⚛️ quantum physics

Identifiability in Quantum State, Process, and Network Tomography

This paper establishes a unified framework using the Fisher Information Matrix to analyze the identifiability of Quantum State, Process, and Network Tomography, demonstrating that while the first two achieve unique parameter reconstruction under Informationally Complete settings, Quantum Network Tomography's identifiability is strictly constrained by network topology and monitor placement, with its rank determined by the path-link incidence matrix.

Original authors: Athira Kalavampara Raghunadhan, Matheus Guedes De Andrade, Don Towsley, Indrakshi Dey, Daniel Kilper, Nicola Marchetti

Published 2026-09-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Athira Kalavampara Raghunadhan, Matheus Guedes De Andrade, Don Towsley, Indrakshi Dey, Daniel Kilper, Nicola Marchetti

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the quiet, controlled world of quantum physics, scientists are constantly trying to map the invisible. They deal with particles that exist in multiple states at once and connections that span across networks of delicate hardware. To make sense of this, researchers use a set of tools called tomography. Think of it as a way to build a complete picture of something hidden by taking many different slices of data from the outside. There are three main types of this work. First, there is the task of figuring out the exact condition of a single quantum object, like a tiny particle of light or an atom. Second, there is the job of understanding how a machine or a channel changes that object as it passes through. Finally, there is the challenge of looking at a whole network of these connections and trying to guess what is happening inside each individual link, even when you can only see the start and the end of the journey. Knowing the difference between these tasks is crucial because the rules for solving them are not the same. While you can often design a perfect experiment to see a single object clearly, a network is constrained by its physical layout, and sometimes the information simply isn't there to be found.

A team of researchers from Trinity College Dublin, the University of Massachusetts Amherst, and South East Technological University has recently clarified exactly where the line is drawn between what can be known and what remains a mystery in these three areas. They focused on a mathematical tool that measures how much useful information a specific set of measurements provides. In their study, they treated the problem of identifying a quantum state, a quantum process, and a quantum network link as variations of the same core puzzle. They found that for single objects and single channels, the solution is straightforward: if you choose a specific, Informationally Complete set of measurements—such as four input states that span the necessary mathematical space—you can determine the unknowns with a finite bound on precision. However, for a network, the situation changes fundamentally. The ability to identify the health of each link depends entirely on the paths the probes take through the system. If the paths overlap in a way that makes two different links look identical from the outside, no amount of extra data can separate them.

The researchers tested their ideas using a specific model of a quantum network, imagining a simple star-shaped setup where three links meet at a central point. They simulated sending test signals, or probes, through different combinations of these links. When they sent probes along two specific routes that covered all three links, they discovered a surprising limitation. Even though every link was touched by a probe, the data collected could not tell them the individual strength of each connection. The math showed that the information was stuck in a loop; the system could not distinguish between a scenario where the first link was strong and the others weak, and a scenario where the first was weak and the others strong, because a specific rescaling transformation (where the first link's parameter is multiplied by a factor while the others are divided by that same factor) leaves every outcome probability unchanged. This meant that the unknown parameters were not identifiable. The researchers confirmed that simply repeating the experiment more times, or using more copies of the signal, would not fix this. It is like trying to solve a puzzle where you have all the pieces but they are all the same color; having more pieces of that color does not help you see the picture.

To solve this, the team showed that the network needed a specific type of probe path: one that went directly to a single link without passing through others. When they added a direct probe to their set of measurements, the mathematical barrier vanished. The data suddenly became capable of distinguishing every link individually. Their analysis proved that the key to solving the network puzzle was not just having enough data, but having the right kind of data structure. They demonstrated that for single objects and channels, the limit is purely about having a complete set of measurement angles. But for networks, the physical arrangement of the probes matters just as much as the measurements themselves. If the paths do not provide independent information about every single link, the system remains blind to certain details, no matter how many times you look.

This work provides a clear rulebook for anyone trying to build or monitor a quantum network. It tells engineers that they cannot rely on volume alone to overcome structural blind spots. If the network topology prevents a probe from isolating a specific link, that link will remain a mystery. The study confirms that while we can perfectly characterize a single quantum device or a single channel with the right experimental design, characterizing a network requires a careful, strategic choice of probe paths that ensures every part of the system is seen from a unique angle. Without this strategic design, the fundamental information needed to understand the network simply does not exist in the data.

Technical Summary: Identifiability in Quantum State, Process, and Network Tomography

Problem Statement
This work addresses the fundamental problem of identifiability across three related quantum parameter-estimation tasks: Quantum State Tomography (QST), Quantum Process Tomography (QPT), and Quantum Network Tomography (QNT). While QST aims to reconstruct an unknown density matrix and QPT characterizes an unknown quantum channel, QNT seeks to infer individual link parameters from end-to-end probe measurements collected at accessible monitor nodes. A critical distinction exists among these problems regarding the conditions required for unique parameter determination. In QST and QPT, the experimenter can typically select an Informationally Complete (IC) measurement set. In contrast, QNT is constrained by network topology and monitor placement, meaning the admissible probe paths dictate the available information. The paper investigates whether the unknown parameters in these systems can be uniquely determined from measurement statistics, utilizing the Fisher Information Matrix (FIM) as a unifying analytical tool.

Methodology
The authors analyze the three tomography problems through a common framework based on the Fisher Information Matrix (FIM), which quantifies the information about unknown parameters contained in measurement outcomes. The FIM is defined as:
$[F(\theta)]_{ij} = \sum_{y} \frac{1}{p_y(\theta)} \frac{\partial p_y(\theta)}{\partial \theta_i} \frac{\partial p_y(\theta)}{\partial \theta_j}$
where $\theta$ is the parameter vector and $p_y(\theta)$ are outcome probabilities. Full rank of the FIM is the necessary condition for a finite Cramér-Rao Bound (CRB) along every parameter direction; rank deficiency indicates directions where statistics leave parameters undetermined.

The study models the systems as follows:

QST: A general two-qubit state described by 15 independent real parameters in the Pauli basis. The FIM is derived for a Positive Operator-Valued Measure (POVM).
QPT: A general single-qubit Completely Positive Trace-Preserving (CPTP) channel in the affine Bloch representation, characterized by 12 real parameters. The analysis considers known input states and Pauli-basis output measurements.
QNT: A quantum network modeled as an undirected graph where links are single-qubit depolarizing channels with unknown parameters $w_l$ . The analysis employs joint Bell-state measurements on cyclic probe paths. The effective Werner parameter for a path $P$ is $X_P = \prod_{e_l \in P} w_l^2$ .

A key mathematical derivation for QNT involves the path-link incidence matrix $A$ . The relationship between path parameters and link parameters is log-linear ( $\log X = 2A \log w$ ). The authors factorize the QNT FIM ( $F_{QNT}$ ) to show that its rank is determined by the rank of the incidence matrix $A$ . Specifically, $F_{QNT} = G^\top C G$ , where $G$ and $C$ are non-singular matrices for valid parameter ranges. Consequently, $\text{rank}(F_{QNT}) = \text{rank}(A)$ .

Key Results
The paper presents a comparative analysis of identifiability conditions and the impact of measurement resources (number of copies, $N$ ):

QST and QPT: Under Informationally Complete (IC) settings, both QST and QPT achieve full-rank FIMs.
- For QST, nine local Pauli settings ( $P \otimes Q$ ) yield a full-rank FIM for 15 parameters ( $\lambda_{\min} = 1.00$ ).
- For QPT, a configuration using four input states spanning $\mathbb{R}^4$ with Pauli-X, Y, and Z output measurements yields a full-rank FIM for 12 parameters ( $\lambda_{\min} = 0.49$ ).
QNT Identifiability: Identifiability in QNT depends strictly on the probe path set covering the network links independently.
- In a three-link star network, using two probe paths ( $P_{12}, P_{13}$ ) results in an incidence matrix with rank 2, despite covering all three links. This leads to a rank-deficient FIM (one zero eigenvalue), rendering the link parameters unidentifiable regardless of the number of copies. The null direction corresponds to a rescaling of link parameters that leaves path probabilities invariant.
- Adding a direct probe path ( $P_1$ ) increases the rank of the incidence matrix to 3, making the FIM full rank and the parameters identifiable.
Effect of Copies ( $N$ ): The paper demonstrates that increasing the number of independent copies per measurement configuration scales the FIM eigenvalues linearly ( $F_N = N F_1$ ) but does not alter the rank.
- If the probe set is insufficient (rank deficient), the FIM remains singular for any $N$ , and no finite CRB exists along the null direction.
- If the probe set is sufficient, increasing $N$ linearly increases eigenvalues, causing the trace of the inverse FIM (representing estimation error) to decay as $1/N$ , thereby improving precision.

Significance and Claims
The authors claim that this work provides a unified FIM-rank-based perspective on identifiability, highlighting a fundamental shift when moving from state/channel characterization to network-level inference. The primary contribution is the demonstration that QNT imposes a structural requirement on top of the measurement-completeness conditions found in QST and QPT.

The paper asserts that reliable network characterization requires not only sufficient measurement resources (copies) but also a probe path topology that provides independent information about the underlying links. Since additional copies cannot compensate for a rank-deficient incidence matrix, the design of probe paths is critical for link inference in scenarios where direct access to every network link is unavailable. These results establish a theoretical basis for assessing identifiability across quantum tomography problems and motivate further research into QNT design constraints imposed by topology and monitor placement.

More like this