Amplify noise to remove it; invert it with quasi-probabilities.
Two ideas dominate quantum error mitigation. Zero-noise extrapolation treats a noisy expectation value as a smooth function of the noise strength and extrapolates to zero; probabilistic error cancellation writes the inverse of the noise as a signed mixture of operations you can run. Both are unbiased only under assumptions, and both pay in variance.
What you’ll learn
Derive the Richardson extrapolation coefficients, the O(λn+1) bias and the variance amplification, including ∑j∣γj∣=2n+1−1 for equally spaced scales.
Justify noise scaling by time rescaling, unitary folding and probabilistic error amplification, and state their assumptions.
Construct quasi-probability decompositions of Pauli channels, prove that PEC is unbiased, and derive γ=e2∑kλk for sparse Pauli–Lindblad noise.
Compare ZNE and PEC by bias, variance and modelling requirements.
Noise as a dial
Let E(λ)=Tr[Oρ(λ)], where λ scales the noise and λ=1 is what the hardware does natively. Temme, Bravyi and Gambetta observed that time-independent Markovian noise can be amplified by stretching time.
DerivationAmplifying noise by stretching time
dtdρ=−i[H(t),ρ]+λL(ρ),0≤t≤T
A drive H(t) and a Lindblad noise generator L of strength λ, assumed independent of the drive.
Hc(t)=c1H(t/c),0≤t≤cT
Run the same pulse c times slower and c times weaker.
τ=t/c:dτdρ=−i[H(τ),ρ]+cλL(ρ)
In rescaled time the coherent part is unchanged while the noise is multiplied by c: the stretched run measures E(cλ).
E(λ)=E(0)+k=1∑nakλk+O(λn+1)
For weak noise E is smooth in λ, and ZNE exploits this expansion.
Hardware-agnostic alternatives scale the noise digitally:
Unitary folding: U↦U(U†U)m gives λ=2m+1 (Giurgica-Tiron et al., 2020). It assumes U† suffers the same noise as U. For depolarizing noise D, which commutes with every unitary channel, the noisy folded gate is exactly D∘U∘D∘U†∘D∘U=D3∘U.
Probabilistic error amplification (PEA): learn a Pauli noise model, then insert extra Pauli errors at random so that every rate is multiplied by exactly λ≥1.
Richardson extrapolation
Measure at n+1 scales λ0<λ1<⋯<λn and combine ER=∑jγjE(λj), choosing the γj so that the first n powers of λ cancel.
DerivationRichardson coefficients and bias
j∑γj=1,j∑γjλjk=0(k=1,…,n)
Keep E(0) and kill a1,…,an: a Vandermonde system with a unique solution.
γj=m=j∏λm−λjλm
These are the Lagrange basis polynomials ℓj(x)=∏m=jλj−λmx−λm evaluated at x=0. Interpolation is exact for polynomials of degree at most n, which is exactly the system above.
ER−E(0)=j∑γjO(λjn+1)=O(∑j∣γj∣λjn+1)
Only the remainder survives: the bias is of order n+1 in the noise.
λj=j+1:γj=(−1)j(j+1n+1),j∑∣γj∣=2n+1−1
For equally spaced scales 1,2,…,n+1 the coefficients are (2,−1), then (3,−3,1), then (4,−6,4,−1).
The variance is the price. With Nj shots at scale λj, N=∑jNj in total, and single-shot variance σ2,
Var(ER)=σ2j∑Njγj2≥Nσ2(j∑∣γj∣)2
Equality holds for the optimal allocation Nj∝∣γj∣ (Cauchy–Schwarz). For equally spaced scales the overhead (2n+1−1)2 grows exponentially with the order n, so in practice n≤2 or 3.
Predict before you look
Four equally spaced noise scales λ=1,2,3,4 (third-order Richardson). What is ∑j∣γj∣?
Show the answer
15
γ=(4,−6,4,−1), so ∑j∣γj∣=15=24−1. Even with optimal shot allocation the variance grows by 152=225 relative to a single run at native noise, before counting the larger noise at high λ.
If the noise is close to global depolarizing, E(λ)=E(0)e−bλ fits better than a polynomial, and then E(0)=E(1)2/E(2) from two points. Choosing the model is the art of ZNE: a wrong model gives a confidently biased answer.
Probabilistic error cancellation
DerivationInverting depolarizing noise
Dp(ρ)=(1−p)ρ+3p(XρX+YρY+ZρZ),Dp(P)=fP,f=1−34p
Each non-identity Pauli P commutes with itself and anticommutes with the other two, so it shrinks by f.
Solve the two equations. The negative weight on the Pauli corrections is the price of "un-depolarizing".
Real devices are characterized more efficiently by twirling the noise of each layer into a Pauli channel and fitting a sparse Pauli–Lindblad model (van den Berg et al., 2023): N=exp(∑kλk(Pk−id)) with Pk(ρ)=PkρPk and only low-weight, local Pk. Its inverse is easy because all the factors commute.
DerivationCost of inverting Pauli–Lindblad noise
eλ(P−id)=e−λ(coshλid+sinhλP)
P2=id, so the exponential series splits into even and odd parts.
Flip the sign of λ to invert one factor, then add the absolute values of its weights.
γ=k∏e2λk=exp(2k∑λk),N∝γ2=exp(4k∑λk)
Each λk is approximately the probability of the error Pk, so the shot overhead is e4×(expected number of errors), matching e4pL for depolarizing noise.
Check your understanding
A learned Pauli–Lindblad model gives ∑kλk=1.5, summed over all layers of a circuit. What is the PEC shot overhead?
e3≈20
e6≈400
e1.5≈4.5
e12≈1.6×105
Show the answer
B.e6≈400
γ=e2⋅1.5=e3, and the shot count scales as γ2=e6≈403.
ZNE versus PEC
Property
ZNE
PEC
Bias
O(λn+1) if the model holds; uncontrolled otherwise
Zero for an exact noise model; model error enters linearly
Variance overhead
(∑j∣γj∣)2, modest at low order
γtot2=e4∑kλk, exponential in the total noise
Needs
A reliable noise dial
An accurate, stable noise model of every layer
Typical use
Large circuits where a small residual bias is acceptable
Smaller circuits where unbiasedness matters
The two combine naturally. PEA uses the learned Pauli–Lindblad model of PEC not to cancel the noise, which is too expensive at scale, but to amplify it exactly, feeding a ZNE fit. This was the strategy of the 127-qubit experiments of Kim et al. (2023).
Key takeaways
Noise can be amplified by time rescaling (for drive-independent Markovian noise), unitary folding or probabilistic error amplification; ZNE then extrapolates E(λ) to λ=0.
Richardson coefficients γj=∏m=jλm/(λm−λj) cancel the first n orders; the variance overhead (∑j∣γj∣)2 grows exponentially with the order.
PEC is exactly unbiased for an exact noise model; its shot overhead is γtot2, with γ=21(3/f−1)≈1+2p per depolarizing gate and γ=e2∑kλk for Pauli–Lindblad noise.
ZNE trades a model-dependent bias for a low overhead; PEC trades unbiasedness for an exponential overhead and heavy characterization.
Check yourself
In the interactive lesson these are scored with your confidence, and what you miss comes back for review.
Question 1
For noise scales λ∈{1,3}, the first-order Richardson coefficients (γ0,γ1) are…
(3/2,−1/2)
(2,−1)
(1/2,1/2)
(3,−2)
Show the answer
A.(3/2,−1/2)
γ0=λ1/(λ1−λ0)=3/2 and γ1=λ0/(λ0−λ1)=−1/2. Check: 3/2−1/2=1 and 23⋅1−21⋅3=0.
Question 2
Why is high-order Richardson extrapolation rarely used?
Because its bias grows with the order
Because it requires an exact noise model
Because ∑j∣γj∣ grows exponentially with the order, amplifying shot noise, and large-λ data are poorly described by the Taylor expansion
Because it only works for a single qubit
Show the answer
C. Because ∑j∣γj∣ grows exponentially with the order, amplifying shot noise, and large-λ data are poorly described by the Taylor expansion
For equally spaced scales ∑j∣γj∣=2n+1−1, so the variance overhead grows like 4n+1; the highest scales also push E(λ) into a regime where low-order polynomials fit badly.
Question 3
PEC is unbiased provided that…
the circuit is shallow
the quasi-probability decomposition inverts the actual noise of every gate exactly
all qi are positive
the noise is amplified to at least λ=3
Show the answer
B. the quasi-probability decomposition inverts the actual noise of every gate exactly
Unbiasedness follows from ∑iqiBi=N−1 exactly. Any mismatch between the model and the real noise leaves a residual bias.
Question 4
For single-qubit depolarizing noise with Pauli error probability p≪1 per gate, PEC over L gates needs about how many times more shots?
1+2pL
e2pL
(1+p)L
e4pL
Show the answer
D.e4pL
γg≈1+2p, so γtot≈e2pL and the shot overhead is γtot2≈e4pL.
Further reading
K. Temme, S. Bravyi & J. M. Gambetta, "Error mitigation for short-depth quantum circuits", Phys. Rev. Lett. 119, 180509 (2017).
Y. Li & S. C. Benjamin, "Efficient variational quantum simulator incorporating active error minimization", Phys. Rev. X 7, 021050 (2017).
S. Endo, S. C. Benjamin & Y. Li, "Practical quantum error mitigation for near-future applications", Phys. Rev. X 8, 031027 (2018).
E. van den Berg, Z. K. Minev, A. Kandala & K. Temme, "Probabilistic error cancellation with sparse Pauli–Lindblad models on noisy quantum processors", Nature Physics (2023).
Z. Cai et al., "Quantum error mitigation", Rev. Mod. Phys. 95, 045005 (2023).