Source-linked AI summary
Fault-tolerant quantum computation against biased noise
Panos Aliferis, John Preskill
TL;DR
The scheme targets reliable quantum computation under a local stochastic biased-noise model. It combines protected GCSS operations, state preparation and injection, recursive decoding, and flagging; the improved analysis gives a 3.51 × 10^-3 lower bound for universal quantum computation at bias 10^4.
Problem
Reliable universal quantum computation requires a threshold analysis for GCSS operations and state injection under a local stochastic biased-noise model.
Method
The scheme protects GCSS operations with concatenated codes, uses gate teleportation and state injection for universality, and improves decoding by retaining C1-syndrome information and raising flags for close votes.
Results
3.51 × 10^-3 is a lower bound on the accuracy threshold for universal quantum computation under bias 10^4, with state-injection error at most 10.4%, below the 14.1% distillation threshold.
Takeaways & Limitations
Reliable universal quantum computation is obtained when protected GCSS operations and sufficiently accurate state injection meet the distillation threshold.
Abstract
from arXiv · showhide
We formulate a scheme for fault-tolerant quantum computation that works effectively against highly biased noise, where dephasing is far stronger than all other types of noise. In our scheme, the fundamental operations performed by the quantum computer are single-qubit preparations, single-qubit measurements, and conditional-phase (CPHASE) gates, where the noise in the CPHASE gates is biased. We show that the accuracy threshold for quantum computation can be improved by exploiting this noise asymmetry; e.g., if dephasing dominates all other types of noise in the CPHASE gates by four orders of magnitude, we find a rigorous lower bound on the accuracy threshold higher by a factor of five than for the case of unbiased noise.
APPENDIX A: COMMENTS ON THE CNOT GADGET
The cnot gadget distinguishes repetition counts for logical measurements, later setting them equal to the repetition-code length. With staggered measurements, this scheduling supports continuous gadget execution without idle storage locations.
- APPENDIX A: COMMENTS ON THE CNOT GADGET: The logical σz measurements can use distinct repetition counts r1 and r2, while the analysis later sets r1 = r2 = n.The repetition-code length n is odd, and the equal choice is reported as optimal or nearly optimal in studied cases.
- APPENDIX A: COMMENTS ON THE CNOT GADGET: Choosing r1 = r2 = n and staggering measurements eliminates storage locations from the cnot gadget.The authors therefore exclude faults at idle storage locations from their failure-probability estimate.
- APPENDIX A: COMMENTS ON THE CNOT GADGET: With r1 = r2 = n, output-block operations finish one time step before input-block operations begin in the cnot gadget.This timing property is evident in the full circuit and allows the next gadget to execute as inputs begin interacting.
- APPENDIX A: COMMENTS ON THE CNOT GADGET: The staggered schedule lets output qubits become ready for the next gadget as corresponding input qubits start their interactions.For qubit 1, output interactions occupy time steps 1 through n, while input interactions begin at time step n+1.
APPENDIX B: THE THRESHOLD FOR GCSS OPERATIONS
The threshold analysis bounds cnot-gadget failure under biased noise by tracking dephasing, non-dephasing, preparation, and measurement faults, with the cnot gadget providing the largest GCSS-gadget bound.
- APPENDIX B: THE THRESHOLD FOR GCSS OPERATIONS: The cnot gadget contains the most fundamental operations among GCSS gadgets, so its failure bound applies to the others.This makes the cnot gadget the limiting case for the GCSS threshold analysis.
- APPENDIX B: THE THRESHOLD FOR GCSS OPERATIONS: Cnot-gadget failure can arise from σx errors on data qubits, σz errors on multiple ancillas, and errors affecting logical measurements.The analysis also accounts for faults in preceding gadgets that can propagate into the current gadget.
- APPENDIX B: THE THRESHOLD FOR GCSS OPERATIONS: A single non-dephasing fault can affect two consecutive gadgets, but its logical error can be propagated forward and charged only to the later gadget.This bounds the contribution of such a fault to one C1-protected gadget failure.
- APPENDIX B: THE THRESHOLD FOR GCSS OPERATIONS: The cnot-gadget failure bound combines non-dephasing cphase faults with dephasing, preparation, and measurement faults.The gate count is evaluated after setting r1 = r2 = r = n.
APPENDIX C: ACCURACY THRESHOLD FOR UNIVERSAL QUANTUM COMPUTATION
The scheme extends reliable C1 ▷C2-protected CSS operations to universal computation through gate teleportation, state injection, and distillation. At bias 10^4, it establishes a universal-computation threshold lower bound of 2.50 × 10^-3.
- APPENDIX C: ACCURACY THRESHOLD FOR UNIVERSAL QUANTUM COMPUTATION: Reliable CSS operations plus high-fidelity |+i⟩ and |T⟩ states enable teleportation of Q, S, and T, yielding universal computation.Q and S generate the Clifford group, while T extends the operations beyond the Clifford group.
- APPENDIX C: ACCURACY THRESHOLD FOR UNIVERSAL QUANTUM COMPUTATION: 14.1% is the tolerated input error probability reported for |T⟩ distillation when each input state is twirled with probability 1/2.The |+i⟩ distillation threshold is stated to be even higher.
- APPENDIX C: ACCURACY THRESHOLD FOR UNIVERSAL QUANTUM COMPUTATION: State injection prepares an encoded |ψ⟩ by teleporting an unprotected |ψ⟩ into a C1 ▷C2 block using an encoded Bell state and Bell measurement.The output is known up to a logical Pauli correction determined by the Bell-measurement outcome.
- APPENDIX C: ACCURACY THRESHOLD FOR UNIVERSAL QUANTUM COMPUTATION: Injection failure can result from recursive decoding errors, Bell-measurement errors, or faults in preparing the single-qubit input state.The encoded Bell-state preparation is treated as flawless when protected GCSS gadgets are reliable.
- APPENDIX C: ACCURACY THRESHOLD FOR UNIVERSAL QUANTUM COMPUTATION: 2.50 × 10^-3 is a lower bound on the accuracy threshold for universal quantum computation at bias 10^4.At this operating point, the reported injection error is ε(P|¯ψ⟩) ≤ 11.5%, below the 14.1% distillation threshold.
APPENDIX D: IMPROVED THRESHOLD VIA FLAGGING AND MESSAGE PASSING
The improved decoder retains C1 syndrome information, flags close majority votes, and uses that information in higher-level C2 decoding. For bias 10^4, this raises the universal-computation threshold lower bound to 3.51 × 10^-3.
- APPENDIX D: IMPROVED THRESHOLD VIA FLAGGING AND MESSAGE PASSING: The improved decoder retains part of the C1 syndrome instead of discarding it after C1 decoding.The retained information is used to optimize decoding of C2 in the concatenated block.
- APPENDIX D: IMPROVED THRESHOLD VIA FLAGGING AND MESSAGE PASSING: A flag is raised when a logical measurement has a close vote, indicating a higher than usual probability of gadget failure.Close votes occur when the winning majority exceeds the minority by one vote.
- APPENDIX D: IMPROVED THRESHOLD VIA FLAGGING AND MESSAGE PASSING: The construction encodes Bell states and cnot-plus-measurement blocks by replacing their basic operations with C1-protected gadgets.The analyzed C2 choice is the concatenated 4-qubit code with σz check operators.
- APPENDIX D: IMPROVED THRESHOLD VIA FLAGGING AND MESSAGE PASSING: Repeating the ancilla-state measurement fewer than n times and postselecting unflagged cases is advantageous in the flagged analysis.For this setting, the accepted conditional failure probability is analyzed separately from the probability of raising a flag.
- APPENDIX D: IMPROVED THRESHOLD VIA FLAGGING AND MESSAGE PASSING: 3.51 × 10^-3 is the improved lower bound on the universal-computation threshold at bias 10^4.The optimum uses r1 = r2 = n = 7 and t = 5; the corresponding injection error is at most 10.4%, below 14.1%.