Quantum Computing
Fold-transversal cultivation cuts the spacetime cost of magic states
Kaavya Sahay, Shruti Puri and colleagues introduce 'fold-transversal surface code cultivation,' a new strategy for preparing the magic states that unlock universal fault-tolerant quantum computation. Published in PRX Quantum, the scheme achieves low spacetime overhead on architectures with nonlocal connectivity — targeting what is widely regarded as the single most resource-hungry step in error-corrected quantum computing, and doing so on exactly the kind of long-range-coupled hardware (neutral atoms, trapped ions, photonics) now scaling up.
Quantum Computing
Provable quantum edge in learning many-body dynamics
A new theory paper devises a physically motivated supervised-learning task — predicting observables of quantum many-body systems evolved under an unknown Hamiltonian — and proves quantum learners solve it efficiently while classical algorithms cannot, unless BQP collapses into P/poly. The quantum procedure combines Hamiltonian learning from short-time samples with Hamiltonian simulation and classical shadows at inference, and the hardness proof embeds a BQP-complete computation into a Feynman-Kitaev-style clock construction. It is one of the more rigorous entries in the search for natural machine-learning tasks with genuine quantum advantage.
Quantum information theory gets machine-checked foundations
Lean-Quantum is a new Lean 4 library providing reusable, machine-checkable infrastructure for finite-dimensional quantum information — states, channels, Choi/Kraus/Stinespring representations, and a large stack of noncommutative trace inequalities — built to be compatible with Mathlib. As its centerpiece, the authors formalize the data processing inequality for the sandwiched Renyi relative entropy, obtaining strong subadditivity as a corollary and delivering the final missing component for a complete formalization of the generalized quantum Stein's lemma. It is a notable step toward AI-assisted, formally verified research in quantum theory.
A lighter-weight alternative to quantum singular value transformation
A new algorithmic framework based on probabilistic mixtures of unitary channels applies arbitrary polynomials of Hermitian operators to quantum states — the same capability as quantum singular value transformation — with a tunable trade-off between sample and query complexity and considerably lower circuit complexity than QSVT with linear-combination-of-unitaries block encodings, which the authors argue lets it scale from NISQ to fault-tolerant machines.
Room-temperature photonic chip benchmarked on machine learning
RP000 is a quantum photonic processor built on standard CMOS-compatible manufacturing that operates at room temperature, encoding quantum systems in the degrees of freedom of single photons. Benchmarked across three quantum-classical machine-learning architectures, the authors report higher accuracy than classical networks of comparable size and superior noise tolerance versus a superconducting quantum processor — self-reported claims worth independent scrutiny, but a concrete design-and-benchmark study.
Hybrid quantum floating point sharpens repeated arithmetic
A hybrid representation of quantum floating-point numbers stores values in a quantum register while a classical register tracks range and approximation tolerances, enabling overflow-free addition and multiplication. Ad hoc examples show up to roughly 90% less approximation error than prior techniques after repeated additions — relevant to the low-depth, high-accuracy arithmetic circuits many practical quantum algorithms need.
Machine learning reads Wigner functions from sparse data
A general machine-learning framework reconstructs Wigner functions of continuous-variable quantum systems directly from sparse phase-space measurements, with provably efficient regression whose measurement cost scales only logarithmically with Hilbert-space dimension for states like cat and binomial codes. On experimental circuit-QED data, the model tracks GKP code states through multiple rounds of quantum error correction and pinpoints the dominant error process with far fewer measurements than conventional estimation.
A smarter compiler squeezes more life from GKP qubits
By redefining the stabilizer group of Gottesman-Kitaev-Preskill codes to include every operation acting trivially on the code space, this work in Quantum opens up many physical implementations of each logical Clifford operation, and provides an algorithm that picks the one least affected by loss during computation. Logical randomized benchmarking shows the compiler increases the lifetime of square-GKP qubits running Clifford circuits compared with a random-walk compiler.
White House convenes the quantum industry
The White House held a summit on American quantum innovation with nearly every major quantum company in attendance, per a first-person account from The Quantum Insider. The piece is long on atmosphere and short on announced deliverables, but the convening itself — following recent federal quantum initiatives and executive orders — signals sustained top-level policy attention on the sector.
Quantum Comms
A concrete candidate for unconditionally uncloneable encryption
Uncloneable encryption — encrypting a classical message into a quantum ciphertext that two adversaries cannot both decrypt — has resisted an unconditional construction since Broadbent and Lord's 2020 random-oracle scheme. A new paper in Quantum proposes a candidate for the simplest case, the uncloneable bit, proving the adversary's advantage shrinks quadratically as 1/2 + 1/(2*sqrt(K)) in the number of keys K for K up to 7, verifying up to K = 17 numerically via the NPA hierarchy, and establishing an asymptotic upper bound of 5/8 with a best-known numerical bound of ~0.598.
Oak Ridge prototypes standardized quantum-network monitoring
As quantum networks approach deployment, a new framework proposes standardized performance metrics — entanglement fidelity, QBER, entanglement rate, timing jitter, and exogenous factors like temperature and vibration — for real-time observability and control. The team demonstrates a non-invasive prototype environmental monitoring system integrated with Oak Ridge National Laboratory's quantum network, groundwork for quantum software-defined networking.
Cheaper multipartite entanglement over Bell-pair networks
A new npj Quantum Information paper presents a protocol for distributing multipartite entanglement across quantum networks built from Bell pairs that is efficient in both physical resources and the classical computation needed to orchestrate it — a practical concern as entanglement-based network testbeds scale beyond point-to-point links.
Stacking entanglement into nuclear qudit memories
Wolf-Ruediger Hannes and Guido Burkard propose a universal scheme in which entanglement is repeatedly transferred from electron qubits into long-lived, high-spin nuclear memory qudits in defect centers, accumulating higher-dimensional entanglement over successive rounds. Published in PRX Quantum, the approach points toward quantum communication protocols that exploit qudits rather than qubits.
Quantum Sensing
When multiparameter quantum sensing loses its quadratic edge
A theory paper establishes a universal geometric obstruction explaining when estimating multiple parameters simultaneously forfeits the quadratic time-scaling advantage of quantum metrology: linear dependence among the commuting components of Hamiltonian derivatives inevitably produces a slow parameter direction with bounded Fisher information. The obstruction is diagnosable via a readily computable Gram matrix, demonstrated on collective spin magnetometry, and circumventable by relegating slow directions to nuisance parameters or applying adaptive control.
A shoebox satellite to sniff out nukes in orbit
Nature reports on a proposed shoebox-sized detector satellite designed to pick up the telltale radiation signature of a hidden nuclear device in space — a compact orbital sensing approach to verifying whether warheads are being stationed in orbit, an arms-control concern with no current monitoring capability.
Post-Quantum Crypto
Lattice crypto meets quantum annealing — no threat yet
A new framework encodes small Module-LWE instances — the hardness core of standardized lattice cryptography — as QUBO problems for quantum annealers, jointly recovering secrets and error vectors from ground states, with a stability analysis linking noise robustness to the QUBO energy gap. The authors quantify embedding overhead at growing dimensions and are explicit that current annealing hardware remains far from cryptographically relevant parameters.
SEALSQ and GlobalFoundries pair up on post-quantum silicon
SEALSQ and GlobalFoundries announced a strategic memorandum of understanding to co-develop secure semiconductor platforms, post-quantum-cryptography-ready silicon, and semiconductor-based quantum computing technologies. It is a press-release MoU with no disclosed terms or timelines, but GlobalFoundries' involvement makes it a notable datapoint in PQC moving into mainstream chip manufacturing.
AI & ML
LLM agents stumble when workflows go multilingual
PolyWorkBench benchmarks LLM agents on 67 multilingual long-horizon workplace tasks spanning commerce, knowledge work, legal analysis, localization, and manufacturing, grading them with a hybrid of structural checks, executable verification, and LLM-based semantic assessment. State-of-the-art agents degrade significantly compared with monolingual settings, with errors compounding across reasoning and execution steps.
Putting AI coding agents under statistical experimental design
Rather than judging coding agents by single benchmark runs, this framework treats them as stochastic model-discovery operators and applies classical experimental design — controlled factors, regression models, multiple response variables — to characterize their autonomous data-modeling behavior. Applied to Codex and Claude Code, it quantifies how reasoning effort shifts output quality, dollar cost, wall-clock time, and process complexity, and whether that shift aligns with a performance-cost utility direction.
Microsoft's Flint gives AI agents a chart language
Microsoft Research released Flint, an open-source visualization language positioned between terse chart specs that produce bland output and full imperative plotting code: AI agents author compact, human-editable specifications that compile to expressive charts. A vendor release, but a concrete, inspectable open-source artifact aimed at the growing agent-tooling stack.
Robotics
LingBot-VLA 2.0 trains on 60,000 hours across 20 robot bodies
LingBot-VLA 2.0 pushes vision-language-action pretraining to around 60,000 hours of data — 50,000 hours of robot trajectories spanning 20 distinct robot configurations plus 10,000 hours of egocentric human video — while expanding the action space beyond dual arms to heads, waists, mobile bases, and dexterous hands. Combined with a predictive-dynamics proxy task for temporal reasoning, the system demonstrates cross-embodiment, long-horizon mobile manipulation on two robotic platforms and is evaluated in a generalist setting on the GM-100 benchmark.
Robots levitate data points with sound
AcoustoBots turns mobile robots into physical data displays: TurtleBot3 platforms carry upward-facing 8x8 ultrasonic phased arrays that acoustically levitate a particle whose height (1-10 cm) encodes a local urban value such as population density or noise. A multi-agent RL policy handles collision-aware navigation while a high-rate acoustic controller keeps the trap stable in motion, achieving 90% and 80% task success in single- and dual-robot trials on a scaled UK map.