QC Study Notes

Parameterized Circuits

3 min read

Parameterized Circuits

qc/programming qc/qiskit

A Parameter is a symbolic placeholder in a circuit — you build the circuit’s structure once, then bind concrete numeric values later, possibly many different sets of values against the exact same compiled circuit.

from qiskit import QuantumCircuit
from qiskit.circuit import Parameter
 
theta = Parameter('θ')
qc = QuantumCircuit(1)
qc.rx(theta, 0)          # circuit built once, angle unbound
 
bound = qc.assign_parameters({theta: 3.14159})   # bind a concrete value

ParameterVector is the same idea for many parameters at once (e.g. one angle per layer of a variational circuit):

from qiskit.circuit import ParameterVector
 
angles = ParameterVector('a', length=4)
qc = QuantumCircuit(4)
for i, angle in enumerate(angles):
    qc.rx(angle, i)

Why this matters

Key insight: transpilation is expensive, and it only depends on circuit structure (gates and connectivity), not on the numeric values inside rotation gates. A parameterized circuit is transpiled once; every subsequent run just rebinds numbers into the already-transpiled template. Variational algorithms (VQE, QAOA) exploit this directly — they sweep over hundreds of parameter values on the same fixed circuit structure, and would be far slower if every iteration re-transpiled from scratch.

You don’t have to bind values yourself before running — primitives accept parameter values directly as part of a PUB (see The Primitives Family), letting you submit many parameter sets as one batched job:

from qiskit_aer.primitives import EstimatorV2
from qiskit.quantum_info import SparsePauliOp
 
obs = SparsePauliOp('Z')
param_sets = [[0.0], [1.57], [3.14]]   # three different theta values, one job
EstimatorV2().run([(qc, obs, param_sets)]).result()[0].data.evs

A missed opportunity, in hindsight

1D Ising Chain and the Mirror Trick’s construct_ising_circuit takes rx_angle as a plain Python float baked in at construction time (rx_angle = π/8 throughout the worked examples here), rebuilding the whole circuit for any other angle. A Parameter there would let the same boxed/Samplomatic-compiled template be reused across many angle values without rebuilding or re-boxing it — exactly the same “structure once, values many times” idea Samplomatic already applies to random Pauli dressings.

Official reference: Qiskit circuit library documentation.

Self-Check

  • Why is a parameterized circuit transpiled once instead of once per parameter value?
  • Could you explain to someone why variational algorithms like VQE and QAOA specifically benefit from this?
  • What would change in 1D Ising Chain and the Mirror Trick’s construct_ising_circuit if rx_angle were a Parameter instead of a float?

Graph View

Backlinks