Multiparameter Quantum Cramér–Rao Bound

• Multiparameter QCRB is a fundamental limit that quantifies the minimum estimation error via the inverse quantum Fisher information matrix.
• It highlights critical trade-offs stemming from noncommuting symmetric logarithmic derivatives and intrinsic measurement incompatibility.
• Advanced strategies like symmetry exploitation and collective measurements enable closer approaches to the QCRB even with singular QFIM scenarios.

Multiparameter Quantum Cramér–Rao Bound (QCRB)

The multiparameter quantum Cramér–Rao bound (QCRB) establishes the fundamental lower limit on the precision achievable in simultaneous estimation of several parameters encoded within a quantum state. Unlike the single-parameter scenario, where the quantum Cramér–Rao bound can always be achieved via optimal projective measurements, the multi-parameter case introduces intrinsic trade-offs due to the incompatibility of quantum observables. The QCRB is tightly connected to the quantum Fisher information matrix (QFIM), and its sharpness, attainability, and structural features depend on both the properties of the QFIM and the commutation relations of the symmetric logarithmic derivatives (SLDs). The modern development of the subject incorporates geometric, information-theoretic, and measurement-theoretic analyses, as well as the role of symmetry, collective measurements, and measurement incompatibility.

1. Formal Definition and Mathematical Structure

Consider a quantum statistical model $\{\rho(\theta)\}$ on a Hilbert space $\mathcal{H}$, where $\theta = (\theta_1,\ldots,\theta_m)^T$ parameterizes the quantum state. For $N$ independent preparations, and any unbiased estimator $\hat \theta$ based on any POVM measurement $\mathcal{M}$, the covariance matrix $\mathrm{Cov}(\hat \theta)$ satisfies

$$\mathrm{Cov}(\hat \theta) \succeq \frac{1}{N} F_Q(\theta)^{-1},$$

where $F_Q(\theta)$ is the quantum Fisher information matrix (QFIM), with entries

$$[F_Q]_{ij} = \frac12 \mathrm{Tr}[\rho(\theta)\{L_i, L_j\}],$$

and the SLDs $L_j$ are Hermitian operators uniquely defined by

$$\frac{\partial \rho(\theta)}{\partial \theta_j} = \frac12\left(\rho(\theta) L_j + L_j \rho(\theta)\right).$$

This matrix inequality is understood in the sense of positive semi-definite ordering.

The equivalence

$$\operatorname{Cov}(\hat \theta)_{ij} \geq [F_Q^{-1}]_{ij}$$

holds for all $(i,j)$, and can be scalarized using a positive semidefinite weight matrix $W$ as

$$\mathrm{Tr}[W\,\mathrm{Cov}(\hat \theta)] \geq \mathrm{Tr}[W F_Q^{-1}].$$

For the case where the QFIM is singular (rank deficient), the Moore-Penrose pseudoinverse $F_Q^+$ replaces $F_Q^{-1}$: $$\mathrm{Cov}(\hat \theta) \succeq \frac{1}{N} F_Q^+.$$

2. Commutation, Incompatibility, and Trade-offs

For multiple parameters, the optimal measurements for each parameter (those that project onto the eigenstates of their respective SLDs) are generally incompatible if the SLDs do not commute: $[L_i,L_j] \neq 0.$ The average (weak) commutativity condition

$$\mathcal{E}_{ij} := (i/4) \mathrm{Tr}[\rho(\theta)[L_i,L_j]] = 0, \quad \forall\, i, j$$

is necessary for the simultaneous attainability of the QCRB for all parameters with a single measurement.

When $\mathcal{E}_{ij}\neq 0$, a trade-off arises: no single measurement can simultaneously reach the single-parameter precision limits for all parameters. The attainable region in the space of estimator variances is defined by convex envelopes (trade-off surfaces) corresponding to tight bounds such as the Holevo or Nagaoka-Hayashi bounds. In such cases, the QCRB is not always achievable, and one typically considers generalized versions such as the Holevo Cramér–Rao bound, formulated as a semidefinite program (Lu et al., 2019, Albarelli et al., 2019).

3. Symmetry-Driven Saturation: Antiunitary Symmetry and Model Design

A sufficient and physically meaningful condition for full saturability of the multiparameter QCRB is the presence of a global antiunitary symmetry (GAS) in the model. An antiunitary operator $A$ on $\mathcal{H}$ satisfies $A^{\dagger}A=I$, $A$ is antilinear, and for all $\theta$, $A \rho(\theta) A^{\dagger} = \rho(\theta)$. In a basis where $A$ acts as complex conjugation, the state is real; all SLDs can be made real-symmetric, and their commutators are purely imaginary and vanish in expectation, guaranteeing that they are simultaneously diagonalizable.

The implications are:

• The mean Uhlmann curvature $\mathcal{E}_{ij}=0$ for all $i,j$ (automatic weak commutativity).
• All SLDs share a common eigenbasis, and a projective measurement in this basis saturates the entire QFIM-based bound, i.e., $\mathrm{Cov} = F_Q^{-1}$ with no trade-off.

Explicit models with this design principle include the Mutually Conjugate Model (MCM) and the Ancilla-Assisted Model (AAMCM) for joint estimation of $(\theta,\phi)$ on the Bloch sphere, demonstrating a factor-of-two improvement in achievable variances over conventional strategies (Wang et al., 22 Nov 2024).

4. Attainability and Optimal Measurement Structure

Asymptotic and Collective Strategies

For full-rank or pure states, the QCRB can be asymptotically approached via collective measurements on many copies, using the machinery of quantum local asymptotic normality (Albarelli et al., 2019). For D-invariant, RLD-attainable, or Gaussian-shift models, the QCRB, Holevo bound, and RLD bounds coincide, and measurements (often Gaussian or collective) exist that saturate the bounds exactly.

Single-Copy Level

The conditions for saturability on a single copy are stricter. For a general mixed-state model, recent theorems establish that necessary and sufficient conditions for saturation at the single-copy level are:

1. The set of projected SLDs onto the support of $\rho$ must be mutually commuting,

$[L_{i,++}, L_{j,++}] = 0,$

where $L_{i,++}$ are support-restricted SLD blocks.

2. There exists a smooth, parameter-dependent unitary $U(\theta)$ on the support solving a nonlinear system of PDEs,

$$U^{\dagger} [ \partial_l U - U V^{\dagger} \partial_l V ]\, \rho_{++} + \rho_{++} [ \partial_l U - U V^{\dagger} \partial_l V ]^{\dagger} U = 0$$

for $l=1,\ldots,p$, with $V$ any smooth basis for $\operatorname{supp}\rho$.

If these hold, a measurement projective simultaneously onto the eigenspaces of $\{L_{i,++}\}$ (and a suitable basis in the null space) saturates the QCRB (Nurdin, 2 May 2024, Nurdin, 18 Feb 2024).

5. Geometric, Information-Theoretic, and Scalarized Perspectives

Geometric quantum estimation interprets the QFIM as the metric tensor of a Riemannian structure on the quantum state manifold, with the Uhlmann curvature quantifying the non-commutativity that blocks QCRB saturation (Li et al., 2022, Goldberg et al., 2021). Scalar versions of the QCRB are formulated using weight matrices, Cartan–Killing metrics, or $f$-mean generalizations:

• The weighted arithmetic mean: $$\frac1n \operatorname{Tr} E \ge \frac{n}{\operatorname{Tr}[F^{-1}]}$$
• The geometric mean: $(\det E)^{1/n} \ge (\det F)^{-1/n}$
• The harmonic mean: $$n/\operatorname{Tr} E^{-1} \ge (\operatorname{Tr} F/n)^{-1}$$ (Lu et al., 2019)

In models with group-theoretic (e.g., $\mathfrak{su}(n)$) structure, the intrinsic scalar QCRB is given as

$$\operatorname{Tr}[g\,\operatorname{Cov}\theta] \ge \frac{1}{4}\operatorname{Tr}[C_\psi(X)^{-1}]$$

where $g$ is the invariant metric and $C_\psi(X)$ the covariance of algebra generators (Goldberg et al., 2021). This form is invariant under parameter reparameterization and state evolution along the unitary group orbit.

6. Examples and Experimental Implementations

Key applications include:

• Simultaneous estimation of multiple optical phases with generalized multi-mode NOON-like states, where explicit QFIMs and tight scaling laws underpin optimal probe design (Zhang et al., 2017).
• Multiparameter estimation of losses in photonic chips using TMSV states and maximum-likelihood strategies, achieving empirical errors within a factor $\lesssim \sqrt{3}$ of the QCRB (2208.00011).
• Recent photonic demonstrations of QCRB saturation by LOEM strategies, utilizing mutually orthogonal probe states and entangling projections, including explicit Heisenberg scaling with the number of sequential parameter interactions (Mi et al., 12 Sep 2025).
• Realization of prioritized estimation trade-offs, where an optimal entangling measurement on two copies allows maximal precision in one parameter without total loss of precision in others (Yung et al., 10 Nov 2025).

The table below summarizes several key model types and their attainability status as formalized in recent literature:

Model Type	Saturation Condition	Measurement Type
Commuting SLDs	$[L_i, L_j]=0$	Joint projective
Global antiunitary	$A\rho A^\dagger = \rho$	Single projective (real)
General mixed, single	Support SLD commutativity + PDE unitary	Block-diag/projective
Gaussian-shift	D-invariant, $F^{RLD}=F^{SLD}$	Gaussian POVM
RLD/Holevo models	Asymptotic/collective	Collective/SDP optimal

7. Singular QFIMs and Effective Parameter Estimation

If the QFIM is singular, the model is over-parameterized due to metrological symmetries: only $k < m$ linearly independent parameter combinations can be estimated. Dimensionality reduction via parameter transformation aligns the QFIM with this effective subspace, yielding a reduced CRB for estimable parameters, while variances in the orthogonal complement diverge (Mihailescu et al., 7 Mar 2025, Namkung et al., 2 Dec 2024).

Bayesian inference in this context reveals posterior distributions that collapse onto ridges/level sets of the effective parameter(s), with the ultimate variance scaling determined by the reduced inverse QFIM.

---

In summary, the multiparameter QCRB framework unifies the limits of quantum parameter estimation under general measurement scenarios. It structurally encodes the interplay between quantum information, measurement incompatibility, group symmetries, and geometric features of parameter space. Modern theoretical and experimental developments now allow for precise tailoring of estimation strategies—often leveraging symmetry or measurement design—to attain or closely approach these ultimate quantum limits.