2
[2026-09-27] · src Sedrakyan, Zhang, Karapetyan, Baktay, Gharibyan, Tepanyan; arXiv:2609.28657 (2026-09-23)
Sedrakyan and coauthors ran vanilla (unboosted) random-circuit sampling on a 61-qubit patch of IBM's 120-qubit Nighthawk r2 chip (ibm_phoenix), through the standard cloud stack with no benchmark-specific calibration: median CZ error ~1.9x10-3, median single-qubit (SX) error ~2.5x10-4, median readout error ~8x10-3. At 36 cycles - the depth where tensor-network contraction cost saturates with system size - they measured a cross-entropy fidelity F_XEB = 2.3x10-3, cross-checked against an independent mirror-benchmarking estimator over more than two decades of fidelity decay. Pricing the classical side with a bounded-fidelity rejection-sampling model calibrated against published Sycamore and Zuchongzhi contraction costs (agreement within 0.7 decades), collecting the reported 106-sample ensemble at that fidelity would cost 1.2x1027 machine operations - over a century on the Frontier supercomputer at 20% efficiency - against 19 s of QPU time. The authors then tested "severing" attacks directly: classical spoofing strategies that partition the circuit and sample the severed pieces independently to fake a high cross-entropy score cheaply. The strongest variant they could construct still falls short of the experimental score by a factor of ten, and the margin for balanced cuts exceeds 106.
Every previous random-circuit-sampling advantage claim - Google's Sycamore, USTC's Zuchongzhi - ran on a bespoke device nobody outside the group could rent, and each was followed by improved classical algorithms, including severing-style spoofing, that closed much of the gap. This is the first such claim on hardware anyone with a cloud account can use, at an operating point its own authors show surviving the best spoofing attack they could find by an order of magnitude - a stronger, more checkable claim than "bigger than a supercomputer" on its own. It buys nothing computationally: vanilla RCS has no known application, and the classical cost model still prices out multi-amplitude reuse and matrix-product methods that a smarter classical algorithm could exploit, which is exactly the pattern that eroded the Sycamore claim. Whether this one holds is now a question for whoever wants to try to spoof it, with the attack surface and the margin stated in the open.
[2026-09-27] · src D'Souza, Chen, Parashar, et al. (Ajoy group); arXiv:2609.28912 (2026-09-24)
D'Souza and coauthors grew a 2:1 cocrystal of 6,13-dihydropentacene and pentacene (DHP/Pc, 33.3 mol% pentacene), reaching a volumetric spin-site density of 9.51x1020 cm-3 (3.3x105 ppm) - more than two orders of magnitude above the previous benchmarks for optically addressable spin ensembles, NV-diamond and pentacene-doped p-terphenyl. Packing spins this densely ordinarily destroys the coherence needed to read them out, through dipolar coupling between neighbors and triplet-exciton transport that carries polarization away before measurement. DHP/Pc instead shows microsecond spin coherence together with room-temperature optically detected magnetic resonance and coherent control. Time-resolved measurements attribute the long-lived triplet population mainly to intersystem crossing, with strong non-thermal sublevel polarization, and DFT calculations point to why density does not cost coherence here: the cocrystal's packing geometry weakens electronic coupling between neighboring pentacene molecules, while dihydropentacene's higher triplet energy blocks the trap-to-trap migration that would otherwise scramble polarization.
NV-diamond's spin density has stood as a rough sensitivity ceiling for ensemble quantum sensors for over a decade, because every attempt to pack spins more densely has traded coherence away in the process. This result reframes that as a materials-design problem rather than a limit: choose the coformer for its triplet energy and packing geometry, not only its spin properties, and density and coherence stop pulling against each other. It is a materials demonstration, not yet a sensing one - no magnetometry or NMR-enhancement figure is reported - so the sensitivity gain the higher density promises is implied rather than measured.
2
[2026-09-27] · src Arban, Complete Conservatory Method — 14 Characteristic Studies
Eleven studies in, each one has asked for a single kind of control held across a whole page. No. 12 asks for three different ones in a row on the same page: a driving forte in continuous sixteenths, a "dolce" episode that wants a completely different, more spacious feel, and a forte triplet outburst that briefly leaves the four-flat home key before the sixteenths return to close it out. The job today is changing gears cleanly at each seam rather than smoothing the whole page into one character.
6 minutes: the one-octave scale in the four-flat key, ascending and descending in continuous sixteenths at quarter = 88, tongued lightly, three times through — then, with no pause, the same scale legato and half the tempo, leaning into a singing tone with no tongue between notes. The point is rehearsing the two extremes the study asks to sit side by side, back to back, before it demands the switch inside a real phrase.
The full page at quarter = 88, three consecutive passes. Success criterion: on all three passes, a listener could name which of the study's three episodes is sounding at any moment without seeing the page — the opening and closing sixteenths driving and even, the "dolce" episode audibly slower-feeling and singing even though the notated tempo hasn't changed, and the triplet outburst landing as a genuine three against the surrounding four rather than blurring into more sixteenths. The triplet episode's brief swing into sharps should also read as a real harmonic jolt, not a passage played as if it were still in the home key.
Art of Phrasing, Melody No. 25. Play it once for notes, then once choosing a single phrase to shape the way the study's "dolce" episode wants to be shaped — a genuine change of character from whatever surrounds it, not just a softer dynamic on the same notes.
Stop after the three clean technical passes, or sooner if the "dolce" episode stops sounding different from the sixteenths around it, or the triplets start smoothing back into the surrounding subdivision — either is fatigue erasing the distinction between the three characters, and a fourth pass built on top of it trains the blur instead of the gear change.
[2026-09-27] · src —
hearClifford Brown & Max Roach — I Don't Stand a Ghost of a Chance With You · Tidal · Spotify
Today's musical target and the study's own "dolce" episode both ask for the same thing: a phrase that sings rather than one that is merely played correctly. Clifford Brown's ballad reading of this standard, with Max Roach, is one of the instrument's standard references for exactly that — wide, confident interval leaps folded into a single unhurried line, with nothing about the width of the leaps disturbing the sense that it's one phrase rather than a series of jumps. Listen for how little the leaps interrupt the line's forward motion, which is the same demand today's "dolce" episode makes of a single page rather than a whole chorus.