The van Deemter Equation: Why There's an Optimal Flow Rate in Chromatography

Plate height: the currency of column efficiency

Chromatographic separations are scored by efficiency — how narrow the peaks are relative to how far apart their apexes sit. The historical bookkeeping comes from the plate theory of Martin and Synge (who won the 1952 Nobel Prize in Chemistry for partition chromatography), which imagines the column as a stack of N discrete equilibration stages, or theoretical plates. The number of plates is measured from a chromatogram as N = 16(t_R/w)² using the peak width at baseline, or N = 5.54(t_R/w₁/₂)² using the width at half-height, where t_R is the retention time.

The problem with N is that a longer column trivially has more plates. To compare columns and, crucially, to study why bands broaden, we normalize by length: the height equivalent to a theoretical plate, H = L/N (also written HETP). H has units of length — micrometers in modern HPLC, hundreds of micrometers in classical GC. A smaller H means each unit of column length delivers more separating power, so the whole enterprise of column and method optimization is a hunt for the conditions that minimize H.

The deep insight of the 1950s was that H is not a fixed property of a column but a function of the mobile-phase velocity u. Peaks broaden through several physically distinct, additive mechanisms, and those mechanisms scale differently with velocity — some grow as you speed up, one shrinks. The van Deemter equation is precisely the statement of how these competing contributions to variance add up into H(u).

The equation and its three terms

In their 1956 paper in Chemical Engineering Science, Jan van Deemter, Frans Zuiderweg, and Anton Klinkenberg wrote the plate height as a sum of three terms:

H = A + B/u + C·u

where u is the linear velocity of the mobile phase (cm/s). The terms are additive because they arise from statistically independent broadening processes, and the variances of independent processes add. Since H is proportional to the spatial variance σ² of the band per unit length, adding variances means adding H-contributions.

• A — eddy diffusion (the multipath term). In a packed bed, molecules thread through the interstitial channels by many different routes. Some paths are short and fast, some long and tortuous, so molecules that started together arrive spread out. Classically A ≈ 2λd_p, where d_p is the particle diameter and λ is a packing-quality factor of order unity. Notice A carries no velocity dependence — it is a flat floor under the curve.
• B/u — longitudinal (axial) molecular diffusion. A concentrated band diffuses along the column axis simply because of the concentration gradient at its edges. The longer a molecule sits in the column, the more it spreads, so this term is inversely proportional to velocity: go faster and there is less time to diffuse. B = 2γD_m, where D_m is the analyte's diffusion coefficient in the mobile phase and γ (≤ 1) is an obstruction factor accounting for the packing blocking free diffusion.
• C·u — resistance to mass transfer. Separation requires molecules to shuttle between the flowing mobile phase and the stationary phase. That transfer is not instantaneous: a molecule caught in the stationary phase is momentarily left behind, while one in the mobile phase races ahead of the band center. The faster the flow, the further apart these two populations drift before re-equilibrating, so this term grows linearly with u. C bundles both a stationary-phase (C_s) and a mobile-phase (C_m) contribution.

Because B/u falls and C·u rises, their sum has a minimum — and A merely raises the whole curve. That competition is the entire reason an optimal velocity exists.

Finding the optimum: calculus on the U-curve

The optimal velocity falls straight out of setting the derivative to zero. Differentiating H = A + B/u + C·u with respect to u:

dH/du = −B/u² + C

Setting this equal to zero gives the classic result

u_opt = √(B/C)

and substituting back yields the minimum attainable plate height

H_min = A + 2√(B·C).

The 2√(B·C) piece is the value B/u + C·u takes at its own minimum (the arithmetic–geometric-mean identity: x/u + y·u ≥ 2√(xy), with equality at u = √(x/y)). Everything below H_min is forbidden by the physics of the packing and the diffusion of the analyte. The A term, being flat, sets an irreducible baseline; the only way to push under it is to change the column itself — smaller or more uniform particles, or an open-tubular geometry where A disappears entirely.

A subtlety worth flagging: because H_min depends on √(B·C), reducing either longitudinal diffusion or mass-transfer resistance lowers the floor, but the two also trade off through u_opt. Making C small (thin films, small particles) both lowers H_min and pushes u_opt higher — which is exactly why modern sub-2-µm HPLC particles let you run fast and stay efficient. The curve becomes flatter and its minimum shifts to higher velocity, so you sacrifice little efficiency by running at high throughput. That flattening is the physical selling point of ultra-high-performance liquid chromatography (UHPLC).

A worked example with real numbers

Take a well-behaved HPLC separation on a 5-µm fully porous C18 particle, and use representative coefficients: A ≈ 8 µm, B ≈ 1 × 10⁻⁵ cm²/s (of order 2γD_m with γ ≈ 0.6–0.7 and D_m ~ 8 × 10⁻⁶ cm²/s for a small molecule in an aqueous–organic eluent), and C ≈ 3 × 10⁻³ s (a lumped mass-transfer coefficient with units of time, since C·u must have units of length).

Then u_opt = √(B/C) = √(1 × 10⁻⁵ / 3 × 10⁻³) = √(3.3 × 10⁻³) ≈ 0.06 cm/s ≈ 3.5 cm/min — a modest flow, consistent with the sub-1-mL/min rates typical of a 4.6-mm-i.d. analytical column. The minimum plate height is H_min = A + 2√(B·C) = 8 µm + 2√(1 × 10⁻⁵ × 3 × 10⁻³) cm. That radical is √(3 × 10⁻⁸) = 1.73 × 10⁻⁴ cm = 1.73 µm, so H_min ≈ 8 + 2(1.73) ≈ 11.5 µm — about 2.3 particle diameters. On a 15-cm column that is N = L/H = 150,000 µm / 11.5 µm ≈ 13,000 plates — a solid, well-behaved efficiency for a 5-µm packing.

Now watch the penalty for running off-optimum. At double the optimal velocity, u = 0.12 cm/s, the B/u term is halved but the C·u term doubles: H = 8 + (1 × 10⁻⁵)/0.12·(10⁴ µm/cm) + (3 × 10⁻³)(0.12)(10⁴) = 8 + 0.87 + 3.5 ≈ 12.3 µm, dropping you to ~12,200 plates. The high-velocity regime is dominated by C·u, which is why analysts who want speed reach for smaller particles rather than simply flooring the pump — you cannot outrun mass-transfer resistance with the same column.

Capillary columns, the Golay equation, and reduced coordinates

The original van Deemter treatment was built for packed columns. When Marcel Golay analyzed the open-tubular (capillary) column in 1958 — the geometry that would come to dominate modern gas chromatography — he found the multipath A term simply vanishes: with a single open channel and no packing, there are no multiple flow paths to broaden the band. The Golay equation keeps only B/u and an enriched C·u, splitting mass-transfer resistance into a mobile-phase term (governed by the column radius r and D_m) and a stationary-phase term (governed by film thickness d_f and D_s):

H = B/u + (C_m + C_s)·u, with C_m ∝ r²/D_m and C_s ∝ d_f²/D_s.

This is why capillary GC columns achieve staggeringly low plate heights and hundreds of thousands of plates: killing the A term removes the flat efficiency floor that packed beds can never escape. It also explains the practical dictum to use thin stationary-phase films and narrow bores — both shrink C — and, in GC, to prefer hydrogen or helium over nitrogen as carrier gas. Hydrogen's large diffusivity flattens the van Deemter curve at high velocity (small C), letting analysts run fast with little efficiency loss, whereas nitrogen gives a lower H_min but a punishingly steep C branch.

To compare columns of different particle size on equal footing, J. Calvin Giddings introduced reduced (dimensionless) coordinates — reduced plate height h = H/d_p and reduced velocity ν = u·d_p/D_m — which John Knox (and Horváth, for pellicular particles) then developed into the standard packed-column benchmark form. The empirical Knox equation, h = A·ν^(1/3) + B/ν + C·ν, replaces the constant A with a ν^(1/3) dependence — a better fit that reflects the coupling between eddy diffusion and mobile-phase mass transfer that J. Calvin Giddings described with his coupled (random-walk) theory in the 1960s. On this scale, an excellent packed column reaches h_min ≈ 2 (i.e., H_min ≈ 2–3 particle diameters) at ν ≈ 3–5, a universal benchmark analysts still use to judge whether a column is well packed.

Limits, subtleties, and where the model bends

The van Deemter equation is a lumped, semi-empirical model, and its clean additivity hides real physics. Giddings' coupling theory showed that eddy diffusion and mobile-phase mass transfer are not truly independent: a molecule in a slow flow channel can diffuse laterally into a faster one, partially averaging out the multipath spread. This coupling is exactly why a purely constant A over-predicts broadening at high velocity, and why the Knox ν^(1/3) term fits real data better. The three-term separation is a convenience, not a law.

Several other effects live outside the equation entirely and must be controlled separately:

• Extra-column band broadening. Variance added in injectors, connecting tubing, and detector cells adds to the observed σ² and inflates the apparent H. With sub-2-µm particles producing peaks only microliters wide, extra-column volume can dominate — the column may be excellent while the system is mediocre.
• Viscous heating and pressure. At the very high linear velocities and pressures of UHPLC, frictional heating creates radial temperature gradients that distort the classic curve; the plate height can rise again for reasons the isothermal van Deemter model never anticipated.
• Overloading and nonlinear isotherms. The derivation assumes a linear partition isotherm (analyte concentration low enough that K is constant). Inject too much and peaks become asymmetric (fronting or tailing), and H loses its meaning as a symmetric-Gaussian width.

Finally, the coefficients are not fundamental constants — B and C both scale with the analyte's diffusion coefficient, so the same column gives different curves for a small solvent molecule and a large protein. For biomacromolecules with tiny D_m, longitudinal diffusion is negligible (B ≈ 0) but mass transfer is brutal (large C), so their optimal velocities are very low and their curves nearly monotonic in the accessible range — a regime where superficially porous 'core–shell' particles, with a solid core that shortens the diffusion path, deliver most of their advantage by slashing C_s.

The three band-broadening terms of the van Deemter equation — physical origin, velocity dependence, and how to shrink each one.

Term	Physical origin	How to reduce it
A — eddy diffusion (multipath)	Analyte molecules take flow paths of different lengths through the packed bed; independent of u	Smaller, more uniform particles; well-packed homogeneous bed; use open-tubular columns (A → 0)
B/u — longitudinal diffusion	Molecular diffusion along the column axis spreads the band; dominates at low u because molecules dwell longer	Run faster; use a low-diffusivity mobile phase (liquid ≪ gas); minimize dead volume
C·u — resistance to mass transfer	Finite time to equilibrate between mobile and stationary phases; band advances while molecules diffuse in/out; dominates at high u	Thin stationary-phase film; small particles; low-viscosity, high-Dm mobile phase; core–shell particles