L17: Error Mitigation & End-to-End Quantum Simulation Workflows for Materials

Course Slides (MSU)

Learning Objectives¶

• Distinguish error mitigation (bias removal in expectation values) from error correction (L13).

• Apply readout error mitigation, zero-noise extrapolation (ZNE), and probabilistic error cancellation (PEC).

• Assemble a full materials pipeline: structure $\to$ active space $\to$ qubit Hamiltonian $\to$ VQE/QPE $\to$ mitigated observable.

• Critically assess published “quantum advantage for materials” claims.

Mitigation vs. Correction¶

• QEC (L13): encode, detect, correct — exponential error suppression, large qubit overhead.

• Mitigation: run noisy circuits, post-process to unbias $\langle O\rangle$ — no extra qubits, but sampling overhead grows exponentially with circuit noise.

Eureka! Mitigation buys accuracy with shots; correction buys it with qubits. NISQ-era materials results live almost entirely on the mitigation side.

Readout Error Mitigation¶

Calibrate the confusion matrix $M_{ij} = p(\text{read } i \mid \text{prepared } j)$ from basis-state preparations; apply $M^{-1}$ (or constrained least squares) to measured distributions.

Scalable variants calibrate qubit-by-qubit (tensored) assuming uncorrelated readout.

Zero-Noise Extrapolation (ZNE)¶

1. Amplify noise by a known factor $\lambda$ (pulse stretching, or gate folding $G \to GG^\dagger G$).

2. Measure $\langle O\rangle(\lambda)$ at $\lambda = 1, 2, 3, \ldots$

3. Extrapolate (linear, exponential, Richardson) to $\lambda \to 0$.

Eureka! ZNE needs no noise model — only the ability to make things worse in a controlled way.

Probabilistic Error Cancellation & Friends¶

• PEC: express the inverse noise channel as a quasi-probability over implementable operations; sample with signs. Unbiased, but variance grows as $\gamma^{2d}$ with depth $d$.

• Symmetry verification: discard shots violating conserved quantities (particle number from L12 ansätze).

• Purification / virtual distillation: use $\rho^2/\operatorname{Tr}\rho^2$ from two copies to suppress stochastic errors.

The End-to-End Materials Pipeline¶

1. Structure (crystal/molecule) $\to$ mean-field calculation (HF/DFT) on CPU.

2. Active space selection: the strongly correlated orbitals (the hard part!).

3. Second quantization $\to$ Jordan–Wigner/BK mapping (L11) $\to$ Pauli Hamiltonian.

4. Solver: VQE (L12) today; QPE (L09) with QEC (L13) in the fault-tolerant era.

5. Mitigation + error bars $\to$ energy, correlators, response functions.

6. Embedding back into the classical description (DMET, DMFT, QM/MM).

Eureka! Steps 1, 2, and 6 are classical materials science. The quantum computer is one solver inside a workflow you already know.

Resource Estimation & Honest Claims¶

• Fault-tolerant costs: logical qubits $\times$ code distance $\times$ T-gate counts (connects L13).

• Compare against the classical state of the art (DMRG, tensor networks, QMC) — not against brute-force diagonalization.

• Checklist for reading papers: system size, mitigation used, error bars, classical baseline.

Mini-Lab / Capstone Launch¶

• Apply readout mitigation + ZNE to the noisy H$_2$ VQE from L12; quantify the improvement.

• Capstone project: pick a small materials-relevant Hamiltonian (Hubbard $2\times2$, LiH, spin chain), build the full pipeline, and present mitigated results with error bars.

Takeaways¶

• Mitigation removes bias from observables at a sampling cost; it is the bridge to the QEC era.

• Quantum computers slot into — not replace — the classical materials toolchain.

• Error bars and classical baselines are non-negotiable in scientific claims.