Scientific computing for the quantum era.
Quanterall is a research and engineering company developing scientific-computing frameworks, distributed AI infrastructure and industrial Digital Twin systems. Our work spans quantum computing, mathematical physics, photonic simulation and distributed systems.
PQCF is our scientific-computing framework. FVLLMP, our distributed AI infrastructure on Erlang/OTP, grew out of TechUPGRADE, our industrial Digital Twin platform, and is being prepared for validation across independently owned nodes. GenHub2 shows how they connect in a real application. Every system is measured against defined tests and published with its methods, results and limitations.
Three kinds of technology,
one company.
Scientific computing · PQCF
- Photonic simulation, symbolic computation, mathematical physics and high-precision numerical methods
- A task and execution architecture for applications: across one owner's machines today; on federation peers, in development
- Open source; every result checked and reproducible
Distributed AI infrastructure · FVLLMP
- An Erlang/OTP platform for independently owned AI nodes: model routing, memory, workload sharing and federation
- Grew out of TechUPGRADE's AI infrastructure; now a pilot of two owners' NVIDIA DGX Spark nodes
- Ten-node validation, then an open-source release: planned
Industrial Digital Twins · TechUPGRADE
- Engineering drawings, live IoT measurements, physical models, CFD and physics-informed neural networks in one platform
- Simulation, scenarios, comparison with the real plant, and AI agents that propose changes for people to approve
- Built by Quanterall for the EU TechUPGRADE project around a thermochemical heat-storage reactor; multi-tenant
GenHub2, the application that connects them → Our research areas → TechUPGRADE's films → TechUPGRADE's technical reports →
The newest results,
each with its details.
What exists, what is simplified,
what is missing.
- Tested
- checked by our own tests against an independent reference
- Benchmarked
- timed against an established tool on the same machine
- Reproduced
- rerun by us from a fresh clone, with a step-by-step page
- Prototype
- working, but not yet in its final place or form
- In development
- implementation underway; not yet complete
- Planned
- intended work that has not started
- Independently reviewed
- evaluated by a reviewer outside the development team: nothing yet
Research, PQCF and GenHub2
- Live circuits on QxSim, our state-vector simulator in Elixir, cross-checked against Qiskit (which code runs what).
- Five codes with full correction cycles; the surface code under gate-level noise on QxStab, our stabilizer simulator, checked against Stim.
- Reproductions of known results, with data: repetition sweeps, a threshold study, benchmarks, photonic schemes (HOM, the KLM gates).
- PQCF, our framework for scientific computing. Version 1.4.0 (8 November), the latest release, contains photonic simulation, the symbolic engine, reconstruction over finite fields and Feynman-integral reduction; its interface has been frozen since 1 November. The task layer and the newer mathematical-physics instruments are tested and reproducible at the commits identified on their respective pages, but are not yet included in a tagged release.
- Research instruments for quantum field theory and QCD developed in PQCF, not yet included in a tagged release: Feynman diagrams and rules, the running of QCD to five loops, and one-loop integrals, compared with FORM, mpmath, CRunDec or LoopTools (18 November).
- GenHub2, virus classification for genetic laboratories, is the first application on PQCF. Its first two phases are complete. Phase B, distributed execution on two Erlang nodes on one DGX Spark: the same results, nearly twice the throughput. The first FVLLMP federation test, task execution on a second DGX Spark: byte-for-byte identical outputs, showing correctness and reproducibility, not yet a performance advantage (17 November). It is live at genhub2.quanterall.com, with public genomes only, and uses TechUPGRADE's simulation engine and 3D visualiser through the platform's external integration API.
- Every published result reproducible from a fresh copy of the code: one list of step-by-step pages, each with the commit it was tested at, and every result set tagged in its repository.
- Shor's modular multiplication is applied as one permutation unitary, not decomposed into gates; N up to 21.
- State-vector codes: one error per run, noiseless gates; the surface code there uses one reused ancilla.
- Research hypothesis tested: the first BEAM decoder design did not achieve the targeted bounded tail latency near compiled speed. The next approach is batching work per process and testing on a dedicated machine (the research question).
- An independent review of the physics, and a paper.
- A release with a DOI. PQCF has a citation file, and every published result set is tagged; the archived, citable release is next.
- Our research question: the first answer is no; next, batches per process and a dedicated machine.
- The repository with a DOI; a first preprint.
- PQCF as a distributed scientific-computing platform: move the federation-peer executor, prototyped through GenHub2, into PQCF, so that scientific tasks can run across independent laboratory nodes. GenHub2's phase C will test it across two laboratories.
- Large-scale symbolic computation, measured against FORM every week; Feynman-integral reduction, now 3 to 7 times faster than FIRE on its own double box; next, larger families, spread across many machines.
FVLLMP: a two-owner pilot, not yet the ten-node validation
- Two NVIDIA DGX Spark nodes, each lending model capacity to the other, in daily use
- Certificates pinned and confirmed by the two owners; credits recorded on both sides and compared
- Containerised tasks for a partner: GenHub2's first 11 tree tasks came back byte for byte identical; through the peer they took 36.9 s, against 15.2 s on the lab's own 10 cores (17 November)
- Charter version 1, signed by both owners; model requests and incoming tasks checked against it on both nodes since 21 November
- Relays over several hops, nodes behind NAT, discovery listings: on simulated federations of two to four nodes
- Owner overrides of a refusal: implemented and tested on simulated nodes; not yet exercised between the two real pilot nodes
- The lab installer and in-place upgrades: on test machines
- Validation across an initial ten nodes
- An open-source release after that testing; the licence is still to be decided
- Charter version 2: under the owners' review, not in force
The progress log → The details of the circuits and codes → How this was built →
One foundation, two platforms,
and the applications on them.
- The foundation: FVLLMP, distributed AI infrastructure on Erlang/OTP: model serving, routing, memory, and a federation of independently owned nodes. Today a two-owner pilot on two NVIDIA DGX Spark; ten-node validation and an open-source release are planned.
- On it, two platforms. TechUPGRADE, industrial Digital Twins, uses FVLLMP's AI infrastructure today: FVLLMP began as that infrastructure. PQCF, scientific computing, runs across one owner's nodes today; its execution on FVLLMP's federation is a prototype in GenHub2 and, natively in PQCF, in development.
- On top, the applications. Our research computations (quantum optics, QCD, Feynman integrals) and GenHub2 run on PQCF; GenHub2 also uses TechUPGRADE's simulation engine and 3D visualiser through the platform's integration API. The digital twins of plants run on TechUPGRADE.
Where FVLLMP came from. FVLLMP began as the AI infrastructure behind TechUPGRADE: the model gateway, semantic routing, memory, development and audit orchestration, and the component-owner agents the Digital Twin platform uses. TechUPGRADE was the first demanding application; the same infrastructure is now being developed into a federation of independently operated nodes.
Quanterall develops FVLLMP; each node in a federation belongs to its owner, who controls it, under agreements between owners. PQCF's architecture → FVLLMP's layers →
What we work on,
and how it is checked.
Feynman diagrams and QCD
- QED and QCD diagrams, equal to FORM's generator in 24 cases
- The three-loop gluon self-energy's 324 diagrams generated, with their weights, in 0.1 s on one core
- QCD's β-function and quark mass anomalous dimension to five loops, as exact coefficients (rationals and ζ-values) equal to the published ones
- Against mpmath to 28 digits, and CRunDec
Feynman-integral reduction
- Integration-by-parts identities, solved modulo primes
- Exact results by finite-field reconstruction
- 3 to 7 times faster than FIRE on the tested problem, its own two-loop double box, on one core: not a general claim
- Next: larger families, across machines
Mathematical physics
- Four quantum-optics problems with closed-form solutions: PQCF agrees to 36 digits; QuTiP to 5–16 digits under the tested default and tightest tolerances
- Holographic QCD: seven computations against their papers
- Series, spectra and ODEs to 25–38 digits
- Symbolic engine measured against FORM every week
Photonic simulation
- Gaussian and Fock-space simulation, boson sampling
- Against The Walrus, Piquasso, Perceval, DeepQuantum, Strawberry Fields
- The fastest exact boson sampler of the four programs that sample, 4 to 20 photons, on one core of our machine; behind on three other kernels
- KLM gates, fusion, GKP and cat states reproduced
Quantum algorithms and error correction
- Ten circuits live on the BEAM, checked against Qiskit
- The surface code under circuit noise, checked against Stim
- A research question, and its first hypothesis falsified
GenHub2
- Virus classification for genetic laboratories
- 593 held-out GenBank genomes from 30 viruses: species right for 96.5% when protein search backs the weak calls; 95.1% by genome similarity alone
- Phase B, two Erlang nodes on one DGX Spark: the same results, nearly twice the throughput
- The first FVLLMP federation test, on a second DGX Spark: identical results, not yet faster
- The first external application on TechUPGRADE: in its own tenant, it uses the simulation engine and the 3D visualiser for viral genomes, mutation positions and 213 protein structures across 29 viruses (from the Protein Data Bank, except 2 AlphaFold predictions, labelled as such); the joint test passed all 14 checks
Four ways in,
depending on why you are here.
For researchers
- The research areas and what each has shown
- Every comparison, including where we lose
- How to reproduce every result
- Our research question
- Collaboration: [email protected]
For developers
- PQCF's architecture
- The repositories: pqcf, genhub2, quantum_computing_simulations
- The user guide
- The Python client
- Building and testing: one Docker image
- FVLLMP: the distributed infrastructure, its layers, and what a node does
- The TechUPGRADE AI-infrastructure film
For laboratories and institutions
- GenHub2: virus classification on the lab's own hardware; live at genhub2.quanterall.com, with public genomes only
- FVLLMP: a node for each lab, linked by agreements between owners; the two-owner pilot, and what is not yet available
- PQCF: scientific computing; execution across one owner's nodes today, on federation peers in development
- Data locality: what stays at the lab (raw reads) and what may travel (derived inputs, named by their hash)
- Security policy and technical documentation
- Research collaboration: [email protected]
For industrial and engineering teams
- TechUPGRADE, our industrial Digital Twin platform: the complete description, or one 16-minute film
- IoT and operational monitoring: a plant's drawing as a live board, alarms, rules and a fleet view
- CFD and physics-informed models, simulation and scenario testing
- AI agents for every component and machine, proposing changes for people to approve
- All the films and the technical reports
- A reference platform, not yet installed at a customer's site: [email protected]
One framework,
from a task to its result.
PQCF brings together a scientific engine and a task and execution architecture. Its scientific engine supports photonic simulation, symbolic computation, finite-field reconstruction, Feynman-integral reduction and mathematical physics. Its task layer represents an application's work as plain data and records execution provenance. Execution across one owner's Erlang nodes is implemented; federation-peer execution in PQCF is in development, building on a working prototype exercised through GenHub2. FVLLMP is the distributed software infrastructure being developed for execution across independent nodes: it originated in the infrastructure work for TechUPGRADE and is being prepared for validation across an initial ten nodes. How PQCF works → Discover FVLLMP →
Three things you can
check for yourself.
after 23 years.
On 23 March 2026, 23 years to the day after incorporation, Quanterall renewed its focus on quantum computing research, returning to the purpose behind its name. The name "Quanter-all" was chosen in 2002 at the Joint Institute for Nuclear Research in Dubna as a statement of intent toward quantum computing; the company was incorporated in Varna on 23 March 2003, and the QC beside its name in the logo stands for quantum computing. Quanterall continues to build on its experience in distributed systems and Erlang while developing research software, scientific-computing frameworks and distributed infrastructure.