{"id":"f8de4c00-b607-4fd1-8481-d087f13bc837","arxiv_id":"2504.21564","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A resource-efficient quantum algorithm simulates Lindblad and non-Markovian open-system dynamics through collision models using near-term Hamiltonian simulation, with explicit trade-offs in circuit depth and qubits.","lead":"The paper designs randomized quantum algorithms that simulate open quantum systems by repeatedly colliding the system with small environment qubits, using only simple Hamiltonian simulation routines and a single extra qubit. This gives a resource-lean way to simulate noise and dissipation on near-term fault-tolerant quantum computers, and compares the cost of different simulation methods.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2's stated circuit depth inverts the Γ and βmax dependence relative to its own proof; the central resource claim does not follow as written.","rationale":"The reader flagged the unproven external bound of Lemma 3 (ν = O(t²mΓ/ε)) as the weakest assumption. That is a legitimate concern, but the most load-bearing issue is internal: even granting Lemma 3, the proof of Theorem 2 derives a circuit depth with Γβmax² in the numerator (Eq. 63), while the theorem statement and Table II place Γβmax² in the denominator (Eq. 57/64). This is not a minor typo; it inverts the parameter dependence of the central resource estimate. Since the abstract and comparisons tout specific asymptotic scalings, the claims as stated are unsupported by the paper's own algebra. The existence of additional inconsistencies in the second-order Trotter entry (Γ vs Γ^{3/4}) reinforces the conclusion that the complexity tables have not been carefully checked. I recommend rejection because the central theorem's stated complexity is not established; the authors would need to correct the theorem, Table II, and revisit the numerical comparisons before the claims can be accepted. I do not object to the algorithmic framework itself, which appears sound, but the quantitative resource claims are the main contribution and are currently unreliable.","tokens_in":37882,"tokens_out":20568,"duration_ms":192627,"concrete_test":"Recompute the circuit depth of Algorithm 2 directly from the K-collision depth formula: substitute K=mν, Δt=t/ν, β=O(sqrt(ν/t)βmax), and ν=O(t²m||O||Γ/ε) into τd = O(β²K²Δt² log(βmt||O||/ε) + mν). If the result is O(m³t³Γβmax²||O||/ε polylog) rather than O(m³t³||O||/(εΓβmax²)), then Theorem 2's statement and Table II need correction. Also re-derive the second-order Trotter entry from Eq. (68) with k=1 and compare the Γ exponent with Eq. (69)/Table II.","verdict_should_be":"REJECT","load_bearing_attack":"The proof of Theorem 2 starts from the K-collision depth of Theorem 1: τd = O(β²K²Δt² log + KτρE), with K=mν, Δt=t/ν, and β ≤ sqrt(ν/t)·O(βmax). This gives τd = O(ν m² t βmax² log + mν). Substituting ν = O(t²m||O||Γ/ε) yields τd = O(m³t³ Γ βmax² ||O||/ε · polylog), which is Eq. (63) in the text. However, the theorem statement and Eq. (64) claim τd = Õ(m³t³||O||/(ε Γ βmax²)), placing Γ and βmax in the denominator. Since ν is proportional to Γ, a larger Γ means more collisions and should increase circuit depth, not decrease it. The stated formula is therefore not a consequence of the derivation. The same inversion appears in Table II for SA-LCU, first-order Trotter, and qDRIFT (Γ in denominator). For second-order Trotter, Eq. (68) gives a Γ^{3/4} dependence that conflicts with Eq. (69) and Table II, which use Γ. Because Theorem 2 is the paper's central quantitative claim, this internal inconsistency undermines the resource comparison and the numerical benchmark if it relies on these formulas.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript develops randomized quantum algorithms for simulating quantum collision models on early fault-tolerant quantum computers. It presents a general framework in which a K-collision Markovian map is implemented by composing Hamiltonian simulations, and then specializes to Lindbladian dynamics via an (m,ν)-collision map, using a bound from Pocrnic et al. to set ν. The authors compare circuit depths for first-, second-, and higher-order Trotterization, qDRIFT, and single-ancilla LCU (SA-LCU), claiming that SA-LCU achieves circuit depth Õ(m³t³‖O‖/(ε Γ βmax²)) while using at most n+2 qubits and no block encodings. They also benchmark CNOT counts on a 10-qubit transverse-field Ising model under amplitude damping, and extend the framework to non-Markovian collisions using partial swaps between environment registers.","tokens_in":38127,"tokens_out":9478,"duration_ms":99359,"significance":"If the resource formulas were correct, the paper would provide a useful qubit-efficient, block-encoding-free route to simulating open-system dynamics via collision models in the early-fault-tolerant regime, and its comparison across Hamiltonian simulation subroutines would be a valuable reference. Strengths include a modular proof structure (Theorem 1, Theorem A3), an explicit re-derivation of the SA-LCU decomposition in Appendix B, and an end-to-end construction that avoids amplitude estimation and block encodings. The reliance on Lemma 3 from Ref. [55] is explicit and not circular. However, the central quantitative claim contains an internal inversion inconsistency, so the asymptotic and numerical comparisons are not currently supported. The numerical benchmark is also not reproducible from the text alone.","major_comments":[{"comment":"The circuit depth stated in Theorem 2 does not follow from the proof. Starting from Eq. (43) with K=mν, Δt=t/ν, β≤O(sqrt(ν/t) βmax), and ν=O(t²m‖O‖Γ/ε), the proof obtains τd = O(m³t³βmax²‖O‖Γ/ε · polylog) in Eq. (63). The theorem then states τd = Õ(m³t³‖O‖/(ε Γ βmax²)), which inverts both Γ and βmax. Since ν is proportional to Γ, a larger Γ means more collision blocks and should increase the circuit depth, not decrease it. This error propagates to Table II, where every row places Γ (and usually βmax) in the denominator; for example, the qDRIFT row should have Γβmax² in the numerator under the same substitution. The central resource comparison is therefore not established as written.","section":"Theorem 2, Eq. (57) and Eq. (64)"},{"comment":"The parameter βmax is defined with dependence on ν, since Eq. (58) contains sqrt(t/(m²ν)) βS and sqrt(t/ν) βEℓ. Because ν itself is set to O(t²m‖O‖Γ/ε), βmax is not independent of Γ, ε, and t, and the substitution into Eq. (63) does not produce a resolved asymptotic bound in fixed problem parameters. Additionally, the proof asserts that βEℓ=1 after Eq. (62) without having imposed this in Sec. IV; if the sub-environment Hamiltonian has a non-unit coefficient, the definition of βmax and the depth estimate change. The theorem should be restated in terms of fixed Hamiltonian parameters with all ν-dependences resolved.","section":"Sec. IV, Eq. (58) and proof of Theorem 2"},{"comment":"The general 2k-order Trotter formula in Eq. (68) has (Γβmax²)^{1/2+1/(4k)} in the numerator, which for k=1 gives Γ^{3/4}βmax^{3/2}. However, Eq. (69) and the second-order row of Table II report Γβmax^{3/2} in the denominator. These two expressions cannot both be correct. The comparison between second-order Trotter and SA-LCU in the following paragraph relies on this conflicting pair of formulas and therefore needs to be redone after the scaling is corrected.","section":"Sec. IVA, Eq. (68), Eq. (69), and Table II"}],"minor_comments":[{"comment":"The numerical benchmark is not reproducible from the manuscript: no code, raw data, or detailed gate-count tables are provided, and the CNOT counts depend on Qiskit optimization choices and on the Trotter-step selection rules. Please release the code or provide a complete tabulation so the claims can be verified.","section":"Sec. IVB, Fig. 3"},{"comment":"In the proof of Theorem A3, Eq. (A7) writes the second term in the sum as another ΦN(YjXj) term; from Eq. (A6) the second term should be the ΦN(XjYj) composition with XK applied on the left and YK† on the right. This appears to be a typographical error, but it should be fixed for clarity.","section":"Appendix D, Eq. (A7)"},{"comment":"The authors explicitly leave open the finite-K convergence of the non-Markovian collision model to a continuous-time non-Markovian master equation. This limitation should be stated more prominently in the abstract or introduction, since without a convergence bound the non-Markovian portion is a circuit construction rather than an end-to-end simulation with a guaranteed approximation error.","section":"Sec. V, after Eq. (89)"},{"comment":"The statement that λ→1/sqrt(Δt) is described as diverging while Δt is small; since Δt=t/ν and ν grows with t and 1/ε, it would be clearer to state the explicit asymptotic regime rather than using the arrow notation, which can be read as a limit in an unspecified parameter.","section":"Sec. IV, paragraph after Eq. (46)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's main contribution is a resource comparison for collision-model simulation, and the inversion error in Theorem 2 and Table II is serious but fixable. The authors should fully re-derive the Γ and βmax dependence, resolve the ν-dependence inside βmax, and then re-run the numerical comparison. No concerns about circularity or novel-result disclosure beyond the need for a more careful positioning relative to Ref. [66]."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the honest take: the paper has one genuinely useful algorithmic idea—the ζ-control trick that keeps LCU weight bounded when composing SA-LCU decompositions—and a clean extension to non-Markovian collisions. But the central resource theorem as stated is internally inconsistent. In the proof of Theorem 2, the depth comes out with Γ and βmax² in the numerator (Eq. 63); the theorem statement and Table II put them in the denominator. Same inversion appears for the Trotter and qDRIFT entries, and the second-order Trotter formula in Eq. (68) gives Γ^{3/4} while Table II uses Γ. This is not a cosmetic typo: it changes the claimed scaling and the relative ranking of methods. The proof is clear enough that the error is identifiable, which is reassuring, but as written the main quantitative claims do not follow.\n\nWhat is actually new: composing SA-LCU for many collisions without exponential α^K growth, and doing it with n+2 qubits and no block encodings. The non-Markovian treatment with partial swaps between environment registers is a sensible extension and the error analysis carries over from the Markovian case. The paper also makes good use of the existing discretization bound from Pocrnic et al.; it does not try to hide that dependence. The numerical benchmark is described in enough detail to be reimplemented, though no code or data file is provided—minor for a theory paper.\n\nThe main fix is to correct the algebra in Theorem 2 and Table II, and re-derive the comparison table. If the signs get flipped back to match the proof, the qualitative ranking may survive (SA-LCU still looks better than first-order Trotter/qDRIFT at high precision) but the quantitative conclusions and the numerical figures need to be redone. The authors should also explicitly reconcile Eq. (68) with the table entries.\n\nWho gets value from this: anyone working on early fault-tolerant simulation of open systems, especially the collision-model community. It is a solid framework submission, not a breakthrough. I would send it to peer review and ask for a revised version; I would not accept it as-is. The ζ-control trick is worth citing once the inconsistency is resolved.","headline":"Solid algorithmic framework with a genuinely useful LCU-composition trick, but the central depth formula is inverted in the theorem and tables—fixable but currently wrong.","tokens_in":38681,"tokens_out":4273,"would_cite":false,"duration_ms":41997,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","81S22"],"pacs":["03.67.Ac","03.65.Yz"],"model":"deepseek-v4-flash","headline":"Open-system dynamics can be simulated with collision models, a single ancilla, and no block encodings.","keywords":["quantum collision models","open quantum systems","Lindblad master equation","Hamiltonian simulation","single-ancilla LCU","qDRIFT","non-Markovian dynamics","early fault-tolerant quantum computing"],"falsifier":"Take the simplest nontrivial case—a single qubit with amplitude damping, $\\mathcal{L}[\\rho]=\\gamma(\\sigma_-\\rho\\sigma_+ - \\frac{1}{2}\\{\\sigma_+\\sigma_-,\\rho\\})$—and numerically compute the induced 1-norm distance $\\|e^{\\mathcal{L}t} - M_{m,\\nu}\\|_{1\\to 1}$ for a range of $t$, $m$, and $\\varepsilon$, finding the smallest $\\nu$ needed for $\\varepsilon$-accuracy. If the minimal $\\nu$ does not grow as $t^2 m \\Gamma/\\varepsilon$ (or the constant is not benign), the claimed $\\tilde{O}(m^3 t^3/\\varepsilon)$ circuit depth for SA-LCU must be revised. The same numerical experiment can be run with the paper's full circuits on a classical simulator for $n=2$ or $3$ qubits to verify the $\\varepsilon$-additive output of Algorithm 2.","tokens_in":37669,"feed_emoji":"⚛️","tokens_out":9079,"duration_ms":81476,"temperature":0.7,"pith_summary":"This paper attempts to establish that quantum collision models—repeated short interactions between a system and a stream of environment qubits—can simulate open-system dynamics on early fault-tolerant quantum computers using only near-term Hamiltonian simulation routines, without block-encoded oracles and with at most $n+2$ qubits. The authors prove that composing $K$ approximate Hamiltonian evolutions and tracing out the environment after each step yields an $\\varepsilon$-accurate estimate of the expectation value of an observable, provided each single-evolution simulation is accurate to $\\varepsilon/(3K\\|O\\|)$. They then use the known correspondence between memoryless collisions and Lindblad dynamics to give an end-to-end algorithm for simulating $e^{\\mathcal{L}t}$, with classical repetitions scaling as $O(\\|O\\|^2 \\log(1/\\delta)/\\varepsilon^2)$ and per-run circuit depth depending on the chosen Hamiltonian simulation method. They compare first- and higher-order Trotterization, qDRIFT, and a single-ancilla linear combination of unitaries (SA-LCU), finding SA-LCU's depth scales as $\\tilde{O}(m^3 t^3 \\|O\\|/(\\varepsilon\\, \\Gamma \\beta_{\\max}^2))$ and numerically that it dominates the other near-term methods at high precision on a 10-qubit spin chain. The same construction extends to memory-retaining collisions with partial swaps between environment qubits, giving non-Markovian simulations with similar costs.","feed_headline":"Open-system dynamics can be simulated with collision models and n+2 qubits","feed_subtitle":"A randomized single-ancilla method avoids block encodings and beats Trotter and qDRIFT at high precision.","key_machinery":"The load-bearing object is the Markovian $K$-collision map $\\Phi_j[\\cdot]=\\mathrm{Tr}_{E_j}[U_j(\\cdot\\otimes \\rho_{E_j})U_j^\\dagger]$ with $U_j=e^{-i\\beta_j H_j \\Delta t}$, composed $K$ times, and its implementation by a single-ancilla LCU called SA-LCU. Each $U_j$ is expanded as an approximate LCU $\\sum_k \\alpha_{jk} W_{jk}$; at every collision two unitaries $X_j$ and $Y_j$ are independently sampled from the distribution $(W_{jk}, \\alpha_{jk}/\\alpha^{(j)})$, applied as controlled and anti-controlled gates on a single ancilla, and the environment register is traced out. The product $\\zeta=\\prod_j \\alpha^{(j)}$ of LCU weights is kept $O(1)$ by choosing the LCU parameter $r=O(\\beta^2 \\Delta t^2 K)$, which prevents the naive exponential growth $\\alpha^K$, and the final measurement of $\\sigma_x\\otimes O$ has expectation $\\mathrm{Tr}[O M_K[\\rho_S]]/\\zeta^2$. This sampling-and-trace machinery is what turns any near-term Hamiltonian simulation subroutine into a collision-model simulator with only one ancilla qubit.","core_discovery":"The central claim is that simulating Lindbladian dynamics does not require specialized block-encoding oracles or many ancillas: a Lindblad generator $e^{\\mathcal{L}t}$ can be approximated by an $(m,\\nu)$-collision map, and that map can be implemented by recycling a single environment register plus one ancilla through $K=m\\nu$ controlled Hamiltonian evolutions, drawing each controlled unitary at random from an LCU decomposition of the collision time-evolution. Provided each collision unitary is simulated to precision $\\varepsilon/(3K\\|O\\|)$, the algorithm outputs an $\\varepsilon$-additive estimate of $\\mathrm{Tr}[O e^{\\mathcal{L}t}[\\rho_S]]$ with probability $1-\\delta$ in $T=O(\\|O\\|^2 \\log(1/\\delta)/\\varepsilon^2)$ coherent runs. For the SA-LCU subroutine the per-run circuit depth is $\\tilde{O}(m^3 t^3 \\|O\\|/(\\varepsilon\\,\\Gamma\\beta_{\\max}^2))$, which the paper shows is better in precision than first-order Trotter and qDRIFT and competitive with second-order Trotter depending on $t$ and $\\varepsilon$; the numerical benchmark on a ten-qubit transverse-field Ising chain under amplitude damping shows SA-LCU needing about 200 times fewer CNOT gates than second-order Trotter and 2000 times fewer than qDRIFT at $\\varepsilon=10^{-4}$. A parallel result for non-Markovian collisions shows that inserting partial-swap channels between consecutive environment qubits preserves the same per-collision error budget and yields a CPTP non-Markovian $K$-collision map with the same asymptotic circuit depth.","pith_inferences":["A direct numerical check of the cited $\\nu=O(t^2 m \\Gamma/\\varepsilon)$ bound on small qubit systems would determine whether the collision-model approach is practical at the quoted depths; this is not done in the paper.","The non-Markovian section proves correctness for finite $K$ but does not give the finite-$K$ distance to a continuous non-Markovian master equation; deriving such a bound would turn the framework into an end-to-end non-Markovian simulator in the same sense as the Lindblad case.","Because the per-collision precision requirement is linear in $K$, extrapolation or error-mitigation techniques could be layered on top of Trotter or qDRIFT to reduce the required $K$, potentially making those subroutines competitive with SA-LCU at high precision."],"forward_implications":["Until error-corrected machines with large ancilla budgets arrive, Lindblad dynamics for $n$-qubit systems can be simulated on early fault-tolerant devices with at most $n+2$ qubits, no block encodings, and no multi-controlled oracle logic.","The same framework can simulate non-Markovian dynamics by adding partial-swap channels between environment qubits, so memory effects are accessible at circuit depths comparable to the Markovian case.","At high precision the single-ancilla LCU method gives shorter circuits than first-order Trotter and qDRIFT; at long evolution times second-order Trotter becomes competitive, so method choice should depend on the target $\\varepsilon$ and $t$.","The $T=O(\\|O\\|^2 \\log(1/\\delta)/\\varepsilon^2)$ sample count is independent of the Hamiltonian simulation subroutine, so the usual Monte Carlo repetition overhead for expectation estimation is not worsened by the collision framework.","Because the per-collision simulation error budget is $\\varepsilon/(3K\\|O\\|)$, any future Hamiltonian simulation algorithm with better depth scaling can be dropped into the framework and immediately improves the end-to-end collision simulation."],"supporting_citations":[{"why":"Supplies the $(m,\\nu)$-collision-map approximation of Lindblad dynamics and the $\\nu=O(t^2 m \\Gamma/\\varepsilon)$ bound used in Theorem 2.","marker":"[55]"},{"why":"Provides the single-ancilla LCU (SA-LCU) Hamiltonian simulation subroutine whose depth and sampling properties the algorithms build on.","marker":"[58]"},{"why":"Defines the qDRIFT randomized Hamiltonian simulation method used as a subroutine option and as a comparison in the complexity tables.","marker":"[4]"},{"why":"Supplies the Trotter error bounds with commutator scaling used for the Trotter circuit-depth entries.","marker":"[53]"},{"why":"One of the sources of the LCU decomposition of time-evolution operators used in Lemma 2.","marker":"[56]"},{"why":"Another source of the qubit-efficient LCU decomposition that SA-LCU relies on.","marker":"[57]"},{"why":"Supplies the partial-swap environment-interaction model that the paper uses for non-Markovian collisions.","marker":"[23]"},{"why":"Establishes that collision models can efficiently reproduce multipartite Markovian dynamics, supporting the Lindblad correspondence.","marker":"[24]"},{"why":"Shows the query-complexity lower bound for simulating Lindblad evolution, which the paper's $t^2/\\varepsilon$ scaling for high-order Trotter matches up to log factors.","marker":"[15]"},{"why":"Presents a direct Hamiltonian-simulation approach to Lindblad dynamics that requires block encodings, providing the contrast motivating the no-block-encoding framework.","marker":"[54]"}],"fun_headline_variants":["Collision models on early fault-tolerant quantum computers without block encodings","Simulate Lindblad dynamics with n+2 qubits and no block encodings","Randomized collision maps beat Trotter and qDRIFT on early fault-tolerant machines","One extra qubit and random collisions simulate open systems efficiently","No specialized oracles: simulate non-Markovian collision models with few qubits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is a bound, taken from another paper, that a collision model with $\\nu=O(t^2 m \\Gamma/\\varepsilon)$ repetitions is within $\\varepsilon$ of the true Lindblad evolution in induced 1-norm; the authors rely on that lemma without proof, and if its true parameter dependence differs, every resource count in the paper changes.","fun_headline_variants_meta":{"raw":{"variants":["Collision models on early fault-tolerant quantum computers without block encodings","Simulate Lindblad dynamics with n+2 qubits and no block encodings","Randomized collision maps beat Trotter and qDRIFT on early fault-tolerant machines","One extra qubit and random collisions simulate open systems efficiently","No specialized oracles: simulate non-Markovian collision models with few qubits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00054,"raw_usage":{"total_tokens":2717,"prompt_tokens":1200,"completion_tokens":1517,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":816,"completion_tokens_details":{"reasoning_tokens":1416}},"tokens_in":816,"tokens_out":1517,"duration_ms":10736,"temperature":1.0,"reasoning_tokens":1416,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T05:00:58.491347+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the simplest nontrivial case—a single qubit with amplitude damping, $\\mathcal{L}[\\rho]=\\gamma(\\sigma_-\\rho\\sigma_+ - \\frac{1}{2}\\{\\sigma_+\\sigma_-,\\rho\\})$—and numerically compute the induced 1-norm distance $\\|e^{\\mathcal{L}t} - M_{m,\\nu}\\|_{1\\to 1}$ for a range of $t$, $m$, and $\\varepsilon$, finding the smallest $\\nu$ needed for $\\varepsilon$-accuracy. If the minimal $\\nu$ does not grow as $t^2 m \\Gamma/\\varepsilon$ (or the constant is not benign), the claimed $\\tilde{O}(m^3 t^3/\\varepsilon)$ circuit depth for SA-LCU must be revised. The same numerical experiment can be run with the paper's full circuits on a classical simulator for $n=2$ or $3$ qubits to verify the $\\varepsilon$-additive output of Algorithm 2.","supporting_citations":[{"cited_title":"Cilluffo, A","cited_arxiv_id":null,"evidence_quote":"Supplies the $(m,\\nu)$-collision-map approximation of Lindblad dynamics and the $\\nu=O(t^2 m \\Gamma/\\varepsilon)$ bound used in Theorem 2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the single-ancilla LCU (SA-LCU) Hamiltonian simulation subroutine whose depth and sampling properties the algorithms build on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Trotter error bounds with commutator scaling used for the Trotter circuit-depth entries."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"One of the sources of the LCU decomposition of time-evolution operators used in Lemma 2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Another source of the qubit-efficient LCU decomposition that SA-LCU relies on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the partial-swap environment-interaction model that the paper uses for non-Markovian collisions."},{"cited_title":"Kandala, K","cited_arxiv_id":null,"evidence_quote":"Establishes that collision models can efficiently reproduce multipartite Markovian dynamics, supporting the Lindblad correspondence."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows the query-complexity lower bound for simulating Lindblad evolution, which the paper's $t^2/\\varepsilon$ scaling for high-order Trotter matches up to log factors."},{"cited_title":"Fischer, Derivation of the quantum-optical master equa- tion based on coarse-graining of time, Journal of Physics Communications2, 091001 (2018)","cited_arxiv_id":null,"evidence_quote":"Presents a direct Hamiltonian-simulation approach to Lindblad dynamics that requires block encodings, providing the contrast motivating the no-block-encoding framework."}],"review_version":1}