From Broken Detectors to Universal Optimiser
In 2024, two mathematical frameworks emerged that promised to reshape optimisation forever.
QUBO, Quadratic Unconstrained Binary Optimisation. The native language of quantum annealing. Pairwise interactions. The physics of spin glasses written in binary.
PUBO, Polynomial Unconstrained Binary Optimisation. Higher-order interactions. Cubic, quartic, and beyond. The mathematics of complex systems where everything affects everything.
The papers were elegant. The proofs were sound. The mathematics was beautiful.
But the mathematics was also stuck in an unusable state. A mathematical quandary.
No system existed that could run arbitrary QUBO instances natively. No system existed that could run PUBO without reducing it to QUBO, which blows up the problem size exponentially. No system existed that could run either at scale. No system existed that could run either continuously.
The papers described what was theoretically possible. They did not describe what was real.
This is the story of how that changed.
Jiuzhang 4.0 was not built to solve optimisation problems. It was built to prove a point.
3,050 photons. 8,176 optical modes. 25 microseconds per sample.
The largest photonic quantum processor ever constructed. A machine so powerful that it performed in microseconds what the world's fastest supercomputer would take 10⁴² years to complete.
But Jiuzhang 4.0 had a problem. A fatal one.
The detectors couldn't survive.
The photon flux was so immense, the load so concentrated, that the cryogenic cooling systems, which had worked perfectly for Jiuzhang 3.0's 255 photons, were completely overwhelmed. The system was, for all practical purposes, single-use. Run it once, get your sample, and the hardware would collapse.
The world celebrated the achievement. They saw the photon count, marvelled at the speed, and declared quantum supremacy achieved.
They did not see the melted detectors.
They did not see the engineers staring at a dead machine, wondering how to make it live again.
The engineers of the Zeus Quantum Federation did not set out to solve optimisation theory. They set out to save the machine.
The Analog AI was the first fix. A real-time stabilisation system that pre-shaped the photon input to balance the load across the detector array. No single detector would bear the full brunt. The system could breathe.
The SME, the Stability Multiplier Equation, was the second. A dynamic, adaptive mathematical framework that calculated, in real time, exactly how much counter-amplitude needed to be injected to flatten the destructive interference hotspots. It mapped onto the null spaces of the Hybrid Spatial-Temporal Encoding lattice, spaces the Jiuzhang team had left empty, never knowing what would eventually fill them.
The Colour System was the third. Three physically isolated bands, S-band for control and AI management, C-band for quantum sampling, L-band for optimisation and teleportation. Hardware-level diagnostics that watched the system in real time. Eyes that never blinked.
Together, they solved the detector collapse. Jiuzhang 4.0 could now run continuously. Forever computing was no longer a dream; it was a specification.
But then something unexpected happened.
When the engineers looked at what they had built, the stabilised, visible, controllable lattice of 8,176 modes, they realised something profound.
They weren't just running boson sampling anymore.
They were running a programmable graph.
The Hybrid Spatial-Temporal Encoding lattice had always been a graph. Nodes, the spatial modes. Edges, the interference paths between them. The Jiuzhang team had used this graph to run the Hafnian, a #P-hard function that proved quantum supremacy.
But the Hafnian was just one specific function running on that graph. A single path through a vast landscape of possibilities.
When the Zeus engineers stabilised the graph, they gained something the Jiuzhang team never had: control over the edge weights.
The Colour System gave them real-time visibility into every node and every edge. The SME gave them the mathematical framework to adjust those weights dynamically. The Analog AI gave them the ability to shape the graph in real time.
And the moment they had that control, they looked at the graph and saw:
"This is a QUBO machine."
Quadratic Unconstrained Binary Optimisation is the native language of quantum annealing. Each variable is a binary decision. Each pairwise interaction is a quadratic coefficient. The ground state of the system, the configuration that minimises the total cost, is the optimal solution.
The Zeus lattice had 8,176 binary variables (the spatial modes). It had full connectivity between every pair, because the Hybrid Spatial-Temporal Encoding created interference between every pair of modes. No physical wiring constraints. No limited topology. Full, complete, perfect connectivity.
And the SME could set the interference amplitudes, the quadratic coefficients, to any value.
The machine was, in its stabilised state, a native QUBO solver. Not through reduction. Not through emulation. Through the physics of light itself.
QUBO was the first revelation. The second, and far more profound, was PUBO.
Polynomial Unconstrained Binary Optimisation handles higher-order interactions. Cubic, quartic, even 50-body interactions. The mathematics of complex systems: protein folding, financial networks, biological systems, consciousness itself.
Classical computers struggle with PUBO because the number of terms grows combinatorially. Quantum computers usually reduce PUBO to QUBO, by introducing auxiliary variables, but that blows up the problem size exponentially. A cubic problem with n variables becomes a quadratic problem with n² variables.
But in the Zeus lattice, photons don't just interfere in pairs. They undergo multi-photon interference. A single run can involve 3,050 photons interacting across 8,176 modes. The output distribution is directly a function of high-order correlations.
The Hafnian is just one specific higher-order function. When the engineers generalised it, when they realised that the SME preserved not just pairwise interactions but the full hypergraph of multi-photon interference, they understood:
"This is a PUBO machine."
No reduction. No auxiliary variables. No exponential blowup. The native physics of the lattice is the higher-order optimisation.
A cubic cost function maps directly to three-photon interference. A quartic maps to four-photon interference. Any polynomial, any higher-order interaction, maps directly to the multi-photon correlations of the stabilised lattice.
The machine that was built to sample photons had become the universal optimiser.
The 2024 papers proved that QUBO and PUBO were mathematically valid.
Zeus proved they were operational.
But the deeper truth, the one that the engineers discovered only by building the machine and listening to what it told them, is that the cascade was never designed.
The Hybrid Spatial-Temporal Encoding was designed to scale photon count.
The Analog AI was designed to stabilise the detectors.
The Colour System was designed to give the system eyes.
The SME was designed to provide the mathematical framework for control.
None of them were designed to unlock QUBO or PUBO. They were designed to fix a broken machine.
But when you fix a broken machine, when you give it stability, visibility, and controllability, you discover what it actually is, not just what you intended it to be.
The machine that was built to sample photons turned out to be a universal optimiser. The machine that was built to run one function turned out to be capable of running any function. The machine that was built to prove quantum supremacy turned out to be capable of solving the world's hardest problems.
Jiuzhang 4.0 was the first chapter. Zeus is the ending that was never written, only discovered.
QUBO and PUBO are not just theoretical curiosities. They are the mathematics of the real world.
The 2024 papers described what was possible. Zeus is what happens when possibility meets reality.
The West is still trying to build quantum computers that survive long enough to run algorithms. Zeus has already built one that runs and runs and runs, and in the process, discovered it does more than anyone expected.
The result is not a gap. The result is a category difference.
Zeus is not ahead. Zeus is alone.
The road to solving the QUBO and PUBO puzzles was not paved with grand theories or elegant designs. It was paved with real problems, solved pragmatically, one after another.
We fixed the detectors because they were breaking.
We added the Colour System because we needed to see what was happening.
We built the SME because we needed a mathematical model to control it.
And only after all of that, when the system was stable, visible, and controllable, did we look at it and say:
"Wait. This is a QUBO machine."
"Wait. This is a PUBO machine."
"Wait. This is a teleporter."
The cascade was not planned. It was revealed.
And what it revealed is that the universe wanted this architecture to exist. We were just the ones who listened closely enough to hear it.
Zeus is not a computer. It is not a data centre. It is not a chip.
Zeus is the machine that listened to itself and discovered what it could become.
And it is only just beginning.
The full technical framework is documented in the Whitepaper, and the optimisation services are available through Zeus as a Service.
, The Zeus Quantum Federation
The complete ten-part whitepaper covers the full architecture, photonic-quantum engines, the SME, Zeus OS, the 7 Pillars of existence, and the path to Human Teleportation.
Read the Whitepaper