Direct observation of magnon-phonon coupling in yttrium iron garnet

Haoran Man,1, ∗ Zhong Shi,2, 3, ∗ Guangyong Xu,4 Yadong Xu,2 Xi Chen,5 Sean Sullivan,5 Jianshi Zhou,5 Ke Xia,6 Jing Shi,2, † and Pengcheng Dai1, 6, ‡ 1Department of Physics and Astronomy, Rice University, Houston, Texas 77005, USA 2Department of Physics and Astronomy, University of California, Riverside, California 92521, USA 3School of Physics Science and Engineering, Tongji University, Shanghai 200092, China 4NIST Center for Neutron Research, National Institute of Standards and Technology, Gaithersburg, Maryland 20899, USA 5Materials Science and Engineering Program, Texas Materials Institute, The University of Texas at Austin, Austin, Texas 78712, USA 6Department of Physics, Beijing Normal University, Beijing 100875, China (Dated: September 13, 2017) The magnetic insulator yttrium iron garnet (YIG) with a ferrimagnetic transition temperature of ∼560 K has been widely used in microwave and spintronic devices. Anomalous features in the spin Seeback effect (SSE) voltages have been observed in Pt/YIG and attributed to the magnon-phonon coupling. Here we use inelastic neutron scattering to map out low-energy spin waves and acoustic phonons of YIG at 100 K as a function of increasing magnetic field. By comparing the zero and 9.1 T data, we find that instead of splitting and opening up gaps at the spin wave and acoustic phonon dispersion intersecting points, magnon-phonon coupling in YIG enhances the hybridized scattering intensity. These results are different from expectations of conventional spin-lattice coupling, calling for new paradigms to understand the scattering process of magnon-phonon interactions and the resulting magnon-polarons.

Spin waves (magnons) and phonons are propagating [8]. Finally, if magnon and phonon lifetime-broadening disturbance of the ordered magnetic moment and lat- is smaller than their interaction strength, the resulting tice vibrations, respectively. They constitute two fun- mixed quasiparticles can form magnon polarons [6, 7]. damental quasiparticles in a solid and can couple to- gether to form a hybrid quasiparticle [1, 2]. Since our Here we use inelastic neutron scattering to study low- current understandings of these quasiparticles are based energy ferromagnetic magnons and acoustic phonons in on linearized models that ignore all the high-order terms the ferrimagnetic insulator yttrium iron garnet (YIG) than quadratic terms and neglect interactions among the with chemical formula Y3Fe5O12 [Figs. 1(a)-1(d)] [12– quasiparticle themselves [3], magnons and phonons are 14]. At zero field and 100 K, we confirm the quadratic believed to be stable and unlikely to interact and break- wave vector dependence of the magnon energy, E = Dq2, down for most purposes [4]. Therefore, discovering and where D is the effective spin wave stiffness constant and understanding how the otherwise stable magnons and q is momentum transfer (in A˚−1 or 1010m−1) away from phonons can couple and interact with each other to in- a Bragg peak [Fig. 1(e)] [13–19]. We also confirm the fluence the electronic properties of solids are one of the linear dispersion of the TA phonon mode [Fig. 1(e)]. central themes in modern condensed matter physics. Upon application of a magnetic field H0, a spin gap of the magnitude gH0 (g ≈ 2 is the Land´eelectron spin g- In general, spin-lattice (magnon-phonon) coupling can factor) opens and lifts up spin waves spectra away from modify magnon in two different ways. First, the static the field-independent phonon dispersion [Figs. 1(f) and lattice distortion induced by the magnetic order may af- 1(g)] [13, 14]. By comparing the zero and 9.1 T field wave fect the anisotropy of magnon exchange couplings, as vector dependence of the spin wave spectra, we find that seen in the spin waves of iron pnictides with large in- instead of splitting and opening up gaps at the spin wave plane magnetic exchange anisotropy [5]. Second, the and acoustic phonon dispersion intersecting points, hy- dynamic lattice vibrations interact with time-dependent bridized magnon polaron scattering at the intersecting spin waves may give rise to significant magnon-phonon points has larger intensity at zero field and magnons re- arXiv:1709.03940v1 [cond-mat.str-el] 12 Sep 2017 coupling [6, 7]. One possible consequence of such cou- main unchanged at other wave vectors as shown schemat- pling is to create energy gaps in the magnon dispersion ically in the bottom panels of Figs. 1(f) and 1(g). This is at the nominal intersections of the magnon and phonon different from the expectations of conventional magnon- modes [8, 9], as seen in antiferromagnet (Y,Lu)MnO3 phonon interaction, where hybridized polaronic excita- [10]. Alternatively, magnon-phonon coupling may give tions at the crossing points should have the sum of sep- rise to spin-wave broadening at the magnon-phonon arate magnon and phonon scattering intensity, and be- crossing points [11]. In both cases, we expect the in- come broader in energy due to the repulsive magnon- tegrated intensity of hybridized excitations at the inter- phonon dispersion curves [8–11]. Our results thus reveal secting points to be the sum of separate magnon and a new magnon-phonon coupling mechanism, calling for phonon scattering intensity without spin-lattice coupling a new paradigm to understand the scattering process of 2

The spin Seebeck effect (SSE), which allows spin cur- rents produced by thermal gradients in magnetic mate- rials to be transmitted and converted to charge voltages a b L magnon scan phonon scan a/2 Fe ‘a’ H in a heavy metal such as Pt, is one of the most techno- Fe ‘d’ 0 Y ‘c’ Q = (4,0,0) + q H logically relevant thermoelectric phenomena to be used a a

a Q = (2,2,0) + q K in ‘spin caloritronic’ devices [24–29]. In the case of a Pt 35 31 TA 0.2 LA

0.1 V) d V) µ µ 0.1 film on the surface of a polished single-crystalline YIG ( 34 ( 30 V c V 0.0 0.0 V)

µ 2.0 2.5 3.0 8.5 9.0 9.5 10.0 µ H (T) ( µ H (T) 0 0 slab (Pt/YIG) [Fig. 1(c)] [30], anomalous features in

V 33 29 magnetic field dependence of the SSE voltages at low H 32 28 1.5 2.0 2.5 3.0 3.5 8.0 8.5 9.0 9.5 10.0 µ H (T) µ H (T) 0 0 temperatures are attributed to the magnon-phonon in- e 8 magnon magnon 0 T Energy scan teraction at the “touching” points between the magnon magnon 9.1 T Q = (2,2,0) + q 6 LA phonon and transverse acoustic (TA) and longitudinal acoustic TA phonon phonon 4 Energy scan Q = (4,0,0) + q (LA) phonon as magnon dispersion curve is lifted by the Energy (meV) 2

0 applied field while phonon is not affected by the field 0 2 4 6 8 10 12 q (108 m-1) [Fig. 1(d)] [31]. While we find no anomaly at the magnon f 0 T 9.1 T g 0 T 9.1 T LA LA and TA/LA acoustic phonon touching points, our data TA TA Energy Energy reveal clear evidence for magnon-phonon interaction at TA TA q q q q zero field, consistent with the formation of magnon po- magnon phonon magnon phonon 0 T 0 T 0 T 0 T 9.1 T 9.1 T 9.1 T 9.1 T larons. χ ’’ χ ’’ Our neutron scattering experiment was carried out at Energy Energy Energy Energy NIST center for neutron research, Gaithersburg, Mary- land [32]. The full body-centered-cubic unit cell of YIG with space group Ia3d comprises eight cubes that are re- lated by glide planes to the basic cube as shown in Fig. FIG. 1: (a) The full unit cell of YIG comprises eight cubes 1(a), where the metallic atomic sites are labelled as ‘a’, that are related by glide planes to the basic cube shown in the figure. (b) The corresponding reciprocal space with the ‘d’, and ‘c’ [13]. Using the cubic lattice parameter of ˚ [H,K, 0] scattering and vertical magnetic field H0. The red a = b = c = 12.376 A, we define momentum transfer and green solid circles mark the positions of reciprocal space Q in three-dimensional (3D) reciprocal space in A˚−1 as where we probe spin waves and acoustic phonons, respectively. Q = Ha∗ + Kb∗ + Lc∗, where H, K, and L are Miller (c) A picture of the Pt/YIG device used for SSE measure- indices and a∗ = aˆ2π/a, b∗ = bˆ2π/a, c∗ = cˆ2π/a [Figs. ments. (d) SSE voltage in the field ranges where anomalous 1(a) and 1(b)]. Consistent with Ref. [31], the magnetic features appear at 100 K. (e) Magnon and phonon disper- field dependence of SSE voltage on our Pt film on YIG sions of YIG at 100 K and different magnetic fields. The black squares and solid red circles are data from 0 T and 9.1 contains two anomalous features at 2.5 T and 9.1 T [Figs. T measurements, respectively. The q ≈ 4.5 × 108 m−1 point 1(c)-1(e)] [32–36]. corresponds to ∆Q = 0.062 in Fig. 2(d). The black and The sample for neutron scattering experiments was ori- red solid lines are quadratic ferromagnetic spin wave fit to ented with a and b(a)-axis of the crystal in the horizontal the data. The blue and red boxes indicate magnon-phonon [H,K, 0] scattering plane [Fig. 1(b)] and mounted inside crossing points. The green and blue solid lines are TA (with a 10 T vertical field magnet. In this geometry, we mea- phonon velocity C ≈ 3.9 × 103 m/s)and LA (C ≈ 7.2 × 103 ⊥ || sured magnon dispersion around (2, 2, 0) and phonon dis- m/s) phonons, respectively [31]. (f,g) The expanded view of the blue and red boxes in (e), respectively. The bottom panels persion around (4, 0, 0). The momentum transfers Q at in (f,g) summarize the results obtained in our measurements these wave vectors are Qmagnon = (2 + ∆Q, 2 + ∆Q, 0) on the magnetic field effect on spin waves, hybridized excita- and Qphonon = (4, ∆Q, 0) for TA phonon [Fig. 1(b)]. tions, and TA phonons. For convenience,√ we calculate relative momentum trans- fer as q = 2π 2∆Q/a for magnon and q = 2π∆Q/a for phonon. We chose (2, 2, 0) for magnetic and (4, 0, 0) for magnon-phonon interactions and the resulting magnon phonon measurements because of their huge differences polarons [31]. in nuclear structure factors [4.75 at (2, 2, 0) versus 50.5 at We chose to study magnon-phonon coupling in YIG (4, 0, 0)], which is directly related to the acoustic phonon because it is arguably the most important material used intensity. Although we expect to find mostly magnetic in microwave and recent spintronic devices [20]. In ad- scattering at (2, 2, 0) and phonon scattering at (4, 0, 0), dition to having a ferrimagnetic ordering temperature the finite Fe3+ magnetic form factor of |F (Q)| means of ∼560 K suitable for room temperature applications, that there are still magnetic contributions to the phonon YIG can be grown with exceptional quality, and has the scattering at (4, 0, 0) (|F (2, 2, 0)|2 / |F (4, 0, 0)|2 ≈ 1.86). lowest Gilbert damping of any known materials and a Magnetic neutron scattering directly measures the narrow magnetic resonance linewidth allowing transmis- magnetic scattering function S(Q,E), which is propor- sion of spin waves over macroscopic distances [21–23]. tional to the imaginary part of the dynamic susceptibil- 3

ΔQ (r.l.u) ΔQ (r.l.u) 0 0.05 0.10 0.15 0 0.05 0.10 0.15 0 T 5K a c b 3 100K 3.5 3.5 4 a ΔQ = 0.062 b ΔQ = 0.112 3.0 0 T 3.0 ’‘(q, E-gH ) χ 0 2 2.5 9.1 T 2.5 9.1 T 9.1T 100K 0T 100K 2 0 T 0T 5K 2.0 2.0 1 ’‘(q, E) (a.u.) χ ’‘(q, 1.5 1.5 Energy (meV)Energy

’‘(q, E) theoretical (a.u.) E) theoretical χ ’‘(q, (ΔQ, ΔQ, 0) q = ΔQ = 0.092 1.0 1.0 0 0 Intensity (a.u.) 1.05 meV Intensity (a.u.) 0 5 10 0 5 10 0 2 4 6 8 0.5 0.5 q (108 m-1) q (108 m-1) Energy (meV) 0.0 Q=(2+ΔQ, 2+ΔQ, 0) 0.0 d 0 T e 9.1 T -2 0 2 4 6 0 2 4 6 8 10 2.0 2.0 Energy (meV) Energy (meV) ΔQ = 0.152 ΔQ = 0.152 3.5 1.8 1.8 c ΔQ = 0.132 2.5 d ΔQ = 0.152 1.6 ΔQ = 0.132 1.6 ΔQ = 0.132 3.0 1.4 1.4 2.5 2.0 ΔQ = 0.112 ΔQ = 0.112 1.2 1.2 2.0 1.5 1.0 ΔQ = 0.092 1.0 ΔQ = 0.092

’’ (a.u.) χ ’’ 1.5 ’’ (a.u.) χ ’’ 0.8 0.8 1.0

ΔQ = 0.062 ΔQ = 0.062 1.0 Intensity (a.u.) 0.6 0.6 Intensity (a.u.) 0.5 0.4 Instrument Resolution 0.4 ΔQ = 0.032 0.5 0.2 0.2 0.0 0.0 ΔQ = 0.012 0.0 Q=(2+ΔQ, 2+ΔQ, 0) 0.0 0 2 4 6 8 10 2 4 6 8 10 Energy (meV) Energy (meV) 0 2 4 6 8 10 0 2 4 6 8 10 Energy (meV) Energy (meV)

FIG. 2: (a) Schematic illustration of the expected magnon FIG. 3: (a,b,c,d) Comparison of the estimated χ00(Q,E) as dispersions at 0 T and 9.1 T for a simple ferromagnet. (b) a function of increasing wave vector at 0 T (black) and 9.1 The expected temperature, magnetic field dependence of low- T (red). The 9.1T data is shifted by 1.05 meV to accommo- energy χ00(Q,E) for simple ferromagnet obtained from SpinW date the field induced energy shift. Light red dots represents software package [39]. Here the magnetic field induced spin the original data position of the 9.1 T data. The horizontal 00 gap gH0 has been subtracted in the 9.1 T χ (q, E−gH0) (red). bars are estimated instrumental energy resolution based on The upper and bottom units are ∆Q and q, respectively. (c) magnon dispersion at 100 K. Our estimated χ00(Q,E) with Q = (2.092, 2.092, 0) at 5 K and 100 K after correcting measured S(Q,E) for the background and Bose-population factor. (d,e) The estimated χ00(Q,E) at 0 T and 9.1 T, respectively, after correcting for background energy of YIG at different temperatures and magnetic and Bose population factor. Scans at different wave vectors fields. Figure 2(c) shows our estimated constant-Q scans are lifted up by 0.3 sequentially. The black and red arrows [Q = (2.092, 2.092, 0) or ∆Q = 0.092 rlu] of χ00(Q,E) at 5 marks the peak positions at 0 T and 9.1 T, respectively. K (filled black squares) and 100 K (filled orange circles). Consistent with the expectation, we see that χ00(Q,E) at these two temperatures are identical within the er- 2 ity χ00(Q,E) through S(Q,E) ∝ |F (Q)| χ00(Q,E)/[1 − rors of the measurement. Figure 2(d) shows constant- E exp(− )], where E is the magnon energy, kB is the kB T Q scans of spin waves of YIG at 100 K and 0 T. At Boltzmann constant [2]. Although YIG is a ferrimag- Q = (2.062, 2.062, 0) or ∆Q = 0.062, χ00(Q,E) has a net, its low-energy spin waves can be well described as clear peak in energy that is slightly larger than the in- a simple ferromagnet [17]. In the hydrodynamic limit of strumentation resolution (horizontal bar). With increas- long wavelength (small-q) and small energies, we expect ing ∆Q, the peak in χ00(Q,E) moves progressively to 2 E = ∆0+gH0+Dq for spin wave dispersion, where ∆0 is higher energies. We have attempted but failed to fit the the possible intrinsic spin anisotropy gap, gH0 is the size spin wave spectra with a simple harmonic oscillator gen- of the magnetic field induced spin gap, and D is in units erally used for a ferromagnet [38]. This may be con- of meVA˚2 [Fig. 2(a)] [13–16]. In addition, for a pure sistent with recent inelastic neutron scattering study of magnetic ordered system without spin-lattice interaction, YIG that reveals the need to use long range magnetic we expect that χ00(Q,E) to be independent of temper- exchange couplings to fit the overall spin wave spectra ature at temperatures well below the magnetic ordering [19]. By fitting the spin wave spectra at zero field with temperature and applied magnetic field after correcting an exponentially modified Gaussian peak function [32], for the field-induced spin gap gH0 [Fig. 2(b)] [37–39]. we obtain the magnon dispersion curve as shown in Fig. 2 To determine if temperature and magnetic field de- 1(e). Fitting the dispersion curve with E = ∆0 + Dq 2 pendence of spin waves in YIG follow these expecta- yields ∆0 ≈ 0 and D = 580 ± 60 meVA˚ , consistent with tions, we measured wave vector dependence of magnon earlier work giving D ≈ 533 meVA˚2 [14]. 4

tical as expected. At the second magnon-phonon cross- 1.2 3.0 a 0 T 0 T ing point with ∆Q ≈ 0.13 [see red arrow in Fig. 1(g)], 1.0 2.5 b Q=(4, 0.2, 0) 9.1 T Q=(4, 0.1, 0) 2.5 T 9.1 T 00 0.8 2.0 the differences between χ (Q,E) at 0 T and 9.1 T are monitor monitor 5 0.6 5 1.5 even more obvious, with intensity at 0 T considerably 0.4 1.0 larger than that at 9.1 T [Fig. 3(c)]. Finally, on moving CNTS/10 0.2 CNTS/10 0.5 to ∆Q = 0.152 well above the magnon-phonon crossing 0.0 0.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Energy (meV) Energy (meV) point wave vectors [Fig. 1(e)], we again see no obvious 00 2.0 5 difference in χ (Q,E) between 0 T and 9.1 T. c 0 T 5 T d FWHM 0 T 2.5 T 9.1 T 4 FWHM 9.1T 1.5 Figure 3 shows that magnetic field dependence of 3 00 1.0 χ (Q,E) is highly wave vector selective, revealing clear 2 magnetic field induced intensity reduction in χ00(Q,E)

0.5 Energy(meV) Integrated Area Integrated 1 at wave vectors associated with magnon-phonon cross- Q=(4, 0.1, 0) Q=(2+ΔQ, 2+ΔQ, 0) 0.0 0 0 2 4 6 8 10 0.00 0.05 0.10 0.15 ing points while having no effect at other wave vectors. Magnetic Field (T) ∆ Q (r.l.u) To confirm the presence of TA phonon and determine its magnetic field effect, we carried out TA phonon mea- surements near (4, 0, 0), which has a rather large nu- clear structure factor compared with (2, 2, 0). Figure 4(a) shows energy scans of at Q = (4, 0.2, 0) and 100 K, which is along the green arrow direction in Fig. 1(g) and far away from the magnon dispersion. The spectra reveal a clear magnetic field independent peak at E ≈ 3 meV, confirming the TA phonon nature of the scattering. Fig- ure 4(b) shows similar energy scan at Q = (4, 0.1, 0) and 100 K, which is along the green arrow direction and near FIG. 4: (a) Energy scan of S(Q,E) at Q = (4, 0.2, 0), a the magnon-phonon crossing point in Fig. 1(f). At 0 position far away from magnon-phonon crossing points, and 100 K to probe TA phonon at 0 T and 9.1 T (b) Energy scan T, we see a peak around E ≈ 1.7 meV consistent with of S(Q,E) to probe magnon-phonon hybridized excitations at dispersions of magnon and TA phonon. With increas- Q = (4, 0.1, 0) near magnon-phonon crossing point at 0 T, 2.5 ing field to 2.5 T and 9.1 T, the intensity of the peak T, and 9.1 T. (c) Magnetic field dependence of the integrated decreases, but its position in energy remains unchanged intensity of magnon-phonon hybridized excitations at 100 K [Fig. 4(b)]. Figure 4(c) shows magnetic field dependence and Q = (4, 0.1, 0). (d) FWHM of the magnon at 0 T and of the integrated intensity, confirming the results in Fig. 9.1 T as a function of ∆Q. 4(b). Since the energy of the magnon should increase with increasing magnetic field, the field independent na- ture of the peak position in Fig. 4(b) suggests that the Upon application of a 9.1 T field at 100 K, we expect mode cannot be a simple addition of magnon and phonon, the magnon dispersion curve to be lifted by gH0 ≈ 1 but most likely arises from hybridized magnon polarons meV. This would be consistent with the observation of [6, 7]. Figure 4(d) shows the full width at half maximum a sharp gap below 1.05 meV in constant-Q scan at Q = (FWHM) of the magnon width at 0 T and 9.1 T. Within (2.012, 2.012, 0) (∆Q = 0.012) [Fig. 2(e)]. Constant- the errors of our measurements, we see no energy width Q scan at Q = (2.032, 2.032, 0) shows similar behavior. change in the measured wave vector region. Figure 2(e) also shows constant-Q scans at identical wave Our results provided compelling evidence for the pres- vectors as those in Fig. 2(d) at 0 T. Using data in Fig. ence of magnon-phonon coupling in YIG at the magnon- 2(e), we plot the magnon dispersion at 9.1 T field in Fig. phonon crossing points at zero field. This is clearly dif- 1(e). Consistent with the expectation, we see a clear gH0 ferent from the SSE measurements, where anomalies are upward shift in magnon energy but the spin wave stiffness only seen at the critical fields that obey “touch” condition D remains unchanged. at which the mangnon energy and group velocity agree To quantitatively determine the magnetic field effect with that of the TA/LA phonons. When the applied on χ00(Q,E) of YIG, we compare χ00(Q,E) at 0 T with field is less than the critical field, the magnon disper- those at 9.1 T. Figure 3(a)-3(d) summarizes the energy sion has two intersections with TA/LA phonon modes. dependence of χ00(Q,E) after down shifting the 9.1 T When the applied field is larger than the critical field, the data by gH0 = 1.05 meV. At ∆Q = 0.062, the scan along magnon dispersoin is separated from the TA/LA phonon the red arrow direction near the magnon-phonon cross- modes. In the theory of hybrid magnon-phonon excita- ing point as shown in Fig. 1(f), we see that χ00(Q,E) at tions [6, 7], the SSE anomalies occur at magnetic fields 9.1 T field is lower in intensity compared with those at and wave vectors at which the phonon dispersion curves 0 T. On moving to ∆Q = 0.10 with no magnon-phonon are tangents to the magnon dispersion, where the ef- crossing, χ00(Q,E) at 0 T and 9.1 T are virtually iden- fects of the magnon-phonon coupling are maximized [40]. 5

While our findings of a novel magnon-phonon coupling at Phys. 20, 1119 (1987). zero field are consistent with the formation of magnon- [17] J. Barker and G. E. W. Bauer, Phys. Rev. Lett. 117, polarons in YIG [6, 7], they are not direct proof that 217201 (2016). magnon-polaron formation alone causes anomalous fea- [18] L. S. Xie, G. X. Jin, L. X. He, G. E. W. Bauer, J. Barker, and K. Xia, Phys. Rev. B 95, 014423 (2017). tures in the magnetic field and temperature dependence [19] A. J. Princep, R. A. Ewings, S. Ward, S. T´oth,C. Dubs, of the SSE. Other effects, such as spin diffusion length, D. Prabhakaran, A. T. Boothroyd, arXiv:1705.06594. acoustic quality of the YIG film, and magnon spin con- [20] A.V. Chumak, V. I. Vasyuchka, A. A. Serga, and B. Hille- ductivity also play an important role in determining the brands. Nature Physics 11, 453461 (2015); SSE anomaly [41]. Regardless of the microscopic origin [21] A. A. Serga, A.V. Chumak, and B. Hillebrands. J. Phys. of the SSE anomaly, our discovery suggests the need to D: Appl. Phys. 43 264002 (2010). understand why magnon-phononok interaction and the [22] A.V. Chumak, , A. A. Serga, and B.Hillebrands. Nat. Commun. 5, 4700 (2014). resulting magnon polarons enhance the hybridized exci- [23] Y. Kajiwara, K. Harii, S. Takahashi, J. Ohe, K. Uchida, tations at the magnon-phonon intersection points. M. Mizuguchi, H. Umezawa, H. Kawai, K. Ando, K. The neutron scattering work at Rice is supported by Takanashi, S. Maekawa, and E. Saitoh, Nature (London) the U.S. DOE, BES de-sc0012311 (P.D.). The materials 464, 262 (2010). work at Rice is supported by the Robert A. Welch Foun- [24] Gerrit E. W. Bauer, Eiji Saitoh, and Bart J. van Wees, dation Grant No. C-1839 (P.D.). The work at UCR (J.S. Nat. Mater. 11, 391 (2012). and Z.S.) is supported as part of the SHINES, an En- [25] K. Uchida, S. Takahashi, K. Harii, J. Ieda, W. Koshibae, K. Ando, S. Maekawa, and E. Saitoh, Nature 455, 778 ergy Frontier Research Center funded by the U.S. DOE, (2008). BES under Award No. SC0012670. YIG crystal growth [26] K. Uchida, J. Xiao, H. Adachi, J. Ohe, S. Takahashi, J. at UT-Austin is supported by the Army Research Office Ieda, T. Ota, Y. Kajiwara, H. Umezawa, H. Kawai, G. MURI award W911NF-14-1-0016. E. W. Bauer, S. Maekawa, E. Saitoh, Nat. Mater. 9, 894 (2010). [27] Ken-ichi Uchida, Hiroto Adachi, Takeru Ota, Hiroyasu Nakayama, Sadamichi Maekawa, and Eiji Saitoh, Appl. Phys. Lett. 97, 172505 (2010). ∗ These authors made equal contributions to this paper [28] C. M. Jaworski, J. Yang, S. Mack, D. D. Awschalom, † Electronic address: [email protected] R. C. Myers, and J. P. Heremans Phys. Rev. Lett. 106, ‡ Electronic address: [email protected] 186601 (2011). [1] W. Heisenberg, Z. Phys. 49, 619 (1928). [29] A. Hoffmann and S. D. Bader. Opportunities at the Fron- [2] S. W. Lovesey, Theory of Thermal Neutron Scattering tiers of Spintronics. Phys. Rev, Appl. 5, 047001 (2015). from Condensed Matter (Clarendon, Oxford, 1984), Vol. [30] Z. Qiu, K. Ando, K. Uchida, Y. Kajiwara, R. Takahashi, 2, Chap. 9. H. Nakayama, T. An, Y. Fujikawa, and E. Saitoh, Appl. [3] L. D. Landau, Sov. Phys. JETP 3, 920 (1957). Phys. Lett. 103, 092404 (2013). [4] M. E. Zhitomirsky and A. L. Chernyshev, Rev. Mod. [31] T. Kikkawa, K. Shen, B. Flebus, R. A. Duine, Ken-ichi Phys. 85, 219 (2013). Uchida, Z. Qiu, G. E.W. Bauer, and E. Saitoh, Phys. [5] Jun Zhao, D. T. Adroja, Dao-Xin Yao, R. Bewley, Shil- Rev. Lett. 117, 207203 (2016). iang Li, X. F. Wang, G. Wu, X. H. Chen, Jiangping Hu, [32] For additional data and analysis, see supplementary in- and Pengcheng Dai, Nat. Phys. 5, 555 (2009). formation. [6] A. Kamra, H. Keshtgar, P. Yan, and G. E. W. Bauer, [33] Z. Jiang, C.-Z. Chang, M. R. Masir, C. Tang, Y. Xu, J. S. Phys. Rev. B 91, 104409 (2015). Moodera, A. H. MacDonald, and J. Shi, Nat. Commun. [7] K. Shen and G. E. W. Bauer, Phys. Rev. Lett. 115, 7, 11458 (2016). 197201 (2015). [34] Y. D. Xu, B. Yang, C. Tang, Z. Jiang, M. Schneider, R. [8] E. Anda, J. Phys. C: Solid State Phys. 9, 1075 (1976). Whig, and J. Shi, Appl. Phys. Lett. 105, 242404 (2014). [9] S. C. Guerreiro and S. M. Rezende, Phys. Rev. B 92, [35] S. M. Wu, F. Y. Fradin, J. Hoffman, A. Hoffmann, and 214437 (2015). A. Bhattacharya, J. Appl. Phys. 117, 17C509 (2015). [10] Joosung Oh, Manh Duc Le, Ho-Hyun Nahm, Hasung [36] M. Collet, L. Soumah, P. Bortolotti, M. Muoz, V. Cros, Sim, Jaehong Jeong, T.G. Perring, Hyungje Woo, Kenji and A. Anane, AIP Adv. 7, 055924 (2017). Nakajima, Seiko Ohira-Kawamura, Zahra Yamani, Y. [37] Pengcheng Dai, Rev. Mod. Phys. 87, 855 (2015). Yoshida, H. Eisaki, S.-W. Cheong, A.L. Chernyshev, and [38] F. Ye, P. C. Dai, J. A. Fernandez-Baca, D. T. Adroja, Je-Geun Park, Nat. Comm. 7, 13146 (2016). T. G. Perring, Y. Tomioka, and Y. Tokura, Phys. Rev. [11] Pengcheng Dai, H. Y. Hwang, Jiandi Zhang, J. A. B 75, 144408 (2007). Fernandez-Baca, S.-W. Cheong, C. Kloc, Y. Tomioka, [39] S. Toth and B. Lake, J. Phys.: Condens. Matt. 27, and Y. Tokura Phys. Rev. B 61, 9553 (2000). 166002 (2015). [12] S. Geller and M. A. Gilleo, Acta Cryst. 10, 239 (1957). [40] B. Flebus, K. Shen, T. Kikkawa, K. Uchida, Z. Y. Qiu., [13] J. S. Plant, J. Phys. C: Solid State Phys. 10, 4805 (1977). E. Saitoh, R. A. Duine, and G. E. W. Bauer, Phys. Rev. [14] V. Cherepanov, I. Kolokolov and V. L’Vov. Phys. Reps. B 95, 144420 (2017). 229, 81 (1993). [41] L. J. Cornelissen, K. Oyanagi, T. Kikkawa, Z. Qiu, T. [15] A. B. Harris, Phys. Rev. 132, 2398 (1963). Kuschel, G. E. W. Bauer, B. J. van Wees, and E. Saitoh, [16] C. M. Srivastava and R. Aiyar, J. Phys. C: Solid State arXiv: 1706.04373v1.