[Paper Review] Quantum filtering: a reference probability approach

Luc Bouten, Ramon van Handel|ArXiv.org|Aug 1, 2005

Quantum Information and Cryptography53 references22 citations

TL;DR

This paper presents a reference probability approach to quantum filtering, using noncommutative probability and Hudson-Parthasarathy calculus to derive Belavkin-Zakai and Belavkin-Kushner-Stratonovich equations for quantum systems with continuous homodyne and photon counting observations. The method employs a nondemolition change of measure via a quantum Girsanov-type transformation, enabling derivation of unnormalized and normalized filtering equations through quantum conditional expectations and a noncommutative Kallianpur-Striebel formula.

ABSTRACT

These notes are intended as an introduction to noncommutative (quantum) filtering theory. An introduction to quantum probability theory is given, focusing on the spectral theorem and the conditional expectation as the least squares estimate, and culminating in the construction of Wiener and Poisson processes on the Fock space. Next we describe the Hudson-Parthasarathy quantum Ito calculus and its use in the modelling of physical systems. Finally, we use a reference probability method to obtain quantum filtering equations, in the Belavkin-Zakai (unnormalized) form, for several system-observation models from quantum optics. The normalized (Belavkin-Kushner-Stratonovich) form is obtained through a noncommutative analogue of the Kallianpur-Striebel formula.

Motivation & Objective

• To develop a systematic, reference probability-based method for deriving quantum filtering equations in continuous time.
• To extend classical filtering techniques—particularly the Zakai and Kallianpur-Striebel formulas—to the quantum domain using noncommutative probability and stochastic calculus.
• To provide a unified derivation of quantum filtering equations for both homodyne and photon counting measurement models in quantum optics.
• To demonstrate that quantum filtering can be reduced to elementary manipulations of quantum conditional expectations under a reference measure, avoiding reliance on the innovations conjecture.

Proposed method

• Uses noncommutative probability theory, focusing on the spectral theorem and quantum conditional expectation as a least squares estimator.
• Constructs Wiener and Poisson processes on Fock space via the spectral theorem and quantum stochastic calculus.
• Applies Hudson-Parthasarathy quantum Itô calculus to model open quantum systems with Markovian dynamics.
• Implements a nondemolition change of measure using a quantum Girsanov transformation, inspired by Holevo’s method, to simplify filtering problems.
• Derives the Belavkin-Zakai equation via quantum conditional expectation under the reference measure, then applies a noncommutative Kallianpur-Striebel formula to obtain the normalized Belavkin-Kushner-Stratonovich equation.
• Validates the approach on two key quantum optical models: homodyne detection with imperfect observations and photon counting with counting process dynamics.

Experimental results

Research questions

• RQ1 How can classical reference probability methods be adapted to derive quantum filtering equations in the noncommutative setting?
• RQ2 What is the role of the spectral theorem and quantum conditional expectation in enabling quantum filtering via least squares estimation?
• RQ3 How can the quantum Girsanov transformation be used to construct a reference measure that simplifies the filtering problem?
• RQ4 What are the explicit forms of the Belavkin-Zakai and Belavkin-Kushner-Stratonovich equations for homodyne and photon counting measurements?
• RQ5 How does the noncommutative Kallianpur-Striebel formula relate the unnormalized and normalized quantum states in filtering?

Key findings

• The Belavkin-Zakai equation for homodyne detection with imperfect observations is derived as $ d\sigma_t(X) = \sigma_t(\mathcal{L}_{L,H}(X))dt + (1+\kappa^2)^{-1}\sigma_t(L^*X + XL)dY_t $.
• For photon counting, the Belavkin-Zakai equation takes the form $ d\sigma_t(X) = \sigma_t(\mathcal{L}_{L,H}(X))dt + (\sigma_t(L^*XL) - \sigma_t(X))(dY_t - dt) $.
• The normalized filtering equation for photon counting is $ d\pi_t(X) = \pi_t(\mathcal{L}_{L,H}(X))dt + \left(\frac{\pi_t(L^*XL)}{\pi_t(L^*L)} - \pi_t(X)\right)(dY_t - \pi_t(L^*L)dt) $.
• The innovations process $ d\overline{Z}_t = dY_t - \pi_t(L^*L)dt $ is shown to be a martingale, confirming that the conditional intensity is $ \pi_t(L^*L) $.
• The reference probability method avoids the innovations conjecture and relies only on quantum conditional expectation and change of measure, making it conceptually simpler than martingale-based derivations.
• The approach is general and applicable to various quantum models, including those with squeezed or thermal input noise, as noted in the discussion of extensions.