Limitations of the dissipative quantum Fisher information in Liouville space Authors: Tatiana Iakovleva, Ruvi Lecamwasam, Jason Twamley Published: 2026 ABSTRACT AbstractOpen quantum system dynamics traditionally described by a quantum master equation can be effectively reformulated in the Liouville space of vectorised density matrices. Despite an extensive use of the Liouville space framework to describe quantum states evolution, its application to quantum metrology is not well explored. Several papers have studied a Liouville-space generalisation of the quantum Fisher information, called the dissipative quantum Fisher information (DQFI). We perform a comparative study between the quantum Fisher information (QFI) in Hilbert space and DQFI, deriving an explicit relation between the two quantities for qubit, qudit, and harmonic oscillator systems. The derived relations are nontrivial, and depend on the state’s purity. We show that for a single qubit, the QFI can be efficiently recovered from the DQFI via a simple mapping, while in higher dimensions the expression becomes convoluted and does not convert to a compact form. We find examples where the DQFI is neither an upper nor lower bound for the Hilbert space QFI, and suggest that the informational content of the DQFI itself cannot be straightforwardly interpreted. Our results clarify the limitations of the Liouville space based approach and structural differences with the QFI in the context of quantum parameter estimation problem.Subject terms: Mathematics and computing, Physics Introduction Parameter estimation plays an important role in modern science, and advances in this field have led to such scientific breakthroughs as detection of gravitational waves1, improvement in magnetometers2, gravimeters3 and other force sensors as well as paving new directions such as dark matter search4. The quantum Fisher information (QFI) is one of the central tools in quantum sensing and metrology5–8. The QFI quantifies, through the Cramer-Rao bound, the ultimate sensitivity with which one can estimate a parameter of interest θ encoded in a quantum system9–12. The QFI has diverse applications in areas including quantum metrology6,13, computing5,14, information theory7, and error correction15, as well as in quantum channel estimation16. The QFI calculation in the unitary encoding case is usually straightforward14,17–19. However, realistic systems couple to an environment and dissipative processes come into play; as a consequence, the evaluation of the QFI becomes more difficult analytically20. The QFI has also been explored in such open system settings as non-Markovian environment21,22 with parameter-dependent dissipators23, non-phase-covariant noise present in the system24, and estimation of a spectral density subject to Gaussian noise channels25. Sensitivity limits for continuously monitored open quantum systems have also been studied in the works26,27, while in28 the authors show how a phase transition can be used to achieve enhanced sensitivity. Despite this rich research background, practical computation of the QFI in generic open quantum systems remains challenging, thus motivating the search for computationally simpler methods. Liouville space is a well-known technique for analytical treatment of the open quantum systems dynamics29. Within this higher-dimensional Liouville space, mixed quantum states are represented as vectors and the dynamics is governed by linear superoperators. The Liouville space formalism can be traced back to works from the 1970s, where it was widely used for master equation treatment30,31, also known as the thermo field dynamics32. The attractiveness of this method lies in its amenability to analytic treatments such as Lie-algebraic technique for open quantum systems dynamics33–35. Recent applications of the Liouville space framework include the real-time dynamics evaluation of thermal states36, studies on the relationship between environmental noise and complexity37 and quantum algorithms38,39. In this paper we will explore whether there is a way to obtain information about the QFI in such cases by mapping the evolution governed by the master equation to Liouville space. Liouville space is a useful mathematical concept, however, to calculate the QFI one needs to convert the resulting quantum state back to Hilbert space. Moreover, the geometric structure of this space has not yet been extensively studied in the context of the quantum parameter estimation problem. In the works40,41, the authors consider the dissipative evolution in the Liouville space of vectorised density matrices and derive a formula for the QFI, the so-called dissipative quantum Fisher information (DQFI). It is important to note that the DQFI can be calculated directly from the Liouville space state vector of the system, unlike the QFI, which generally requires computation of the symmetric logarithmic derivative (SLD) operator. In this sense, the DQFI may offer a computational advantage, much like the the Hilbert-Schmidt statistical speed42,43, which has been shown to provide an efficient alternative in parameter estimation tasks where the QFI tends to be computationally demanding. In this work, we aim at a conceptual understanding of the DQFI. The DQFI is derived precisely in analogy to the QFI. However, it is is not well-known what information the DQFI contains, if the QFI and DQFI show qualitatively similar behaviour, and why differences arise between the two. We find that the DQFI can show very different qualitative and quantitative behaviour compared to the QFI. In general, the relation between them can be very complicated. As a result of our studies we remain uncertain as to the usefulness of the DQFI for metrological purposes. Preliminaries In this section we introduce the concept of Liouville space. Further, we formulate the Liouville dissipative quantum Fisher information and show how it is different from the Hilbert space QFI. Introduction to Liouville space In this work we will denote the Hilbert space QFI and DQFI as and respectively. Following this, all derivatives throughout the paper are taken with the respect to the parameter θ, unless specified otherwise, which we omit for visual clarity. Consider an open system governed by a master equation in the Lindblad form Here H is the system’s Hamiltonian, is the anti-commutator, and are the Lindblad operators representing the interaction with the environment with coupling strengths . The idea behind the Liouville space formalism is to re-write this in a Schrödinger-like form29 It is always possible to write this, because the Lindblad equation Eq. (1) is linear in ρ. By making the linearity explicit, Eq. (2) is more amenable to analytic methods. For example, a formal solution can be immediately obtained as where is the initial state of the system. The Schrödinger-like form also allows us to apply Lie-algebraic methods which have been developed for unitary systems19,35,44. For a general density matrix the Liouville state vector is defined as To demonstrate the vectorisation process, we take a general pure state . Then the corresponding vectorised density matrix is given by The Liouvillian superoperator is defined as where we have defined the Liouville commutator and anti-commutator as Note that denotes the Hermitian conjugate of X, and its transpose. The open system dynamics are then modeled by Eq. (2)29,31. Since is non-Hermitian, the norm of will change with time. In fact, the norm of the Liouville state vector corresponds to the purity of the state Quantum Fisher information Let us now consider the quantum Fisher information. Suppose we have a pure state depending on a parameter θ. If θ is varied, the first order change to our state is described by The component of this which is parallel to represents a global phase shift which is not measurable. Thus the measurable change is described by The quantum Fisher information is then11 given by Substituting Eq. (8) gives us the quantum Fisher information (QFI) We use a superscript Q to denote that this is the standard quantum Fisher information, defined in terms of the Hilbert space state. This derivation shows that the QFI has a geometric meaning, as the perpendicular distance between rays in Hilbert space corresponding to and . Mathematically, this corresponds to the Fubini-Study metric on Hilbert space. However, this is defined only on pure states. For mixed quantum states we must use more complicated expressions, defined using the symmetric logarithmic derivative11. An analogue of the quantum Fisher information using Liouville-space state was introduced in40,41. This effectively substitutes the normalised Liouville space state into Eq. (9). It is necessary to normalise the state, because under the non-unitary evolution, the norm is not preserved, whereas the Fubini-Study metric is defined on rays in Hilbert space. The normalised state is denoted with a subscript N Then in terms of this, the so-called dissipative quantum Fisher information (DQFI) is defined as We note that, despite the straightforward mathematical analogy with the Fubini-Study metric in Hilbert space, it is not clear that this has the same geometrical meaning. For a general evolution, we cannot directly assign the same physical meaning as a distinguishability measure to the DQFI. The metrological usefulness of the DFQI remains an open question. For further details about Liouville space we refer the reader to a comprehensive review on the topic29. Evaluation of the DQFI To better understand what information is captured by the DQFI, we rewrite Eq. (11) in terms of the density matrix ρ, i.e. we transform vectors back to the original Hilbert space, but keep the DQFI structure. First, note that the norm N of the state depends on the parameter of interest θ. Thus when calculating the derivative of the normalised state this dependence should be taken into account where , and . Substituting this into Eq. (11) we obtain The inner product of two vectors in Liouville space can be expressed through the density matrices in Hilbert space as 29. Applying this to our density matrices, noting that both are Hermitian, we can rewrite the DQFI as The QFI is often written in terms of the Symmetric Logarithmic Derivative (SLD) operator L, defined as Then the QFI is given by11 . With this we can rewrite Eq. (14), as40 This gives us an expression for the DQFI in terms of the density matrix. Let us now compare Eq. (15) with the standard QFI. Pure states Let us first consider a pure state , where the parameter θ is encoded as a unitary evolution of ρ. To find the SLD operator, we differentiate both sides of to obtain . This gives Carefully examining Eq. (15), we calculate each term in this expression. We first find From this we evaluate as where we have used the normalisation condition of the pure quantum state , so that , which gives . Next, since we are dealing with pure states, the second term in Eq. (15), is exactly the QFI The rightmost term in Eq. (15) vanishes as . Combining everything we arrive at which agrees with the results found in40,41. Thus for pure states undergoing unitary evolution, the DQFI is always equal to twice the QFI. Mixed states The motivation of the DQFI is to understand the evolution of open quantum systems. A mixed state can be represented in terms of the density matrix eigenbasis as . Then Eq. (15) can be written as We define the normalised purity weights as In terms of these the last two terms in Eq. (21) can be written as a weighted variance of the SLD Using this we can express the dissipative quantum Fisher information as This gives us an expression for the dissipative quantum Fisher information in terms of the density matrix of the state. This expression shows how different states contribute to the DQFI based on the weights defined in Eq. (22), indicating that purer states make a larger contribution. Let us compare this with the expression for the quantum Fisher information . Note that here we have the usual variance, weighted by the probabilities and the last term vanishes since . Comparing Eq. (23) and Eq. (25), we see that in Liouville space the squared probabilities in play the role of the weights in the variance. Note that the variance in case of the DQFI is taken with the above defined purity biased weights, which we denote by the subscript w. This purity-weighted variance is different from the Hilbert space variance. In the pure state case one can show that , so the purity weighted variance becomes double the standard variance. Comparing the QFI with the DQFI Eq. (24), there does not appear to be a clear relationship between the two. To gain more insight, in the following sections we will study the relationship between the two for several physical systems. Results We will now explicitly calculate the QFI and DQFI for qubits and qudits. This will shed light on the physical meaning of the DQFI. Moreover, qubits and qudits are important systems, and Liouville space is a powerful tool to study their dynamics under the effects of decoherence and noise. We will derive an explicit formula for the QFI in terms of the Liouville space vector, allowing us to analyze metrology with these systems entirely in Liouville space. We will consider a completely general evolution and parameter dependence. Qubit system A single qubit density matrix can be written as , where are the Pauli matrices, and \vec r is the Bloch vector. Then Eq. (14) becomes Using the Pauli matrix identity we can simplify Eq. (26) and obtain the DQFI in Liouville space for a qubit in terms of the Bloch vector \vec r as Let us compare this with the quantum Fisher information. A general formula for the QFI of a qubit is6 Using the explicit dependence of the density matrix on the Bloch vector, we can rewrite Eq. (28) as following We will now proceed to find an analytic relation between the Eq. (29) and Eq. (27). Dividing Eq. (29) by P/2 we get then we can show We can further simplify Eq. (31), by noting that the numerator in the rightmost term of Eq. (31) can be expressed as , which is related to the change of purity . Inserting this into Eq. (31), we arrive at an expression for the relation connecting the QFI in Hilbert space to the DQFI in terms of the quantum state purity For pure states, we again have . However in general, this expression shows that contains information not just about metrology, but also the purity and its rate of change. We must thus be cautious when using the DQFI to study metrological problems. We can also rewrite Eq. (32) in terms of the Liouville space vector to obtain: where is given by Eq. (11). This equation allows us to compute the QFI utilising only Liouville space quantities. It shows a simple relationship between the DQFI and QFI in Hilbert space, however, it is not clear whether is useful for quantum metrology. Qudit system Let us now consider a general d-dimensional system. Qudits, while used less often than qubits, have a great potential in quantum computing and metrology. For example, it has been shown that utilising d-dimensional space of a qudit one can improve magnetometry45, and quantum computing46. A qudit state can be represented by a generalised Bloch vector6 where . Here are generalised Gell-Mann matrices. They a basis for the set of d× d traceless Hermitian matrices, and satisfy where and are completely symmetric and anti-symmetric rank 3 tensors. It follows that . The purity of the state is then Since the SLD is Hermitian , we can also expand this in the basis of generalised Gell-Mann matrices where are scalar coefficients, and . Then the QFI is We find coefficients and \vec l by substituting decompositions of the SLD (Eq. (37)) and the density matrix into . Next, we use Eq. (15) to obtain the resulting dissipative quantum Fisher information where the matrix M is defined as For a qubit case we have the following parameters d=2, and , and we recover the previous result Eq. (32). Again we can see that the DQFI contains information beyond the QFI. The first and the second terms in Eq. (39) are similar to the qubit case and proportional to the QFI and purity change correspondingly. However, the third and the fourth terms represent the higher-dimensional structure of the SU(d). This calculation also shows that the simple relationship that existed for a qubit does not hold in higher dimensions. Does the DQFI provide a bound on the QFI? In the previous section, we established a connection between the DQFI and QFI for qubit and qudit systems. It is natural to ask whether the DQFI can be considered a bound on the QFI in Hilbert space. In the work40, the authors derive inequalities relating the QFI and DQFI. However, these inequalities involve complex functions of the quantum state. In this section we study two more general examples showing an unexpected behaviour of the DQFI, demonstrating that it does not bound the QFI either from above or below. Thermometry with a thermal state of a quantum harmonic oscillator Temperature is one of the most important physical quantities, and the metrological task of thermometry has been extensively studied. To estimate the temperature of an environment, traditionally one places a probe inside the medium and, after allowing it to thermalise, infers the temperature by measuring the probe’s state. Since the temperature is not a quantum observable (i.e. there is no operator corresponding to temperature), one must estimate it with the help of quantum parameter estimation theory. We will consider estimating the temperature of the thermal state of a quantum harmonic oscillator. First, we examine the QFI for estimating the temperature T of the thermal state of a quantum harmonic oscillator. The thermal state is given by where β =1/T, and we are using units where . Then the QFI is given by47,48 Now, let us calculate the DQFI, which is given by Eq. (11) as The vectorised and normalised thermal state is with given by Eq. (41), and the state’s norm is given by . Now let us calculate . We note that the Fock number states |n〉 are independent of the temperature T, so we only need to take the derivative of the eigenvalues Substituting the derivatives into Eq. (43) and summing over n, we arrive at the expression for the DQFI for temperature estimation Using , we can express the DQFI as Rewriting the purity in terms of a hyperbolic function , or equivalently , we obtain Eq. (48) shows that for this problem the DQFI is always less than the QFI. Equality is reached only for states with P=1 (at zero temperature T=0), or P=0 (infinite temperature T=∞). We plot the QFI and DQFI in fig. 1. We can see that the curves have different shapes; for the same value of ω, maxima of the QFI and DQFI occur at different temperatures. However, it is also clear that for this problem, the DQFI is always smaller than the QFI. It is reasonable to ask if this is a general principle. In the next section we will demonstrate that this is false. Parameter estimation of the temperature T of the thermal state of a harmonic oscillator. Blue curve shows the DQFI, and orange the QFI. The dashed black line corresponds to the lower bound on the QFI according to40. The upper bound is not present in the plot as it diverges. The oscillator’s frequency is ω =0.6. The DQFI is always lower than the QFI for all ω, attaining equality when T=0 and T=∞. In40, the authors derived explicit lower and upper bounds on the QFI in terms of the DQFI. These bounds are given by an complex expression involving the eigenvalues of the density matrix, the DQFI, and, in the case of the upper bound, the symmetric logarithmic derivative. We plot these bounds for the two examples considered in this work. In the thermal state example, we note that the upper bound diverges, because the lowest eigenvalue . Thus, this bound does not provide any information. On the other hand, the lower bound presents a very different quantitative behaviour and cannot serve as a good estimate of the QFI. From this example, we can observe that the bounds defined in40 are neither tight nor accurate estimators of the QFI. This is due to the fact that the DQFI contains a direct contribution of the state’s purity change. Our earlier findings indicate that this purity information is not easy to separate in general. Two coupled qubits Let us now consider a case where the purity of a subsystem can oscillate in time. Consider two qubits, initially in the separable state , evolving under the Hamiltonian with the propagator where θ is the parameter we are estimating. This unitary evolution can be seen as a phase control gate on two qubits, or a very simplistic Ising model, discarding any transverse terms. Since , one can easily compute the time evolved state as Then the reduced density matrix of the first qubit is with the purity oscillating in time as . It is easy to see that the derivative with respect to θ is . To find the QFI, we expand the SLD operator in the Pauli basis as for scalar coefficients . As before we can substitute this and Eq. (51) into the definition of the SLD, and equate coefficients of the different Pauli matrices. This gives From this we find the QFI to be Now let us consider the DQFI. The normalised state after the evolution is where is the purity we calculated earlier. Next, we differentiate the state and calculate the DQFI according to Eq. (11) We then plot the QFI and DQFI in fig. 2. One can observe that the DQFI in Liouville space oscillates and periodically reaches zero, while the QFI in Hilbert space steadily grows over time. Moreover, we note in some regions , while in others . This example shows that the DQFI in Liouville space cannot be, in general, considered as an upper or lower bound on the QFI. When the evolution of the system changes the purity of the state, it is highly questionable what information the DQFI provides about parameter estimation. Comparison of the DQFI and Hilbert space QFI for the coupling constant θ between two qubits. [Left] Evolution of the state on the Bloch ball. The Bloch vector periodically oscillates on the x-axis back and forth between . Arrows show the Bloch vector at time t=0 (orange), t=π /6 (blue), and t=π /2 (red). [Middle] Comparison between the QFI and DQFI. The two plots exhibit a completely different behaviour. Solid (dashed) black lines correspond to the upper (lower) bound according to40. The upper bound has a critical point at times when the state becomes pure. [Right] Purity of the qubit A as a function of time. The state periodically becomes pure at times for integer n. Additionally, we plot the bounds derived in40 in fig. 2. We observe an oscillatory behaviour of the lower bound similar to that of the DQFI, which is very different from the QFI. The upper bound diverges at times when the state of the qubit becomes pure, and the corresponding minimum eigenvalue . This shows how the purity contribution significantly affects the DQFI and the bounds derived from it, but has no such effect on the QFI. Conclusion In this work we investigated the dissipative quantum Fisher information (DQFI), defined by analogy with the QFI for pure states, applied to the Liouville space state vector. The motivation for this is to study metrology of open quantum systems, a crucial task for the development of real-world quantum technology, and tests of fundamental physics. However, fundamentally it is the QFI that has physical meaning. We have found analytical relationships between the two for metrology problems in qubits, qudits, thermal states of harmonic oscillators, and coupled two-level systems. These show that there is a highly nontrivial relationship between the two. Thus given knowledge of the DQFI, it is not at all straightforward to translate this to knowledge of the QFI, and hence parameter estimation. Moreover, the DQFI does not act as a bound on the QFI, as it can be lower, higher, or even oscillate between lower and higher in the same problem. Bounds on the QFI in terms of the DQFI were derived in40. However these can show significant divergence from the QFI, both quantitatively and qualitatively. Thus precise physical meaning and utility of the DQFI remains elusive. Our work demonstrates that the DQFI contains a mixture of information about changes in the parameter, and also changes in purity of the state. This is interesting, given that the two quantities are introduced with identical motivation to geometrically quantify the distinguishability of perturbations to the quantum state. We argue that this throws doubt on if the DQFI is indeed the correct informational quantifier for metrology with open quantum systems. Liouville space is a useful tool to study the dynamics of open quantum systems. For lower-dimensional systems, such as a single qubit or two coupled qubits, one can remain in Liouville space, exploiting the linear structure of dynamics. However, in general cases, the connection between the QFI and DQFI is nontrivial and may still require knowledge of the symmetric logarithmic derivative, limiting the computational advantage of the Liouville space approach. In comparison, other methods for open quantum system’s dynamics, like Stinespring dilation theorem, involve purification to a larger Hilbert space, followed by projection back to the original space. In such framework, one may require to perform optimization over pure states to calculate the QFI. Moreover, these maps bypass the necessity to consider measurements in Liouville space, and allow one to evaluate the QFI in Hilbert space where the physical interpretation of statistical distance and measurement precision is physically valid. This work considered only single parameter estimation, and we studied only time-independent Liouvillians. An extension to a multiparameter estimation and time-dependent operators can be seen as a natural generalisation of this work. Additionally, one could consider a non-Markovian dynamics. Moreover, a key advantage of Liouville space is its amenability to Lie algebraic techniques, and it would perhaps be interesting to study Lie algebraic reformulation of the DQFI. Acknowledgements This work was supported by the Okinawa Institute of Science and Technology Graduate University. Author contributions T.I. conducted the theoretical research. R.L. and J.T. contributed to the conceptualization and discussion. All authors contributed to writing and reviewing the manuscript. Funding R. L. was supported by a Competitive Research Program NRF-CRP30-2023-002. Data availability All data generated or analysed during this study are included in this published article. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement All data generated or analysed during this study are included in this published article.