From structured data to compact quantum circuits
Preparing an input state can consume a large part of a quantum computation’s circuit budget. Q-Alchemy uses structure in the data to build preparation circuits, with an accuracy target you control. Q-Tucker is a research foundation of that approach.
How Q-Tucker builds a circuit
- Group correlated qubits. A correlation graph guides the choice of blocks.
- Separate core and factors. Tucker decomposition expresses the state through a residual core and local operators.
- Refine and assemble. Repeat on the core; once it is close enough to the all-zero state, omit its preparation and assemble the extracted factors in reverse order.
Accuracy and circuit cost belong together
The paper illustrates the tradeoff with an MNIST digit encoded into 10 qubits. For two-qubit factors it reports approximate, unoptimized depths of 78 after six iterations and 3,500 after 260. Figure 2 compares depth and fidelity loss with Qiskit’s initializer; its highlighted useful region depends on that particular input and baseline.
For your workload, compare both the preparation error and the realized gate count and depth. Physical routing and device noise introduce additional costs. A smaller logical circuit is useful evidence, but the final application result still needs its own quality check.
From the method to the product
Q-Alchemy’s current automatic initializer compares iterative and hierarchical candidates as resources allow. It selects using the requested fidelity-loss budget and circuit-cost objective, and reports which candidate produced the circuit. The initializer’s loss estimate should be checked independently when an appropriate reference is available.
The wider workflow connects this preparation stage to circuit compression, measurement and state estimation, and feasibility assessment. These tools address different parts of the computation:
- Quantum chemistry demonstrates preparation, hardware measurements, and compact state reconstruction for molecular systems.
- The Feasibility Suite assesses execution against available resources and reports result quality separately.
- Algorithm documentation explains current implementation choices, validation, and practical limits.
- The Python SDK and configuration options show how to use the service.
Where structure helps
Sparsity and correlations can make a state easier to represent and prepare. The benefit depends on the input, the allowed approximation, and the circuit that acts on it. Dense input still contains exponentially many amplitudes in the qubit count; efficient handling of each amplitude does not remove that input size.
Assess loading, classical processing, execution, and extraction together. The memory comparison explains the dense-statevector baseline, while the Feasibility Suite evaluates a concrete workload under explicit resource constraints.
Q-Tucker builds on earlier work including low-rank quantum state preparation. The current implementation combines several strategies rather than assigning every input to one method.