Learn · Explore · Research

Quantum science.
One idea at a time.

Understand the principles, explore the mathematics, and see what changes when you try an idea for yourself.

Courses, interactive experiments, and research explained with care.

Explore the courses Read the latest news ↓

Experiment · Single-Photon Interference

Mach–Zehnder Interferometer

Single photons split into a superposition across two arms, acquire an optical phase shift \(\phi\), and interfere at the second beam splitter. A photon interferes only with itself.

Single-photon Mach–Zehnder interferometer A single-photon source emits wavepackets through a 50:50 beam splitter BS1, splitting into an upper path with an adjustable phase shifter and a lower path with mirrors. Both paths recombine at beam splitter BS2 and are recorded at detectors D0 and D1. Source BS₁ M₂ M₁ φ = 0° BS₂ D₀ D₁ Upper arm · |0⟩ Lower arm · |1⟩
Detector D₀ (P = cos²(φ/2)) 100%
0 clicks
Detector D₁ (P = sin²(φ/2)) 0%
0 clicks

At φ = 0°, destructive interference completely extinguishes detection at D₁: every single photon emerges at D₀.

Each photon enters in superposition, travels both arms simultaneously, and interferes with itself at the second beam splitter. Blocking an arm destroys the superposition, yielding an equal 50 / 50 distribution at both detectors regardless of φ.

Choose a direction

A place to begin.
A way to go deeper.

From the field

Discoveries & perspectives

Visit the news desk ↗

A claimed refutation of the Exponential Time Hypothesis

OpenAI says an AI system has found a subexponential-time algorithm for counting solutions of fixed-width k-SAT formulas, which would refute the Exponential Time Hypothesis. A manuscript and partial Lean proof are under external review; P versus NP is unaffected.

Read the article

Latest

News

All news ↗

Reported developments, with links to the sources and a clear account of what was found.

Q-CTRL raises $290 million for quantum control software

The Sydney company Q-CTRL has closed a US$290 million Series C led by Salesforce Ventures, which values it at more than US$1 billion. It is a large bet on the software that makes noisy quantum hardware and quantum sensors usable, not a new technical result.

IBM plans a 520-qubit Nighthawk for 2027 and a two-chip, 1,040-qubit system

At its developer conference IBM reported results from its 120-qubit Nighthawk r2 and announced Nighthawk r3, a 520-qubit processor designed for 10,000 operations, two of which it plans to link into one 1,040-qubit system in 2027. The new chip and the linked system are plans; the delivered results are speedups over IBM's own earlier hardware.

Google AI agents design quantum programs and prove them correct

Google DeepMind and Google Quantum AI report Gemini-based agents that design quantum circuits and prove them correct in the Lean theorem prover, cutting estimated costs by 290 times for an arcsin routine and 2.5 to 11 times for fusion, photochemistry and drug-metabolism simulations. The figures are resource estimates against partly author-adjusted baselines, and the Lean library is not yet public.

Background

Explainers

All explainers ↗

Take time with the ideas behind the headlines, one explanation at a time.

MIP* = RE, refereed and machine-checked

The 2020 proof that entangled provers can certify the halting problem, refuting Connes' embedding conjecture, now has a final version accepted by the Annals of Mathematics and a Lean 4 formalisation of its main theorem, registered on 9 October and built with heavy use of Claude. Its consequences for operator algebras remain non-constructive.

AI-built proof claims to settle the Aaronson–Ambainis conjecture

Physical Superintelligence and David Turturean have released a paper and Lean 4 formalisation claiming to prove the 2008 Aaronson–Ambainis conjecture, which limits quantum speedups on unstructured problems. The background, the key lemma, and what the Lean 4 verification covers.

How can measurements drive a quantum computer?

In measurement-based quantum computing, an entangled resource state is prepared first and the computation is carried out by measuring single qubits, with later measurements adapted to earlier outcomes. The model, a worked two-qubit example, and what it demands of real hardware.

Register / AI-assisted research

Claims and open problems

Resolution claims attributed to named groups, and problems that remain open, in physics, mathematics, computer science, and quantum information, with scope notes and sources.

12 problems tracked · Sources reviewed

Papers

Recently selected papers

All selected papers ↗

Technical summaries, methods, and limitations for researchers and advanced readers.

  1. Loss of the quadratic speedup in Grover search under local and coherent noise

    Grover search needs about \pi\sqrt N/4 iterations, and every one gives noise a chance to act. Earlier results are spread over different noise models, error placements and figures of merit, and are mostly thresholds rather than explicit success probabilities. How fast does each kind of noise erase the quadratic speedup, and why are some errors worse than others?

  2. When can local unitaries generate a complete orthogonal basis?

    All four Bell states are related by local unitaries, which is what dense coding uses. Is this special, or can every pure state of a given system be completed to an orthonormal basis using local unitaries alone? A system where this holds for all states is called universally locally encodable (ULE). It was known for 2\times2, 4\times4 and 8\times8, conjectured for two qutrits, and known to fail for four or more qubits.

  3. Explicit Absolutely Maximally Entangled states in dimensions twice a prime power

    An absolutely maximally entangled state AME(N,d) has every reduction to half the parties maximally mixed. Tensoring AME(N,m) with AME(N,n) gives AME(N,mn), but a local dimension 2q would need an AME state of qubits, and those exist only for N=2,3,5,6. Stabilizer constructions face the same obstruction. Which local dimensions admit AME states of nine and more parties, and can the missing factor of two be supplied another way?