Free, glass-box, GPL-3.0 — steady, batch, dynamic, control and properties
The process simulator that shows its work.
Choupo is a free and open-source chemical process simulator: chemical engineering flowsheet simulation software for steady-state and dynamic processes, written in C++17 and run in your browser through WebAssembly, with nothing to install. Build steady, batch and controlled processes; inspect thermodynamics; fit property models; and follow the balances and nonlinear solves all the way to the answer — flash and phase equilibrium, MESH distillation, CSTR, PFR and Gibbs reactors, heat exchangers, compressors, reverse-osmosis and nanofiltration membranes, electrodialysis, crystallisation and drying.
EduTools — nothing to install, no case to open
None of them needs a case open first. Most run Choupo's engine on their own worked example in your browser and set the method beside the engine's answer; others work a closed form on the page, quote a cited source or a recorded run, or start from assumptions you set. Each card says where its numbers come from, and in the app every result carries the same mark.
What a Choupo case is on disk and how the engine puts it together, on the real flash01 case: system/ says HOW, constant/ says WITH WHAT, 0/ says FROM WHERE. Click any file to see what it says and the line of the engine that reads it, follow the assembly move by move against the run's own log, see that topology is not state and that a stream is fixed by N_c + 2 numbers, see what a run writes back -- then break a copy of the case and read Choupo's refusal in its own words.
Open in Choupo →Where a run's log goes and how loud it is (verbosity 0 to 4, measured), then the log itself, read on real runs in the browser: the header and what was assembled, a unit's Newton iterations, a convergence verdict in five numbers, a recycle whose units stay quiet until the tear converges, the balances, and the two blocks at the end. Every line is coloured by the marker the engine printed it with, and a click names the line of the engine that printed it. And the three messages a student must tell apart -- an advisory, a divergence, a refusal -- with the exit codes 0, 1 and 2 and what each asks you to do.
Open in Choupo →That the Buckingham Pi theorem only REDUCES the number of variables -- f = Phi(Re, eps/D) is still an arbitrary function -- and that the simplicity of a correlation comes from elsewhere: asymptotic regimes where a group drops out or enters as a power (laminar 64/Re, fully rough flow), the separation of scales in which a fast slave follows a slow master (Haken's two-variable model, the boundary layer), and the narrow range and generous tolerance a power law is fitted with -- the dimensional matrix reduced exactly, the Moody plane and its local slopes, a one-decade fit whose exponent drifts, and Churchill's expression rebuilt from its two asymptotes as the interpolation it is, every curve the engine's own -- and why the laminar-turbulent transition band is a region of instability where Re alone does not decide the regime, from Reynolds' 1883 apparatus to the measured onset of sustained turbulence.
Open in Choupo →First, what a temperature IS: the property that decides the direction of heat between two bodies in thermal contact, intensive, and not an amount of energy; equal temperatures define thermal equilibrium where heat can pass. Then that a temperature reading is the END OF A MEASUREMENT CHAIN, not a number nature printed on the system: a thermometer measures T through a signal, a calibration and its assumptions, and decimal places are not uncertainty. The entropy definition, K vs T90 and pyrometry are an optional third part.
Open in Choupo →The entropy BALANCE first -- variation, transfer (with heat or matter, never with work) and generation (never negative), worked on two blocks exchanging heat -- then that an entropy value in a simulator is a LEDGER: a measured third-law datum (s_298), a temperature line (integral cp/T), a pressure line (-R ln P/P0), a mixing line (-R sum y ln y) and a model line (s_R) -- each computed in one named place. The floor on the work of a complete ideal-gas separation (T times the mixing line) with its conditions, how an adiabatic machine GENERATES entropy (dS_gen), and which lines cancel when two states are compared.
Open in Choupo →That energy is conserved but the capacity to produce work is not: the same heat from two reservoirs, B_Q = Q(1 - T0/T), buys different work. Then PHYSICAL flow exergy prices a stream in work, b_ph = (h - h0) - T0*(s - s0), its entropy leg SIGNED, against a restricted dead state the case DECLARES (the engine refuses to choose your environment; chemical exergy is refused, never silently zeroed). Why the two LEGS re-add on screen, why b_ph at the restricted dead state is zero identically, why b is datum-independent but NOT model-independent, and Gouy-Stodola: every dS_gen an adiabatic machine publishes costs T0 times itself in exergy destroyed.
Open in Choupo →That a thermometer is never right by itself -- it is right the way a witness is credible: agreeing with itself (Regnault's comparability), agreeing with others (convergence), and being cross-examined by something that does not share its errors. The gas thermometer solved live by the engine (and why its idealised reading is Z times T), ITS-90's own fixed points and the instruments that DEFINE the scale (not the only ones that measure on it), what a quoted platinum melting point is worth, and Hasok Chang's four circles, labelled as his interpretation.
Open in Choupo →That every number in a component's data file has a PROVENANCE -- measured, fitted, or estimated -- and the reflex question to ask of each: measured by whom, fitted over what window, estimated from which groups at what typical error. The anatomy of a .dat file by what CONSUMES each field, and the standing rule that estimation is a curation-time act, never a runtime one.
Open in Choupo →Never trust a thermophysical record merely because it exists: test its internal consistency and read its declared validity. Thirteen records cross-examined against themselves, live -- water's own file disagrees with itself by 0.70 K, and a predictive glycerol correlation that asked to be validated misses by a factor of 19.
Open in Choupo →That every mixture model answers one question -- what did it KNOW about this pair before it priced it: nothing (ideal), the pair's own measured data (Wilson/NRTL), the groups (UNIFAC), the quantum charge surface (COSMO-SAC), or molecular theory with association (PC-SAFT). The engine grades four of them against 12 measured ethanol/water points live; a fit wins where its evidence was taken and loses where it was not, and the predictive rungs buy reach at a price the page shows in kelvin.
Open in Choupo →The full derivation of COSMO-SAC 2002 as the engine runs it: the quantum charge surface, the σ-profile, the misfit and hydrogen-bond exchange energy, the IMPLICIT segment activity coefficient solved as a damped fixed point, the residual as a move between environments, and Staverman-Guggenheim for size. Every equation cited to the line of CosmoSac.cpp that runs it, every constant quoted with its units, every assumption named where it is made — and the licence honesty that the shipped σ-profiles are synthetic teaching surrogates.
Open in Choupo →That Wilson, NRTL and UNIQUAC are one idea carried three distances -- a molecule's neighbourhood is not the bulk composition -- and that each exists because of something the previous one could not do: Wilson STRUCTURALLY cannot produce a liquid-liquid split, NRTL buys that with a third parameter, UNIQUAC separates size from energy and opens the door UNIFAC walks through. Every equation cited to the engine line that runs it, every assumption named where it is made -- including the one an implementation has to make, that a pair with no parameters does not fail but runs on a SUBSTITUTED neutral value (tau = 0 in NRTL, Lambda = 1 set directly in Wilson, which zero energy alone would not give), whose binary is then ideal -- announced and refused unless the case authorised it.
Open in Choupo →That UNIFAC is not a new theory but UNIQUAC's residual term moved one level down, from molecules to their groups: r and q assembled from group counts, a liquid treated as a solution OF GROUPS, one interaction number per ordered main-group pair, and the pure-minus-mixture subtraction that makes a group activity into an activity coefficient. And that its ideal fallback is a DIFFERENT failure from NRTL's despite identical arithmetic -- NRTL's gap is nobody fitted your pair and closes with data, UNIFAC's is a hole in the published table and does not.
Open in Choupo →That PC-SAFT builds a residual HELMHOLTZ energy from named contributions -- hard chain, dispersion, association -- so one surface yields both phases and their densities (an activity model needs a separate vapour model, and a Henry standard state for a dissolved supercritical gas). The temperature-dependent segment diameter, the four zeta moments, the perturbation series in packing fraction, van der Waals one-fluid mixing with k_ij as a CORRECTION rather than the whole interaction, and Wertheim TPT1's site-fraction fixed point closed into a_assoc; then the derivatives that give Z, P and ln phi, and why two density roots are candidate phases, not an equilibrium. And the lesson this repository paid for: the association SCHEME is part of the fit -- water curated as 4C instead of the paper's 2B passed a pure-density anchor by coincidence while the mixture flash collapsed.
Open in Choupo →That in this case the pH axis of a speciation diagram is a RESULT, not a knob: 44 separate equilibrium calculations, each a real neutral mixture, each landing wherever its own charge balance puts it -- and the two crossovers arriving near the two pK's without either crossover having been imposed (the K's are data the case reads). An imposed-pH diagram is a different, valid specification that must say what carries the charge. Also the two things the hand-drawn version cannot show: the three fractions of the total carbon do not sum to one, because the titrant is not a spectator (the plotted curves are renormalised onto the three and the shortfall is printed beneath them), and the crossovers sit below the thermodynamic pK because the engine works in activities.
Open in Choupo →That the enthalpy of real hydrogen rises 20.6 kJ/mol from 300 to 1000 K at 1 bar and 1.3 kJ/mol from 1 to 1000 bar at 300 K — small against temperature, never zero; that the degree sign names a RULE OF REFERENCE (the standard pressure and the reference phase) and not a temperature, so a standard property is a function of T; that a process at 150 bar uses the standard property plus the equation of state's residual, without assuming an ideal gas; that the zero of enthalpy is a third, separate choice, and a datum built from conserved quantities (the elements, the usual one) makes species' zeros consistent across every reaction; and why equilibrium is a minimum of G, not G = 0, with Kirchhoff, Gibbs–Helmholtz and van 't Hoff kept apart.
Open in Choupo →Which solvent, or which blend of two, for a given substance: Hansen's three parameters with the temperature they were placed at, the distance of a solvent from the solute's sphere and the relative energy difference where a radius was measured, the volume-fraction mixture rule as a checked rule of thumb, and the phase check that refuses to average a pair the activity model finds two-phase -- the student predicts the best composition before the engine's curve is shown, on a synthetic pair whose minimum sits at 50:50 by construction, then meets Hansen's own polystyrene with hexane + acetone (better than either) and hexane + nitromethane (the rule likes it, UNIFAC refuses it), beside the dissolution ratings he recorded; and the sentence that a small distance is a compatibility indicator, never a saturation concentration.
Open in Choupo →That putting ENTHALPY on the vertical axis turns the energy balance into a point on the diagram: the difference point, whose enthalpy is a net flow over a flow and belongs to no stream, and from which every operating line is a ray. The lever rule then reads the internal reflux off the page with a ruler, total reflux is the difference point at infinity, and minimum reflux is a ray coinciding with a tie line. Above all: two parallel straight enthalpy curves are a sufficient condition for constant molar overflow, so a departure from them shows where the internal flows can vary -- how much that moves the reflux or the stage count needs both models solved for the same specifications, which this page (one ray, no staircase) does not do.
Open in Choupo →That a dissolved acid gas is not one substance: H2S(aq), HS- and S2- coexist and the pH decides the split, so a stream carries BOTH bases -- the apparent components that close the mass balance and a declared speciation of the liquid. That the equilibrium is a small Newton (mass action, master balances, electroneutrality when the pH is solved) whose K's move with temperature by van't Hoff on each record's own enthalpy. That a reactive TRAY is just that flash, reached through one K-value entry point, so the column's Jacobian never sees an ion. And the prediction a reader gets wrong: stripping the acid gas makes the pH FALL, because the ammonia leaves too and the sulfide does not follow the carbonate out.
Open in Choupo →The single equilibrium stage, and the two curves that decide it: the material balance is a straight line that PIVOTS about the feed on the diagonal with slope -L/V, and where it cuts the equilibrium curve is the flash. Every other construction in this list is this step repeated, moved to other axes, or integrated.
Open in Choupo →Operating lines, the q-line and the staircase: how reflux R and feed quality q set the number of ideal stages.
Open in Choupo →Where the first number you type into a rigorous column comes from: total reflux fixes N_min, minimum reflux fixes R_min, and Gilliland — a fit, not a derivation — interpolates N(R) between those two asymptotes, with the rigorous MESH column's own stage count for the same separation drawn beside it.
Open in Choupo →The absorption factor A = L/(mV): how solute recovery scales with stage count when both lines are straight in the same basis — judged against the engine's stagewise absorber, which is itself a model (Henry's law, constant total flows), not a general rigorous absorber.
Open in Choupo →The graphical Rayleigh integration for a simple still: the area under 1/(y*−x) between the charge and the pot IS ln(W0/W) — drawn from the engine's own equilibrium curve, judged against the engine's time integration of the same still.
Open in Choupo →The S-shaped breakthrough curve: the mass-transfer zone consumes bed capacity long before the bed saturates — the ideal square wave drawn at the engine's stoichiometric time.
Open in Choupo →How a laboratory breakthrough curve sizes a plant bed: t_b and t_st read off the data, the capacity they imply checked against the isotherm, LUB = L(1 − t_b/t_st) carried to full scale at the same velocity — L_full = L_es + LUB, the diameter from the design flow, and why a longer bed wastes a smaller fraction.
Open in Choupo →The two pure batch operations behind a UF/NF membrane and the one line that differs between them (Q_d = 0 or Q_d = Q_p): each mode's own clock -- the volume concentration factor and the diavolume, never seconds -- the closed forms a student integrates by hand for each, and why they part from the run. Every one of them pulls the rejection out of the integral, and the engine publishes the observed rejection at every instant to show it moving. Hermia fouling as a claim about a MECHANISM, on a twin case that differs by nothing else, and the wash-time trade whose optimum is read off the engine's own flux rather than borrowed from a gel model this engine does not carry.
Open in Choupo →One point rules the whole column: the difference point Δ = F − E₁ = R_j − E_{j+1} lies on every operating line, so the triangle alternates tie-lines with lines through Δ — and the rigorous cascade's own stages are laid on it as the test.
Open in Choupo →Where to start when the problem stops being 'solve this flash' and becomes 'what equipment do I even choose?' -- twenty-six chapters of process-design heuristics, from flowsheet structure and reactor type through separation trains, membranes, solids, exchangers and pinch to hydraulics, relief and cost, each one click from the chapter that states it. And the posture that makes a heuristic useful rather than dangerous: a rule is a starting bet, not a verdict, this simulator is glass-box so you can falsify it in an afternoon, and anything that could touch safety follows the named standard instead.
Open in Choupo →The humid-gas state map: saturation, relative-humidity and adiabatic-saturation / wet-bulb lines locate the common drying and conditioning paths.
Open in Choupo →Hot and cold composite curves: the pinch splits the problem and fixes Q_H,min / Q_C,min before any exchanger is drawn — the in-view cascade cross-checked against the engine's targets.
Open in Choupo →Effectiveness vs NTU at a capacity ratio: why counter-current wins and when extra area stops paying — the run's exchanger placed on its own curve.
Open in Choupo →Merkel's one diagram: saturated-air enthalpy above, the operating line below, and the shaded gap between them is the driving force the packing must buy.
Open in Choupo →The two classical drying plots: X against time, and the rate against moisture — where, for a material with a constant-rate period (not every material has one), the critical moisture is VISIBLE as the corner at which a flat rate starts to fall toward the isotherm's equilibrium.
Open in Choupo →That a Claus stoichiometry CANNOT be written by hand, because '3/x S_x' contains an unknown that is itself an equilibrium -- and that the way out is to stop specifying reactions: declare the atoms and the candidate species and minimise G subject to A n = b, which collapses N unknowns into one element potential per ELEMENT plus a total. That an adiabatic temperature is a RESULT of the air rate, not a specification. That the 2 : 1 H2S : SO2 a conventional Claus plant is controlled to is never entered as a ratio: the declared air rate forces it through the O and H balances, while the furnace's 1.04 is explained by the hydrogen on the same line. And that a species with no formation data has no Gibbs energy and cannot play at all -- which is why the case writes two components of its own, one new and one an overlay filling a gap the catalogue record honestly declares.
Open in Choupo →One chart, two areas: the PFR's integral under 1/(−r) against the CSTR's rectangle at the outlet rate — why a CSTR needs more volume for the same conversion under positive-order kinetics.
Open in Choupo →That a design climbs from what the atoms allow to what thermodynamics allows and only then to what the rate delivers, each rung needing more information than the last; that a declared conversion is accepted even when equilibrium forbids it, because nothing on that rung knows K; that the equilibrium and Gibbs reactors agree when the species permit one reaction; that the temperature approach places a design short of the limit before any rate law exists; and that a plug-flow bed rises toward the limit and cannot cross it, reaching it when hot and falling short when cool — on one water-gas-shift feed at 10 bar.
Open in Choupo →That a Gibbs reactor declares NO reactions — it minimises G subject to the element balances, which is its power and its trap in one sentence; that it sits early because nobody has the KINETICS yet, not because of the project stage; that the approach to equilibrium evaluates the CHEMISTRY at T + ΔT while enthalpy, Psat and the energy balance stay at the physical T; and THE SIGN — positive is conservative only for an EXOTHERMIC reaction, which is why the author declares a MAGNITUDE and the ENGINE assigns the direction from the reaction's thermicity and announces it: a reformer gets T − ΔT without anyone typing a minus.
Open in Choupo →That the degree sign of a standard reaction enthalpy fixes the standard STATE (1 bar, each species in its reference phase), not the temperature, so there is a standard value at every temperature and the table's is the one at 298.15 K; that Kirchhoff's law moves it by the reaction heat capacity; that for ammonia the engine prices −45.9 kJ per mole of NH₃ at 25 °C and −53.4 at 500 °C, so the 25 °C value under-states a converter's heat by about 14 %; and that an energy balance on the elements datum never needs a reaction enthalpy at all.
Open in Choupo →That a plant with a recycle has TWO ratios: the fresh-feed ratio, which the steady-state plant balance ties to the stoichiometry, the extent and the losses, and the reactor-inlet ratio, which the loop sets — a fresh-feed error divided by the purge fraction, so one part in ten thousand moves the ammonia converter's H₂/N₂ visibly; that a SELECTIVE loss (gas dissolved in the product) shifts it even with a stoichiometric feed; and that which reactant is limiting is decided by safety, selectivity, cost or equilibrium, in that order, leaving one variable to optimise.
Open in Choupo →That an ideal-gas equilibrium landscape is fixed by two signs you can read off the reaction — the thermicity (van 't Hoff) and the mole change (pressure) — and the engine draws the five cases of that table from the same Gibbs kernel; that a temperature approach is a HORIZONTAL move along the curve whose effect is the local slope, so one magnitude is a large shortfall on ammonia and nothing on a combustion plateau; that the sign is assigned per cell from the feed's OVERALL thermicity, so a minor endothermic product inside an exothermic feed (thermal NO) is moved BEYOND its equilibrium — the GLOBAL caveat as a curve; and that the three overall routes to fixed nitrogen — a sugar as the electron donor, water as the donor, hydrogen as the donor — price on ONE formation surface with every species on a declared standard state (`reactionGibbs`), the biological routes carrying no H₂ at all; the equilibrium limit and the extent bookkeeping transfer to an enzyme, the temperature approach does not.
Open in Choupo →That "equilibrium" answers two different questions: a Gibbs reactor is told SPECIES and opens every pathway they permit, an equilibrium reactor is told REACTIONS and moves only along them; that declaring every independent reaction (R = N − rank A) reproduces the Gibbs reactor, so leaving ONE off the list is a claim about the catalyst — on one syngas at 525 K and 50 bar, methane with every pathway open and methanol with methanation closed; that the declared answer is a constrained minimum of the same Gibbs energy, the Gibbs answer its bound; and that the approach temperature is the second knob, its sign assigned PER REACTION from the way each one runs — a shift running backward takes its K below T while the synthesis beside it takes its K above.
Open in Choupo →That syngas is a PLATFORM -- carbon, hydrogen and oxygen handed to a catalyst as CO and H2 -- and that two numbers decide what it is good for: H2/CO and the module (H2 - CO2)/(CO + CO2); the four ways to make it (steam reforming, partial oxidation, autothermal reforming, gasification) and what each gives; why reforming is run hot and at low pressure, and why industry still compresses the feed instead; carbon on the catalyst at low steam-to-carbon; the water-gas shift as the ratio knob, and why two stages when equilibrium needs one; the ratio each product asks (methanol, Fischer-Tropsch, oxo, ammonia) and why the module is fixed by the atoms of the feed, not by the shift; cleaning the gas and pure hydrogen by PSA; water gas, producer gas, town gas, and CO2 plus green hydrogen -- each part run by the engine in the browser, and what the equilibrium model does not do.
Open in Choupo →Heat generated against heat removed: a straight line can cut a sigmoid three times, so in this example the same reactor with the same feed has three steady states — and the middle one fails the open-loop slope test, a necessary condition only: the page computes no dynamics.
Open in Choupo →The concentration field INSIDE one pellet: turn the rate constant and watch it go from nearly flat — the whole pellet working — to a boundary layer hugging the surface over a dead core, with the effectiveness factor beside it as the consequence, and the closed form drawn over the numerical answer so the method is watched being checked.
Open in Choupo →The operating point is an intersection: the pump model's rise falling with flow (a constant-power model, not a centrifugal pump's measured curve, and no basis for selecting one) against the pipe system's demand rising with it — crossed where the engine's own columns cross.
Open in Choupo →What a valve taking a large pressure drop does to the stream, and why its severity depends on the fluid's state, its vapour pressure, the valve's F_L and x_T, its geometry and the flow: the green-ammonia separator liquid let down 140 -> 20 -> 1 bar through the engine's isenthalpic valve -- the dissolved gas first, then the ammonia flashing, the temperature falling to the two-phase equilibrium state at the outlet pressure and the volume growing about forty times; the purge gas's mass flow choked at a pressure ratio of 70; the vena contracta against the vapour pressure (cavitation versus flashing, F_L yours) and the liquid choke; staged and tortuous-path trims, where the valve is installed, hardened trims, and power-recovery turbines instead of a valve; and why the noise follows the dissipated power.
Open in Choupo →Five valves around a two-product column with a total condenser, three inventories, two left for composition: the cross-section shows which measurement drives which valve, the cited authorities disagree in the open, and the sensor tray is computed from the column's own profile — the slope criterion and the sensitivity criterion, which pick different trays.
Open in Choupo →Whether a loop will be stable BEFORE it is switched on, and how much room is left, under the conditions the page states (a linear, time-invariant, stable open loop whose phase crosses −180° once): the engine sweeps a real tank one frequency per run and publishes the measured amplitude ratio and phase lag, the first-order closed form is laid over those points as the test, and then gain, lag, integrator, dead time and the PID terms are stacked into an open loop whose two crossover frequencies give the gain and phase margins. The page states which half Choupo computes and which half it does not.
Open in Choupo →Why noisy plant meters do not close a mass balance, and what to believe instead -- read through Bayes from the first page: a prior (what was known, at least the exact balances), a likelihood (what the meters say) and a posterior (what is believed now), whose most probable point is the weighted least squares answer and whose width is narrower than its own meter's. Then plant a gross error on any meter and watch the frequentist chi-square and measurement tests reject the snapshot and point at the meter, and the error smear over its neighbours until the meter is taken out of service; add yesterday's estimate as a prior; and set three answers side by side -- classical, Bayes with a flat prior, Bayes with yesterday's prior -- to see that Bayes represents and updates what is known.
Open in Choupo →How an owner buys a process design in stages separated by gates (Front-End Loading, FEL), with FEED / basic engineering as the engineering work inside FEL-3, and why sources number the stages differently; what each stage decides, at what simulation fidelity, with which deliverables and which AACE 18R-97 estimate class; the range of ranges drawn as overlapping bands around your own point estimate; why separated bands are sufficient, not necessary, to choose between two options (correlated errors, the difference estimated directly, expected cost and risk); and where Choupo stands at each stage, from the ammonia ladder to the green-ammonia base case.
Open in Choupo →That a plot plan is a set of decisions about distance -- safety (the domino effect), pipe and access (a tube bundle must be pulled out along its own axis); what a separation distance is, why it is keyed on pairs of equipment kinds and must say whether it is measured edge to edge or centre to centre, and why Choupo ships no spacing table (the published ones grant no redistribution, so the case declares its own, with a source); a guided exercise on the ammonia loop's declared plan, checked by the engine in the browser: a short distance announced, an overlap refused; the prevailing wind (ignition sources upwind of releases) and the hazardous-area zones 0, 1 and 2 as envelopes the case declares; and what the model does not do -- pipe cost and pressure drop (the engine routes and measures the pipes, and the page shows their lengths), structures, drainage.
Open in Choupo →That a pharmaceutical plant is laid out inside a building: the cleanliness grades of EU GMP Annex 1 and how this example declares its oral-solid-dose rooms grade D by analogy or controlled not classified (a declared choice, not a universal requirement); the pressure cascade (air from clean to less clean with the doors closed, the case's declared step -- a control, not a proof that nothing crosses); cascade, bubble and sink airlocks, and why a contained room's airlock is a sink; gowning for the grade one enters; containment of a potent ingredient (negative pressure, the exposure band Choupo does not know); the personnel, material and waste flows, one way each and kept apart as this example organises them; the solvent suite as a DECLARED ATEX zone checked geometrically against ignition rooms (a classification needs the sources, their grade and the ventilation); and escape as reachability (a lift is not an exit) -- each as an exercise on a teaching building the engine checks in the browser, a broken rule announced and an impossible building refused; and what the model does not do (air changes, doors, travel distances).
Open in Choupo →Why the material of every wall decides a plant -- corrosion, strength at temperature, toughness, cost; the Pareto of the materials chemical plants are built from, as a teaching ordering: carbon steel first and why, austenitic stainless (304/316, molybdenum against chlorides), duplex, nickel alloys, titanium, copper alloys in exchangers, aluminium in the cold, linings, non-metallics and refractories; the metallurgy behind it -- ferrite and austenite, carbon in iron, the ductile-brittle transition, creep, the weld and its heat-affected zone, sensitisation and the L grades, passivity; how a material is chosen -- corrosion allowance and the wall, the temperature window, chloride stress-corrosion cracking, pitting, hydrogen embrittlement and high temperature hydrogen attack, ammonia stress-corrosion cracking of carbon steel, and the material factor in the cost; pump seals, gaskets, packing and elastomer compatibility; and what Choupo does with a material, on a bench where the engine re-sizes and re-prices one vessel for every material record in the tree.
Open in Choupo →That routes to a product are screened on paper in order of cost, cheapest filter first: the stoichiometric gross margin, the atom economy, the equilibrium limit at the catalyst's temperature, the heat to remove, and whether the product is where the atoms go when every species is allowed; that direct oxidation of methane wins the first four and fails the fifth, because the combustion products lie far below methanol in Gibbs energy; that the remaining questions (Safety, Environmental, Legal, Economics, Control, Throughput) need no rate constant; and that the rate law is bought last, for the one route that survives.
Open in Choupo →The importance of economic appraisal, contractual robustness and downside analysis, read from risks and outcomes recorded in public documents on a world-scale PTA plant: why the price of PTA, the paraxylene it is made from and the conversion cost are analysed together, with the spread as the central measure of the margin; a capital cost stated seven times with different perimeters; an offtake contract whose 450 kt/yr is a maximum, not a guaranteed purchase, against a 700 kt/yr plant; a simplified pre-tax, unfinanced appraisal you move, with its break-even spread and a tornado, labelled an illustrative teaching scenario; the risks the lender's risk department recorded, each read as risk identified, condition recommended, what the documents record about the decision, and later outcome; and ten loss and price figures, each defined, that overlap and must not be added.
Open in Choupo →Patent or trade secret as a decision you take -- can it be reverse-engineered, can infringement be detected, how long it lasts against the 20-year term, what filing costs, whether the secret survives staff turnover -- and why a catalyst recipe and a new molecule land on opposite sides; the clock a filing starts (priority year, publication at 18 months, the PCT's 30 months, divisionals); intellectual property as a weapon (blocking and fencing, thickets, decoys, continuations, evergreening, utility models), defensive publication as the end of novelty for everyone, prior-user rights and dated deposits, and the counter-moves (freedom-to-operate searches, third-party observations, opposition); cross-licences, pools, standard-essential patents and assertion entities; employee inventions; the limits (sham litigation, abuse of dominance, inequitable conduct, misappropriation); worked cases on a process patent whose recipe does not work, a turbine whose blades melt, Whittle's turbojet patent and the Wright and Watt patents; a case study of the reverse-osmosis membrane, from Loeb and Sourirajan's cellulose acetate to Cadotte's composite and the FilmTec litigation, told from the patents and the Federal Circuit opinions; and an exercise against a rival who reads your filings. Not legal advice.
Open in Choupo →Why every protective layer is itself a system that can fail (Gall, a paediatrician, beside Kletz, a chemist: 'what you don't have, can't leak'), read through the Flixborough Court of Inquiry and the Bhopal protection that was out of service; Stoessel's cooling-failure scenario run by the engine on a reaction that is fictitious by declaration -- the accumulation, the MTSR with the dosed reactor's mass correction, the worst instant right after a dose, the class of one scenario against the class of the process, the margin to the reaction mass's bubble point along the run and an assumed T_D24 -- on a batch, the same reaction dosed, and a dosing vessel sized so that its whole content cannot take the reactor past its limits (a pump limits the rate, not the total); and a 50 bar ammonia loop that converts less per pass, passes about four times the volume of gas, carries about four times as much ammonia in the recycle stream, and earns less, so that the engine cannot call it safer without sizing the equipment and the release scenarios.
Open in Choupo →Why a relief device is sized for the credible cause (blocked outlet, external fire, cooling or reflux failure, runaway, tube rupture, thermal expansion) that needs the largest required relief area, not the largest flow; operating, design pressure, MAWP and set pressure kept apart, and the allowed accumulation; the devices and the back pressure behind them; the published gas, liquid and fire-case equations worked on fluid properties the engine computes, and where the liquid one stops holding -- an acetone drum in a fire, and an ammonia loop whose 50 bar version, on the same declared blocked-outlet flow, needs a relief valve about four and a half times larger in the sizing arithmetic (a comparison under one teaching hypothesis, not of two real relief systems) -- where the relief goes after the valve, and why a runaway vent is often two-phase and, under a restricted energy estimate, must pass more mass.
Open in Choupo →How far a pool fire's radiation reaches, built from the fuel up: the heat of combustion from the catalogue's formation enthalpies, the Burgess burning rate calibrated on one public-domain NIST measurement and checked on another, the flame height by Thomas and by Heskestad against measured heights, the emissive power by energy and by Mudan and Croce's smoke correlation (and why it does not fit a methanol flame), the point-source and solid-flame models side by side -- apart near the fire, the same law far away -- and the still-air distance to four published thresholds, each with the receiver and exposure its source states, set beside the separation the acetic-acid plant's layout declares -- a consequence for one scenario, not a risk and not a certified siting distance.
Open in Choupo →Why a chemical plant is designed to limit the consequences when something goes wrong: the bund as a passive barrier and its place among the layers of protection (it does not prevent the failure, it stops its propagation); every cubic metre of a spill accounted for -- in the tank, in the bund, over the wall, through the drain; the geometric capacity kept apart from the net volume Choupo's layout check computes and from the useful capacity, in which the failing tank's own footprint is free; the volume the tanks displace, a sloped floor, a freeboard and rain; the 110 % / 25 % rule as a teaching example traced to HSE guidance and an English regulation, with their limits, never as a universal obligation; a rupture, a leak, rain first and a drain left open, integrated in time and drawn in section until the bund overtops; low and wide against high and narrow, and how a badly conceived bund raises some risks; the pool a flammable liquid makes, qualitatively; local or remote containment; construction, linings and compatibility without an invented table; and an exercise on three tanks of 20, 15 and 10 m3, dry and after rain.
Open in Choupo →A recycle has no first unit, so one stream must be assumed -- and WHICH one is decided by the order you declared the units in, not by how many loops you can see. Move a unit and watch the streams flip from forward to backward, watch the tear count fall when a reorder lets one cut open two cycles, and then read Choupo's own refusal when the declaration is withdrawn: it finds the cycle, names every unit on it, and still will not choose the cut for you.
Open in Choupo →How one accelerator serves a tear vector holding a molar flow, a set of mole fractions and a temperature at once: it never compares them. Each variable gets its OWN secant slope and its own coefficient, both dimensionless, both unchanged if you re-express the variable -- and the page lets you check that on screen. The price is drawn beside it: n scalar secants are blind to the coupling Newton's Jacobian sees, and the convergence TEST, unlike the step, is not scale-free.
Open in Choupo →The hard part of a constrained quadratic is not the algebra -- it is guessing which constraints the answer is pressed against. Watch the working set change: a constraint ADDED when a step runs into it, DROPPED when its multiplier goes negative, on the very problem Choupo asserts to 1e-9 before every constrained optimisation. Then meet it for real, reconciling a laboratory water analysis that does not balance on charge, where each law's Lagrange multiplier IS the number of standard deviations it moved each measurement.
Open in Choupo →That Levenberg-Marquardt is the easy half. Watch the damping being repriced up and down on a real NRTL fit -- and then meet the apparatus that makes the answer worth quoting: three measurements withheld from the fitter by data flow rather than by rule, an acceptance band declared with its reason BEFORE anything was fitted, and a verdict published as a WORD. Turn the band and watch that word change while the residual does not move -- which is exactly why a real curation declares it first. The same run is `validated` and NOT individually identifiable at once.
Open in Choupo →That Choupo's Gibbs reactor never solves for its amounts: it solves for ONE potential per element and the log of the total, M + 1 unknowns however many species, with every species' stationarity built into the change of variables -- so no amount can ever go negative. Watch the engine's own Newton, iterate by iterate: its first step cut to 1/64 by the line search, then the digits doubling. Check by hand that each species' chemical potential is the sum of its atoms' potentials, which puts every reaction at equilibrium at once. See a liquid and a solid added not by a bigger Newton but by a test on the gas answer and a one-dimensional search. And meet the feed on which the least-squares seed makes the Jacobian singular at iteration 0, and the primal re-seed the engine announces.
Open in Choupo →66 live tools, generated from the application's own tool registry — this page cannot list a tool the app does not have, or miss one it does.
What is inside Choupo
Every number here is generated from the published Choupo-2608 tree, not written by hand. See the full model catalogue for names, status and source.
Counts exclude READMEs, templates, generated outputs and private data — only what is active and distributed in the release. Regenerate with bin/curate/release_inventory.py.
The glass-box contract
Choupo keeps equations, inputs, iterations, balances and provenance close together. A model is not merely a name in a dropdown: the run tells you which correlation was selected, where its parameters came from and whether the numerical problem closed.
Cases remain plain-text dictionaries on disk. The browser runs the same C++ engine through WebAssembly; it is a runner and visualiser, not a second source of process physics.
isothermalFlash | benzene / toluene | 370 K, 1 bar (flash01) Rachford-Rice Newton it V/F |g(V)| 0 0.50000000 3.76e-02 1 0.30001863 7.82e-04 2 0.30398070 5.66e-07 3 0.30398357 2.95e-13 converged K_benzene 1.6514 from Psat / P K_toluene 0.6735 from Psat / P material balance CLOSED energy balance reported
How it compares
Three process simulators a chemical engineering student is likely to meet, compared on facts each one publishes about itself.
| Choupo | DWSIM | DWSIM Pro | Aspen Plus | |
|---|---|---|---|---|
| Licence | GPL-3.0, open source | GPL-3.0, open source | commercial subscription | commercial (academic licences through institutions) |
| Cost to a student | free, no account | free | paid plans | through the institution's licence |
| Where it runs | any modern browser (WebAssembly, computes on your machine) — or native Linux build | desktop install: Windows, Linux, macOS | browser, vendor-hosted cloud | desktop install, Windows |
| Source code | open | open | closed extension of DWSIM | closed |
| Case files | plain-text dictionaries, diffable in git | see each product's documentation | ||
| Solver trace in the run | every Newton iteration, residual, balance and data source printed | see each product's documentation | ||
Checked 2026-10-07 against each product's public pages: DWSIM — github.com/DanWBR/dwsim and dwsim.org; DWSIM Pro — dwsim.org (“Why DWSIM Pro?”); Aspen Plus — aspentech.com, and a university installation page for its licensing and platform (kb.wisc.edu: “Aspen is a Windows only program”, licensed through a licence server). If a cell is out of date, open an issue and it will be corrected. Aspen Plus is a trademark of Aspen Technology, Inc.; DWSIM and DWSIM Pro belong to their respective owners. Names are used only to identify the products compared; no affiliation or endorsement is implied.
Capabilities
The breadth is now larger than the old landing page suggested. These are implemented workflows with runnable cases, not a roadmap.
Recycle networks, balances, design specifications and non-ideal separations, from first flash to multi-unit plants.
Evaluate, compare and regress pure-fluid and mixture models with provenance and validity visible beside the result.
Detailed workflows in areas usually compressed into opaque blocks, with inventories and transport assumptions exposed.
Recipe-driven campaigns and continuous dynamics with fixed or adaptive integration, controllers and energy ledgers.
Representative workflows
Each case is runnable, readable and designed around a question that can be checked in its output.
Move beyond shortcut columns into stagewise material, equilibrium, summation and heat equations, including azeotropic and reactive systems.
Combine electrolyte speciation, activity-based saturation indices, concentration polarisation and module-by-module recovery.
Carry one adsorption record from measured-data regression into working capacity, batch uptake and conservative fixed-bed transport.
Compare MSMPR moments and finite-volume crystal-size distributions while closing dissolved, solid and campaign inventories.
See adaptive integration through a thermal event, then distinguish the plant integration grid from a fixed digital-controller sampling grid.
Evidence, not badges
Choupo separates numerical convergence, conservation, model validation and data provenance. They are different questions and receive different verdicts.
Most tutorial cases ship an expected golden-master file (383 of 393 in this release); the rest are sweeps, deliberate refusal cases and composite sub-sectors, which are still run and still guarded against NaN/inf. bin/runTests re-runs the whole suite and flags any KPI drift, across all 4 engines.
A passing golden proves an answer has not moved, never that it is right. Agreement with published measurement is claimed only where a case carries an explicit anchor row citing its primary source — a separate and much smaller set.
Choupo is educational and research software, not certified process-design or safety software. Engineering use requires independent validation of models, data and results.
No installation required
The browser application executes the C++ engine locally through WebAssembly. Choose a tutorial, inspect its files, run it and read the results.
Run it
The browser needs nothing installed and works everywhere. Build from source when you want the native binaries, custom cases and the regression suite.
The C++ solver is compiled to WebAssembly and runs locally in your browser tab. Nothing is installed, no account is needed and no case data leaves your machine.
Recommended way to start on any operating system.
# Nothing to install. Open: https://www.choupo.org/app/ # The same C++ engine as the native build, # executed in your browser via WebAssembly.
1 · Browser — recommended to start. Just open the app; nothing to install.
2 · WSL2 — recommended for the native build, development and the regression tests.
There is no published native .exe installer. The Unix commands below run inside WSL2 (Ubuntu), not in PowerShell or CMD.
# In PowerShell (admin), then reboot: wsl --install -d Ubuntu # Inside the Ubuntu (WSL2) shell: sudo apt update && sudo apt install -y git build-essential git clone https://github.com/choupo-admin/choupo.git cd choupo make all source etc/bashrc runCase tutorials/steady/flash/flash01_benzene_toluene bin/runTests # The graphical interface (GUI), once: sudo apt install -y nodejs npm emscripten make wasm-gui # then bin/runGui, and open in Windows: # http://127.0.0.1:5173/ bin/runGui
1 · Browser — nothing to install; open the app.
2 · Native build from source — the standard C++17 toolchain, no external libraries.
A native macOS build is not yet part of the automated test matrix; treat it as unverified (both Apple Silicon and Intel) until a workflow confirms it.
# Command Line Tools (clang + make):
xcode-select --install
git clone https://github.com/choupo-admin/choupo.git
cd choupo
make all
source etc/bashrc
runCase tutorials/steady/flash/flash01_benzene_toluene
bin/runTests
The native target. Only a C++17 compiler and GNU make are required — zero external libraries.
The graphical interface on your own computer needs three more tools — Node.js, npm and Emscripten — because it runs the engine compiled to WebAssembly. bin/runGui checks each one, says in one sentence what is missing, and opens a browser window at http://127.0.0.1:5173/. After pulling a new version, run make wasm-gui again.
# Debian / Ubuntu deps: sudo apt install -y git build-essential git clone https://github.com/choupo-admin/choupo.git cd choupo make all source etc/bashrc runCase tutorials/steady/flash/flash01_benzene_toluene bin/runTests # The graphical interface (GUI), once: sudo apt install -y nodejs npm emscripten make wasm-gui # then, every time: bin/runGui bin/runGui --kill # stops it
The browser app runs in Safari and can be added to the Home Screen (Share → Add to Home Screen) to open full-screen.
Reading cases, running tutorials and inspecting results work on iPad. Building or editing flowsheets is not recommended on small phone screens.
There is no native iOS app. Choupo on iPhone and iPad means the WebAssembly build in Safari — the same engine, in the browser.
# iPad / iPhone — open in Safari: https://www.choupo.org/app/ # Share -> Add to Home Screen for a # full-screen, offline-capable launcher.
Documentation
Run, organise and understand Choupo cases, reports and workflows.
Open the User Guide (Choupo-dev)Select, evaluate and curate property models without hiding uncertainty.
Open the Property Guide (Choupo-dev)Inspect every implementation or report a reproducible problem.
Open GitHubThese guides track the development line. Documentation matching the stable release is the tag's own: docs at v2608.
Cite Choupo
Geraldes, V. (2026). Choupo — Open-source Chemical Process Simulator (version 2608) [Computer software]. https://www.choupo.org/
People and governance
Choupo was conceived and created by Vítor Geraldes, chemical engineer and process-modelling researcher. The Properties Guide and Theory Guide are coauthored with Pedro Mendes and Miguel Rodrigues. Their authorship and copyright are preserved in those two works.
Suggestions, bug reports, validation evidence and data corrections are welcome. Choupo is centrally developed; external code contributions and pull requests are not accepted. Suggestions do not imply acceptance, priority or an implementation timeline.