When a nanotube vibrates while electrons hop on and off a tiny island of charge (a quantum dot) inside it, the tube's resonance frequency drops sharply every time the dot's charge changes. The depth of that dip is how experimentalists measure the "electromechanical coupling," which is how strongly one electron pushes on the tube. The reported number depends on which physical model is fitted to the dip. Particle physicists face the same problem when they extract decay properties from fitted amplitude models, and they report a separate "model" error bar that is often the largest one. Nanomechanics does not. This note proposes two things. The first is to measure the coupling from the area under the frequency dip rather than its shape: in the slow-electron limit that area is fixed by a counting argument, since exactly one electron is added, regardless of how the dip is broadened. The second is to quantify, with simulated data generated from the most complete available model, how much model error remains when that limit is relaxed.
The motivation is concrete. Ultrastrong coupling between electron tunneling and mechanical motion reports $g_m/2\pi = 0.80 \pm 0.04$ GHz, which corresponds to a coupling-to-frequency ratio of $2.72 \pm 0.14$. It reaches this value by fitting a rate-equation model and a semiclassical electrostatic model. The error bar is a fit error only. Meanwhile, the BESIII amplitude analysis in Amplitude Analysis of the Decays η′→π+π−π0 and η′→π0π0π0 finds a model error of $1.84\times10^{-4}$ against a statistical error of $0.60\times10^{-4}$ on the same quantity. If anything like that ratio holds in nanomechanics, then the claimed ultrastrong ratios, including the re-analyzed literature values of 1.7 and 1.25, carry uncertainties nobody has stated.
Core Claim
Treat the dot as a two-state charge system whose occupation $\langle n\rangle$ depends on the lead electrochemical potential $\mu$, measured relative to the dot level. Let $F$ be the force one electron exerts on the tube, $m$ the effective mass, and $\omega(\mu)$ the measured resonance frequency. In the adiabatic limit, where electrons tunnel much faster than the tube oscillates, the spring constant softens by $F^2\,\partial\langle n\rangle/\partial\mu$. Integrating across one charge transition then gives
$$\int \left[\omega_m^2-\omega^2(\mu)\right]d\mu \;=\; \frac{F^2}{m}\,\Delta n \;=\; 2\hbar\,\omega_m\, g_m^2 ,$$
where $\Delta n = 1$ and we use the convention $g_m = F x_{zp}/\hbar$ with $x_{zp}=\sqrt{\hbar/2m\omega_m}$. The right-hand side contains no lineshape information. Thermal broadening, tunnel broadening, and disorder change the shape of the dip but not its area in $\omega^2$.
This is the nanomechanical analogue of the form-factor-independent ratio in Radiative decays of X(3872) in DD̄* molecule scenario. There, the absolute widths swing from 0.01 to 5.94 keV with an unknown cutoff, but a ratio of two widths does not. We borrow the idea of choosing an observable built to cancel the parameter nobody can compute; we are not borrowing the physics.
The sum rule does not remove all model dependence, and the proposal is built around measuring what remains. Three residual sources remain. The first is the lever arm, which converts gate voltage to $\mu$ and is calibrated separately. The second is non-adiabatic tunneling: when the tunnel rate $\Gamma$ is comparable to $\omega_m$, the in-phase spring response is reduced by roughly $\Gamma^2/(\Gamma^2+\omega_m^2)$ for sequential tunneling. The third is ultrastrong self-consistency: at $g_m/\omega_m > 1$, the tube's equilibrium position shifts with charge, which re-maps the detuning axis and can produce bistability. The claim is that these residuals are smaller than, and better controlled than, the lineshape dependence of a direct dip fit. We also claim they can be mapped as explicit functions of $g_m/\omega_m$, $\Gamma/\omega_m$, and $k_BT/\hbar\Gamma$.
Research Program
The work has three parts.
A simulation generator. We build a toy-data generator from a quantum master equation treatment of a charge-coupled oscillator, following Quantum noise in the electromechanical shuttle: Quantum master equation treatment. That paper shows that discrete tunneling renormalizes the oscillation frequency, including the non-adiabatic and coherent effects that the rate-equation and semiclassical models drop. The generator produces frequency-versus-gate curves with realistic readout noise across a grid of regimes. The grid is anchored on the device parameters of Vigneau et al. ($\omega_m/2\pi = 294.5$ MHz, ratios up to about 2.7) and on the readout figures of Resonant Optomechanics with a Vibrating Carbon Nanotube and a Radio-Frequency Cavity (quality factor 700 to 1000, about 6 pm/√Hz).
Injection and recovery across model pairs. This is how amplitude analyses assess model error, and it differs from simply reporting how much several fits disagree. Data are generated under the richest model with a known injected $g_m$. They are then analyzed with each simpler model and with the sum rule. For every regime we record bias, the width of the reported interval, and coverage, meaning the fraction of toys in which the reported interval contains the true value.
A reportable error budget. The output separates statistical, calibration (lever-arm and readout gain), and model components for $g_m$. It comes with a regime map showing where each extraction method can be trusted. We then apply it to published softening curves. Because the file lists no public data repository, this requires digitizing published figures or obtaining raw sweeps from the authors.
Second-system test (if time permits). We would test a graphene version: an analogous integral over the Landau-level sawtooth in Modulation of mechanical resonance by chemical potential oscillation in graphene. That fit uses a single broadening parameter and neglects the electrostatic tensioning term. The open question is whether a sum rule can separate the quantum-capacitance term from that tensioning term.
Relation to adjacent work
Machine Learning and Kalman Filtering for Nanomechanical Mass Spectrometry also trains and tests on synthetic resonator data. However, it uses one fixed physical model and studies detection of frequency jumps. Here the generating model is deliberately different from the fitting model, because that mismatch is the object of study.
Propagation of Myocardial Fibre Architecture Uncertainty on Electromechanical Model Parameter Estimation propagates structural uncertainty into fitted cardiac parameters. That is the closest methodological cousin. It works with continuum tissue models, however, not with competing microscopic theories of a quantum device.
Baselines
The naive reference is a single-model fit reporting only its covariance error. This is the practice in the source paper, which reports $g_m/2\pi = 0.80 \pm 0.04$ GHz in Ultrastrong coupling between electron tunneling and mechanical motion. We run it on identical toys and expect it to under-cover wherever its lineshape is wrong. The proposal beats it if our intervals reach nominal 68% coverage in regimes where the naive interval does not.
The strongest existing method is the amplitude-analysis practice of fitting several alternative models and taking their spread as the model error. Examples include the Fit A/B/C S-wave variations in Search for CP violation in the Dalitz-plot analysis of D±→K+K−π± and Measurement of interfering K+K− and K−K+ amplitudes in the decay D0→K+K−π0, and the separate model term in BESIII. We implement it with the rate-equation, semiclassical-electrostatic, and master-equation fits on the same toys, the same noise realizations, and the same optimizer budget. No reported result was found on Lacuna for the coverage or bias of any $g_m$ extraction, so the bar is set inside this study. The proposal beats this baseline if it achieves coverage at least as good with a narrower interval, or matching coverage in regimes where the spread approach over-covers or under-covers.
Two ablations isolate the key ingredient. The first applies the sum rule without the non-adiabatic correction, which shows where the counting argument alone fails. The second fixes the lever arm at its nominal value, which separates calibration error from model error.
Why not just…?
Why not just fit several models and quote the spread, as particle physicists do? This is the real threat. If the spread across three fits already covers the truth at about 68% everywhere, the sum rule's value reduces to a narrower interval, and we will say so. The spread has no coverage guarantee, though. It can be small precisely when all the candidate models share the same wrong assumption. The rate-equation and semiclassical models both assume instantaneous charge equilibrium, so they can agree with each other and both be biased once $\Gamma \sim \omega_m$. The injection-recovery coverage map settles this directly.
Why not just fit the full master-equation model and skip the simpler ones? It is costly to fit to every gate sweep. More importantly, it still needs a lineshape and a lever arm, so it moves the model dependence elsewhere rather than removing it. We include it as a fitting model on toys generated with perturbed electrostatics. That tests whether the "most complete" model is also robust when its own assumptions are slightly wrong.
Why not just train a neural network or use Bayesian model averaging? A neural network trained on simulations inherits the simulator's assumptions. The Kalman-filtering study also found that purely data-driven detectors failed on closely spaced events where physics-based state tracking succeeded (Machine Learning and Kalman Filtering for Nanomechanical Mass Spectrometry). Bayesian model averaging concentrates its weight on one model as data accumulate, even when none of the models is true, so it should under-cover in exactly the mismatched regimes. We run it as an additional comparison on the same toys.
Why not measure charge and motion directly? In the pinched-off nanotube, the direct current is zero, and readout depends on the cavity coupling model (Resonant Optomechanics with a Vibrating Carbon Nanotube and a Radio-Frequency Cavity). External probes would also disturb the graphene electronic state that the drum senses. Direct measurement is not available in the regimes that matter.
Proposed Experiments
The decisive check is a set of coverage and bias maps over $g_m/\omega_m \in [0.5, 3]$ and $\Gamma/\omega_m \in [0.1, 100]$ at several temperatures, built from about $10^3$ toys per grid point. The proposal succeeds if three things hold together. The lineshape-only error of the sum-rule estimate stays below the fit error of the single-model baseline across the adiabatic region. The corrected sum rule keeps near-nominal coverage into the region $\Gamma \sim \omega_m$, where the multi-fit spread fails. And the location of the breakdown in the ultrastrong regime is identified and stated.
A second check re-analyzes the published softening curves of the device in Ultrastrong coupling between electron tunneling and mechanical motion and of the two re-analyzed literature devices. For each, it asks whether $0.80$ GHz and the ratios 2.72, 1.7, and 1.25 survive once a model term is added, and whether that term is comparable to the stated $\pm 0.04$ GHz.
A falsifying outcome would also be informative. Suppose dip-shape fits and sum-rule areas disagree in real data by more than the model budget predicts. That would point to physics missing from all three models, such as coherent charge-motion superposition, rather than to an error in fitting.
References
- Ultrastrong coupling between electron tunneling and mechanical motion
- Amplitude Analysis of the Decays η^{'}→π^{+}π^{-}π^{0} and η^{'}→π^{0}π^{0}π^{0}.
- Radiative decays of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mi>X</mml:mi><mml:mo stretchy="false">(</mml:mo><mml:mn>3872</mml:mn><mml:mo stretchy="false">)</mml:mo></mml:math> in <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mi>D</mml:mi><mml:msup><mml:mover accent="true"><mml:mi>D</mml:mi><mml:mo stretchy="false">¯</mml:mo></mml:mover><mml:mo>*</mml:mo></mml:msup></mml:math> molecule scenario
- Quantum noise in the electromechanical shuttle: Quantum master equation treatment
- Resonant Optomechanics with a Vibrating Carbon Nanotube and a Radio-Frequency Cavity
- Modulation of mechanical resonance by chemical potential oscillation in graphene
- Machine Learning and Kalman Filtering for Nanomechanical Mass Spectrometry
- Propagation of Myocardial Fibre Architecture Uncertainty on Electromechanical Model Parameter Estimation: A Case Study
- Search for<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mi>C</mml:mi><mml:mi>P</mml:mi></mml:math>violation in the Dalitz-plot analysis of<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:msup><mml:mi>D</mml:mi><mml:mo>±</mml:mo></mml:msup><mml:mo>→</mml:mo><mml:msup><mml:mi>K</mml:mi><mml:mo>+</mml:mo></mml:msup><mml:msup><mml:mi>K</mml:mi><mml:mo>−</mml:mo></mml:msup><mml:msup><mml:mi>π</mml:mi><mml:mo>±</mml:mo></mml:msup></mml:math>
- Measurement of interfering<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:msup><mml:mi>K</mml:mi><mml:mrow><mml:mo>*</mml:mo><mml:mo>+</mml:mo></mml:mrow></mml:msup><mml:msup><mml:mi>K</mml:mi><mml:mo>−</mml:mo></mml:msup></mml:math>and<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:msup><mml:mi>K</mml:mi><mml:mrow><mml:mo>*</mml:mo><mml:mo>−</mml:mo></mml:mrow></mml:msup><mml:msup><mml:mi>K</mml:mi><mml:mo>+</mml:mo></mml:msup></mml:math>amplitudes in the decay<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:msup><mml:mi>D</mml:mi><mml:mn>0</mml:mn></mml:msup><mml:mo>→</mml:mo><mml:msup><mml:mi>K</mml:mi><mml:mo>+</mml:mo></mml:msup><mml:msup><mml:mi>K</mml:mi><mml:mo>−</mml:mo></mml:msup><mml:msup><mml:mi>π</mml:mi><mml:mn>0</mml:mn></mml:msup></mml:math>