betaFeedback

WorksPhysics and Astronomy

Nonmonotonic Magnetic Friction from Collective Rotor Dynamics

Abstract

Amontons' law postulates a monotonic relationship between frictional force and the normal load applied to a sliding contact. This empirical rule, however, fails in systems where internal degrees of freedom - such as structural or electronic order - play a central role. Here, we demonstrate that friction can emerge entirely from magnetically driven configurational dynamics in the absence of physical contact. Using a two-dimensional array of rotatable magnetic dipoles sliding over a commensurate magnetic substrate, we observe a pronounced non-monotonic dependence of friction on the interlayer separation, and thus on the effective load. The friction peaks at an intermediate distance where competing ferromagnetic and antiferromagnetic interactions induce dynamical frustration and hysteretic torque cycles during sliding. Molecular dynamics simulations and a simplified two-sublattice model confirm that energy dissipation is governed by collective magnetic reorientations and their hysteresis. Our results establish the occurrence of sliding-induced changes in collective magnetic order, which has a strong impact on friction, and thus open new possibilities for contactless friction control, magnetic sensing, and the design of reconfigurable, wear-free frictional interfaces and metamaterials.

Introduction

Amontons’ law, a time-honored and foundational empirical rule in tribology, posits a monotonic relationship between frictional force and normal load1. While widely applicable, this principle offers limited insight into the microscopic mechanisms underlying friction. Recent studies have revealed significant deviations from this law in systems where internal degrees of freedom — such as structural, electronic, or magnetic order — play a critical role2–7. These exceptions, often linked to phase transitions or symmetry breaking at sliding interfaces8–10, underscore the potential of friction as a sensitive probe of interfacial dynamics. Magnetic contributions to friction have gained growing attention, particularly in contactless systems where dissipation is governed by configurational dynamics rather than mechanical abrasion11–13. Previous theoretical and numerical work has established a connection between magnetic ordering and friction14–27, further supported by experiments using magnetic atomic force microscopy11,28. However, despite the atomic force microscopy’s ability to detect magnetic interactions, it lacks the spatial resolution and sensitivity to fully resolve dynamic excitations at a microscopic spin level that govern friction at the nanoscale. Consequently, direct experimental evidence for the connection between spin dynamics and sliding friction is lacking.

In this study, we overcome this limitation by employing a two-dimensional array of rotatable magnetic dipoles that interact with a commensurate magnetic substrate. This macroscopic yet spatially resolved system allows us to directly track changes of collective magnetic configurations during sliding and correlate them with frictional behavior. Strikingly, we observe a non-monotonic dependence of friction on interlayer separation, with maximum dissipation occurring in a regime of frustrated magnetic alignment between ferromagnetic and antiferromagnetic order. This anomalous peak reveals a frictional mechanism rooted in sliding-induced magnetic hysteresis, absent of any mechanical contact. Our results demonstrate that magnetic friction can arise purely from internal reorientation dynamics and that frictional forces can serve as a sensitive and direct measure of magnetic order — accessible beyond the limitations of conventional techniques. These findings open pathways for designing tunable, wear-free frictional metamaterials29–31 and advancing magnetic sensing technologies based on configurational control32.

Experimental setup

We investigate the friction between two parallel layers in the xyxy-plane, each consisting of periodically arranged, millimeter-sized magnets in a square lattice with the same lattice constant PP. The top layer (slider) comprises 7×77 \times7 neodymium-iron-boron (NdFeB) ring magnets with an outer diameter of 4 mm4\ \mathrm{mm}, an inner diameter of 1.2 mm1.2\ \mathrm{mm}, and a height of 4 mm4\ \mathrm{mm} (Figure 1a). Each magnet is mounted on a non-magnetic metal axis, allowing its magnetic moment m⃗ij\vec{m}_{ij} to freely rotate within the xzxz-plane, forming an angle θij\theta_{ij} relative to the xx-axis (Figure 1a). Here, the indices ii and jj denote the position of each rotor within the lattice. The entire rotor array is mounted on an aluminum frame attached to a three-axis resistive force sensor with a relative force resolution of approximately 0.02 N0.02\ \mathrm{N}. More details of the force sensor are given in Sec. S1 of the Supporting Information (SI).

Image not extracted

Figure 1. Experimental setup, total friction, and order parameter. (a) Magnetic rotor unit design: Ring-shaped NdFeB magnet mounted on a metal shaft enabling rotation around the yy-axis with the orientation of the magnetic moment characterised by the angle θ\theta. (b) Illustration of the two interacting magnetic layers at separation (height) hh with square lattice geometry (lattice constant P=16mmP = 16\mathrm{mm}). The top layer (7x7 magnets) is composed of magnetic rotors while the magnetic moments in the bottom layer (16x9 magnets) are fixed in yy-direction. The relative horizontal displacement between the layers during sliding is characterized by Δx\Delta x. (c) Mechanical structure for the sliding friction experiments: The setup includes a motorized stage, brass rollers for maintaining consistent separation, and a force sensor for direct measurement of the friction. The orientations of the rotors are directly observed using an overhead camera. (d) Total lateral friction ⟨Fx⟩\langle F_x\rangle as a function of the height hh of the slider. The magnitude −⟨Fx⟩-\langle F_x\rangle is presented, since ⟨Fx⟩\langle F_x\rangle is negative. (e) Averaged orientation correlation order parameter ⟨Σ⟩\langle\Sigma\rangle shown as a function of height hh.

The bottom layer (substrate) is positioned parallel to the slider and consists of a 16×916 \times9 square lattice of cylindrical magnets. These magnets, made of the same NdFeB material, have a height and diameter of 4 mm4\ \mathrm{mm} (Figure 1b). Unlike the slider, the magnetic moments of the substrate magnets are fixed and oriented along the xx-axis. The slider is translated relative to the substrate along the xx-direction using a motorized stage at a constant velocity of v=8 mm/sv = 8\ \mathrm{mm/s}, ensuring quasi-static conditions where the magnetic moments in the top layer effectively remain in equilibrium at each position during the sliding process.

To measure the orientations θij\theta_{ij} of the rotors, the magnets are labeled with colored markers, which are tracked using a color-sensitive overhead camera (Figure 1c). The vertical separation hh between the layers (height of the slider) is maintained using brass rollers mounted on the metal frame of the slider, which compensate for the magnetic attraction between the layers. Variations in hh are achieved by inserting spacers of different thicknesses between the rollers and the substrate. At the start of each experiment, the two layers are aligned parallel, with their magnetic lattices perfectly superimposed at the initial position. The slider is then translated forward and backward by seven lattice periods (Δxmax⁡=7×16 mm\Delta x_{\max} = 7 \times16\ \mathrm{mm}) along the xx-direction. During the translation, we record the xx-component of the total force FxF_x exerted on the slider using the force sensor, as well as the orientations θij\theta_{ij} of the rotors. A typical example measurement of FxF_x and a snapshot of the rotor orientations θij\theta_{ij} during sliding are shown in Figure 1c. Additional technical details of the setup components are provided in the Methods section.

Total friction and average magnetic order

fig:1d shows how the measured mean sliding force ⟨Fx⟩\langle F_x\rangle (the brackets correspond to the average over an integer number of lattice spacings PP) varies as a function of the slider height hh. This quantity is the total friction corresponding to the sliding process. Note that, while qualitatively consistent, the measured friction can strongly depend on small vertical misalignments. More information on its accuracy is given in Sec. S2.

Although the magnetic interaction, i.e. the normal load between the two layers monotonically decreases with increasing hh, ⟨Fx⟩\langle F_x\rangle exhibits a clear peak around h0=9.0h_0 = 9.0 mm. This is in strong contradiction with Amonton's empirical first law, which claims that sliding friction is proportional to the load33. The effective load due to magnetic attraction and the resulting differential friction coefficient for the setup are presented in Sec. S3.

The key to understanding this unusual friction behavior lies in analyzing how the magnetic order within the top layer evolves during the sliding process, as theoretical predictions suggest that changes in magnetization and magnetic excitations can couple to magnetic interaction-based friction — and vice versa[34, 35] [17, 19, 21, 22]. To characterise the “relative” orientations of neighboring magnetic moments, we introduce the displacement-dependent order parameter

Σ(Δx)=1Z∑i=17∑j=16cos⁡(θij−θij+1)(1)\Sigma(\Delta x) = \frac{1}{Z} \sum_{i=1}^{7} \sum_{j=1}^{6} \cos(\theta_{ij} - \theta_{ij+1}) \tag*{(1)}

where Z=7×6Z = 7 \times6 is a normalization factor. Accordingly, Σ\Sigma varies between +1 and -1. Here, +1 corresponds to a state in which all magnetic moments are aligned parallel to each other, representing a ferromagnetic (FM) order, whereas -1 corresponds to a state where neighboring moments regarding the yy-direction are aligned anti-parallel, indicating an antiferromagnetic (AFM) order. When plotting the mean value ⟨Σ⟩\langle\Sigma\rangle of the order parameter (averaged over an integer number of lattice spacings) vs. the layer distance hh, we find a gradual change from FM to AFM ordering. Notably, ⟨Σ⟩=0\langle\Sigma\rangle= 0 (i.e., no average magnetic order) is located exactly at that distance where the peak in the friction force is observed (Fig. 1e). Similar friction peaks near structural rearrangements have been theoretically predicted8 and experimentally measured[36] [9] for materials externally heated close to a phase transition. This suggests a strong connection between the frictional behavior and changes of magnetic ordering, which will be discussed in the following.

Competing intra- and interlayer magnetic interactions and layer interaction force

Based on the hh-dependent mean magnetic order parameter, we categorize the system into three distinct regimes with (i) ferromagnetic (FM) ⟨Σ⟩≥0.95\langle\Sigma\rangle\ge0.95, (ii) competing (CP) −0.70<⟨Σ⟩<0.95-0.70 < \langle\Sigma\rangle< 0.95 and (iii) antiferromagnetic (AFM) ⟨Σ⟩≤−0.70\langle\Sigma\rangle\le-0.70 ordering, respectively. Figure 2a compares how the relative orientation of magnetic moments change during the sliding in the different regimes. To illustrate the differences in more detail, we discuss the orientational changes of two neighboring moments marked in blue and red located at (i,j)=(4,3)(i,j) = (4,3) and (4,4)(4,4), during the sliding process, respectively (Fig. 2b).

For small hh, the magnetic field of the bottom layer is the dominant influence on the rotors, resulting in their parallel alignment. Since all rotors are exposed to the same magnetic field due to the commensurate conditions, this eventually leads to collective rotation during sliding of the rotors and thus to FM ordering. In contrast, at large hh, the influence of the bottom layer weakens and the interactions between neighboring rotors prevail. This leads to AFM order with an alternating rotor orientation along the yy-axis while being parallel along the xx-axis (fig:2a, right column). During sliding, only a small collective “wiggling” of the rotors is observed (fig:2b, right column). At intermediate distances, where magnetic inter- and intra-layer interactions are comparable, the rotors periodically alternate between FM and AFM order in a rather discontinuous fashion (middle column in Figure 2b). The dynamics of the moments corresponding to the three different regimes are presented in Video 1.

Image not extracted

Figure 2. Magnetic moment configurations and dynamic responses at different heights hh. The orientation angle θij\theta_{ij} of two example moments are depicted as a function of the displacement Δx\Delta x for three selected heights h=6.5h = 6.5 mm (left), 8.5 mm (middle), and 11 mm (right) representing the different dynamical states (FM, CP, AFM). (a) Experimental snapshots of the 7×77 \times7 rotor array configurations at different displacements Δx\Delta x. The local magnetic field acted by the substrate on the moments is indicated above each snapshot. (b, c) Experimental and simulation results for the tagged θij\theta_{ij} showing the displacement-dependent dynamic

To corroborate our experiments, we additionally performed Molecular Dynamics (MD) simulations, where the magnets are modelled as point dipoles with their configuration and rotational degrees of freedom exactly matching the conditions in our experiments. The magnetic dipole strength has been determined from the magnetization of NdFeB provided by the manufacturer. To allow rapid relaxation of the dipoles and to maintain quasi-equilibrium conditions during the sliding process, we added an orientational friction term to the equations describing the angular motion of the magnetic moments. For more details, see the Methods section.

response of two selected rotors from the array at each height. (d) Orientation correlation order parameter Σ\Sigma as a function of the Displacement Δx\Delta x.

Figure 2c shows the corresponding results of the simulations matching the experimental data in Figure 2b. In the FM and AFM regime, the dynamics of the experiments and the simulations are in good agreement. In the CP regime, the jumps between parallel and antiparallel configurations appear less periodic in the experiments, which is due to some small variations in the purchased magnets and small deviation in the structure dimensions limited by the printing resolution of the 3D printer used to manufacture the base of the slider (Methods). For a similar reason, the flipping of magnetic moments between parallel and antiparallel direction is not perfectly reproduced in the simulations because the system is highly unstable at the tipping point and thus susceptible to small perturbations. However, both the simulations and the experiments exhibit the dynamical transition between the FM and AFM ordering in the CP regime. A direct comparison between the experiments and the simulations are shown in Video 2.

In Figure 2d, we compare the displacement-dependent order parameter Σ(Δx)\Sigma(\Delta x) obtained from experiments and simulations. Consistent with above, the orientation correlation of magnetic dipoles hardly changes in the FM and AFM regime but strongly alternates in the CP regime. Such variations are also responsible for the small value of ⟨Σ⟩\langle\Sigma\rangle at intermediate distances (Figure 1e). The full height dependence of ⟨Σ⟩\langle\Sigma\rangle obtained from simulations is shown in Figure 1e as a solid line and fits excellently to the experimental data. The detailed dynamics of the magnetic moments during the transition between the three regimes is presented in Video 3.

Figure 3 shows the experimentally measured forces FxF_x (symbols) during the sliding process for the three different regimes with heights h=6.5 mmh = 6.5\ \mathrm{mm} (FM), 8.5 mm8.5\ \mathrm{mm} (CP), and h=11 mmh = 11\ \mathrm{mm} (AFM), respectively. Notably, FxF_x contains both, magnetic FxmagF_x^{\mathrm{mag}} and mechanical FxmechF_x^{\mathrm{mech}} friction forces, where the latter is mostly caused by the brass rollers compensating for the magnetic attraction (load) between the two layers. Since we are mainly interested in the magnetic friction, both contributions need to be separated, which is discussed in the following.

Image not extracted

Figure 3. Height-dependent magnetic interaction force between the two layers. Total force acting on the top layer as a function of the displacement Δx\Delta x obtained from the force sensor (symbols) and the simulations (lines). The left, middle, and right columns represent the FM (h=6.5 mmh = 6.5\ \mathrm{mm}), the CP (h=8.5 mmh = 8.5\ \mathrm{mm}), and the AFM regimes (h=11 mmh = 11\ \mathrm{mm}), respectively. The dashed lines are the mechanical contributions of the friction obtained from the simulations.

The mechanical part contains two contributions F1mechF^{\mathrm{mech}}_{1} and F2mechF^{\mathrm{mech}}_{2}. The first part is due to the magnetic attraction FzmagF^{\mathrm{mag}}_{z} between the layers leading to contact friction due to the brass rollers. Using Amonton's law, we approximate F1mech=μFzmagF^{\mathrm{mech}}_{1}=\mu F^{\mathrm{mag}}_{z} with the friction coefficient μ=0.09\mu=0.09 being experimentally determined by variation of the load between the layers (Sec. S4). In addition, another small load-independent source of friction exists which is likely caused by a small error in the initial zero-load conditions caused by vertical misalignment when calibrating the setup (see also Sec. S2). This can be seen for largest distances (i.e., h=11.0 mmh=11.0\,\mathrm{mm}) in fig:1c and Figure 3, where Fzmag≈0F^{\mathrm{mag}}_{z}\approx0 and therefore F1mech≈0F^{\mathrm{mech}}_{1}\approx0. The small deviation from ⟨Fx⟩=0\langle F_x\rangle=0 and Fx=0F_x=0 is here attributed to F2mech≈−0.34 NF^{\mathrm{mech}}_{2}\approx-0.34\,\mathrm{N}. More details on the mechanical friction model can be found in Sec. S4.

The solid lines in Figure 3 correspond to simulations where the above mechanical contributions have been taken into account. Despite the simple dipole approximation, the simulations are in excellent agreement with the experiments. In the FM regime, we observe large oscillations in Fx(Δx)F_x(\Delta x) due to the strong interaction between the slider and the bottom layer. Such fluctuations gradually decrease with decreasing magnetic interaction strength and essentially vanish for h=11 mmh=11\,\mathrm{mm}. Notably, in the FM and AFM case, FxF_x almost symmetrically fluctuates around the mechanical contribution FxmechF^{\mathrm{mech}}_x (marked by the dashed lines in Figure 3 and approximated from the simulation data). This suggests that the mean magnetic friction ⟨Fxmag⟩\langle F^{\mathrm{mag}}_x\rangle (averaged over an integer number of lattice periods) almost vanishes and only ⟨Fxmech⟩\langle F^{\mathrm{mech}}_x\rangle contributes to the total friction ⟨Fx⟩\langle F_x\rangle. This is in strong contrast to the CP regime where ⟨Fxmag⟩\langle F^{\mathrm{mag}}_x\rangle clearly deviates from zero, which results in large magnetic friction.

Non-monotonic magnetic friction and its relationship to hysteresis

Sum of torques TysubT_{y}^{\mathrm{sub}} acted by the substrate field on the top layer obtained from the simplified model. (e) Sim of torques TysubT_{y}^{\mathrm{sub}} plotted over the angle Θ\Theta with ∣M⃗∣cos⁡Θ=x^⋅M⃗\lvert\vec{M}\rvert\cos\Theta=\hat{x}\cdot\vec{M} between the movement direction x^\hat{x} and the magnetization M⃗\vec{M} of the top layer (see inset) to highlight the hysteresis. To depict the correct area that describes the dissipated energy, TysubT_{y}^{\mathrm{sub}} is scaled with the sign β\beta, which is positive when Θ(Δx)\Theta(\Delta x) decreases and negative when Θ(Δx)\Theta(\Delta x) increases (see Sec. S11). (f) Magnetic friction ⟨Fxmag⟩\langle F_{x}^{\mathrm{mag}}\rangle and average order parameter ⟨Δ⟩\langle\Delta\rangle calculated from the simplified model.

Subtracting the mechanical friction from the sensor measurements allows us now to isolate the magnetic friction force. Figure 4a shows the experimental (symbols) and numerical results (lines) of ⟨Fxmag⟩\langle F_{x}^{\mathrm{mag}}\rangle as a function of hh. Although the magnetic attraction, i.e. the load on the bottom layer, continuously decreases with increasing distance hh (see Sec. S3), the friction displays a maximum at h0≈9.0 mmh_{0}\approx9.0\ \mathrm{mm} (CP regime). Such behaviour is in strong contrast to the monotonic load-dependence of mechanical friction (Amaton’s law, dashed line). The magnetic friction peak can also be successfully reproduced by reconstructing the experimental friction using the measured orientations θij(Δx)\theta_{ij}(\Delta x) and calculating ⟨Fxmag⟩\langle F_{x}^{\mathrm{mag}}\rangle under the assumption of magnetic dipole interactions (see Section S5). This confirms that the observed peak in magnetic friction is independent of the applied model for the mechanical contribution of the friction.

Image not extracted

Figure 4. Non-monotonic magnetic friction and its underlying mechanism. (a) Magnetic friction ⟨Fxmag⟩\langle F^{\mathrm{mag}}_x\rangle in dependence of the height hh. The additional subpeaks and plateaus of the simulation results are finite size effects, where the moments on the edges rotate differently than the rest of the moments. (b) Visualization of the magnetic substrate field B⃗sub\vec{B}_{\mathrm{sub}} in dependence of the displacement Δx\Delta x and the height hh obtained from the simulations. (c) Schematic sketch of the simplified model, where the slider is divided in two sublattices. (d)

To understand the unusual load-dependence on magnetic friction, we consider the Hamiltonian

H=H0+Hsub+H0,sub(2)\mathcal{H}=\mathcal{H}_{0}+\mathcal{H}_{\mathrm{sub}}+\mathcal{H}_{0,\mathrm{sub}} \tag*{(2)}

of an idealized system, where Fxmech≡0F_{x}^{\mathrm{mech}}\equiv0. Here, H0\mathcal{H}_{0} is the total energy of the slider, Hsub\mathcal{H}_{\mathrm{sub}} denotes the magnetic exchange energy between the layers, and H0,sub\mathcal{H}_{0,\mathrm{sub}} the internal magnetic energy of the bottom layer (being constant). The energy transfer from and to the top layer during sliding is then given by the time derivative of H0\mathcal{H}_{0}. Using the rotors’ equations of motion, we find (see Sec. S6 for the derivation)

dH0dt=∑i,j[f(θ˙ij)θ˙ij−∂Hsub∂θijθ˙ij].(3)\frac{\mathrm{d}\mathcal{H}_{0}}{\mathrm{d}t}=\sum_{i,j}\left[f(\dot{\theta}_{ij})\dot{\theta}_{ij}-\frac{\partial\mathcal{H}_{\mathrm{sub}}}{\partial\theta_{ij}}\dot{\theta}_{ij}\right]. \tag*{(3)}

The first term on the r.h.s. accounts for the energy dissipation rate owing to the rotor’s mechanical friction within their shafts, where f(θ˙ij)f(\dot{\theta}_{ij}) is the microscopic shaft friction. The second term considers the power required for rotation of the magnets in presence of the substrate field B⃗sub\vec{B}_{\mathrm{sub}}. Because the B⃗sub\vec{B}_{\mathrm{sub}} rotates as a function of Δx\Delta x (see Figure 4b), it exerts torques T⃗ijsub=m⃗ij×B⃗sub\vec{T}_{ij}^{\mathrm{sub}}=\vec{m}_{ij}\times\vec{B}_{\mathrm{sub}} on the magnetic moments m⃗ij\vec{m}_{ij}. The torque components parallel to the rotating axis [Tijsub]y=−∂Hsub/∂θij\left[T_{ij}^{\mathrm{sub}}\right]_{y}=-\partial\mathcal{H}_{\mathrm{sub}}/\partial\theta_{ij} leads to the rotation of the moments as experimentally observed (Figure 2a/Figure 2b). To simplify Eq. (3), we use of the commensurability between both layers, which renders H0\mathcal{H}_{0} periodic with respect to the lattice constant PP. Accordingly, when taking the average over PP and considering a steady state, ⟨dH0/dt⟩=0\langle\mathrm{d}\mathcal{H}_{0}/\mathrm{d}t\rangle=0, and we find

∑i,j⟨f(θ˙ij)θ˙ij⟩=∑i,j⟨∂Hsub∂θijθ˙ij⟩=−∑i,j⟨[Tijsub]yθ˙ij⟩.(4)\sum_{i,j}\langle f(\dot{\theta}_{ij})\dot{\theta}_{ij}\rangle=\sum_{i,j}\left\langle\frac{\partial\mathcal{H}_{\mathrm{sub}}}{\partial\theta_{ij}}\dot{\theta}_{ij}\right\rangle=-\sum_{i,j}\left\langle\left[T_{ij}^{\mathrm{sub}}\right]_{y}\dot{\theta}_{ij}\right\rangle. \tag*{(4)}

This implies that the averaged energy dissipation rate (l.h.s) is only given by T⃗ijsub\vec{T}^{\mathrm{sub}}_{ij} and ϑ˙ij\dot{\vartheta}_{ij} (r.h.s). We will discuss these two parameters below.

The experimental observation that the magnetic moments of the slider exhibit within good approximation always either ferromagnetic or antiferromagnetic order (see Figure 2a/Figure 2b) allows for a simplified model, where the moments of the slider are arranged within two sublattices with different magnetic orientations (Figure 4c). Because the moments within each sublattice are always parallel, the top layer is fully characterized by two rotational degrees of freedom φ\varphi and ϑ\vartheta. Considering the overdamped limit where the slider energy H0\mathcal{H}_0 is only determined by magnetic interactions HI\mathcal{H}_I within the slider and applying only nearest neighbor interactions, H0\mathcal{H}_0 can be splitted into the interaction energy of moments within the same sublattice

Hself=asin⁡2(φ)+asin⁡2(ϑ)(5)\mathcal{H}_{\mathrm{self}} = a\sin^2(\varphi) + a\sin^2(\vartheta) \tag*{(5)}

and the magnetic exchange energy across the two sublattices

Hint=bcos⁡(φ−ϑ),(6)\mathcal{H}_{\mathrm{int}} = b\cos(\varphi-\vartheta), \tag*{(6)}

with H0=Hself+Hint\mathcal{H}_0 = \mathcal{H}_{\mathrm{self}} + \mathcal{H}_{\mathrm{int}}. The full derivation is detailed in Sec. S7. The amplitudes aa and bb are derived for the geometry of our setup and matched to the specifications of the magnets. Because the substrate magnetic field corresponds to a rotating field in the comoving frame of the top layer, we approximate

Hsub=c(h)cos⁡(φ−kΔx)+c(h)cos⁡(ϑ−kΔx)(7)\mathcal{H}_{\mathrm{sub}} = c(h)\cos(\varphi-k\Delta x) + c(h)\cos(\vartheta-k\Delta x) \tag*{(7)}

with k=2π/Pk = 2\pi/P and interaction strength c(h)∼1/h3c(h) \sim1/h^3, again matched to the interactions in the experiments. Assuming f(φ˙)≈−Γφ˙f(\dot{\varphi}) \approx-\Gamma\dot{\varphi} and f(ϑ˙)≈−Γϑ˙f(\dot{\vartheta}) \approx-\Gamma\dot{\vartheta} with Γ\Gamma a generalized friction coefficient associated with the microscopic change of the magnetization, we arrive at the equations of motion Γφ˙=−∂H/∂φ\Gamma\dot{\varphi}=-\partial\mathcal{H}/\partial\varphi and Γϑ˙=−∂H/∂ϑ\Gamma\dot{\vartheta}=-\partial\mathcal{H}/\partial\vartheta, which can be solved numerically (Methods). The corresponding total dissipated energy can be obtained by integrating Eq. (4), which can then be used to calculate the friction. Details are given in Sec. S8.

Figure 4d shows how the sum of torques Tysub=−∂Hsub/∂φ−∂Hsub/∂ϑT_y^{\mathrm{sub}}=-\partial\mathcal{H}_{\mathrm{sub}}/\partial\varphi-\partial\mathcal{H}_{\mathrm{sub}}/\partial\vartheta varies for different heights. For h=6.5 mmh=6.5\ \mathrm{mm} the torque oscillates smoothly which leads to an approximately constant angular velocity ω≈kv\omega\approx kv as experimentally confirmed in Figure 2a/Figure 2b (for h→0h\to0, ω=kv\omega=kv becomes perfectly constant, see Sec. S9). Because TysubT_y^{\mathrm{sub}} alternates symmetrically around Tysub=0T_y^{\mathrm{sub}}=0 during one lattice constant, we obtain from Eq. (4) ⟨∂Hsub/∂φ φ˙⟩+⟨∂Hsub/∂ϑ ϑ˙⟩≈−⟨Tysub⟩ω≈0\langle\partial\mathcal{H}_{\mathrm{sub}}/\partial\varphi\ \dot{\varphi}\rangle+\langle\partial\mathcal{H}_{\mathrm{sub}}/\partial\vartheta\ \dot{\vartheta}\rangle\approx-\langle T_y^{\mathrm{sub}}\rangle\omega\approx0 for the average energy dissipation rate, i.e. a vanishing magnetic friction (due to the mechanical shaft friction of the magnets, the friction is not exactly zero). For intermediate distances, h=9.0 mmh=9.0\ \mathrm{mm}, however, TysubT_y^{\mathrm{sub}} no longer symmetrically oscillates around Tysub=0T_y^{\mathrm{sub}}=0, which results in a strong magnetic friction as observed in our experiments. For h=11.0 mmh=11.0\ \mathrm{mm}, the torque is small while the rotational velocity of the moments is close to zero (c.f. Figure 2a/Figure 2b). This leads to a vanishing magnetic friction for large hh, and, thus, explains the experimentally observed maximum in the magnetic friction ⟨Fxmag⟩\langle F_x^{\mathrm{mag}}\rangle as a function of hh (Figure 4a). Analytic approximations for ⟨Fxmag⟩\langle F_x^{\mathrm{mag}}\rangle based reflecting the discussed behavior of the different regimes, are presented in Sec. S9.

The asymmetric variation of the torque as a function of Δx\Delta x leads to the signature of a magnetic hysteresis behavior similar to that observed in a ferromagnet (see Sec. S10). Such a behavior was also reported earlier for magnetic friction measurements28, as well as in simulations on energy dissipation in Ising layers[37, 38] [14]. The observed hysteresis is shown in Figure 4e, where we plotted TysubT_y^{\mathrm{sub}} obtained from numerical integration of the simplified model as a function of

M⃗=∑ijm⃗ij\vec{M} = \sum_{ij} \vec{m}_{ij}

the angle Θ\Theta between the slider’s magnetization and the xx-axis (inset Figure 4d). Within the chosen representation, the area within the hysteresis loops corresponds to the dissipated energy during sliding (see Sec. S11 for a derivation), showing the same nonlinear dependence on hh. The signature of the hysteresis can also be understood from the simplified Hamiltonian of our system, which we discuss in Sec. S12.

Calculating ⟨Fxmag⟩\langle F_x^{\mathrm{mag}}\rangle from our simplified model (Methods), the result is indeed in excellent qualitative agreement with our results thus far (Figure 4f). Notably, our simplified model also confirms the observed connection between ⟨Fxmag⟩\langle F_x^{\mathrm{mag}}\rangle and the averaged order parameter ⟨Σ⟩\langle\Sigma\rangle (Figure 4f). This underlines the direct link between the measured changes in the magnetic arrangements and the resulting friction. Using the MD simulations, we can confirm all our findings based on the simplified model for the full system taking all 7×77 \times7 rotational degrees of freedom into account (see Sec. S13).

Discussion and Outlook

Our results demonstrate that magnetic friction can be precisely tuned by controlling the microscopic magnetic order at the interface. This aligns with theoretical predictions[34] [8, 14] and recent atomic force microscopy measurements[36] [9], which suggest that friction increases sharply in systems near structural rearrangements or phase transitions, regardless of their origin. Unlike these previous studies, where such changes in order are externally initiated by temperature control or applied fields, the interactions during sliding themselves drive the transition within the studied magnetic layers, with peak dissipation arising from dynamic switching between ferromagnetic and antiferromagnetic states. The macroscopic nature of our setup provides unobstructed access to the structural dynamics. This offers unique insights into the interplay between sliding and structural order, while overcoming the limitations of atomic force microscopy-based techniques, which cannot directly monitor the dynamics of underlying phase transitions and excitations. Expanding our model system to incommensurate magnetic configurations and mismatching magnetic layers, this macroscopic nature of the setup could also be exploited to explore the dynamics and propagation of topological magnetic solitons[39, 40, 41, 42], which are hard to access experimentally in conventional magnetic materials without modifying the underlying structure[43].

The observed breakdown of Amontons’ law establishes friction as a sensitive probe of collective magnetic dynamics and paves the way for wear-free, tunable interfaces. Beyond macroscopic rotor arrays, the principles demonstrated here are broadly applicable to low-dimensional magnets, spintronic materials, and XY model-type systems with similar energetic symmetries, where frictional anomalies could signal or enable magnetic switching27. These findings open avenues for the design of friction metamaterials[44, 45] [29, 30, 31] — engineered surfaces that dissipate energy in a programmable manner. By tailoring magnetic hysteresis through carefully arranged micromagnets32, such systems could enable reconfigurable, energy-efficient interfaces controlled by internal degrees of freedom.

Acknowledgement

We acknowledge stimulating discussions with Andrea Vanossi. We thank Dieter Barth for advice and technical assistance in building the force testing setup, Davide Bossini and Veit-Loretz Hueta for fruitful discussions, Laurence Carls for help with communications, and Thomas Franosch for his valuable input regarding the calculation of the energy transfer rates. The authors acknowledge support by the local computing resources through the core facility SCCKN.

Methods

Experiments: A single ring NdFeB magnet (N35 grade with nickel surface coating, HKCM Article No. 9963-73617) serves as the rotor, and a cylindrical magnet of the same kind (HKCM Article No. 9962-61814) serves as the substrate. It is radially magnetized so that its dipole moment can freely rotate about a fixed metal pin, providing one rotational degree of freedom. The metal pin is an 8 mm-long dowel, 1 mm in diameter, made of 18-8 stainless steel (McMaster-Carr part number 91585A062).

A commercial fused deposition modeling (FDM) 3D printer (Bambu Lab X1 Carbon) and standard PLA filament were used to fabricate the plastic probe, to which the pin was then affixed. This setup allows the cylindrical magnet to rotate around the pin while remaining attached to the probe.

To maintain a consistent gap between the probe and the substrate, we incorporated brass rollers that serve as direct mechanical supports during sliding. These rollers help stabilize the probe against the strong magnetic attraction between the 7x7 rotor array and the substrate. Each brass roller has a cylindrical form similar to the rotors, and additional 3D-printed spacers of varying heights ensure discrete probe-substrate separations. This arrangement enables smooth gliding of the probe across the substrate surface.

Simulations: To obtain numerical data for our setup, we integrate the dynamics of the angles θij\theta_{ij} with the Verlet algorithm[46]

θij(t+Δt)=2θij(t)−θij(t−Δt)+1If(θ˙ij(t))Δt2+1ITijtot(t)Δt2\theta_{ij}(t+\Delta t)=2\theta_{ij}(t)-\theta_{ij}(t-\Delta t)+\frac{1}{I}f(\dot{\theta}_{ij}(t))\Delta t^2+\frac{1}{I}T_{ij}^{\mathrm{tot}}(t)\Delta t^2

using a time step of Δt=3×10−4\Delta t=3\times10^{-4} s. Here, II is the moment of inertia of the rotors of the experiments, and Tijtot=m⃗ij×B⃗T_{ij}^{\mathrm{tot}}=\vec{m}_{ij}\times\vec{B} is the magnetic torque acting on the moment m⃗ij\vec{m}_{ij} due to the total magnetic field B⃗\vec{B} generated by the moments of the substrate and the other moments on the slider. We assume dipole interaction, which means all moments m⃗\vec{m} generate a field[47]

B⃗dipole(r⃗)=μ04π∣r⃗∣3[3(m⃗⋅r⃗)r⃗∣r⃗∣2+m⃗],\vec{B}_{\mathrm{dipole}}(\vec{r})=\frac{\mu_0}{4\pi|\vec{r}|^3}\left[\frac{3(\vec{m}\cdot\vec{r})\vec{r}}{|\vec{r}|^2}+\vec{m}\right],

where μ0\mu_0 is the vacuum permeability. The geometry and parameters of the simulations are matched to the experimental values. For the friction term in the integrator for θij\theta_{ij}, we assume f(θ˙ij(t))=−γθ˙ij(t)f(\dot{\theta}_{ij}(t))=-\gamma\dot{\theta}_{ij}(t), where the friction coefficient γ\gamma is a free parameter that is adjusted for stability and to match the experimental results. The total magnetic interaction force between the top and the bottom layer is calculated by summing over all dipole interaction forces F⃗ij=∇(m⃗ij⋅B⃗sub)\vec{F}_{ij}=\nabla(\vec{m}_{ij}\cdot\vec{B}_{\mathrm{sub}}) acting between moments of the slider and the substrate. More details are given in Sec. S14.

Simplified model: Calculating the partial derivatives of the Hamiltonian H\mathcal{H} of the simplified model regarding the two degrees of freedom φ\varphi and ϑ\vartheta and taking the shaft friction into account, we obtain two coupled, nonlinear differential equations. These equations are solved numerically using a standard solver based on the explicit Runge-Kutta (4,5) formula[48, 49]. The dissipated energy EdissE_{\mathrm{diss}} can be obtained by formulating Eq. (2) for the two remaining degrees of freedom, integrating the right-hand side regarding the time needed to translate one lattice constant PP, and inserting the solutions for φ\varphi and ϑ\vartheta (see also Sec. S7 of the SI). The observables are calculated via ⟨Fxmag⟩=Ediss/P\langle F_x^{\mathrm{mag}}\rangle=E_{\mathrm{diss}}/P and ⟨Σ⟩=⟨cos⁡(φ−ϑ)⟩\langle\Sigma\rangle=\langle\cos(\varphi-\vartheta)\rangle. More information regarding the numerical solution of the simplified model is given in Sec. S15.

References

  1. [34]Iglói, F., Pleimling, M. & Turban, L. Nonequilibrium phase transition in a driven Potts model with friction. Phys Rev E Stat Nonlin Soft Matter Phys 83, 041110 (2011).
  2. [35]Hucht, A. Nonequilibrium phase transition in an exactly solvable driven Ising model with friction. Phys Rev E Stat Nonlin Soft Matter Phys 80, 061138 (2009).pubmed.ncbi.nlm.nih.gov/20365149
  3. [36]Wang, W., Dietzel, D., Liu, C. & Schirmeisen, A. Nanoscale Friction across the First-Order Charge Density Wave Phase Transition of 1T-TaS. ACS Appl Mater Interfaces 15, 4774–4780 (2023).
  4. [37]Acharyya, M. & Chakrabarti, B. K. Response of Ising systems to oscillating and pulsed fields: Hysteresis, ac, and pulse susceptibility. Phys Rev B Condens Matter 52, 6550–6568 (1995).pubmed.ncbi.nlm.nih.gov/9981884
  5. [38]Ortín J. & Goicoechea J. Dissipation in quasistatically driven disordered systems. Phys. Rev. B 58, 5628 (1998).DOI
  6. [39]Kuchkin, V. M., Barton-Singer, B., Bessarab, P. F. & Kiselev, N. S. Symmetry-governed dynamics of magnetic skyrmions under field pulses. Commun. Phys. 8, 26 (2025).DOI
  7. [40]Bezsmer tna, O. et al. Magnetic Solitons in Hierarchical 3D Magnetic Nanoarchitectures of Nanoflower Shape. Nano Lett 24, 15774–15780 (2024).
  8. [41]Jin, X.-W., Yang, Z.-Y., Liao, Z.-M., Jing, G. & Yang, W.-L. Unveiling stable one-dimensional magnetic solitons in magnetic bilayers. Phys. Rev. B. 109, 014414 (2024).DOI
  9. [42]Souza, J. C. B., Vizarim, N. P., Reichhardt, C. J. O., Reichhardt, C. & Venegas, P. A. Skyrmion soliton motion on periodic substrates by atomistic and particle-based simulations. EPL 148, 56002 (2024).
  10. [43]Castro, I. et al. Modeling the spatial resolution of magnetic solitons in magnetic force microscopy and the effect on their sizes. Sci Rep 15, 11944 (2025).arxiv.org/abs/2406.04007
  11. [44]Jeong, E., Calius, E. & Ramezani, M. Design and analysis of a 3D frictional mechanical metamaterial for efficient energy dissipation. Adv. Mater. Technol. 10, 2400614 (2025).DOI
  12. [45]Lv, W., Yu, P. & Li, D. An energy dissipation metamaterial based on Coulomb friction and vibration. Int. J. Mech. Sci. 263, 108764 (2023).DOI
  13. [46]Levesque, D. & Verlet, L. Molecular dynamics and time reversibility. J. Stat. Phys. 72, 519–537 (1993).DOI
  14. [47]Boyer, T. H. The force on a magnetic dipole. Am. J. Phys. 56, 688–692 (1988).DOI
  15. [48]Dormand, J. R. & Prince, P. J. A family of embedded Runge-Kutta formulae. J. Comput. Appl. Math. 6, 19–26 (1980).DOI
  16. [49]Shampine, L. F. & Reichelt, M. W. The MATLAB ODE Suite. SIAM J. Sci. Comput. 18, 1–22 (1997).

Paper details

vlnc:W_2xqae8ng
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents