Oracles and Bell links meet film stress (09/23/26)

TL;DR Bayesian adaptive thresholding—updating the qubit-state posterior after each analog readout and tightening the feedback threshold as ground-state confidence grows—resets a dispersively readout transmon to $99.44\pm 0.04\%$ fidelity with deterministic duration under real-time FPGA feedback. Versus repeat-until-success (RUS)—re-measure and conditional $\pi$ until a fixed threshold accepts ground—the protocol keeps five-qubit simultaneous-reset fidelity at $98.78\pm 0.19\%$ without stochastic wall-clock stretch.

• Main result: Real-time FPGA feedback on a dispersively readout transmon reaches ground-state initialization fidelity $99.44\pm 0.04\%$ after a small number of adaptive reset rounds.
• Protocol: Bayesian adaptive thresholding folds the full analog measurement history into the state estimate and raises the acceptance bar as confidence that the qubit is in $|g⟩$ increases, rather than discarding soft information at a fixed discriminator.
• Contrast with RUS: Fixed-threshold or repeat-until-success (RUS) loops either waste soft readout data or make reset duration stochastic; adaptive thresholding keeps wall-clock reset length deterministic while still using the soft record.
• Multi-qubit scale: On five qubits reset in parallel, the same adaptive scheme sustains simultaneous-reset fidelity $98.78\pm 0.19\%$, showing the method is usable as a shared cQED initialization primitive.
• Hardware implication: Because the loop runs on FPGA feedback already common in circuit QED, the gain is software-side—better use of existing continuous readout records—without waiting for $T_1$ thermalization or adding new reset hardware.

TL;DR Sputtered $\mathrm{Nb}$ resonators with nearly identical transport critical temperature $T_c\approx 9\mathrm{K}$ show stress-dependent microwave loss and frequency shift that a single-gap Mattis–Bardeen fit (standard BCS microwave conductivity) misattributes to an effective $T_c\approx 2.4\mathrm{K}$. The spectra instead require a low-gap superconducting component that grows with compressive film stress—so DC screening can miss the quasiparticle channel that sets cQED loss.

• Main result: Across a biaxial-stress series of sputtered $\mathrm{Nb}$, transport $T_c$ ($9.0\pm 0.2\mathrm{K}$) and residual resistance ratio stay nearly fixed, yet highly compressive films show stronger temperature-dependent resonator loss and larger frequency shifts.
• Failed single-gap fit: Fitting the microwave response with one Mattis–Bardeen gap returns an effective critical temperature $2.4\pm 0.2\mathrm{K}$, about four times below the transport $T_c$.
• Two-component picture: The data match an additional low-gap superconducting component whose microwave weight increases with compressive stress.
• Hardware implication: Stress-tuned low-energy excitations can dominate kinetic inductance and quasiparticle loss in $\mathrm{Nb}$ cQED films even when DC $T_c$ and RRR look “good.”

TL;DR There is a classical oracle relative to which $\mathsf{BQP}$ (bounded-error quantum polynomial time) is not contained in $\mathsf{IP}$ (classical interactive proofs with a single prover), so efficient quantum computation need not admit a relativizing classical interactive verification protocol. The witness is Forrelation—deciding whether a Boolean function $f$ correlates with the Fourier spectrum of another function $g$—which a quantum algorithm solves with few queries while no polynomial-communication classical interactive protocol with a polynomial-query verifier can.

• Main result: Relative to a classical oracle $\mathcal{O}$, ${\mathsf{BQP}}^{\mathcal{O}}⊈{\mathsf{IP}}^{\mathcal{O}}$, resolving the long-open question of whether containment of efficient quantum computation in classical interactive proofs relativizes.
• Hard problem: Forrelation distinguishes a quantum-easy distribution ${\mathcal{D}}_n$ (signs of Forrelated Gaussians) from uniform ${\mathcal{U}}_n$; a one-query $\mathsf{BQP}$ algorithm succeeds with constant advantage, while every $\mathsf{IP}$ verifier has acceptance gap at most $2^{-\Omega (n)}$.
• Proof idea: Interactive acceptance is modeled by Avg-Max circuits (trees of average and maximum gates with decision-tree leaves); these are approximated by smooth convex functions with small ${\ell }_1$ gradient and Hessian, which the Aaronson–Ambainis Forrelation distribution fools.
• Corollary: Combined with Aaronson et al.'s relativizing $\mathsf{BQP}\subseteq \mathsf{MIP}$ proof, the same oracle separates $\mathsf{IP}$ from $\mathsf{MIP}$ (multi-prover interactive proofs) and from $\mathsf{QIP}$ (quantum interactive proofs), answering a question from Fortnow's thesis.
• Implication: Any prover-efficient classical interactive protocol for $\mathsf{BQP}$ must use non-relativizing ingredients (e.g., cryptography or a complete problem that fails under all oracle reductions), partially explaining stalled progress on unconditionally sound doubly efficient classical verification of quantum computation.

TL;DR An event-ready asymmetric link—one trapped ${}^{40}{\mathrm{Ca}}^{+}$ ion entangled with a photon sent over $25\mathrm{km}$ of telecom fiber—violates the Clauser–Horne–Shimony–Holt (CHSH) Bell inequality at $|S|=2.75_{-0.15}^{+0.16}$, where $S$ is the CHSH correlator that certifies nonlocality from two-party measurement statistics. That surplus above the device-independent quantum key distribution (DIQKD) asymptotic threshold $|S|=2.362$ yields a conservative $\sim 69$ secret bits from $10908$ heralded Bell events, showing DIQKD—key security without trusting the measurement boxes—can run on a heterogeneous, repeater-compatible atom–photon network.

• Architecture: A single trapped ${}^{40}{\mathrm{Ca}}^{+}$ ion anchors one DIQKD party while a remote station measures the partner photon after $25\mathrm{km}$ of spooled telecom fiber, in an event-ready (heralded) asymmetric atom–photon layout.
• Link fidelity: Double quantum frequency conversion plus active polarization stabilization keep the atom–photon quantum correlation intact across the full fiber path.
• Main result: The measured CHSH parameter reaches $|S|=2.75_{-0.15}^{+0.16}$, exceeding the model threshold $|S|=2.362$ needed for a positive asymptotic secret-key fraction.
• Key yield: Under that same asymptotic DIQKD model, the run estimates about $69$ secret-key bits from $10908$ detected Bell states.
• Network implication: The setup sketches a path to DIQKD on heterogeneous, repeater-compatible quantum networks that mix matter memories with telecom photons.