Darviq-Quantum on real quantum hardware
Run the same circuits on a real quantum computer through the Qiskit adapter, and compare its noisy results with the simulator's exact ones.
PlannedExperiments in quantum computing, robotics, AI and orbital mechanics. They're research, not products, and they keep the engineering sharp.
A quantum computer simulated from first principles, with four classic quantum algorithms checked against known answers.
A simulated humanoid in MuJoCo, built from human body measurements, balanced by a hand-designed controller and by one learned through trial and error.
A simulator for an autonomous interplanetary transfer: two-body Keplerian orbits and Hohmann transfers from J2000.0 planetary data, with a mission autopilot whose flights end in one of three outcomes: arrival, mission failure or crew lost.
A qubit can be 0 and 1 at the same time, and n qubits hold 2n numbers at once. Darviq-Quantum keeps all of those numbers explicitly and applies quantum gates to them as plain matrix arithmetic in numpy, with no quantum SDK. On top of that simulator sit four textbook algorithms, each tested against the answer a classical computer gives.
The simulator core is under 300 lines of Python, the whole toolkit about 860. Every gate is a small matrix you can read, so nothing is hidden behind a framework. A laptop handles about 20 to 24 qubits; each extra qubit doubles the memory, which is exactly why real quantum hardware matters.
Deutsch-Jozsa tells whether a hidden function is constant or balanced in a single query, where a classical computer may need 2n-1 + 1. Bernstein-Vazirani recovers a hidden bit string in one query instead of one per bit.
Finds one marked item among 256 in 12 queries with 99.99% certainty. A classical search would expect to check about 128. The speedup is quadratic, and the simulation also shows its catch: run past the optimum and the answer fades again.
The building block of Shor's factoring algorithm, the one that threatens today's encryption. The simulator's output matches numpy's classical Fourier transform to within 3 × 10-16.
| Queries | Quantum | Classical |
|---|---|---|
| 0 | 0.39% | 0.4% |
| 1 | 3.48% | 0.8% |
| 2 | 9.46% | 1.2% |
| 3 | 17.97% | 1.6% |
| 4 | 28.47% | 2.0% |
| 5 | 40.32% | 2.3% |
| 6 | 52.76% | 2.7% |
| 7 | 65.03% | 3.1% |
| 8 | 76.37% | 3.5% |
| 9 | 86.07% | 3.9% |
| 10 | 93.52% | 4.3% |
| 11 | 98.26% | 4.7% |
| 12 | 99.99% | 5.1% |
| 13 | 98.62% | 5.5% |
| 14 | 94.22% | 5.9% |
| 15 | 87.06% | 6.2% |
| 16 | 77.60% | 6.6% |
| 17 | 66.42% | 7.0% |
| 18 | 54.21% | 7.4% |
| 19 | 41.75% | 7.8% |
| 20 | 29.80% | 8.2% |
| 21 | 19.10% | 8.6% |
| 22 | 10.33% | 9.0% |
| 23 | 4.03% | 9.4% |
| 24 | 0.59% | 9.8% |
75 automated tests compare every algorithm with its known answer across different qubit counts and random inputs: recovered secrets bit for bit, Fourier outputs against numpy, Grover's success probability against theory. An optional Qiskit adapter expresses the same circuits for real quantum hardware; running them there is the next experiment.
Not by making laptops faster: quantum computers are specialists. They are expected to win at a few kinds of problem where nature itself is quantum, or where the maths has a structure they can exploit, and to stay worse than ordinary computers at almost everything else.
Simulating molecules exactly is beyond classical computers for all but tiny cases, and it is the most widely expected first real win: drug candidates, better battery chemistry, catalysts for cleaner fertiliser production. Chemistry is quantum, so a quantum computer is the natural tool.
A large, error-corrected quantum computer running Shor's algorithm would break RSA and elliptic-curve encryption, which protect most of the internet. Data stolen today can be decrypted later, so the move has already begun: NIST published post-quantum encryption standards in 2024, and major browsers and CDNs already use them for much of their traffic.
Logistics, finance and scheduling are often named, but the honest view is mixed: speedups like Grover's are quadratic, not exponential, and today's hardware overheads can eat them. Expect real but narrower wins, mostly in hybrid quantum-classical systems.
Few organisations will own a quantum computer. They will rent time on one, the way they rent servers: Amazon Braket, Azure Quantum and IBM Quantum already offer real machines as cloud services, alongside the classical systems that prepare and check their work.
Today's machines have hundreds to over a thousand physical qubits, but they are noisy, and useful work needs error-corrected "logical" qubits built from many physical ones. Since 2024, experiments have shown error correction getting better as it scales up, the key milestone, and the major roadmaps aim for the first large error-corrected machines around the end of this decade. India's National Quantum Mission, approved in 2023, is funding domestic machines on a similar horizon.
What to do about it now: for most organisations, not buying quantum computers but taking stock of where encryption is used and planning the switch to post-quantum algorithms. That is a cloud and infrastructure job, and it is one of the services we offer.
Run the same circuits on a real quantum computer through the Qiskit adapter, and compare its noisy results with the simulator's exact ones.
PlannedFactor small numbers like 15 and 21 with the quantum Fourier transform already in the toolkit.
PlannedTrained by evolution strategies in the same MuJoCo simulation. Much better than the classical controller at pushes from behind, worse from the front; next, make it mirror-symmetric.
Built · mixed resultsRecover from larger pushes by taking a step, not just shifting weight at the ankles.
PlannedAdd the pull of a third body and solar radiation pressure for longer missions.
PlannedWe're always glad to compare notes, or to help plan a move to post-quantum encryption.