This is an IAST prediction tool, not a regression tool. It never fits parameters to your data: you supply them — from the built-in library, from a JSON file, or by typing them in — and the app evaluates the isotherms, predicts the binary mixture and, if you paste measurements, tells you how far the prediction sits from them.
The five-minute path
- Pick a sorbent / gas pair from the dropdown and press
Load. Both components arrive with published parameters and a citation.
- Set the temperature and the P min / P max window to your real operating conditions.
- Check the two component cards — model choice and parameters — then press Calculate.
- Read the pure isotherms, then scroll to the IAST panel for the binary prediction.
- Optionally paste measured data to see how the prediction compares, then press Report.
Bringing your own system
Press Import JSON and the dialog offers a blank template (every field, with its units and the model options spelled out) as well as a filled template built from the shipped library. Fill either one in and load it back through the same button. The dialog also has Choose JSON file… for a library you already have. Save / Load write and read a whole case as JSON — both components with their models and parameters, the conditions and the validation data. Browser storage keeps only the lightweight settings (temperature, pressure window, y list, trust limit, heat-map quantity, axis mode, theme, case name and which library entry was chosen), so a refresh restores your working conditions, but a hand-edited parameter set should be saved to a file.
y list is the set of gas-phase mole fractions of component A drawn as the curve family on the competitive-loading chart and written to its CSV. q max left blank auto-scales the loading axis; type a number to pin it.
Choosing a model
Twelve models are available per component, and the two components need not use the same one: linear (Henry), single-site and dual-site Langmuir, anti-Langmuir, quadratic, single-site Langmuir–Freundlich, Tóth, Freundlich, BET, type V, Sips + Henry, and a free-form custom expression. The panel below each card renders the live equation and the meaning of every parameter, so the model is never a black box.
- Langmuir (SSL) — one site type, saturating. The only model for which extended (competitive) Langmuir is thermodynamically consistent with IAST, and only when both qs are equal.
- Dual-site Langmuir — heterogeneous surfaces, the workhorse for CO2 on zeolites.
- Freundlich / Sips — empirical heterogeneity. Note the Henry limit is not finite when n ≠ 1.
- Tóth — heterogeneous with correct Henry limit; ψ is evaluated by quadrature rather than in closed form.
- BET, anti-Langmuir — multilayer and cooperative behaviour; both have a pole you must stay below.
Temperature dependence
Affinity constants use b(T) = b₀ exp(-ΔH / RT). Pick the basis (pressure or concentration) that matches how the parameters were reported — the label switches from ΔH to ΔU on a concentration basis, because that is what the fit actually contains.
Comparing against measurements
Paste rows of P_bar, y_A, q_exp into the validation box, set the data temperature, and press Plot & compare with data (Clear removes the overlay). Nothing is fitted: the points are overlaid on the prediction produced by the parameters currently in the two cards, and R², RMSE and AARD quantify the gap. If the agreement is poor, adjust the parameters or the model yourself — the app will not do it for you. Rows with y = 1 or y = 0 test the pure-component parameters you entered; intermediate y tests the IAST prediction itself, since IAST has no adjustable binary parameter.
What each card answers
| Card | Question | Read it like this |
|---|
| x–y equilibrium | How strongly is A enriched in the adsorbed phase? | Distance above the diagonal is the separation. The curve can never cross the diagonal. |
| ψ-construction | Where does selectivity reverse? | Crossings of the two ψ curves are exactly the pressures where S = 1. |
| Selectivity vs composition | Is S stable across composition and pressure? | Curves that fan out mean S depends strongly on feed composition — bad news for a fixed-cycle design. |
| Heatmap (P × y) | What happens anywhere in the P–y plane? | The drop-down switches between selectivity S₁₂, q₁*, q₂* and x₁. Click to pin a point; the two slice cards re-cut through it. |
| Slice q*(P) at fixed y₁ | q*(P) along an isopleth | The working capacity between adsorption and regeneration pressure. |
| Slice q*(y₁) at fixed P | q*(y) at one pressure | Sensitivity of loading to feed composition swings. |
| 3-D loading surface | qA(P, y) as a whole | A rotatable 3-D surface; drag to spin it, hover a vertex for its exact P, y and q. |
Diagnostics
The summary above the cards reports the Henry selectivity H₁/H₂, any pressure where the ψ curves cross (a selectivity reversal), the worst extrapolation ratio pᵢ⁰/P over the sweep, and per component the Henry limit (finite, zero or divergent), any pole in the isotherm, whether q(P) is monotonic, and whether ψ is analytic or by quadrature. The P trust field sets the pressure beyond which you consider the pure isotherm untrustworthy; the summary flags cases where IAST needs a hypothetical pressure p0i above that limit.
A flagged extrapolation does not mean the answer is wrong — it means the answer rests on isotherm behaviour you never measured. Extend the fitting range or widen the trust limit deliberately, not by accident.
Interacting
- Scroll to zoom, drag to pan. Every chart behaves the same way — the two charts at the top of the page and the seven IAST cards share one renderer.
- The toolbar in the corner of a chart holds pan, box zoom (then drag a rectangle), zoom in, zoom out, reset to the full range, pop out and save as PNG.
- Hover a curve to read exact values: the crosshair follows the cursor, the readout snaps to the nearest abscissa, and a dot is placed on every series there, so one hover reads both components at once.
- On the heat map the tooltip gives the cursor's own P and y alongside the plotted quantity, and clicking pins that (P, y) to drive the two slice cards.
- The 3-D surface works differently: drag rotates it, scroll zooms, reset returns to the default view, and hovering snaps to the nearest vertex to report its exact P, y and q.
- Anything that changes a chart's extent — q max, the pressure window, a different case — rescales the axes; a zoom you set by hand survives an ordinary repaint such as a theme switch.
Pop out
The ⤢ button opens the chart full screen. Type and line weights grow with the canvas, so a pop-out is genuinely more readable rather than the same chart stretched. Zoom and pan carry over, and you can download from there.
Chemical names
Component names are typeset as formulas wherever they appear on a chart: CO2 becomes CO₂, CH3OH becomes CH₃OH, Ca(OH)2 and CuSO4·5H2O keep their counts as subscripts, and an ionic charge such as SO42- is set as SO₄²⁻. Sloppy input is tolerated — co2, h2o and even water resolve to the right formula. Anything that is not a formula (Zeolite 13X, MOF-74, Methane) is left exactly as you typed it.
Styling
Every chart is drawn by the same built-in canvas renderer — no external plotting library and nothing fetched from the network. One profile is used throughout: bold padded axis titles, ticks pointing inward on all four sides, no grid, no legend frame, and a legend that places itself in whichever corner covers the least data. Type, line weights and margins scale with the canvas, so a chart stays legible inside a narrow card and gains presence when popped out.
Images
The PNG button redraws the chart at 5× device resolution on a white background with black axes. The layout is identical to what is on screen — nothing reflows, and the current zoom or 3-D orientation is exactly what you get — so the file is simply a sharper version of the figure you were just reading.
Data
CSV exports exactly the grid that was plotted, so a figure and its data always agree. Export CSV in the IAST panel header dumps every card at once and each card has its own CSV button; Export All (PNG + CSV) does all nine charts and every data set together (downloads are staggered — allow multiple files in your browser).
Text
Export params JSON gives a reproducible parameter block. The Report button builds a printable document — with its own Print / Save as PDF action — in which every chart is embedded as a high-resolution image alongside the model equations, parameter tables, fit statistics and a written interpretation of each card, including the caveats that apply to your particular pair.
What IAST cannot do
- No composition azeotropes. Because S = p20/p10 depends on spreading pressure alone, a binary IAST x–y curve can never cross the diagonal. Experimental data that do cross (the classic C2H4–C3H8 on 5A case) indicate real non-ideality; that calls for VST, not for re-fitting the pure isotherms.
- Ideal adsorbed phase. Strong lateral interactions, molecular sieving and pore-size exclusion are all outside the theory.
- Extrapolation. IAST evaluates each pure isotherm at a hypothetical pressure p0i that is generally higher than the total pressure. At low y for a weakly held component this can be orders of magnitude beyond your data.
Numerical traps
- Poles. Anti-Langmuir diverges at P = 1/b; BET at P = Ps. Keep
P max below the pole or the solver will not converge.
- Freundlich and Sips with n < 1 have an infinite Henry slope; ψ is computed by quadrature and the low-pressure end is the least reliable region.
- Degenerate pairs. If ψ1 ≡ ψ2 the components are IAST-indistinguishable and every card collapses to the trivial answer — the app says so rather than pretending to a prediction.