⚛️ quantum physics

Reconstructing Quantum States and Expectations via Dynamical Tomography

This paper characterizes the feasibility of reconstructing quantum states and expectations via dynamical tomography using Krylov-based methods, providing deterministic and randomized tests for Markovian dynamics and demonstrating the approach's capabilities and limits through applications to spin chains and electron-nuclear systems.

Original authors: Marco Peruzzo, Tommaso Grigoletto, Francesco Ticozzi

Published 2026-02-17

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Marco Peruzzo, Tommaso Grigoletto, Francesco Ticozzi

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

Imagine you are trying to figure out the exact shape of a mysterious, invisible object hidden inside a dark room. You can't see it, and you only have one small flashlight to shine on it.

In the world of quantum physics, this is the problem of Quantum State Tomography. Scientists need to know the exact "shape" (state) of a quantum system (like an atom or a group of qubits), but they usually can only measure a few specific properties. If the system is huge, measuring every single part directly would take forever and require impossible amounts of equipment.

This paper, "Reconstructing Quantum States and Expectations via Dynamical Tomography," proposes a clever trick: Don't just look at the object; watch it move.

Here is the breakdown of their idea using simple analogies:

1. The Problem: The "One-Flashlight" Limit

Normally, to fully map a complex object, you need a whole set of different flashlights shining from every angle (a complete set of observables).

The Old Way: If you have a giant, complex machine with 100 parts, you might need 100 different sensors to see everything. This is expensive and slow.
The Limitation: What if you only have one sensor? In the past, scientists thought, "If you only have one flashlight, you can never fully map the whole object."

2. The Solution: The "Dancing Shadow" (Dynamical Tomography)

The authors say: "Wait! If we know exactly how the object moves (its dynamics), we can use that movement to our advantage."

Imagine the object is a dancer in the dark. You only have one camera (one observable) that can only see the dancer's left hand.

Static View: If the dancer stands still, you only know where the left hand is. You have no idea about the right hand, the head, or the feet.
Dynamic View: If the dancer starts spinning, jumping, and waving, their left hand will eventually move to positions where it reveals information about the rest of their body. By watching the trajectory of that single left hand over time, you can mathematically reconstruct the position of the entire dancer.

In physics terms, they let the quantum system evolve (change) according to known rules (like a Hamiltonian or a specific type of noise). By measuring the same observable at different moments in time, they effectively get "new angles" of the system without needing new sensors.

3. The Magic Tool: "Observability"

How do they know if watching the dancer is enough? They use a concept from engineering called Observability.

The Analogy: Think of a car engine. If you listen to the sound of the exhaust (one measurement), can you tell if the pistons are firing correctly?
- If the engine is "observable," the sound of the exhaust changes in a unique way for every possible internal problem. You can diagnose the whole engine just by listening.
- If it's "unobservable," two different broken engines might make the exact same exhaust sound. You can't tell them apart.
The Paper's Contribution: They developed a mathematical test (using something called "Krylov subspaces," which is like a specialized map of all possible movements) to prove exactly when a system is "observable" and when it isn't.

4. The Surprising Twist: Noise is Good!

Usually, in quantum physics, "noise" (dissipation, or the system losing energy to the environment) is seen as a bad thing that destroys delicate information.

The Paper's Discovery: Surprisingly, for this specific task, noise is a superpower.
The Analogy: Imagine trying to solve a puzzle.
- Unitary (Clean) Dynamics: The pieces slide around perfectly but never change shape. If you only have one piece to look at, you might get stuck in a loop where you can't see the rest of the picture.
- Dissipative (Noisy) Dynamics: The pieces get shuffled, stretched, and mixed up by the environment. This "messiness" actually helps spread information from the hidden parts of the system out to the part you can measure.
Result: The authors prove that for many systems, you can reconstruct the entire state of a complex network of qubits just by measuring one single spot, provided there is enough "noise" (dissipation) in the system. Without noise, you might need to measure many spots.

5. The "Smart Selection" Algorithm

Even with this trick, you can't measure at every single moment in time (that would take forever). You need to pick the best times to measure.

The Analogy: You are taking photos of a spinning top. Taking a photo every millisecond is wasteful. Taking a photo every hour misses the action. You want to take photos at the exact moments the top reveals a new, unique angle.
The Paper's Tool: They created a step-by-step recipe (an algorithm) that picks the best observable and the best time to measure it. It works like a detective: "Okay, I've seen the left hand move up. Now, what is the next thing I should watch to learn something I don't already know?" This saves time and resources.

6. Real-World Examples

They tested their theory on two real-world scenarios:

A Chain of 4 Spins: Like a row of magnets. They showed that with just a little bit of noise, they could figure out the state of the whole chain by looking at just the middle two magnets.
Electron-Nuclear Systems (Diamonds): This relates to "NV centers" used in quantum sensors. They showed that even if you can only measure the electron (which is easy), you can figure out what the nucleus (which is hard to measure) is doing, just by watching how they interact over time.

Summary

This paper is like a guidebook for quantum detectives. It tells us:

You don't need a million sensors to map a quantum system.
If you know the rules of how the system moves, you can use time as a substitute for extra sensors.
Noise (usually the enemy) can actually be your best friend in revealing hidden information.
There is a smart way to choose when to look so you get the most information with the least effort.

It turns the problem of "not having enough tools" into a game of "using time and motion to do the heavy lifting."

1. Problem Statement

Quantum State Tomography (QST) is the fundamental task of reconstructing an unknown quantum state $\rho$ via repeated measurements. Standard QST requires an informationally complete (IC) set of observables, meaning the number of required measurements scales prohibitively with system size (typically $d^2 - 1$ for a $d$ -dimensional Hilbert space).

The paper addresses the scenario where only a limited set of observables is available for measurement, but the dynamics of the system are known. The core question is: Can we reconstruct the initial state (or specific expectation values) by letting the system evolve under a known dynamical map before measuring the limited set of observables? This approach is termed Dynamical Quantum State Tomography (DQST).

The authors aim to:

Establish rigorous conditions for when DQST is feasible.
Develop practical algorithms to select optimal measurement times and observables.
Analyze the limits of DQST for unitary (closed) vs. dissipative (open) dynamics.
Address the problem of reconstructing expectations of unmeasurable target observables without full state reconstruction.

2. Methodology

The authors employ a framework rooted in Linear System Theory and Control Theory, specifically utilizing observability analysis.

Vectorization and Linear Representation:
The quantum evolution of observables in the Heisenberg picture is mapped to a linear system. By vectorizing operators (stacking columns of matrices into vectors), the evolution of an observable $X_i$ is described by $\dot{x}_i(t) = \hat{L} x_i(t)$ (continuous time) or $x_i[t+1] = \hat{\Phi} x_i[t]$ (discrete time), where $\hat{L}$ and $\hat{\Phi}$ are super-operators derived from the Lindblad generator or the quantum channel.
The measurement output is $y(t) = r_0^\dagger x(t)$ , where $r_0$ is the vectorized initial state.
Observability Analysis:
The problem of DQST is equivalent to the observability of the linear system $(\hat{L}, C)$ or $(\hat{\Phi}, C)$ , where $C$ represents the measurement operators.
- Observable Subspace ( $\mathcal{O}$ ): The span of all evolved measurement operators $\{ \Phi_t(X_i) \}$ .
- Non-observable Subspace ( $\mathcal{N}$ ): The orthogonal complement of $\mathcal{O}$ .
- Feasibility Condition: DQST is feasible for all states if and only if the system is observable, i.e., $\mathcal{O} = \mathcal{B}(\mathcal{H})$ (the space of all bounded operators), or equivalently, $\mathcal{N} = \{0\}$ .
Krylov Subspace Methods:
Instead of evaluating trajectories over infinite time, the authors use Krylov subspace properties to determine observability. For a system of dimension $d$ , the observable subspace is generated by applying the generator $L$ (or map $\Phi$ ) at most $d^2 - 1$ times to the initial set of observables. This leads to the construction of Observability Matrices (analogous to the Kalman rank condition) whose rank determines feasibility.
Heuristic Selection Algorithm (AOT):
To minimize experimental resources, the authors propose an Algorithm for Observable and Time selection (AOT). This iterative heuristic selects the next observable and time step that maximizes the projection onto the orthogonal complement of the currently spanned subspace (maximizing "new" information), thereby improving the signal-to-noise ratio and reducing the Mean Squared Error (MSE).

3. Key Contributions

A. General Feasibility Framework

The paper proves the equivalence of three conditions for a quantum system:

The system is observable (in the control theory sense).
Every state is Uniquely Dynamically Determined among All (UDDA).
DQST is feasible for every initial state.
This framework holds for general Completely Positive Trace-Preserving (CPTP) evolutions, not just Markovian ones.

B. Markovian Dynamics and Genericity

Simplified Tests: For time-homogeneous Markovian dynamics, the authors provide deterministic tests using the rank of the continuous-time or discrete-time observability matrices (involving powers of the generator or the map).
Parametric Genericity: They prove that if a system is observable for one set of parameters in a parametric family, it is observable for almost all parameter sets (with respect to Lebesgue measure). This allows for randomized testing of feasibility.

C. Unitary vs. Dissipative Dynamics

Unitary (Closed) Systems: The authors revisit the "no-go" results. For a system of dimension $d$ , at least $d$ linearly independent observables are required for DQST under unitary dynamics. Specifically, for multipartite systems (e.g., qubit networks), DQST from single-site observables is generally impossible for $N > 3$ qubits under Hamiltonian dynamics.
Dissipative (Open) Systems: In contrast, the paper demonstrates that purely dissipative dynamics (Lindblad generators with zero Hamiltonian) can render a system observable even with a single observable (plus the identity). This highlights the beneficial role of noise/dissipation in enhancing tomographic capabilities.

D. Reconstructing Unmeasurable Expectations

The paper addresses Problem 3: predicting the expectation of a target observable $Z$ that cannot be directly measured.

Condition: The expectation of $Z$ is reconstructible if and only if $Z$ lies within the observable subspace $\mathcal{O}$ generated by the available measurements and their evolution.
Implication: Full state reconstruction is not necessary to predict specific properties; one only needs the target observable to be in the span of the measured data.

E. Optimization and Error Analysis

The authors derive an upper bound for the Mean Squared Error (MSE) of the state estimate, showing it depends on the trace of the inverse of the observability Gramian.
The AOT algorithm is shown to significantly reduce the MSE compared to random selection or worst-case scenarios by iteratively selecting the most "orthogonal" (informative) measurement sequences.

4. Results and Examples

The framework is validated through two numerical examples:

4-Spin Chain:
- Hamiltonian Dynamics: With nearest-neighbor interactions, the system was found not observable using single-site measurements, confirming theoretical bounds.
- Dissipative Dynamics: Adding a local Lindblad dissipation term made the system observable. The AOT algorithm successfully identified a set of measurement times and operators (requiring a maximum evolution time of $\approx 2\tau$ , where $\tau$ is the decay time) to reconstruct the state.
- Performance: The squared error scaled linearly with the number of measurement repetitions ( $1/N$ ), validating the statistical model.
Coupled Electron-Nuclear System (NV Center Model):
- Setup: A simplified model of a Nitrogen-Vacancy center where only the electronic spin is measurable.
- Result: The full state of the system was not observable (observable subspace dimension 8 < 64).
- Partial Reconstruction: However, the target observable (nuclear spin $Z$ ) was found to lie within the observable subspace. The authors successfully reconstructed the expectation value of the nuclear spin using only measurements on the electron spin at two specific time points ( $t=0$ and $t=50$ ), demonstrating the utility of the "shadow tomography" approach.

5. Significance

Resource Efficiency: The work provides a theoretical and practical pathway to perform tomography on large quantum systems where full IC measurements are experimentally infeasible.
Role of Dissipation: It fundamentally shifts the perspective on noise, showing that engineered dissipation can be a resource for state reconstruction, enabling full tomography from single-site measurements in regimes where unitary dynamics fail.
Control-Theoretic Bridge: By rigorously applying observability analysis to quantum systems, the paper unifies quantum information theory with control theory, providing robust tools (Krylov methods, PBH tests) for feasibility analysis.
Practical Algorithms: The introduction of the AOT heuristic offers a concrete, computationally efficient method for experimentalists to design measurement protocols that maximize information gain, addressing a previously open problem in the literature.

In conclusion, this paper establishes that knowledge of dynamics can substitute for a lack of measurement resources, provided the system's evolution is sufficiently mixing (often aided by dissipation) and the measurement strategy is optimized using observability principles.