Optimizing Circuit Depth

On near-term hardware, qubits decohere. Every layer of gates the circuit must execute is more time during which the fragile quantum state decays, so circuit depth — the number of sequential gate layers on the critical path — is a first-order determinant of how reliable a result will be. Reducing depth is therefore one of the most valuable optimizations a transpiler performs.

Depth versus gate count

Gate count and depth are different metrics. Count is how many gates there are in total; depth is the length of the longest chain of gates that must run one after another. Gates acting on disjoint qubits run in the same layer, so a wide circuit can have low depth despite a high count:

\text{depth} = \text{length of the critical path in the gate dependency graph}.

Two gates can share a layer exactly when they touch no common qubit. The optimizer models the circuit as a directed acyclic graph of gate dependencies and asks for the minimum number of layers — a graph-scheduling problem.

Two levers for depth

Parallelization (scheduling). Pack independent gates into the same time step. A column of single-qubit gates on different qubits is depth 1, not depth $n$ . Greedy "as-soon-as-possible" scheduling slides each gate to the earliest layer where its qubits are free.

Cancellation and merging. Removing redundant gates shortens the critical path directly. If a chain contains $X X = I$ , deleting the pair removes two layers of depth; merging a run of $z$ -rotations into one removes the rest. Depth reduction and gate-count reduction reinforce each other here.

Why equivalence still rules

As with every optimization, reducing depth must not change the unitary. Deleting an $XX$ pair is safe because it equals the identity; reordering two gates is safe only if they commute. The grader compares full unitaries precisely so that a shallower circuit is accepted only when it computes the same operation.

Try it

The given circuit applies $X$ twice in a row before the CNOT — a redundant pair that inflates depth from 2 to 4. Produce the depth-optimized equivalent. The grader checks the full two-qubit unitary, so any circuit equal to $H_0$ followed by $\mathrm{CX}(0,1)$ passes.

Run your code to see the quantum state.