Lazy evaluator for Bloq execution

[vtomole]

Qualtran represents quantum programs hierarchically as Bloqs. However, executing a large program through Cirq generally requires decomposing it and constructing a complete circuit before simulation.

For very large or repetitive algorithms, the fully expanded circuit may be impractical to store even though the Qualtran representation is compact. It would be useful to evaluate a Bloq incrementally, decomposing only the current subgraph and forwarding executable leaf operations to a simulator or hardware.

Qualtran already evaluates individual Bloq decompositions lazily. But I don't think this extends to execution: recursively decompose and execute only the currently demanded portion of the graph without first constructing a fully flattened circuit.

[mpharrigan]

What does "execution" mean here? Numerical/tensor simulation, classical simulation, or something else?

[vtomole]

Classical simulation to begin with. For context, I’m trying to understand how Qualtran is intended to support eventual execution on hardware.

For example, #1517 proposes converting Bloqs into ZX diagrams and ultimately into some form of ISA. I’m unclear how that workflow would run, say, a circuit with more than 1,000 logical qubits and 10^10 logical gates on hardware.

[mpharrigan]

The classical simulation protocol will "walk the call graph lazily". It doesn't flatten large circuits.

You can invoke with my_bloq.call_classically(...) or use the slightly-lower-level API in qualtran.simulation.classical_sim