Information Geometry
Quantum Information Geometry
Classical information geometry works with commutative distributions: the Fisher metric is unique. Quantum mechanics requires non-commutative density matrices, and the Petz theorem (1996) shows there are infinitely many valid quantum analogs of the Fisher metric.
- NIST atomic clocks: quantum Cramer-Rao bound via SLD metric limits precision to around 10^{-18} relative error - a fundamental geometric limit on quantum parameter estimation
- Quantum sensors: entangled N-qubit states achieve Heisenberg scaling F_Q = O(N^2) vs. classical O(N) - geometrically: states with higher variance of the generator H
- Quantum state tomography on IBM Quantum: Umegaki divergence D(rho||sigma) certifies how close a prepared state is to the target; monotonicity under channels provides the theoretical guarantee
- Variational quantum algorithms (VQE): quantum natural gradient uses the quantum Fisher metric for optimizing parametric circuit parameters
Предварительные знания
- Density matrices and quantum states
- Classical Fisher information matrix
- Operator algebras and matrix functions
Quantum Fisher Metric and Density-Matrix Geometry
Quantum states are described by density matrices: positive semidefinite Hermitian matrices rho with Tr(rho) = 1. The space M_n^+ is closed and convex. Non-commutativity of matrices means there is no unique quantum analog of the Fisher metric: Petz proved in 1996 that every monotone Riemannian metric on M_n^+ is generated by an operator-monotone function f with f(1) = 1. There are infinitely many such functions, hence infinitely many valid quantum Fisher metrics.
For a qubit rho = (I + n * sigma)/2 (Bloch vector n, |n| <= 1) the SLD Fisher information for rotation about axis H is F_Q = 4 Var_rho(H) = 1 - (n * h)^2, where h is the unit axis vector. At the equator of the Bloch sphere F_Q = 1 (maximum). In the eigenstate of H: F_Q = 0 - the state carries no information about a rotation around its own axis.
What does the quantum Cramer-Rao bound through F_Q constrain?
Quantum Cramer-Rao bound (Helstrom): for any unbiased quantum estimator Var(theta_hat) >= 1/F_Q. For NIST atomic clocks this gives a relative precision limit near 10^{-18}. The bound is achieved by the optimal POVM on many-copy states.
Quantum Metrology and Heisenberg Scaling
Quantum metrology studies the fundamental precision limits for estimating physical parameters using quantum probes. The quantum Cramer-Rao bound Var(theta_hat) >= 1/F_Q gives a lower bound on estimation variance for any measurement strategy (POVM). For N probe particles the classical shot noise limit gives F_Q = O(N), while entangled states can achieve Heisenberg scaling F_Q = O(N^2).
NIST optical atomic clocks achieve precision near 10^{-18} relative uncertainty, approaching the quantum Cramer-Rao bound. Squeezing and entanglement in these clocks push precision beyond the standard quantum limit. The theoretical framework is exactly the SLD Fisher information: maximizing F_Q over all probe states and all measurement strategies.
Why can entangled N-qubit states achieve F_Q = O(N^2) while separable states are limited to F_Q = O(N)?
F_Q = 4 Var_rho(H). For a separable state rho = otimes rho_i and collective H = sum h_i, variances add: Var(H) = sum Var(h_i) = O(N). For an entangled state like GHZ, quantum correlations make Var(H) = O(N^2). Geometrically: the entangled state traverses a longer arc on the Bloch sphere for a unit change in theta, giving more distinguishable states per unit parameter change.
Petz Theorem and Classification of Monotone Metrics
The Petz theorem (1996) provides a complete classification of Riemannian metrics on density matrix space M_n^+ that are monotone under quantum channels. Unlike the classical case where Cencov's theorem guarantees uniqueness of the Fisher metric, the quantum case has infinitely many valid metrics - one for each operator-monotone function.
The choice of monotone metric in a quantum estimation problem determines the optimal measurement class. The SLD metric is achieved by projective measurements in the eigenbasis of the SLD operator L_i. For multi-parameter estimation, the optimal POVMs for different parameters may be incompatible - a quantum consequence of the uncertainty principle.
What does the Petz theorem say about monotone metrics on M_n^+?
Unlike the classical case (unique Fisher metric by Cencov), quantum monotone metrics are infinitely many - parametrized by operator-monotone functions. Non-commutativity of density matrices destroys uniqueness. The Petz theorem gives the complete description of the whole class: every such metric arises from some f, and every qualifying f gives a valid metric.
Connections to other topics
Quantum information geometry connects quantum physics, operator algebras, and optimal estimation theory.
- Quantum metrology — Related topic
- Quantum channels — Related topic
- Variational quantum algorithms — Related topic
Итоги
- Umegaki divergence D(rho||sigma) = Tr[rho(log rho - log sigma)]: quantum KL, monotone under quantum channels
- SLD: d_i rho = (L_i rho + rho L_i)/2; SLD metric g^{SLD}_{ij} = Tr[rho {L_i, L_j}] / 2 - the largest monotone metric
- Quantum Cramer-Rao: Var(theta_hat) >= 1/F_Q with F_Q = 4 Var_rho(H) for unitary families
- Petz theorem: all monotone metrics on M_n^+ are parametrized by operator-monotone functions f with f(1) = 1
- Petz alpha-divergence interpolates between Umegaki (alpha -> 1) and reverse divergence (alpha -> -1)