Local ergotropy dynamically witnesses many-body localized phases Preprint: APS/123-QED F. Formicola 1,∗,†, G. Di Bello 1,∗,†, G. De Filippis 2,3, V. Cataudella 2,3, D. Farina 1, C. A. Perroni 2,3 Abstract Many-body localization is a dynamical phenomenon characteristic of strongly interacting and disordered many-body quantum systems which fail to achieve thermal equilibrium. From a quantum information perspective, the fingerprint of this phenomenon is the logarithmic growth of the entanglement entropy over time. We perform intensive numerical simulations, applied to a paradigmatic model system, showing that the local ergotropy, the maximum extractable work via local unitary operations on a small subsystem in the presence of Hamiltonian coupling, dynamically witnesses the change from ergodic to localized phases. Within the many-body localized phase, both the local ergotropy and its quantum fluctuations slowly vary over time with a characteristic logarithmic law analogous to the behaviour of entanglement entropy. This showcases how directly leveraging local control, instead of local observables or entropies analyzed in previous works, provides a thermodynamic marker of localization phenomena based on the locally extractable work. Numerous studies have investigated dynamical localization phenomena, in which quantum many-body systems fail to reach thermodynamic equilibrium Nandki ; Abanin ; Alet ; Sierant . Many-body localization (MBL), characteristic of strongly interacting and disordered many-body systems, is qualitatively different from localization of non-interacting systems. The hallmark of MBL is the development of non-local correlations, which, however, lead only to a logarithmic growth of entanglement entropy over time serbyn2013universal ; vedi_parametri ; lukin2019probing . Since directly measuring the entanglement entropy is notoriously challenging (see lukin2019probing ), it is fundamental to use alternative experimental markers schreiber2015observation ; smith2016many ; Choi ; Kohlert ; gong2021experimental to characterize MBL which still has many elusive aspects Sierant . The properties of MBL can be relevant not only for fundamental research, but also for applications. One promising field of application is that of quantum batteries campaioli2024colloquium ; le2018spin ; ferraro2018high . Recent works have focused not only on the charging process and energy storage andolina2018charger ; farina2019 ; mitchison2021charging ; dou2022cavity ; yao2022optimal ; mazzoncini2023optimal , but also on work extraction andolina2019 ; arjmandi2022performance , where the main investigated quantity is the ergotropy joshi2022experimental ; hu2022optimal , defined as the maximum amount of work extractable from an isolated quantum system via cyclic unitary transformations allahverdyan ; alicki2013entanglement . Assuming \hat{H} to be the system Hamiltonian and \hat{\rho} the statistical operator describing the quantum state, the ergotropy \mathcal{E} is a functional of \hat{\rho} and \hat{H} : \begin{split}\mathcal{E}\hskip 1.42262pt&[\hat{\rho},\hat{H}]\coloneqq\max\limits_{\hat{U}\in\hskip 1.42262pt\mathcal{U}(d)}\Tr{\hat{H}\bigg(\hat{\rho}-\hat{U}\hspace{0.5mm}\hat{\rho}\hspace{0.8mm}\hat{U}^{\dag}\bigg)}.\end{split} (1) The maximization has to be carried out over the set \mathcal{U} of all available unitary operators \hat{U} acting on the system d -dimensional Hilbert space \mathcal{H} , yielding a quantity which does not vary over time for an isolated system campaioli2024colloquium ; allahverdyan ; lenard1978thermodynamical ; pusz1978passive . It was shown in Ref. Polini that the dynamical behaviour of the ergotropy of subsets of quantum cells of a quantum spin chain distinguishes Anderson localization (AL), MBL and ergodic (ERG) phases. The low level of entanglement in MBL phase ensures significantly superior work extraction capabilities compared to ERG phase. However, the protocol used in Ref. Polini requires an energy cost for switching on/off the interaction between the subsystems, and, as a consequence, a form of non-local control. Moreover, a clear signature distinguishing AL and MBL phases comes only from the comparison of their temporal energy fluctuations Polini . The main aim of the present work is to show that a dynamical signature of localization phenomena is provided by a recently defined quantity, the local ergotropy, the maximum work extractable by local unitary operations, i.e., without turning off any Hamiltonian coupling SalviaGiovannetti . We analyse the one-dimensional disordered XXZ Heisenberg model, which is one of the most commonly used to investigate localization effects Abanin ; Sierant , and which is potentially realizable in laboratory murmann2015antiferromagnetic . Our results are obtained using numerical techniques based on matrix product state (MPS) representation of quantum many-body states and operators perez2006matrix ; white1992density ; schollwock2011density . We employ the time dependent variational principle (TDVP) to compute the time evolution haegeman2011time ; haegeman2016unifying , and a Bayesian optimization algorithm (BOA) frazier2018tutorial to evaluate the maximum extractable work via local operations. We investigate the dynamics of several quantities, focusing on the local ergotropy and its quantum fluctuations for a small subsystem within the spin chain. The subsystem can be interpreted as an open quantum battery experiencing interaction with its surrounding (the rest of the chain) under non-weak coupling conditions. This differs from alternative perspectives where full chains are considered as quantum batteries grazi2024prldimer and generally require forms of nonlocal control to extract work. Over time, we have found that, in the ERG case, the local ergotropy rapidly drops to zero, in the AL phase, it stabilizes to a constant value, and, in the MBL phase, it slowly decreases following a characteristic logarithmic law. Moreover, the quantum fluctuations exhibit a similar logarithmic time behaviour which, like for entanglement entropy, turns out to be a robust feature. Model and numerical methods. — We consider a one-dimensional disordered XXZ Heisenberg model with nearest-neighbour interactions between 1/2 spins. The Hamiltonian reads as \begin{split}\hat{H}=&\frac{J_{\perp}}{2}\sum\limits_{i=1}^{N-1}\bigg\{\hat{S}_{i}^{+}\hat{S}_{i+1}^{-}+\hat{S}_{i}^{-}\hat{S}_{i+1}^{+}\bigg\}\\ &+J_{z}\sum\limits_{i=1}^{N-1}\hat{S}_{i}^{z}\hat{S}_{i+1}^{z}+\sum\limits_{i=1}^{N}h_{i}\hskip 2.84526pt\hat{S}_{i}^{z},\end{split} (2) where \hbar=1 . For each chain site i , we have the Pauli operator \hat{S}_{i}^{z}=\frac{1}{2}\hat{\sigma}_{i}^{z}=\frac{1}{2}\big(\ket{\uparrow}\bra{\uparrow}_{i}-\ket{\downarrow}\bra{\downarrow}_{i}\big) and the ladder operators \hat{S}_{i}^{-}=\ket{\downarrow}\bra{\uparrow}_{i} and \hat{S}_{i}^{+}=\hat{S}_{i}^{-}{}^{\dagger} . J_{\perp} is the spin-flip transverse coupling energy, J_{z} is the longitudinal coupling energy, and h_{i} are on-site random energy values, related to an effective magnetic field, extracted by a flat probability distribution of width 2W , such that h_{i}\in\big[-W,W\big]. (3) Hard-wall boundary conditions are imposed for simplicity on a chain with size N in the subspace S^{z}_{tot}=0 , being S_{tot}^{z}=\sum_{i=1}^{N}S_{i}^{z} the z-component of the total spin. This model can be mapped to a one-dimensional disordered spinless electron model with nearest-neighbour particle-particle interactions, through the Jordan-Wigner transformation suppmat . We remark that this is one of the most investigated models for localization phenomena, as it exhibits ERG, AL and MBL phases Abanin ; Alet ; Sierant . In the following, we will use J_{\perp} as reference energy scale. Being the full compound SE always in a pure state its ergotropy ( 1 ) (later on referred as global ergotropy) coincides with its excitation energy (constant in time) suppmat . However, having a unitary operator acting on the whole Hilbert space means that an experimenter has full control over the whole isolated system. This is sometimes an unrealistic assumption. We shall now consider a different scenario, in which the isolated quantum system is partitioned into two subsystems S and E . Hence, the total Hamiltonian \hat{H} reads: \hat{H}=\hat{H}_{S}\otimes\hat{I}_{E}+\hat{I}_{S}\otimes\hat{H}_{E}+\hat{V}_{SE}, (4) where \hat{H}_{S} and \hat{H}_{E} are the Hamiltonians for the subsystems S and E , respectively, while \hat{V}_{SE} is the interaction operator. In the following, we will assume that the experimenter has control only over the subsystem S , having the d_{s} -dimensional Hilbert space \mathcal{H}_{S} . This leads us to consider S as an open system, in contact with E , which plays the role of an environment. The system SE is initially in a high energy state (a charged state of the potential battery) which is not an eigenstate of the Hamiltonian \hat{H} , such that \hat{\rho}(t) is the resulting time-evolving statistical operator. AL and MBL are dynamical phenomena, where non-stationary quantities are used as markers for localization Abanin ; Sierant . We preferably choose the first two spins of the chain ( i=1,2 ) as subsystem S , with the remaining N-2 spin to be the environment E . The interaction energy between S and E is then related to the bond between the spins at i=2 and i=3 (see upper panel of Fig. 1 ). The main aim is to investigate whether localization effects can be inferred by actions on just two spins, representing the minimal subsystem with interacting degrees of freedom Iemini2016 . This, for instance, differs from the analytical study of local ergotropy of Ref. SalviaGiovannetti on the same model, where the subsystem S was identified with the first spin of the chain, under some approximations and without focusing on the different phases of the model. We perform numerical simulations using MPS techniques to represent the quantum states and operators from ITensor library fishman2022itensor . For all the simulations presented in this work, the system is initialized in the Néel state: \ket{\psi}=\ket{\uparrow, \downarrow, \uparrow, \downarrow, \dots, \uparrow, \downarrow} , typically used for MBL dynamics Abanin ; Sierant . Indeed, Néel state has high mean energy in comparison with the ground state and it is a product state in the spin basis. Adopting the 2-TDVP haegeman2011time ; haegeman2016unifying , we compute the time-evolved quantum state. Since the model is disordered, every simulation starts from a different stochastically extracted configuration of disorder. The quantities at each time step t are obtained by averaging over all the realizations of the disorder. A closed formula for the local ergotropy is known only for a subsystem made of a single spin-1/2 SalviaGiovannetti . Therefore, for our subsystem with two 1/2 spins, we approach the problem numerically. For each realization of disorder and time step, we use a BOA to approximate the optimal local operator for the work extraction task suppmat . The approach mimics in spirit the one of Ref. Grazia and, rigorously speaking, leads to a lower bound for local ergotropy (for simplicity, also referred as local ergotropy in the numerical results). Results. —The entanglement entropy for an isolated system partitioned into two interacting subsystems S and E is defined as the Von-Neumann (VN) entropy of one of these two: \begin{split}S_{ent}(SE)&=-\Tr_{S}\big\{\hat{\rho}_{S}(t)\ln{\hat{\rho}_S(t)}\big\}=S_{VN}(t).\end{split} (5) being \hat{\rho}_{S}(t)=\Tr_{E}\hskip 2.84526pt\big\{\hat{\rho}(t)\big\} the subsystem density operator. It is well known in the literature Abanin ; serbyn2013universal ; vedi_parametri that, for a quantum spin chain partitioned into two equal halves ( N/2,N/2 ), the time evolution of entanglement entropy is a marker for MBL and AL phases. Localized systems show a time behaviour logarithmically dependent on the strength of the interactions: S_{ent}(t)\sim\ln( J_z t/\hbar) , Abanin ; eisert-arealaw ; universal-slow-growth ; zeng2023logarithmic ; toniolo2024dynamics . Conversely, for ergodic systems, a ballistic growth over time is observed: S_{ent}(t)\sim v\hskip 1.42262ptt . Further details on time-dependent entanglement entropy for this partition are provided in the End Matter, where the ERG case is analyzed, finding perfect agreement with established results. The following numerical results analyze the case in which the partition is different. As depicted in the upper panel of Fig. 1 , our subsystem now consists of 2 spins, while the remaining N-2 sites form the environment. In the lower panel of Fig. 1 , we find a qualitative agreement between the entropy dynamics of this two-spin partition and the standard N/2,N/2 case, which is included for comparison in Appendix A of End Matter. At initial time, the Néel state has zero entanglement entropy. Then, there is an initial quick rise of the entanglement, until J_{\perp}t\simeq 1 , corresponding to the expansion of wave packets to a size comparable to the localization length. After this expansion, in the AL phase ( J_{z}=0 ), the entanglement entropy saturates to a constant, but, in the MBL phase ( J_{z}\neq 0 ), it increases logarithmically for any non-zero interaction strength. As discussed in Appendix B of End Matter, these results are robust upon increasing the chain size N . \begin{overpic}[images/Fig1.png] \put(-2.4,37.6){\scalebox{1.5}{{(a)}}} \end{overpic} \begin{overpic}[images/Fig2.png] \put(-2.5,47.0){\scalebox{1.5}{{(b)}}} \end{overpic} Figure 1 : (a) Quantum spin chain divided into the subsystem S (first two spins), and environment E (remaining spins). The initial Néel state is depicted. (b) Entanglement entropy as a function of time, for the 2,(N-2) partition, comparing the MBL phase ( J_{z}/J_{\perp}=0.2 , green triangle) and AL phase ( J_{z}/J_{\perp}=0.0 , blue circle). Results are obtained by averaging over 10^{3} disorder realizations, with chain length N=8 , and disorder strength W/J_{\perp}=5 . We show now how local ergotropy can be used as a direct probe for discerning ergodic and localized phases. One can consider a unitary quench on the subsystem S while this is still interacting with the environment. Upon optimization, this approach leads to the recently investigated Grazia ; perciavalle2024extractable local ergotropy introduced in Ref. SalviaGiovannetti , that reads as, \begin{split}&\mathcal{E}_{S}\hskip 1.42262pt[\hat{\rho}(t),\hat{H}]\coloneqq\\ &\max\limits_{\hat{U}_{S}\in\hskip 1.42262pt\mathcal{U}(d_{S})}\Tr{\hat{H}\bigg(\hat{\rho}(t)-(\hat{U}_S \otimes\hat{I}_E)\hspace{0.5mm}\hat{\rho} (t)\hspace{0.5mm}(\hat{U}_S^{\dag} \otimes\hat{I}_E)\bigg)},\end{split} (6) and is a time-dependent quantity (see also castellano2024ELE ; castellano2025PE for more recent extended versions). In Fig. 2 , we compare the local ergotropy as a function of time in the three different phases: AL, MBL, and ERG. Figure 2 : Local ergotropy as a function of time in the three phases: AL (blue circle) for J_{z}/J_{\perp}=0.0 and W/J_{\perp}=5 ; MBL (green triangle) for J_{z}/J_{\perp}=0.2 and W/J_{\perp}=5 ; and ERG (red diamond) for J_{z}=0.2 and W=0 . The inset provides a zoomed-in view of the local ergotropy as a function of time for the two localized phases. Results are obtained by averaging over 10^{3} disorder realizations, with chain length N=8 . Since the system is initially prepared in the high-energy Néel state, we can interpret the subsystem as a charged battery. The post-work extraction state depends on the specific procedure responsible for work extraction. We have mostly simulated the system in the regime W/J_{\perp}\gg 1 , which ensures that the system is localized and the disorder energy scale dominates the dynamics. Fig. 2 shows that in the ERG case ( W=0 , no disorder), after a time transient analogous to that observed in the entanglement entropy ( J_{\perp}t\simeq 1 ), the local ergotropy rapidly decreases to zero. This decay is characterized by small oscillations due to finite-size effects (more on this in Appendix B of End Matter suppmat ). Consequently, ergodic batteries undergo a rapid discharge, rendering them incapable of sustaining work extraction over long durations. In contrast, within localized phases, local ergotropy remains much higher. The order of magnitude of the local ergotropy is about W/2 in the regime W/J_{\perp}\gg 1 suppmat . It can be concluded that, within the observed time scales, localized batteries facilitate the energy storage for a much longer duration and the extraction of a greater amount of work in comparison with ergodic batteries. Focusing now specifically on the difference between the two localized phases, it can be observed that after the initial transient, the spins entangle only weakly with each other, as characteristic of localization. From this point on, a clear distinction emerges between the AL and MBL phases: in the AL phase, local ergotropy stabilizes at a constant value, just like the entanglement entropy. On the other hand, in the MBL case, local ergotropy shows a logarithmic decrease over time, in a manner analogous yet opposite to the entanglement entropy, which instead logarithmically increases. It is thus evident that the time behaviour of local ergotropy can be used as witness for localization phenomena (see also End Matter and suppmat ). Indeed, \begin{split}\mathcal{E}_{S}=(e^{c}_{S}+e^{c}_{int}+e^{c}_{E})-(e^{f}_{S}+e^{f}_{int}+e^{f}_{E}),\end{split} (7) where e^{c}_{S} ( e^{f}_{S} ) represents the energy of the subsystem S , e^{c}_{E} ( e^{f}_{E} ) is the energy of the environment E , and e^{c}_{Eint} ( e^{f}_{int} ) the interaction energy between S and E , for the global charged (post-work extraction) state. While the difference e^{c}_{E}-e^{f}_{E} can be analytically checked that is zero SalviaGiovannetti , in localized phases also the difference e^{c}_{int}-e^{f}_{int} is expected to be small. Therefore, in localized phases, the local ergotropy is controlled by the difference e^{c}_{S}-e^{f}_{S} derived at each time step from the global system evolution. Nevertheless, local ergotropy turns out to be a dynamical witness of the system phases. Thus it can serve as a marker of localization dynamics, directly leveraging local control rather than local projective measurements or entropies. The second investigated procedure for work extraction involves disconnecting the subsystem from the environment before extracting work from the subsystem S . This procedure has the energy cost \Delta_{SO}(t)=-\Tr_{SE}\hskip 2.84526pt\big\{\hat{V}_{SE}\hskip 2.84526pt\hat{\rho}(t)\big\}, (8) where \Tr denotes the trace operation. The subsystem ergotropy is the maximum amount of work extractable via unitary operations \hat{U}_{S} acting only on the subsystem S , which is now isolated: \mathcal{E}_{SS}\hskip 0.56905pt[\hat{\rho}_{S}(t),\hat{H}_{S}]\coloneqq\max\limits_{\hat{U}_{S}\in\hskip 0.56905pt\mathcal{U}(d_{S})}\Tr_{S}{\hat{H}_{S}\bigg(\hat{\rho}_{S}(t)-\hat{U}_{S}\hskip 1.42262pt\hat{\rho}_{S}(t)\hskip 1.42262pt\hat{U}_{S}^{{\dagger}}\bigg)}. (9) This is also a time dependent quantity. Finally, the total amount of work extractable from S , including the previous disconnection, is defined as switch-off ergotropy SalviaGiovannetti , \mathcal{E}_{SO}\hskip 1.42262pt[\hat{\rho}(t),\hat{H}]\coloneqq\mathcal{E}_{SS}\hskip 1.42262pt[\hat{\rho}_{S}(t),\hat{H}_{S}]-\Delta_{SO}(t). (10) One important consideration is that turning off the interactions implicitly requires control over E as well. We compare the two aforementioned protocols for local and switch-off ergotropy. Figure 3 : Local ergotropy \mathcal{E}_{S} (green circle) and switch-off ergotropy \mathcal{E}_{SO} (orange triangle), both in units of J_{\perp} , as functions of time for the system in MBL phase. The inset provides a view over the difference between local ergotropy and the switch-off ergotropy \mathcal{E}_{S}-\mathcal{E}_{SO} . Results are obtained by averaging over 2.6\cdot 10^{3} disorder realizations, with chain length N=12 , disorder strength W/J_{\perp}=5 , and J_{z}/J_{\perp}=0.2 . The plot in Fig. 3 for the switch-off ergotropy can be related to the results previously found in Polini . Moreover, it reveals that the switch-off ergotropy is smaller than the local ergotropy, albeit slightly. This is because the energy term \Delta_{SO} in Eq. ( 8 ) is small in localized phases. Since the longitudinal interactions are small compared to the other energy scales ( J_{z}\ll W,J_{\perp} ), the difference between the local and the switch-off ergotropy remains small in localized phases. We also perform work extraction without BOA, observing that it is equivalent to the switch-off protocol, allowing a gain of 1\% on the local ergotropy value suppmat . Figure 4 : Relative quantum fluctuations of the local ergotropy as a function of time in the MBL ( J_{z}/J_{\perp}=0.2 , green triangle) and AL phase ( J_{z}/J_{\perp}=0.0 , blue circle). Results are obtained by averaging over 10^{3} disorder realizations for the AL case and approximately 2.6\cdot 10^{3} disorder realizations for the MBL case, both with chain length N=8 , and disorder strength W/J_{\perp}=5 . We have finally computed quantum fluctuations of the local ergotropy following Ref. Quantum-correlations-francica : \begin{split}\sigma^{2}(t)=\Tr[(\hat{H^{\prime}}-\hat{H})^{2}\hat{\rho}(t)]-\left\{\Tr[(\hat{H^{\prime}}-\hat{H})\hat{\rho}(t)]\right\}^{2},\end{split} (11) in which \hat{H}^{\prime}=\hat{U}_{S}^{{\dagger}}\hat{H}\hat{U}_{S} . Numerical results in Fig. 4 show that the fluctuations behaviour over time also represents a marker for MBL and AL phases. In analogy with the behaviour of the entanglement entropy, after a transient of ballistic growth, the fluctuations saturate to a constant value in the AL case, while they grow logarithmically in the MBL case. For both phases, the short-time value is due to the quantum fluctuations of the initial state and the disorder strength. In the Appendix C of End Matter, we show that increasing the disorder strength W enhances the value of local ergotropy while, remarkably, reducing relative quantum fluctuations. Disorder-induced fluctuations turn out to be smaller than quantum fluctuations suppmat . We find that disorder fluctuations are larger in the AL phase than in MBL phase, consistent with previous results in the literature Polini . Moreover, while disorder fluctuations increase with W , the relative disorder fluctuations remain nearly constant, unlike quantum fluctuations. Conclusions. —We have demonstrated that local ergotropy and its quantum fluctuations can serve as thermodynamic markers of localization phenomena. Specifically, in the ERG phase, the extractable work rapidly drops to zero, in the AL phase, it reaches a stationary constant value, and, in MBL case, it decreases logarithmically over time. A summary of the different dynamical signatures of localization phenomena is reported in Table 1 . Our findings open the way to characterization of different localized phases in many-body systems via direct local work extraction in the presence of Hamiltonian coupling. The analysis presented in this work can be extended to MBL systems in the presence of an electric field without disorder taylor2020experimental and generalized to other models, such as the Sachdev–Ye–Kitaev one campaioli2024colloquium ; rossini2020 and open ones de2023signatures ; di2024environment . Acknowledgements.— D.F. and G.D.F. acknowledge financial support from PNRR MUR Project No. PE0000023-NQSTI. C.A.P. acknowledges funding from IQARO (Spin-orbitronic Quantum Bits in Reconfigurable 2DOxides) project of the European Union’s Horizon Europe research and innovation programme under grant agreement n. 101115190. G.D.B., G.D.F. and C.A.P. acknowledge funding from the PRIN 2022 project 2022FLSPAJ “Taming Noisy Quantum Dynamics” (TANQU). C.A.P. acknowledges funding from the PRIN 2022 PNRR project P2022SB73K “Superconductivity in KTaO3 Oxide-2DEG NAnodevices for Topological quantum Applications” (SONATA) financed by the European Union - Next Generation EU. The authors acknowledge stimulating discussions with A. Russomanno, P. Lucignano, and M. Sassetti. Table 1 : Dynamical signatures of localization phenomena (yellow cells for established previous results). † † nicematrix-placeholder: NiceTabular (nicematrix) End Matter We detail some points relevant for the analysis of the disordered XXZ Heisenberg chain. In particular, we provide additional informations on the entanglement entropy for the chain partition N/2,N/2 , on the scaling behaviour of physical quantities with chain size N , and on the quantum fluctuations of local ergotropy in the case when the subsystem S consists of two spins. Appendix A Entanglement entropy for the partition N/2,N/2 Here we report the analysis on the time behaviour of the entanglement entropy in the case of the chain partitioned into two exact halves, N/2,N/2 , to highlight the differences between the three phases: MBL, AL and ERG. Figure 5 : Entanglement entropy for the N/2,N/2 partition, as a function of time in the three phases: AL (blue circle, J_{z}/J_{\perp}=0 , W/J_{\perp}=5 ), MBL (green triangle, J_{z}/J_{\perp}=0.2 , W/J_{\perp}=5 ), and ERG phase (red diamond, J_{z}/J_{\perp}=0.2 , W=0 ). The dotted line indicates the saturation value of entanglement entropy for the system. The inset provides a zoomed-in view of the local ergotropy as a function of time in the two localized cases. Results are obtained by averaging over 10^{3} disorder realizations, with chain length N=8 . These results are qualitatively identical to those obtained in the 2,(N-2) case reported in Fig. 1 , with only small differences at long times. In fact, considering a subsystem length of N/2 , the finite-size saturation value for N\geq 3 is S_{N/2,N/2}^{max}=(N/2)\ln 2>S_{2,(N-2)}^{max}=2\ln 2 . In each of the three phases, the entropy starts from zero because of the initial Néel state. For a first transient of time, each phase exhibits a ballistic growth. After that, in the ergodic phase (red diamond in Fig. 5 ) entropy keeps growing towards the saturation value (dotted line in Fig. 5 ). In the two localized phases, the entropy growth halts around J_{\perp}t\sim 1 . In the MBL phase, entropy grows logarithmically, while, in the AL phase, it remains constant. These results are fully consistent with previous theoretical and experimental results. Appendix B Dependence of physical quantities on size N Here we examine how several physical quantities scale with chain lengths N , as robustness to size increase is a key feature of MBL phenomenology Sierant . Figure 6 : Entanglement entropy for N/2,N/2 partition, and imbalance \mathcal{I} as functions of time for different sizes: N=8 (green circle), N=10 (brown diamond), and N=12 (gold triangle). Results are obtained by averaging over 1.6\cdot 10^{3} disorder realizations, with coupling J_{z}/J_{\perp}=0.1 , and disorder strength W/J_{\perp}=5 . In Fig. 6 , we report the entanglement entropy, in the case of N/2,N/2 partition, as a function of time for different sizes. Moreover, we show the time behaviour of the imbalance defined as the normalized charge difference between odd and even sites, that in the spin representation reads: \mathcal{I}=(\langle S^{z}_{o}\rangle-\langle S^{z}_{e}\rangle)/(\langle S^{z}_{o}\rangle+\langle S^{z}_{e}\rangle+N/2) (where sites first index is 1, see Refs. schreiber2015observation ; Iemini2016 ). From N=8 to N=12 , the plots of both quantities show only minor differences. The plots in the upper panel of Fig. 7 reveal that local ergotropy changes slightly with increasing N in the MBL phase. Only at intermediate values of time ( J_{\perp}t\simeq 1 ), there is a deviation with varying N . This could be attributed to the fact that the local ergotropy is obtained also through a BOA. In the localized phases, N=8 is close to the thermodynamic limit within the parameter ranges analyzed in this work. However, in the ERG phases finite-size effects are more pronounced. Finally, in the lower panel of Fig. 7 , we report the entanglement entropy, in the case of 2,(N-2) chain partition, as a function of time for different sizes. In analogy with the entanglement entropy for the N/2,N/2 chain partition (see Fig. 6 ), very small differences are obtained with increasing N . \begin{overpic}[images/Fig8.png] \put(-5.1,43.0){\scalebox{1.5}{{(a)}}} \end{overpic} \begin{overpic}[images/Fig9.png] \put(-5.1,46.6){\scalebox{1.5}{{(b)}}} \end{overpic} Figure 7 : (a) Local ergotropy and (b) entanglement entropy for 2,(N-2) partition as functions of time in the MBL phase ( J_{z}/J_{\perp}=0.2 , W/J_{\perp}=5 ) for various lengths: N=8 (green circle), N=12 (brown diamond) and N=16 (gold triangle). Results are obtained by averaging over 10^{3} disorder realizations for N=8,12 and around 6.6\cdot 10^{2} for N=16 . Appendix C Quantum fluctuations We report here details relative to quantum fluctuations of local ergotropy additional to those presented in Fig. 4 . In Fig. 8 (a) we plot the local ergotropy with increasing disorder strength W . The local ergotropy depends mostly on W for large disorder W/J_{\perp}\gg 1 suppmat . More precisely, it is of the order of W/2 . Thus, as disorder strength increases, local ergotropy and its quantum fluctuations also increase. However, as shown in Fig. 8 (b) the relative quantum fluctuations decrease with increasing W , underscoring the robustness of localization. \begin{overpic}[images/Fig10.png] \put(-5.8,43.6){\scalebox{1.5}{{(a)}}} \end{overpic} \begin{overpic}[images/Fig11.png] \put(-5.8,46.6){\scalebox{1.5}{{(b)}}} \end{overpic} Figure 8 : (a) Local ergotropy and (b) its relative quantum fluctuations as functions of time for different values of the disorder strength W in the MBL phase ( J_{z}/J_{\perp}=0.2 ): W/J_{\perp}=5.0 (green circle), W/J_{\perp}=8.0 (magenta triangle). Results are obtained by averaging over 10^{3} disorder realizations, with chain length N=8 .