Quantum papers — 2026-09-16

Today's focus is on making quantum engineering reliable because turning quantum phenomena into useful technology requires high levels of coordination and characterization that current methods struggle with. We are developing Quantum-Harbor, a virtual laboratory designed for scientific artificial intelligence agents to interact with and perform actions on quantum systems in a controlled environment. This allows us to verify both what the agent does and what it concludes.

This framework is then used to create QIQCBench, a benchmark consisting of forty-nine expert tasks covering areas like calibration and error correction. This benchmark reveals a significant gap between an agent showing capability and actually operating reliably.

Moving down in importance is the work on shuttling compilers for trapped-ion quantum computers. This addresses the manual effort needed to write routing logic for qubit movement. We fine-tuned five large language models on schedules generated by hand heuristics for linear and branched one-dimensional trap architectures.

We found that in twelve percent of compilations, running the best schedule ten times reduced operations by up to twenty-one percent compared to original heuristic baselines after a simple post-processing step. Furthermore, a single run of one of these fine-tuned LLMs successfully generated a valid shuttling schedule for an unseen four-way branched architecture. However, two other unseen architectures proved impossible for any LLM to solve.

Finally, there is research into cryptographic foundations concerning Hamiltonian phase state assumptions and their relationship to one-way functions. We showed that if the Hamiltonian phase state assumptions are true, then one-way functions must exist. This means we cannot use these states to instantiate genuine Microcryptography.

It does provide a new toolbox for building efficiently verifiable one-way puzzles by exploiting tailored state certification protocols. This work connects to prior findings by showing that if the state certification protocol allows for efficient classical post-processing, we get an efficiently verifiable one-way puzzle, which in turn implies the existence of one-way functions.

Today's papers

Evaluating Verified Autonomy in Quantum Engineering Scientific artificial intelligence agents can be reliably tested using a virtual laboratory and benchmarks to measure their performance. Shuttling Compiler for Trapped-Ion Quantum Computers Based on Fine-Tuned Large Language Models uses language models to automatically generate the movement schedules for qubits in trapped-ion quantum computers. Instantiating Microcrypt Obstacles and opportunities via tailored state certification explores how Hamiltonian phase states relate to constructing classical cryptography. Towards Surrogate Based Dequantization of Quantum Reinforcement Learning studies classical methods that can mimic the performance of quantum reinforcement learning algorithms. O(n) alternative to Quantum Fourier Transform with efficient neural net classical post-processing proposes a shallower circuit design for Shor's algorithm using a neural network for post-processing. [paper]
Evaluating Verified Autonomy in Quantum Engineering Scientific artificial intelligence agents can be reliably tested using a virtual laboratory and benchmarks to measure their performance. Shuttling Compiler for Trapped-Ion Quantum Computers Based on Fine-Tuned Large Language Models uses language models to automatically generate the movement schedules for qubits in trapped-ion quantum computers. Instantiating Microcrypt Obstacles and opportunities via tailored state certification explores how Hamiltonian phase states relate to constructing classical cryptography. Towards Surrogate Based Dequantization of Quantum Reinforcement Learning studies classical methods that can mimic the performance of quantum reinforcement learning algorithms. O(n) alternative to Quantum Fourier Transform with efficient neural net classical post-processing proposes a shallower circuit design for Shor's algorithm using a neural network for post-processing. [paper]
Evaluating Verified Autonomy in Quantum Engineering Scientific artificial intelligence agents can be reliably tested using a virtual laboratory and benchmarks to measure their performance. Shuttling Compiler for Trapped-Ion Quantum Computers Based on Fine-Tuned Large Language Models uses language models to automatically generate the movement schedules for qubits in trapped-ion quantum computers. Instantiating Microcrypt Obstacles and opportunities via tailored state certification explores how Hamiltonian phase states relate to constructing classical cryptography. Towards Surrogate Based Dequantization of Quantum Reinforcement Learning studies classical methods that can mimic the performance of quantum reinforcement learning algorithms. O(n) alternative to Quantum Fourier Transform with efficient neural net classical post-processing proposes a shallower circuit design for Shor's algorithm using a neural network for post-processing. [paper]
Evaluating Verified Autonomy in Quantum Engineering Scientific artificial intelligence agents can be reliably tested using a virtual laboratory and benchmarks to measure their performance. Shuttling Compiler for Trapped-Ion Quantum Computers Based on Fine-Tuned Large Language Models uses language models to automatically generate the movement schedules for qubits in trapped-ion quantum computers. Instantiating Microcrypt Obstacles and opportunities via tailored state certification explores how Hamiltonian phase states relate to constructing classical cryptography. Towards Surrogate Based Dequantization of Quantum Reinforcement Learning studies classical methods that can mimic the performance of quantum reinforcement learning algorithms. O(n) alternative to Quantum Fourier Transform with efficient neural net classical post-processing proposes a shallower circuit design for Shor's algorithm using a neural network for post-processing. [paper]
Evaluating Verified Autonomy in Quantum Engineering Scientific artificial intelligence agents can be reliably tested using a virtual laboratory and benchmarks to measure their performance. Shuttling Compiler for Trapped-Ion Quantum Computers Based on Fine-Tuned Large Language Models uses language models to automatically generate the movement schedules for qubits in trapped-ion quantum computers. Instantiating Microcrypt Obstacles and opportunities via tailored state certification explores how Hamiltonian phase states relate to constructing classical cryptography. Towards Surrogate Based Dequantization of Quantum Reinforcement Learning studies classical methods that can mimic the performance of quantum reinforcement learning algorithms. O(n) alternative to Quantum Fourier Transform with efficient neural net classical post-processing proposes a shallower circuit design for Shor's algorithm using a neural network for post-processing. [paper]

The papers

Important terms

Quantum-Harbor: A virtual laboratory created for scientific artificial intelligence agents to safely interact with and manipulate quantum systems. It's designed to verify both what the agent does and what it concludes.
QIQCBench: A benchmark consisting of forty-nine expert tasks, covering areas like error correction and calibration. It highlights the difference between an agent showing skill and actually working reliably.
Shuttling Compilers: Research focused on automating the manual process of writing routing logic for moving qubits in trapped-ion quantum computers. This uses fine-tuned large language models to generate schedules.
Hamiltonian Phase State Assumptions: Assumptions about the phase state of a Hamiltonian in cryptography. The research shows that if these assumptions are true, then one-way functions must exist.
One-Way Functions: Mathematical functions that are easy to compute but extremely hard to reverse without specific secret information. This work connects their existence to specific cryptographic protocols.