{"id":"33487098-dffc-4171-b27b-c06f1fa3afa5","arxiv_id":"2607.16430","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"StsDOpt solves large coupled T&D AC-OPF by smoothing subproblem responses and parallelizing them, cutting memory and sometimes solve time.","lead":"A decomposition solver, StsDOpt, splits combined transmission-and-distribution AC optimal power flow into a master problem plus parallel distribution subproblems, using barrier smoothing to pass derivative information between stages. Tests up to ~660k variables show large memory reductions and modest speedups over a monolithic solver, though the 'no relaxations' claim is internally contradicted by the paper's own penalty smoothing.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Penalty relaxation of boundary power balance (12)-(13) directly contradicts the 'no relaxations' claim; exactness depends on unspecified penalty weights, so reported optimality gaps may be for a different problem.","rationale":"The reader's weakest assumption identified the penalty relaxation of Eqs. (12)-(13) and the unspecified penalty factors as the core gap between the solved problem and the claimed original problem. My review confirms this is the most load-bearing concern: it directly invalidates the paper's central 'no approximations or relaxations' claim, and the numerical validation does not close the gap because no boundary-feasibility metric is reported. The proposed concrete test is minimal and decisive: it checks whether the penalized solution actually satisfies the original constraints for realistic penalty values and whether the objective is stable. The reader's verdict of CONDITIONAL is appropriate; the paper could be revised by reporting penalty parameters, demonstrating boundary mismatch goes to zero, and softening the overclaim. Since I agree with the reader's assessment, the verdict should remain unchanged.","tokens_in":15221,"tokens_out":2444,"duration_ms":21962,"concrete_test":"For a large test case (e.g., Case500-CaseR1R2R3R4R5GC), re-run StsDOpt with the user-visible penalty factor varied over at least three orders of magnitude (e.g., 1e1, 1e3, 1e5). At each setting, record the objective value and the maximum boundary mismatch for (12)-(13), specifically max over boundary links of |sum_phi p_D,phi + p_T| and the reactive analogue, using the final primal solution. Then attempt to solve the same case with the constraints (12)-(13) enforced exactly (no penalty relaxation). If the mismatch does not converge to zero as the penalty grows, or if the objective value changes by more than 0.005%, the reported optimality gap is not for the original coupled T&D AC-OPF and the 'no relaxations' claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central differentiator is that StsDOpt 'does not rely on approximations or relaxations' (Abstract; Section I-B). Yet Section II-B explicitly relaxes the active/reactive boundary balance constraints (12)-(13) and penalizes their difference, asserting only that 'sufficiently large' penalty factors make the flows match. No penalty values, update schedule, or proof of exact recovery is given. The numerical optimality gaps in Table I compare StsDOpt against the integrated monolithic solution of the original problem, but the decomposition actually solves a penalized problem; a 0.000% gap could mean the penalty is large enough, or that both solvers find the same local solution of two different problems. Since the value functions are nonsmooth, the barrier smoothing is an additional approximation, though it converges as mu -> 0. The penalty relaxation, by contrast, is not part of any documented continuation and may leave a nonzero boundary mismatch at the reported solutions. No test case reports the maximum violation of (12)-(13). Thus the claim of exactness is unsupported, and the 'no relaxations' headline is internally contradicted. The implicit-function-theorem smoothness assumption in Eq. (5) is a further unverified condition, but the penalty issue is more fundamental because it changes the problem being solved.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes StsDOpt, a smoothed two-stage decomposition algorithm for large-scale transmission and distribution AC-OPF. The master transmission problem is solved with an SQP method and the distribution subproblems with barrier-based interior-point solvers, using implicit-function-theorem derivatives to pass smooth value-function information to the master. The method is implemented in Julia/PowerModelsITD and evaluated on test cases up to roughly 52k nodes, with multithread and multiprocess variants. The central claims are that the method avoids approximations or relaxations, handles nonlinear nonconvex unbalanced multiphase models, reduces wall-clock time, and lowers memory footprint compared with monolithic integrated solution.","tokens_in":15536,"tokens_out":3802,"duration_ms":36720,"significance":"If the exactness and scalability claims hold, this would be a valuable contribution: it would provide a practical parallel solver for a difficult nonconvex T&D AC-OPF class without convexification, with reduced memory requirements and open-source integration into PowerModelsITD. The numerical study is internally consistent, the package versions are specified, and the use of only second-order derivatives (unlike the prior work in [18]) is a concrete step forward. However, the central 'no approximations or relaxations' claim is contradicted by the penalty relaxation in Section II-B, and the exactness of the decomposition is not established. The paper is therefore promising but needs substantial revision before its main claim can be accepted.","major_comments":[{"comment":"The abstract and Section I-B state that StsDOpt 'does not rely on approximations or relaxations.' Yet Section II-B explicitly relaxes the boundary active/reactive power balances (12)-(13) and penalizes their difference, asserting only that 'sufficiently large' penalty factors make the flows match. No penalty values, update schedule, or proof of exact recovery is given, and no test case reports the maximum violation of (12)-(13). Because Table I computes the optimality gap against the integrated solution of the original problem, a 0.000% gap could mean the penalized problem has a nearby optimum. The exactness claim is therefore unsupported as written. Please provide penalty values and boundary-mismatch statistics, or a proof that the penalized problem recovers the original solution, or revise the no-relaxation claim.","section":"Section II-B, Eqs. (12)-(13); Table I"},{"comment":"The generic subproblem in Eq. (7) includes a hard boundary equality x^T_{i,beta_D} - y^D_{i,beta_T} = 0, but the T&D implementation appears to replace the power-flow part of this coupling with the relaxed constraints (12)-(13) while retaining voltage/angle equalities (14)-(17). The relation between Eq. (7) and Eqs. (12)-(17) is not stated. This ambiguity is load-bearing because the exactness of the decomposition depends on which constraints are enforced in the master/subproblem and which are penalized. Please clarify the precise formulation that is actually solved.","section":"Section II-B, Eq. (7) vs. Eqs. (12)-(15)"},{"comment":"The derivative computation via the implicit function theorem requires the Jacobian of the barrier KKT system, ∇_v F_i, to be nonsingular along the solution path. For barrier problems this requires constraint qualifications and strict complementarity-type conditions that are not verified for the reported T&D instances. Active-set changes can make the value function nonsmooth even for finite μ. Because the master SQP relies on these exact second derivatives, the paper should state the conditions under which Eq. (5) is valid and what the solver does when those conditions fail (e.g., a fallback to quasi-Newton or derivative-free steps).","section":"Section II-A, Eq. (5)"}],"minor_comments":[{"comment":"Grammar: 'Traditional OPF optimize consider transmission systems' and similar phrases should be edited for clarity.","section":"Abstract and Section I"},{"comment":"Case500-CaseR1R2R3R4R5GC is described as using the IEEE 118-bus system, but the case name and the variable counts suggest a 500-bus transmission system. Please correct the description.","section":"Section IV-B, test case 10"},{"comment":"'PowerModelITD' appears to be a typo for 'PowerModelsITD'.","section":"Section I-B, contribution 6"},{"comment":"The notation uses hat{f}_i for both the value function in Eq. (7) and the distribution cost function in Eq. (9); please use distinct symbols to avoid confusion.","section":"Eq. (9)"},{"comment":"The 'Iterations' column appears to report different quantities for the integrated solver versus the decomposition (master/subproblem iterations); please clarify what is counted.","section":"Table I"}],"recommendation":"major_revision","confidential_remarks":"The paper's core algorithmic idea is plausible and the numerical study is internally consistent, but the 'no relaxations' claim is directly contradicted by the manuscript's own penalty relaxation of Eqs. (12)-(13). The authors should be asked to either prove exact recovery, report the penalty parameters and boundary mismatches, or substantially temper the claim. This is fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline is that StsDOpt is a credible, well-engineered solver for multi-phase unbalanced T&D AC-OPF with real memory and some speed benefits on large instances, but the paper overstates its own cleanliness: the 'no approximations or relaxations' claim is contradicted by the penalty relaxation of (12)-(13), and the missing penalty parameters and lack of boundary-mismatch reporting leave the exactness question open.\n\nWhat's genuinely new is the parallel (multithread/multiprocess) implementation of the smoothed two-stage decomposition from [18] in the PowerModelsITD framework, with only second-order derivatives needed. The numerical results are internally consistent and show a clear trend: for small cases monolithic wins, but for cases with tens of thousands of nodes and many feeders, the decomposition reduces memory from ~50 GiB to ~7 GiB and cuts wall-clock time by roughly half. That is a practical gain for HPC deployments.\n\nThe soft spots are in the framing and reproducibility. The abstract and Section I-B say the method 'does not rely on approximations or relaxations,' yet Section II-B explicitly relaxes the boundary power balance equations and penalizes their difference. The barrier smoothing is a standard continuation, so that part is defensible, but the penalty relaxation is not documented: no penalty values, no update schedule, no reported maximum violation of (12)-(13). Without that, the 0.000% optimality gap is uninformative — it could be comparing two different problems that happen to have the same objective. Also, the accuracy benchmark is only against the authors' own monolithic solver, and no code or data are provided. The implicit function theorem condition is a minor issue; it's the standard assumption and likely holds.\n\nThis is not a fatal flaw. The algorithm is sensible and the engineering is solid. But the writing needs to be honest about the penalty relaxation, and the experiments need to report boundary mismatch and penalty weights. If those are fixed, the paper is a useful contribution for practitioners in T&D co-optimization. I'd send it to peer review — the referees can hold the authors to that standard — but I would not take the 'no relaxations' claim at face value.","headline":"Solid parallel decomposition solver for T&D AC-OPF, but the 'no relaxations' claim is contradicted by the penalty relaxation and the missing penalty parameters undercut the exactness story.","tokens_in":16023,"tokens_out":2555,"would_cite":true,"duration_ms":22577,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["90C30","90C26","90C55","90C31"],"pacs":[],"model":"deepseek-v4-flash","headline":"A two-stage decomposition solver for AC optimal power flow matches the monolithic optimum while cutting memory and wall-clock time on large grids.","keywords":["AC optimal power flow","transmission-distribution coordination","two-stage decomposition","barrier smoothing","sequential quadratic programming","interior-point methods","distributed energy resources","nonconvex optimization"],"falsifier":"Run the largest reported multiprocess test case, record the final active and reactive power mismatch at every transmission–distribution boundary bus, and check whether the values are zero to within numerical tolerance. If any mismatch is nonzero, the solved point is not feasible for the original coupled AC-OPF problem, contradicting the exactness claim.","tokens_in":15100,"feed_emoji":"⚡","tokens_out":3052,"duration_ms":30859,"temperature":0.7,"pith_summary":"This paper claims that the coupled transmission–distribution AC-OPF problem can be decomposed into a transmission-level master problem and many distribution-level subproblems without sacrificing exactness. The key move is to replace each subproblem's nonsmooth value function with its log-barrier counterpart, making the subproblem response differentiable in the shared boundary variables. The authors then use an implicit-function-theorem sensitivity calculation to feed exact first and second derivatives into a sequential quadratic programming master solver. In numerical experiments up to roughly 52,000 buses, the decomposed solver produces objective values within 0.005% of the monolithic solution while solving faster on large cases and using a small fraction of the master-node memory.","feed_headline":"Two-stage AC-OPF solver matches monolithic optimum within 0.005%","feed_subtitle":"Smoothing feeder responses lets a master problem get exact gradients; large grids solve faster and use far less memory.","key_machinery":"The load-bearing mechanism is the barrier-value-function smoothing: each inequality-constrained distribution subproblem is reformulated with a log-barrier term and a copy of the shared boundary variables. The dual variables on the copy constraints provide the exact gradient of the smoothed value function, and the implicit function theorem applied to the subproblem's KKT system yields exact second derivatives. A penalty relaxation of the active and reactive boundary power-balance constraints keeps the subproblems feasible; the penalty factors are assumed 'sufficiently large' to enforce power matching at the optimum.","core_discovery":"The central claim is that the smoothed two-stage decomposition, denoted StsDOpt, solves the nonlinear, nonconvex, multi-phase unbalanced T&D AC-OPF problem without reliance on approximations or relaxations. The paper establishes that substituting the barrier formulation for each distribution subproblem makes the subproblem's optimal value a smooth function of the transmission boundary variables, so that exact gradients and Hessians can be computed from the KKT system of the subproblem. The master SQP solver then iterates with these derivatives. Experimental results show that this approach, parallelized across feeders, yields objective values nearly identical to the monolithic integrated solv","pith_inferences":["The reported optimality gaps are computed against a monolithic solve of the same relaxed boundary model; if the penalty relaxation is not exact, the comparison may be against the wrong baseline, so the true gap to the original coupled feasible problem could be larger.","The implicit-function-theorem derivative calculation assumes nonsingularity and strict complementarity of the subproblem KKT system; at points where the active set changes or the matrix becomes singular, the master solver may stall or need a fallback derivative approximation.","The scalability study uses repeated identical copies of the same distribution feeder, which may understate the difficulty of real heterogeneous feeders; a test with widely varying feeder sizes and characteristics would stress the master's ability to coordinate very different subproblem curvatures.","Since penalty parameters are not reported, a practical user would need to tune them per instance; the method's effectiveness in deployment may hinge on a systematic penalty-selection rule that the paper does not provide."],"forward_implications":["The exact decomposition eliminates the need for an AC-feasibility-check post-processing step that is required by methods based on convex relaxations.","On the largest tested case (about 52,500 nodes), the multiprocess version reduces wall-clock solve time from roughly 225 to 117 seconds and master-node memory from about 51 GiB to about 7 GiB.","The smoothing approach naturally handles multiphase, unbalanced distribution feeders instead of single-phase balanced approximations.","Because subproblems are independent, the method can be distributed across compute nodes, enabling solution of problems that do not fit in a single shared-memory machine."],"fun_headline_variants":["Smooth two-stage AC-OPF matches monolithic to 0.005%","Exact gradients for decomposed T&D AC-OPF via smoothing","Parallel T&D AC-OPF solver avoids relaxations, cuts time","StsDOpt: exact feeder gradients for large-scale AC-OPF","Decomposed AC-OPF with smoothing matches full solution"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The boundary power-balance equations are relaxed and enforced only through a penalty whose magnitude is never stated, and the paper asserts without proof that 'sufficiently large' penalty factors make the boundary flows match exactly at the optimum.","fun_headline_variants_meta":{"raw":{"variants":["Smooth two-stage AC-OPF matches monolithic to 0.005%","Exact gradients for decomposed T&D AC-OPF via smoothing","Parallel T&D AC-OPF solver avoids relaxations, cuts time","StsDOpt: exact feeder gradients for large-scale AC-OPF","Decomposed AC-OPF with smoothing matches full solution"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000548,"raw_usage":{"total_tokens":2476,"prompt_tokens":786,"completion_tokens":1690,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":530,"completion_tokens_details":{"reasoning_tokens":1594}},"tokens_in":530,"tokens_out":1690,"duration_ms":57931,"temperature":1.0,"reasoning_tokens":1594,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T20:57:12.372689+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the largest reported multiprocess test case, record the final active and reactive power mismatch at every transmission–distribution boundary bus, and check whether the values are zero to within numerical tolerance. If any mismatch is nonzero, the solved point is not feasible for the original coupled AC-OPF problem, contradicting the exactness claim.","supporting_citations":[],"review_version":1}