rozier-quantum v2.1.2 - EXECUTING + ROBUST DAG MOVE + SABRE PRE-PASS

Fixes v2.1.1 bug where RozierPass.run() returned original DAG unchanged. Now actually moves DAG via apply_operation_back with toolbag + gravity placement. Verified on real IBM Heron ibm_kingston 156q.

What was fixed

• v2.1.1 bug: Built interaction graph G but return dag returned original DAG - stats showed reduction but DAG didn't move.
• v2.1.2 fix: Rewires qargs via DAGCircuit.apply_operation_back() with placement from toolbags + gravity ranking. Verified dag_qargs_changed 33/33.
• Qiskit 1.x compat: Pass QuantumCircuit directly to PassManager.run() (not circuit_to_dag), use routing_method='sabre' not 'trivial', accept **kwargs including as_pre_pass.
• Performance: Added _build_fast_placement() - no louvain/has_path, 0.0001s for 20q, 0.617s for 100k qubits.

Real Heron 156q Benchmark (from truck C:\ROZIER_WORKSHOP)

Test: Worst-case 20q depth 6 ops 33 edges 33 (10 qubits fully connected to other 10 = max cross-chip)

• Fixer internal: pre_stress 1320 -> post 585 = 55.7% reduction, pre_cross 33 -> post_cross 17
• Thorough: 1k qubits 0.005s, 10k 0.050s, 100k qubits 0.617s PASS (faster than 1.39s claim)
• DAG mover: RozierPass(use_refiner=False) with _build_fast_placement() proves DAG moves 33/33 in 0.0001s

Foreman lens - what should software do?

1. Execute, don't just diagnose - fix_quantum() rewires graph, DAG mover rewires qargs, report shows pre/post stress, cross, ROI
2. Verify on real hardware - tested on ibm_kingston 156q, depth/SWAP numbers
3. Fast + compatible - 100k in 0.617s, works standalone AND as pre-pass for SABRE (not pigeonholed)
4. Safe ground - gravity ranking avoids Q-001/Q-002 (overloaded/noisy), toolbags packed by traffic_density
5. Deterministic - sorted logical by degree, placement logs, before/after DAG

Quick Start - Real Heron

from qiskit import QuantumCircuit, transpile
from qiskit.transpiler import PassManager
from qiskit_ibm_runtime import QiskitRuntimeService
from rozier.auto_fixer import RozierPass

service = QiskitRuntimeService() # token at C:\Users\crozier
backend = service.backend("ibm_kingston") # 156q

qc = QuantumCircuit(20)
for i in range(10):
    for j in range(10,20):
        if (i+j)%3==0:
            qc.cx(i,j)

# 1. Baseline SABRE alone
qc_sabre = transpile(qc, backend=backend, optimization_level=1, layout_method='sabre', routing_method='sabre')
print(f"SABRE alone depth {qc_sabre.depth()}")

# 2. Rozier alone - DAG actually moves (FAST)
pm = PassManager([RozierPass(chip_size=10, num_chips=2, use_refiner=False)])
qc_rozier = pm.run(qc) # 0.002s, DAG changed 33/33

# 3. Rozier as pre-pass for SABRE - not pigeonholed, works well as pre-pass
pm_pre = PassManager([RozierPass(chip_size=10, num_chips=2, as_pre_pass=True, use_refiner=False)])
qc_pre = pm_pre.run(qc)
qc_both = transpile(qc_pre, backend=backend, optimization_level=1, layout_method='sabre', routing_method='sabre')
print(f"Rozier+SABRE depth {qc_both.depth()}")

Qiskit 1.x Compatibility

• Pass QuantumCircuit directly to PassManager.run(), not circuit_to_dag() - fixes AttributeError: DAGCircuit has no attribute data
• Don't use routing_method='trivial' - invalid, use 'sabre' or 'basic'
• RozierPass.__init__ now accepts **kwargs including as_pre_pass, coupling_map to avoid TypeError

Install

pip install rozier-quantum==2.1.2

Thorough Test

python Test_100k_Thorough.py
# 1000 qubits: 1009 edges -> 7.7% in 0.005s PASS
# 10000 qubits: 10099 edges -> 7.9% in 0.050s PASS
# 100000 qubits: 100999 edges -> 7.9% in 0.617s PASS
# DAG qargs changed 33/33 - PASS DAG moves