{"id":"26a06708-49b3-47c6-8147-12ff37910d22","arxiv_id":"2502.09402","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A generalized convex envelope method computes multiphase solid-liquid-vapor equilibria for multicomponent mixtures without a priori phase information.","lead":"This paper extends an existing computational method, the convex envelope method, so it can compute phase equilibria involving solid, liquid, and vapor phases for mixtures with up to four components. The method builds a geometric hull of the mixture's Gibbs energy surface to find stable phase splits without knowing the number or type of phases in advance.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Abstract overclaims 'any phase equilibrium': Eq. 5 makes solid Δg affine in composition, so the demonstrated CEM can only stabilize pure-component solids, excluding solid solutions and compounds that Section 4 defers to future work.","rationale":"The reader's weakest assumption identifies exactly the load-bearing point: Eq. 5 restricts solid phases to pure components, so the abstract's 'any phase equilibrium' and 'arbitrary aggregate states' are not supported by the demonstrated model. I agree with that reading. The paper does have independent support: it provides an open-source implementation, the convex-envelope construction is deterministic, and the quantitative comparisons for VLE and SLE up to four components show low mean deviations against experimental data. Those merits justify a conditional rather than a reject verdict. The concern does not change the reader's verdict, because the reader already assigned CONDITIONAL based on the same limitation and on the missing convergence study. The conditional acceptance should require either a rewording of the abstract to state the pure-solid assumption explicitly or a demonstration with a non-linear solid Gibbs-energy model, as proposed in the concrete test.","tokens_in":15020,"tokens_out":14186,"duration_ms":149515,"concrete_test":"Use the published CEM code to compute a binary SLE for a system with known solid solubility, e.g., an ideal solid solution defined by Δg_mix,solid = RT Σ x_i ln x_i plus the Eq. 5 pure-component terms, at a T and p inside the solid-solution single-phase region. If the solid branch of the implementation is hardwired to Eq. 5, it will fail to return an interior solid phase, falsifying the implementation-level claim. If the code accepts a user-defined g_solid and returns the solid-solution phase, rerun the same test with a congruently melting compound as an additional fixed-composition solid point; the question is whether the abstract's 'any phase equilibrium' is a property of the supplied model, not of the CEM. The test's outcome distinguishes a wording/validation issue from an algorithmic gap.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the extended CEM calculates 'any phase equilibrium ... over the whole composition space' for arbitrary aggregate states. For this to hold, the Gibbs model supplied for every conceivable phase must cover the cases covered by the claim. In the actual construction, Eq. 5 defines Δg_mix,solid as an affine function of x_i (pure-liquid reference plus linear interpolation of melting terms). An affine Gibbs surface has zero curvature, so it cannot have an interior minimum; the convex envelope can therefore place stable solid phases only at the pure-component vertices. That excludes solid solutions, congruently melting compounds, hydrates/solvates, and other solid complexes. The paper itself acknowledges this in Section 4, where 'only pure components are allowed to form solid phases' and more advanced solid behavior is listed as future work. The sentence after Eq. 5 says the method 'also works for more complex thermodynamics ... as long as g=g(T,p,x) can be calculated', which rescues the algorithm as a framework, but not the abstract's claim as stated, because all quantitative results and the published implementation are based on Eq. 5. Thus the load-bearing generality assertion rests on an unvalidated and, in the demonstrated form, false universal statement about solid phases.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper extends the convex envelope method (CEM) of Göttl et al. (2023) to T,p flash calculations that can, in principle, consider liquid, vapor, and solid aggregate states simultaneously. The method discretizes the composition simplex, evaluates the molar Gibbs energy of mixing for each phase model at every grid point, constructs the convex envelope of the union of these graphs, classifies heterogeneous simplices, and then computes phase splits for arbitrary feed compositions by a linear problem. Numerical demonstrations cover binary, ternary, and quaternary VLE and SLE systems using literature NRTL parameters, a differential-evolution-based NRTL parameter-fitting procedure, and a proof-of-concept combination with the HANNA machine-learning activity-coefficient model. The authors claim that the method computes any phase equilibrium of a mixture with arbitrary number of components and phases over the whole composition space without prior knowledge of the phase split.","tokens_in":15255,"tokens_out":5559,"duration_ms":52295,"significance":"If the central claim were fully supported, the paper would offer a robust and initialization-free alternative to conventional flash algorithms, particularly attractive for process synthesis because the expensive convex-hull construction is decoupled from the fast per-feed phase-split query. The paper has concrete strengths: the algorithm avoids nonlinear solves and is deterministic; the implementation is open source; the method has no fitted parameters of its own beyond the discretization parameter δ; and the HANNA example demonstrates a pathway from molecular structure to complete phase diagrams. The reported errors for the studied systems are small, and the method reproduces literature SLE/VLE topology, including multiphase regions. However, the significance is moderated by the fact that the main claimed generality rests on a solid-phase model that only stabilizes pure-component solids, and by validation that mostly compares literature-fitted models to the data used to fit them.","major_comments":[{"comment":"The abstract's claim that 'any phase equilibrium ... can be calculated over the whole composition space' is not supported by the demonstrated implementation. Equation (5) defines Δg_mix,solid as an affine function of composition, which has zero curvature and therefore cannot produce stable solid phases at interior compositions. Consequently, solid solutions, congruently melting compounds, and solid solvates are excluded, as the authors acknowledge in Section 4 ('only pure components are allowed to form solid phases'). The sentence after Eq. (5) stating that the method works for more complex thermodynamics as long as g(T,p,x) is computable rescues the framework but not the paper's central claim, because all quantitative results use Eq. (5). Please revise the abstract, introduction, and conclusion to distinguish the algorithmic framework (which can accept arbitrary phase models) from the implemented and validated special case (pure-component solids), or add demonstrations with a nonlinear solid-phase model.","section":"Abstract; Section 4"},{"comment":"The 'whole composition space' claim is also qualified by the finite discretization. Tables 1, 4, 6, and 7 report missing feeds (MF) near azeotropes and multiphase boundaries, and Section 4 states that interpolation errors cannot be fully eliminated. The paper should either characterize these missing feeds quantitatively (for example, by reporting the distance to the phase boundary or by a discretization-refinement study) or soften the claim to 'approximate phase equilibria computed on a discretized grid.' As written, the abstract and Section 5 overstate the completeness of the calculation.","section":"Section 2.2; Section 4; Tables 1-7"},{"comment":"The quantitative validation conflates the accuracy of the CEM with the accuracy of the underlying NRTL models. The MD values in Tables 1–7 compare CEM outputs, obtained with literature-fitted NRTL parameters, to the same experimental data that were used to fit those parameters; they therefore mainly measure the model–data mismatch, not the error introduced by the convex-envelope construction. To support the claim of high accuracy of the method itself, the paper should compare CEM results against a reference flash algorithm (for example, Michelsen stability analysis plus phase-split calculation) using identical thermodynamic parameters, or explicitly restrict the claimed accuracy to agreement with the underlying models and the literature data.","section":"Section 3.2; Eq. (10)"}],"minor_comments":[{"comment":"The phrase 'One the other hand' should read 'On the other hand.'","section":"Section 1"},{"comment":"The caption contains 'methyl M TBE'; this should be 'MTBE.'","section":"Figure 2 caption"},{"comment":"The sentence describing Figure 3 c) and d) contains the garbled phrase 'at 303.15 K in is given in c)'; it needs grammatical correction.","section":"Section 3.1"},{"comment":"The phrase 'deducing a high accuracy' should read 'indicating a high accuracy,' and 'mising feeds' should be 'missing feeds.'","section":"Section 3.2"},{"comment":"In the methanol–toluene row, the reported value A12 = 3.348 × 10^37 appears inconsistent with the other fitted values and likely represents an unphysical optimum or a typographical error; please verify this entry.","section":"Table 8"},{"comment":"The symbol T is used both for temperature and for the set of data points in Eq. (9); please introduce distinct notation to avoid ambiguity.","section":"Eq. (9)"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the journal's scope, and the algorithmic framework is a sound extension of prior CEM work. My principal concern is that the abstract and conclusions promise a generality that the implemented solid-phase model and the validation do not deliver; this is fixable in revision by recalibrating the claims and adding a reference-algorithm comparison. I would not recommend rejection because the limitation is explicitly acknowledged in Section 4 and the open-source implementation is a useful contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quirin,\n\nThis paper does something real: it takes the convex envelope method for liquid-liquid equilibria and generalizes it to handle vapor and solid phases in one framework. The CEM discretizes the composition simplex, computes Gibbs energy of mixing for each phase model, builds a convex hull, and reads off phase splits from the hull facets. Their Python implementation is open source, and they validate on binary, ternary, and quaternary VLE and SLE data with mean deviations below 0.02. That part is solid.\n\nWhat's actually new is the unification. Prior convex hull methods handled binary/ternary VLE or SLE separately; this one handles all three aggregate states together, automatically determines number and type of phases, and scales to four components. The parameter fitting section and the HANNA integration are nice extras—makes the tool practical for solvent screening and process synthesis.\n\nNow the soft spots. The abstract says 'any phase equilibrium ... can be calculated over the whole composition space.' That's an overclaim. The solid model in Eq. 5 is affine in composition, so the convex envelope can only produce pure-component solid phases. No solid solutions, no compounds, no hydrates. The paper itself acknowledges this in Section 4, so it's a fixable wording issue, but as stated it's false. Second, they don't show a convergence study with respect to the discretization parameter δ. The quantitative validation is against experimental data that was used to fit the underlying NRTL parameters, so the reported MD mixes model error and method error. They mention this in the text, but a proper residual breakdown would be more convincing. Third, missing feeds near azeotropes are reported, which is honest but shows the method isn't failsafe.\n\nNone of these are fatal. The core idea is sound, the code is available, and for the tested systems the results are accurate. The main fix is to tone down the abstract and discuss the solid model's scope more prominently. A convergence study and a clearer separation of model error from discretization error would make it much stronger.\n\nBottom line: this deserves a serious referee. Send it to review, ask for revisions, and it could be a useful reference for process simulation. I'd bring it to a reading group if anyone here works on phase equilibrium methods.","headline":"A useful, sound extension of the convex envelope method for VLE/SLE up to quaternary mixtures, but the abstract's 'any phase equilibrium' overclaims given the pure-component-only solid model.","tokens_in":15765,"tokens_out":4014,"would_cite":true,"duration_ms":34886,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper extends the convex envelope method for T,p flash calculations to mixtures where vapor and solid phases can coexist, so the number of phases and their aggregate states are determined automatically.","keywords":["convex envelope method","T,p flash calculation","phase equilibrium","tangent plane criterion","vapor-liquid equilibrium","solid-liquid equilibrium","Gibbs energy of mixing","NRTL parameter fitting"],"falsifier":"Run the method on a binary system with a well-documented solid solution (e.g., KCl–KBr) at a temperature where the solid solution exists; the linear solid model produces no solid phase of intermediate composition, so the calculated solid-liquid equilibrium will miss the solid-solution region entirely, whereas the experimental diagram shows it.","tokens_in":14799,"feed_emoji":"⚗️","tokens_out":5172,"duration_ms":41996,"temperature":0.7,"pith_summary":"This paper extends the convex envelope method (CEM) for T,p flash calculations from liquid-only systems to mixtures in which vapor and solid phases can also coexist. The central claim is that by computing the Gibbs energy of mixing for every conceivable aggregate state over a discretized composition space and taking the convex envelope of the combined graph, the phase equilibrium can be obtained for any feed without specifying in advance how many phases form or what they are. The method thus turns the flash calculation into a two-step process: a one-time, possibly expensive construction of a piecewise-linear phase diagram, followed by essentially immediate phase-split answers for any feed composition. The authors validate the approach against literature data for vapor-liquid and solid-liquid equilibria with up to four components, report mean deviations below 0.02, and demonstrate the same machinery for fitting NRTL parameters. If the claim holds, phase-equilibrium calculations for process simulation no longer need a starting guess for the number or type of phases.","feed_headline":"One convex envelope maps vapor, liquid, and solid equilibria","feed_subtitle":"T,p flashes no longer need a guess for how many phases form; phase splits emerge from the Gibbs-energy hull.","key_machinery":"The central object is the convex envelope of the combined graph G = {(a, Δg_mix,s(x)) | x in discretization D, s in {solid, liquid, vapor}} ⊆ $R^{{n+1}}$, where a is the cartesian coordinate of composition x and Δg_mix,s is the Gibbs energy of mixing of aggregate state s. Phase splits are identified by 'heterogeneous simplices' of this envelope: simplices containing a line segment that either connects non-neighboring grid points (the classic LLE criterion) or connects points of different aggregate states (the new generalization). The envelope is constructed with a standard convex-hull algorithm, and once stored, any feed composition is answered by locating the simplex containing it and solving a tiny linear problem. The solid-phase model, given as a linear interpolation of pure-component melting terms, keeps solid phases pinned to pure-component vertices.","core_discovery":"The discovery is a generalization of the tangent-plane criterion: instead of separating phases one aggregate state at a time, the method builds one convex envelope over the Gibbs energy of mixing graphs of liquid, vapor, and solid phases simultaneously. A phase split is recognized whenever the convex envelope connects two points that are either non-neighboring in the composition grid or belong to different aggregate states. For a solid phase the paper assumes the Gibbs energy of mixing is linear in composition, so solids occur only at pure-component vertices; with that model in hand, the convex envelope of the combined graph yields any number of coexisting phases, their aggregate states, and their compositions for any feed, with no prior knowledge of the phase split. This extends the mathematical framework that was proven for liquid phases in prior work to arbitrary aggregate states.","pith_inferences":["My inference: if the linear-solid assumption is lifted, the convex-envelope framework should extend to solid solutions and stoichiometric compounds, because the machinery only needs a computable g(T,p,x) for every point in composition space.","My inference: the decoupling of expensive hull construction from fast flash evaluation suggests the method could serve as a precomputed thermodynamic library for real-time flowsheet optimization, including in reinforcement-learning process design.","My inference: the reported missing feeds near azeotropes are a discretization artifact; adaptive refinement of the composition grid near multiphase boundaries could presumably eliminate them without the cost of global fine grids."],"forward_implications":["T,p flash calculations can be performed for any feed without a prior guess of the number or type of phases, removing a common failure mode in process simulation.","The same framework covers VLE, LLE, SLE, VLLE, and combinations thereof, so unified phase-diagram construction replaces separate case-by-case solvers.","The method can fit gE-model parameters directly against phase-equilibrium data (e.g., NRTL), because the flash result is a continuous function of the model parameters.","Combined with predictive property models, the CEM can construct complete T,x,y diagrams from molecular structure alone, enabling solvent and process design without experimental parameters."],"supporting_citations":[{"why":"Supplies the general mathematical framework for the CEM with an arbitrary number of components, which this work extends to multiple aggregate states.","marker":"(Göttl et al., 2023)"},{"why":"Establishes the tangent plane criterion that underlies the CEM's detection of phase splits.","marker":"(Baker et al., 1982)"},{"why":"Earlier CEM applied to liquid phase equilibria; provides the two-step discretization-and-envelope concept adopted here.","marker":"(Ryll et al., 2012)"},{"why":"Gives the solid-phase Gibbs energy expression that the paper's linear solid model is based on.","marker":"(Reyes et al., 2001)"},{"why":"Provides the experimental VLE data for methanol-benzene and MTBE systems used in the quantitative validation.","marker":"(Oh et al., 2003)"},{"why":"Provides the experimental SLE data for ternary chloronitrobenzene systems used in the quantitative validation.","marker":"(Li et al., 2016)"},{"why":"Provides the VLE data for methyl acetate–isopropyl acetate and the quaternary mixture used in validation.","marker":"(Xiao et al., 2013)"},{"why":"The machine-learning HANNA model that predicts activity coefficients from SMILES strings, combined with the CEM to construct a full phase diagram.","marker":"(Specht et al., 2024)"}],"fun_headline_variants":["No phase-count guess needed: convex envelope finds all equilibria","One convex hull reveals vapor, liquid, and solid splits","Generalized convex envelope: any phases, no prior count","Convex envelope now covers vapor and solid phases too","Phase splits from Gibbs-energy hull without phase counting"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The solid-phase Gibbs energy of mixing is assumed to be linear in composition, so solid phases can appear only as pure components; if a system forms solid solutions or compounds, the convex envelope built from this model will not capture them.","fun_headline_variants_meta":{"raw":{"variants":["No phase-count guess needed: convex envelope finds all equilibria","One convex hull reveals vapor, liquid, and solid splits","Generalized convex envelope: any phases, no prior count","Convex envelope now covers vapor and solid phases too","Phase splits from Gibbs-energy hull without phase counting"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000868,"raw_usage":{"total_tokens":3721,"prompt_tokens":868,"completion_tokens":2853,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":484,"completion_tokens_details":{"reasoning_tokens":2774}},"tokens_in":484,"tokens_out":2853,"duration_ms":19020,"temperature":1.0,"reasoning_tokens":2774,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T21:36:39.801328+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the method on a binary system with a well-documented solid solution (e.g., KCl–KBr) at a temperature where the solid solution exists; the linear solid model produces no solid phase of intermediate composition, so the calculated solid-liquid equilibrium will miss the solid-solution region entirely, whereas the experimental diagram shows it.","supporting_citations":[{"cited_title":", author Pierce, A","cited_arxiv_id":null,"evidence_quote":"Establishes the tangent plane criterion that underlies the CEM's detection of phase splits."},{"cited_title":", author Blagov, S","cited_arxiv_id":null,"evidence_quote":"Earlier CEM applied to liquid phase equilibria; provides the two-step discretization-and-envelope concept adopted here."},{"cited_title":", author Conesa, J","cited_arxiv_id":null,"evidence_quote":"Gives the solid-phase Gibbs energy expression that the paper's linear solid model is based on."},{"cited_title":", author Wang, Q.l","cited_arxiv_id":null,"evidence_quote":"Provides the VLE data for methyl acetate–isopropyl acetate and the quaternary mixture used in validation."},{"cited_title":", author Nagda, M","cited_arxiv_id":null,"evidence_quote":"The machine-learning HANNA model that predicts activity coefficients from SMILES strings, combined with the CEM to construct a full phase diagram."}],"review_version":1}