How do mixed-dimensional circuits—some qu3its, some qu8its together—actually compile down to real hardware gates?

# Compiling Mixed-Dimensional Quantum Circuits to Hardware Gates

The core challenge is that most quantum hardware speaks only one language—typically qubits (2-level systems)—but you want to run a program that mixes qudits of different dimensions (a 3-level and an 8-level system, say). How do you bridge that gap?

State Preparation as the Foundation

The first step is getting your mixed-dimensional state *into* the hardware. Rather than hand-craft circuits, the emerging approach uses decision diagrams—a compact data structure that represents the quantum state you want to prepare—to automatically generate the circuit [AG-2024.06-1578]. Think of a decision diagram as a flowchart that describes your state; the compilation tool walks that flowchart and emits gates. Importantly, this method works even when your qudits have mismatched dimensions, because the decision diagram naturally encodes which subsystem is which level [AG-2024.06-1578].

Standardized Language and Gate Compilation

Once you have a state prepared, you need a systematic way to express mixed-dimensional circuits and translate them to hardware. MQT Qudits, an open-source framework, does exactly this [AG-2024.10-655]. It provides:

• A standardized circuit specification language for mixed-dimensional systems (so qu3its and qu8its can be described unambiguously together)
• Compilation to hardware gate sets—the crucial step where abstract circuit operations are converted into the specific native gates your device actually supports
• Efficient circuit simulation to predict what will happen

The framework handles the messy practical question: if your hardware natively supports certain multi-level operations (e.g., a 3-level gate on one qudit and a 2-level CNOT elsewhere), the compiler decomposes your high-level circuit into those primitives while respecting device connectivity.

General Circuit Optimization Strategies

Beyond qudits specifically, the broader compilation pipeline follows principles now well-established [AG-2024.06-2419]:

1. Qubit/qudit mapping: Assign logical qudits in your algorithm to physical qudits on the chip

2. Routing: Insert SWAP gates (or their qudit equivalents) to satisfy hardware connectivity constraints

3. Gate synthesis: Decompose each logical operation into native gates with minimal depth and error

Recent work shows that reinforcement learning can find near-optimal decompositions orders of magnitude faster than classical search methods—achieving strong results on Clifford and permutation circuits up to many qubits [AG-2024.05-2287]. This is relevant because once you've specified a mixed-dimensional gate, you still need to synthesize it efficiently.

The Hardware Design Question

Finally, there's a subtler point: which gates should your hardware support in the first place? For qubit systems, researchers use gate expressivity as a design criterion—not just average fidelity, but whether the gate set can efficiently implement algorithmic structures [AG-2024.02-2255]. For mixed-dimensional systems, this question is even richer, because you can choose gate sets of different dimensionalities. Tools like YAQQ explore this design space systematically, evaluating gate sets against target workloads to find which native gates minimize circuit depth and error [AG-2024.06-2270].

Summary

Mixed-dimensional circuit compilation is a three-layer problem: (1) prepare the state using decision diagrams, (2) compile abstract operations to a standardized hardware gate set using MQT Qudits or similar tools, and (3) optimize the resulting circuit using synthesis and routing algorithms, informed by which native gates your hardware actually supports. The field is young—much of this work appeared in 2024–2025—but the frameworks exist and are becoming practical.