Non-Degenerate Sensitivity: Theory & Applications

Updated 5 July 2026

Non-degenerate sensitivity is a cross-disciplinary concept that lifts degeneracy to reveal a system's underlying, diagnostically useful response structure.
In quantum optics, it ensures that pump shaping can jointly correct non-collinear and non-degenerate biphoton correlations, preserving key interference patterns.
In variational analysis and NSDP, it underpins the identification of stable active structures, thereby guaranteeing reliable convergence and effective sensitivity analysis.

Non-degenerate sensitivity is not a single universally standardized notion. In current research usage, it denotes a family of situations in which sensitivity remains structurally meaningful once a degenerate description is lifted. In quantum optics, it concerns the persistence of pump-controlled biphoton correlations when the signal and idler are non-degenerate in wavelength or non-collinear in angle. In topological dynamics, it appears as nontrivial sensitivity induced by non-singleton regionally proximal fibers. In variational analysis, it names the stable regime in which nearby approximate critical points identify a smallest active set or active manifold. In nonlinear semidefinite programming, it motivates weaker variants of constraint nondegeneracy that preserve convergence and sensitivity-analysis consequences while relaxing classical basis-dependent linear-independence conditions (Lib et al., 2020, Li et al., 2020, Drusvyatskiy et al., 2012, Andreani et al., 2020).

1. Terminological scope

The expression is best treated as a cross-disciplinary label rather than a single definition. The common thread is that degeneracy obscures the local structure relevant to perturbation, whereas non-degenerate sensitivity singles out a regime in which perturbations produce diagnostically useful behavior.

Domain	Non-degenerate structure	Sensitivity content
SPDC and quantum wavefront shaping	Non-degenerate signal and idler wavelengths; non-collinear emission	Pump shaping still controls coincidence correlations after scattering
Minimal group actions	Regionally proximal classes or fibers with multiple points	$n$ -sensitivity, thick $n$ -sensitivity, and blockily thickly $n$ -sensitivity
Variational analysis	Locally minimal identifiable set; identifiable manifold	Nearby approximate critical points eventually lie in the same active structure
NSDP	Nondegeneracy or weaker kernel-based variants	KKT, convergence, and sensitivity-analysis benefits survive under weaker CQs

This suggests that “non-degenerate” may refer either to the physical configuration itself, as in non-degenerate SPDC, or to the regularity condition that makes sensitivity analysis well posed, as in optimization and NSDP. The literature therefore uses the phrase in a structurally analogous but not definitionally identical way (Lib et al., 2020, Li et al., 2020, Drusvyatskiy et al., 2012, Andreani et al., 2020).

2. Quantum-optical usage: SPDC under scattering

In spontaneous parametric down-conversion, the key non-degenerate-sensitivity question is whether wavelength-separated photons remain jointly correctable by shaping only the classical pump. For a thin crystal pumped monochromatically, the biphoton amplitude is written as $G$ 2 so the two-photon state is tied to the pump angular spectrum and to energy conservation. With a diffuser modeled by $G$ 3 the coincidence rate takes the form $G$ 4 and the crucial identity $G$ 5 maps the coincidence pattern onto the pump-beam intensity. The reported implication is that pump-shaping remains effective even when the entangled photons are both non-collinear and non-degenerate, which is presented as a strong indication of the robustness of the pump-to-correlation correspondence in SPDC (Lib et al., 2020).

The experimental platform is explicit: a continuous-wave 404 nm laser is shaped by a phase-only spatial light modulator, imaged onto a 2 mm long PPKTP nonlinear crystal, and the generated photons are imaged onto a thin polymer-on-glass diffuser with divergence angle $0.25^\circ$ . The pump is measured with a CMOS camera; the photons are measured in the far field with filtered single-photon detectors; coincidence measurements are performed by scanning the signal detector while keeping the idler detector fixed. In the non-degenerate case, filters centered at 766 nm for the idler and 850 nm for the signal yield photons separated by nearly 100 nm, yet the pump speckle, the degenerate two-photon speckle, and the non-degenerate two-photon speckle are reported to have similar shapes. A common misconception is that such wavelength separation should require separate correction of each photon arm. The stated result is the opposite: shaping the pump once can correct the pair jointly, because the two photon scattering amplitudes combine into the pump-frequency response (Lib et al., 2020).

The same paper also identifies the practical limits. In the non-collinear setting, pump shaping works only within the memory effect or isoplanatic patch of the medium. In the non-degenerate setting, the observed speckle scale depends on which arm is scanned and which is fixed; in the reported measurement, the scale is governed by the 850 nm scanning detector. The lower signal-to-noise ratio in the non-degenerate data is attributed to wavelength-dependent losses and to an additional factor-of-two reduction caused by using a simple beam splitter instead of a dichroic mirror (Lib et al., 2020).

A later bulk-crystal Type-I SPDC treatment sharpens the same issue from an imaging perspective. It reports that degenerate SPDC is approximately separable in frequency and transverse momentum, whereas non-degenerate SPDC exhibits strong frequency-angle mixing and wavelength-dependent camera scaling, $G$ 6 In that framework, narrowband filtering of the signal arm improves spatial resolution only in the non-degenerate case, and a frequency-non-degeneracy plus walk-off-aware reconstruction recovers the intrinsic far-field anti-correlation ridge with corrected slope $m \approx -0.994 \pm 0.002$ (Polednicek et al., 9 Oct 2025).

3. Dynamical-systems usage: regionally proximal structure and stronger sensitivity

For minimal actions of a countable discrete group $G$ on a compact metric space, the relevant background objects are the maximal equicontinuous factor $G$ 7 and the regionally proximal relation $Q(X,G)$ . The fiberwise set $G$ 8 measures how far the system is from equicontinuity. In this terminology, the paper does not introduce non-degenerate sensitivity as a separate new definition; instead, it interprets sensitivity as “non-degenerate” when a fiber or regionally proximal class contains enough distinct points to witness separation (Li et al., 2020).

The core structural theorems are exact. Under minimality and the hypothesis that $X\times X$ has a dense set of minimal points, $G$ 9 $Q(X,G)$ 0 and blockily thickly $n$ -sensitivity holds if and only if there exist $x\in X$ and pairwise distinct $n$ 0 such that $n$ 1 is a minimal point of $n$ 2 (Li et al., 2020).

This gives a precise non-degeneracy dictionary. The diagonal case $n$ 3 corresponds to equicontinuity. A nontrivial $n$ 4 yields ordinary sensitivity. Uniformly large fibers yield thick sensitivity. Minimal $n$ 5-tuples inside one fiber yield blockily thick sensitivity. A plausible implication is that “non-degenerate sensitivity” in this setting is best understood as a fiber-size and fiber-geometry statement rather than as an independent chaos notion (Li et al., 2020).

4. Variational analysis: identifiable sets and manifold reduction

In variational analysis, the relevant object is an identifiable set for a set-valued mapping $Q(X,G)$ 1 A subset $n$ 6 is identifiable at $n$ 7 for $n$ 8 if every sequence $Q(X,G)$ 2 eventually satisfies $n$ 9. The central notion is the locally minimal identifiable set, characterized equivalently by being a locally maximal necessary set and by being both identifiable and necessary. The paper explicitly develops this as a framework for distinguishing non-degenerate from degenerate sensitivity behavior (Drusvyatskiy et al., 2012).

The non-degenerate regime is the one in which local sensitivity is concentrated on a smallest stable active structure. For lower semicontinuous $n$ 0, identifiability can be transferred to the subgradient mapping $n$ 1, and nearby approximate subgradients then identify a stable active set. A major structural result states that if $n$ 2 is prox-regular and identifiable, then $Q(X,G)$ 3 locally around $n$ 3. In the manifold case, identifiable manifolds are locally minimal and locally stable, and partial smoothness becomes equivalent to manifold identifiability together with the strong interiority condition $Q(X,G)$ 4 The paper also links the tangent geometry of $n$ 4 to the critical cone $Q(X,G)$ 5 (Drusvyatskiy et al., 2012)

Degenerate sensitivity corresponds to failure of this stabilization. The paper notes that locally minimal identifiable sets may fail to exist even for simple convex functions, and it gives explicit examples, including $Q(X,G)$ 6 This is not merely a failure of strict complementarity. It is a failure of the identifiable preimages $n$ 5 to stabilize to a smallest local set, so no single active manifold captures nearby perturbation behavior (Drusvyatskiy et al., 2012).

5. Nonlinear semidefinite programming: weaker nondegeneracy for sensitivity analysis

For NSDP, $Q(X,G)$ 7 classical constraint nondegeneracy is $Q(X,G)$ 8 If $n$ 6 is an orthonormal basis of $n$ 7, this is equivalent to the surjectivity of $Q(X,G)$ 9 onto $n$ 8, and also to the linear independence of the $n$ 9 derivative vectors $X\times X$ 0 The paper then proves an equivalent all-bases reformulation: nondegeneracy holds if and only if, for every orthonormal basis $0.25^\circ$ 0 of $0.25^\circ$ 1, the $0.25^\circ$ 2 diagonal vectors $0.25^\circ$ 3 are linearly independent (Andreani et al., 2020).

This basis-sensitive reformulation is the entry point for weaker notions. Weak-nondegeneracy requires that for every sequence $0.25^\circ$ 4, there exist orthonormal eigenvector matrices for the $0.25^\circ$ 5 smallest eigenvalues of $0.25^\circ$ 6 with a limit basis $0.25^\circ$ 7 such that $0.25^\circ$ 8 is linearly independent. Weak-Robinson’s CQ is the analogous positive-linear-independence condition. Sparse-nondegeneracy instead exploits a reduced matrix $X\times X$ 1 and asks for linear independence only on structurally relevant entries, while GS-nondegeneracy weakens this further by ignoring entries whose gradients vanish at the point (Andreani et al., 2020).

The stated motivation is explicit: classical nondegeneracy is often too strong for diagonal or block-diagonal structure, whereas the weaker notions are designed to preserve the main algorithmic and sensitivity-analysis benefits of nondegeneracy. The paper reports that weak-Robinson’s CQ is sufficient for KKT at local minimizers and for feasible cluster points of an external penalty method; sparse-nondegeneracy implies Robinson’s CQ; and GS-nondegeneracy is invariant under perturbations $0.25^\circ$ 9 with $m \approx -0.994 \pm 0.002$ 0 and $m \approx -0.994 \pm 0.002$ 1 (Andreani et al., 2020).

The phrase should not be conflated with every occurrence of “non-degenerate” near a sensitivity claim. In atomic-interferometric detector physics, a realistic gain-EIT negative-dispersion medium is realized using non-degenerate Zeeman sublevels of $m \approx -0.994 \pm 0.002$ 2, and the operative result is an enhancement of the quantum-noise-limited sensitivity-bandwidth product on the order of $m \approx -0.994 \pm 0.002$ 3– $m \approx -0.994 \pm 0.002$ 4; here “non-degenerate” refers to Zeeman splitting, not to a general theory of non-degenerate sensitivity (Zhou et al., 2016).

Likewise, in CSP theory, sensitivity means that every tuple in every constraint extends to a solution, and the main theorem states that a finite idempotent algebra $m \approx -0.994 \pm 0.002$ 5 has a $m \approx -0.994 \pm 0.002$ 6 variable near unanimity term operation if and only if every $m \approx -0.994 \pm 0.002$ 7-instance over $m \approx -0.994 \pm 0.002$ 8 is sensitive. This is an algebraic strengthening of local consistency, not a non-degenerate-sensitivity notion in the variational or dynamical sense (Barto et al., 2020).

A further neighboring example arises in non-Hermitian sensing. There the central distinction is between At-ED and Off-ED operation in a double-chain Hatano–Nelson model: exact exceptional deficiency gives $m \approx -0.994 \pm 0.002$ 9, while Off-ED removes the geometric singularity, restores $G$ 0, and preserves exponential sensitivity scaling $G$ 1. This is structurally close to the broader theme that exact degeneracy can spoil usable sensitivity, but the paper frames the problem as a sensitivity-noise trade-off near spectral singularities rather than as “non-degenerate sensitivity” itself (Li et al., 3 Jun 2026).

Across these literatures, the recurring lesson is precise. Degeneracy may increase raw response, but it can also make sensitivity ambiguous, noisy, or structurally unstable. Non-degenerate sensitivity, in the strict or extended senses documented here, identifies the regime in which perturbation response is still governed by a recoverable structure: the pump-frequency response in SPDC, the regionally proximal fiber in minimal dynamics, the identifiable set in nonsmooth optimization, or the reduced kernel geometry in NSDP.