198
Part Five · Convergence

Noise as a feature: using device noise to model biological environments

Everywhere else in this book, hardware noise is the enemy: it limits circuits, demands error correction, and costs millions of qubits. Biology, though, happens in noisy environments, and simulating a molecule in a protein means simulating the noise too, which classically costs extra. Why not let the quantum computer’s own decoherence play the role of the protein’s? Several groups have tried, with some success. The idea is appealing and hard to make rigorous.

On the plate: matching a device’s noise to a target environment.

The idea

An open biological system evolves by a master equation with dephasing and relaxation (Chapter 161). A real qubit evolves by approximately the same kind of equation, with relaxation rate 1/T11/T_1 and pure dephasing rate γφ=1/T2−1/2T1\gamma_\varphi = 1/T_2 - 1/2T_1 (Chapter 135). If a simulated time step is mapped to a physical time chosen so that the qubit’s accumulated dephasing equals the target environment’s, the device’s own noise implements the bath. Groups simulating radical-pair dynamics on IBM processors reported that leaning on the native noise reproduced noisy quantum beats better than some simple classical approximations, with good agreement with experimental data. Debated

Match the bath

Device T1=T_1 = 100 µs and T2=T_2 = 80 µs; target dephasing over one simulated time unit 0.50. The plate computes the device’s pure dephasing rate and the physical idle time per simulated time unit needed to match. Then it adds the catch: device parameters drift by tens of percent over hours, and the noise spectrum is not the one the protein has. The target curve (white) and the device-emulated curve (amber, with its drift band) separate. Debated

The critique

Device noise is uncontrolled. It includes relaxation, dephasing, crosstalk, leakage and slow drift, with spectra set by fabrication defects, not by biology. Matching a single rate is easy; matching a structured, temperature-dependent spectral density with memory is not. And results can be hard to verify: if a noisy simulation agrees with experiment, is it because the noise models the protein, or because the observable is insensitive to the details? Controlled noise injection, as in the analog experiments of Chapter 195, is more defensible than relying on native noise. The approach is a clever use of imperfect hardware, not yet a reliable method. Debated

What you now have

The idea of using hardware decoherence as a simulated bath, how to compute a device’s pure dephasing rate and match a target, why drift and spectral mismatch limit it, and why controlled noise injection is preferable.

· · ·

Three questions to carry with you.

How is pure dephasing obtained from T₁ and T₂?tap
γ_φ = 1/T₂ − 1/(2T₁).
Why is native noise hard to trust as a bath?tap
It drifts and has the wrong spectrum and memory for a protein environment.
What is more defensible?tap
Deliberate, calibrated noise injection.

And one to work out. With T1=100T_1 = 100 µs and T2=80T_2 = 80 µs, what is γφ\gamma_\varphi in µs⁻¹?

γ_φ =

Sources: H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems (OUP, 2002). P. Krantz et al., Appl. Phys. Rev. 6, 021318 (2019) (T₁, T₂ and drift in superconducting qubits). A. Potočnik et al., Nat. Commun. 9, 904 (2018) (controlled noise injection). Radical-pair simulations on IBM hardware: see the sources of Chapter 196. The drift band (±20%) is illustrative.

Next · Chapter 199
Quantum annealing for biological optimization problems

Turn protein design into a magnet problem and let an annealer search it.

Turn the page