loading...
running...
How to use
Pick a preset, or tick compounds and set the mobile phase and column. The animation shows the column with silica beads and the coloured bands moving down it, the UV flow cell at the outlet and the chromatogram being drawn live at the speed you chose (the Fast-forward button completes the run). Switch to TLC plate to see the same chemistry on a thin-layer plate under a 254 nm lamp. Drag to orbit, wheel to zoom. Space pauses, R restarts. Things to try: (1) at 40 % methanol compare the predicted and observed retention times; (2) lower the organic fraction to 25 % and watch resolution rise and late peaks broaden; (3) tick all six compounds, run isocratic at 40 %, then run the 10-90 % gradient preset; (4) raise the flow to 3 mL/min and read the plate number N of a peak, then compare the van Deemter graph; (5) shrink the particle to 1.7 um and shorten the column to 50 mm - the pressure readout tells you what it costs; (6) raise the temperature and watch k fall; (7) inject 50 uL and compare peak widths of early and late peaks.
The glass column, ring stand, clamps, stopcock, solvent bottles, fraction-collector turntable with 12 tubes, rack, bench and a magnified silica bead with grafted C18 chains were modelled in Blender 5.2 with baked PBR textures (glass, brushed steel, silica, wood, plastic) and exported as one Draco GLB. Three.js r160 places the colour bands from the same band-transport model as the 2D page, fills the fraction tubes with the eluate that leaves the column and animates the carousel.
The science
Chromatography separates a mixture because its components spend different fractions of their time stuck to the stationary phase (here octadecyl-silica, C18) and moving with the mobile phase (water/methanol). A compound that is retained moves more slowly. The page models one analytical column, 4.6 mm wide, packed with porous particles (total porosity 0.65).
Retention. The retention factor k is the ratio of time spent in the stationary and in the mobile phase. A band moves at the mobile-phase velocity divided by (1 + k), so in an isocratic run (constant composition)
tR = t0 (1 + k), t0 = ε A L / F, RF = 1/(1 + k) (thin-layer plate)
In reversed-phase liquid chromatography k falls exponentially with the organic fraction φ (the linear solvent-strength model of Snyder) and with temperature (van 't Hoff):
ln k = ln kw − S φ − (ΔH/R)(1/T − 1/298.15 K)
The values of ln kw, S and ΔH are plausible, illustrative numbers for small molecules of this kind chosen for the demo, not measured data of those exact compounds on one specific column. The display colours are also only a visual aid: most of these compounds are colourless and are seen by a UV detector.
Band broadening. Each band is a Gaussian whose variance grows as it travels. The plate height H is the variance added per unit length, from the Knox equation with the Giddings correction of the mass-transfer term for retained solutes:
h = H/dp = A ν1/3 + B/ν + C f(k) ν, ν = u dp/Dm, N = L/H
The three terms are eddy dispersion, longitudinal diffusion and mass transfer. Their sum has a minimum, the optimum velocity, and smaller particles lower the whole curve (and flatten it at high speed). Dm scales with T/η and the viscosity η of methanol/water has a maximum near 40 % organic.
Resolution. Two peaks are baseline-separated at Rs ≥ 1.5:
Rs = (tR2 − tR1) / (2 (σ1 + σ2)) ≈ (√N/4) ((α − 1)/α) (k/(1 + k))
The three factors are efficiency (column length, particle size, flow), selectivity α = k2/k1 (chemistry, temperature) and retention. The fourth graph is the optimisation map: the worst-pair Rs for every isocratic composition. The best composition for the early pair is a weak solvent, but then the last compounds take hours: the general elution problem.
Gradient elution. If φ rises with time, a band sees a falling k as it moves. The model integrates the front of the programme (it travels at the mobile-phase velocity after the dwell volume has been flushed) and the band centre x, dx/dt = vm/(1 + k(φ(t − x/vm))). The back of a band meets stronger solvent earlier than its front, which compresses it; the variance obeys
dσ²/dt = 2 (∂v/∂x) σ² + H v, ∂v/∂x = − S k/(1 + k)² dφ/dt
so late peaks come out as narrow as early ones. Back-pressure follows Darcy's law, ΔP = Φ η L u/dp² with a flow-resistance factor Φ = 500: halving the particle size quadruples it, so a UHPLC column needs a pump rated for hundreds of bar. Injection of a plug of V microlitres adds a variance (V/(εA))²/12 to every band.
Numerical checks run in the page. (1) The band ODE reaches the exit at t0(1 + k); (2) the detector peak area equals the injected mass times t0/L for any k; (3) the resolution measured from simulated half-height widths equals the resolution computed from each peak's own plate number; (4) the van Deemter curve has an interior minimum; (5) and (6) the Darcy scaling with particle size and flow; (7) a gradient run from the ODE matches an independent marching integral; (8) the van 't Hoff ratio for 50 C against 25 C.
Simplifications. Linear (Henry) isotherm with no overload or tailing; Gaussian bands; no extra-column volume except injection and a first-order detector time constant; no pump pulsation, noise or baseline drift; the mobile-phase composition is the same across the column cross-section; constant flow-resistance factor; pH and ionisation (so important for acids such as aspirin and ibuprofen) are ignored. It teaches the scaling laws, it is not a method-development tool.