Quantum Process Tomography (Intro)

Overview

Quantum Process Tomography (QPT) answers a different question than benchmarking:

• “What operation did the device actually implement?”

Instead of summarizing with a single average number, tomography aims to reconstruct a detailed description of a gate or channel.

Why it matters:

• It can reveal specific error structure (wrong axis, crosstalk-like behavior, unwanted rotations).
• It can guide debugging and calibration.

But it comes with a major cost: tomography scales poorly with system size.

Intuition

Tomography as “system identification”

In classical engineering, you might identify a black box by:

• feeding in known inputs
• measuring outputs
• fitting a model

Quantum process tomography is the same idea:

• prepare a set of known input states
• apply the unknown process
• measure outputs in enough ways to infer what happened

Why it’s hard

A quantum process can affect amplitudes and phases. To reconstruct it, you need many experiments with many settings. As you add qubits, the number of degrees of freedom grows extremely fast.

So tomography is powerful for:

• 1-qubit or 2-qubit gates
• careful lab debugging

It is usually not a “daily metric” for large multi-qubit devices.

Formal Description

We keep this conceptual: what QPT tries to reconstruct and why it scales badly.

What QPT reconstructs (in words)

QPT aims to determine a full description of a quantum process:

• how it maps input states to output states

This includes both:

• ideal unitary behavior (the intended gate)
• non-ideal behavior (noise, leakage, unwanted coupling)

Why it scales poorly

To reconstruct a process, you need to probe it with enough input states and enough measurement settings.

As the number of qubits $n$ increases:

• the size of the state space grows like $2^n$
• the number of parameters needed to describe a general process grows even faster

So the number of experiments required becomes enormous.

When tomography is still useful

Despite scaling issues, QPT is useful when:

• you have a small subsystem and want a detailed diagnosis
• you are developing or calibrating a specific gate
• you want to compare two implementations and see how they differ, not just “how much”

Worked Example

Imagine you are implementing a single-qubit X-rotation. RB might tell you:

• “average error is low”

Tomography might reveal:

• the rotation axis is slightly tilted (you are rotating partly around Y)
• there is a consistent over-rotation by a small angle

Those are very different failure modes. They can have different fixes (calibration vs hardware stability), and tomography can help distinguish them.

Turtle Tip

Common Pitfalls

Quick Check

What’s Next

After characterizing gates, you also need to characterize measurement. Next we cover measurement calibration and readout correction: how to build and use confusion matrices to interpret observed bitstrings.