Rareytec Property Platform — Help

How the platform estimates, regresses and exports thermophysical properties for CHEMCAD. Every value is traceable to the method or source that produced it.

1. Overview

For each component the platform assembles the constants and temperature-dependent correlations CHEMCAD needs, choosing for every slot between: your imported data, the company data bank, the shared core bank, a physical estimate, or a manual entry. You pick per slot; the choice is stored per project. The result is exported as a CHEMCAD neutral file (.nf) or written directly into a CHEMCAD simulation (.ccsim).

2. Projects & components

A project holds a set of components and (optionally) mixtures. Add components by search, manual identifiers, structure drawing, a CHEMCAD neutral/flat file, an IK-CAPE/PPDS file, a ThermoML archive, or by attaching a CHEMCAD .ccsim (its full property set is harvested in). Every added component is run through identification against the core registry, so the same substance is recognised across sources by InChIKey.

3. The CHEMCAD parameter block

The parameter block lists every scalar constant (MW, Tb, Tc, Pc, Vc, ω, and the secondary parameters ZRA, UNIQUAC r/q, dipole moment, …). Each row shows the current value and where it comes from. Priority when nothing is chosen: your project choice → core DB → estimate → component identity. Click a row to see all candidate values (imported / company / core / estimate / internet) and pick or type one; that choice then applies to this project only. The provenance label always names the method that produced the number — e.g. ZRA reads “Yamada-Gunn from estimated omega”, not a blanket method name.

4. Estimation methods

  • Tb, Tc, Pc, Vc — the Rarey-Nannoolal group-contribution methods, directly from structure.
  • T-dependent curves — vapour pressure and liquid viscosity by Rarey-Nannoolal, liquid thermal conductivity by Rarey-Govender, surface tension by Rarey-Olivier — each limited to its measured range of validity (section 4.1).
  • Acentric factor ω — derived: ω = −1 − log10(Ps(0.7·Tc)/Pc) on the Rarey-Nannoolal vapour-pressure curve.
  • ZRA (Rackett) — Yamada-Gunn: ZRA = 0.29056 − 0.08775·ω.
  • ΔHvap(Tb) — Clausius-Clapeyron on the Rarey-Nannoolal Ps(T) curve (skipped for vapour-associating species — carboxylic acids, HF).
  • UNIQUAC r/q — UNIFAC group contribution, as CHEMCAD itself does when the DECHEMA values are absent.
  • Mixture activity coefficients (predicted) — the UNIFAC family (original, Modified UNIFAC (Dortmund), NIST-modified, UNIFAC 2.0, PSRK) and COSMO-SAC (2010) from stored σ-profiles — fully implemented and fast (Newton segment solver, <1 ms per evaluation); available in every predicted-model list on the binary and ternary pages.

Derived properties inherit whatever you chose for their inputs (e.g. a chosen Tc feeds the ω and ZRA estimates).

4.1 Range of validity

A group-contribution method returns a number for any structure you give it, whether or not the number means anything. The platform therefore limits each method to the range in which it was shown to work: outside it a value is either flagged with a warning or withheld altogether. The limits below are not conventions — they were measured against the NIST/TRC ThermoML archive, using each component's experimental normal boiling point so that the method's own error is not confused with the error of an estimated Tb.

Temperature

The temperature-dependent methods are simple functions of the reduced temperature and cannot follow a liquid property into the critical region — or, for viscosity, far below its reference temperature Tv. The figure gives the measured mean deviation against T/Tc; the shaded band is where all three methods are inside their accepted window, and the dashed lines are the points beyond which a value is refused.

all three inside their window 5 10 20 50 100 200 500 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 VIS TCN SFT VIS refused TCN refused SFT refused reduced temperature T / Tc mean deviation [%]
Mean relative deviation from experiment against reduced temperature (logarithmic scale). Tc from the Rarey-Nannoolal critical method. 8794 thermal-conductivity, 3009 surface-tension and 18547 viscosity points.
MethodValue returnedWarningWithheld Mean deviation inside / outside
Thermal conductivity T/Tc ≤ 0.90> 0.75> 0.90 5–6 % below 0.6 → 35–56 % above 0.85
Surface tension T/Tc ≤ 0.93> 0.85> 0.93 5–6 % below 0.75 → 169 % above 0.95
Liquid viscosity 0.40 ≤ T/Tc ≤ 0.95 < 0.50 or > 0.90< 0.40 or > 0.95 15–19 % (median) in range → 491 % below 0.40

Viscosity fails in the opposite direction to the other two: it is extrapolated downwards, below Tv, where the exponential form runs away — the classic symptom is a glycerol-like value of 106 mPa·s at room temperature. Its deviations are larger throughout because viscosity spans orders of magnitude, so a relative measure is harsher on it than on the others. Tc is itself estimated unless you supply one, which is why the inner limit warns rather than withholds.

Molecule size

Group contribution assumes a molecule is the sum of its parts. That assumption is weakest at both ends of the size range: in a very small molecule almost every atom is a neighbour of every other, so the groups interact rather than add; in a very large one the accumulated group count amplifies any single bad contribution, and such molecules tend to be the exotic, poorly represented ones.

the reliable band 0 5 10 15 20 5 10 15 20 25 Tb TCN SFT heavy atoms in the molecule (n) mean deviation [%]
Mean relative deviation against the number of heavy (non-hydrogen) atoms — the same n that appears inside these equations. Marker area is proportional to the number of compounds in the bin; bins with fewer than 5 compounds are omitted. Boiling point: 8777 compounds.
Heavy atoms nTb mean dev. CompoundsAssessment
2–310.4 / 5.0 %57 groups overlap; treat with caution
44.5 %152 usable, still above the plateau
5–16 2.9–4.8 % 7373 the reliable band
17–227.4–10.7 %711 degrading
≥ 238.9–18.2 %191 large / exotic; mean driven by outliers — the median stays near 3 %

The same shape appears in thermal conductivity (22.6 % at n = 3, 4–6 % from n = 5) and surface tension (16.7 % → 4–6 %). For the smallest molecules an experimental value is almost always available and should be preferred over any estimate.

Structure

Independently of temperature and size, a method is applicable only if every group in the molecule carries a fitted contribution. Where a structure needs a group the method never regressed, no value is produced — the platform says so rather than dropping the group silently, which would return a plausible but meaningless number. Fragmentation that cannot cover the whole molecule is reported the same way.

5. Temperature-dependent correlations & regression

Each T-dependent property (vapour pressure, density, ΔHvap, heat capacities, viscosities, thermal conductivities, surface tension) is carried as a correlation in a DIPPR / Antoine form. You can adopt an imported/bank correlation, enter coefficients by hand, or regress the curve to the component's experimental points.

Regression modes
  • Direct — least-squares fit of the chosen form to the points, in ln (relative) space so low-magnitude points are not drowned out.
  • Anchored — for sparse data (or a single Tb-style value): take the temperature dependence of the Rarey-Nannoolal Ps(T) method and shift it by a constant to pass through the data, then express it in the chosen form.
Degrees of freedom

Never more than n − 1 constants are fit for n data points; the remaining constants are held at zero. In particular, an Antoine fit to exactly 3 points fixes C = −Tb/8 — the denominator of the Rarey-Nannoolal Ps(T) method, log10P = (ΔB0+ΔB)·(Trb−1)/(Trb−⅛) — and fits only A and B.

Fit objective & experimental error

The final fit minimises the deviation weighted by the probable experimental error of each point, σ, rather than a plain least-squares:

σ = max( ThermoML reported uncertainty , σfloor + Y·0.1% + |dY/dT|·0.01 K )

σfloor is a typical absolute measurement error for the property (e.g. 1 mbar for vapour pressure); the 0.1 % relative and 0.01 K temperature-uncertainty terms (the latter propagated through the local slope dY/dT) apply to all properties. An author's reported uncertainty is used only where it is larger than this model — it can down-weight a point flagged as poor, but a small (often over-optimistic) stated uncertainty is never allowed to make one point dominate the fit.

Exponential properties

Vapour pressure and liquid viscosity vary exponentially with temperature (ln Y ≈ linear in 1/T). Their plots offer a log Y vs 1/T view, on which a good correlation is a straight line — the quickest way to spot a bad point or an over-extrapolated curve. (ΔHvap, density, vapour viscosity, heat capacities, vapour thermal conductivity and surface tension are not exponential.)

The “CHEMCAD (simulator)” curve

When a project is linked to a CHEMCAD simulation in COM (live) mode, every T-dependent property page offers a CHEMCAD curve (simulator) overlay. It computes the property inside CHEMCAD over COM and draws what the simulator actually calculates with its selected models. This is deliberately separate from the stored correlation: with a special enthalpy / K-value / property model (for example an equation of state such as SRK), the vapour pressure, ΔHvap, heat capacity, etc. no longer come from the DIPPR / Antoine coefficients stored in the file — so the page lets you compare three things at once: the correlation stored in the .ccsim (our evaluation of it), the simulator's live calculation, and the experimental data. Any disagreement between the first two is exactly what you want to see before trusting a unit-op result.

Each point is obtained from the simulator's own engine: a pure bubble flash gives Ps(T); liquid properties (density, viscosity, thermal conductivity, surface tension, Cp, ΔHvap) are read from a slightly sub-cooled stream (liquid properties are essentially pressure-independent, so this equals the saturated-liquid value); vapour properties (vapour viscosity and thermal conductivity) are read at the vapour pressure but capped at 6 bar, and above Tr = 0.9 a flat 6 bar is used instead of Ps (near the critical point the saturation pressure is high and unreliable, and the saturated vapour is far from ideal). The ideal-gas heat capacity is the vapour Cp evaluated at a very low pressure, where the real-gas departure vanishes. Values are returned in SI and shown in the property's own units. A computed curve is cached per component, so re-opening the page is instant.

Requires the linked simulation to be reachable over COM (the CHEMCAD host) and its property set to have been read in (Read sim properties after linking).

6. Binary interaction parameters

For each component pair you can regress an NRTL (or other gE-model) parameter set to VLE/LLE/hE/azeotrope data, review it against the CHEMCAD databank BIPs harvested from an attached simulation, and select the set the project should use. Chosen BIP sets ride along in both exports.

6.1 Vapour-phase treatment (γ-φ)

Bubble pressures and vapour compositions use the γ-φ formulation yi φiV P = xi γi Pisat. The activity coefficient γi comes from the chosen gE model; the vapour-phase reality is handled as follows:

  • Carboxylic acids — the vapour does not behave as free monomers: two acid molecules form a hydrogen-bonded dimer, and near saturation most of the vapour is dimer (acetic acid at 373 K has an apparent molar mass near 100 instead of 60). The platform models this with the Marek-Standart chemical theory — the vapour is an ideal mixture of reacting monomers and self-/cross-dimers in chemical equilibrium — using the DDB dimerisation constants for the 17 parameterised acids. It is enforced (not optional) for those acids: it is applied when a curve is drawn and inverted when experimental points are reduced to activity coefficients, so γ stays a true liquid-phase quantity. When it is in effect it is named on the plot and in the regressed parameter set's description; an associating species with no dimerisation data (HF, acids outside the set) is reported and its estimated curve withheld rather than shown wrong.
  • All other components — the physical vapour non-ideality (fugacity coefficient φiV) is presently taken as ideal (φ = 1, i.e. modified Raoult), which is accurate at the low-to-moderate pressures where these correlations are used. A physical correction (Redlich-Kwong, the usual simulator convention) is planned; PSRK / VTPR predictive models are a separate later addition.

7. Exports: .nf vs .ccsim

Two ways to hand the data to CHEMCAD — they differ in how the T-dependent correlations travel:

CHEMCAD neutral (.nf)CHEMCAD sim (.ccsim)
Constantswritten as valueswritten into the sim's property bank
Correlations written as a (T, value) table — CHEMCAD regresses its own coefficients the regressed coefficients are written directly
BIPsNRTL lines in the blockwritten into the job's BIP file
Resultimport into a new/existing CHEMCAD jobopen the file — components + thermo are already there (no flowsheet)
Modifying an uploaded simulation

A .ccsim you upload can be re-downloaded with your changed constants and BIPs written in. A CHEMCAD databank component can't be altered in place, so any component whose parameters you change is cloned into a user component: it keeps its slot in the simulation but is given its own identity, so CHEMCAD uses your values instead of its databank ones. When you open the file, CHEMCAD recognises it as a user component and offers to save it to a user database. Its binary interaction parameters travel with it, and the flowsheet is untouched — streams and units reference components by their position in the component list, not by name or id, so re-identifying one component never disturbs the flowsheet.

Neither export touches a CHEMCAD licence or dongle — the files are written directly.

8. Data sources

The clean core is the public NIST/TRC ThermoML archive plus our own estimations; the company and project layers hold your imported and measured data. External sources (Dortmund Data Bank availability, KDB, NIST WebBook, PubChem, …) are queried per project and not folded into the core. See Data Sources & Licensing for the full source tiers and the licensing posture.

9. Claude assistant

The assistant (orange button, or the Claude operator mode) reads your projects and data, computes with the platform's own engine, and — inside your own project — acts directly: no confirmation clicks. Before the first change of a turn it takes a dated restore point of your project database, and every action is logged. Its chat is stored with the project and restored when you reopen it; the button starts a fresh session.

  • Components — add single components by exact name or SMILES; add a whole homologous series (e.g. the first 100 oligomers of an acid + epoxide) in one request. Identification is exact-or-nothing: novel structures become honest custom components, never a "similar" database hit.
  • Properties — pin Rarey-Nannoolal estimates into the parameter block for all components at once; your own chosen values and core-DB values are never overwritten, and out-of-range methods are reported, not silently filled.
  • UNIFAC groups — list the group assignments (original UNIFAC, modified UNIFAC (Dortmund), PSRK) per component.
  • Kinetic reactions — build a reaction ladder with one shared rate constant k(T) = k0·exp(−Ea/RT); k0 and Ea stay editable on the project page under Kinetic Reactions (deliberately separate from the electrolyte equilibrium block).

10. Outlook — planned

On the roadmap, not yet available:

  • Uncertainty-weighted regression — use each experimental point's reported (ThermoML) uncertainty as the fit weight, falling back to the typical absolute-error model where a point carries none.
  • Full custom-component export — write a complete property set into the .ccsim for components not found in the CHEMCAD bank (today those export with minimal data and may need completion in CHEMCAD).
  • More mixture models — electrolyte (eNRTL / PSRK-LIFAC) support for aqueous and ionic systems, alongside the existing NRTL / UNIQUAC / UNIFAC.
  • User-configurable units — choose your preferred units for every displayed variable (temperature, pressure, composition, energy, …) instead of the current fixed SI-based display.
  • Entrainer / solvent selection — systematic screening and ranking of candidate entrainers for extractive and azeotropic distillation and of solvents for liquid-liquid extraction, built on the predictive models (selectivity, capacity, boiling-point and azeotrope/LLE constraints).
  • More data-source connectors — extending the current Dortmund Data Bank / KDB / NIST WebBook / PubChem set.

Questions or a method you'd like documented in more depth — contact tec.rarey.net.