PHYSICAL REVIEW LETTERS 130, 266901 (2023)

Nonlinear Exceptional Points with a Complete Basis in Dynamics

Kai Bai ,1 Jia-Zheng Li,1 Tian-Rui Liu,1 Liang Fang,1 Duanduan Wan,1,* and Meng Xiao 1,2,†
1
Key Laboratory of Artificial Micro- and Nano-structures of Ministry of Education and School of Physics and Technology,
Wuhan University, Wuhan 430072, China
2
Wuhan Institute of Quantum Technology, Wuhan 430206, China

(Received 13 January 2023; revised 13 April 2023; accepted 8 June 2023; published 29 June 2023)

Exceptional points (EPs) are special spectral singularities at which two or more eigenvalues, and their
corresponding eigenvectors, coalesce and become identical. In conventional wisdom, the coalescence of
eigenvectors inevitably leads to the loss of completeness of the eigenbasis. Here, we show that this scenario
breaks down in general at nonlinear EPs (NEPs). As an example, we realize a fifth-order NEP (NEP5 )
within only three coupled resonators with both a theoretical model and simulations in circuits. One stable
and another four auxiliary steady eigenstates of the nonlinear Hamiltonian coalesce at the NEP5 , and the
response of eigenfrequency to perturbations demonstrates a fifth-order root law. Intriguingly, the
biorthogonal eigenbasis of the Hamiltonian governing the system dynamics is still complete, and this
fact is corroborated by a finite Petermann factor instead of a divergent one at conventional EPs.
Consequently, the amplification of noise, which diverges at other EPs, converges at our NEP5. Our finding
transforms the understanding of EPs and shows potential for miniaturizing various key applications
operating near EPs.

DOI: 10.1103/PhysRevLett.130.266901

Introduction.—The exotic physics at singularities is parity-time symmetry, and EPs is rare [35–39]. On the other
always under the scrutiny of theoretical and experimental hand, signal amplification [13,16–18] and chiral states
investigations. Exceptional points (EPs) are unique singu- transfer [19–22] require operations pretty close to the
larities in non-Hermitian systems [1–3] where two or more EPs, which demands tedious and precise parameter control,
eigenvalues and their corresponding eigenvectors coalesce especially when higher-order EPs get involved [13,40–42].
[3–5]. After being experimentally demonstrated in micro- Moreover, the unavoidable noise is dramatically increased
wave cavities [6], EPs were subsequently observed in near conventional linear EPs due to the loss of the com-
various systems [7–18]. Many exciting physics and novel pleteness of eigenbasis [43–46]. The above conflicts seem
phenomena related to EPs have been elucidated [19–26]. irreconcilable and generally require brilliant schemes for
For example, the spectra exhibit algebraic singularities at EP-related operations [22,47]. It is thus natural to ask: Can
the EPs, which are responsible for amplifying a detected the completeness of the eigenbasis be revived while the key
signal [13,16]. Meanwhile, a dynamic loop near an EP features of EPs are still preserved?.
leads to chiral state transfer [19–22], and this unique feature Here in this work, we present a nonlinear route to over-
enables various nontrivial functionalities [21–25]. come this dilemma. In analogy with laser theory [29,48], a
These EP-related advances have enriched our under- Hamiltonian with nonlinear saturable gain has some steady
standing of complex systems with nonconservative elements eigenmodes whose frequencies satisfy a polynomial equa-
(gain, loss, and nonreciprocal coupling) and brought many tion with real coefficients. Thus, the solutions of this
fascinating potential applications. However, some funda- polynomial equation are either real or come in complex
mental challenges remain, especially when considering conjugate pairs. This nonlinear Hamiltonian can be mapped
miniaturizing the device in practical schemes. On the one into a higher dimensional parity-time (PT) symmetric linear
hand, previous studies mainly focused on linear systems, Hamiltonian [31,49]. Under this mapping, the phase tran-
where system parameters are independent of the fields’ sition point, EP, of the PT-symmetrical linear Hamiltonian
amplitudes. The detectable signal dissipates rapidly in a corresponds to a unique singularity of the nonlinear
passive system. Including gain elements increases the Hamiltonian. Corresponding to the coalescence of the
system’s complexity in structural optimization and noise PT-symmetrical linear Hamiltonian eigenmodes at the
control. Meanwhile, a proper description of systems with EPs, some (auxiliary) steady eigenmodes of the nonlinear
gain elements unavoidably involves nonlinear effects such Hamiltonian become identical at this singularity. Therefore,
as gain saturation and Kerr nonlinearity [27,28]. Then the we name it nonlinear exceptional point (NEP) with an
outcome deviates from the prediction of the linear model additional “N” denoting nonlinearity. Across the NEPs, the
[29–34]. Nevertheless, research combining nonlinearity, number of steady eigenmodes with real frequency can

0031-9007=23=130(26)=266901(6) 266901-1 © 2023 American Physical Society

 PHYSICAL REVIEW LETTERS 130, 266901 (2023)
(a) (b)
change. Meanwhile, the dynamical evolution of the non-
linear system is governed by the instantaneous Hamiltonian,
whose eigenstates are generally different from the steady
eigenmodes to form the NEP. (Note that as the nonlinear
Hamiltonian depends on the instantaneous state of the
system, different eigenmodes can correspond to different (c)
instantaneous Hamiltonians.) Remarkably, this fact allows
us to design a NEP while the instantaneous Hamiltonian
exhibits a complete basis in dynamics.
To illustrate the physics discussed above, we propose a
general scheme to construct a NEP with a complete basis in
dynamics. However, to keep our formulas and text neat and
concise, we choose a fifth-order NEP (NEP5 ) as an example
and leave the general form to the Supplemental Material,
Sec. 1 [50]. We construct this NEP5 based on three coupled FIG. 1. (a) Schematic of our model. The nonlinear gain
resonators combined with a nonlinear saturable gain, and gA ðjψ A jÞ depends on jψ A j. (b) Eigenvalues versus the external
verify our conclusions with both a model Hamiltonian and perturbation ϵ, where the solid red line and four dashed (cyan,
circuit simulations. At the NEP5 , one steady stable and four blue, purple, and green) lines represent the stable and four
auxiliary eigenmodes become identical, and the response of auxiliary modes, respectively. (c) The critical behavior of the
eigenfrequencies demonstrates a fifth-order root law. Owing stable mode near the NEP5 . Here the circles are from the red line
to the feedback mechanism of the saturable gain, the system in (b), and the solid line represents a straight line with a slope 1=5.
can fall back to one of the stable eigenmodes in a short time The parameters used are ωB ¼ 1.18, ωC ¼ 1.53, lB ¼ 2.87,
and remains at that eigenmode thereafter. Consequently, gC ¼ 1.25, κ2 ¼ 2.
the temporal dynamics of the system at the NEP5 is go-
verned entirely by the instantaneous Hamiltonian anchored and gC represent the linear loss in resonator B and the linear
by the stable eigenmode. Intriguingly, this instantane- gain in resonator C, respectively. Here the nonlinearity of the
ous Hamiltonian is diagonalizable, i.e., the eigenbasis is gain gC is ignored, which is appropriate when the stable
complete. The Petermann factor (PF), which measures value of jψ C j is much smaller than the saturation amplitude
the nonorthogonality of the eigenbasis in an experimen- of the gain profile in resonator C. κ 1 and κ 2 are the corres-
tally friendly manner, converges. Thus, the adverse ef- ponding coupling strength. All the parameters are normal-
fects of noise are largely suppressed from a fundamental ized by κ1, and we set ωA to 0 since any global frequency
perspective. shift is irrelevant. Focusing on the stable state reached, the
Hamiltonian exhibiting NEP5 .—Our system is sketched eigenfrequency ω satisfies a fifth-order equation
in Fig. 1(a). The nonlinear Schrödinger equation is
pðωÞ ¼ ωjDetðH B−C − ωIÞj2
R R
Hjψ R i jψ i ¼ ωjψ i; ð1Þ
þ Im½DetðHC − ωIÞ  Im½DetðHB−C − ωIÞ
where ω is the eigenfrequency. jψ R i ≡ ðψ A ; ψ B ; ψ C ÞT is the þ Re½DetðHC − ωIÞ  Re½DetðHB−C − ωIÞ
right eigenstate with superscript T short for transpose, and ¼ ω5 þ x4 ω4 þ x3 ω3 þ x2 ω2 þ x1 ω þ x0 ¼ 0; ð3Þ
ψ A , ψ B , and ψ C representing the field amplitude of the red
(left), blue (middle), and light red (right) resonators A, B, and the corresponding saturated gain value gs is deter-
and C, respectively. And the tight-binding Hamiltonian mined by
Hjψ R i can be written as
0 1 gs Re½DetðHB−C − ωIÞ − ωIm½DetðHB−C − ωIÞ
ωA þ igA ðjψ A jÞ κ1 0
B C − Im½DetðHC − ωIÞ ¼ 0; ð4Þ
Hjψ R i ¼ @ κ1 ωB − ilB κ2 A; ð2Þ
0 κ2 ωC þ igC here I is the identity matrix.
 
where ωA , ωB , and ωC represent the corresponding resonant ωB − ilB κ2
HB−C ¼ ð5Þ
frequencies, respectively. gA ðjψ A jÞ denotes a nonlinear κ2 ωC þ igC
saturable gain which decreases with the increasing of
jψ A j. In optics, a commonly used gain saturation model is represents the tight-binding Hamiltonian of a subsystem
gA ðjψ A jÞ ¼ Γ=ð1 þ jψ A j2 Þ − γ 0 with Γ representing the consisting of the resonators B and C; HC ¼ ðωC þ igC Þ
pump strength and γ 0 representing the intrinsic loss. lB denotes the on site term of the resonator C. fx0 ; …; x4 g are

266901-2
 PHYSICAL REVIEW LETTERS 130, 266901 (2023)

real functions in the ðωB ; ωC ; lB ; gC ; κ 2 Þ space. Their explicit (a) (b)

forms are provided in the Supplemental Material, Sec. 2

[50]. Here, the gain saturation form is irrelevant to the order
of pðωÞ. We analyze all the possible fifth-order multiple
roots of pðωÞ and choose the combination of parameters
where the roots are stable. Let ϵ represent the external
perturbation of the system parameters to those at the fifth-
order multiple roots ω0 . Near ω0, the response of eigen-
pffiffiffi
frequency Δω ≡ ω − ω0 for ϵ is proportional to 5 ϵ (c)
if ∂ϵ pðωÞ ≠ 0. (Detailed proof is provided in the
Supplemental Material, Sec. 2 [50].) The improved respon-
sivity, ∂ϵ Δω, is highly desirable in amplifying a detected
perturbation ϵ [13,16]. In the main text, ϵ is imposed on ωA
for demonstration purposes, i.e., ωA þ ϵ. We note that
single-mode systems with Kerr nonlinearity can improve
the sensing in noisy environments [51]. In addition to (d) (e)
extraordinary sensitivity enhancement, our approach based
on NEPs with multimode systems is robust and flexible to
achieve various applications explored in conventional linear
EPs [19–25].
To understand the underlying physics, we extend the
solutions search of the polynomial [Eq. (3)] to the complex
plane. In Fig. 1(b), the solid red lines correspond to stable
modes and the other lines to the other four auxiliary modes.
Figure 1(c) shows the critical component of the Δω versus ϵ FIG. 2. (a) Orthogonality function χ for a Hermitian system
pffiffiffi
for the stable mode (open circles), which fits well with 5 ϵ (dot-dashed purple line), a PT-symmetric Hamiltonian in
(the red line). Substituting the ω obtained above and the Eq. (S11) (dashed gray line), and our system (solid red line).
corresponding gain gs into Eq. (1), the eigenstates can be Here, χ does not reach 1 as the parameters are not far enough
obtained (see Fig. S3 of the Supplemental Material [50]). It from all the possible EPs in the parameter space. (b) PF of the
is clear that one stable eigenmode and four auxiliary stable state (solid red line) at the NEP5 is finite, contrasting
sharply with a divergent one for the linear system (dashed gray
eigenmodes become identical at this special singularity line). (c) Evolution of Reðψ A Þ (red line) and gA ðjψ A jÞ (dark green
NEP5 . The five eigenvalues in Fig. 1(b) are either real or line) starting from a small initial state ð10−3 ; 10−3 ; 10−3 ÞT . In a
come in complex conjugate pairs. These eigenvalues and short time, the nonlinear system will reach a stable state. The
the corresponding eigenstates can be mapped into a PT- gA ðjψ A jÞ (dark green) matches gs obtained from Eq. (4) (dark
symmetric linear Hamiltonian [31,49]. Compared with the green circles). (d) The real part of the eigenvalues of H s . (e) The
conventional linear systems, the parameters required to amplitude and phase distribution of the stable state. The param-
reach an EP5 are reduced from 12 in conventional systems eters for the linear system are provided in the Supplemental
Material, Sec. 2 [50]. Parameters of the nonlinear system in (a),
to five in our system. Note that, for the four auxiliary
(b),(d),(e) are the same as those in Fig. 1. The parameters used in
eigenmodes in Fig 1, the corresponding gs are complex. In (c) are ωA ¼ 2.9, ωB ¼ 4.18, ωC ¼ 4.53, lB ¼ 2.87, gC ¼ 1.25,
contrast, gA ðjψ A jÞ for most of the common gain models κ2 ¼ 2, and gA ¼ 5=ð1 þ jψ A j2 Þ − 3.
[29–31] is purely real. Thus, the four auxiliary eigenvalues
(eigenstates) are unphysical, but their presence is crucial to
understanding the NEP5 . with α > β and n denoting the dimension. All eigenstates
In conventional wisdom, the orthogonality of eigenstates are orthogonal to each other for Hermitian Hamiltonians,
gradually diminishes as one approaches the EPs, and the i.e.,hψ Rα jψ Rβ i ¼ δα;β , and thus χ ≡ 1 as depicted by the
eigenbasis is incomplete. The loss of orthogonality can be purple dot-dashed line in Fig. 2(a). In contrast, all eigen-
captured with [46] states become identical, i.e., hψ Rα jψ Rβ i ¼ hψ Rα jψ Rα i ¼ 1 at
conventional EPs in linear systems. As a result, χ ¼ 0 at a
χ ¼ 1 − fjhψ Rα jψ Rβ ijg; ð6Þ linear EP5 as shown by the gray dashed line in Fig. 2(a).
The loss of orthogonality in the vicinity of EPs dramatically
where each eigenstate is prenormalized as hψ Rα jψ Rα i ¼ 1, increases the adverse effects of noise and has triggered
f•g represents the algebraic average of the set f•g, and the an ongoing debate in the EP-related sensing proto-
subscript α; β ¼ f1; …; ng denote different eigenstates cols [13,16,43–46,56]. The enhancement of noise for an

266901-3
 PHYSICAL REVIEW LETTERS 130, 266901 (2023)

eigenmode jψ R i is conventionally characterized by the nonlinear saturable gain. We note that there have been a few
PF [43–45], a measure of nonorthogonality: delicate schemes to compensate for the adverse conse-
quence due to the loss of complete basis: shifting from EPs
PF ≡ 1=jhψ L jψ R ij2 : ð7Þ to transmission peak degeneracies in accelerometers [47]
and stabilizing noise at a nonlinear exceptional nexus [31].
The corresponding left eigenstate hψ L j is also prenormal- Here, the NEP5 with a complete basis offers a way out of
ized. Since the left and right eigenstates are self-orthogonal the dilemma from its fundamental origins, i.e., noise is
at conventional linear EPs, the PF diverges when approach- converged. Meanwhile, χ and the PF of NEPs can be
ing a conventional linear EP as shown by the dashed gray further tuned by including nonreciprocal coupling (see
line in Fig. 2(b). Fig. S6 of the Supplemental Material [50]).
The above scenario for conventional linear EPs can Another promising potential application of EPs is the
break down at NEPs. The (auxiliary) steady eigenmodes of chiral state transfer during the dynamical encirclement of
Eq. (1) can correspond to different Hamiltonians as Hjψ R i EPs. Given the same initial state, the final state does not
depends on jψ R i. Hence the (auxiliary) steady eigenmodes depend on the details of the trajectory but on the direction of
becoming identical at NEPs does not necessarily lead to the the winding (clockwise or anticlockwise) [19–25,58].
coalescence of the eigenmodes of the instantaneous Dynamical encirclement of a NEP5 in the parameter space
Hamiltonian, which governs the system dynamics. More can also lead to chiral state transfer similar to conventional
specifically, each solution of Eq. (3) can correspond to a linear EPs. For demonstration purposes, the trajectory of
different gain value as given by Eq. (4). However, the encirclement (loop) is defined as lB þ δxðtÞ and ωB þ δyðtÞ.
nonlinear system will reach one of the steady stable modes We set δxðtÞ ¼ r cosðτ2πt=TÞ and δyðtÞ ¼ r sinðτ2πt=TÞ
(if it exists) in a short time due to the feedback mechanism with r representing the radius of the loop, T denoting the
[29–31,57], and the nonlinear gain coefficient gA ðtÞ is cycle period, and τ ¼ 1 for the winding direction.
anchored by the stable mode at a stable value gs . Thereafter, Figures 3(a) and 3(b) show the steady-state frequencies
the temporal dynamics is governed by a 3 × 3 instanta- versus δx and δy in the parameter space. The light red and
neous Hamiltonian Hs with the saturable gain replaced by light blue regions represent stable and unstable states,
gs and subjected to noise-imposed fluctuations. Generally, respectively. The system starts at a higher-frequency stable
Hs does not exhibit an EP point though (auxiliary) steady state, as marked by the bold green arrows. If τ ¼ 1 [see
eigenmodes coalesce at the NEPs. Hence, the PF factor, Fig. 3(a)], the state can adiabatically evolve to the lower
which characterizes the noise dynamics within the space frequency state, provided the circling process is slow
governed by Hs, does not diverge.
To unveil this fact more explicitly, we still use the three-
(a) (c) (e)
resonator system in Fig. 1(a). The red and dark green
lines in Fig. 2(c) show the Reðψ A Þ and gA by solving the
time-dependent nonlinear Schrödinger equation i∂t jψ R i ¼
Hjψ R i jψ R i starting from a small initial state. After a short
time, the gain stays at a stable value gs (green circles) as
(b) (d) (f)
defined in Eq. (4). We emphasize that the stable state
reached is also an eigenstate of Hs . Besides the stable state,
there are another two eigenstates of Hs . Figure 2(d) shows
the real part of the eigenvalues of Hs , with the solid red line
representing the same stable mode as shown in Fig. 1(b)
and the dot-dashed and dashed gray lines denoting the other
two eigenvalues of H s . Figure 2(e) shows the normalized
field amplitudes fjψ A j; jψ B j; jψ C jg and relative phases FIG. 3. Steady-state frequencies (a),(b) and the corresponding
fθA ; θB ; θC g of the stable mode. The fields of the other amplitudes (c),(d) around the NEP5 (marked by the black star). In
two eigenstates are provided in Fig. S5. It is clear that the (c),(d), the solid green, dot-dashed blue, and dashed purple lines
three eigenstates of Hs do not coalesce at the NEP5 . represent the cases without noise, and the corresponding light
Substituting the three eigenstates into Eq. (6), we see that lines represent one typical simulation with noises. The chirality
function Λ (e) and the corresponding standard deviations σ (f) are
χ is nonzero at the NEP5 [the red line in Fig. 2(a)]. In other
shown as a function of the encircling radius r. In (e),(f), the open
words, the eigenbasis of Hs is still complete at the NEP5 , cyan circles and gray aplstars are obtained from simulations of
which is in stark contrast to linear Hamiltonians at conven- over 200 independent noises, and the red and blue lines are
tional EPs. We also calculate the PF factor for the stable for eye guiding. The parameters are ωB ¼ 1.25, ωC ¼ 1.59,
state as shown with the red line in Fig. 2(b). These results lB ¼ 3.06, gC ¼ 1.40, κ 2 ¼ 2.18, r ¼ 0.01, T ¼ 20000, and
verify our core conclusion that the completeness of the gA ¼ 5=ð1 þ jψ A j2 Þ − 3. The amplitude of the noise is given
eigenbasis can be recovered and the PF is finite with a by D ¼ 0.2.

266901-4
 PHYSICAL REVIEW LETTERS 130, 266901 (2023)

enough. In contrast, the state experiences a nonadiabatic applications such as (quantum) sensing [13,16,31,47,63],
transition at p1 and evolves to the same state when winding chiral state transfer [19–22], etc. [64,65]. In addition, our
in the clockwise direction [see Fig. 3(b)]. To quantify findings will enrich the physics of the EPs of Bloch
the chirality of a parametric path, we adopt [20] and extend Hamiltonians under continuous deformations [66–68]
the definition of chirality function as Λ ¼ ρCW ρCCW , where and provide new insights into the nonlinear non-
the subscripts CW and CCW denote the clockwise and Hermitian systems [69].
counterclockwise directions, respectively. Here the eigen-
state population ρ describes the relative weight of the The authors thank Chong Chen for helpful discussions.
This work is supported by the National Key Re-
occupation coefficients and is defined as ρðtÞ ¼ ½jcu ðtÞj2 −
search and Development Program of China (Grant
jcl ðtÞj2 =½jcu ðtÞj2 þ jcl ðtÞj2 , where cu;l ðtÞ represents the
No. 2022YFA1404900), the National Natural Science
projection of the instantaneous state jψðtÞi on the upper Foundation of China (Grants No. 12274330,
or lower energy state. Figures 3(c) and 3(d) show the No. 12274332), and the Knowledge Innovation Program
evolution of instantaneous amplitudes in the presence of of Wuhan-Shuguang (Grant No. 2022010801020125).
noise (light lines), which are bounded around the case where
there is no noise (bold lines). Different from circling other
linear EPs, wherein the wave amplitude either exponentially
increases or decreases (for passive systems) with time, here *
the wave amplitude maintains almost at the same order since [email protected]
†
[email protected]
the evolution is on the stable state surface. Figures 3(e)
[1] W. D. Heiss, J. Phys. A Math. Theor. 45, 444016 (2012).
and 3(f) show the chirality function Λ and the corresponding [2] I. Rotter, J. Phys. A Math. Theor. 42, 153001 (2009).
standard deviations σ with a fixed noise as a function of the [3] Y. Ashida, Z. Gong, and M. Ueda, Adv. Phys. 69, 249
encircling radius r. Ideally, Λ should be −1 independent of r; (2020).
in the presence of noise, Λ deviates from −1. Compared with [4] M. Berry, Czech. J. Phys. 54, 1039 (2004).
circling an EP in a typical linear system [the blue line in [5] W. D. Heiss, J. Phys. A 37, 2455 (2004).
Fig. 3(e)], the chirality function Λ [the red line in Fig. 3(e)] is [6] C. Dembowski, H.-D. Gräf, H. L. Harney, A. Heine, W. D.
much more robust against noise as it remains around −1 for a Heiss, H. Rehfeld, and A. Richter, Phys. Rev. Lett. 86, 787
much smaller r. Moreover, the standard deviation σ con- (2001).
verges [red line in Fig. 3(f)] as we approach the NEP5 . When [7] W. Tang, X. Jiang, K. Ding, Y.-X. Xiao, Z.-Q. Zhang, C. T.
Chan, and G. Ma, Science 370, 1077 (2020).
r is large, the feedback mechanism is the dominant factor for
[8] S.-B. Lee, J. Yang, S. Moon, S.-Y. Lee, J.-B. Shim, S. W.
the robustness of the chiral state transfer against noise; when Kim, J.-H. Lee, and K. An, Phys. Rev. Lett. 103, 134101
r is small enough, both the feedback mechanism and the (2009).
reviving of the complete basis contribute to the robustness. [9] Y. Choi, S. Kang, S. Lim, W. Kim, J.-R. Kim, J.-H. Lee, and
A more detailed analysis is provided in the Supplemental K. An, Phys. Rev. Lett. 104, 153601 (2010).
Material, Sec. 4 [50]. Here, the parametric steering [10] B. Zhen, C. W. Hsu, Y. Igarashi, L. Lu, I. Kaminer, A. Pick,
process with an ultrasmall r enables the miniaturization S.-L. Chua, J. D. Joannopoulos, and M. Soljačić, Nature
of various key applications based on the chiral state transfer (London) 525, 354 (2015).
[21–25]. [11] T. Gao, E. Estrecho, K. Y. Bliokh, T. C. Liew, M. D. Fraser,
Conclusions.—In summary, we propose NEPs with the S. Brodbeck, M. Kamp, C. Schneider, S. Höfling, Y.
Yamamoto, F. Nori, Y. S. Kivshar, A. G. Truscott, R. G.
critical features of EPs. It is remarkable to find that the
Dall, and E. A. Ostrovskaya, Nature (London) 526, 554
completeness of the eigenbasis at the NEP recovers in (2015).
temporal dynamics, and this fact is also corroborated by a [12] A. Regensburger, C. Bersch, M.-A. Miri, G. Onishchukov,
finite PF. The adverse consequences due to the loss of D. N. Christodoulides, and U. Peschel, Nature (London)
completeness are largely degraded from a fundamental 488, 167 (2012).
aspect. Since nonlinearity fundamentally changes the [13] H. Hodaei, A. U. Hassan, S. Wittek, H. Garcia-Gracia, R.
dimension and dynamics of the system, we name it NEP El-Ganainy, D. N. Christodoulides, and M. Khajavikhan,
to distinguish it from conventional EPs. Our model can be Nature (London) 548, 187 (2017).
implemented within diverse classical and quantum systems, [14] K. Ding, G. Ma, M. Xiao, Z. Q. Zhang, and C. T. Chan,
and we show their existence in the circuits in the Phys. Rev. X 6, 021007 (2016).
Supplementary Material, Sec. 5. Besides the circuit system, [15] S. Wang, B. Hou, W. Lu, Y. Chen, Z. Q. Zhang, and C. T.
Chan, Nat. Commun. 10, 832 (2019).
other promising platforms include microwave cavities
[16] W. Chen, Kaya Özdemir, G. Zhao, J. Wiersig, and L. Yang,
[6,16], exciton-polariton billiards [11], optical micro- Nature (London) 548, 192 (2017).
cavities [8], coupled atom-cavity systems [9], cold atoms [17] L. Feng, Z. J. Wong, R.-M. Ma, Y. Wang, and X. Zhang,
[59,60], nitrogen-vacancy centers [61], superconduc- Science 346, 972 (2014).
ting circuits [62], etc. [7,10,12,63]. The nonlinear EPs [18] H. Hodaei, M.-A. Miri, M. Heinrich, D. N. Christodoulides,
discussed in our work exhibit great potential in EP-related and M. Khajavikhan, Science 346, 975 (2014).

266901-5
 PHYSICAL REVIEW LETTERS 130, 266901 (2023)

[19] J. Doppler, A. A. Mailybaev, J. Böhm, U. Kuhl, A. Girschik, [44] S.-Y. Lee, J.-W. Ryu, J.-B. Shim, S.-B. Lee, S. W. Kim, and
F. Libisch, T. J. Milburn, P. Rabl, N. Moiseyev, and S. K. An, Phys. Rev. A 78, 015805 (2008).
Rotter, Nature (London) 537, 76 (2016). [45] H. Wang, Y. H. Lai, Z. Yuan, M. G. Suh, and K. Vahala, Nat.
[20] H. Nasari, G. Lopez-Galmiche, H. E. Lopez-Aviles, A. Commun. 11, 1610 (2020).
Schumer, A. U. Hassan, Q. Zhong, S. Rotter, P. [46] C. Chen, L. Jin, and R.-B. Liu, New J. Phys. 21, 083002
LiKamWa, D. N. Christodoulides, and M. Khajavikhan, (2019).
Nature (London) 605, 256 (2022). [47] R. Kononchuk, J. Cai, F. Ellis, R. Thevamaran, and T.
[21] H. Xu, D. Mason, L. Jiang, and J. G. E. Harris, Nature Kottos, Nature (London) 607, 697 (2022).
(London) 537, 80 (2016). [48] A. E. Siegman, Lasers (University Science Books,
[22] A. U. Hassan, B. Zhen, M. Soljačić, M. Khajavikhan, and Sausalito, California, 1986).
D. N. Christodoulides, Phys. Rev. Lett. 118, 093002 (2017). [49] C. M. Bender and P. D. Mannheim, Phys. Lett. A 374, 1616
[23] X. L. Zhang, T. Jiang, and C. T. Chan, Light Sci. Appl. 8, 88 (2010).
(2019). [50] See Supplemental Material at http://link.aps.org/
[24] Y. Choi, C. Hahn, J. W. Yoon, S. H. Song, and P. Berini, Nat. supplemental/10.1103/PhysRevLett.130.266901, which in-
Commun. 8, 14154 (2017). cludes Refs. [19–25,29–31,43–45,51–55], for further
[25] B. Peng, Şahin Kaya Özdemir, M. Liertzer, W. Chen, J. details.
Kramer, H. Yılmaz, J. Wiersig, S. Rotter, and L. Yang, Proc. [51] K. J. H. Peters and S. R. K. Rodriguez, Phys. Rev. Lett. 129,
Natl. Acad. Sci. U.S.A. 113, 6845 (2016). 013901 (2022).
[26] H.-Z. Chen, T. Liu, H.-Y. Luan, R.-J. Liu, X.-Y. Wang, X.-F. [52] M. Lyubarov, Y. Lumer, A. Dikopoltsev, E. Lustig, Y.
Zhu, Y.-B. Li, Z.-M. Gu, S.-J. Liang, H. Gao, L. Lu, L. Ge, Sharabi, and M. Segev, Science 377, 425 (2022).
S. Zhang, J. Zhu, and R.-M. Ma, Nat. Phys. 16, 571 (2020). [53] T. J. Milburn, J. Doppler, C. A. Holmes, S. Portolan, S.
[27] D. Smirnova, D. Leykam, Y. Chong, and Y. Kivshar, Appl. Rotter, and P. Rabl, Phys. Rev. A 92, 052124 (2015).
Phys. Rev. 7, 021306 (2020). [54] J. Schindler, Z. Lin, J. M. Lee, H. Ramezani, F. M. Ellis, and
[28] V. V. Konotop, J. Yang, and D. A. Zezyulin, Rev. Mod. T. Kottos, J. Phys. A Math. Theor. 45, 444029 (2012).
Phys. 88, 035002 (2016). [55] https://www.analog.com/en/design-center/design-tools-and-
[29] S. Assawaworrarit, X. Yu, and S. Fan, Nature (London) 546, calculators/ltspice-simulator.html.
387 (2017). [56] H.-K. Lau and A. A. Clerk, Nat. Commun. 9, 4320
[30] H. Wang, S. Assawaworrarit, and S. Fan, Opt. Lett. 44, 638 (2018).
(2019). [57] A. Cerjan and A. D. Stone, Phys. Rev. A 90, 013840
[31] K. Bai, L. Fang, T.-R. Liu, J.-Z. Li, D. Wan, and M. Xiao, (2014).
Natl. Sci. Rev. 10, nwac259 (2022). [58] X.-L. Zhang, S. Wang, B. Hou, and C. T. Chan, Phys. Rev.
[32] S. Mandal, R. Banerjee, E. A. Ostrovskaya, and T. C. H. X 8, 021066 (2018).
Liew, Phys. Rev. Lett. 125, 123902 (2020). [59] J. Li, A. K. Harter, J. Liu, L. de Melo, Y. N. Joglekar, and L.
[33] B. Zhu, Q. Wang, D. Leykam, H. Xue, Q. J. Wang, and Y. D. Luo, Nat. Commun. 10, 855 (2019).
Chong, Phys. Rev. Lett. 129, 013903 (2022). [60] Y. Jiang, Y. Mei, Y. Zuo, Y. Zhai, J. Li, J. Wen, and S. Du,
[34] R. Banerjee, S. Mandal, and T. C. H. Liew, Phys. Rev. Lett. Phys. Rev. Lett. 123, 193604 (2019).
124, 063901 (2020). [61] W. Liu, Y. Wu, C.-K. Duan, X. Rong, and J. Du, Phys. Rev.
[35] M. S. Kirsch, Y. Zhang, M. Kremer, L. J. Maczewsky, S. K. Lett. 126, 170506 (2021).
Ivanov, Y. V. Kartashov, L. Torner, D. Bauer, A. Szameit, [62] M. Naghiloo, M. Abbasi, Y. N. Joglekar, and K. W. Murch,
and M. Heinrich, Nat. Phys. 17, 995 (2021). Nat. Phys. 15, 1232 (2019).
[36] D. Smirnova, D. Leykam, Y. Chong, and Y. Kivshar, Appl. [63] L. Ding, K. Shi, Q. Zhang, D. Shen, X. Zhang, and W.
Phys. Rev. 7, 021306 (2020). Zhang, Phys. Rev. Lett. 126, 083604 (2021).
[37] S. Mittal, G. Moille, K. Srinivasan, Y. K. Chembo, and M. [64] M. Brandstetter, M. Liertzer, C. Deutsch, P. Klang, J.
Hafezi, Nat. Phys. 17, 1169 (2021). Schöberl, H. E. Türeci, G. Strasser, K. Unterrainer, and S.
[38] S. Mittal, E. A. Goldschmidt, and M. Hafezi, Nature Rotter, Nat. Commun. 5, 4034 (2014).
(London) 561, 502 (2018). [65] B. Peng, K. Özdemir, S. Rotter, H. Yilmaz, M. Liertzer, F.
[39] S. Xia, D. Kaltsas, D. Song, I. Komis, J. Xu, A. Szameit, Monifi, C. M. Bender, F. Nori, and L. Yang, Science 346,
H. Buljan, K. G. Makris, and Z. Chen, Science 372, 72 328 (2014).
(2021). [66] B. Zhen, C. W. Hsu, Y. Igarashi, L. Lu, I. Kaminer, A. Pick,
[40] Z. Lin, A. Pick, M. Lončar, and A. W. Rodriguez, Phys. Rev. S.-L. Chua, J. D. Joannopoulos, and M. Soljačić, Nature
Lett. 117, 107402 (2016). (London) 525, 354 (2015).
[41] A. Pick, B. Zhen, O. D. Miller, C. W. Hsu, F. Hernandez, [67] Y. Xu, S.-T. Wang, and L.-M. Duan, Phys. Rev. Lett. 118,
A. W. Rodriguez, M. Soljačić, and S. G. Johnson, Opt. 045701 (2017).
Express 25, 12325 (2017). [68] Z. Yang, C.-K. Chiu, C. Fang, and J. Hu, Phys. Rev. Lett.
[42] H. Jing, K. Özdemir, H. Lü, and F. Nori, Sci. Rep. 7, 3386 124, 186402 (2020).
(2017). [69] T. Karzig, C.-E. Bardyn, N. H. Lindner, and G. Refael,
[43] A. E. Siegman, Phys. Rev. A 39, 1253 (1989). Phys. Rev. X 5, 031001 (2015).

266901-6