Noise Models (Conceptual Overview)

Overview

To make progress with noisy devices, we need a way to talk about noise that is:

• simple enough to reason about
• accurate enough to predict main effects

That’s what noise models are. A noise model is not “the true microscopic physics.” It is a controlled approximation that captures how imperfections change the output distribution of circuits.

Noise models matter because they enable:

• sanity checks and debugging of experiments
• comparisons between devices
• the design of mitigation techniques

This page introduces three common conceptual models:

• depolarizing noise
• amplitude damping
• phase damping

No Kraus operators and no master equations yet.

Intuition

A good way to build intuition is to ask: what does noise do to a qubit’s Bloch-sphere picture?

• Some noise shrinks the Bloch vector toward the center (loss of purity).
• Some noise pulls the state toward $|0\rangle$ (energy loss).
• Some noise smears out phase (loss of coherence around the equator).

Noise models package these effects into simple “rules of thumb.”

Formal Description

We define each model by what it tends to do to states.

Depolarizing noise

Intuition:

• with some probability, the state is replaced by something close to “completely random.”

Effect in words:

• superpositions lose contrast
• measurement outcomes drift toward uniform randomness

Depolarizing noise is often used as a simple stand-in for “generic errors per gate.” It is not always physically accurate, but it is useful for building first intuition.

Amplitude damping

Amplitude damping models energy relaxation. Intuition:

• population leaks from $|1\rangle$ to $|0\rangle$

Effect in words:

• states are biased toward the ground state
• excited-state probability decays over time

This connects to the $T_1$ concept.

Phase damping

Phase damping models dephasing. Intuition:

• relative phase becomes uncertain while energy populations are mostly unchanged

Effect in words:

• superpositions lose interference
• computational-basis probabilities may look stable, but phase-sensitive experiments degrade

This connects to the $T_2$ concept.

Why models are useful (and limited)

Useful:

• they let you predict qualitative outcomes
• they support approximate calculations
• they guide mitigation ideas

Limited:

• real noise can be nonstationary, correlated, and hardware-dependent
• models may miss coherent (systematic) errors

So you treat models as tools, not truth.

Worked Example

Imagine you want to run a simple interference experiment:

1. prepare $|0\rangle$
2. apply H to get $|+\rangle$
3. apply H again to return to $|0\rangle$
4. measure

Ideally, you always measure 0.

Now consider conceptual noise effects:

• Depolarizing noise might sometimes randomize the state, producing occasional 1 outcomes.
• Phase damping would reduce coherence, making interference less perfect and increasing the chance of measuring 1.
• Amplitude damping would bias outcomes toward 0, which can mask some errors (you might see many 0s even if coherence is lost).

This shows why different noise types have different signatures.

Turtle Tip

Common Pitfalls

Quick Check

What’s Next

Next we can go deeper in two directions:

• how to measure these effects experimentally (calibration and characterization)
• how to reduce their impact (error mitigation and, later, error correction)

Before that, we’ll extend the noise story to multi-qubit settings and discuss why noise becomes harder when qubits interact.