Predictive Indicator of Critical Point in Equilibrium and Nonequilibrium Magnetic

Systems
Tianyi Zhang1 , Caihua Wan1, 2∗
and Xiufeng Han1,2,3∗
1
Beijing National Laboratory for Condensed Matter Physics,
University of Chinese Academy of Sciences, Chinese Academy of Sciences, Beijing 100190, China
2
Center of Materials Science and Optoelectronics Engineering,
University of Chinese Academy of Sciences, Beijing 100049, China
3
Songshan Lake Materials Laboratory, Dongguan, Guangdong 523808, China

Determining critical points of phase transitions from partial data is essential to avoid abrupt
arXiv:2510.20360v1 [cond-mat.stat-mech] 23 Oct 2025

system collapses and reducing experimental or computational costs. However, the complex physical
systems and phase transition phenomena have long hindered the development of unified approaches
applicable to both equilibrium and nonequilibrium phase transitions. In this work, we propose pre-
dictive indicators to determine critical points in equilibrium and nonequilibrium magnetic systems
based on frequency-dependent response function. For equilibrium phase transition, such as magne-
tization switching under magnetic field, the static magnetization response function to a perturbing
magnetic field diverges at the critical field, serving as a noise-resilient predictive indicator that also
reflects the transition order and critical exponents. In contrast, for nonequilibrium phase transition,
such as magnetization switching driven by spin-transfer torque, static response fails to signal criti-
cality. Instead, the dynamic response at ferromagnetic resonance frequency diverges at the critical
point, which is also robust against thermal noise. We further demonstrate that these static and
dynamic indicators can be unified in the frame of first-order linear differential systems, offering a
generalizable strategy for predicting criticality in both equilibrium and nonequilibrium transitions.

Phase transition is an emergent [1–3] phenomenon that free energy, which belongs to equilibrium phase transi-
widely exists in physical systems [4–7], biological sys- tions. On the other hand, in cases where magnetization
tems [8–12], and socio-economic networks[13]. Predicting is switched using spin-transfer torque (STT) [28], spin-
the critical point of phase transitions helps avoid sudden orbit torque (SOT) [29, 30], or laser [31–35], the phase
system collapses, reduces the cost of experiments and transition arises from dynamic instability caused by the
simulations, and aids in understanding emergent behav- injection of energy flow, which belongs to nonequilibrium
iors. For example, in spintronics, predicting the criti- phase transitions. Therefore, magnetic systems serve as
cal current [14–19] is essential for stabilizing magnetic an ideal platform for studying both types of phase tran-
random-access memory (MRAM) devices. Several works sitions. However, a unified method for predicting the
have proposed significant methods for predicting critical critical points of equilibrium and nonequilibrium phase
points, such as mean-field theory [20], Landau’s theory transitions in magnetic systems is currently lacking and
of phase transitions [21], renormalization group [22], ma- remains an open question.
chine learning [23–25], dynamical equation analysis [14– In this study, we propose two appropriate prediction
19], and critical slowing down and variance increase anal- indicators for critical points in equilibrium and nonequi-
ysis [8, 26, 27]. These methods require prior knowledge librium magnetic systems. We calculate the cases of
of the system’s Hamiltonian or dynamical equations, or magnetic field and STT driven ferromagnet switching
necessitate dense sampling near the critical point. For with perpendicular magnetic anisotropy, corresponding
unknown systems, determining critical point based on to equilibrium and nonequilibrium phase transitions, and
partial data is an interesting and challenging task. study the response function of magnetization to a trans-
There are distinctions in the phase transition mecha- verse disturbance field. Additionally, we explored how
nisms, critical behaviors, and research methods between to utilize the static response to determine the transition
equilibrium and nonequilibrium phase transitions, while order and critical exponents of equilibrium phase tran-
studies on their connections are relatively scarce. Com- sitions. Finally, we general these two indicators to first-
plex research systems and phase transition phenomena order linear dynamical systems.
have hindered the development of general methods for
Predicting the critical magnetic field for magnetization
determining the critical points of both equilibrium and
switching is crucial for designing new magnetic materi-
nonequilibrium phase transitions. Magnetic systems,
als and improving magnetic device stability. For a fer-
however, exhibit both types of phase transitions and have
romagnet with perpendicular magnetic anisotropy sub-
clear physical pictures. On the one hand, the magneti-
jected to a constant magnetic field Bz in the z-direction
zation can be switched by magnetic field, and the phase
and an oscillatory perturbative field Bx = Bx0 cos(ωt) in
transition occurs due to changes in the minimum of the
the x-direction, the double-well potential model describ-
ing the system’s energy is illustrated in Fig. 1a. If the
initial state is spin down, as the Bz gradually increases,
∗ [email protected]; [email protected] the energy of the spin-down state gradually increases.
 2

When it exceeds the barrier, the magnetization under- of |m

e x (0)|, but the variance increases as the tempera-
goes a switching, which is a equilibrium phase transition ture rises. This demonstrates the temperature robust-
process, and the state of the system is determined by ness of |m
e x (0)| as a predictive indicator for magnetization
minimizing the free energy. In order to study the effect switching.
of Bx on the magnetization, we calculated the response
function of mx to Bx , which is defined as (a) Spin up Spin down
(b)
1000

mx (t) 750

|χ0 (ω)|
χ0 (t − t′ ) = (1) 500
Bx (t′ )
250
Its Fourier transform χ
e0 (ω) reflects the amplitude of the
ω/

2.0
mx oscillation at frequency ω. By first linearizing the Bz > 0 ω

1.0
5 2.0
Landau-Lifshitz-Gilbert (LLG) equation up to the first- c
0.0 0.5 1.0 1.
Bx = Bx0 cos(ωt) Bz / Bc
order terms of mx and my , then performing the Fourier (c) (d) 5%
transformation on the equation, and finally calculating |mx(ωc)|

|mx(0)| (×10-2)
|mx(ωc)| (×10-2)
15
4%
the frequency response of mx with respect to Bx , one |mx(0)| 10
10 3%

RE
can obtain
2%
5 5
2

µS α(1+α2 ) 1%
α Bz m z + Km2z
+ B z mz + K − i γ ω
χ
e0 (ω) = h 0 0%
i2 0
2 ]) 2 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
α (Bz mz + Km2z ) − i µs (1+α ω + (Bz + Kmz )
γ Bz / Bc Bz / Bc
(2) (e) (f)
RE

|mx(0)|-0.02 (×10-6)
Here K is anisotropy field, µs is saturation magnetiza- 0.8 4.4% 4.1
tion and γ is gyromagnetic ratio. For a detailed deriva- Bz / Bc 0.6 3.3%
tion, please refer to Supplemental material A [36]. The 4.0
0.4 2.2%
dependence of |e χ0 (ω)| on ω and Bz is shown in Fig. 3.9
0.2 1.1%
1(b). Here, Bc = K represents the critical magnetic field,
and ωc = γK/µs represents the ferromagnetic resonance 0.0% 3.8
2 4 6 8 10 0 100 200 300
frequency in the absence of an external magnetic field, Bx0 / Bc (×10-3) T
Bx0 = 0.01Bc . As shown in Fig. 1(b), |e χ0 (0)| increases
rapidly as Bz approaches Bc , which suggests that we can FIG. 1. (a) Energy landscape of a double-well potential
use |me x (0)| as a predictive indicator for the magnetiza- representing the equilibrium magnetic system under Bx =
tion switching. Substituting ω = 0 and mz = −1 into Bx0 cos(ωt) and Bz > 0. (b) The dependence of response
Eq. (2), we can obtain function |eχ0 (ω)| on the dc magnetic field Bz and ac pertur-
bation magnetic field’s frequency ω. (c) The dependence of
1 |m
e x (ω)| on Bz at ω = 0 and ωc for the lower half branch of the
χ
e0 (0) = (3) hysteresis loop with Bx0 = 0.01Bc . (d) Relative error (RE)
−Bz + K
between |m e x (0)| and |m
e x0 (0)|. (e) The RE as a function of
The relationship between |e χ0 (0)| and |e
χ0 (ωc )| with re- Bz and Bx0 . (f) The |m e x (0)| dependence on the temperature
T at Bz = 0.5Bc .
spect to Bz for the lower half branch of the hystere-
sis loop is shown in Fig. 1(c). We can observe that
|e
χ0 (0)| increases rapidly as Bz approaches Bc , while In the previous section, it is demonstrated that the
|e
χ0 (ωc )| gradually decays, which is consistent with the response of an equilibrium magnetic system to a static
results of past experiments [37, 38] and theoretical stud- perturbation magnetic field is amplified near the critical
ies [39–41]. This is because the increase in Bz de- point. Interestingly, by utilizing this property, we can
stroys the resonance condition. The relative error RE = determine the transition order and the critical exponent
|e
χ(0) − χ e0 (0)| / (e
χ(0) + χ
e0 (0)), where χe(0) is simulated of the equilibrium phase transition. For the first-order
response function using LLG equation, with respect to phase transition system shown in Fig. 1(a), by linearizing
Bz varies as shown in Fig. 1(d). Near Bc , RE increases the equilibrium LLG equation, solves analytically for mz
rapidly, because the deviation of magnetization becomes under a small uniform transverse perturbation Bx from
larger near Bc , violating the assumption |mx | ≪ 1. The −Bx0 to Bx0 , and computes the variances of mz as
dependence of RE on Bz and Bx0 is shown in Fig. 1(e). 4
h
4
i
We can observe that as Bz approaches Bc and Bx0 grad- Var (mz ) = 2Bx0 / 45 (Bz − Bc ) (4)
ually increases, the RE increases, which is also due to
the increase magnetization deviation. Finally, thermal And the max value is
noise is introduced and the temperature T dependence Var (mz )max = 1 (5)
of |me x (0)| is calculated at Bz = 0.5Bc , as shown in Fig.
1(f). Each temperature is repeated 20 times. It shows For a detailed derivation, please refer to Supplemental
the temperature has almost no effect on the mean value material B [36]. The dependence of Var(mz ) on Bz is
 3

shown in Fig. 2(a). Worth noting, the maximum value with perpendicular magnetic anisotropy subjected to a
of Var(mz ) does not depend on Bx0 , but only on the constant STT Jstt polarized in the z-direction, a con-
difference in magnetization before and after the critical stant magnetic field Bz in the z-direction and an oscilla-
point. For the case of a second-order phase transition, tory perturbation magnetic field Bx = Bx0 cos(ωt) along
we take the ferromagnetic-to-paramagnetic transition in the x-direction, the double-well potential model is illus-
two-dimensional Ising model as an example. As shown in trated in Fig. 3(a). Unlike magnetic field-driven mag-
Fig. 2(b), a 50×50 Ising model is subject to a constant netization switching, STT-driven magnetization switch-
magnetic field h0 and a perturbation temperature ∆T , ing occurs by destabilizing the system from its equilib-
which follows a uniform distribution from −∆T0 to ∆T0 . rium point rather than altering the system’s energy land-
The magnetization mz and Var (mz ) as a function of tem- scape. When the STT is sufficiently large to push the sys-
perature are shown in Fig. 2(c) with exchange interaction tem over the energy barrier, the magnetization switches,
energy J = 1, Boltzmann’s constant kB = 1, h0 = 0 and which belongs to nonequilibrium phase transition pro-
∆T0 = 0.2. Near the critical temperature Tc = 2.269, the cess. And conclusions drawn from previous studies on
magnetization approximately follows mz ∝ (Tc − T ) ,
α
(a) (b) +1
1.0
where α is the critical exponent ideally equals to 0.125 -1
[42]. The dependence of Var(mz ) on T is: 0.8
h0

Var(mz)
0.6
1
Var (mz ) = (Tc − T )
2α−2
∆T02 (6)
0.4 ΔT
3 0.2
0.0
And the max value 0.0 0.5 1.0
Bz / B c
Var (mz )max ∝ (∆T0 )
2α
(7) (c) (d)
1.0 -3 data
8

Var(mz) (×10-3)
By comparing Eqs. (4) and (6), Eqs. (5) and (7), two in- -6 Linear fitting
0.8 mz

ln(Var(mz))
teresting conclusions can be extracted. Firstly, while the 0.6 Var(mz) -9
4
mz

peak value of the variance remains independent of per- 0.4 k = -1.75891

-12
turbation strength in first-order transitions, it strongly 0.2 -15
depends on it in second-order cases. Secondly, in the 0.0 0
-18
case of second-order transitions, the critical exponent α -5 -4 -3 -2 -1 0
1.6 2.0 2.4 2.8
can be extracted by examining the scaling behavior of the T ln(Tc-T)
variance near the critical point. Specifically, the scaling
behavior of Var(mz ) is plotted on a logarithmic scale and FIG. 2. (a) The perturbation-induced variance Var(mz ) of
performed a linear fit in the vicinity of the transition crit- the lower half part of hysteresis loop as a function of Bz . (b)
ical point, as shown in Fig. 2(d). The estimate α ≈0.121, A 50×50 Ising model subject to a constant magnetic field h0
which is in close agreement with the theoretical predic- and a perturbation temperature ∆T . (c) A combined plot of
tion 0.125. Deviations from linearity observed far from magnetization mz and Var(mz ) as a function of T with h0 =
0 and ∆T0 = 0.2. (d) The relationship between Var(mz ) and
the critical point arise due to the breakdown of the ap-
√
1/8 temperature T near the critical point Tc ≈ 2.269.
proximation mz = 1 − sinh−4 ln(1 + 2)Tc /T

≈
1/8
(1 − T /Tc ) , which holds only in the regime (Tc −
T )/Tc ≪ 1.
Compared to equilibrium phase transitions, due to
the lack of a unified theoretical framework and com-
plexity of research systems and phenomena, nonequi- equilibrium phase transitions valid. Therefore, by using
librium phase transitions have been relatively less stud- the same method as in Eq. (2), and taking into account
ied. However, ferromagnetic systems can serve as a sim- the influence of STT in the LLG equation, we can obtain
ple and clear physical model. Consider a ferromagnet the response function as

µs α(1+α2 )
α Jstt mz + α Bz mz + Km2z + Bz mz − αJstt mz + K − i γ ω
χ
e0 (ω) =   2 (8)
2) 2
Jstt mz + α (Bz mz + Km2z ) − i µs (1+α
γ ω + (Bz − αJstt + Kmz )

Substituting ω = 0 and mz = −1 into Eq. (8), we can obtain

K − Bz
χ
e0 (0) = 2 2 (9)
Jstt + (Bz − K)
 4

Detailed derivation is in Supplemental material C [36]. dictive and stabilizing mechanism for reliable spin torque
From Eq. (9), it can be seen that as the absolute value oscillator operation.
of Jstt increases, the static response function decreases
monotonically. Clearly, it cannot serve as an indicator (a) Spin up Spin down (b) 1000
for predicting phase transitions. Therefore, we analyze 800
the situation at the critical current, substituting Jstt =

| χ0 (ω)|
600
Jc = −α(Bz + Kmz ) and mz = −1 in Eq. (8). To 400
simplify the physical picture, we consider the case where Bx Bx 200

Bz = 0. It follows that Jstt Jstt

2.0
ω/
Jstt ≠ 0 ω 1.5 2.0

1.0
K + iα µγs ω c 0.5 1.0 / J c
χ
e0 (ω)|Jstt =Jc =  (10) Bx = Bx0 cos(ωt) 0.0 J stt
2 
(c)2 (d)

(α2 + 1) K 2 − µγs ω
|mx(ωc)|

|mx(0)|-10-5 ×(10-9)
40%

|mx(ωc)| (×10-2)
0.5
Therefore, when ω = ωc , the response function tends to |mx(0)|
-0.0

RE
diverge at the critical point. And |e χ0 (ωc )| is 1 20%
√ -0.5
1 + α2 K 0%
|χ̃0 (ωc ) | = q (11) 0 -1.0
2 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(αK − Jstt ) (Jstt + αK) + 4K 2
Jstt / Jc Jstt / Jc
The dependence of the response function on ω and Jstt is (e) RE (f)
0.8 90%

|mx(ωc)| (×10-4)
shown in Fig. 3(b). It can be observed that |e χ0 (ωc )| in- 9.9

Jstt / Jc
creases rapidly as Jstt approaches Jc , indicating that we 0.6 60%
can use |m e x (ωc )| as a predictive indicator for STT-driven 0.4
9.8
magnetization switching. The dependence of |m e x (0)| and 30%
0.2
|me x (ωc )| on Jstt for the lower half of the hysteresis loop
0.0% 9.7
is shown in Fig. 3(c). As Jstt approaches Jc , |m e x (ωc )| 2 4 6 8 0 100 200 300
-3
increases rapidly, while |m e x (0)|gradually decays, which Bx0 / Bc (×10 ) T
is exactly the opposite of the situation where the magne-
tization switched by magnetic field. It is consistent with FIG. 3. (a) Energy landscape of a double-well potential rep-
the results of past experiments [43, 44]. Interestingly, it resenting the nonequilibrium magnetic system under Bx =
shows that when the magnetic system is in a resonant Bx0 cos(ωt) and spin transfer torque strength Jstt . (b) The
state, its response to external disturbances is amplified, dependence of response function |e χ0 (ω)| on the Jstt and ac
perturbation magnetic field’s frequency ω. (c) The depen-
while its response to internal disturbances is suppressed.
dence of |me x (ω)| on Jstt at ω = 0 and ωc for the lower half
We calculated the relative error RE, as shown in Fig. branch of the hysteresis loop with Bx0 = 10−5 Bc . (d) Rela-
3(d). Near Jc , RE increases rapidly because the magne- tive error (RE) between |m e x (ωc )| and |me x0 (ωc )|. (e) The RE
tization deviation from the equilibrium state increases. It as a function of Jstt and Bx0 . (f) The |m e x (ωc )| dependence
is worth noting that for the same Bx0 , the magnetization on the temperature T at Jstt = 0.5Jc .
in the resonant state deviates from the equilibrium state
much more than not in the resonant state. Therefore, in Based on the above analysis, it is demonstrated that
the calculations of Fig. 3(c)–(f), Bx0 = 10−5 Bc , which is for magnetization switching driven by magnetic fields and
1/1000 of that in Fig. 1. The dependence of RE on Jstt STT, the response amplitude of mx under perturbation
and Bx0 is shown in Fig. 3(e). It can be seen that as Jstt fields at frequencies of 0 and ωc can be served as pre-
and Bx0 increase, RE increases rapidly, which is also be- dictive indicators for critical points. It is interesting to
cause of the increased magnetization deviation. Finally, generalize this approach to more general systems. For a
we calculated the effect of thermal noise on |m e x (ωc )| at system exhibiting both equilibrium and nonequilibrium
Jstt = 0.5Jc , as shown in Fig. 3(f). By repeating the phase transitions, assume it follows a first-order linear
calculation 20 times at each temperature, it can be ob- differential equation,
served that as the temperature increases, the mean value
of |me x (ωc )| remains essentially unchanged, but the vari- dX(t)
ance gradually increases, which is similar to the situation = AX(t) + B(t) (12)
dt
when the magnetic field switched the magnetization, re-
flecting the temperature robustness of |m e x (ωc )| as a crit- In the equation, X is an n-dimensional vector composed
ical prediction indicator. Dynamic response under weak of order parameters, A is the dynamic coefficient matrix,
transverse perturbation provides as a sensitive indicator B is input vector that determines the steady state
of approaching the critical point in MRAM devices, of- of the system. The eigenvalues λi (i = 1, 2, · · · , n) of
fering an early warning of write/read failure, instability A are ordered according to their real parts, with the
and inhomogeneity. This method also establishes a pre- assumption that Re (λ1 ) ≥ Re (λ2 ) ≥ · · · ≥ Re (λn ).
 5

When the system is at the critical point of an equilibrium driven by magnetic fields and STT in magnetic systems.
phase transition, it must hold that Re (λ1 ) = 0. Since In the case of magnetic field-driven switching, by apply-
the system is in equilibrium and there is no external ing a transverse static disturbance field Bx0 , the response
energy input, oscillations cannot occur at the critical mx diverges as Bz approaches the critical point Bc , serv-
state. Therefore, Im (λ1 ) = 0 holds true and det(A) = ing as a noise-resilient predictive indicator. By replacing
0, which implies that for a static small perturbation, the static disturbance with a uniformly distributed dis-
the response of X, which is A−1 B, tends to diverge as turbance and performing multiple measurements to ob-
system approaches the critical point. For example, in the tain the variance of the order parameter, we found that
case of magnetic field-driven


magnetization switching, for first-order phase transitions, the maximum variance
λ1,2 = −γ/ µs 1 + α2

[α (−Bz + K) ± i (Bz − K)], depends only on the difference of the order parameter
B(t) = γ/ µs 1 + α2 (α, 1)T Bx0 , at the critical across the critical point and is independent of the dis-
magnetic field Bc = K, Re (λ1 ) = Im (λ1 ) = 0. However, turbance magnitude. However, for second-order phase
for nonequilibrium phase transitions, the situation is transitions, the maximum variance is proportional to the
totally different. At the critical point, there still exists 2α-th power of the disturbance magnitude. Near the crit-
Re (λ1 ) = 0, but injection of external energy flow enables ical point, the variance is proportional to the 2(α-1)-th
the oscillatory behavior, which is a main difference power of the difference between the critical point and the
between equilibrium and nonequilibrium phase transi- variable. Using this property, we calculated the tempera-
tions. Therefore, Im (λ1 ) may not equals to 0 at critical ture critical exponent of a 50 × 50 two-dimensional Ising
point and Hopf bifurcation may occur. For a constant model and get 0.121, which is close to the theoretical
perturbation, the response of X may not tend to diverge value of 0.125. For STT-driven magnetization switch-
at the critical point. For instance, in the case of STT- ing, applying a transverse disturbance field oscillating at
driven magnetization


switching discussed above, λ1,2 = the ferromagnetic resonance frequency, the response mx
−γ/ µs 1 + α2 [−JST T + α (K − Bz ) ± i (Bz − αJST T at frequency ωc rapidly increases as Jstt approaches the
−K)], and at the critical current Jc = α (K − Bz ), critical point Jc , also serving as a noise-resilient predic-
Re (λ1,2 ) = 0, Im (λ1,2 ) = ±iγ (K − Bz ) /µs . However, tive indicator. Finally, we demonstrate that this frame-
if the perturbation has a frequency of ωc = | Im (λ1 ) |, work is applicable to general first-order linear systems.
the response of X will gradually diverge as system Our approach offers predictive indicators for determining
approaches the critical point. Therefore, X(0) e and critical points in both equilibrium and nonequilibrium
X(ω
e c ) can serve as predictive indicators for critical point phase transitions, which is computationally and exper-
in equilibrium and nonequilibrium phase transition, re- imentally cost-effective, easy to implement, and robust
spectively. Several significant works previously reported against thermal fluctuations. Our work has potential ap-
are also consistent with this conclusion. For instance, plications in enhanced magnetic sensor, MRAM device
the static response function in the desynchronized- test, dynamic control of spin torque oscillators, designing
synchronized equilibrium transition of the Kuramoto new magnetic materials and serving as warning signals of
model [45, 46] diverges at the critical point. It is worth critical point in complex systems.
noting that in the quantum system (for example, Ryd-
berg atoms[47]), we also observed the divergent behavior
ACKNOWLEDGMENTS
of dynamic indicators near the critical point, indicating
the potential application of our theory in quantum
physics. However, in topological phase transitions This work is financially supported by the National Key
without local order parameters, the proposed method in Research and Development Program of China (MOST)
this work is not applicable. And how to predict critical (Grant No. 2022YFA1402800), the National Natu-
points in these systems is worthy of further investigation. ral Science Foundation of China (NSFC) (Grants No.
T2495210, No. 51831012, No. 12134017, No. 12374131,
and No. W2412079 ) and the Chinese Academy of
In summary, we have proposed two prediction indi- Sciences President’s International Fellowship Initiative
cators for the critical points of magnetization switching (PIFI Grant No. 2025PG0006).

[1] P. W. Anderson, More is different, Science 177, 393 [4] J. M. Kosterlitz and D. J. Thouless, Ordering, metasta-
(1972). bility and phase transitions in two-dimensional systems,
[2] S. Strogatz, S. Walker, J. M. Yeomans, C. Tarnita, E. Ar- Journal of Physics C: Solid State Physics 6, 1181 (1973).
caute, M. De Domenico, O. Artime, and K.-I. Goh, Fifty [5] J. M. Kosterlitz, The critical properties of the two-
years of ‘more is different’, Nature Reviews Physics 4, dimensional xy model, Journal of Physics C: Solid State
508 (2022). Physics 7, 1046 (1974).
[3] A. Rupe and J. P. Crutchfield, On principles of emergent [6] D. J. Amit, H. Gutfreund, and H. Sompolinsky, Storing
organization, Physics Reports 1071, 1 (2024). infinite numbers of patterns in a spin-glass model of neu-
 6

ral networks, Physical Review Letters 55, 1530 (1985). [23] L. Wang, Discovering phase transitions with unsuper-
[7] R. Zivieri, Critical phenomena in ferromagnetic antidot vised learning, Physical Review B 94, 195105 (2016).
lattices, AIP Advances 6 (2016). [24] Y. Huang, S. Bathiany, P. Ashwin, and N. Boers, Deep
[8] V. Dakos, M. Scheffer, E. H. van Nes, V. Brovkin, learning for predicting rate-induced tipping, Nature Ma-
V. Petoukhov, and H. Held, Slowing down as an early chine Intelligence 6, 1556 (2024).
warning signal for abrupt climate change, Proceedings of [25] S. Panahi, L.-W. Kong, M. Moradi, Z.-M. Zhai, B. Glaz,
the National Academy of Sciences 105, 14308 (2008). M. Haile, and Y.-C. Lai, Machine learning prediction of
[9] T. P. Hughes, J. T. Kerry, A. H. Baird, S. R. Con- tipping in complex dynamical systems, Physical Review
nolly, A. Dietzel, C. M. Eakin, S. F. Heron, A. S. Hoey, Research 6, 043194 (2024).
M. O. Hoogenboom, G. Liu, M. J. McWilliam, R. J. [26] M. Scheffer, J. Bascompte, W. A. Brock, V. Brovkin,
Pears, M. S. Pratchett, W. J. Skirving, J. S. Stella, and S. R. Carpenter, V. Dakos, H. Held, E. H. van Nes, M. Ri-
G. Torda, Global warming transforms coral reef assem- etkerk, and G. Sugihara, Early-warning signals for critical
blages, Nature 556, 492 (2018). transitions, Nature 461, 53 (2009).
[10] M. Berdugo, M. Delgado-Baquerizo, S. Soliveres, [27] M. Scheffer, S. R. Carpenter, T. M. Lenton, J. Bas-
R. Hernández-Clemente, Y. Zhao, J. J. Gaitán, N. Gross, compte, W. Brock, V. Dakos, J. van de Koppel, I. A.
H. Saiz, V. Maire, A. Lehmann, M. C. Rillig, R. V. Solé, van de Leemput, S. A. Levin, E. H. van Nes, M. Pascual,
and F. T. Maestre, Global ecosystem thresholds driven and J. Vandermeer, Anticipating critical transitions, Sci-
by aridity, Science 367, 787 (2020). ence 338, 344 (2012).
[11] B. M. Flores, E. Montoya, B. Sakschewski, N. Nasci- [28] D. C. Ralph and M. D. Stiles, Spin transfer torques,
mento, A. Staal, R. A. Betts, C. Levis, D. M. Lapola, Journal of Magnetism and Magnetic Materials 320, 1190
A. Esquı́vel-Muelbert, C. Jakovac, C. A. Nobre, R. S. (2008).
Oliveira, L. S. Borma, D. Nian, N. Boers, S. B. Hecht, [29] C. Song, R. Zhang, L. Liao, Y. Zhou, X. Zhou, R. Chen,
H. ter Steege, J. Arieira, I. L. Lucas, E. Berenguer, J. A. Y. You, X. Chen, and F. Pan, Spin-orbit torques: Mate-
Marengo, L. V. Gatti, C. R. C. Mattos, and M. Hirota, rials, mechanisms, performances, and potential applica-
Critical transitions in the amazon forest system, Nature tions, Progress in Materials Science 118, 100761 (2021).
626, 555 (2024). [30] X. Han, X. Wang, C. Wan, G. Yu, and X. Lv, Spin-orbit
[12] Z. Yang, J. Liang, and C. Zhou, Critical avalanches torques: Materials, physics, and devices, Applied Physics
in excitation-inhibition balanced networks reconcile re- Letters 118, 120502 (2021).
sponse reliability with sensitivity for optimal neural rep- [31] A. Kirilyuk, A. V. Kimel, and T. Rasing, Laser-induced
resentation, Physical Review Letters 134, 028401 (2025). magnetization dynamics and reversal in ferrimagnetic al-
[13] J. Moran, F. P. Pijpers, U. Weitzel, J.-P. Bouchaud, loys, Rep Prog Phys 76, 026501 (2013).
and D. Panja, Critical fragility in sociotechnical systems, [32] W. Cheng, X. Li, H. Wang, X. Cheng, and X. Miao,
Proceedings of the National Academy of Sciences 122, Laser induced ultrafast magnetization reversal in tbco
e2415139122 (2025). film, AIP Advances 7, 056018 (2017).
[14] J. Z. Sun, Spin-current interaction with a monodomain [33] M. Deb, P. Molho, B. Barbara, and J.-Y. Bigot, Control-
magnetic body: A model study, Physical Review B 62, ling laser-induced magnetization reversal dynamics in a
570 (2000). rare-earth iron garnet across the magnetization compen-
[15] K.-S. Lee, S.-W. Lee, B.-C. Min, and K.-J. Lee, Thresh- sation point, Physical Review B 97, 134419 (2018).
old current for switching of a perpendicular magnetic [34] Y. Peng, D. Salomoni, G. Malinowski, W. Zhang,
layer induced by spin hall effect, Applied Physics Let- J. Hohlfeld, L. D. Buda-Prejbeanu, J. Gorchon,
ters 102, 112410 (2013). M. Vergès, J. X. Lin, D. Lacour, R. C. Sousa, I. L. Pre-
[16] T. Taniguchi, S. Mitani, and M. Hayashi, Critical current jbeanu, S. Mangin, and M. Hehn, In-plane reorientation
destabilizing perpendicular magnetization by the spin induced single laser pulse magnetization reversal, Nature
hall effect, Physical Review B 92, 024428 (2015). Communications 14, 5000 (2023).
[17] T. Taniguchi and H. Kubota, Instability analysis of spin- [35] J.-X. Lin, Y. Le Guen, J. Hohlfeld, J. Igarashi, Q. Remy,
torque oscillator with an in-plane magnetized free layer J. Gorchon, G. Malinowski, S. Mangin, T. Hauet, and
and a perpendicularly magnetized pinned layer, Physical M. Hehn, Femtosecond-laser-induced reversal in in-plane-
Review B 93, 174401 (2016). magnetized spin valves, Physical Review Applied 22,
[18] D. Zhu and W. Zhao, Threshold current density for per- 044051 (2024).
pendicular magnetization switching through spin-orbit [36] See supplemental material at [url] for details in response
torque, Physical Review Applied 13, 044078 (2020). function in the case of magnetization switched by mag-
[19] X. Zhang, Z. Xu, and Z. Zhu, Revisiting the analyti- netic field, analytical value of perturbation-induced vari-
cal solution of spin-orbit torque switched nanoscale per- ance and determine the order of the phase transition and
pendicular ferromagnets, Physical Review B 110, 184428 the critical exponent based on the perturbation-induced
(2024). variance and response function in the case of magnetiza-
[20] J. W. Negele, The mean-field theory of nuclear struc- tion switched by spin transfer torque, .
ture and dynamics, Reviews of Modern Physics 54, 913 [37] S. A. Mathews, C. Musi, and N. Charipar, Transverse
(1982). susceptibility of nickel thin films with uniaxial anisotropy,
[21] I. S. Aranson and L. Kramer, The world of the complex Scientific Reports 11, 3155 (2021).
ginzburg-landau equation, Reviews of Modern Physics [38] M. Jungfleisch, A. Chumak, A. Kehlberger, V. Lauer,
74, 99 (2002). D. Kim, M. Onbasli, C. Ross, M. Kläui, and B. Hille-
[22] B. Nienhuis and M. Nauenberg, First-order phase tran- brands, Thickness and power dependence of the spin-
sitions in renormalization-group theory, Physical Review pumping effect in y 3 fe 5 o 12/pt heterostructures mea-
Letters 35, 477 (1975). sured by the inverse spin hall effect, Physical Review B
 7

91, 134407 (2015). [44] S. Petit, C. Baraduc, C. Thirion, U. Ebels, Y. Liu, M. Li,
[39] R. Chantrell, The magnetic properties of textured fine P. Wang, and B. Dieny, Spin-torque influence on the high-
particle systems, Texture, Stress, and Microstructure 11, frequency magnetization fluctuations in magnetic tunnel
107 (1989). junctions, Physical review letters 98, 077203 (2007).
[40] R. D. Cookson, Transverse susceptibility studies of [45] D. C. Roberts, Linear reformulation of the kuramoto
recording media, Ph.D. thesis, University of Central Lan- model of self-synchronizing coupled oscillators, Physical
cashire (2002). Review E 77, 031114 (2008).
[41] C. T. Moorfield, An investigation of non-linear transverse [46] D. Garcı́a-Gudiño, E. Landa, J. Mendoza-Temis,
susceptibility as a basis for improving the measurement of A. Albarado-Ibañez, J. C. Toledo-Roy, I. O. Morales, and
anisotropy in magnetic powdered samples., Ph.D. thesis, A. J. P. O. Frank, Enhancement of early warning prop-
University of Central Lancashire (2020). erties in the kuramoto model and in an atrial fibrillation
[42] C. N. Yang, The spontaneous magnetization of a two- model due to an external perturbation of the system,
dimensional ising model, Physical Review 85, 808 (1952). PLoS ONE 12, e0181953 (2017).
[43] M. Wang, W. Cai, D. Zhu, Z. Wang, J. Kan, Z. Zhao, [47] J. Zhang, L.-H. Zhang, B. Liu, Z.-Y. Zhang, S.-Y. Shao,
K. Cao, Z. Wang, Y. Zhang, T. Zhang, et al., Field- Q. Li, H.-C. Chen, Z.-K. Liu, Y. Ma, T.-Y. Han, Q.-F.
free switching of a perpendicular magnetic tunnel junc- Wang, C. S. Adams, B.-S. Shi, and D.-S. Ding, Early
tion through the interplay of spin–orbit and spin-transfer warning signals of the tipping point in strongly interact-
torques, Nature electronics 1, 582 (2018). ing rydberg atoms, Physical Review Letters 133, 243601
(2024).