Equilibrium formulation#

This document specifies required phase, reaction and reactive-phase mathematics. Equation definitions own the displayed relations; algorithms own their numerical solution. A proposed or registered equation is not a capability claim. Declared-phase specifications and the explicit native tangent-plane test/fixed-T/P discovery path have retained numerical qualification for the stated neutral and admitted ionic/reacting cases. These retained cases have stated limits; PH/PS, TV and UV use the bounded neutral formulations below. Mechanical/local stability checks and truthful global-claim limits apply to both declared and discovered results.

Energy, reference states and coordinates#

At fixed temperature and pressure, fluid equilibrium minimizes the sum of phase Helmholtz energies plus pressure-volume work subject to conserved amounts and phase electroneutrality. Sources: Gross–Sadowski2001 Appendix A, Pereira2012 Eqs. 2–8, Ascani2022 Eq. 5, and the retained Tsanas2019/2022 studies under research/. Use EqID: phase_helmholtz_amount_volume and EqID: system_gibbs_helmholtz_coordinates. All phase values and derivatives come from the same defining energy.

A phase carries amount (mol), normalized composition, and volume (m³), or an equivalent packing-fraction coordinate (EqID: phase_packing_volume). The declared neutral ideal-gas base uses the log inverse unit-amount volume coordinate from EqID: system_gibbs_helmholtz_coordinates; its occupied fraction is zero. It admits one vapor phase at fixed positive T/P. Caloric conditions, solved-state caloric actions, source-reference transfer and phase searches are deferred and refused before calculation. Evaluate intensive properties at unit amount before scaling extensive contributions. An incipient phase has zero amount but a defined intensive composition and volume per mole; do not evaluate a logarithm of its zero extensive amount. The positive-amount logarithmic coordinate transformation is not defined at zero.

Eliminating volume by a global admissible minimum of Helmholtz energy plus pressure-volume work gives the Gibbs formulation. A mechanically stable stationary density root identifies only a local branch; it does not establish the global minimum. Joint volume/composition coordinates avoid an inner root choice but still require stability assessment of the final candidate.

The ideal mixing/density term supplies nonreactive equilibrium at fixed T. Reactive or caloric work additionally requires declared standard-state offsets and their temperature dependence. Use EqID: standard_pressure_density_conversion, EqID: ideal_gas_helmholtz, EqID: total_enthalpy, and EqID: total_entropy. Reaction correlations constrain a reaction subspace. A chosen gauge may define a reference convention but cannot supply missing absolute caloric evidence.

Ideal liquid chemistry#

IdealMoleFraction declares species identifiers and charges without EOS parameters. At fixed positive T/P, one free liquid uses EqID: ideal_mole_fraction_reaction and EqID: ideal_reaction_amount_coordinates. Scalar constants are dimensionless mole-fraction constants; laws identify ideal-mole-fraction explicitly. EOS-basis laws, reaction/source references, supplied species Gibbs energies and affinity targets are refused. The positive logarithm reference in ReactionLogPolynomial may lie outside its evaluation interval; evaluation outside the interval remains refused.

The default activity_basis="mole_fraction" retains the equations above. With activity_basis="molarity", EqID: ideal_molarity_reaction requires a finite positive solution_volume_m3 and an absolute Amounts feed. Scalar K is the dimensionless concentration constant on the 1 mol/L standard; correlations declare ideal-molarity-1-mol-per-litre. No basis conversion is inferred. unit_activity_solvent optionally names one neutral species on the positive liquid support. Its activity is one; its amount and stoichiometry remain in the balances. Volume and unit-solvent declarations are refused on the default basis.

A neutral nonnegative mol or mole-fraction feed defines the default conserved inventories. Every reacting species requires a positive attainable inventory; an inventory face forcing one absent is refused. The native inventory bounds limit log amounts, including amounts greater than the feed reference where attainable. A supplied positive normalized composition guess is a native cold seed. Previous-result starts are refused before compilation or reaction-law evaluation.

One native row expression supplies values and exact derivatives to the existing feasibility/Ipopt lifecycle and post-solve polish. The returned phase amount is the current species total in mol, with normalized true species fractions. The objective is zero; it is not physical Gibbs energy. Density and packing arrays are empty. Endpoint Jacobian rank is checked. EOS-only solved-state actions, stability searches and caloric conditions are unavailable for this model.

The explicit reaction_activities(model, result) call reads dimensionless species activities and their logarithms at each returned phase composition and density, including a finite candidate that misses its requested criterion. It performs no calculation or storage during an unrequested solve, continuation or fit. The model must match the result, which must retain its original declaration and physical arrays. Records follow phase/reaction order; species arrays use component order. A zero phase amount does not remove its defined composition.

EqID: reaction_activity_quotient governs the products-positive log_quotient. Unreferenced Helmholtz reactions, including ideal gas and species-derived targets, return density-standard activities. Ideal mole-fraction reactions return exactly the mole fractions. Molarity reactions instead use returned phase amounts times fractions and the shared fixed-volume expression, with unit solvent activity where declared. Concentration pH is an analysis quantity under a declared individual-ion convention, not an Engine result. Source references retain their declared pressure and pure/dilute species standards; common molality subtracts the solvent-mass/standard-molality factor for every solute. basis, reference_pressure_pa and the species solvent/molality fields identify these conventions. Ionic values use the EOS single-ion convention; only charge-neutral combinations have convention-independent meaning. Values come from the state and reference owner, independently of targets and residuals.

species.causes names unavailable component quantities; cause names an unavailable quotient. A finite log value survives activity exponentiation underflow or overflow, with the activity absent. A structural zero has activity zero and log activity negative infinity when its reference exists. Zero stoichiometric coefficients are skipped, and conflicting infinite terms make the quotient unavailable. On an EOS model each row also carries the pressure-basis quotient \(\ln Q_P=\ln Q_\rho+\Delta\nu\ln(RT/P^\circ)\), with \(P^\circ\) the reaction’s reference_pressure_pa or else the thermochemistry record’s standard pressure (standard_pressure_pa; with neither, pressure_cause names the absence). At equilibrium it equals \(\ln K_P\) in every phase. log_pressure_quotient_actions lists its partial derivatives at the phase state along \(T\), \(P\) and \(\mathbf e_j-\mathbf x\) (amount \(j\) at fixed \(T\), \(P\)); an action that cannot be evaluated is empty and pressure_cause names the refusal. solid_activities returns 1 for each present pure solid and no value for an absent one. pH, a persistent cache and a fitting payload are not supplied.

For the MDEA example, EqID: mdea_reaction_conserved_amounts supplies independent inventory checks. These are numerical verification of conserved hydronium chemistry, not physical validation of ideal ionic activities, reproduction of Figure 15 or agreement with the paper’s nonconserved printed water balance. Hot-condition extrapolation remains deferred; only declared evaluation intervals are admitted.

Multiplier units and signs#

Chemical potentials are dimensional (J/mol). Balance multipliers lambda are conjugate to the declared conserved-moiety row basis; lambda dot B_i has units J/mol. Charge numbers z_i are dimensionless, so the charge multiplier varphi_p in chemical stationarity has units J/mol. With stationarity written as mu_i minus lambda dot B_i minus z_i varphi_p = 0, a physical electric potential psi_p in volts corresponds to varphi_p = -F psi_p, where F is the Faraday constant. Only potential differences are determined without a gauge.

The dual of reduced molar energy uses dimensionless multipliers alpha, with potential differences divided by RT. Optimizer callback multipliers Lambda_opt multiply the implemented constraint functions (including their declared scales); they are distinct from chemical multipliers embedded in the primal coordinates. The weighted-Hessian falsifier must vary both independently.

Constraints and specification#

For one neutral component at fixed positive temperature, free pressure and one liquid/one vapor phase with one free amount and one amount pinned at zero, EqID one_component_two_density eliminates composition normalization, feed balance, pressure and the common chemical multiplier exactly. The driver’s ordinary attempt solves the two log densities; this changes neither the full defining rows nor their observable derivatives. The returned pressure is the vapor pressure, avoiding subtraction of large liquid terms at low pressure. The lift sets each composition to one, the free amount to the feed total, and \(\ln\eta_p=\ln[C(T,\theta)\rho_p]\) in declaration order.

The shared full-row, topology, local stability and density-branch admission checks still govern publication. A row-converged equal-density pair is not coexistence. Rejection retains the original-start Ipopt fallback and attempted work. Full derivative preparation remains the sole source of solved-state actions and parameter-fitting columns; the limited isothermal projection is not a temperature or parameter derivative. The scoped method and numerical error estimates are defined in ALG-EQUILIBRIUM and ADR 0009.

The balance matrix spans conserved moieties; its nullspace is the admitted reaction span. Reactions must conserve charge. Independent balance rows replace reaction extents as primal coordinates; extents may be reconstructed afterward. At each phase use EqID: chemical_potential_stationarity, EqID: pressure_equality, EqID: phase_normalization, EqID: phase_support, EqID: phase_electroneutrality and EqID: material_balance.

One charge row can be omitted only for an ion-containing phase of nonzero amount when neutral feed, charge-conserving reactions and balances imply it. An explicit charged Vapor declaration that would select Vapor as the omitted-charge phase is refused with charged vapor cannot be the omitted-charge phase. Internal charged Vapor trials instead carry their directly constructed pure-charge balance. Their numerical log-charge representation is confined to Newton correction and classification (EqID tangent_plane_trial_balance); physical constraints and Ipopt derivatives are unchanged.

A zero-amount phase does not carry that implication. All returned charged phases must satisfy electroneutrality. Support removes coordinates; explicitly empty support must not silently mean all species. A support unable to form a neutral ionic phase is inadmissible.

A neutral feed is one whose mole fractions x_feed = n_feed / N meet the state rule of full-eos.md §12, |Σ z_i x_i| ≤ 10⁻¹² (absolute); any other feed is refused as charged. The row-free phase p₀ carries the feed charge Q amplified by N/n₀, which is unbounded, so an admitted feed is solved with its ions of the excess sign scaled by min(C, A)/max(C, A), where C = Σ_{z_i>0} z_i n_i and A = Σ_{z_i<0} |z_i| n_i. A feed with C = A, including one with no ions, is left unchanged. The scaling preserves signs and zero amounts, changes the charge by exactly −Q and the amounts by Σ|Δn_i| ≤ |Q| ≤ 10⁻¹² N (every ion has |z_i| ≥ 1); each balance row against the declared feed moves by at most max|B|·10⁻¹². Ions of a single sign at that charge are scaled to zero. The charge left in a returned phase is the solver’s own: the charge component of the balance-row residual in p₀ and each charge row’s residual elsewhere. A charge row bounds its phase’s charge by the row criterion. In p₀ the balance residual is normalized by N, so its charge is that criterion amplified by N/n₀, like the feed charge. Neither is bounded by the state rule; the largest measured is 3.2×10⁻¹⁴, 31 times under the state rule.

For k pinned phase amounts, f free intensives and e independent extensive specification rows, rows minus unknowns = k + e − f; a square system requires f = k + e. Ordinary specifications retain Π − k ≤ rank B. PH/PS or TV adds one independent balance, permitting Π − k ≤ rank B + 1. Generic-point rank does not prove existence, uniqueness or regularity at the solution. Reject dependent reactions, charged feed, impossible supports, inconsistent pins and invalid domains.

PH/PS declares exactly one extensive H_spec [J] or S_spec [J/K], free T and fixed P and free phase amounts. EqIDs: extensive_enthalpy_specification, extensive_entropy_specification. A neutral declaration with complete component records uses them directly, with interval admission at every consuming state. Ordinary density-basis reactions retain Δh = t − Tt′ and Δcp = −Tt″: constant affinity t = a has Δh = a, while constant K has Δh = 0. A source-referenced declaration, or one whose supports include record-less species, uses the record-anchored Gibbs completion (reference_species_calorics) at every free-T point: it requires consistent records, determined neutral combinations and admitted record and source-law intervals. Charged declarations follow either path.

TV declares positive total volume_m3 [m³], fixed T, free P, free phase amounts and neutral nonreacting declared supports (extensive_volume_specification). The driver solves Helmholtz stationarity as a feasibility system and admits a minimum on the total-volume-conserving tangent through the shared physical Hessian. A homogeneous declaration returns its local branch; it does not discover other phases. TV needs no ideal thermal records. Automatic TV/UV phase discovery remains unavailable. UV additionally declares finite internal_energy_j [J], both T and P free, one or two free fluid phases and complete ideal records strictly inside their intervals (extensive_internal_energy_specification). It reuses the volume and caloric geometry and qualifies the entropy Lagrangian on the inventory/normalization/U/V tangent, retaining T and excluding P and chemical duals. The reported physical potential is -S_tot/R; missing records and exact interval endpoints are refused.

The resolved feed n_feed is the amount basis of the whole problem, and returned phase amounts are on that same basis. An amounts feed is absolute mol and is never normalized. A mole-fraction feed resolves to n_i = x_i, mol per mol of feed; a mass-fraction feed resolves to n_i = w_i / M_i, mol per kg of feed. A fraction feed must sum to one within the composition-sum policy value; it is rejected otherwise and never renormalized. A molality feed is absolute mol: solute entries are mol per kg of solvent times the stated solvent mass, and solvent entries are relative molar amounts rescaled so the solvent mass equals the stated value. One declared solvent needs no amount. Several declared solvents with no positive solvent amount are rejected, because their ratio is a physical input; a declared solvent left at zero beside a supplied one stays zero.

A pinned amount is feed-relative, N_p = v_p Σ n_feed with 0 ≤ v_p ≤ 1. When the declared reaction space conserves total moles, rank([B; e]) = rank(B) with e ∝ (1, …, 1), the pins additionally satisfy Σ v_p ≤ 1, and Σ v_p = 1 when every phase is pinned; a nonconserving reaction space admits no universal total-mole sum rule. Reaction participation is sign independent. A participant is removed only when a globally nonnegative conserved functional has an exactly zero feed inventory and a positive coefficient for that species. Exact native dot products verify the functional against every original reaction and feed component; an LP objective rounded to zero alone does not establish absence. Positive trace inventories retain their species. An all-pinned nonreactive declaration compiles square by count but its retained balance and normalization rows are dependent, Σ_i balance_i = Σ_p v_p normalization_p, so it reports generic rank loss and no nonlinear solve is run.

On a certified face, restrict state coordinates to the surviving species A and select independent echelon row operations L from B[:, A]. Keep the full-column map L B for balances and feed controls; only state rank and total-mole conservation use its restriction to A. Preserve discarded inventory restrictions for action admission. The feasible reaction space is ker(B[:, A]), embedded with zero amounts outside A, rather than reactions with stoichiometric columns deleted. Full reaction declarations and species reference/caloric completion stay fixed before restriction to A. Removed species return literal zero amounts and fixed-face actions. Feed directions must preserve vanished conserved inventories; individual removed feed slots need not independently vanish. Supplied phase composition guesses follow the same support restriction before the existing supported-composition normalisation.

For a declared liquid/vapor boundary, an incipient phase means the limit of an emerging positive phase on the same exact inventory face. For each nonnegative certificate w, conservation gives 0 = Σ_p N_p wᵀx_p. Every positive phase therefore has wᵀx_p = 0, and the closed composition face contains its incipient limit. A zero amount alone does not imply this restriction for an arbitrary intensive phase. Restrict both finite and incipient supports before differentiation; the continuous energy extension x ln x → 0 is differentiated only along surviving coordinates. No chemical row for a removed species survives. A discarded balance row vanishes on A, so its potential contributes nothing to surviving stationarity. The original reaction/reference completion and ker(B[:, A]) therefore define the same constrained energy and stationary variations on this face.

Automatic incipient restriction is admitted only with exactly two phases of opposite liquid/vapor kinds, fixed positive T, free P, exactly one Pinned(0) amount and one free amount, and a nonempty effective incipient support whose every species is uncharged. Outside this domain, an intersection with certified absence retains zero_inventory_incipient_support, naming the species, phase and excluded condition. Other empty supports remain refused. The reduced square Jacobian must be regular at the solved endpoint; generic rank alone does not qualify it. Singular or coalesced endpoints and charged incipient extensions are unqualified. This face equivalence neither proves uniqueness nor establishes global stability. Incipient supports without certified absence retain their existing admission, and finite additions may connect separately solved faces.

The conserved-inventory bound call admits no additional charge, energy or chemical conditions. On the union of compiled supports it maximizes species amount under \(Bn=b\), \(n\ge0\), with structural zero outside that union. Amount results are in mol; their separate feed-normalized value is \(u_i/N_{\mathrm{feed}}\). When reactions change total moles this value differs from a mole-fraction maximum. For an explicit supported fraction floor \(f\ge0\), \(S f\le1\), fraction bounds use \(y=n/N\), \(\tau=N_{\mathrm{feed}}/N\): \(By=(b/N_{\mathrm{feed}})\tau\), \(\sum y_i=1\), \(y_i\ge f\) on support, \(y_i=0\) outside, \(\tau\ge0\). These are aggregate inventory fractions. They bound an individual liquid fraction only for a sole finite liquid with a zero-amount vapor, as in the retained Born cases. They establish neither thermodynamic equilibrium nor stability, and no derivative of the potentially nonsmooth maximum is supplied.

When no target is supplied, participating species thermochemistry supplies the density-standard reaction Gibbs series through order three, preserving missing records and excluded-interval causes. This source admits ordinary fixed-T/P problems without solids, caloric conditions or source-reference transfer. Attaching records alone does not select absolute species standards: with neither reactions nor a supplied vector the compiler uses zero species standards.

Ordinary supplied reaction targets are a constant affinity, a scalar K, or the P3 correlation ln K(T) on the EOS density basis, activity (ρ x_i / 1 mol m⁻³) exp(μ_i^res / RT). A correlation on another basis requires the explicit neutral reference declaration below. An ordinary scalar K is a temperature-independent target: its series is −RT ln K, −R ln K, 0, 0, so it states zero density-basis reaction enthalpy and differs from a correlated target by Δ_r h / T at first temperature order. A temperature-dependent source constant is therefore stated as a correlation on the EOS basis, not as its value at one temperature; the engine does not detect that misuse. The two native conversions, an ideal-gas pressure basis (a − Δν ln(R T_ref ρ₀ / P₀), c − Δν) and a standard-state shift with temperature-independent Δh and Δs (a − Δs / R, b + Δh / R, old minus new), have an exact polynomial image of a + b/T + c ln(T/T_ref) + dT. A condensed-phase reference (molality or infinite dilution, solution molarity, a pure-liquid mole-fraction reference) transfers through the residual potential of the reference state and has none. A supplied constant standard Gibbs vector must satisfy every reaction target at the evaluation temperature by the same criterion as the inferred vector, and a free temperature admits it only when the targets are constant. Supplied species thermochemistry is compared with each target through Gibbs, enthalpy and heat capacity on the density basis; a mismatch is a reference inconsistency, not a preference for either input.

The implemented neutral reference uses one pure neutral solvent and a positive charge-neutral all-solute dilution path. An explicit source-basis enum selects common-molality or raw mole-fraction correlations; a nonempty source ID records provenance. EqID: neutral_reaction_reference_transfer converts each source at its declared fixed or system reference pressure, then the existing reaction/SVD owner compiles the density-basis target. References refresh at each current nonlinear temperature, and system-pressure references also follow each current pressure. A fixed source pressure retains thermal motion and has zero derivatives with respect to system pressure. The liquid selector chooses the dense branch when several stable roots exist; a unique stable root follows the density owner’s unspecified-label semantics. Reference qualification must therefore establish the physical solvent state for the application; the selector supplies no universal phase classification. This qualified finite terminal reference supports fixed-T/P central solves and feed actions. Its exact first and second pressure derivatives also support square pinned free-pressure cells and pressure/feed actions through order two (EqID: neutral_reaction_reference_pressure_jet). Its first temperature derivative at the reference pressure supports order-one temperature and mixed T/P actions (EqID: standard_state_transfer_temperature_derivative); its second and mixed temperature-pressure derivatives extend temperature actions, alone or mixed with feed or pressure, to order two (EqIDs: standard_state_transfer_temperature_second_derivative, standard_state_transfer_temperature_pressure_derivative). Feed-only actions at fixed T/P retain order three. Declared-branch continuation along fixed T, fixed P and feed, or a straight physical EOS/Born parameter path, uses the same first-order chains for its tangent. All other declarations and source conventions stay fixed on the parameter path. Third-order actions involving pressure or temperature remain unavailable. Raw referenced lever requests retain their existing refusal. Referenced one-free-intensive pinned cells use current-point T/P row jets and A1/A2 with at most one parameter-bearing direction, including T/P–parameter terms. No external direction may drive its solved intensive; simultaneous free T/P still lacks the additional constraint for a regular square chart. Record-anchored Gibbs h/s/u, frozen c_p/c_v and Gibbs-conjugate volume use the shared caloric composition and its consumed reference columns. The retained prescribed-T/P evidence and subsequent free-intensive F6 qualification have separate source identities and domains. Ordinary unreferenced reactions retain their existing consistency checks and derivative behavior.

Bubble, dew, pure saturation, fixed-fraction and ordinary flash can be constructed as data specifications of these rows. Names may help users construct a problem; they must not select application-specific physics. A single declared phase and one-component coexistence are ordinary cells of the same specification. Admission and numerical qualification must cover phase/support, reaction, charge, pin and free-intensive combinations explicitly; structural acceptance of a cell is not evidence that it solves.

Pure stoichiometric solids#

A pure solid s (ADR 0013) is not a fluid phase: it has no composition, density root, pressure, normalization or charge row. It declares a charge-neutral dissolution stoichiometry ν_s over model components, exactly one sourced ln K_s or P3 law with its reaction reference, the reference’s source standard pressure P° (required) and its molar volume V_s. EqIDs: solid_standard_gibbs, solid_saturation_index, solid_amount_complementarity, solid_material_balance. A free solid adds one amount coordinate n_s ≥ 0 with no upper bound and no stationarity row in the optimizer; it is counted as a complementarity pair, and one endpoint classification adds either its saturation row or n_s = 0, so the classified declaration is square. A Pinned(0) solid adds no coordinate and reports SI_s only. Solids require fixed T and P and a neutral reference; PH/PS, critical, envelope, azeotrope, pinned free-intensive cells, discovery, stability trials, continuation, and caloric and stability fit observations refuse them. The ASL equilibrium record (v8) carries declared-present free solids with a ln K(T) law beside one liquid and returns their amounts; a solid’s onset (n_s = 0 at the solved state) is PhaseBoundaryNotDifferentiable, and the system enthalpy is refused while a solid is present, since a solid has no caloric record. The ASL property record supplies each declared solid’s ℓ_s = ln K_s^ρ(T, P°) + V_s(P − P°)/RT − Δν ln(P/(RT ρ°)), ρ° = 1 mol/m³, the right side of Σ_i ν_{s,i}(ln x_i + ln φ_i) = ℓ_s for a present solid, with exact (T, P, θ) derivatives. Composition, amount-ratio, partial-pressure and phase-density fit observations admit them (ALG-FIT). Solid solutions and kinetics are not expressible.

Stationarity, feasibility and stability#

For free phase amounts at fixed T/P, the rows are first-order conditions of the constrained Gibbs minimum. Phase potential equality and zero reaction affinity follow from stationarity: EqID: interphase_potential_equality and EqID: reaction_affinity. The positive-definite reduced Hessian (EqID: reduced_hessian_stability) is a sufficient strict local condition on a regular constraint manifold, not a necessary condition for every local minimum or a global certificate.

Pinned amounts or an independent caloric row, with free intensives, define a feasibility specification using intensive chemical-potential and pressure rows. They are not all the KKT rows of the original fixed-T/P minimization. Their algorithm must satisfy the actual constraint Jacobian and, when advertised, weighted Hessian requirements in ALG-EQUILIBRIUM.

For free phase amounts the classifying stationarity residuals are the first-order conditions in the solved logarithmic chart, N_p x_i μ̃_i / N_feed for composition and N_p / N_feed times the pressure row for packing, which are the quantities the constrained minimizer controls; the unscaled chemical and pressure rows are retained as diagnostics and are certified only to the criterion times N_feed / (N_p x_i), which is unbounded where an amount or a fraction vanishes. The amount-stationarity component (phase Gibbs energy equal to the balance dual) is not a classified row; it is bounded by the optimizer’s dual tolerance and checked by the retained Gibbs-dual identities. A feasibility specification classifies its unscaled rows, which are its constraints. Report achieved balance, charge, normalization, pressure, stationarity/affinity and phase-distinctness diagnostics on declared scales. A stationary candidate is not automatically a stable equilibrium. A requested distinct phase set that coalesces must report the degeneration explicitly, including a pair whose identities swap between consecutive states of a followed branch (passage through a critical point) other than an envelope’s requested ln K straddle.

Discovery and global claims#

The certified inventory-face argument for incipient phases also applies to trial phases: nonnegativity and the zero conserved inventory force each certified-absent species to have zero fraction in every phase of positive amount, so generated trial supports exclude those species.

EqID: tangent_plane_distance defines the reduced molar distance from the candidate’s conserved supporting plane. Sources: Michelsen1982 Part I; Pereira2012 Eq. 12; Perdomo2025 §3.3 Eqs. 37–40; Ascani2022 §3.3 Eqs. 33–35. The reactive extension follows the homogeneous energy and conserved-moiety identities, EqID: tangent_plane_gibbs_dual_identity. Temperature is in K, pressure in Pa, amounts in mol and molar volume in m³/mol; reduced energies and chemical multipliers are in RT units. The plane uses the result’s chemical balance multipliers, recovered for all-free cells or primal for pinned cells, not optimizer constraint multipliers. Freeze actual standard terms, including reference transfer and gauge, at the solved T/P snapshot.

Only explicit Discover() requests run discovery. Only explicit stability_test(...) calls test an accepted declared result. Declared solves, continuation, parameter fitting and ASL never invoke either implicitly. Discover once at initialization, then declare the returned phases for subsequent calculations. Explicit testing of pinned/free-intensive results releases pins only for diagnosing Gibbs descent at their solved fixed T/P; it says nothing about feasibility along the original pinned path. Zero-amount phases cannot supply a donor. Discovery with free T/P remains excluded.

Full discovery remains the default. Explicit Discover(mode="fast") is a nonreacting, at-most-two-phase finite search on neutral or electrolyte fluids. An admitted split establishes local admission and resolved finite descent from the selected parent, without testing that endpoint for another phase. A completed fast schedule can report only no admitted split in the fast trial set. Unavailable trials, unresolved finite descent, exhausted budgets or a refused flash give an incomplete result with diagnostic candidates and no usable public state arrays. The returned omissions include unexecuted starts, post-flash stability and three-or-more-phase topologies; salt-rich or remote minima may remain undiscovered. Neither outcome certifies global stability. ALG-STABILITY owns the schedule.

The native test compiles a one-free-phase unit-amount trial through the shared exact-curvature driver, without reactions or feed-moiety proportions. Its internal reduced balance space spans normalization and charge, retaining the row transformation (EqID: tangent_plane_trial_balance). The stationary trial identity uses the multiplier contraction with its transformed right-hand side; species ordering need not leave a total-amount row first. The frozen standard shift has constant coefficients but retains exact log-chart curvature (EqID: tangent_plane_shift_log_curvature). A normalized reacting trial is a search composition, not necessarily reaction equilibrium: EqID: normalized_trial_reaction_affinity. The full added declaration restores every original reaction, reference input and affinity row.

An explicit stability_test(...) with StabilityOptions.evaluate_supplied_trials = true instead evaluates each supplied composition_guess, without minimizing composition or creating a finite split. Entries, including repetitions and unavailable points, retain caller order. It uses the same unit-amount trial, frozen standards, chemical plane and EqID: tangent_plane_distance. Its point is the supplied full-species vector after native reconstruction, log/exp chart preparation and final reconstruction; supported entries must be positive and structural zeros remain zero. The returned composition is the actual energy-evaluation point. The total input displacement and final reconstruction displacement are separate.

A supplied packing coordinate is replayed through the chart and retains its pressure defect; without one, the existing Liquid/Vapor density root selects packing at the initially reconstructed composition. Reconstruction preserves that packing when recomputing molar volume, without a second root solve. The returned packing, density and pressure row describe the final point. Packing outside the domain or nonpositive supported composition yields an unavailable row with its existing cause, never an invented TPD. Charged-family admission below applies even when packing is supplied. An explicit charged Vapor support requires DensityTemperature and must match its own lowest stable pressure root after final reconstruction, within the existing root window and residual criterion, with positive finite pressure slope. A sole root on the Liquid side is unavailable; a sole root with no loop evidence retains the density owner’s Unspecified meaning. This admission supplies no charged-Vapor discovery or physical vapor-validation claim. Donor slope and estimated uncertainty remain available when a positive full-support donor exists, but optimizer-only values are absent. Exhausting the evaluation sequence, a zero or negative TPD, or a donor estimate establishes no stability conclusion. ALG-STABILITY owns these operations.

Raw optimizer points are not physically feasible by construction. Native normalization and ionic group rescaling followed by a two-pivot arithmetic correction admit a nearby trial (EqID: electroneutral_trial_reconstruction). Recompute its molar volume from composition and packing and reevaluate shared energy. Retain raw residuals, displacement and reconstructed point separately. Repair remaining candidate balance/charge defects in supported amount variables, then check positivity, support and domain and reevaluate; keep this corrected candidate separate from the public result. No failed initialization supplies a trial point. Arithmetic reconstruction does not assert exact-real feasibility.

A negative plane value requires a strictly positive donor with the entire trial support and finite withdrawal capacity. The actual fixed-molar-volume donor slope includes chemical and pressure corrections (EqID: tangent_plane_donor_descent); present phases obey the full chemical-plus-pressure identity (EqID: present_phase_tangent_plane_identity). Association derivative floors, Euler defect, feasibility correction, supplied standard-term uncertainty and arithmetic contribute separately to an estimated slope uncertainty (EqID: tangent_plane_descent_uncertainty). A resolved negative slope and plane value also require a retained finite inventory-conserving split whose computed Gibbs decrease exceeds its propagated estimate (EqID: conserved_finite_phase_split). Optimizer convergence or positive local trial inertia is not necessary for this descent evidence.

The adopted claim is numerically resolved descent with estimated uncertainty. A proved instability witness additionally requires justified feasibility and total error bounds; current association floors are not such enclosures. Missing error terms, near-zero margins and failed reconstruction remain indeterminate. A negative trial without an eligible donor remains an unresolved plane trial. Completed finite testing reports only no witness in the recorded trial set, with unresolved trials named; it never certifies global stability. Restricted supports and density-label starts limit that set, and positive local curvature does not exclude a remote split. ALG-STABILITY owns enumeration, conserved addition/removal and separate work budgets.

Charged trial supports are admitted under every declared bulk-permittivity form. The existing DensityTemperature (Schreckenberg) and DipolarPerturbation (Rueben) forms reach the vacuum limit, \(\epsilon_r\to1\) as \(\rho_m\to0\), by their definitions in full EOS §6.2 and §6.4, EqID: density_correlated_bulk_permittivity and EqID: dipolar_perturbation_bulk_permittivity; their charged trials count at every density. IonSuppression (including solvent-only, ion-fraction and ion-specific choices) and ComponentMixing are density independent. Their electrolyte terms are defined for condensed phases, and a free-volume trial search cannot infer that domain from a Liquid label or mechanical pressure-root admission alone. Under these forms a charged trial point counts only when it is a stable density root of its own composition at the test T and P that the density owner’s side classification places on the liquid side of that composition’s own mechanical loop, the classification declared Liquid phases already satisfy.

ALG-STABILITY owns the seed and end-point checks. A vapor-side root, a sole root without a loop sample and a point away from every stable root are unavailable with the cause “charged trial outside the condensed domain of density-independent permittivity”; they are never witnesses or stability evidence. Neutral trials, vacuum-limit forms, declared solves, continuation, fitting and ASL retain their existing domain and calculations. No density cutoff, new tolerance, optimizer-success filter, parameter substitution or model change is introduced. The condition is necessary for this search, not physical validation or a global stability certificate, and it establishes condensed density only, not the correlation’s fitted range. The previously evaluated solvent-only gas-density descent at 2.32 mol/m³ lies on a sole root without a loop and stays refused.

The exact global dual infimum (EqID: helmholtz_lagrangian_dual, EqID: dual_inner_problem) can supply a lower bound. A local inner minimizer is an upper bound on that infimum; substituting it as a certified lower bound is invalid. A finite outer approximation alone does not close the global gap. EqID: outer_approximation_bound and the historical tunneling/candidate relations describe source methods, not a selected current discovery algorithm. The adopted finite native controller does not implement those global-bound methods. Justified global bounds, replacement at the balance-rank phase limit and generalized multi-donor/reactive withdrawal remain separately scoped work. Budget exhaustion never means proven stable. The qualification notebook (private research repository) retains neutral comparisons, admitted density-dependent ionic/reacting checks, density-independent charged-search refusals and unchanged declared work. These are numerical verification; physical validation and a global certificate do not follow from this doctrine.

Reuse of discovered declarations is numerically qualified for the retained enumerated cases. It is not a guarantee for every amount scale or initial point. A 100 mol neutral discovery reproduces the one-mol intensive split and extensive Gibbs scaling. Cold declared solving and supplied-result public reuse at that scale also keep the split.

Sensitivity and validity#

EqID: implicit_sensitivity applies on a smooth fixed branch and support with a square full-rank row Jacobian. Solve its linear system rather than forming an inverse. EqID: implicit_second_action defines second directional actions, the third-order relation and the diagonal coefficient recurrence that realises them; admission at an accepted result uses the classified residual of the solved chart. All direct and implicit terms of the requested observable must be included. Energy derivative order is determined by the observable, as tabulated in the algorithms specification.

At rank loss, phase coalescence, support changes, branch switches or a nonsmooth ionic boundary, report the unavailable action and its reason while preserving valid central values. Conditioning and residuals qualify each linear solve. Do not return an unexplained empty array as a derivative.

Neutral nonreactive critical equations (EqID: critical_equations) are necessary conditions: a solution is a critical-equation candidate with T, P, density and composition, not a stability certificate, and its pressure-closed quantities that divide by P_rho = 0 are unavailable. Envelope cells (EqID: envelope_local_condition) trace a feed with a zero-pinned incipient phase (ALG-EQUILIBRIUM). Solid phases and kinetics remain separately justified extensions; this fluid formulation does not imply their support.