SciPost Submission Page
Boundary Chaos: Exact Entanglement Dynamics
by Felix Fritzsch, Roopayan Ghosh, Tomaž Prosen
This Submission thread is now published as
Submission summary
Authors (as registered SciPost users):  Felix Fritzsch 
Submission information  

Preprint Link:  https://arxiv.org/abs/2301.08168v3 (pdf) 
Date accepted:  20230718 
Date submitted:  20230605 15:52 
Submitted by:  Fritzsch, Felix 
Submitted to:  SciPost Physics 
Ontological classification  

Academic field:  Physics 
Specialties: 

Approach:  Theoretical 
Abstract
We compute the dynamics of entanglement in the minimal setup producing ergodic and mixing quantum manybody dynamics, which we previously dubbed {\em boundary chaos}. This consists of a free, noninteracting brickwork quantum circuit, in which chaos and ergodicity is induced by an impurity interaction, i.e., an entangling twoqudit gate, placed at the system's boundary. We compute both the conventional bipartite entanglement entropy with respect to a connected subsystem including the impurity interaction for initial product states as well as the socalled operator entanglement entropy of initial local operators. Thereby we provide exact results in a particular scaling limit of both time and system size going to infinity for either very small or very large subsystems. We show that different classes of impurity interactions lead to very distinct entanglement dynamics. For impurity gates preserving a local product state forming the bulk of the initial state, entanglement entropies of states show persistent spikes with period set by the system size and suppressed entanglement in between, contrary to the expected linear growth in ergodic systems. We observe similar dynamics of operator entanglement for generic impurities. In contrast, for Tdual impurities, which remain unitary under partial transposition, we find entanglement entropies of both states and operators to grow linearly in time with the maximum possible speed allowed by the geometry of the system. The intensive nature of interactions in all cases cause entanglement to grow on extensive time scales proportional to system size.
Author comments upon resubmission
Dear Editor,
we thank the Referees for carefully evaluating our manuscript, for their positive assesment, and their suggestions. We comment on their remarks in detail below.
Reply to Referee 1, Pieter Claeys:
1) The Referee points out an inconsistency of Eq. (21) and asks for the connections with the derivation cited in Ref. [65].
The permutation defined in (new) Eq. (21) differs from the cited reference for the following reason. The impurity interaction acts in the second layer of the circuit which generates the evolution of states in the Schrödinger picture whereas the folded impurity acts in the first layer of the supercircuit generating the evolution of operators in the Heisenberg picture. This leads to slight differences in the definition of the permutations defined in Eq. (21) and (64), respectively. In particular Eq. (21) does not correspond to Ref. [65] where the evolution of operators is described. We added a corresponding clarifying comment below the respective equations and moreover provide a simpler definition of the permutations $\sigma_\delta$.
2) The Referee asks for the entanglement dynamics, when the impurity interaction both supports a vacuum state and is Tdual.
In the minimal case of qubits demanding both Tduality and the existence of a vacuum state forces the impurity interaction to be of the form $\ket{0}\bra{0} \otimes v + \ket{0}\bra{0} \otimes w$ with diagonal singlequbit gate v and arbitrary singlequbit gate w (or with the two qubits swapped). This does not lead to entangling dynamics for the initial product states considered. For larger local Hilbert space dimension there are examples leading to nontrivial entanglement dynamics. A simple example is given by folded Tdual gates, for which the vectorized identity deals as a vacuum state. This however, is a special example, as in this case the vacuum state is also conserved upon evolution in "spatial" direction (of the gates entering the construction of the transfer matrices). Whether there are Tdual gates (for $q>2$) with vacuum state, which do not originate from such a folding procedure and which give rise to nontrivial entanglement dynamics is an open question. We now mention this after introducing Tdual gates, i.e., below Eq. (43).
3) The Referee criticizes that the discussion of the entanglement dynamics of states from generic impurity interactions is too brief.
We expanded the discussion of the entanglement dynamics for this situation, including a more detailed description of the origin of the plateaus and the subleading contributions as well as comments on different R\'enyi orders $n$. We find qualitatively similar behavior as shown in Fig. 3 with positive probability, when sampling the impurity interaction Haar random. Similarly, we also find gates, for which the subleading contribution is essentially irrelevant, with (larger) positive probability, which we now also mention in the text
Moreover, the Referee asks for the fate of the exact results upon perturbing away from Tduality or the existence of a vacuum state.
For the latter case one can obtain a rough picture as follows: For an impurity which supports an vacuum state the leading part of the transfer matrices' spectra lead to the reduced density matrix being a pure state. Perturbing away from such gates with vacuum states will cause the contribution from the leading eigenvalue to slightly differ from a pure states and hence yields nonzero entanglement thereby setting the value of $R_n$ on the plateaus. The subleading contributions, which encode the full entanglement dynamics in the unperturbed case (gates with vacuum), still give an important contribution upon small perturbation. However, they now add to nonzero entanglement entropy of the leading contribution ('on top of the plateaus') instead of the zero leading entanglement entropy in the unperturbed case. In this sense, the entanglement entropies depicted in Fig. 3, in which subleading contributions are clearly visible, indicates some proximity of the chosen gate to gates with vacuum state. We incorporated this picture into the text, when explaining the observed entanglement dynamics from generic impurities qualitatively.
In contrast, the lack of control over the subleading part of the spectrum and the associated eigenvectors of transfer matrices does not allow for a similar picture for perturbations of Tdual gates both for the entanglement dynamics of states and operators.
Additionally, as suggested by the Referee, we added Appendix C, in which we list all the impurity interactions used for numerical computations.
4) The Referee points out that the transfer matrices in the tensor network (25) correspond to the rows rather than the columns.
We changed the description of the network accordingly.
5) The Referee asks about numerical results for R\'enyi entropies of higher order $n>2$.
We have checked different orders $n$ for the R\'enyi entropies numerically. For the case of gates with vacuum and states as well as generic gates and operators, the entanglement due to the leading eigenvalues of transfer matrices is that of a pure state and hence independent from $n$. Similarly the exponential scaling with $\delta$ of subleading terms is independent from $n$. The prefactor (which we cannot compute explicitly) does weakly depend on $n$. For both operators and states in the Tdual case as well as for states in the generic case we numerically find almost no differences for different $n$. In particular the value of the plateaus for fixed $\tau>0$ does depend only very weekly on $n$. Subleading contributions are qualitatively similar for different $n$ but our analytical methods do not allow for quantifying the subleading contributions. We now comment on this when discussing numerical results, i.e., below Eq. (41) and (61) as well as in the extended Sec. 3.2.3. for states and below Eq. (69) and (109) for operators.
6) We corrected the typos pointed out by the Referee.
Reply to Referee 2:
13) As suggested by the Referee, we corrected the typos pointed out by the Referee, stated the values for $L$, $l$ and $t$ below Eq. (25), and explicitly mention the $\tau$ dependence of the subleading eigenvalue $\lambda_0$ below Eq. (40).
4) The Referee asks for the reasons to choose the interaction parameter $J=\pi/4  0.05$, when discussing Tdual gates.
This is done to ensure that the system is chaotic. In fact the point $J=\pi/4$ for qubits corresponds to the most chaotic gates (in the sense of Ref [73]). We choose $J$ slightly away from this point, to be in a more generic, but still chaotic regime. We now state this below Eq. (42) and in the caption of Fig. 2.
List of changes
1) We reformulated Eqs. (21) and (64) and comment on the origin of their differences.
2) We add discussion on impurity interactions which are both Tdual and support a vacuum state below Eq. (43)
3) We expand the discussion in Sec. 3.2.3. on the entanglement dynamics for generic impurity interactions.
4) We changed 'columns' to 'rows' in the description of the network (25) and added the concrete values of $L$, $l$, and $t$.
5) We added comments on numerical results for R\'enyi entropies of order $n>2$ below Eq. (41) and (61) as well as in the extended Sec. 3.2.3. for states and below Eq. (69) and (109) for operators
6) We explicitly mention the $\tau$ dependence of subleading eigenvalues of transfer matrices below Eq. (40)
7) We comment on the choice of the parameter $J$ in Eq. (42) below the equation and in the caption of Fig. 2
8) We corrected the typos pointed out by the Referees
9) We added Appendix C, in which we list all impurity interactions used for numerical computations.
Published as SciPost Phys. 15, 092 (2023)
Reports on this Submission
Report 1 by Pieter W. Claeys on 2023614 (Invited Report)
Report
I would like to thank the authors for their detailed response. All my questions and comments have been appropriately addressed and I am happy to recommend this paper for publication in SciPost Physics.