Rate Theory and Band Broadening

A chromatographic peak leaving a column is always a bell curve, never a spike. A plug of solute is injected as a sharp pulse. Why does it emerge, after passing through a metre of packed silica, as a smooth Gaussian? This peak is dozens of seconds wide. The answer is not chemistry in the usual sense. It is kinetics and random motion. This is the same physics that spreads a drop of ink in still water. Here, though, it is sped up, structured, and made productive. This happens through the flow of a mobile phase past a bed of particles. Plate Theory and Column Efficiency answers how many theoretical plates a column has. It does this by treating the column as a stack of discrete equilibration stages. This is an elegant fiction borrowed from distillation theory. It tells you that efficiency corresponds to a plate height $H = L/N$ . Smaller $H$ means sharper peaks. What it cannot tell you is why $H$ takes the value it does. Nor can it explain why $H$ changes with flow rate, particle size, or temperature. Plate theory is bookkeeping; rate theory is mechanism. Rate theory reopens the column and asks what physically happens to a solute molecule as it migrates. It dodges through irregular channels between particles. It jitters by ordinary molecular diffusion along the flow axis. It must also repeatedly hop between the moving mobile phase and the stationary phase. This hop takes a finite, non-zero time. Each of these three processes independently smears the band. Remarkably, their variances simply add. This single addition rule is the seed from which the whole quantitative theory of column efficiency grows. It is the direct ancestor of the van Deemter equation. This is the working formula every separation scientist reaches for when optimizing flow rate. This article builds that kinetic picture from the ground up. It covers the geometry of eddy diffusion in a packed bed, the random walk of longitudinal diffusion, and the coupled two-phase kinetics of mass transfer. Each contribution is derived as a variance-per-unit-length. These are combined into a plate height, and checked against realistic numbers for a real HPLC column. The destination — foreshadowed but not fully unpacked here — is the van Deemter equation itself. It also leads to the practical question of extra-column broadening that plagues real instruments.

Key takeaways

A chromatographic band broadens because of three independent, additive random processes, not one.
Eddy diffusion ( $A$ -term): molecules take different-length paths through the packed bed's random channel network.
Longitudinal diffusion ( $B$ -term): ordinary Fickian diffusion spreads solute along the flow axis, worse at low velocity.
Mass-transfer resistance ( $C$ -term): finite equilibration time between mobile and stationary phases lags some molecules behind others.
Because band-spreading mechanisms are statistically independent, their variances — not their widths — add directly.
Key idea: plate height is a sum of independent variance contributions per unit column length, $H = \sum_i \sigma_i^2/L$ , and this sum is what rate theory calculates term by term.

Where this lives today

Rate theory is not a historical curiosity; it is the working design language of every modern separation platform. Pharmaceutical quality-control laboratories choose particle size, column length, and flow rate for HPLC assays by directly consulting van Deemter-type curves, trading a small loss of theoretical plates for a tenfold reduction in run time during high-throughput drug release testing. Instrument manufacturers engineer ultra-high-performance liquid chromatography (UHPLC) systems around sub-2-µm fully porous or superficially porous ("core-shell") particles precisely because rate theory predicts — and experiment confirms — that shrinking $d_p$ suppresses both the eddy and mass-transfer terms simultaneously, flattening the $H$ -vs- $u$ curve so that analysts can run far faster without sacrificing resolution. In biopharmaceutical characterization, monoclonal antibody separations by size-exclusion or ion-exchange chromatography are dominated by the mass-transfer term because these large biomolecules diffuse slowly ( $D_M$ is small, scaling roughly as $M^{-1/3}$ to $M^{-1/2}$ ), which is why protein separations are run at markedly lower optimal linear velocities than small-molecule separations — a direct, quantitative consequence of the $B/u$ and $Cu$ balance derived here. Environmental and forensic laboratories running gas chromatography–mass spectrometry rely on the same theory in reverse: because gas-phase diffusion coefficients are enormous compared to liquids, the longitudinal-diffusion term dominates GC van Deemter curves, which is why capillary GC columns are operated at comparatively high linear velocities and why helium or hydrogen carrier gases (higher diffusivity, flatter high-velocity branch) are preferred over slower nitrogen when speed matters. Column and instrument manufacturers also use rate theory to diagnose and design against extra-column effects — dead volume in injectors, detectors, and connecting tubing that adds its own Gaussian variance to the true column-generated band, a topic that becomes essential once instrument hardware, not just the packed bed, must be optimized for the narrow peaks that modern small-particle columns can, in principle, deliver.

Ask the tutor: How do these three terms combine graphically into the van Deemter curve, and how do real instruments locate the practical optimum flow rate?

The physical picture

Picture an idealized packed column: uniform spherical particles, diameter $d_p$ , packed into a tube through which mobile phase flows at linear velocity $u$ . A pulse of solute is injected as a plane at the inlet. As it moves down the column, three physically distinct things happen to it simultaneously, and each one turns the initially infinitesimally thin plane into a wider distribution. First, eddy diffusion (also called multipath or "A-term" broadening). The packed bed is not a bundle of straight parallel tubes; it is a random three-dimensional maze of interstitial channels around roughly spherical particles. Two solute molecules starting side by side can take routes of noticeably different length and different local velocity — one squeezes through a narrow gap and gets briefly retarded, another finds an open channel and shoots ahead. Purely from the geometry of the packing, some molecules statistically travel a longer path than others in the same elapsed time, producing spreading that in the simplest model is independent of how fast the mobile phase moves. Second, longitudinal (axial) molecular diffusion ("B-term"). Solute is initially a sharp concentration spike; concentration gradients on either side of it drive ordinary diffusion, spreading the band forward and backward along the direction of flow, exactly as a dissolved dye disperses in a stationary beaker of solvent, just superimposed on bulk flow. Because it is diffusion happening over a fixed transit time, this term gets worse the longer the molecule sits in the column — i.e., worse at low flow velocity. Third, resistance to mass transfer ("C-term"). A molecule does not equilibrate between mobile and stationary phase infinitely fast. It takes real time to diffuse into stagnant mobile phase trapped in particle pores ( $C_M$ , mobile-phase mass transfer) and real time to partition into and back out of the stationary film or bonded phase ( $C_S$ , stationary-phase mass transfer). While one molecule is momentarily "parked" in the stationary phase, others sweep past it in the moving stream; when it re-emerges it is now behind molecules that never paused. This lag is a kinetic effect, and unlike the A-term it gets worse at high flow velocity, because the mobile phase does not wait for slow equilibration to finish.

Figure

Three independent broadening mechanisms superimposed on a packed-bed cross-section

The crucial statistical fact that ties these three pictures together is that each mechanism acts as an independent random process superimposed on the same flowing band. When independent random variables add, it is their variances, $\sigma^2$ , that add — not their standard deviations, and not their widths measured in any other way. This is exactly the Central Limit Theorem machinery already implicit in treating the eluting peak as Gaussian in plate theory; rate theory simply asks what physical process contributes how much variance per unit length traveled.

Figure

Variance addition: independent broadening mechanisms combine in quadrature

The core ideas

The observed band variance per unit length traveled is the plate height, and rate theory decomposes it additively:

H = H_{\text{eddy}} + H_{\text{diffusion}} + H_{\text{mass transfer}}

where:

$H$ — plate height (cm), $H = \sigma_L^2/L$ with $\sigma_L^2$ the band variance in length units and $L$ the column length
each term on the right is the variance contribution per unit column length of one physical mechanism

Reading: total plate height is a sum, not a maximum or a product — every mechanism contributes independently and simultaneously, and none can be "turned off" by suppressing another.

Eddy diffusion. In the simplest (Van Deemter) treatment, packing irregularity contributes a plate-height term independent of velocity:

H_{\text{eddy}} = 2\lambda d_p

where:

$\lambda$ — dimensionless packing-quality factor (smaller for more uniform, well-packed beds; typically $0.5$ – $1$ for good HPLC packings)
$d_p$ — average particle diameter (cm or µm)

Reading: tighter, more uniform, smaller particles give a smaller, flow-independent contribution to plate height — this is the physical reason sub-2-µm UHPLC particles are so much more efficient than 5-µm HPLC particles even before considering the other terms.

Longitudinal diffusion. Molecular diffusion in the mobile phase, acting over the residence time $t = L/u$ , spreads the band according to the Einstein relation $\sigma^2 = 2D_M t$ . Dividing by $L$ and using $t=L/u$ gives a term that shrinks as velocity increases:

H_{\text{diffusion}} = \frac{2\gamma D_M}{u}

where:

$\gamma$ — obstruction factor (dimensionless, $<1$ ) accounting for the tortuous path diffusion must take around packed particles (typically $0.6$ – $0.8$ )
$D_M$ — solute diffusion coefficient in the mobile phase (cm² s⁻¹)
$u$ — linear mobile-phase velocity (cm s⁻¹)

Reading: the faster the mobile phase moves, the less time diffusion has to act, so this term is large at low $u$ and vanishes as $u\to\infty$ — the classic reason very slow flow rates give broad, low-efficiency peaks.

Mass-transfer resistance. Finite equilibration kinetics between phases contribute a term that grows with velocity, because faster flow means the mobile phase advances more per unit equilibration time:

H_{\text{mass transfer}} = C\,u, \qquad C = C_S + C_M

where:

$C_S$ — stationary-phase mass-transfer resistance coefficient (s), reflecting the finite rate of diffusion/partitioning into the stationary film or bonded phase
$C_M$ — mobile-phase mass-transfer resistance coefficient (s), reflecting diffusion into and out of stagnant mobile phase in particle pores
$u$ — linear velocity (cm s⁻¹)

Reading: raise the flow rate to save time, and this term inflates band width linearly — the fundamental trade-off between speed and resolution in every chromatographic method. Putting the three together gives the van Deemter-type sum

H = 2\lambda d_p + \frac{2\gamma D_M}{u} + (C_S+C_M)\,u

which already displays the qualitative shape — flat/constant $A$ , a $1/u$ branch, and a linear-in- $u$ branch — that produces the characteristic U-shaped $H$ -vs- $u$ curve central to optimizing chromatographic flow rate.

Derivation ALongitudinal diffusion contributes

2\gamma D_M/u

to plate height

Goal. Starting from Fick's laws applied to a moving band, show that ordinary axial molecular diffusion contributes a plate-height term inversely proportional to linear velocity.

Step 1 — diffusion of an initially sharp band. Model the injected solute as a plane source at $x=0,\,t=0$ in a frame co-moving with the mobile phase at velocity $u$ . In that moving frame, only diffusion acts, governed by Fick's second law,

\frac{\partial C}{\partial t} = D_M \frac{\partial^2 C}{\partial x^2}

The solution to this equation for a point (plane) source is the time-dependent Gaussian

C(x,t) = \frac{n_0}{\sqrt{4\pi D_M t}}\exp\!\left(-\frac{x^2}{4D_M t}\right)

This is precisely the fundamental solution (Green's function) of the diffusion equation, and by inspection its variance is $\sigma_x^2 = 2D_M t$ — the Einstein–Smoluchowski relation for one-dimensional diffusion.

Step 2 — introduce tortuosity. Diffusion inside a packed bed does not proceed through open solvent; it must wind around particles, lengthening the effective diffusive path relative to free solution. This is captured empirically by multiplying the free-solution diffusion coefficient by an obstruction factor $\gamma<1$ :

\sigma_x^2 = 2\gamma D_M t

Step 3 — convert residence time to velocity. A solute band spends time $t = L/u$ traversing a column of length $L$ at linear velocity $u$ (for an unretained or lightly retained species this is a fair first approximation; for retained solutes one uses the appropriate retention time, but the scaling with $u$ is unchanged). Substituting,

\sigma_x^2 = \frac{2\gamma D_M L}{u}

Step 4 — convert variance to plate height. By definition $H \equiv \sigma_x^2/L$ , so dividing through by $L$ isolates the per-unit-length contribution:

H_{\text{diffusion}} = \frac{2\gamma D_M}{u}

Check — limiting case. As $u\to\infty$ (instantaneous transit), $H_{\text{diffusion}}\to 0$ : diffusion has no time to act, matching the physical expectation that fast flow "outruns" diffusion. As $u\to 0$ , $H_{\text{diffusion}}\to\infty$ , correctly predicting catastrophic peak broadening for a column left standing with stopped flow — a well-known practical hazard.

Go deeper · Derivation B · 2 minMass-transfer resistance contributes a term linear in velocity

Derivation BMass-transfer resistance contributes a term linear in velocity

Goal. Show, via a simplified two-site (mobile/stationary) kinetic exchange model, that finite-rate partitioning produces a plate-height contribution proportional to $u$ .

Step 1 — set up the exchange kinetics. Treat the solute as continuously exchanging between the moving mobile phase and a stationary phase (or stagnant pore fluid) with first-order forward and reverse rate constants $k_1$ (mobile→stationary) and $k_{-1}$ (stationary→mobile). A molecule that instantaneously equilibrated would never lag the moving average; delay arises because the sojourn time in the stationary phase, of order $\tau \sim 1/k_{-1}$ , is finite.

Step 2 — relate sojourn time to displacement lag. While parked in the stationary phase for time $\tau$ , the mobile phase (and molecules currently in it) advances a distance $u\tau$ relative to the parked molecule. This lag, accumulated over many random excursions in and out of the stationary phase during the total residence time $t_R$ , behaves as a random walk in displacement with a single-event variance contribution scaling as $(u\tau)^2$ per excursion, and a number of excursions during transit scaling as $t_R/\tau$ .

Step 3 — combine into a variance. The accumulated positional variance from many independent excursions of this kind scales as

\sigma_x^2 \sim (u\tau)^2 \times \frac{t_R}{\tau} = u^2 \tau\, t_R

Using $t_R \propto L/u$ (transit scales inversely with velocity for fixed column length) gives

\sigma_x^2 \sim u^2 \tau \cdot \frac{L}{u} = u\,\tau L

Step 4 — convert to plate height and identify the coefficient. Dividing by $L$ ,

H_{\text{mass transfer}} = \tau u \equiv C u

where the full theoretical expressions (obtainable from rigorous stochastic or moment-based treatments, e.g. Giddings' generalized non-equilibrium theory) resolve $C$ into a stationary-phase term depending on the diffusion coefficient in the stationary film, $D_S$ , and film/pore geometry, and a mobile-phase term depending on $D_M$ and particle geometry:

C_S \propto \frac{d_f^2}{D_S}, \qquad C_M \propto \frac{d_p^2}{D_M}

with $d_f$ the stationary-phase film thickness and $d_p$ the particle diameter — thinner films, smaller particles, and faster intraparticle/intrafilm diffusion all reduce $C$ .

Check — limiting case. As equilibration becomes infinitely fast, $\tau \to 0$ , so $C\to 0$ and $H_{\text{mass transfer}}\to 0$ for any $u$ : an idealized instantaneously-equilibrating column has no kinetic broadening at all, recovering the equilibrium picture implicit in simple plate theory. Conversely, for a fixed nonzero $\tau$ , $H_{\text{mass transfer}}$ grows without bound as $u\to\infty$ , correctly predicting that pushing flow rate arbitrarily high eventually ruins efficiency no matter how good the packing.

Worked examples

Example 1 — Foundational: computing each term and the total plate height at one velocity. A packed HPLC column has particles of diameter $d_p = 5.0\ \mu\text{m} = 5.0\times10^{-4}\ \text{cm}$ , packing factor $\lambda = 0.75$ , obstruction factor $\gamma = 0.70$ , solute diffusion coefficient in the mobile phase $D_M = 1.0\times10^{-5}\ \text{cm}^2\,\text{s}^{-1}$ , and a combined mass-transfer coefficient $C = C_S+C_M = 0.012\ \text{s}$ . The linear velocity is $u = 0.20\ \text{cm s}^{-1}$ . Find $H$ and the number of plates for a 15.0 cm column. Eddy term:

H_{\text{eddy}} = 2\lambda d_p = 2(0.75)(5.0\times10^{-4}\ \text{cm}) = 7.5\times10^{-4}\ \text{cm}

Diffusion term:

H_{\text{diffusion}} = \frac{2\gamma D_M}{u} = \frac{2(0.70)(1.0\times10^{-5}\ \text{cm}^2\text{s}^{-1})}{0.20\ \text{cm s}^{-1}} = 7.0\times10^{-5}\ \text{cm}

Mass-transfer term:

H_{\text{mass transfer}} = Cu = (0.012\ \text{s})(0.20\ \text{cm s}^{-1}) = 2.4\times10^{-3}\ \text{cm}

Total:

H = 7.5\times10^{-4} + 7.0\times10^{-5} + 2.4\times10^{-3} = 3.17\times10^{-3}\ \text{cm}

N = \frac{L}{H} = \frac{15.0\ \text{cm}}{3.17\times10^{-3}\ \text{cm}} = 4.73\times10^{3}\ \text{plates}

Notice that at this velocity the mass-transfer term already dominates (about 76% of $H$ ); the eddy term is next (24%), and diffusion is nearly negligible (2%) — a pattern typical of HPLC run at reasonably fast flow, in contrast to GC or very slow HPLC, where the diffusion term matters much more.

Example 2 — Applied: finding the optimum velocity and minimum plate height. Using the same column ( $\lambda=0.75$ , $d_p = 5.0\times10^{-4}$ cm, $\gamma=0.70$ , $D_M = 1.0\times10^{-5}\ \text{cm}^2\text{s}^{-1}$ , $C = 0.012\ \text{s}$ ), find the flow velocity $u_{\text{opt}}$ that minimizes $H$ , and the corresponding minimum plate height $H_{\min}$ . Write $H(u) = A + B/u + Cu$ with $A = 7.5\times10^{-4}\ \text{cm}$ and $B = 2\gamma D_M = 1.4\times10^{-5}\ \text{cm}^2\text{s}^{-1}$ . Minimize by setting $dH/du = 0$ :

\frac{dH}{du} = -\frac{B}{u^2} + C = 0 \quad\Longrightarrow\quad u_{\text{opt}} = \sqrt{\frac{B}{C}}

u_{\text{opt}} = \sqrt{\frac{1.4\times10^{-5}\ \text{cm}^2\text{s}^{-1}}{0.012\ \text{s}}} = \sqrt{1.167\times10^{-3}\ \text{cm}^2\text{s}^{-2}} = 0.0342\ \text{cm s}^{-1}

This optimum velocity is considerably slower than the 0.20 cm/s used operationally in Example 1 — a common real-world situation, since chemists usually sacrifice some theoretical efficiency for the sake of shorter run times. Evaluate $H$ at $u_{\text{opt}}$ :

H_{\min} = A + 2\sqrt{BC} = 7.5\times10^{-4} + 2\sqrt{(1.4\times10^{-5})(0.012)}

\sqrt{(1.4\times10^{-5})(0.012)} = \sqrt{1.68\times10^{-7}} = 4.10\times10^{-4}\ \text{cm}

H_{\min} = 7.5\times10^{-4} + 2(4.10\times10^{-4}) = 7.5\times10^{-4}+8.20\times10^{-4} = 1.57\times10^{-3}\ \text{cm}

For the same 15.0 cm column, $N_{\max} = L/H_{\min} = 15.0/1.57\times10^{-3} = 9.55\times10^{3}$ plates — roughly double the plate count obtained at the faster, more practical velocity of Example 1, illustrating the classic speed/efficiency trade-off quantified by this minimum.

Example 3 — Exam-level: back-calculating a mass-transfer coefficient from two experimental $(u,H)$ pairs, then predicting resolution loss at high flow. A chromatographer measures plate height at two velocities on a column with known $A = 6.0\times10^{-4}$ cm and known $B = 1.6\times10^{-5}\ \text{cm}^2\text{s}^{-1}$ (from independent diffusion-coefficient and obstruction-factor data). At $u_1 = 0.10\ \text{cm s}^{-1}$ , the measured plate height is $H_1 = 2.10\times10^{-3}$ cm. Determine $C$ , then predict $H$ at $u_2 = 0.50\ \text{cm s}^{-1}$ , and finally state the fractional loss in plate count between the two velocities for a 25.0 cm column. Step 1: solve for $C$ from the first data point.

H_1 = A + \frac{B}{u_1} + Cu_1

2.10\times10^{-3} = 6.0\times10^{-4} + \frac{1.6\times10^{-5}}{0.10} + C(0.10)

2.10\times10^{-3} = 6.0\times10^{-4} + 1.60\times10^{-4} + 0.10\,C

2.10\times10^{-3} - 7.60\times10^{-4} = 0.10\,C \quad\Longrightarrow\quad 1.34\times10^{-3} = 0.10\,C

C = 1.34\times10^{-2}\ \text{s}

Step 2: predict $H_2$ at $u_2 = 0.50$ cm/s.

H_2 = 6.0\times10^{-4} + \frac{1.6\times10^{-5}}{0.50} + (1.34\times10^{-2})(0.50)

H_2 = 6.0\times10^{-4} + 3.20\times10^{-5} + 6.70\times10^{-3} = 7.33\times10^{-3}\ \text{cm}

Step 3: plate counts and fractional loss.

N_1 = \frac{25.0}{2.10\times10^{-3}} = 1.190\times10^{4}, \qquad N_2 = \frac{25.0}{7.33\times10^{-3}} = 3.411\times10^{3}

\text{fractional loss} = \frac{N_1-N_2}{N_1} = \frac{1.190\times10^{4}-3.411\times10^{3}}{1.190\times10^{4}} = 0.713

Running the column five times faster costs about 71% of the plate count — a dramatic illustration of why the $Cu$ term, not the eddy or diffusion terms, ultimately governs high-throughput/fast-LC method development, and why real methods are always a negotiated compromise between $u_{\text{opt}}$ and acceptable analysis time. Since $N\propto R_s^2$ at fixed selectivity and retention, this loss in plates corresponds to a resolution scaled down by roughly $\sqrt{N_2/N_1}=\sqrt{0.287}\approx0.536$ — resolution falls to about 54% of its low-flow value.

Limiting cases, scope & common traps

Regime / limit	What happens to $H$	Physical reason
$u \to 0$ (very slow flow, or stopped-flow)	$H \to \infty$ (dominated by $B/u$ )	Longitudinal diffusion has unlimited time to act
$u \to \infty$ (very fast flow)	$H \to \infty$ (dominated by $Cu$ )	Mass transfer cannot keep pace; equilibrium lag grows without bound
$u = u_{\text{opt}} = \sqrt{B/C}$	$H$ at global minimum, $H_{\min}=A+2\sqrt{BC}$	Diffusion loss and mass-transfer loss exactly balance
$d_p \to 0$ (very small particles, e.g. UHPLC sub-2-µm)	$A$ and $C_M$ both shrink	Shorter diffusion paths in eddies and pores
Perfectly uniform packing ( $\lambda\to 0$ )	$A\to 0$ , only $B/u+Cu$ remain	No multipath effect if every channel is identical
Instantaneous equilibration ( $C\to 0$ )	Mass-transfer term vanishes; recovers ideal plate-theory limit	No kinetic lag between phases
Gas chromatography vs. HPLC	$D_M$ in GC is ~10⁴ times larger than in liquids	$B$ -term dominates GC optimum curves; optimum $u$ shifts higher

Common trapCommon misconceptions

Two errors recur constantly. First, students sometimes add plate-height contributions as standard deviations ( $H_{\text{total}} = H_1+H_2+H_3$ is correct for variance-equivalent $H$ values only because $H\equiv\sigma^2/L$ is already a variance-per-length; do not additionally take square roots anywhere in combining the three terms — the formula is already variance-additive by construction). Second, many assume smaller particles only help through the eddy term; in fact reducing $d_p$ shrinks both $A$ (shorter multipath channels) and $C_M$ (shorter intraparticle diffusion distances), which is why modern sub-2-µm UHPLC columns show dramatically flatter, lower van Deemter curves rather than merely a lower plateau.

The chemist behind it

The kinetic (rate) theory of chromatography, superseding the purely equilibrium-stage plate model, emerged from a collaboration between a physical chemist trained in reaction kinetics and statistical mechanics and colleagues applying random-walk and stochastic-process mathematics to separations. The 1956 paper by J. J. van Deemter, F. J. Zuiderweg, and A. Klinkenberg, written at Royal Dutch Shell's laboratories while studying gas–liquid chromatography for petroleum analysis, first assembled the eddy-diffusion, longitudinal-diffusion, and mass-transfer terms into the compact $A + B/u + Cu$ form that bears van Deemter's name. Their achievement was to recognize that a packed column could be treated not as a cascade of ideal equilibrium stages, but as a continuous kinetic system, borrowing the mathematics of random walks and diffusion that had already matured in statistical physics.

Origins · 1 minJan Jozef van Deemter and how this was discovered

Jan Jozef van Deemter

1918-1985 · Industrial physical chemist, Royal Dutch Shell, applying kinetic and stochastic theory to chromatographic separations in the petroleum industry

History

- 1941 — A. J. P. Martin and R. L. M. Synge propose the plate model as an equilibrium-stage analogy from distillation theory, for which they later share the Nobel Prize.

- 1952 — Martin and A. T. James introduce practical gas–liquid chromatography, exposing clear velocity-dependent efficiency effects unexplained by the static plate model.

- 1956 — van Deemter, Zuiderweg, and Klinkenberg publish the rate-theory equation combining eddy diffusion, longitudinal diffusion, and mass-transfer resistance terms.

- 1965 — J. Calvin Giddings develops a more rigorous "coupled" (generalized non-equilibrium) theory refining the eddy-diffusion term and unifying random-walk and rate-based descriptions.

- 1977 onward — Csaba Horváth and coworkers extend rate theory quantitatively to liquid chromatography and high-performance packings, setting the stage for the small-particle, high-efficiency columns of modern HPLC and UHPLC.

Why do we add variances instead of peak widths when combining broadening mechanisms?

Peak width (for example, width at half-height or baseline width) is a linear measure of spread, but the statistical processes causing spreading — random path differences, random diffusive steps, random equilibration delays — are each modeled as independent Gaussian-generating processes. A fundamental property of independent random variables is that their variances add, not their standard deviations; this is a consequence of how variances of sums of independent variables behave under the Central Limit Theorem. Since plate height is defined as $H=\sigma^2/L$ , a variance-per-length quantity, adding plate-height terms directly and correctly reflects the underlying variance addition. Adding widths (or standard deviations) directly would overestimate the true combined spread, because it ignores the way independent random fluctuations partially "cancel" in aggregate rather than stacking linearly.

How does rate theory relate to the plate number N from simple plate theory?

Plate theory treats the column as $N$ discrete equilibrium stages and defines $H=L/N$ purely operationally, from the observed retention time and peak width, without explaining why $H$ has the value it does. Rate theory supplies the missing mechanism: it derives $H$ from the physical processes of eddy diffusion, longitudinal diffusion, and mass-transfer kinetics, each contributing an independent, calculable variance per unit length. The two frameworks are fully consistent — the $H$ measured experimentally from a real Gaussian peak using plate-theory formulas is the same $H$ that the van Deemter-type sum predicts from first principles — but rate theory explains why $N$ depends on flow rate, particle size, and diffusion coefficients, which plate theory alone cannot do.

Why does the mass-transfer term get worse at high flow rate but the diffusion term gets better?

These two terms represent competing effects of time spent in the column. Longitudinal diffusion is a spreading process that needs time to act; less time (higher velocity) means less diffusive spreading, so the $B/u$ term shrinks as $u$ grows. Mass-transfer resistance, in contrast, is a lag effect — a molecule momentarily immobilized in the stationary phase or a stagnant pore falls behind the bulk flow, and the faster the bulk flow moves during that fixed equilibration time, the further behind it falls. Because the moving-phase displacement per equilibration event scales directly with $u$ , the $Cu$ term grows with velocity. The opposing trends are exactly what produces a U-shaped $H$ -versus- $u$ curve with a well-defined minimum.

Does eddy diffusion depend on flow velocity at all?

In the simplest van Deemter treatment, no — the $A$ term $2\lambda d_p$ is written as velocity-independent, reflecting only packing geometry. However, more rigorous treatments (notably Giddings' coupled theory) recognize that eddy diffusion and mobile-phase mass transfer are not fully separable processes: molecules taking a "slow" channel can, given enough time, diffuse laterally into a "fast" channel, partially averaging out the path-length differences. This coupling makes the effective eddy contribution mildly velocity-dependent at very low or very high flow, blending smoothly into the mass-transfer term rather than remaining a strictly flat plateau. For introductory purposes, treating $A$ as constant is an excellent approximation across the normal operating range.

Why do smaller particles reduce both the A and C terms, and is there a limit to how small particles can go?

Smaller particles shrink the characteristic length scale of the packing's random channel network, directly reducing the spread of path lengths responsible for eddy diffusion, and they also shrink the diffusion distances solute molecules must travel into and out of stagnant intraparticle pore fluid, directly reducing $C_M$ . Both effects lower $H$ , which is why UHPLC columns built from sub-2-µm particles achieve dramatically higher efficiency per unit length than traditional 5-µm HPLC packings. The practical limit is not chemistry but engineering: smaller particles pack more densely and resist flow far more strongly, requiring the column to be driven at very high back-pressure, which in turn demands specialized pumps, fittings, and detectors capable of withstanding pressures well above traditional HPLC limits.

Is the Gaussian shape of a chromatographic peak a consequence of rate theory or of plate theory?

Both frameworks predict a Gaussian shape, but for related though distinct reasons. Plate theory derives Gaussian character as the limiting binomial/Poisson distribution of a solute redistributing across many discrete equilibrium stages, becoming Gaussian by a discrete-limit argument as the number of stages grows large. Rate theory instead derives Gaussian character from the physical mechanisms themselves — molecular diffusion intrinsically produces a Gaussian solution to Fick's second law, and the random-walk excursions underlying mass-transfer lag likewise converge to a Gaussian distribution via the Central Limit Theorem. The two pictures converge on the same observable peak shape because they describe the same underlying random process from two different mathematical vantage points.

Why is the optimum velocity often not used in real, high-throughput methods?

The optimum velocity u_opt minimizes plate height and therefore maximizes efficiency for a given column, but it is frequently quite slow, and slow flow means long analysis times. In pharmaceutical quality control, environmental screening, and other high-throughput settings, laboratories deliberately operate above u_opt, accepting a modest loss of theoretical plates in exchange for a much shorter run time, provided the resulting resolution still clears the required specification. Method developers use the shape of the H-versus-u curve itself to choose this compromise deliberately: because the curve rises only gradually just above u_opt but very steeply at very high velocity, there is often a "sweet spot" noticeably faster than u_opt that sacrifices little efficiency while saving substantial time.

Rate Theory and Band Broadening

Where this lives today

The physical picture

The core ideas

Worked examples

Limiting cases, scope & common traps

The chemist behind it

Before this

Where this leads