WorksPhysics and Astronomy
Exceptional-point Sensing with Quantum Advantage
Abstract
Exceptional-point sensors exploit singular non-Hermitian dynamics to enhance their response to weak perturbations. Recently, however, a no-go result has been established for a gain-loss-balanced dimer coupled to independent, phase-insensitive reservoirs: the exceptional-point susceptibility amplifies the signal and reservoir noise equally, precluding any quantum advantage. Here, we circumvent this limitation by engineering non-Hermitian dynamics in particle-hole Nambu space. A single common reservoir acts on modes forming a quantum-mechanics-free subsystem and generates an enhanced response protected from additional noise. We demonstrate quantum-metrological advantage for weak-signal sensing and static-parameter estimation. A continuous interpolation of the operator basis at fixed exceptional-point drift, with a singular midpoint, reveals how reservoir noise vanishes toward the protected limit. Our results provide a route to exceptional-point sensing with genuine quantum-metrological advantage through reservoir engineering.
Introduction
Introduction— Exceptional points (EPs) are non-Hermitian degeneracies at which eigenvalues and eigenvectors coalesce [1]. Near a second-order EP, a generic perturbation produces a square-root resonance splitting, motivating sensing proposals and experiments on different platforms [2, 3, 4, 5, 6, 7, 8, 9, 10, 11]. Enhanced spectral response, however, does not by itself imply improved measurement precision [12, 13, 14]: an effective non-Hermitian equation of motion is only the reduced description of a larger open quantum system. The physical implementation of gain, loss, drive, and readout also determines the quantum fluctuations against which the signal must be resolved.
Several studies have shown that the enhanced response near an EP can be accompanied by a corresponding enhancement of quantum noise [13, 14, 15, 16, 17, 18, 19, 20]. Loughlin and Sudhir made this limitation explicit for a gain-loss-balanced dimer [21]: noise from its independent gain and loss reservoirs is amplified by the same EP susceptibility as the signal, eliminating the apparent quantum-limited advantage.
Here, we show that this no-go result can be circumvented using a single engineered reservoir coupled to particle-hole Nambu modes that form a quantum-mechanics-free subsystem (QMFS) [22, 23, 24, 25]. The engineered reservoir generates the EP-enhanced signal response without concomitantly amplifying the reservoir noise: the correlated Langevin forces cancel exactly in the measured collective observable while adding in its conjugate partner. We demonstrate the resulting quantum advantage for both weak-signal sensing and static-parameter estimation, establish its equivalence to the auxiliary-cavity realizations of Refs. [26, 27, 28, 29, 30], and identify the general Hamiltonian-level principle underlying this construction. This provides a genuinely dissipative route to EP-enhanced quantum sensing. It differs from coherent QMFS emulation without engineered dissipation [31]; a recent four-mode superconducting implementation demonstrated enhanced sensitivity in a non-quantum-limited regime [32].
Gain-loss-balanced dimer
We first give a brief review of Loughlin and Sudhir’s no go theorem for a gain-loss-balanced dimer with independent, phase-insensitive reservoir and focus on the weak-signal measurement. The non-Hermitian dynamics for the dimer is
\dot{b}=\gamma b-g a.
The normal mode frequencies are and coalesce at the EP with . The corresponding collective EP eigenmode is the difference of and , namely,
A minimal Hamiltonian embedding for such non-Hermitian dynamics requires a coherent coupling between and , promoting the amplitudes to the quantum operators,
and also couplings to two independent reservoirs with input noise operators and , respectively. The resulting quantum Langevin equations of motion are given by
\dot{\hat{b}}=\gamma\hat{b}-g\hat{a}+\sqrt{2\gamma}\hat{b}_{\mathrm{in}},
where we have included a weak signal coupled to the mode via a coupling strength of : . At the EP with , the collective EP mode obeys
which, in the frequency domain, reads
The first term in the bracket is the signal term while the second one is the noise term. Both terms are multiplied by , which is divergent at the EP eigen-frequency . This shared amplification factor for both terms is the no-go mechanism: increasing the EP susceptibility amplifies both the signal and the quantum noise, preventing any enhancement of the signal-to-noise ratio in the quantum-limited regime.
Non-Hermitian dynamics in Nambu space— The Hamiltonian embedding of the non-Hermitian dynamics is drastically different if we consider the particle-hole Nambu space instead of the space spanned by . As illustrated in Figure 1, the minimal Hamiltonian embedding for only requires a common reservoir coupled to the EP mode , without the need of additional coherent coupling between and (the derivation is given in Appendix A). The corresponding Langevin equations are

Figure 1. Hamiltonian embeddings for the gain-loss-balanced dimer (left) and the noise-protected EP mode (right). For the latter, one common reservoir with input/output fields is coupled to and ; the common input noise cancels from the EP mode , allowing the EP response to yield a quantum advantage.
The homogeneous part of the above equation again has a second-order EP: the two eigenfrequencies coalesce at 0, and only one normal mode remains independent.
In contrast to Eq. (5) and Eq. (6), both and are driven by the same noise operator from the common reservoir. Therefore, the new EP mode in the Nambu space, defined as
satisfies , which in the frequency domain is
Therefore, such a quantum-noise-protected EP mode has a divergent signal response at the EP pole without additional noise. In this example, this ideal divergence at is regularized by finite time, residual loss, and nonlinearities, which are practical imperfections rather than the fundamental limitation that amplifies both signal and noise in the gain-loss-balanced dimer discussed earlier and also in Ref. [21].
Quantum advantage for weak-signal detection— Eq. (3) and Eq. (7) already show the EP mode defined in the Nambu space demonstrates quantum-noise protection compared to the one for the gain-loss-balanced dimmer. We now show explicitly that such quantum advantage persists when we finally read out the outgoing field of the reservoir. According to the standard input-output relation [33], we have
Using the homodyne detection to measure the phase quadrature of outgoing field, defined as in the frequency domain, we obtain
For vacuum input noise, in the single-sided convention [34, 35], and therefore the weak-signal measurement imprecision, normalized by the signal response, is
The imprecision vanishes as approaches instead of being flattened out for the gain-loss-balanced dimer.
Quantum advantage for parameter estimation— We now consider the quantum advantage for parameter estimation in the Nambu space. Let a small parameter to be estimated perturbs the off-diagonal coupling:
Here we have neglected additional noise terms of the order , which does not change the conclusion on the quantum advantage (details are provided in Appendix B). Introducing
which is conjugate to , the above equations can be rewritten as
Eliminating leads to
where we have defined . The resonant frequencies are which shows the standard square-root splitting common for EP sensors. Together with Eq. (8), the resulting noise spectrum for the output phase quadrature of around is
where we have assumed that the frequency of interest is sufficiently close to and in the approximation. It can be converted into the following frequency noise spectrum [21], given a coherent output amplitude ,
If we measure the frequency splitting, the sensitivity is given by
The resulting parameter-estimation imprecision , scales as
This shows the quantum advantage for the EP sensing, which is in contrast to the gain-loss-balanced dimer as shown in Ref. [21], the imprecision is independent of , namely,
which is because the near-resonant frequency noise spectrum amplitude is proportional to and exactly cancels the EP sensitivity scaling.
Working principle— The quantum advantage discussed above is intimately related to QND measurements with QMFS, which underlies its working principle. This can be better understood by considering the EP mode and its conjugate pair , which we defined in Eq. (6) and Eq. (13), respectively. They satisfy . The continuum field of the common reservoir for and is only coupled to via , so that we have . Since there is no coherent coupling between and and no free Hamiltonian for , is commuting with itself at different times, namely,
Therefore, the Hermitian quadratures of : and are indeed (QND) observables which form the QMFS [22]. As illustrated in Figure 2, such a Hamiltonian structure also means that the coupling of to the common reservoir not only induces a one-way coupling from to since has no dependence on . While will contain since , which means the input noise from the reservoir is only driving without directly driving . This makes the EP mode protected from quantum noise of the reservoir, which is the essence of noise-protected EP sensing.

Figure 2. Illustration of the working principle for a quantum-noise protected EP sensor. The measured EP variable carries the signal, while common-reservoir input noise is routed to the conjugate sector . The one-way coupling protects from the input noise; a perturbation closes the reverse path and produces the square-root mode frequency splitting.
Minimal realization of a common reservoir— The Hamiltonian embedding for can be realized by coherently coupling the EP mode to an auxiliary mode with a fast dynamics:
and is in turn coupled to the external continuum field which defines and . The corresponding equations of motion for and are
Adiabatically eliminating the fast auxiliary mode by ignoring in Eq. (25), we have , leading to
where . Identifying with , they recover the equations of motion shown in Eq. (11) and Eq. (12). Such a PT-symmetric EP sensor model was first proposed in Refs. [26, 27, 28, 29, 30], which was motivated by reducing the quantum noise of gravitational-wave detectors and dark matter search. This has also been the main mechanism for realising dissipative pairing interactions in bosonic lattice [36].
Interpolating between Hamiltonian embeddings— Interestingly, there is a continuous interpolation between the gain-loss-balanced dimer and the Nambu-space model via a fixed EP drift by continuously varying the operator basis. For , define
Thus in the gain-loss-balanced dimer and in the Nambu-space model. With the signal temporarily omitted, impose the equations of motion:
where and are uncorrelated vacuum inputs, each with single-sided quadrature spectrum one. Since and , the system dynamics retains the same second-order EP throughout. The interpolation is not canonical: , so the commutator matrix is and . The EP variable therefore becomes a QMFS variable only at the particle-hole endpoint with .
For canonical input operators with commutator matrix , preservation of requires
which fixes the noise commutator but does not completely fix . One possible spectral factorization of the commutator matrix is
where , and with .
Restoring the weak signal sensing gives , where . In the single-sided vacuum convention used above, the phase-quadrature force has spectrum
It decreases monotonically from to zero, vanishing as near the protected endpoint. Thus the reservoir force is suppressed while the EP susceptibility and signal coupling remain fixed. This spectrum characterizes the chosen reservoir realization, rather than a universal noise bound inferred from alone.
For , an explicit input-output realization uses the coherent Hamiltonian
and two independent continuum fields of the reservoir , with and . Each field has the free Hamiltonian and propagation convention of Appendix A. Define the system coupling combinations
and their corresponding interaction Hamiltonian with the continuum field is given by
The Heisenberg equation for any system operator is
With and solution for the continuum fields, we arrive at Eq. (29), including exactly the noise matrix in Eq. (31). The sign of fixes the reservoir phase to match that matrix.
The coherent term combines beam-splitter and two-mode-squeezing interactions. At , it reduces to Eq. (2) with , while and : the independent-reservoir dimer is recovered. At , both and vanish, leaving . This recovers the common-reservoir interaction; , so the reservoir force on vanishes.
One subtlety arises when . In this case, we have is Hermitian, but the basis discards its conjugate quadrature . Then the commutation matrix is singular, and imposing the same gives the non-Hermitian equation , which cannot be a physical quadrature equation. In addition, the divergences of and as when are also the signatures for the failure of the reduced two-dimensional description —so a regular interpolation must retain missing conjugate quadrature and use an enlarged drift matrix. Away from this singularity, the explicit embeddings demonstrate how identical EP drift matrices can support a quantum-noise behavior with metrological advantages.
Discussion and conclusion— The Loughlin–Sudhir no-go theorem applies to a gain-loss-balanced EP dimer whose signal mode is directly driven by independent, phase-insensitive reservoir noise. The common-reservoir sensor shows that this is not a universal limitation of EP metrology: its EP variable forms a protected quantum-mechanics-free subspace, while reservoir noise drives only the conjugate partner. Earlier three-mode PT-symmetric proposals realize this model in the fast-auxiliary-mode limit. The fixed-drift comparison further identifies the Hamiltonian and reservoir embedding, together with the operator commutators, as essential to the quantum-noise behavior; the non-Hermitian drift alone does not determine the metrological performance. Practical imperfections such as finite observation time, internal loss, and nonlinear saturation will bound the achievable enhancement, but they are distinct from the noise projection responsible for the independent-reservoir no-go result.
*Acknowledgement—*Y. M. and C. W is supported by the National Natural Science Foundation of China under Grant No. 12474481 and No. 12441503, National Key R&D Program of China (2023YFC2205801). X. G., D. Z., M. L. Y., L., and H. M. are supported by the National Natural Science Foundation of China under Grant No. 12441503 and National Key R&D Program of China (2023YFC2205800).
Minimal continuum embedding
Here we provide the details for the Langevin equations Eqs. (11)–(12). We use a continuum field with and propagation speed set to unity. The local ininteraction and free-field Hamiltonians are [33, 37]
Here is a continuum coupling, distinct from the coherent dimer coupling . The field equation is
Integrating across the interaction point gives
With , , and the symmetric junction convention,
The system equations consequently read
Setting reproduces Eqs. (11)–(12) without the signal, and . Adding supplies the signal term. This normalization fixes both damping and readout consistently.
Physical completion of the perturbed sensor and its noise scaling
In this appendix, we will show the exact Markovian input-output realization of Eqs. (11) and (12) for without neglecting the additional noise term of the order of . Define
Introduce independent continuum fields and , each with the free Hamiltonian and propagation convention of Appendix A, and with vacuum inputs and . Their equal-time commutators are , with all interchannel commutators zero. In terms of and , choose the local interaction
The relative sign fixes a reservoir phase convention. The equations of motion for the continuum fields are
Integrating across the junction, with inputs at and outputs at , yields
The symmetric junction fields are consequently
Since and , the system Heisenberg equations before eliminating the local fields are
Substituting these junction fields and returning to gives the Langevin equations
Compared with (11) and (12), we have an additional noise terms from .
Equivalently, we can rewrite the above equations of motion using and and the monitored output are
At the second channel disappears and the protected common-reservoir model is recovered.
In the frequency domain, the solution of away from the poles is
Here denotes the Fourier transform of , equivalently , and we have defined
The output of the readout mode follows form the input-output relation: , which reduces to
where the output transfer coefficients are
They obey . With independent vacuum inputs, the two output quadratures have equal noise spectra and no anomalous field correlation; this realization is phase insensitive. In the single-sided vacuum convention, the phase quadrature of has the following noise spectrum
It recovers the scaling with respect to shown in Eq. (17) with an unimportant factor of 2 difference, when we consider the frequency sufficiently close to the EP.
References
References
- [1]M.-A. Miri and A. Alù, Science 363, eaar7709 (2019).
- [2]J. Wiersig, Physical Review Letters 112, 203901 (2014).
- [3]W. Chen, Ş. K. Özdemir, G. Zhao, J. Wiersig, and L. Yang, Nature 548, 192 (2017).
- [4]X.-W. Luo, C. Zhang, and S. Du, Physical Review Letters 128, 173602 (2022).
- [5]J. Wiersig, Physical Review Research 5, 033042 (2023).
- [6]Z. Sun, X. Zhao, L. Zhang, C. Liang, B. Li, W. Lu, W. Zhang, R. You, J. Zhao, and Z. You, Advanced Optical Materials 12, 2302420 (2024).
- [7]K. Bai, T.-R. Liu, L. Fang, J.-Z. Li, C. Lin, D. Wan, and M. Xiao, Physical Review Letters 132, 073802 (2024).
- [8]L. Hu, Y. Li, K. Zhu, H. Chen, and Z. Guo, Applied Physics Letters 126, 091702 (2025).
- [9]Y.-P. Ruan, J.-S. Tang, Z. Li, H. Wu, W. Zhou, L. Xiao, J. Chen, S.-J. Ge, W. Hu, H. Zhang, C.-W. Qiu, W. Liu, H. Jing, Y.-Q. Lu, and K. Xia, Nature Photonics 19, 109 (2025).
- [10]D.-Y. Chen, L. Dong, and Q.-A. Huang, Nature Electronics 9, 489 (2026).
- [11]A. Almanakly, R. Assouly, H. H. Kang, M. Gingras, B. M. Niedzielski, H. Stickler, M. E. Schwartz, K. Serniak, M. Hays, J. A. Grover, and W. D. Oliver, Physical Review Letters 136, 183601 (2026).
- [12]W. Langbein, Physical Review A 98, 023805 (2018).
- [13]H.-K. Lau and A. A. Clerk, Nature Communications 9, 4320 (2018).
- [14]C. Chen, L. Jin, and R.-B. Liu, New Journal of Physics 21, 083002 (2019).
- [15]M. Zhang, W. Sweeney, C. W. Hsu, L. Yang, A. D. Stone, and L. Jiang, Physical Review Letters 123, 180501 (2019).
- [16]H. Wang, Y.-H. Lai, Z. Yuan, M.-G. Suh, and K. Vahala, Nature Communications 11, 1610 (2020).
- [17]W. Wang, Y. Zhai, D. Liu, X. Jiang, S. Vashahri Ghamari, and J. Wen, Quantum Science and Technology 10, 025016 (2025).
- [18]C. Wang, D. Hu, S. Zorzetti, A. Grassellino, A. Romanenko, and Z. Zhang, Physical Review Letters 136, 250801 (2026).
- [19]J. Wiersig and S. Rotter, Reports on Progress in Physics 89, 067501 (2026).
- [20]W. Wang, Y. Liang, P. Han, X. Wu, D. Liu, and M. Gu, (2026), arXiv:2607.18799 [quant-ph].arxiv.org/abs/2607.18799
- [21]H. A. Loughlin and V. Sudhir, Physical Review Letters 132, 243601 (2024).
- [22]M. Tsang and C. M. Caves, Physical Review X 2, 031016 (2012).
- [23]M. J. Woolley and A. A. Clerk, Physical Review A 87, 063846 (2013).
- [24]C. F. Ockeloen-Korppi, E. Damskägg, J.-M. Pirkkalainen, A. A. Clerk, M. J. Woolley, and M. A. Sillanpää, Physical Review Letters 117, 140401 (2016).
- [25]L. Mercier de Lépinay, C. F. Ockeloen-Korppi, M. J. Woolley, and M. A. Sillanpää, Science 372, 625 (2021).
- [26]X. Li, M. Goryachev, Y. Ma, M. E. Tobar, C. Zhao, R. X. Adhikari, and Y. Chen, arXiv:2012.00836 [quant-ph].arxiv.org/abs/2012.00836
- [27]X. Li, J. Smetana, A. S. Ubhi, J. Bentley, Y. Chen, Y. Ma, H. Miao, and D. Martynov, Physical Review D 103 (2021).
- [28]C. Wang, C. Zhao, X. Li, E. Zhou, H. Miao, Y. Chen, and Y. Ma, Physical Review D 106 (2022).
- [29]Y. Chen, M. Jiang, Y. Ma, J. Shu, and Y. Yang, Phys. Rev. Res. 4, 023015 (2022).
- [30]Y. Chen, C. Li, Y. Liu, J. Shu, Y. Yang, and Y. Zeng, Reports on Progress in Physics 88, 057601 (2025).
- [31]Y.-X. Wang and A. A. Clerk, Physical Review A 99, 063834 (2019).
- [32]R. Assouly, H. H. Kang, A. Almanakly, M. A. Gingras, B. M. Niedzielski, H. Stickler, M. E. Schwartz, K. Serniak, M. Hays, J. A. Grover, and W. D. Oliver, (2026), arXiv:2606.16060 [quant-ph].arxiv.org/abs/2606.16060
- [33]C. W. Gardiner and M. J. Collett, Physical Review A 31, 3761 (1985).
- [34]V. B. Braginsky and F. Y. Khalili, Quantum Measurement, edited by K. S. Thorne (Cambridge University Press, Cambridge, 1992).
- [35]A. A. Clerk, M. H. Devoret, S. M. Girvin, F. Marquardt, and R. J. Schoelkopf, Reviews of Modern Physics 82, 1155 (2010).
- [36]A. Pocklington, Y.-X. Wang, and A. A. Clerk, Physical Review Letters 130, 123602 (2023).
- [37]Y. Chen, Journal of Physics B: Atomic, Molecular and Optical Physics 46, 104001 (2013).