent signal is recognized. Most notably, a clear voltage signal appears when TT is decreased down to 2 K. The signal is an odd function of HH, reflecting the symmetry of the ISHE[25]. The sign of the V~SSE\widetilde{V}_{\mathrm{SSE}} signal is opposite to that of the magnon-mediated SSE[26, 27], which is confirmed in a similar setup for a ferrimagnetic insulator Y3_3Fe5_5O12_{12}/Pt sample in Figure 2e and Figure 2f. The result indicates that spin current carriers in these two systems have different spin polarization directions with respect to HH, consistent with the triplon-current senario. Note that the sign of the SSE signals in spin nematic LiCuVO4_4/Pt (ref. 28) is also opposite to the present CuGeO3_3/Pt case. We also note that, unlike the magnon-mediated SSE in ferro/ferrimagnets, the triplon-mediated SSE dose not scale with the M(H)M(H) curve. For the magnon-mediated SSE[27], M(H)M(H) represents the increase in the spin-current polarization. In contrast, the magnetization increase in the SP phase means larger density of broken dimers and the scattering probability between triplons and broken dimers becomes lager. The unusual HH-dependence of the triplon SSE originates from the different properties between ferro/ferrimagnets and SP phases, and it is one of the characteristics of triplon SSE. (see Supplementary Figure 2 for details).

Comparison between magnetization and SSE signal at 2 K, 6 K and 15 K

Supplementary Figure 2. Comparison between magnetization MM and the SSE signal ∣VSSE∣\lvert V_{\mathrm{SSE}}\rvert. a-c, at 2 K, 6 K and 15 K, respectively.

As shown in Figure 3a and Figure 3b, the temperature dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} shows that the signal magnitude increases monotonically as TT decreases from TSPT_{\mathrm{SP}} to 3 K. When T<3T < 3 K, the SSE voltage remains almost unchanged down to 2 K. At higher temperatures, the dimerized ground state is weakened because the thermally excited phonon disturbs the spontaneous displacement of the Cu2+\mathrm{Cu}^{2+} ion. In other words, the singlet ground state and triplon excitation is no longer a good picture for the low-energy excitation, and the triplon lifetime shrinks toward higher temperatures and completely disappears at TSPT_{\mathrm{SP}}. This effect has been observed as an increase in the peak width of the triplon excitation in neutron scattering experiments[29]. We note that, by phenomenologically incorporating the triplon-phonon scattering term in a calculation, the TT dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} at moderate temperatures can be explained qualitatively (see Supplementary Note G4 for details).

Figure 3. TT dependence of VSSEV_{\mathrm{SSE}}. a and b, HH dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} at selected temperatures for CuGeO3_3/Pt, with ∇T\nabla T applied along the spin chain. c, HH dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} for CuGeO3_3/Pt, with ∇T\nabla T applied perpendicular to the spin chain. d and e, HH dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} at selected temperatures for Cu0.99_{0.99}Zn0.01_{0.01}GeO3_3/Pt and Cu0.97_{0.97}Zn0.03_{0.03}GeO3_3/Pt, respectively. In a–e, the SSE signal is normalized by the heating power and the detector resistance. f and g, TT dependence of magnetic susceptibility for Cu0.99_{0.99}Zn0.01_{0.01}GeO3_3 and Cu0.97_{0.97}Zn0.03_{0.03}GeO3_3, respectively. h, TT dependence of SSE signal in different samples at μ0H=2\mu_0 H = 2 T. Error bars represent standard deviation.

Influence of crystal orientation and Zn-doping on the triplon SSE. Let us turn to the HH dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} in CuGeO3\mathrm{CuGeO}_{3}/Pt. In the entire temperature range, as HH increases from zero, V~SSE\widetilde{V}_{\mathrm{SSE}} increases first and takes its maximum at μ0H∼2\mu_{0}H \sim2 T. As HH further increases, V~SSE\widetilde{V}_{\mathrm{SSE}} is then suppressed. As shown theoretically in the following, triplon scattering by inevitable impurities and defects in the nominally pure CuGeO3\mathrm{CuGeO}_{3} sample can explain the observed HH dependence of V~SSE\widetilde{V}_{\mathrm{SSE}}.

To show that the observed voltage is related to the triplon in the SP phase, we performed another control experiment with ∇T\nabla T applied perpendicular to the spin chain configuration. Figure 3c shows V~SSE(H)\widetilde{V}_{\mathrm{SSE}}(H) for ∇T⊥c\nabla T \perp c in a reference sample prepared with the same procedure as ∇T∥c\nabla T \parallel c.

The observed SSE signal in ∇T⊥c\nabla T \perp c is only 1/7 in magnitude as compared to the ∇T∥c\nabla T \parallel c configuration, consistent with the highly anisotropic transport of triplons. This result corroborates that the origin of the observed signal is the 1-D triplon transport. Moreover, additional reference experiments are performed for two nonmagnetic Zn-doped CuGeO3\mathrm{CuGeO}_{3} samples, in which the triplon spin current should be suppressed by Zn-induced scattering. Figure 3d and Figure 3e show the V~SSE(H)\widetilde{V}_{\mathrm{SSE}}(H) for Cu0.99Zn0.01GeO3\mathrm{Cu}_{0.99}\mathrm{Zn}_{0.01}\mathrm{GeO}_{3}/Pt and Cu0.97Zn0.03GeO3\mathrm{Cu}_{0.97}\mathrm{Zn}_{0.03}\mathrm{GeO}_{3}/Pt in the same temperature range as the CuGeO3\mathrm{CuGeO}_{3}/Pt measurements. Clearly, V~SSE\widetilde{V}_{\mathrm{SSE}} is suppressed in the 1% Zn-doped sample and disappeared in the 3% Zn-doped sample.

Figure 3f and Figure 3g show χ\chi for Cu0.99Zn0.01GeO3\mathrm{Cu}_{0.99}\mathrm{Zn}_{0.01}\mathrm{GeO}_{3} and Cu0.97Zn0.03GeO3\mathrm{Cu}_{0.97}\mathrm{Zn}_{0.03}\mathrm{GeO}_{3}, respectively. For Cu0.99Zn0.01GeO3\mathrm{Cu}_{0.99}\mathrm{Zn}_{0.01}\mathrm{GeO}_{3}, the SP transition is observed at 13.2 K. The decrease of TSPT_{\mathrm{SP}} in doped CuGeO3\mathrm{CuGeO}_{3} agrees with the result from the previous reports[30]. On the other hand, the SP transition is not visible in χ(T)\chi(T) in Cu0.97Zn0.03GeO3\mathrm{Cu}_{0.97}\mathrm{Zn}_{0.03}\mathrm{GeO}_{3}. Instead, a clear cusp is observed in χc(T)\chi_{c}(T) (with H∥cH \parallel c-axis) at TN∼4.3T_{\mathrm{N}} \sim4.3 K, indicating phase transition to an antiferromagnetic phase. The TNT_{\mathrm{N}} is consistent with the previously reported value for Zn-3% doped CuGeO3\mathrm{CuGeO}_{3} (ref. [30]). The doping effect implies that non-magnetic Zn breaks the Cu-spin chains and introduces unpaired Cu free spins, called solitons, into the spin chain. Solitons disturb the dimerization, and local antiferromagnetic correlation may lead to long-range antiferromagnetic order[31] at low temperatures in Cu0.97Zn0.03GeO3\mathrm{Cu}_{0.97}\mathrm{Zn}_{0.03}\mathrm{GeO}_{3}. As shown in Figure 3h, V~SSE\widetilde{V}_{\mathrm{SSE}} is significantly suppressed in the ∇T⊥\nabla T \perp spin-chain configuration and the Zn-doped CuGeO3\mathrm{CuGeO}_{3} samples in all TT range. In the doped samples, the spin-excitation gap is still present in the SP phase, but their transport may be blocked by Zn. The result suggests that the triplon spin current is strongly influenced by Zn-induced scattering. Curie-Weiss fitting of χ(T)\chi(T) at low TT (< 3 K)[13, 32] (see Supplementary Note F for details) reveals that the density of free S=1/2S = 1/2 spins is estimated to be 0.02% in the nominally pure CuGeO3_3 sample, so the effect of triplon scattering by impurities cannot be ignored. Therefore, we attempted to formulate the triplon spin current in the manner of Boltzmann equation, which is capable of incorporating scattering effects.

Theoretical model for the triplon SSE. Figure 4 compares the HH dependence of the observed SSE voltage signal with theoretical results. As we already discussed, ∣V~SSE∣|\widetilde{V}_{\mathrm{SSE}}| is expected to monotonically grow with increasing HH because the triplon density of Sz=1S_z = 1, i.e., the dominant carrier of spin current, increases with lowering the spin gap up to the critical field HmH_{\mathrm{m}}. However, the observed broad peak (see Figure 4a) and the Zn-doped results shown in Figure 3 strongly suggest the importance of triplon scattering processes. To theoretically analyze triplon currents in CuGeO3_3, we here apply the Boltzmann equation[33]. The spin current JsJ_s computed by the Boltzmann equation (See Supplementary Note G2) is given by

Figure 4. Magnetic field dependence of SSE voltage. a, Magnetic field (μ0H\mu_0 H) dependence of the magnitude of ∣V~SSE∣|\widetilde{V}_{\mathrm{SSE}}| at 2.5 K. b, Calculation results of the μ0H\mu_0 H dependence of the triplon spin current (see text and Supplementary Note G2 for details).

Js=∑Sz∫h2πℏSz(vSz(k))2τkkBΔTεSz(k)(kBT)2exp⁡(εSz(k)/kBT)(exp⁡[εSz(k)/kBT]−1)2,(1)J_s = \sum_{S_z} \int\frac{h}{2\pi} \hbar S_z \left(v_{S_z}(k)\right)^2 \tau_k k_{\mathrm{B}}\Delta T \frac{\varepsilon_{S_z}(k)}{(k_{\mathrm{B}}T)^2} \frac{\exp\left(\varepsilon_{S_z}(k)/k_{\mathrm{B}}T\right)}{\left(\exp\left[\varepsilon_{S_z}(k)/k_{\mathrm{B}}T\right]-1\right)^2}, \tag*{(1)}

where ΔT\Delta T is the temperature difference along the spin chain direction. vSz(k)v_{S_z}(k), τk\tau_k and εSz(k)\varepsilon_{S_z}(k) are the group velocity, relaxation time, and energy dispersion for the triplons[17], respectively. We assume that the triplon can be viewed as a bosonic particle and the final part of Equation (1) stems from the Bose distribution function. The Sz=0S_z = 0 branch of the triplon does not contribute to the spin current. At H=0H = 0, the distribution functions of Sz=±1S_z = \pm1 branches are identical, which cancel out the total spin current. By varying TT and HH, the thermal excitation and the Zeeman effect alter the triplon distributions. Assuming a constant τk\tau_k, the population difference between Sz=±1S_z = \pm1 increases monotonically at a fixed T(<TSP)T(< T_{\mathrm{SP}}) due to the Zeeman effect, resulting in an almost HH-linear increase in JsJ_s. The HH-linear behavior can also be reproduced by using a tunnel spin-current formula,[10, 28, 34, 35] which is another microscopic theory for thermal spin current (see Supplementary Note G3).

As shown in Figure 4a, ∣V~SSE(H)∣|\widetilde{V}_{\mathrm{SSE}}(H)| takes a maximum at around 3 T. To explain the non-monotonic behavior of V~SSE\widetilde{V}_{\mathrm{SSE}}, we considered two mechanisms of triplon scattering processes. One is the scattering by nonmagnetic impurities, where the scattering probability 1/τk,non−mag1/\tau_{k,\mathrm{non-mag}} is constant with respect to HH. Another source of scattering is free spins of broken dimers and unpaired Cu2+\mathrm{Cu}^{2+} (S=1/2S = 1/2). The density of scatterers and scattering potential are assumed to depend on HH as a Brillouin function. The total scattering probability is then the sum of the two scattering mechanisms, 1/τk,total=1/τk,non−mag+1/τk,mag1/\tau_{k,\mathrm{total}} = 1/\tau_{k,\mathrm{non-mag}} + 1/\tau_{k,\mathrm{mag}}. Assuming that the scattering effect by magnetic impurities is dominant, we can reproduce the experimentally obtained HH dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} shown in Figure 4b. In the low HH regime, where the spin of magnetic impurities is not aligned and the scattering effect is weak, the Zeeman effect dominates the generation and transport of the triplon current. With increasing HH, the triplon mode of Sz=+1S_z = +1 (Sz=−1S_z = -1) shifts downward (upward) due to the Zeeman effect, and the imbalance between Sz=±1S_z = \pm1 triplons becomes larger, resulting in an increase in the total triplon current. As HH is further increased, the impurities are magnetized and magnetic scattering dominates the triplon transport, resulting in a broad peak as shown in Figure 4b. The result agrees with the experimentally obtained data shown in Figure 4a in a semi-quantitative level.

Discussion

Finally, we here examine other possible origins for the observed thermoelectric voltage. One candidate could be the paramagnetic SSE due to free Cu spins. Several papers have addressed the spin current transport in paramagnets, including Gd3Ga5O12\mathrm{Gd}_{3}\mathrm{Ga}_{5}\mathrm{O}_{12} (refs. [9] [36]), DyScO3\mathrm{DyScO}_{3} (ref. [9]) and La2NiMnO6\mathrm{La}_{2}\mathrm{NiMnO}_{6} (ref. 37). Paramagnon is believed to be responsible for spin transport in paramagnets as a result of short-range magnetic correlation or long-range dipole interactions. The spin diffusion length of Gd3Ga5O12\mathrm{Gd}_{3}\mathrm{Ga}_{5}\mathrm{O}_{12} is estimated to be about 1.8 μm1.8\ \mu\mathrm{m} at 5 K5\ \mathrm{K}36. However, in our CuGeO3\mathrm{CuGeO}_{3} sample with the free spin density of 0.02%0.02\%, the average distance between free spins is estimated to be around 0.29 nm/0.02%∼1.5 μm0.29\ \mathrm{nm}/0.02\% \sim1.5\ \mu\mathrm{m} (the distance between neighboring Cu2+\mathrm{Cu}^{2+} ions along the cc-axis is 0.29 nm0.29\ \mathrm{nm}38). This means that the paramagnetic spins are too dilute for spin correlation and spin current transport. Furthermore, the suppression of VSSEV_{\mathrm{SSE}} in the Zn-doped samples, which exhibit larger paramagnetic moments than CuGeO3\mathrm{CuGeO}_{3}, also rules out the possibility of paramagnetic SSE.

The normal and anomalous Nernst effects may also contribute to an electric voltage under a temperature gradient, but these effects contradict with the suppression of the signal in the high temperature spin-liquid phase; the vanishing signals in the ∇T⊥c\nabla T \perp c configuration and the Zn-doped samples (see Figure 3a-Figure 3e). Furthermore, the normal Nernst effect should be linear with respect to the applied magnetic field and can not explain the observed magnetic field dependence of VSSEV_{\mathrm{SSE}} shown in Figure 3a and Figure 3b.

Note that the voltage observed at 15 K15\ \mathrm{K} in the CuGeO3\mathrm{CuGeO}_{3}/Pt shows almost no HH-dependence, as shown in Figure 2c. Above TSPT_{\mathrm{SP}}, the elementary excitation of spin chains are gapless spinons[21]. A spinon spin current, a spin current carried by spinons, was found in Sr2CuO3\mathrm{Sr}_{2}\mathrm{CuO}_{3} (ref. 10). The spinon spin current may contribute to a small HH-linear voltage signal above TSPT_{\mathrm{SP}} (see Supplementary Note C for details). In the spin-Peierls phase, the gapless spinon is replaced by the gapped triplon[21].

In summary, we observed the triplon spin-Seebeck effect in CuGeO3\mathrm{CuGeO}_{3}/Pt. Due to the triplon excitation in the system, the sign of the observed SSE is opposite to that of the conventional magnon SSE. The triplon SSE signal persists down to 2 K in the spin-Peierls phase and the magnetic field dependence of the triplon SSE is consistent with microscopic calculation. Our result shows that the spin-Seebeck effect also acts as a probe for spin excitations in gapped spin systems, and can also be applied to other materials with exotic spin excitation, such as a spin ladder system SrCu2O3\mathrm{SrCu}_{2}\mathrm{O}_{3} (ref. 39) and a spin dimer system SrCu2(BO3)2\mathrm{SrCu}_{2}(\mathrm{BO}_{3})_{2} (ref. 40).

Methods

Sample fabrication. CuGeO3\mathrm{CuGeO}_{3} single crystals were grown by a traveling solvent floating zone (TSFZ) method. Samples were cut into size of ∼7\sim7 mm ×3\times3 mm for further measurements. 40nm-thick ferrimagnetic insulator Y3Fe5O12\mathrm{Y}_{3}\mathrm{Fe}_{5}\mathrm{O}_{12} single crystals were grown on Gd3Ga5O12\mathrm{Gd}_{3}\mathrm{Ga}_{5}\mathrm{O}_{12}(111) substrates by magnetron sputtering. Prior to the deposition, Gd3Ga5O12\mathrm{Gd}_{3}\mathrm{Ga}_{5}\mathrm{O}_{12} substrates were annealed in the air at 825 ∘C825\,^{\circ}\mathrm{C} for 30 min in a face-to-face configuration. To crystalize the as-grown amorphous Y3Fe5O12\mathrm{Y}_{3}\mathrm{Fe}_{5}\mathrm{O}_{12}, the samples were post-annealed in air at 825 ∘C825\,^{\circ}\mathrm{C} for 200 s. For SSE measurements, on-chip devices with Pt detector and SiO2\mathrm{SiO}_{2}/Au heater were fabricated on top of samples (both pure/doped CuGeO3\mathrm{CuGeO}_{3} and Y3Fe5O12\mathrm{Y}_{3}\mathrm{Fe}_{5}\mathrm{O}_{12}) by an electron-beam lithography, magnetron sputtering and lift-off technique (see Supplementary Note A).

SSE measurement. We adapted the on-chip heating method for SSE, as illustrated in Fig. 1b. The resistance between the Au and Pt layers is much greater than MΩM\Omega at room temperature. A longitudinal temperature gradient was generated by applying an a.c. current with the frequency of f=13f = 13 Hz to the Au heater. Since the SSE is driven by Joule heating, the frequency of the SSE signal on the Pt wire is 2f2f. The SSE signal was detected via a lock-in method.

Magnetization measurement. The magnetization of CuGeO3\mathrm{CuGeO}_{3} was measured in a Quantum Design Physical Properties Measurement System (PPMS) with a vibrating sample magnetometer (VSM) option.

Acknowledgements This is a post-peer-review, pre-copyedit version of an article published in Nature Communications. The final authenticated version is available online at: https://doi.org/10.1038/s41467-021-25494-7. We thank D. Hirobe, T. Kikkawa, G. E. W. Bauer and Y. Hirayama for valuable discussions. This research was supported by JST ERATO Spin Quantum Rectification Project (No. JPMJER1402), JPSP KAKENHI ( the Grant-in-Aid for Scientific Research (S) No. JP19H05600 and JP17H06137; the Grant-in-Aid for Scientific Research (A) No. JP16H02125; the Grant-in-Aid for Scientific Research (B) No. JP20H01830 and No. JP19H02424; the Grant-in-Aid for Scientific Research (C) No. JP17K05513; the Grant-in-Aid for Challenging Research (Exploratory) No. JP19K22124 and No. JP20K20896; the Grant-in-Aid for Scientific Research on Innovative Areas (Research in a proposed research area) No. JP20H04631, No. JP20H05153, No. JP19H05825 and No. JP19H04683; the Grant-in-Aid for Research Activity startup No. JP20K22476), JST CREST (No. JPMJCR20C1 and No. JPMJCR20T2) and MEXT (Innovative Area “Nano Spin Conversion Science” (No. 26103005)). Y.C., Y.T. and K.O. are supported by GP-Spin at Tohoku University. Y.C. is also supported by the Japan Society for Promotion of Science through a research fellowship for young scientists (No. JP18J21304).

Author contributions Y.C. designed the experiments in discuss with Y.S. and E.S. Y.T., Y.N., T.M. and M.F. grew single crystals for this study. Y.C. and K.O. nano-fabricated devices. Y.C. collected and analyzed the experimental data. M.S. preformed the theoretical calculations. E.S. supervised this work. Y.C., M.S., Y.S. and E.S. wrote the manuscript. All authors discussed the results and commented on the manuscript.

Competing Interests The authors declare that they have no competing interests.

Data availability The data that support the findings of this study are available from the corresponding authors upon reasonable request.

Code availability The codes used in theoretical simulations and calculations are available from the corresponding authors upon reasonable request.

Correspondence Correspondence and requests for materials should be addressed to Y.C (e-mail: [email protected]) and M.S (e-mail:[email protected]).

Image not extracted

Figure 1.

Image not extracted

Figure 1 (PDF p. 16). Spin chain and spin-Peierls (SP) phase of CuGeO3_3. a, Crystalline structure of CuGeO3_3. The Cu spin chains are shown as dotted lines. Spin-1/2 spin chains along the cc-axis in CuGeO3_3 are formed by staking CuO2_2 chains. JcJ_c and JbJ_b denote the exchange interactions along the cc and bb-axis, respectively. b, When temperature (TT) falls down to the SP transition temperature TSPT_{\mathrm{SP}}, CuGeO3_3 undergoes a SP transition. After the transition, the Cu spins dimerize and the ground state becomes a spin singlet state ((∣↑↓⟩−∣↓↑⟩)/2(\left|\uparrow\downarrow\right\rangle-\left|\downarrow\uparrow\right\rangle)/\sqrt{2} with S=0S=0). Elementary excitation in the SP phase is a triplon with S=1S=1. c, Schematic band structure of triplon in CuGeO3_3[19, 21]. Three triplet states (∣↓↓⟩\left|\downarrow\downarrow\right\rangle, (∣↑↓⟩+∣↓↑⟩)/2(\left|\uparrow\downarrow\right\rangle+\left|\downarrow\uparrow\right\rangle)/\sqrt{2} and ∣↑↑⟩\left|\uparrow\uparrow\right\rangle) are threefold degenerated at zero field. By applying a magnetic field (HH), the Zeeman effect lift the degeneracy. d, The ground state and excitation state of a ferromagnet. Elementary excitation is a magnon with S=1S=1. The magnons reduce the magnetization (M\mathbf{M}) along the magnetic field (H\mathbf{H}) as δSz\delta S^z. e, TT dependence of magnetic susceptibility (χ\chi) of CuGeO3_3 under an applied magnetic field of 1 T along the bb-axis. The SP transition temperature TSPT_{\mathrm{SP}} is illustrated as the arrow in the figure. f, Magnetic field - temperature (HH - TT) spin phase diagram for the CuGeO3_3 sample obtained from the M(H)M(H) and χ(T)\chi(T) measurement results. The diamond symbols represent data taken from a previous study$[24]. The dotted curves are guide for the eyes.

Supplementary information for

Experimental details of spin-Seebeck effect measurements

Single crystalline CuGeO3\mathrm{CuGeO}_{3} prepared by a floating zone method[1] was an elliptical cylinder with the height of ∼3\sim3 cm (along the aa-axis), the long axis of 7 mm (along the cc-axis) and a short axis of 3 mm (along the bb-axis). Crystal orientations are determined by using a Laue camara. Firstly, a 340 μm340\,\mu\mathrm{m} long, 1.5 μm1.5\,\mu\mathrm{m} wide and 5 nm thick Pt wire was fabricated on CuGeO3\mathrm{CuGeO}_{3} by an e-beam lithography and lift-off process. Subsequently, a SiO2\mathrm{SiO}_{2}/Au wire with the length of 360 μm360\,\mu\mathrm{m} and the width of 7.5 μm7.5\,\mu\mathrm{m} was fabricated on the top of the Pt wire. The thickness of the SiO2\mathrm{SiO}_{2} and Au are 8 nm and 80 nm, respectively. Au and Pt layers are insulated by a SiO2\mathrm{SiO}_{2} layer.

In the SSE measurement, a sinusoidal current (f=13f = 13 Hz) was applied to the Au layer with a current source (Keithley 6221, Tektronix, Inc.). The voltage signal showing up along the Pt wire was recorded by a lock-in amplifier (NF 5640, NF Corporation). All SSE measurements were performed in a PPMS (Physical Property Measurement System, Quantum Design, Inc.).

Estimation of TSPT_{\mathrm{SP}} and HmH_{\mathrm{m}} of the CuGeO3_3 sample

Image not extracted

Figure 1 (PDF p. 25). Temperature (TT) and magnetic field (HH) dependence of the magnetization (MM) of CuGeO3_3. a, TT dependence of magnetic susceptibility (χ=M/H\chi= M/H) at different HH values. b, HH dependence of MM at different TT values. Data are obtained with HH applied along the bb-axis.

Magnetization (MM) measurements were carried out using the VSM (vibrating sample magnetometer) option of PPMS. The χ(T)=M(T)/H\chi(T) = M(T)/H results obtained under various HH values are shown in Supplementary Fig. 1a. Based on the results, we determined the spin-Peierls (SP) transition temperature TSPT_{\mathrm{SP}} at each HH. The transition field (HmH_{\mathrm{m}}) from the SP phase to the magnetic phase at each TT was determined by measuring M(H)M(H). A steep increase in M(H)M(H) was detected at μ0H∼13\mu_{0}H \sim13 T for T<TSPT < T_{\mathrm{SP}}, as shown in Supplementary Fig. 1b. HmH_{\mathrm{m}} is defined as the field where dM(H)/dHdM(H)/dH is maximized.

Comparison between magnetization and SSE signal

A comparison between M(H)M(H) and SSE at different temperatures are shown in Supplementary Figure 2. Unlike the case of the magnon SSE in most ferro/ferri-magnets, VSSE(H)V_{\mathrm{SSE}}(H) in CuGeO3_3 does not scale with M(H)M(H). The magnetization of CuGeO3_3 at T<TSPT < T_{\mathrm{SP}} mainly consists of impurity spins and unpaired spins on thermally broken dimers. As shown in Supplementary Figure 2 and Supplementary Figure 2, the SSE signal does not scale with MM-HH at T=2T = 2 K and 6 K. The MM-HH in the low field range reflects the paramagnetic term from free (impurity) spins, and these free spins do not play a role in VSSE(H)V_{\mathrm{SSE}}(H) because they are well decoupled. At T=6T = 6 K, the magnetic component from broken dimers overwhelms the paramagnetic component from impurity spins. Since the broken dimers are also little important for spin-current carriers in SSE, VSSEV_{\mathrm{SSE}} and MM should show totally different magnetic field dependence.

For T=15T = 15 K >TSP> T_{\mathrm{SP}}, the system is theoretically predicted to be a type of paramagnets: a Tomonaga-Luttinger (TL) spin liquid, where the equilibrium state is still a zero magnetizated state in the absence of fields and the spin excitation is a gapless spinon instead of triplon. For higher temperatures (kBT>Jk_{\mathrm{B}}T > J), the system experience a crossover from the TL spin liquid to a paramagnetic state. We also expect a spinon SSE2SSE^{2} in the intermediate temperature range (TSP<T≪J/kBT_{\mathrm{SP}} < T \ll J/k_{\mathrm{B}}). From the paper by Hirobe, et al (ref. 2), the spinon SSE signal in TL spin liquids was found to be much smaller than the ferromagnetic SSE, linear with respect to the magnetic field. The SSE signal measured at 15 K >TSP> T_{\mathrm{SP}} satisfies all these properties, and thus may be an indication of the spinon SSE.

Also, the anomaly in the SSE is hardly observed at HmH_{\mathrm{m}}. This is because the net spin current above HmH_{\mathrm{m}} is still determined by low-energy S=+1S = +1 triplons and high-energy S=−1S = -1 triplons, similar to the case of H<HmH < H_{\mathrm{m}} (see Supplementary Figure S3 which illustrates the triplon bands above HmH_{\mathrm{m}}). As shown in Supplementary Figure S3, for H>HmH > H_{\mathrm{m}}, the excitation energy of the S=+1S = +1 triplon is lower than the energy of the ground state, and the S=+1S = +1 triplon condenses. This is called the triplon BEC [3]. The condensed triplon shows static magnetization but does not carry spin current.

A schematic illustration of triplon BEC

Figure S3. A schematic illustration of triplon BEC. a, triplon band structure with magnetic field (HH) less than the transition field (HmH_{\mathrm{m}}) from the SP phase to the magnetic phase. b, triplon band structure with H>HmH > H_{\mathrm{m}}. Green, black and red curves represent three triplet states ∣↓↓⟩\lvert\downarrow\downarrow\rangle, (∣↑↓⟩+∣↓↑⟩)/2(\lvert\uparrow\downarrow\rangle+ \lvert\downarrow\uparrow\rangle)/\sqrt{2}, and ∣↑↑⟩\lvert\uparrow\uparrow\rangle respectively.

The comparison between SSEs in an antiferromagnetic insulator and the SP magnet CuGeO3CuGeO_{3} would be useful to deeply understand the feature of the triplon SSE. In the case of antiferromagnet/Pt system, the SSE signal changes abruptly near the spin-flop transition[4, 5]. At the spin-flop transition, both the ground state and the magnon-band structure drastically change and a net magnetization suddenly appears because the spin-flop transition is of a first-order type. Due to this significant change of the magnon band, the SSE signal largely changes at the spin-flop transition.

On the other hand, although the triplon BEC occurs for H>HmH > H_{\mathrm{m}} of CuGeO3CuGeO_{3}, the band structure of the triplon excitation is very similar to that for H<HmH < H_{\mathrm{m}}. Therefore, there was no significant change in the triplon SSE signal around HmH_{\mathrm{m}}.

Angular and heater-power dependence of the spin-Seebeck signal

Angular and heating power dependence of the SSE signal

Supplementary Figure 4. Angular and heating power dependence of the SSE signal. a, Angular dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} at μ0H=3\mu_0 H = 3 T and T=2T = 2 K, with a heating current (II) of 2 mA. The red curve indicates a sinusoidal fit. b, Heater-power dependence of V~SSE\widetilde{V}_{\mathrm{SSE}} at μ0H=3\mu_0 H = 3 T and T=2T = 2 K. The red line is a linear fit to the data for the heater currents ranging from 0 to 2 mA. Error bars represent 95% confidence interval of the sinusoidal fitted amplitude.

Supplementary Fig. 4a shows V~SSE(β)\widetilde{V}_{\mathrm{SSE}}(\beta) in CuGeO3_3/Pt at T=2T = 2 K and μ0H=3\mu_0 H = 3 T. Here, HH is always perpendicular to the Pt wire and the out-of-plane magnetic field angle β\beta is defined as shown in Supplementary Fig. 4a. V~SSE(β)\widetilde{V}_{\mathrm{SSE}}(\beta) is well fitted by a sin⁡(β)\sin(\beta) function, which is consistent with the angular dependence of the inverse spin-Hall effect (ISHE) voltage: VISHE∝[Js×σ]a∝sin⁡(β)V_{\mathrm{ISHE}} \propto[\mathbf{J}_{\mathrm{s}} \times\boldsymbol{\sigma}]_{a} \propto\sin(\beta). Here, Js\mathbf{J}_{\mathrm{s}} and σ\boldsymbol{\sigma} denote the spatial direction of the spin current and the spin-polarization vector of the spin current, respectively[6]. [Js×σ]a[\mathbf{J}_{\mathrm{s}} \times\boldsymbol{\sigma}]_{a} is the aa-axis component of Js×σ\mathbf{J}_{\mathrm{s}} \times\boldsymbol{\sigma}.

The heater-power dependence of VSSEV_{\mathrm{SSE}} is shown in Supplementary Fig. 4b. The amplitude of VSSEV_{\mathrm{SSE}} at each heater-power is estimated by fitting VSSE(β)V_{\mathrm{SSE}}(\beta) using a sin⁡(β)\sin(\beta) curve at μ0H=3\mu_{0}H = 3 T.

When the current applied to the Au heater is less than 2 mA, the heater-power dependence of VSSEV_{\mathrm{SSE}} is well fitted by a linear function. For higher heater power, however, VSSEV_{\mathrm{SSE}} deviates from the linear dependence. This non-linear dependence was also observed in the magnon spin-Seebeck effect in Y3_3Fe5_5O12_{12}/Pt systems[7]; the non-linearity is attributed to the strong non-equilibrium excitation of magnons in Y3_3Fe5_5O12_{12}. Another possibility for the non-linearity in the CuGeO3_3 case is temperature rise of the CuGeO3_3 sample due to the Joule heating, since VSSEV_{\mathrm{SSE}} should decrease with increasing TT. Therefore, to avoid the undesirable non-equilibrium/heating effects, we fixed the heater current at 1.5 mA for all the measurements shown in the main text.

Temperature and field dependence of the Pt and Au resistivity

VSSEV_{\mathrm{SSE}} is proportional to the heating power of the Au heater. In the case of a constant current (IAuI_{\mathrm{Au}}), VSSEV_{\mathrm{SSE}} is proportional to the resistance of Au (RAuR_{\mathrm{Au}}). VSSEV_{\mathrm{SSE}} is also proportional to the resistance of Pt (RPtR_{\mathrm{Pt}}) for a constant spin current JsJ_s injection: VSSE∝Js×RPtV_{\mathrm{SSE}} \propto J_s \times R_{\mathrm{Pt}}. Thus, a change in RPtR_{\mathrm{Pt}} and RAuR_{\mathrm{Au}} with respect to TT and HH will cause extrinsic variations in VSSEV_{\mathrm{SSE}}. Supplementary Figs. 5a and b show the TT dependence of RR of the Pt wire and Au heater, respectively. As TT is reduced, both RPtR_{\mathrm{Pt}} and RAuR_{\mathrm{Au}} decrease monotonically. As shown in the insets of Supplementary Figs. 5a and b, the variation of RPtR_{\mathrm{Pt}} and RAuR_{\mathrm{Au}} at low TT (T<20T < 20 K) is less than 1%. The HH-dependence of RPtR_{\mathrm{Pt}} and RAuR_{\mathrm{Au}} at T=2T = 2 K is shown in Supplementary Figs. 5c and S5d. RPtR_{\mathrm{Pt}} and RAuR_{\mathrm{Au}} change less than 1% up to 6 T in both metals. According to the R(T)R(T) and R(H)R(H) results, the contribution of the resistance change to the observed SSE signal is negligibly small.

Image not extracted

Figure 5. Temperature (TT) and magnetic field (HH) dependence of resistance (RR) of detector and heater. a-b, TT dependence of RR of Pt and Au. c-d, The magnetoresistance of Pt and Au at 2 K. The resistivity change for both metals up to 6 T is below 1%.

As for the actual temperature at the interface, we can use the Pt detector layer as a thermometer (i.e. Pt resistance thermometer) to investigate the temperature change with and without the Au heating current (e.g. ref.[8]). When the heater current is applied to the Au heater, the resistance of the Pt detector changes, as shown in Supplementary Figure 6. Using dR/dTdR/dT and

Temperature difference at the interface

Supplementary Figure 6. Temperature difference at the interface. a, The temperature (TT) dependence of the resistance (RR) of Pt with and without the heating current (IAuI_{\mathrm{Au}}). b-d, The temperature dependence of dR/dT\mathrm{d}R/\mathrm{d}T, ΔR\Delta R and ΔT\Delta T. The dotted curve is a guide for the eyes. Error bars represent standard deviation.

ΔR≡R(IAu=1.5 mA)−R(IAu=0),\Delta R \equiv R(I_{\mathrm{Au}} = 1.5\,\mathrm{mA}) - R(I_{\mathrm{Au}} = 0),

we estimated the typical temperature difference across the device ΔT=ΔR/(dR/dT)\Delta T = \Delta R/(dR/dT) to be 0.2-0.5 K, as shown in Supplementary Figure 6. For T>5T > 5 K, ΔT\Delta T is as small as 0.25 K and 0.5 K even at T=2T = 2 K. In the constant power setup, a lower value of thermal conductivity κ\kappa leads to a larger temperature difference and causes a larger observed SSE signal. In the case of CuGeO3_3, because of the suppression of phonon transport, κ\kappa monotonically decreases for T<5T < 5 K (ref. 9), which is consistent with our results.

Impurity density in Cu0.99_{0.99}Zn0.01_{0.01}GeO3_3 and nominally pure CuGeO3_3

The density of impurities in CuGeO3_3 can roughly be estimated from the χ(T)\chi(T) curve. In the SP phase, the magnetic susceptibility can be decomposed into three terms [10, 11]:

χ(T)=χ0+χPara(T)+χSP(T)(2)\chi(T) = \chi_0 + \chi_{\mathrm{Para}}(T) + \chi_{\mathrm{SP}}(T) \tag*{(2)}

where χ0\chi_0, χPara\chi_{\mathrm{Para}}, and χSP\chi_{\mathrm{SP}} stand for constant diamagnetic contribution, paramagnetic contribution from impurity-induced free spins (unpaired Cu spins) obeying χPara=C/(T−Θ)\chi_{\mathrm{Para}} = C/(T - \Theta), and spin-Peierls contribution, respectively. For Cu0.99_{0.99}Zn0.01_{0.01}GeO3_3 sample, C=5.1×10−7C = 5.1 \times10^{-7} μB\mu_{\mathrm{B}}Cu ⋅\cdot Oe−1^{-1} were obtained by fitting the χ(T)\chi(T) data by using Supplementary Eq. (1) at a low TT range (<3< 3 K), assuming that the spin-Peierls contribution is zero (χSP=0\chi_{\mathrm{SP}} = 0)1. The fitted result of χPara\chi_{\mathrm{Para}} is shown as a function of TT in Supplementary Fig. 7a. Assuming that all S=1/2S = 1/2 Cu2+\mathrm{Cu}^{2+} ions contribute to the paramagnetic term, Call=S(S+1)g2μB23kB=7.06×10−5μBCu⋅Oe−1C_{\mathrm{all}} = \frac{S(S+1)g^2\mu_{\mathrm{B}}^2}{3k_{\mathrm{B}}} = 7.06 \times10^{-5}\mu_{\mathrm{B}}\mathrm{Cu}\cdot\mathrm{Oe}^{-1} with g=2.1g = 2.1 for CuGeO3\mathrm{CuGeO}_{3} (ref. 12). By comparing CallC_{\mathrm{all}} with the fitted CC result, we conclude that the density of free spins is about C/Call∼0.73%C/C_{\mathrm{all}} \sim0.73\% of all Cu atoms, which is very close to the 1% Zn-doping. Preforming the same analysis on the CuGeO3\mathrm{CuGeO}_{3} sample (see Supplementary Fig. 7b), we obtain an impurity density level (C/CallC/C_{\mathrm{all}}) of 0.02% in the CuGeO3\mathrm{CuGeO}_{3} sample. Therefore, even in nominally pure samples, the effect of triplon scattering by impurities cannot be ignored.

Magnetic susceptibility at low temperature for Cu0.99Zn0.01GeO3 and CuGeO3

Supplementary Figure 7. Magnetic susceptibility (χ\chi) at the low temperature (TT) range. a and b, temperature (TT) dependence of magnetic susceptibility (χ\chi) for Cu0.99Zn0.01GeO3\mathrm{Cu}_{0.99}\mathrm{Zn}_{0.01}\mathrm{GeO}_{3} and CuGeO3\mathrm{CuGeO}_{3} at a low TT range. The impurity induced paramagnetic term χPara\chi_{\mathrm{Para}} for each sample is estimated from the fitting to the Curie-Weiss law (black curves).

Theoretical analysis of SSE in CuGeO3_3

In this section, we discuss our theoretical analysis for SSE in CuGeO3_3. First, we review a quantum sine-Gordon (SG) model as low-energy effective theory for CuGeO3_3. The SG model describes triplon excitation in the spin-Peierls phase. Then, based on the results of the SG model, we compute a SSE spin current in CuGeO3_3. To this end, we apply two microscopic methods: the Boltzmann equation and the tunnel spin-current theory based on Keldysh Green’s function.

Quantum sine-Gordon field theory as the low-energy effective model for CuGeO3_3

Here, we explain low-energy effective theory for CuGeO3_3. The model Hamiltonian for each antiferromagnetic spin chain in CuGeO3_3 is

HSP=∑jJ1(1+δ(−1)j)Sj⋅Sj+1+J2Sj⋅Sj+2−B∑jSjz(3)H_{\mathrm{SP}} = \sum_j J_1 \left(1 + \delta(-1)^j\right) \mathbf{S}_j \cdot\mathbf{S}_{j+1} + J_2 \mathbf{S}_j \cdot\mathbf{S}_{j+2} - B \sum_j S_j^z \tag*{(3)}

where Sj\mathbf{S}_j is the electron spin on the jj-th site, and B≡gμBμ0HB \equiv g\mu_{\mathrm{B}}\mu_0 H with the gg factor g=2.1g = 2.1 (ref. 12), Bohr magneton μB\mu_{\mathrm{B}} and an external magnetic flux density μ0H\mu_0 H. For CuGeO3_3, the exchange coupling constants of nearest neighboring spins and next nearest neighboring spins are respectively estimated to be J1≃120.6J_1 \simeq120.6 (ref. K13) and J2≃0.36J1J_2 \simeq0.36J_1 (ref. 14). A finite dimerization parameter, δ\delta, emerges in the spin-Peierls phase due to the lattice deformation. First, we focus on the zero-field case. In the small-J2J_2 case of α≡J2/J1<0.241≡αc\alpha\equiv J_2/J_1 < 0.241 \equiv\alpha_{\mathrm{c}} (ref. 15), the spin chain belongs to the universality class of a Tomonaga-Luttinger (TL) liquid, i.e., the ground state is a non-magnetic spin liquid and the low-energy excitations are described by gapless spinons. For α>αc\alpha> \alpha_{\mathrm{c}}, the ground state is spontaneously dimerized and an excitation gap opens. A dimerization induces a lattice distortion through the spin-phonon coupling, and, as a result, δ\delta becomes finite. In the dimerized (i.e., spin-Peierls) phase with a finite δ\delta, the lowest excitation branch is given by the triply degenerated S=1S = 1 mode, namely, triplons. The low-energy effective Hamiltonian [16, 17, 18] is given by

Heff=∫dx v2π[1K(∂xϕ)2+K(∂xθ)2]+J1δda0sin⁡(4πϕ)+λa0cos⁡(16πϕ).(4)H_{\mathrm{eff}} = \int dx\, \frac{v}{2\pi} \left[\frac{1}{K}(\partial_x\phi)^2 + K(\partial_x\theta)^2\right] + \frac{J_1\delta d}{a_0}\sin(\sqrt{4\pi}\phi) + \frac{\lambda}{a_0}\cos(\sqrt{16\pi}\phi). \tag*{(4)}

Where ϕ\phi and θ\theta are a pair of dual boson fields. vv and KK are, respectively, the spinon velocity and the TL-liquid parameter, and a0a_0 is the lattice constant. The spin-rotation SU(2) symmetry makes the TL-liquid parameter fixed to be K=1/2K = 1/2. The quadratic-boson (i.e., free-boson) part corresponds to the TL-liquid Hamiltonian. The sine term stems from the bond alternation δ\delta and the non-universal constant d(>0)d(> 0) can be numerically determined [17, 19] while the cosine term cos⁡(16πϕ)\cos(\sqrt{16\pi}\phi) exists even in the uniform spin chain without δ\delta. The sine term with scaling dimension KK is more relevant than the cosine term with dimension 4K4K. If α\alpha is close enough to αc\alpha_c, the coupling constant λ∝α−αc\lambda\propto\alpha- \alpha_c. In the uniform case of δ=0\delta= 0, the λ\lambda term is marginally irrelevant (relevant) for λ<0\lambda< 0 (λ>0\lambda> 0), which corresponds to a gapless TL liquid (a spontaneously dimerized ground state with a spin gap). For a SP phase with δ≠0\delta\ne0, a small λ\lambda term can be negligible in the sense of renormalization group, and the low-energy Hamiltonian is given by a sum of the TL-liquid part and a single cosine term cos⁡(4πϕ)\cos(\sqrt{4\pi}\phi). This is called quantum sine-Gordon (SG) model and is an integrable quantum field theory [16, 18, 20, 21]. Since α\alpha is close to αc\alpha_c in CuGeO3_3, we can expect that its low-energy physics is described by a SG model. The SG model is known to well describe the low-energy physics of the spin-Peierls Hamiltonian with high accuracy [16, 17, 18, 19], and the model enables us to compute various physical quantities in a quantitative level [22, 23].

The sine term is relevant in the SU(2) case we focus on, and therefore the field ϕ\phi is locked at the energy minimum of the sine potential in the ground state, i.e., ⟨ϕ⟩=π(n−14)\langle\phi\rangle= \sqrt{\pi}(n-\frac{1}{4}) for δ>0\delta> 0 and ⟨ϕ⟩=π(n+14)\langle\phi\rangle= \sqrt{\pi}(n+\frac{1}{4}) for δ<0(\delta< 0 (n$: integer). The elementary excitations in the SG model are soliton, antisoliton, and breathers (bound state of soliton and antisoliton) [17, 20, 21]. The soliton and antisoliton are the domain-wall excitations connecting two neighboring ground states: For instance, a soliton texture located around x=x0x=x_0 is given by ⟨ϕ⟩=π(n−14)\langle\phi\rangle= \sqrt{\pi}(n-\frac{1}{4}) for x≪x0x \ll x_0 and ⟨ϕ⟩=π(n+1−14)\langle\phi\rangle= \sqrt{\pi}(n+1-\frac{1}{4}) for x≫x0x \gg x_0. The mass gap of these excitations depends on the parameters vv, KK, and J1δd/a0J_1\delta d/a_0 if we neglect the marginal cosine term cos⁡(16πϕ)\cos(\sqrt{16\pi}\phi). In the SU(2) case, soliton, antisoliton, and first breather have the same mass m1∼δ2/3m_1 \sim\delta^{2/3}. Second breather’s mass m2m_2 is given by m2=2m1m_2=\sqrt{2}m_1. In the language of spin-Peierls chains, the triply-degenerated excitations of soliton, antisoliton, and first breather correspond to spin-1 triplons around the wave number k=πk=\pi, while the second breather is spin-singlet excitation around k=πk=\pi. The energy bands of triplons, i.e., soliton (Sz=1S_z=1), first breather (Sz=0S_z=0), and anti-soliton (Sz=−1S_z=-1), are respectively given by

ϵS(kπ)=m12+v2kπ2−B,for Sz=1;ϵB(kπ)=m12+v2kπ2,for Sz=0;ϵAS(kπ)=m12+v2kπ2+B,for Sz=−1.\begin{aligned} \epsilon_{\mathrm{S}}(k_\pi) &= \sqrt{m_1^2+v^2k_\pi^2}-B,\quad&&\text{for } S_z=1;\\ \epsilon_{\mathrm{B}}(k_\pi) &= \sqrt{m_1^2+v^2k_\pi^2},\quad&&\text{for } S_z=0;\\ \epsilon_{\mathrm{AS}}(k_\pi) &= \sqrt{m_1^2+v^2k_\pi^2}+B,\quad&&\text{for } S_z=-1. \end{aligned}

where kπ=k−πk_\pi=k-\pi. The BB dependence of ϵAS\epsilon_{\mathrm{AS}} and ϵS\epsilon_{\mathrm{S}} reflects the Zeeman splitting.

Finally, we mention values of the parameters in the SG model for CuGeO3_3. We have K=1/2K=1/2 in the SU(2) model, and the spin gap (triplon gap) m1≃2.36m_1 \simeq2.36 meV (≈27.4\approx27.4 K) is estimated from an ESR experiment[12] at a low temperature. The velocity vv may be approximated by that of J1J_1J2cJ_2^c spin chains since α\alpha of CuGeO3_3 is closed to αc\alpha_c. The velocity has been numerically estimated as v=1.174J1a0v = 1.174J_1a_0 [17, 24]. Combining this value of vv and the exact result of the SG model[17, 20], we can compute the soliton mass as m1/J1≃2.13δ2/3m_1/J_1 \simeq2.13\delta^{2/3}. This relation, J1≃120.6J_1 \simeq120.6 K, and m1≃27.4m_1 \simeq27.4 K lead to δ≃0.034\delta\simeq0.034 in CuGeO3_3 at a sufficiently low temperature. If temperature TT grows up to the same order as the lowest gap ϵS(0)=m1−B\epsilon_{\mathrm{S}}(0) = m_1 - B, the fluctuation of δ\delta increases mainly due to the effect of phonons, and as a result, the description of the SG model gradually becomes invalid.

Approach based on Boltzmann equation

Using the above result, let us compute the thermal spin current in SSE for CuGeO3_3. The SG model predicts that when magnetic field BB is increased in the SP phase, the difference of the soliton density and anti-soliton one (i.e., spin current carrier) monotonically increases at a fixed temperature due to the Zeeman splitting. Nevertheless, the experimental result shows a non-monotonic BB dependence of the SSE voltage. It indicates the importance of scattering processes among triplons, impurities, phonons, etc.

If we consider a triplon as a wave packet located at rr with the wave vector k\mathbf{k}, the triplon distribution function fη(k,r,t)f_\eta(\mathbf{k},r,t) follows the Boltzmann equation,

∂∂tfη+vη(k)⋅∇rfη+dkdt⋅∇kfη=∂∂tfη∣scattering,(5)\frac{\partial}{\partial t}f_\eta+ \mathbf{v}_\eta(\mathbf{k})\cdot\nabla_r f_\eta+ \frac{d\mathbf{k}}{dt}\cdot\nabla_k f_\eta= \left.\frac{\partial}{\partial t}f_\eta\right|_{\mathrm{scattering}}, \tag*{(5)}

where the index η\eta corresponds to soliton (Sz=−1S_z=-1) and anti-soliton (Sz=1S_z=1). The right hand side is the scattering rate of triplons. The group velocity of triplon is vη(k)=1ℏ∂ϵη(k)∂k\mathbf{v}_\eta(\mathbf{k}) = \frac{1}{\hbar}\frac{\partial\epsilon_\eta(\mathbf{k})}{\partial\mathbf{k}}. Since breathers are spin-singlet quasi particles and do not contribute to spin current. Therefore, it is enough to consider soliton and anti-soliton for computing the spin current.

The real compound CuGeO3\mathrm{CuGeO}_{3} has a weak but finite three dimensionality, but we expect that the triplon distribution along the spin-chain (xx) direction is essential for the SSE in CuGeO3\mathrm{CuGeO}_{3}. Therefore, we will replace k\mathbf{k} with the wave number kk along the xx direction hereafter. In the SSE measurement, a temperature gradient is applied along the xx axis to the sample. If the system length LxL_x is sufficiently large, we may approximate the gradient as the constant ∂T/∂x=−Thigh−TlowLx≡ΔTx\partial T/\partial x = -\frac{T_{\mathrm{high}}-T_{\mathrm{low}}}{L_x} \equiv\Delta T_x. For a small ΔTx\Delta T_x, the system approaches to a non-equilibrium steady state following local equilibrium, and the distribution function is approximated as

fη=fη(0)(k)+gη(k,x,t),(6)f_{\eta}=f_{\eta}^{(0)}(k)+g_{\eta}(k,x,t), \tag*{(6)}

where fη(0)(k)f_{\eta}^{(0)}(k) is the equilibrium distribution function and gη(k,x,t)g_{\eta}(k,x,t) is the deviation from the equilibrium. We assume that fη(0)f_{\eta}^{(0)} can be described by a Bose distribution function,

fη(0)(k)=1eβ(ϵη(k)−μ)−1.(7)f_{\eta}^{(0)}(k)=\frac{1}{e^{\beta(\epsilon_{\eta}(k)-\mu)}-1}. \tag*{(7)}

Triplons are not equal to bosons or fermions, but the approximation of Supplementary Eq. (7) is expected to work well if triplon density is low enough. The non-equilibrium steady state satisfies ∂fη/∂t=dk/dt=0\partial f_{\eta}/\partial t = dk/dt = 0. For the scattering term, we adapt relaxation time (τk\tau_k) approximation [25]:

∂∂tfη∣scattering=−fη(k,x,t)−fη(0)(k)τk=−gη(k,x,t)τk.(8)\left.\frac{\partial}{\partial t}f_{\eta}\right|_{\mathrm{scattering}}=-\frac{f_{\eta}(k,x,t)-f_{\eta}^{(0)}(k)}{\tau_k}=-\frac{g_{\eta}(k,x,t)}{\tau_k}. \tag*{(8)}

Under the assumption that ∂T/∂x\partial T/\partial x and gg are small, we may omit the ∂g/∂T\partial g/\partial T term in Supplementary Eq. (5) and therefore it is simplified as

gη(k)=τkvη(k)kBΔxT∂∂(kBT)fη(0)(k)(9)g_{\eta}(k)=\tau_kv_{\eta}(k)k_{\mathrm{B}}\Delta_xT\frac{\partial}{\partial(k_{\mathrm{B}}T)}f_{\eta}^{(0)}(k) \tag*{(9)}

From these discussions, the total spin current with SzS^z polarization along the xx direction is

Js=∑Sz=1,−1∑kℏSzvSz(k)fSz(k)=∑Sz∑kℏSzvSz(k)gSz(k)=∑Sz∑kℏSzvSz(k)2τkkBΔxTϵSz(k)(kBT)2eβϵSz(k)(eβϵSz(k)−1)2=∑Sz∫B.Z.dk2πℏSzvSz(k)2τkkBΔxTϵSz(k)(kBT)2eβϵSz(k)(eβϵSz(k)−1)2.(10)\begin{aligned} J_s &= \sum_{S_z=1,-1}\sum_k \hbar S_z v_{S_z}(k) f_{S_z}(k) \\ &= \sum_{S_z}\sum_k \hbar S_z v_{S_z}(k) g_{S_z}(k) \\ &= \sum_{S_z}\sum_k \hbar S_z v_{S_z}(k)^2 \tau_k k_B \Delta_x T \frac{\epsilon_{S_z}(k)}{(k_B T)^2}\frac{e^{\beta\epsilon_{S_z}(k)}}{(e^{\beta\epsilon_{S_z}(k)}-1)^2} \\ &= \sum_{S_z}\int_{\mathrm{B.Z.}}\frac{dk}{2\pi}\hbar S_z v_{S_z}(k)^2\tau_k k_B\Delta_x T\frac{\epsilon_{S_z}(k)}{(k_B T)^2}\frac{e^{\beta\epsilon_{S_z}(k)}}{(e^{\beta\epsilon_{S_z}(k)}-1)^2}. \tag*{(10)} \end{aligned}

This is Equation (1) in the main text.

To numerically calculate the HH dependence of JsJ_s, it is necessary to determine the relaxation time τk\tau_k based on appropriate impurity scattering mechanisms. If we focus on the elastic scattering between a triplon and an impurity, τk\tau_k is estimated as [25]

τk(ϵ)−1∼nimpD(ϵ)V2,(11)\tau_k(\epsilon)^{-1}\sim n_{\mathrm{imp}}D(\epsilon)V^2, \tag*{(11)}

where nimpn_{\mathrm{imp}}, D(ϵ)D(\epsilon) and VV are impurity density, the density of states (DOS) of the triplons and impurity potential, respectively. In CuGeO3_3, Jc∼120J_c\sim120, Jb∼0.1JcJ_b\sim0.1J_c and Ja∼−0.01JcJ_a\sim-0.01J_c (ref. 13), and the two-dimensionality is strong compare to other one-dimensional systems. Therefore, We may approximate D(ϵ)D(\epsilon) at the bottom of the band as a constant. Consider the elastic scattering from both nonmagnetic and magnetic impurities, the total relaxation time can be approximated as

τk,total−1=τk,non−mag−1+τk,mag−1,(12)\tau_{k,\mathrm{total}}^{-1}=\tau_{k,\mathrm{non-mag}}^{-1}+\tau_{k,\mathrm{mag}}^{-1}, \tag*{(12)}

if the impurity densities are small enough. Combining these theoretical arguments and experimental results on the doped CuGeO3_3, we assume that the relaxation times from nonmagnetic and magnetic impurities is given by

τk,non−mag−1=nimpC0V02.(13)\tau_{k,\mathrm{non-mag}}^{-1}=n_{\mathrm{imp}}C_0V_0^2. \tag*{(13)}
τk,mag−1=(nimpBS(T,H))⋅Cmag⋅(VmagBS(T,H))2.(14)\tau_{k,\mathrm{mag}}^{-1}=\left(n_{\mathrm{imp}}B_S(T,H)\right)\cdot C_{\mathrm{mag}}\cdot\left(V_{\mathrm{mag}}B_S(T,H)\right)^2. \tag*{(14)}

C0C_0 and CmagC_{\mathrm{mag}} are constants and BS(T,H)B_S(T,H) is the Brillouin function with S=1/2S=1/2. The introduction of BSB_S is reasonable because it is expected that magnetically-polarized impurities more strongly refute triplons compared to weakly polarized ones. The numerical result of Js(H)J_s(H) at T=2.4T=2.4 K is shown in the main text as Figure 4, in which CmagVmag≫C0V0C_{\mathrm{mag}}V_{\mathrm{mag}}\gg C_0V_0 and the TT dependent triplon gap m1(T)m_1(T) measured in the neutron scattering are used. Due to the Zeeman-splitting induced density difference between soliton and antisoliton, the sign of spin current is opposite to that in ferromagnetic insulators. We verify that the broad peak structure of JsJ_s as a function of HH is stable against small changes of τk,mag−1\tau_{k,\mathrm{mag}}^{-1}. Moreover, the broad peak still survives even if the equilibrium distribution fn(0)f_n^{(0)} is changed into the fermion one in the regime of a low triplon density.

Approach based on the tunnel spin current

Another microscopic approach for SSE is the formula of tunnel spin current on the interface from the magnet (CuGeO3_3) to the metal (Pt)[2, 26, 27, 28]. An advantage of this approach is that it can be applied to a broad class of magnets even with non-magnon type excitations such as spinons, triplons, magnon pairs, etc. However, it cannot take into account the effects of scattering among magnetic excitations during a flow of spin currents.

In this approach, we focus on the interface between CuGeO3_3 and Pt. We assume that the dominant interaction at the interface is Heisenberg type exchange interaction between localized spin of CuGeO3_3 and conducting-electron spin of Pt. Such an exchange interaction has been used to describe interfacial spin transfer between an insulator magnet and a metal[2, 26, 27, 28]. The interface Hamiltonian is given by

Hint=Jint∑r∈interfaceSr⋅sr,(15)H_{\mathrm{int}} = J_{\mathrm{int}} \sum_{r \in\mathrm{interface}} \mathbf{S}_r \cdot\mathbf{s}_r, \tag*{(15)}

where Sr\mathbf{S}_r and sr\mathbf{s}_r respectively denote the localized spin and the conducting-electron spin at site rr on the interface. If we perturbatively treat this interface interaction using Keldysh Green’s function method[27, 29], the tunneling spin current is calculated as

Is∝−Jint2∫−∞∞dωIm⁡XR−+(ω)Im⁡χR−+(ω)[coth⁡(ω2kBTmetal)−coth⁡(ω2kBTmagnet)](16)I_s \propto-J_{\mathrm{int}}^2 \int_{-\infty}^{\infty} d\omega\operatorname{Im} X_R^{-+}(\omega) \operatorname{Im} \chi_R^{-+}(\omega) \left[ \operatorname{coth}\left(\frac{\omega}{2k_{\mathrm{B}}T_{\mathrm{metal}}}\right) - \operatorname{coth}\left(\frac{\omega}{2k_{\mathrm{B}}T_{\mathrm{magnet}}}\right) \right] \tag*{(16)}

up to the leading order of the interface coupling constant JintJ_{\mathrm{int}}. Here, ω\omega, TmetalT_{\mathrm{metal}} and TmagnetT_{\mathrm{magnet}} are respectively frequency, the spatially averaged value of temperature in the metal (Pt), and that in the CuGeO3_3. XR−+(ω)X_R^{-+}(\omega) and χR−+(ω)\chi_R^{-+}(\omega) are respectively the retarded part of the local dynamical susceptibility of the CuGeO3_3 at T=TmagnetT = T_{\mathrm{magnet}} and that of the Pt at T=TmetalT = T_{\mathrm{metal}}. The indices −+-+ denotes the transverse spins S±S^{\pm}. −Im⁡XR−+(ω)-\operatorname{Im} X_R^{-+}(\omega) with ω>0\omega> 0 may be viewed as the DOS of Sz=+1S_z = +1 magnetic excitations, while Im⁡XR−+(ω)\operatorname{Im} X_R^{-+}(\omega) with ω<0\omega< 0 as that of Sz=−1S_z = -1 magnetic excitations[2]. On the other hand, the magnetic energy scale in CuGeO3_3 is much smaller than that of the kinetic energy of electrons in the metal (Pt), and therefore the susceptibility of the metal can be approximated as[30, 31]

Im⁡χR−+(ω)≃ℏωD(ϵF)2+⋯ ,(17)\operatorname{Im}\chi_R^{-+}(\omega) \simeq\hbar\omega D(\epsilon_F)^2 + \cdots, \tag*{(17)}

where D(ϵF)D(\epsilon_F) is the DOS of conduction electrons at Fermi surface ϵ=ϵF\epsilon= \epsilon_F in the metal. Namely, Im⁡χR−+(ω)\operatorname{Im}\chi_R^{-+}(\omega) is an odd function of ω\omega. Therefore, the sign of the tunnel spin current reflects the difference between the weights of Sz=−1S_z = -1 and Sz=+1S_z = +1 modes. If the temperature difference ΔT=(Tmagnet−Tmetal)/2\Delta T = (T_{\mathrm{magnet}}-T_{\mathrm{metal}})/2 is sufficiently small, the TT-dependent factor in Supplementary Eq. (16) is approximated by

coth⁡(ω2kBTmetal)−coth⁡(ω2kBTmagnet)≃−ω(kBTave)21sinh⁡2(ω2kBTave)ΔT+⋯(18)\coth\left(\frac{\omega}{2k_{\mathrm{B}}T_{\mathrm{metal}}}\right)-\coth\left(\frac{\omega}{2k_{\mathrm{B}}T_{\mathrm{magnet}}}\right)\simeq-\frac{\omega}{(k_{\mathrm{B}}T_{\mathrm{ave}})^2}\frac{1}{\sinh^2\left(\frac{\omega}{2k_{\mathrm{B}}T_{\mathrm{ave}}}\right)}\Delta T+\cdots \tag*{(18)}

where TaveT_{\mathrm{ave}} is the averaged temperature (Tmagnet+Tmetal)/2(T_{\mathrm{magnet}}+T_{\mathrm{metal}})/2. Substituting Supplementary Eq. (17) and (20) into Supplementary Eq. (16), we arrive at the simplified formula

Is∝πJint2D(ϵF)2ΔT1(kBT)2∫−∞∞dωIm⁡XR+(ω)ω2sinh⁡2(ω2kBT)(19)I_{\mathrm{s}}\propto\pi J_{\mathrm{int}}^2D(\epsilon_{\mathrm{F}})^2\Delta T\frac{1}{(k_{\mathrm{B}}T)^2}\int_{-\infty}^{\infty}d\omega\operatorname{Im}X_{R}^{+}(\omega)\frac{\omega^2}{\sinh^2\left(\frac{\omega}{2k_{\mathrm{B}}T}\right)} \tag*{(19)}

where for simplicity we have replaced TaveT_{\mathrm{ave}} with TT.

Here, we emphasise that the formula of this tunnel spin current has suceeded in well explaining several SSEs in standard ferro(i)magnets[26], an one-dimensional spin liquid[2], a spin-nematic magnet[28], and a compensated ferrimagnet[32]. We also note two points about the formula of Supplementary Eq. (16). The first thing is that triplons in the magnetic insulator CuGeO3_3 cannot be injected to the metal Pt, while the spin of triplons is transfered to that of conducting electrons in Pt through the interface exchange JintJ_{\mathrm{int}}. Namely, spin angular momentum can tunnel from CuGeO3_3 to Pt and vice versa. This is in contrast with a tunnel charge current in bilayer systems consisiting of two metals. The second is about the possibility of the TT and HH dependences of JintJ_{\mathrm{int}}. The energy scale of chemical bonds at the interface is usually higher than those of the ranges of TT and HH (0 - 15 K and 0 - 14 T). Therefore, we have here assumed that JintJ_{\mathrm{int}} is independent of TT and HH.

The remaining task is to compute the susceptibility for a spin-Peierls chain Im⁡XR+(ω)\operatorname{Im}X_{R}^{+}(\omega). If CuGeO3_3 in the spin-Peierls phase is described by the SG model, we can use so-called the formfactor method [21]. Following it, we calculate the dynamical susceptibility in (k,ω)(k,\omega) space as

XR−+(kπ,ω)≃−Zω+ϵAS(kπ)+iγ+Zω−ϵS(kπ)+iγ+⋯(20)X_R^{-+}(k_\pi,\omega) \simeq\frac{-Z}{\omega+\epsilon_{\mathrm{AS}}(k_\pi)+i\gamma}+\frac{Z}{\omega-\epsilon_{\mathrm{S}}(k_\pi)+i\gamma}+\cdots \tag*{(20)}

around k=πk=\pi. Here the first and second terms are respectively the contribution of antisoliton (Sz=−1S_z=-1) and soliton (Sz=1S_z=1), and γ→+0\gamma\to+0 is the infinitesimal factor. The renormalization factor ZZ generally depends on kk and ω\omega, but it can be viewed as a constant if the system is ideally described by the integrable SG model. The local susceptibility XR−+(ω)X_R^{-+}(\omega) is related to XR−+(kπ,ω)X_R^{-+}(k_\pi,\omega) as XR−+(ω)=N−1∑kπXR−+(kπ,ω)X_R^{-+}(\omega)=N^{-1}\sum_{k_\pi}X_R^{-+}(k_\pi,\omega) with NN being the total site number. As we mentioned, in finite temperature case, the neutron-scattering experiments showed that the mass gaps of soliton and antisoliton decrease and their life time becomes shorter. This behavior cannot be reproduced within the SG model, but such TT dependence can be treated by replacing γ\gamma and m1m_1 of the bands ϵS,AS\epsilon_{\mathrm{S,AS}} with the line width Γ(T)\Gamma(T) and the TT-dependent mass m1(T)m_1(T), respectively. Namely, we may describe the finite-temperature susceptibility XR−+(kπ,ω)X_R^{-+}(k_\pi,\omega) as

XR−+(kπ,ω)≃−Zω+ϵAS(kπ,T)+iΓ(T)+Zω−ϵS(kπ,T)+iΓ(T)+⋯(21)X_R^{-+}(k_\pi,\omega) \simeq\frac{-Z}{\omega+\epsilon_{\mathrm{AS}}(k_\pi,T)+i\Gamma(T)}+\frac{Z}{\omega-\epsilon_{\mathrm{S}}(k_\pi,T)+i\Gamma(T)}+\cdots \tag*{(21)}

where the bands ϵS,AS(kπ,T)\epsilon_{\mathrm{S,AS}}(k_\pi,T) are defined as ϵS(kπ,T)=m1(T)2+v2kπ2−B\epsilon_{\mathrm{S}}(k_\pi,T)=\sqrt{m_1(T)^2+v^2k_\pi^2}-B and ϵAS(kπ,T)=m1(T)2+v2kπ2+B\epsilon_{\mathrm{AS}}(k_\pi,T)=\sqrt{m_1(T)^2+v^2k_\pi^2}+B.

Using Supplementary Eq. (19) and Supplementary Eq. (21), we can calculate the HH and TT dependence of the tunnel spin current in a low-TT and low-HH regime. The calculated result shows that a spin current linearly emerges as the field BB increases and its sign is opposite to that of SSE in ferromagnetic insulators. These results in the small-BB regime are consistent with those from the Boltzmann equation.

SSE in a moderate temperature regime

To compute the TT-dependence of JsJ_s in a relatively high temperature range (5 K - 10 K), we have to include the TT-dependence of triplon bands and scattering processes in the Boltzmann equation approach. To this end, we need information on the TT-dependence of the excitation gap of triplons m1(T)m_1(T) and the lifetime of triplons 1/τk(T)1/\tau_k(T). By fitting neutron scattering results of spin-gap m1(T)m_1(T) (ref. 33) with m1(T)=m1(0)(1−T/TSP)αm_1(T)=m_1(0)(1-T/T_{\mathrm{SP}})^{\alpha}, we obtain m1(0)∼2.05m_1(0)\sim2.05 meV and α≃0.12\alpha\simeq0.12. In the numerical calculations, m1m_1 in Supplementary Eq. (4)–(6) is replaced by the fitted function of m1(T)m_1(T). The TT-dependence of the lifetime of triplons is obtained as the peak width Γ(T)\Gamma(T) of triplon excitation from inelastic neutron scattering experiments^{33,34}. Here, the explicit TT-dependence of 1/τk1/\tau_k is set as a new term 1/τk(T)1/\tau_k(T), and the total relaxation time is

τk,total−1=τk,non−mag−1+τk,mag−1+τk(T)−1.(22)\tau_{k,\mathrm{total}}^{-1}=\tau_{k,\mathrm{non-mag}}^{-1}+\tau_{k,\mathrm{mag}}^{-1}+\tau_k(T)^{-1}. \tag*{(22)}

1/τk,non−mag1/\tau_{k,\mathrm{non-mag}} is independent of TT and 1/τk,mag1/\tau_{k,\mathrm{mag}} only weakly depends on TT via the Brillouin function.

The experimental results of triplon peak width Γ(T)\Gamma(T) obtained from Ref. [34, 35] are well fitted by a power law Γ(T)=Γ(0)T4\Gamma(T)=\Gamma(0)T^4, where Γ(0)\Gamma(0) is the fitting parameter. The scattering with thermally excited phonons and other triplons may be two origins of the TT-dependence of Γ\Gamma but the detail of microscopic mechanism is not clear. In this work, we treat these TT-dependent scattering processes phenomenologically. Specifically, we set τk(T)\tau_k(T) as

τk(T)−1=CthVth2Γ(0)(TTSP)4.(23)\tau_k(T)^{-1}=C_{\mathrm{th}}V_{\mathrm{th}}^2\Gamma(0)\left(\frac{T}{T_{\mathrm{SP}}}\right)^4. \tag*{(23)}

Where CthC_{\mathrm{th}} and VthV_{\mathrm{th}} are TT-independent constants.

Calculated result of the TT-dependent SSE is shown in Supplementary Figure 8. We have a qualitative agreement between experiment and theory in the moderate temperature range (approximately 5 K - 10 K).

Comparison with theory

Supplementary Figure 8. Comparison with theory. a-c Comparison between calculation results of the TT dependence of the triplon spin current and the observed SSE signal at μ0H=1,3,5\mu_0 H = 1, 3, 5 T, respectively. Error bars represent standard deviation.

Finally, we emphasize that the result of the temperature dependence is justified only in the moderate temperature range. The reason why our result is less reliable in the high-TT range (T→TSPT \rightarrow T_{\mathrm{SP}}) is as follows: our theory is based on the Sine-Gordon model, which can well describe the low-energy physics of the SP phase. When TT or HH increases, the dimerized ground state is partially broken down and the SG model gradually becomes less reliable. For T→TSPT \rightarrow T_{\mathrm{SP}}, the critical nature of the SP transition becomes relevant, and the triplon picture is no longer applicable. This is also why theoretical calculations in Supplementary Notes G1-G3 focused only on the low temperatures and low magnetic fields, in which the singlet ground state and triplon excitation are well established. On the other hand, the relation of Γ(T)∝T4\Gamma(T) \propto T^4 is not experimentally supported in the low-TT range. Thus, Supplementary Figure 8 shows the temperature range from 5 K to 11 K.

References

  1. [1]Fujita, M. et al. Temperature Dependence of Spin Excitations in the Frustrated Spin Chain System CuGeO3. Journal of the Physical Society of Japan 82, 084708 (2013).DOI
  2. [2]Hirobe, D. et al. One-dimensional spinon spin currents. Nature Physics 13, 30–34 (2017).DOI
  3. [3]Giamarchi, T., Rüegg, C. & Tchernyshyov, O. Bose-Einstein condensation in magnetic insulators. Nature Physics 4, 198–204 (2008).
  4. [4]Seki, S. et al. Thermal Generation of Spin Current in an Antiferromagnet. Phys. Rev. Lett. 115, 266601 (2015).DOI
  5. [5]Wu, S. M. et al. Antiferromagnetic Spin Seebeck Effect. Phys. Rev. Lett. 116, 097204 (2016).DOI
  6. [6]Maekawa, S., Valenzuela, S., Saitoh, E. & Kimura, T. Spin Current (OUP Oxford, Oxford, 2012).
  7. [7]Oyanagi, K., Kikkawa, T. & Saitoh, E. Magnetic field dependence of the nonlocal spin Seebeck effect in Pt/YIG/Pt systems at low temperatures. AIP Advances 10, 015031 (2020).DOI
  8. [8]Wu, S. M., Fradin, F. Y., Hoffman, J., Hoffmann, A. & Bhattacharya, A. Spin Seebeck devices using local on-chip heating. Journal of Applied Physics 117, 17C509 (2015).DOI
  9. [9]Salce, B. et al. Thermal conductivity of pure and Si-doped CuGeO3. Physics Letters A 245, 1560–1563 (1998).DOI
  10. [10]Hase, M., Terasaki, I. & Uchino­kura, K. Observation of the spin-Peierls transition in linear Cu2+ (Spin-1/2) chains in an inorganic compound CuGeO3. Physical Review Letters 70, 3651–3654 (1993).pubmed.ncbi.nlm.nih.gov/10053928
  11. [11]Grenier, B. et al. Magnetic susceptibility and phase diagram of CuGe₁−ₓSiₓO₃ single crystals. Physical Review B 57, 3444–3453 (1998).DOI
  12. [12]Nojiri, H. et al. Submillimeter Wave ESR Study of Spin Gap Excitations in CuGeO₃. Journal of the Physical Society of Japan 68, 3417–3423 (1999).DOI
  13. [13]Nishi, M., Fujita, O. & Akimitsu, J. Neutron-scattering study on the spin-Peierls transition in a quasi-one-dimensional magnet CuGeO₃. Physical Review B 50, 6508–6510 (1994).pubmed.ncbi.nlm.nih.gov/9977042
  14. [14]Kamieniarz, G., Bieliński, M. & Renard, J. P. Susceptibility behavior of CuGeO₃: Comparison between experiment and the quantum transfer-matrix approach. Physical Review B 60, 14521–14524 (1999).DOI
  15. [15]Okamoto, K. & Nomura, K. Fluid-dimer critical point in S = 1/2 antiferromagnetic Heisenberg chain with next nearest neighbor interactions. Physics Letters A 169, 433–437 (1992).
  16. [16]Giamarchi, T. Quantum Physics In One Dimension (Clarendon Press, Oxford, 2003).
  17. [17]Takayoshi, S. & Sato, M. Coefficients of bosonized dimer operators in spin-½ XXZ chains and their applications. Physical Review B 82, 214420 (2010).DOI
  18. [18]Gogolin, A., Nersesyan, A. & Tsvelik, A. Bosonization and Strongly Correlated Systems (Cambridge University Press, 2004).
  19. [19]Hikihara, T., Furusaki, A. & Lukyanov, S. Dimer correlation amplitudes and dimer excitation gap in spin-1/2 XXZ and Heisenberg chains. Physical Review B 96, 134429 (2017).DOI
  20. [20]Lukyanov, S. & Zamolodchikov, A. Exact expectation values of local fields in the quantum sine-Gordon model. Nuclear Physics B 493, 571–587 (1997).DOI
  21. [21]Kuzmenko, I. & ESSLER, F. H. L. Dynamical correlations of the spin-1/2 Heisenberg XXZ chain in a staggered field. Physical Review B 79, 024402 (2009).DOI
  22. [22]Katsura, H., Sato, M., Furuta, T. & Nagaosa, N. Theory of the Optical Conductivity of Spin Liquid States in One-Dimensional Mott Insulators. Physical Review Letters 103, 177402 (2009).pubmed.ncbi.nlm.nih.gov/19905782
  23. [23]Sato, M., Katsura, H. & Nagaosa, N. Theory of Raman Scattering in One-Dimensional Quantum Spin-1/2 Antiferromagnets. Physical Review Letters 108, 237401 (2012).DOI
  24. [24]Okamoto, K. & Nakamura, T. Critical properties of the spin-Heisenberg chain with frustration and bond alternation. Journal of Physics A: Mathematical and General 30, 6287–6298 (1997).DOI
  25. [25]Abrikosov, A. Fundamentals of the Theory of Metals (Dover Publications, Mineola, New York, 2017).
  26. [26]Adachi, H., Ohe, J.-i., Takahashi, S. & Maekawa, S. Linear-response theory of spin Seebeck effect in ferromagnetic insulators. Physical Review B 83, 533 (2011).DOI
  27. [27]Jauho, A.-P., Wingreen, N. S. & Meir, Y. Time-dependent transport in interacting and noninteracting resonant-tunneling systems. Physical Review B 50, 5528 (1994).pubmed.ncbi.nlm.nih.gov/9976896
  28. [28]Hirobe, D. et al. Magnon Pairs and Spin-Nematic Correlation in the Spin Seebeck Effect. Physical Review Letters 123, 117202 (2019).DOI
  29. [29]Haug, H. & Jauho, A. Quantum Kinetics In Transport And Optics Of Semiconductors (Springer, Berlin, Heidelberg, 2007).
  30. [30]Moriya, T. The Effect of Electron-Electron Interaction on the Nuclear Spin Relaxation in Metals. Journal of the Physical Society of Japan 18, 516–520 (1963).DOI
  31. [31]Shastry, B. S. & Abrahams, E. What does the Korringa ratio measure? Physical Review Letters 72, 1933–1936 (1994).pubmed.ncbi.nlm.nih.gov/10055741
  32. [32]Geprägs, S. et al. Origin of the spin Seebeck effect in compensated ferrimagnets. Nature Communications 7, 1–6 (2016).DOI
  33. [33]Lussier, J. G., Coad, S. M., McMorrow, D. F. & Paul, D. M. The temperature dependence of the spin-Peierls energy gap in CuGeO₃. Journal of Physics: Condens. Matter 8, L59–L64 (1996).
  34. [34]Regnault, L. P., Ain, M., Hennion, B., Dh Ahmad, G. & Revcolevschi, A. Inelastic-neutron-scattering investigation of the spin-Peierls system CuGeO₃. Physical Review B 53, 5579–5597 (1996).pubmed.ncbi.nlm.nih.gov/9984167
  35. [35]Kikuchi, J., Yasuoka, H., Hase, M., Sasago, Y. & Uchinokura, K. Cu nuclear quadrupole resonance study of cugeo3. Journal of the Physical Society of Japan 63, 872–875 (1994).DOI

Paper details

vlnc:W_7sbyq9d8
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