A quantum exceptional point depends on what you measure

What does a mathematical exceptional point actually let you observe? In quantum systems that exchange energy or information with their surroundings, the answer depends on which equations contain the degeneracy—and how those equations connect to a particular measurement.

In a theoretical preprint, Yusui Chen and colleagues examine that connection in two models with environmental memory. Their central result is not a new universal signal. It is a demonstration that the same underlying degeneracy can leave different signatures in different quantities, or lose its most distinctive contribution altogether.

From memory equations to a state

An exceptional point occurs when eigenvalues and their associated eigenvectors merge in a non-Hermitian operator. Such operators arise in descriptions of open quantum systems. Their evolution can contain powers of time multiplying an exponential: a quadratic factor times a decaying exponential, for example, rather than an exponential alone.

But the paper studies exceptional points in equations for coefficients encoding environmental memory. Those coefficients are not themselves automatically measurable properties of the quantum system.

The authors use non-Markovian quantum-state diffusion, or NMQSD. This represents the system through stochastic pure states driven by noise with correlations over time. Averaging the corresponding projectors gives the reduced density operator, which describes the system without separately retaining its environment.

For both models, the researchers rewrite nonlinear coefficient equations as larger, finite linear systems. This exposes their exceptional points and identifies which associated modes the prescribed initial condition actually reaches. They then reconstruct the quantum state and examine whether a chosen output retains the characteristic time dependence.

This separates three questions: does the degeneracy exist, does the preparation reach it, and does the readout reveal it? A divergence of a coefficient also need not establish an exceptional point: the chosen coefficient coordinates can fail while the physical state remains regular.

Qubits: one degeneracy, different signals

The first model comprises qubits sharing a common reservoir, restricted to states with zero or one excitation. The bath begins in its vacuum state, independently of the system, and has a specified exponentially decaying correlation.

Within these restrictions, the relevant memory operator is independent of the noise. This allows the authors to derive a closed, exact master equation without a weak-coupling or memoryless approximation.

They prove that every defective decaying mode in this class retains its full pole order in a suitable unconditional coherence between an excited state and the zero-excitation reference. Here coherence describes the phase-sensitive relationship between those alternatives. “Suitable” matters: the result guarantees an appropriate preparation and readout, not visibility in every observable.

A two-qubit example revisits a known third-order exceptional-point configuration. With the same initial preparation, one phase-sensitive component contains a quadratic-in-time factor multiplying exponential decay. Another contains only a linear factor because its quadratic contribution cancels. The exceptional point has not changed; the projection used to observe it has.

The paper also shows how separately prepared ensembles could reconstruct the memory coefficients when model parameters are known and the amplitude map is invertible. Small singular values and numerical differentiation can amplify errors, limiting that reconstruction.

An oscillator: averaging changes the interpretation

The second model is a position-coupled harmonic oscillator in an optomechanical setting with fixed photon number. There is no cavity drive or optical loss. Unlike the qubit case, its exact memory operator contains a term depending on the noise history.

Two second-order coefficient blocks produce a response containing a linear-in-lag factor times an exponential. This result concerns the conditional displacement response to the mathematical NMQSD noise history, after a long preparation time and at fixed lag. It is not an established applied-force susceptibility or a continuously monitored experimental trajectory.

Other state properties behave differently. On the regular, normalizable branch studied with bath decay slower than the oscillator frequency, a purely algebraic coefficient tail cancels from the Gaussian width.

Complete averaging introduces another distinction: the reduced-state variance includes both the conditional width and the weighted spread of state centers. In one reported example, conditional momentum variance is about 0.48, below the initial vacuum value of 0.5, while center spread raises the total to about 0.56. Conditional narrowing therefore does not establish reduced-state narrowing.

What the work establishes

These are model-specific analytical results supported by numerical consistency and refinement checks, not new experimental observations. The finite-time variance comparisons do not isolate an exceptional-point-specific enhancement, and the coefficient degeneracies do not automatically identify exceptional points of the reduced-state generator or information backflow.

The source is an arXiv preprint; peer-reviewed status is not established here. This account uses its supplied PDF text extraction, which omits figure images and may damage equations, tables or layout. Within those limits, its message is precise: identifying a degeneracy is only the start. Preparation, state reconstruction, readout and averaging determine what it means for a quantum signal.