Quantum Probability Estimation (QPE)

• Quantum Probability Estimation is a framework that extracts phase, energy, and randomness from quantum systems using measurement-theoretic and algorithmic methods.
• It encompasses multiple protocols, including standard QPE, curve-fitted QPE, and statistical methods, each tailored for resource optimization and robust error mitigation.
• The methodology employs quantum estimation factors, Hadamard tests, and entanglement-assisted techniques to achieve high-precision phase and eigenvalue estimation in quantum computing.

Quantum Probability Estimation (QPE) encompasses a set of measurement-theoretic and algorithmic frameworks for extracting probabilistic or spectral information from quantum systems. Its role spans fault-tolerant quantum algorithms, quantum randomness generation, precise Hamiltonian spectroscopy, and statistical certification of quantum experiment outputs. QPE algorithms and protocols are central to quantum computing, realizing optimal scaling in phase and eigenvalue estimation, efficiently generating certified randomness even in device-independent and adversarial scenarios, and enabling resource-efficient simulation and signal processing at quantum scale.

1. Mathematical Foundations of Quantum Probability Estimation

QPE protocols formalize the extraction of phase, energy, or randomness information from quantum circuits in both single-trial and multi-trial scenarios. In the randomness-generation context, quantum probability estimation generalizes classical probability estimation schemes—such as probability estimation factors (PEFs) chained into supermartingales (Zhang et al., 2018)—to the quantum regime using quantum estimation factors (QEFs) (Knill et al., 2018). Let $C,Z$ be classical registers and $E$ a quantum system. A QEF $F: C \times Z \to \mathbb{R}_+$ with power $\beta > 0$ for “$C|Z$” and a model $\mathcal{C}(CZ)$ satisfies

$$\sum_{c,z} \operatorname{Tr}[\rho(c,z)] F(c,z) \#_\alpha(\rho(c,z)\| \rho(z)) \leq \operatorname{Tr}[\rho]$$

where $\alpha = 1+\beta$, and $\#_\alpha$ denotes the sandwiched Rényi power of order $\alpha$, $E$0. QEF chaining provides statistical confidence bounds—directly yielding high-confidence lower bounds on the smooth conditional min-entropy in non-i.i.d., adaptive, or device-independent settings (Knill et al., 2018, Zhang et al., 2018).

For quantum phase estimation (in the circuit sense), the QPE protocol is realized by preparing an eigenstate $E$1 of a unitary $E$2 (with $E$3), applying a Hadamard layer, controlled-$E$4 operations, and inverse QFT, yielding probabilities

$E$5

from which the target $E$6 is estimated (Lim et al., 2024).

2. Algorithmic Frameworks and Resource Scaling

QPE exists in several paradigms each with distinct resource, scaling, and performance tradeoffs:

Standard QPE (QFT-QPE)

• Utilizes $E$7 ancilla/recording qubits and a system register (Witzel et al., 2023).
• Applies controlled-$E$8 gates and inverse QFT; output is an $E$9-bit estimate of the phase $F: C \times Z \to \mathbb{R}_+$0.
• Precision tradeoff: circuit depth $F: C \times Z \to \mathbb{R}_+$1, total runtime $F: C \times Z \to \mathbb{R}_+$2, variance $F: C \times Z \to \mathbb{R}_+$3 in the Heisenberg limit for $F: C \times Z \to \mathbb{R}_+$4 samples ($F: C \times Z \to \mathbb{R}_+$5) (Lim et al., 2024, Kimura et al., 13 Mar 2026).

Curve-Fitted QPE

• Hybrid quantum-classical scheme: Uses all observed QFT-QPE bitstrings to fit the exact theoretical outcome PMF, extracting $F: C \times Z \to \mathbb{R}_+$6 via nonlinear least squares (Lim et al., 2024).
• Achieves the Cramér–Rao lower bound for precision, $F: C \times Z \to \mathbb{R}_+$7.
• Retains quantum resource usage as standard QPE; classical post-processing adds negligible overhead and requires no iterative quantum-classical loop.

Hadamard Test-Based QPE (HT-QPE) and Statistical QPE

• Replaces QFT with sequences of Hadamard tests to estimate $F: C \times Z \to \mathbb{R}_+$8 at various $F: C \times Z \to \mathbb{R}_+$9 (Blunt et al., 2023).
• Statistical phase estimation methods reconstruct CDFs or perform Bayesian inference over measurement data for robust, error-mitigated eigenvalue estimation (Lin et al., 2021).
• Precision scaling: for $\beta > 0$0 the overlap with the target state, resource product obeys $\beta > 0$1 (Kimura et al., 13 Mar 2026).

Entanglement-Assisted and Bayesian Multiphase Estimation

• Utilization of optimal entangled initial states or parallel Bayesian estimation approaches to simultaneously estimate multiple eigenphases (Gebhart et al., 2020, Sakuma et al., 2024). Heisenberg scaling and exploitation of phase correlations are achieved, attaining variances $\beta > 0$2 in total photon or circuit resources.

Window-Assisted and Spectral Leakage Mitigation

• Spectral leakage, intrinsic to finite-register QPE, is reduced using window/taper functions or entangled input states, sharpening the probability peaks and reducing systematic errors, which is critical for observable estimation tasks (Apel et al., 8 Aug 2025, Sakuma et al., 2024).

3. Fisher Information Theory and Fundamental Limits

The achievable error $\beta > 0$3 in quantum phase estimation is fundamentally lower bounded by the Fisher information $\beta > 0$4 inherent in the measurement statistics. For estimation of a specific phase $\beta > 0$5: $\beta > 0$6 with $\beta > 0$7 for QFT-QPE and $\beta > 0$8 for HT-QPE, $\beta > 0$9 for uniform or arithmetic sampling. Thus,

• QFT-QPE: $C|Z$0
• HT-QPE: $C|Z$1

This establishes that QFT-QPE outperforms in the low-overlap regime ($C|Z$2), while HT-QPE is superior when $C|Z$3. State-of-the-art implementations such as QMEGS and curve-fitted QPE saturate these theoretical bounds (Kimura et al., 13 Mar 2026).

4. Practical Implementations and Error Mitigation

Empirical QPE protocols must contend with decoherence, gate infidelity, limited circuit depth, and measurement errors:

• Statistical QPE employs cumulative distribution function techniques, optimized Fourier series kernels, and advanced error mitigation (zero-noise extrapolation, randomized compiling) to achieve practical ground-state accuracy at reduced resource cost. Randomized compiling effectively converts coherent errors to stochastic noise, enabling efficient error removal (Blunt et al., 2023).
• Variational Compilation reduces gate depth in controlled evolutions, supporting high-fidelity execution on near-term and fault-tolerant hardware.
• Resource-Optimal Designs: For molecular systems, QPE cost is governed by basis choice, time-evolution strategy (Trotterization vs qubitization), and fermion-to-qubit encoding (Ku et al., 2 Oct 2025).
  • Trotter QPE: $C|Z$4 scaling in number of orbitals $C|Z$5 and precision $C|Z$6.
  • Qubitization: $C|Z$7 for $C|Z$8 electrons, $C|Z$9 orbitals in the plane-wave basis.

5. Extensions: Randomness Generation and Device-Independent Certification

QPE-based protocols enable certified randomness generation even with quantum or classical side information. The QEF framework:

• Provides sound, non-IID, and fully adaptive randomness certification (Knill et al., 2018, Zhang et al., 2018).
• Achieves asymptotic optimality at constant error (i.e., quantum conditional entropy rate) and supports exponential randomness expansion with only $\mathcal{C}(CZ)$0 entropy seed for $\mathcal{C}(CZ)$1 output bits.
• Is directly applicable to finite statistics in device-independent quantum randomness and cryptographic settings.

Improvements with QEF optimization in Bell tests provide up to two orders of magnitude reduction in data requirements compared to previous approaches based on entropy accumulation, enabling faster and more robust randomness extraction in experimental settings.

6. Open Problems and Current Research Frontiers

Current research continues to target:

• Extension of QPE protocols for simultaneous multiphase estimation, including Bayesian and entangled approaches with noise resilience and tight scaling (Gebhart et al., 2020, Sakuma et al., 2024).
• Advanced post-processing methods (curve-fitting, MLE, Bayesian inference) to extract maximal information from measurement outcomes, often attaining theoretical error lower bounds (Lim et al., 2024, Kimura et al., 13 Mar 2026).
• Window-assisted and entanglement-assisted input states to mitigate spectral leakage and improve resolution in complex spectra (Apel et al., 8 Aug 2025, Sakuma et al., 2024).
• Resource optimization for large-scale electronic structure and dynamical response simulations, adapting QPE variants to different Hamiltonian structures and basis choices (Ku et al., 2 Oct 2025).
• The formal certification of QPE circuit correctness using interactive theorem-proving systems, establishing mathematically rigorous guarantees for protocol performance and output distributions (Witzel et al., 2023).

7. Summary Table: Principal QPE Variants and Features

Method	Scaling in Precision $\mathcal{C}(CZ)$2	Resource Features	Typical Application Domains
QFT-QPE (standard)	$\mathcal{C}(CZ)$3	Depth exponential in $\mathcal{C}(CZ)$4	Fault-tolerant quantum chemistry, simulation
Curve-fitted QPE	$\mathcal{C}(CZ)$5	Readily saturates CRLB; low classical overhead	Spectroscopy, Bayesian/precision phase estimation
HT/Statistical QPE	$\mathcal{C}(CZ)$6	Lower depth, flexible error mitigation	Early fault-tolerant, error-prone hardware
Window/Entangled QPE	$\mathcal{C}(CZ)$7	Leakage-mitigated, improved error/bias	Observable estimation, spectrum extraction
Bayesian Multiphase	$\mathcal{C}(CZ)$8, $\mathcal{C}(CZ)$9 resources	Correlations in posterior, parallel resources	Multiphase and correlated signal estimation

Standard QPE, curve-fitted QPE, and advanced statistical approaches provide the theoretical and practical backbone for quantum measurement and randomness certification, while algorithmic innovations and Fisher-information theoretic analyses guide current efforts to saturate the ultimate performance limits in quantum probability estimation.

---

References:

• (Lim et al., 2024, Zhang et al., 2018, Knill et al., 2018, Kimura et al., 13 Mar 2026, Gebhart et al., 2020, Blunt et al., 2023, Ku et al., 2 Oct 2025, Witzel et al., 2023, Lin et al., 2021, Apel et al., 8 Aug 2025, Sakuma et al., 2024).