# Thermal gradient driven domain wall dynamics

###### Abstract

The issue of whether a thermal gradient acts like a magnetic field or an electric current in the domain wall (DW) dynamics is investigated. Broadly speaking, magnetization control knobs can be classified as energy-driving or angular-momentum driving forces. DW propagation driven by a static magnetic field is the best known example of the former in which the DW speed is proportional to the energy dissipation rate, and the current-driven DW motion is an example of the latter. Here we show that DW propagation speed driven by a thermal gradient can be fully explained as the angular momentum transfer between thermally generated spin current and DW. We found DW-plane rotation speed increases as DW width decreases. Both DW propagation speed along the wire and DW-plane rotation speed around the wire decrease with the Gilbert damping. These facts are consistent with the angular momentum transfer mechanism, but are distinct from the energy dissipation mechanism. We further show that magnonic spin-transfer torque (STT) generated by a thermal gradient has both damping-like and field-like components. By analyzing DW propagation speed and DW-plane rotation speed, the coefficient () of the field-like STT arising from the non-adiabatic process, is obtained. It is found that does not depend on the thermal gradient; increases with uniaxial anisotropy (thinner DW); and decreases with the damping, in agreement with the physical picture that a larger damping or a thicker DW leads to a better alignment between the spin-current polarization and the local magnetization, or a better adiabaticity.

## I Introduction

Manipulating domain walls (DW) in magnetic nanostructures has attracted much attention because of its potential applications in data storage technology parkin () and logic gates DA2005 (). The traditional DW control knobs, namely magnetic fields and spin-polarized currents, have certain drawbacks in applications. In the magnetic-field-driven DW motion, energy dissipation is the main cause of DW propagation whose speed is proportional to the energy dissipation rate xrw1 (); Fe2009 (), and the magnetic field tends to destroy unfavorable domains and DWs, instead of driving a series of DWs synchronously DA2003 (); GS2005 (); MH2006 (). An electrical current drives a DW to move mainly through the angular momentum transfer so that it pushes multiple DWs LB1996 (); Slonczewski (); SZ2004 (); GT2004 () in the same direction. To achieve a useful DW speed, it requires high electrical current densities that result in a Joule heating problem Jole2 (); Expecom1 (); Expecom2 (). To avoid these problems, thermal gradient has been proposed as an alternative DW control parameter spincal () since the thermal gradient is capable of inducing speedy DW motion Jiang13 () with low energy consumption MagSTT (). Unlike an electric current that is only applicable to metallic systems, the thermal gradient can be used in both magnetic metals and magnetic insulators. In addition, the thermal gradient generated spin current might be a promising means to harvest the dissipated heat of electric circuits spincal (); Nat.com2017 ().

To understand the mechanism behind thermal-gradient-driven DW dynamics, there are microscopical theories XSW11 (); NOWAK11 (); KOVALEV12 (); XGWANG12 (); PYAN2015 () and macroscopic thermodynamic theories XSW14 (); NOWAK14 (). Briefly speaking, the microscopic theories suggest that magnons populated in the hotter region diffuses to the colder region to form a magnon spin current. The magnon spin current passes through a DW and exerts a torque on the DW by transferring spin angular momentum to the DW. Thus, magnons drive the DW propagating toward the hotter region of the nanowire, opposite to the magnon current direction MagSTT (); XSW11 (); NOWAK11 (). The thermodynamic theories anticipate that a thermal gradient generates an entropy force which always drives the DW towards the hotter region in order to minimize the system free energy. The macroscopic theories do not provide any microscopic picture about DW dynamics although a thermal gradient is often considered as an effective magnetic field to estimate DW speed XSW14 (); NOWAK14 () from field-driven DW theories. Thus, one interesting issue is whether a thermal gradient in DW dynamics acts like a magnetic field or an electric current. DW propagation speed should be sensitive to both DW width and types of a DW (transverse DW) under an energy-driving force while the speed should be insensitive to the DW and DW structure in the angular-momentum-driving force. This is the focus of the current work.

In this paper, we investigate DW motion along a uniaxial wire with the easy axis along the wire direction under a thermal gradient. We found that the DW always propagates to the hotter region with an accompanied DW-plane rotation. DW propagation speed and DW-plane rotation speed increases as the magnetic easy-axis anisotropy and damping decreases. We show that DW motion can be attributed to the angular momentum transfer between magnonic spin current and the DW. Thus we conclude that a thermal gradient interacts with DW through angular-momentum transfer rather than through energy dissipation. Similar to an electric current Nonad (), a thermal gradient can generate both damping-like (or adiabatic) STT and field-like (or non-adiabatic) STT. From the damping-dependence and anisotropy-dependence of the average DW velocity and DW-plane rotation angular velocity, we extract field-like STT coefficient (). It is found that is independent of thermal gradient; is bigger for a thinner DW; and decreases with the damping coefficient. We also show that in the presence of a weak hard-axis anisotropy perpendicular to the wire, the DW still undergoes a rotating motion. The DW propagation speed increases slightly while the DW-plane rotation speed decreases with the strength of the hard-axis anisotropy.

## Ii Model and method

We consider a uniaxial nanowire of length and cross-section along the -axis (easy axis) with a head-to-head DW at the center, as shown in Fig. 1. , is much smaller than the DW width , and is much smaller than . A thermal gradient is applied along the wire. The highest temperature is far below the Curie temperature . It is assumed that the saturation magnetization and exchange constant are temperature-independent. The magnetization dynamics is governed by the stochastic Landau-Lifshitz-Gilbert (LLG) equation Brown1963 (); TLGIL (),

(1) |

where and are respectively the magnetization direction and the saturation magnetization. is the Gilbert damping constant and is measured in the units of in which is the gyromagnetic ratio. is the effective field measured in the units of , where is the exchange constant, () denote Cartesian coordinates , , , is the easy-axis anisotropy, and is the dipolar field. is the thermal stochastic field.

The stochastic LLG equation is solved numerically by MUMAX3 package Mumax () in which we use adaptive Heun solver. The time step is s when the cell size is () and s for unit cells smaller than () . The saturation magnetization and exchange constant are used to mimic permalloy in our simulations. The thermal field follows the Gaussian process characterized by following statistics

(2) |

where and denote the micromagnetic cells, and , represent the Cartesian components of the thermal field. and are respectively temperature and the Gilbert damping at cell , and is the cell size. is the Boltzmann constant Brown1963 (). The numerical results presented in this study are averaged over 15 ensembles (for simulated DW speed and DW-plane rotation speed) and 4000-5000 ensembles (for spin current calculation).

Under the thermal gradient , magnetization at different positions deviate from their equilibrium directions differently and small transverse components and are generated. The transverse components vary spatial-temporally and depend on the local temperature. This variation generates a magnonic spin current NOWAK11 (). This magnonic spin current can interact with spin textures including DWs. Neglecting the small demagnetic and stochastic thermal fields, the spin current along the direction can be defined from the spin continuity equation derived from Eq. (1) as follows,

(3) |

where

(4) |

is the total spin current density along -direction due to both exchange interaction and Gilbert damping where length is in the units of . The spin current components along , and due to the exchange interaction are , and . The first term of the Eq. (4), , dominates the total spin current. The second term due to the damping effect contributes a small negative value to total spin current density. The total spin current , expressed in terms of , and , can be directly computed from simulations.

Integrating the component of Eq. (3) over a space enclosed the DW in the center and noticing the absence of the first term on the right, we can relate the DW velocity with as

(5) |

where the velocity unit of is added and , mean the total spin current on the left and right sides of the DW. The equation clearly shows that the DW propagates opposite to the spin current. This is the theoretical DW velocity under the assumption of angular momentum conservation, and it will be compared with the directly simulated DW velocity below.

## Iii Results

### iii.1 Average spin current and DW velocity

To substantiate our assertion that DW propagation under a thermal gradient is through angular-momentum effect instead of energy effect, we would like to compare the DW velocity obtained from micromagnetic simulations and that obtained from total spin current based on Eq. (5). Eq. (4) is used to calculate . Fig. 2(a) is spatial distribution of the ensemble averaged with DW at for various thermal gradients. The sudden sign change of at the DW center is a clear evidence of strong angular-momentum transfer from spin current to the DW. Technically, magnetization of the two domains separated by the DW point to the opposite directions, thus the spin current polarization changes its sign. In calculating DW velocity from Eq. (5), the spin currents before entering DW and after passing DW are the averages of over and , where is the DW width which is 16 nm in the current case. The thermal gradient dependence of is shown in Fig. 2(b) (solid squares). compares well with the velocity (open squares) obtained directly from simulations by extracting the speed of the DW center along -direction (the small difference can be attributed to the contribution from the demagnetic and stochastic fields).

It is noted that almost coincides with except a small discrepancy at very high thermal gradient when the nonlinear effects is strong. These observations are consistent with magnonic STT XSW11 (); NOWAK11 (); XGWANG12 (); PYAN2015 (). It is observed that the DW-plane rotates around the -axis counter-clockwise for head-to-head DW and clockwise for tail-to-tail DW during DW propagation. DW rotation speed (squares) is shown in Fig. 2 (c)) as a function of .

### iii.2 Damping and anisotropy dependence of DW dynamics

An energy-effect and angular-momentum-effect have different damping-dependence and anisotropy-dependence of DW dynamics. To distinguish the roles of energy and the angular-momentum in thermal-gradient driven DW dynamics, it would be useful to probe how the DW dynamics depends on and . Damping have two effects on the spin currents: one is the decay of spin current during its propagation so that the amount of spin angular momentum deposited on a DW should decrease with the increase of the damping coefficient. As a result, the DW propagation speed and DW-plane rotation speed should also be smaller for a larger . Indeed, this is what we observed in our simulations as shown in Fig. 3(a) for DW speed and DW-plane rotation speed (open squares for , solid circles for , and stars for ). The model parameters are (), nm, K/nm and . The second damping effect is that the larger helps the spin current polarization to align with the local spin. This second effect enhances the adiabatic process that is important for non-adiabatic STT or field-like torque discussed in the next subsection. Therefore, dependence of DW dynamics supports the origin of thermal driven DW dynamics to be the angular-momentum effect, not the energy effect that would lead to a larger and for a larger xrw1 (); Fe2009 (); RW2010 (); Euro (); AM2007 () instead of a decrease observed here.

Here we would like to see how the DW dynamics depends on uniaxial anisotropy . Fig. 4 shows both (open squares), (filled squares) and (circles) for = 2048 nm, and . The DW propagation speed, decreases with while DW-plane rotational speed increases with . These results seem follow partially the behavior of magnetic-field induced DW motion, in which DW propagation speed is proportional to DW width () or decrease with , and partially electric current driven DW motion, in which DW-plane rotational speed increases with . Thus, one may tend to conclude that a thermal gradient behaves more like a magnetic field rather than an electric current from the DW width dependence of DW propagation speed, opposite to our claim of the angular-momentum effects of the thermal gradient. It turns out, this is not true. The reason is that magnon spectrum, , has a gap in a system with magnetic anisotropy. The larger is, the bigger the energy gap will be. Thus, it becomes harder to thermally excite magnon. As a result, the spin current decreases as increases. To see whether the thermal-gradient driven DW motion is due to the angular-momentum transfer or not, one should compare whether and maintain a good agreement with each other as varies. Indeed, a good agreement between and is shown in Fig. 4. This conclusion is also consistent with existing magnonic STT theories RW2010 (); Euro (); AM2007 ().

### iii.3 Separation of adiabatic and non-adiabatic torques

We have already demonstrated that a thermal gradient interacts with DW through magnonic STT rather than through energy dissipation. It is then interesting to know what kind of STTs a thermal gradient can generate. Specifically, whether a magnonic spin current generates damping-like (adiabatic), or field-like (Non-adiabatic) torques, or both just like an electric current Nonad () does. To extract the STT generated from a thermal gradient, we approximate DW dynamics by the motion of its collective modes of DW center and the titled angle of DW-plane. Subject to both damping-like and field-like torques, using the travelling-wave ansatz RW2010 (); Euro (); AM2007 (), where , one can derive the equations for X and ,

(6) |

From the above two equations, one can straightforwardly find DW propagating speed and DW-plane rotation speed,

(7) |

One can extract and equivalent electric current density from and obtained in simulations. For , , the and are obtained and plotted in Fig. 5 as a function of . It is evident that linearly increases with and is independent of as it should be. We then fixed K/nm, and repeat simulations and analysis mentioned above for various and . Fig. 6 (a) and (b) shows as a function of and . From the figure, it is evident that decreases with . This is because the larger damping favors the alignment of spin current polarization with the local spin so that the non-adiabatic effect, , becomes smaller. increases with for the similar reason: Larger means a thinner DW so that it is much harder for the spin current polarization to reverse its direction after passing through the thinner DW, i.e. a stronger non-adiabatic effect.

## Iv Summary and discussion

We have studied the thermal gradient-driven DW dynamics in an uniaxial nanowire. In reality, there is always certain hard anisotropy in a wire whose cross-section is not a perfect ellipse. Thus, it is interesting to see how the above results will change in a weak biaxial nanowire with a small hard anisotropy , say along -direction. Our simulations show that a DW still propagates towards the higher temperature region in a similar way as that in a uniaxial wire. Interestingly, as shown in Fig. 2(d) for the -dependence of (filled squares) and (open squares), DW speed increases slightly with . This may be due to the increase of torque along -direction AM2007 () since is proportional to . This is also consistent with the early results for the uniaxial wire that (which includes stochastic thermal field and demagnetic fields) is always larger than (where the transverse fields are neglected). At the meanwhile, decreases with .

In summary, our results show that the thermal gradient always drives a DW propagating towards the hotter region and the DW-plane rotates around the easy axis. DW velocity and DW-plane rotation speed decrease with the damping coefficient. The DW velocity obtained from simulations agrees with the velocity obtained from angular momentum conservation when the magnon current density ) from the simulation is used to estimate the amount of angular momentum transferred from magnon current to the DW. All the above findings lead to the conclusion that the thermal gradient interacts with DW through angular-momentum transfer rather than through energy dissipation. Furthermore, we demonstrated that the magnonic STT generated by a thermal gradient has both damping-like and field-like components. The field-like STT coefficient is determined from DW speed and DW-plane rotation speed. does not depend on the thermal gradient as expected, but increases with a decrease of DW width. This behavior can be understood from the expected strong misalignment of magnon spin polarization and the local spin so that non-adiabatic torque (also called field-like torque) is larger. For the same reason, a larger Gilbert damping results in a better alignment between spin current polarization and the local spin, thus should decrease with . Since typical magnon energy consumption is millions times smaller than that by electrons, the thermal gradient can be a very interesting control knob for nano spintronics devices MagSTT (), especially those made from magnetic insulators.

This work was supported by the National Natural Science Foundation of China (Grant No. 11774296) as well as Hong Kong RGC Grants No. 16300117 and No. 16301816. X.S.W acknowledges support from NSFC (Grant No. 11804045) and China Postdoctoral Science Foundation (Grant No. 2017M612932 and 2018T110957). M. T. I acknowledges the Hong Kong PhD fellowship.

## References

- (1) S. S. P. Parkin, M. Hayashi, and L. Thomas, Science 320, 190 (2008).
- (2) D. A. Allwood, G. Xiong, C. C. Faulkner, D. Atkinson, D. Petit, and R.P. Cowburn, Science 309, 1688 (2005).
- (3) X. R. Wang, P. Yan, J. Lu, and C. He, Ann. Phys. (N. Y.) 324, 1815 (2009).
- (4) X. R. Wang, P. Yan, and J. Lu, Europhys. Lett. 86, 67001 (2009).
- (5) D. Atkinson, D. A. Allwood, G. Xiong, M. D. Cooke, C. C. Faulkner, and R. P. Cowburn, Nat. Mater. 2, 85 (2003).
- (6) G. S. D. Beach,C.Nistor, C. Knutson, M. Tsoi, and J. L. Erskine, Nat. Mater. 4, 741 (2005).
- (7) M. Hayashi, L. Thomas, Ya. B. Bazaliy, C. Rettner, R. Moriya, X. Jiang, and S. S. P. Parkin, Phys. Rev. Lett. 96, 197207 (2006).
- (8) L. Berger, Phys. Rev. B 54, 9353 (1996).
- (9) J. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996).
- (10) S. Zhang and Z. Li, Phys. Rev. Lett. 93, 127204 (2004).
- (11) G. Tatara and H. Kohno, Phys. Rev. Lett. 92, 086601 (2004).
- (12) A. Yamaguchi, T. Ono, S. Nasu, K. Miyake, K. Mibu, and T. Shinjo, Phys. Rev. Lett. 92, 077205 (2004).
- (13) M. Hayashi, L. Thomas, Y.B . Bazaliy, C. Rettner, R. Moriya, X. Jiang, and S. S. P. Parkin, Phys. Rev. Lett. 96, 197207 (2006).
- (14) A. Yamaguchi, A. Hirohata, T. Ono, and H. Miyajima, J. Phys. Condens. Matter 24, 024201 (2012).
- (15) G. E. W. Bauer, E. Saitoh, and B. J. van Wees, Nat. Mater. 11, 391 (2012).
- (16) W. Jiang, P. Upadhyaya, Y. B. Fan, J. Zhao, M. S. Wang, L. T.Chang, M. R. Lang, K. L. Wong, M. Lewis, Y. T. Lin, J. S. Tang, S. Cherepov, X. Z. Zhou, Y. Tserkovnyak, R. N. Schwartz, and K. L. Wang, Phys. Rev. Lett. 110, 177202 (2013).
- (17) X. R. Wang, P. Yan, and X. S. Wang, IEEE Trans. Magn. 48, 11 (2012).
- (18) C. Safranski, I. Barsukov, H. K. Lee, T. Schneider, A. Jara, A. Smith, H. Chang, K. Lenz, J. Lindner,Y. Tserkovnyak, M. Wu, and I. Krivorotov, Nat. Com. 8, 117 (2017).
- (19) P. Yan, X. S. Wang, and X. R. Wang, Phys. Rev. Lett. 107, 177207 (2011).
- (20) A. A. Kovalev and Y. Tserkovnyak, Europhys. Lett. 97, 67002(2012).
- (21) D. Hinzke and U. Nowak, Phys. Rev. Lett. 107, 027205 (2011).
- (22) X. G. Wang, G. H. Guo, Y. Z. Nie, G. F. Zhang, and Z. X. Li, Phys. Rev. B 86, 054445 (2012).
- (23) P. Yan, Y. Cao, and J. Sinova, Phys. Rev. B 92,100408 (2015).
- (24) F. Schlickeiser, U. Ritzmann, D. Hinzke, and U. Nowak, Phys. Rev. Lett. 113, 097201 (2014).
- (25) X. S. Wang and X. R. Wang, Phys. Rev. B 90, 014414 (2014).
- (26) Jun-ichiro Kishine and A. S. Ovchinnikov Phys. Rev. B 81, 134405 (2010).
- (27) W. F. Brown, Phys. Rev. 130, 1677 (1963).
- (28) T. L. Gilbert, IEEE Trans. Magn. 40, 3443 (2004).
- (29) A. Vansteenkiste, J. Leliaert, M. Dvornik,M. Helsen, F. Garcia-Sanchez and B. Van Waeyenberge, AIP Advances 4, 107133 (2014).
- (30) S. R. Etesami, L. Chotorlishvili, A. Sukhov, and J. Berakdar, Phys. Rev. B 90, 014410 (2014).
- (31) A. Mougin, M. Cormier, J. P. Adam1, P. J. Metaxas and J. Ferre, EPL 78, 57007 (2007).
- (32) R. Wieser, E. Y.Vedmedenko, P. Weinberger, and R. Wiesendanger, Phys. Rev. B 82, 144430(2010).
- (33) Z. Z. Sun, J. Schliemann, P.Yan, and X.R.Wang, Eur. Phys. J. B 79, 449â453 (2011).