Some wireless chargers keep themselves tuned without any tracking electronics. The transmitter coil is driven by a switch that flips the drive voltage every time the coil current changes sign. The resulting oscillation locks onto whichever natural frequency ("mode") the coupled coils prefer, so the frequency follows the coupling as a receiver moves. This works well with one receiver. With several receivers, and especially when they sit in symmetric positions, two modes of the coil network can have nearly the same frequency. We conjecture that the switch then hops between them: the drive frequency jitters, output power ripples, and the switches heat up. This is the same failure as chattering in sliding-mode control, where a sign-switching controller oscillates rapidly around its target. We propose to treat it as a spectral problem. Hopping should appear when the frequency gap between the two competing modes becomes smaller than the modes' own linewidth. The fix we test is to open that gap on purpose, through a small deliberate asymmetry in coil geometry chosen the way a numerical analyst regularizes an ill-conditioned Hessian.
The key observation links two of the four directions. For a lossless inductive-capacitive coil network, the mode frequencies are the eigenvalues of the network's quadratic energy function, which is literally a Hessian. Work on Hessian regularization gives a result that carries over directly. Adding a uniform shift ($H+\lambda I$, the "ridge" fix) does nothing to separate two nearly equal eigenvalues. Only a targeted perturbation does that, in the spirit of eigenvalue filtering. In coil terms, adding the same damping or detuning to every receiver cannot stop the hopping. A geometric perturbation aimed at the degenerate pair of modes can, and at a small efficiency cost. If this holds, it gives a design rule for multi-receiver self-tuning power transfer, a regime where the existing theory assumes stability it has not shown.
Core Claim
Parity-time-symmetric wireless power transfer (PT-WPT) balances energy gain in the transmitter against energy loss in the receiver, so that the network's resonant frequencies stay real and self-adjust as coupling changes. Shu and Zhang show this can be done with a nonlinear negative resistor: a half-bridge converter whose gain saturates. That setup delivered about 19 W at about 60% efficiency over roughly 34 m with no retuning (Single-Wire Electric-Field Coupling Power Transmission Using Nonlinear Parity-Time-Symmetric Model with Coupled-Mode Theory). That paper does not analyze closed-loop stability or switch-level chattering.
Zhu et al. extend the coupled-mode model to one transmitter and $n$ receivers. They show that higher-order systems lose their real, stable frequencies under arbitrary coupling unless an exact impedance-matching condition holds, and they never build the matching in hardware (Parity–Time Symmetric Model and Analysis for Stable Multi-Load Wireless Power Transfer). Their work motivates ours: it identifies instability in the multi-load regime but models ideal linear parts with no switching dynamics.
Our claim has three parts. First, a switch that flips on current zero-crossings is a relay feedback law of the form $u=-K\,\mathrm{sgn}(i)$. In a multi-receiver network it chatters between modes whenever two modes overlap within one linewidth. Second, the size of that overlap is set by the local spectral gap of the network's energy Hessian. Third, deliberately breaking the geometric symmetry of the coils is the targeted regularizer that restores a gap floor. Uniform loss or uniform detuning is the ridge regularizer, and it provably cannot.
The mechanism fits in one relation. Let $M$ be the inductance matrix and $K=C^{-1}$ the inverse-capacitance matrix, so that the mode frequencies $\omega$ solve $K v=\omega^2 M v$. Let $P$ be an $M$-orthonormal basis for two nearly degenerate modes near frequency $\omega_0$, and let $\delta K$ be a perturbation of the network. First-order degenerate perturbation theory, the standard result that a perturbation splits a degenerate pair only through its restriction to that pair's subspace, gives
$$\Delta\omega_{12}\approx\frac{1}{2\omega_0}\,\big|\,\mu_1-\mu_2\,\big|,\qquad \mu_{1,2}=\mathrm{eig}\big(P^\top\delta K\,P\big).$$
A uniform shift $\delta K=\lambda M$ makes $P^\top\delta K P=\lambda I$, so it gives $\mu_1=\mu_2$ and no split at all. This is the circuit counterpart of the observation that ridge damping shifts every eigenvalue equally and flattens useful curvature (Regularizing the Hessian Spectrum for Stable Optimization). The split comes only from the traceless part of the projected perturbation.
Our quantitative hypothesis is that mode-hopping largely stops once $\Delta\omega_{12}\gtrsim\omega_0/Q_L$, where $Q_L$ is the loaded quality factor, a measure of how sharply each mode resonates. The reasoning is that a relay oscillator can only tell two modes apart if they are separated by more than the bandwidth over which the loaded tank "remembers" its phase between switching events.
Research Program
The first step is a hybrid model of a transmitter plus two or three receivers. It combines coupled-mode equations for the coils with an explicit event-driven switch: zero-crossing detection, a finite gate delay, and dead-time. Using this model, we will map where mode-hopping occurs across receiver positions. We measure hopping as switching-period jitter and as switching events that land on the wrong mode. We then test whether the boundary of the hopping region follows the predicted curve $\Delta\omega_{12}=\omega_0/Q_L$ better than it follows the exceptional-point boundary. An exceptional point is where two modes merge completely; Zhang et al. give the governing condition as $|\gamma_1-\gamma_2|=2|\mu|$ (Gain-gain and gain-lossless PT-symmetry broken from PT-phase diagram).
We will also record a second quantity: the condition number of the eigenvector matrix. As two modes approach merging, their eigenvectors become nearly parallel. Graefe et al. show that an oscillator with purely real frequencies can still diverge transiently in this situation, because it is non-Hermitian (Classical and quantum dynamics in the (non-Hermitian) Swanson oscillator). That result constrains our design: we cannot assume "unbroken PT phase" means "no bursts". So we check whether hopping bursts line up with transient amplification in that non-normal regime.
The second step is the design problem. We parameterize small geometric asymmetries of the transmitter coil, such as winding offset, an eccentric turn, or an asymmetric ferrite segment. Each is a physical perturbation $\delta K,\delta M$. We then choose the perturbation that minimizes efficiency loss subject to $\Delta\omega_{12}\ge g_{\min}$ over the whole envelope of receiver positions.
The minimum over the envelope is non-smooth and non-convex. The natural solver is the adaptive trust-region eigenvalue filtering of Chen et al., which switches between eigenvalue clamping and absolute-value filtering based on how well its quadratic model is predicting progress. Chen et al. report that it converges orders of magnitude faster than plain clamping on ill-conditioned elastic meshes, at about 6% extra cost per iteration (Trust-Region Eigenvalue Filtering for Projected Newton). Here we reuse that method only as a design solver; it is not the contribution.
The idea of geometric asymmetry comes from Direction 2. Deformed microdisks and asymmetric claddings lift degeneracies of optical modes and shape emission (Research Progress of Gallium Nitride Microdisk Cavity Laser, Structural Asymmetry for Directional Wave Control). Lee et al. show that asymmetric coupling can stand in for balanced gain and loss in a $2\times 2$ passive system (Passive $\mathcal{PT}$-Symmetric Couplers without Complex Optical Potentials). Our use differs from both. Here asymmetry serves neither emission shaping nor replacing gain. It is a minimal, targeted perturbation that guarantees a gap in a multi-mode power network whose gain is still active.
The third step is hardware. We will build a bench prototype with a self-oscillating half-bridge transmitter and two to three receivers on a positioning rig. We fabricate a symmetric transmitter and the optimized asymmetric one, and measure the same quantities as in simulation.
Baselines
The naive reference point is the symmetric multi-receiver PT-WPT system driven by a saturable switching negative resistor: the Shu–Zhang nonlinear PT mechanism (Single-Wire Electric-Field Coupling Power Transmission Using Nonlinear Parity-Time-Symmetric Model with Coupled-Mode Theory) extended to the Zhu multi-load topology. It is the strongest existing self-tuning approach, and its single-link result of about 19 W at about 60% efficiency shows it is practical. There is no reported result on Lacuna for mode-hopping or chattering in multi-receiver PT-WPT, so the comparison numbers must come from our own measurements. Beating this baseline means cutting mode-hopping events and output-power ripple substantially over the receiver-position envelope, while keeping DC-to-load efficiency within a few points of the symmetric design.
The ablation that isolates the key ingredient is uniform regularization at matched cost. We add equal series resistance, or equal capacitive detuning, to every receiver, tuned so the efficiency loss equals that of the geometric design. The perturbation theory predicts this leaves the degenerate pair unsplit. If uniform regularization suppresses hopping as well as the targeted design does, our core claim is false.
The strongest non-geometric competitor is the topological approach reviewed by Wu et al. It uses chains of coils with alternating coupling strengths (Su-Schrieffer-Heeger chains) and 2D corner states to suppress frequency splitting, and it maintains efficiency under up to 30% coil displacement (Research progress of robust magnetic resonance wireless power transfer based on topological physics). The review covers static configurations and does not examine switch-driven chattering. We will build the smallest topological array that fits the same footprint and coil count, drive it with the same half-bridge, and compare hopping rate and efficiency. To beat it, the asymmetric design should match its hopping suppression with fewer coils or higher efficiency.
On the control side, the strongest chattering remedy is to make the drive continuous with a super-twisting law. This integrates the sign function so that the actuator receives a continuous command; Kuang et al. use it on a wafer stage, reaching 4.95 μm RMS error versus 9.12 μm for standard sliding-mode control (Precise Motion Control of Wafer Stages via Adaptive Neural Network and Fractional-Order Super-Twisting Algorithm). Those figures come from motion control, not wireless power. We will implement super-twisting gate timing in the same digital controller with the same tuning budget, and compare both separately and combined with the geometric fix.
Why not just…?
Why not just lock the frequency with a phase-locked loop onto one mode? A phase-locked loop is a feedback circuit that forces the drive to track a chosen frequency, and forcing it onto one mode could suppress hopping. This is a real threat. The cost is losing the self-tuning that motivates PT-WPT in the first place: the tracking loop lags fast coupling changes, and it can lock onto the wrong branch when modes swap near a merging point. We will include a phase-locked mode-lock baseline and test it on fast receiver-motion profiles. If it matches the geometric design's ripple and efficiency under motion, we will report that the geometric approach earns its place only where tracking hardware is unacceptable.
Why not just add damping or detune every receiver equally? Degenerate perturbation theory says a perturbation proportional to identity cannot split a degenerate pair. This is the circuit version of ridge regularization's known weakness. The matched-cost uniform ablation tests the claim directly. A positive result there would falsify the proposal rather than merely weaken it.
Why not just smooth the switching with a boundary layer or a linear amplifier? A boundary layer replaces the sign function with a smooth ramp near zero, and a linear amplifier avoids switching altogether. Both remove the relay, so they should remove hopping. The chattering literature notes that boundary-layer smoothing gives up the switching law's robustness to disturbances (Parallel-Connected Inverter Control Method in Material Micromachining). In power stages it also moves the switches out of efficient hard-switching operation. We will measure the efficiency penalty of a smoothed drive against the geometric fix at equal hopping suppression.
Why not just detune receiver capacitors differently, instead of changing geometry? Unequal capacitor detuning is a diagonal but non-uniform perturbation, so it can split the pair. It is a legitimate competitor. It also moves each receiver off the shared operating frequency, and the split it produces depends on position rather than being fixed. We will optimize it under the same gap constraint and compare the efficiency cost across the position envelope. If it performs as well as geometry, the claim narrows to "targeted spectral perturbation, whatever its physical form", which is still the core finding.
Proposed Experiments
The decisive simulation check is a map over receiver position and gate delay. On it, the observed hopping boundary should track $\Delta\omega_{12}=\omega_0/Q_L$ better than it tracks the exceptional-point condition. The decisive hardware check has three parts. At matched efficiency loss, the geometrically asymmetric transmitter should reduce switching-period jitter and output-power ripple across the envelope, while uniform damping should not. The asymmetric design should also match or approach the topological array with fewer coils.
Secondary checks go beyond the main comparison. They test whether hopping bursts line up with high eigenvector condition numbers, as the Swanson-oscillator result would predict. They test whether super-twisting gate timing adds to the geometric fix or makes it redundant. Finally, they report the trust-region design solver's convergence against plain eigenvalue clamping on the design problem, along with DC-to-load efficiency and switch temperature for every configuration.
References
- Single-Wire Electric-Field Coupling Power Transmission Using Nonlinear Parity-Time-Symmetric Model with Coupled-Mode Theory
- Parity–Time Symmetric Model and Analysis for Stable Multi-Load Wireless Power Transfer
- A note on Platt's probabilistic outputs for support vector machines
- Gain-gain and gain-lossless PT-symmetry broken from PT-phase diagram
- Classical and quantum dynamics in the (non-Hermitian) Swanson oscillator
- Trust-Region Eigenvalue Filtering for Projected Newton
- Research Progress of Gallium Nitride Microdisk Cavity Laser
- Engineering of metal-clad optical nanocavity to optimize coupling with integrated waveguides
- Passive<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi mathvariant="script">PT</mml:mi></mml:math>-symmetric couplers without complex optical potentials
- Research progress of robust magnetic resonance wireless power transfer based on topological physics
- Precise Motion Control of Wafer Stages via Adaptive Neural Network and Fractional-Order Super-Twisting Algorithm
- Parallel-Connected Inverter Control Method in Material Micromachining