Stability and phase discovery#
A converged declared split satisfies the equations for the phases you allowed. It does not establish that another phase cannot lower the Gibbs energy. Request a stability test or fixed-T/P phase discovery explicitly. The water/isobutanol notebook compares a declared split with discovery and inspects finite trials. See the stability algorithm and equilibrium formulation for trial definitions.
Test a declared result#
stability_test evaluates a solved snapshot at its solved temperature and
pressure with the actual standards frozen. Trial supports, starting points
and work limits determine which possibilities are examined. The driver
performs no implicit trial searches during ordinary declared solves,
continuation, parameter fitting or ASL evaluation.
Tangent-plane distance measures reduced molar energy above the candidate’s conserved supporting plane. Inspect trial feasibility, density availability, charge balance and energy descent together. An optimizer message alone does not establish a feasible lower-energy phase. The native records distinguish unresolved trials and numerical uncertainty from an admitted descending split. The equilibrium reference and native stability records expose these calls and evidence fields.
Discover at fixed temperature and pressure#
A Problem with phases=Discover() asks the native search to consider topology
changes at fixed T/P. Supply a feed and explicit search budgets. Optional trial
declarations restrict the starts and supports examined. The search uses the
shared equilibrium driver for admitted subproblems and retains its work and
stopping causes.
Discover(mode="fast") runs a smaller search: it stops at the first resolved
split and solves one two-phase declaration, without a final stability search.
Its record lists the start families and topologies it left unsearched. Use the
default full mode when a fast result must be checked for a further split.
See water/isobutanol discovery for the bounded numerical comparison.
Once a topology is found, it may seed later declared calculations within a qualified domain. Reuse preserves the phase supports and reference choices; a previous discovery does not remove the need to check a new state’s residuals, branch identity and requested stability evidence.
Charged trial admission#
Charged trials must be electroneutral and satisfy the permittivity model’s domain. Density-dependent forms with the vacuum limit and density-independent condensed-phase forms have different trial admission rules. Under the latter, a charged trial counts only at a stable root on the liquid side of its own composition’s mechanical loop. An off-root, vapor-side or loop-less trial can remain unavailable with an explicit cause.
See electrostatic domains and the stability algorithm. The neutral notebook does not qualify every charged or reacting search. A refused trial is not evidence of the physical absence of that phase.
What a finite search establishes#
A resolved feasible Gibbs descent is evidence against the candidate’s stability, within the reported numerical uncertainty. Finding no such descent in a finite set of local trials is weaker: retain the number of trials, unresolved attempts, tested supports and exhausted budgets.
Local stability searches are not global proofs. Positive mechanical slopes and local curvature also do not exclude a distant lower-energy split. A critical-equation candidate satisfies necessary equations; it is not by itself a stability certificate. Use the wording supported by the actual search and read the limits before promoting a phase prediction.