Simple Algorithm
Quantum Computing · Rydberg State · 30 min

Tomography & Gate Set Tomography

Self-consistent characterization of noisy operations.

State tomography trusts its measurements. Process tomography trusts its state preparations and measurements. But on a real device every preparation and measurement is built from the same imperfect gates you are trying to characterize. Gate set tomography breaks this circle by fitting everything at once, and pays for it with a fundamental ambiguity: the gauge freedom.

What you’ll learn

  • Represent channels by Pauli transfer matrices, count their parameters, and describe quantum process tomography.
  • Explain why SPAM errors make process tomography self-inconsistent, and define the gate-set model.
  • Derive linear gate set tomography, exhibit the gauge freedom, and distinguish gauge-invariant from gauge-variant quantities.
  • Explain how long germ sequences give 1/L1/L precision, and how model violation tests Markovianity.

Process tomography

A channel E\mathcal E on nn qubits (d=2nd = 2^n) is fixed by its action on a basis of operators. Use the Pauli strings Pi∈{I,X,Y,Z}⊗nP_i\in\{\id,X,Y,Z\}^{\otimes n}, for which Tr⁡(PiPj)=d δij\Tr(P_iP_j) = d\,\delta_{ij}:

Predict before you look

How many real parameters does a general CPTP map on two qubits have?

Show the answer

240 parameters

A Hermiticity-preserving map on d×dd\times d matrices has d4d^4 real parameters (its PTM is a real d2×d2d^2\times d^2 matrix), and trace preservation fixes the first row, removing d2d^2. With d=4d = 4: 256−16=240256-16 = 240. For one qubit it is 12. Complete positivity is an inequality and removes no parameters.

Quantum process tomography (Chuang & Nielsen 1997) prepares an informationally complete set of inputs (for example all tensor products of ∣0⟩,∣1⟩,∣+⟩,∣+i⟩\ket0,\ket1,\ket+,\ket{+i}, which is 4n4^n states), applies E\mathcal E, performs state tomography on each output in 3n3^n Pauli bases, and inverts linearly or by maximum likelihood. That is 12n12^n configurations, each with its own shot noise. State tomography itself is treated in the Measurement topic's Rydberg State lesson on quantum state tomography.

The SPAM problem and the gate-set model

Process tomography assumes that the input states and measurement effects are known exactly. On real hardware they are made with gates: ∣+⟩\ket+ is "∣0⟩\ket0 followed by a π/2\pi/2 pulse", and an XX-basis measurement is "a pulse followed by a ZZ readout". Errors in those pulses masquerade as errors of the process under study, and the reconstructed map can even fail to be completely positive. The fix is to characterize preparations, measurements and gates together.

For a qubit, the trace-preserving gauge group has 16−4=1216-4 = 12 parameters. A gate set with one preparation (3 parameters), one two-outcome measurement (4) and three trace-preserving gates (12 each) has 43 parameters, of which only 43−12=3143-12 = 31 are physically meaningful.

Check your understanding

Which of these quantities can, in principle, be determined by experiment?

  1. The process fidelity of the gate GxG_x with its target
  2. The eigenvalues of the superoperator GxG_x
  3. The Bloch vector of the prepared state ρ\rho
  4. The off-diagonal entries of the Pauli transfer matrix of GxG_x
Show the answer

B. The eigenvalues of the superoperator GxG_x

Eigenvalues are invariant under similarity transformations G↦MGM−1G\mapsto MGM^{-1}, so they are gauge-invariant and observable, as are all circuit probabilities and RB decay rates. The other three change under a gauge transformation, so their reported values depend on a convention, usually fixed by gauge optimization: choosing the MM that brings the estimate closest to the target.

Linear GST: the gauge made concrete

Choose fiducial sequences F1,…,Fd2F_1,\ldots,F_{d^2}: short gate strings such that the effective states Fj∣ρ⟩ ⁣⟩F_j|\rho\rangle\!\rangle and effective effects ⟨ ⁣⟨E∣Fi\langle\!\langle E|F_i are informationally complete. Then measure the probabilities of the circuits "prepare, FjF_j, optional gate GkG_k, FiF_i, measure".

DerivationLinear gate set tomography
  1. A=(⟨ ⁣⟨E∣F1⋮⟨ ⁣⟨E∣Fd2),B=(F1∣ρ⟩ ⁣⟩ ⋯ Fd2∣ρ⟩ ⁣⟩)A = \begin{pmatrix}\langle\!\langle E|F_1\\\vdots\\\langle\!\langle E|F_{d^2}\end{pmatrix},\qquad B = \Big(F_1|\rho\rangle\!\rangle\ \cdots\ F_{d^2}|\rho\rangle\!\rangle\Big)

    Stack the effective measurement covectors as the rows of AA and the effective prepared states as the columns of BB. Both are d2×d2d^2\times d^2 and invertible by informational completeness, but unknown.

  2. g~ij=⟨ ⁣⟨E∣FiFj∣ρ⟩ ⁣⟩  ⇒  g~=AB,(P~k)ij=⟨ ⁣⟨E∣FiGkFj∣ρ⟩ ⁣⟩  ⇒  P~k=AGkB\tilde g_{ij} = \langle\!\langle E|F_iF_j|\rho\rangle\!\rangle\;\Rightarrow\;\tilde g = AB,\qquad(\tilde P_k)_{ij} = \langle\!\langle E|F_iG_kF_j|\rho\rangle\!\rangle\;\Rightarrow\;\tilde P_k = AG_kB

    The measured "Gram matrix" (no gate in the middle) and the data for gate GkG_k.

  3. g~−1P~k=B−1A−1A Gk B=B−1GkB\tilde g^{-1}\tilde P_k = B^{-1}A^{-1}A\,G_k\,B = B^{-1}G_kB

    The data determine every gate up to the same unknown similarity transformation BB: this is the gauge freedom.

  4. g~−1(⟨ ⁣⟨E∣Fi∣ρ⟩ ⁣⟩)i=B−1∣ρ⟩ ⁣⟩,(⟨ ⁣⟨E∣Fj∣ρ⟩ ⁣⟩)j=⟨ ⁣⟨E∣B\tilde g^{-1}\big(\langle\!\langle E|F_i|\rho\rangle\!\rangle\big)_i = B^{-1}|\rho\rangle\!\rangle,\qquad\big(\langle\!\langle E|F_j|\rho\rangle\!\rangle\big)_j = \langle\!\langle E|B

    The state and the effect come out in the same gauge, so {B−1∣ρ⟩ ⁣⟩, ⟨ ⁣⟨E∣B, B−1GkB}\{B^{-1}|\rho\rangle\!\rangle,\ \langle\!\langle E|B,\ B^{-1}G_kB\} is a complete, self-consistent estimate made without any assumption about SPAM.

Long sequences, precision and model violation

Long-sequence GST measures circuits of the form fiducial, germ repeated, fiducial. Each germ gg is a short gate string, repeated until the total length reaches LL (typically L=1,2,4,…L = 1,2,4,\ldots). Germs are chosen so that every error parameter of the gate set is amplified by some germ.

DerivationWhy repetition gives 1/L precision
  1. G=Rx(π2+ϵ),G4L=Rx(2πL+4Lϵ)G = R_x\big(\tfrac\pi2+\epsilon\big),\qquad G^{4L} = R_x\big(2\pi L+4L\epsilon\big)

    A small over-rotation ϵ\epsilon. Four repetitions ideally give the identity (up to sign), so the error accumulates coherently.

  2. P(1)=sin⁡2(π/2+(4L+1)ϵ2)=12+12sin⁡((4L+1)ϵ)≈12+(4L+1)ϵ2P(1) = \sin^2\Big(\frac{\pi/2+(4L+1)\epsilon}{2}\Big) = \frac12+\frac12\sin\big((4L+1)\epsilon\big)\approx\frac12+\frac{(4L+1)\epsilon}{2}

    Prepare ∣0⟩\ket0, apply GG once to reach the equator (where sensitivity is greatest), then G4LG^{4L}, and measure.

  3. δϵ≈2 δP4L+1≈1(4L+1)N\delta\epsilon\approx\frac{2\,\delta P}{4L+1}\approx\frac{1}{(4L+1)\sqrt N}

    With NN shots, δP≈1/(2N)\delta P\approx1/(2\sqrt N) near P=12P = \tfrac12. The precision improves as 1/L1/L, not only as 1/N1/\sqrt N.

  4. time  ∝  NL  ∼  1δϵinstead of1δϵ2\text{time}\;\propto\;NL\;\sim\;\frac{1}{\delta\epsilon}\qquad\text{instead of}\qquad\frac{1}{\delta\epsilon^2}

    A length-LL circuit takes time ∝L\propto L, so at fixed NN the cost of reaching precision δϵ\delta\epsilon scales like 1/δϵ1/\delta\epsilon (Heisenberg-like), as long as the error stays coherent and stable over the whole sequence.

Robust phase estimation (Kimmel, Low & Yoder 2015) uses the same idea to calibrate rotation angles with Heisenberg-like scaling. The catch is that amplification only works if the errors are Markovian and stable during the long sequences.

MethodTrustsLearnsCost
State tomographythe measurementsone stateexponential in nn
Process tomographypreparations and measurementsone channel (d4−d2d^4-d^2 parameters)12n12^n configurations
Gate set tomographyonly the Markovian modelall preparations, measurements and gates, up to gaugethousands of circuits for one qubit; rarely used beyond two
Randomized benchmarking(nearly) gate-independent noiseone average error rate per gate setpolynomial
Pauli-noise learning, cycle benchmarkingPauli twirlingPauli error rates of whole layerspolynomial for sparse models

Key takeaways

  • Channels are d2×d2d^2\times d^2 Pauli transfer matrices with d4−d2d^4-d^2 free parameters; process tomography estimates them by assuming perfectly known preparations and measurements.
  • Gate set tomography models preparations, measurements and gates together; any gauge transformation Gk↦MGkM−1G_k\mapsto MG_kM^{-1}, with ρ\rho and EE transformed accordingly, leaves all data unchanged.
  • Linear GST recovers B−1GkBB^{-1}G_kB from the Gram matrix of fiducial circuits, the gauge freedom made explicit. Only gauge-invariant quantities (probabilities, spectra, RB decays) are truly observable.
  • Long germ sequences amplify coherent errors for 1/L1/L precision, and the goodness of fit (NσN_\sigma) tests whether the noise is Markovian at all.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

In the Pauli transfer matrix of a qubit channel, what does a non-zero entry in the first column below the top, such as R30≠0R_{30}\ne0, signal?

  1. The channel is not trace-preserving
  2. The channel is not completely positive
  3. The channel is not unital: it maps I\id to something other than I\id, as relaxation toward ∣0⟩\ket0 does
  4. The channel is entangling
Show the answer

C. The channel is not unital: it maps I\id to something other than I\id, as relaxation toward ∣0⟩\ket0 does

The first column is the image of I\id. Trace preservation constrains the first row; a non-zero entry lower in the first column means E(I)≠I\mathcal E(\id)\ne\id, as for amplitude damping.

Question 2

In linear GST the data give g~−1P~k=B−1GkB\tilde g^{-1}\tilde P_k = B^{-1}G_kB. What does this imply?

  1. GkG_k is recovered exactly, because BB cancels
  2. Each gate is recovered up to a common similarity transformation, the gauge, which no amount of data can remove
  3. The method fails unless the fiducials are perfect
  4. BB can be determined by measuring the fiducials separately
Show the answer

B. Each gate is recovered up to a common similarity transformation, the gauge, which no amount of data can remove

BB, the matrix of effective prepared states, is unknown and enters as a similarity transformation shared by all gates, the state and the effect. Every gauge-equivalent gate set predicts the same data.

Question 3

Why do GST and robust phase estimation use long, repeated gate sequences?

  1. To beat the 1/N1/\sqrt N shot-noise scaling by taking more shots
  2. To make the noise Markovian
  3. To twirl the noise into a Pauli channel
  4. Coherent errors accumulate linearly with repetition, so small error parameters can be estimated with precision ∝1/L\propto1/L
Show the answer

D. Coherent errors accumulate linearly with repetition, so small error parameters can be estimated with precision ∝1/L\propto1/L

An over-rotation ϵ\epsilon repeated LL times becomes LϵL\epsilon, so a measurement with fixed statistical error resolves ϵ\epsilon to about 1/(LN)1/(L\sqrt N).

Question 4

A GST fit returns Nσ=250N_\sigma = 250. What is the most reasonable conclusion?

  1. The data are inconsistent with any Markovian gate set of this form: look for drift, leakage or non-Markovian noise
  2. The gates are very bad
  3. The fit has converged to the target gate set
  4. More gauge optimization is needed
Show the answer

A. The data are inconsistent with any Markovian gate set of this form: look for drift, leakage or non-Markovian noise

NσN_\sigma compares the fit's log-likelihood with its expected χ2\chi^2 distribution. A value in the hundreds means the model cannot explain the data, whatever the quality of the gates.

Further reading

  • I. L. Chuang & M. A. Nielsen, "Prescription for experimental determination of the dynamics of a quantum black box", Journal of Modern Optics 44, 2455 (1997).
  • S. T. Merkel et al., "Self-consistent quantum process tomography", Physical Review A 87, 062119 (2013).
  • R. Blume-Kohout et al., "Demonstration of qubit operations below a rigorous fault tolerance threshold with gate set tomography", Nature Communications 8, 14485 (2017).
  • E. Nielsen, J. K. Gamble, K. Rudinger, T. Scholten, K. Young & R. Blume-Kohout, "Gate set tomography", Quantum 5, 557 (2021).
  • D. Greenbaum, "Introduction to quantum gate set tomography", arXiv:1509.02921 (2015).