Optimizing Efficiency and Flow Rate

Pharmaceutical quality control is the most immediate home of this optimization: a compendial HPLC assay method must resolve an active pharmaceutical ingredient from its process impurities and degradation products to a specified $R_s$, usually within a target run time set by production throughput, and method validation reports routinely include the $\alpha$-versus-$N$-versus-$k$ trade-off analysis explicitly, choosing mobile-phase pH or organic-modifier content to hit a target $\alpha$ before ever specifying particle size or flow rate. Biopharmaceutical characterization — separating closely related glycoforms or charge variants of a monoclonal antibody by ion-exchange or hydrophobic-interaction chromatography — depends almost entirely on selectivity engineering (small changes in salt gradient slope shifting $\alpha$) precisely because $N$ gains from particle size are comparatively weak for these large, slowly diffusing biomolecules, whose $B$ and $C$ terms are dominated by very small diffusion coefficients. UHPLC instrumentation itself is a direct commercial embodiment of the kinetic performance limit calculation worked in Example 3: pump manufacturers rate systems to specific pressure ceilings (400, 600, 1000, 1300 bar) precisely because that ceiling, combined with a target analysis time, sets the particle size that maximizes achievable plate count — the calculation is not academic, it is literally how instrument specifications are chosen. Superficially porous ("core-shell") particles achieve much of the small-particle efficiency benefit (reduced $C$-term, since diffusion path length into a thin porous shell is shorter than into a fully porous particle) while keeping backpressure closer to that of conventional fully-porous 3–5 µm particles, directly exploiting the $A$/$C$-term particle-size scaling derived above without paying the full $1/d_p^2$ pressure penalty. Environmental and forensic laboratories running high sample throughput on GC and LC systems use exactly this framework to decide between a long, high-efficiency column (better separation, slower per-sample time) and a short, fast column relying more on selectivity (faster per-sample time, adequate separation) — a direct, everyday instance of the $\sqrt{N}$-versus-run-time trade quantified in Example 2. Looking further, once $N$, $\alpha$, and $k$ have each been pushed close to their practical limits for an isocratic or isothermal run, the next lever chromatographers reach for is making $k$ and $\alpha$ vary during the run — Gradient and Temperature Programming — which lets early peaks elute at low $k$ (fast) while late peaks still get enough retention to resolve, escaping the single-run-condition constraint that this article's optimization has assumed throughout.

Ask the tutor: If I have a fixed pump pressure ceiling and a fixed maximum analysis time, how do I actually solve for the single best particle diameter, rather than just comparing two candidates as in the worked example?

Resolution between two adjacent peaks is the master output variable of any chromatographic method, and it is built from three factors that are, to good approximation, separable:

Each factor lives in a different physical domain. The efficiency term $\sqrt{N}$ depends on how well the column packs, how uniform the flow paths are, and how fast solutes diffuse — the terrain of the van Deemter equation. The selectivity term depends on thermodynamics: differences in partition coefficients between the two solutes, governed by the stationary-phase chemistry, temperature, and mobile-phase composition. The retention term depends on how long the solute dwells in the stationary phase relative to the mobile phase, governed mostly by solvent strength. These are not fully independent in practice — changing mobile-phase composition to shift $k$ often nudges $\alpha$ too — but treating them as separable is the standard first-order model and it is quantitatively very useful. The efficiency term is where flow rate enters, through the plate height $H = L/N$ and the van Deemter relationship:

Here $A$ is eddy diffusion (path heterogeneity in the packed bed), $B/u$ is longitudinal (axial) molecular diffusion, and $Cu$ is resistance to mass transfer between phases. At low velocity, diffusion dominates and $H$ is large; at high velocity, mass-transfer lag dominates and $H$ grows again. There is a minimum $H$ — hence maximum $N$ per unit length — at some optimum velocity $u_{opt}$. This curve is the terrain on which flow-rate optimization is fought, and it shifts favorably (lower minimum $H$, higher $u_{opt}$) as particle diameter shrinks, because both $A$ and $C$ scale with $d_p$ or $d_p^2$. The practical tension is this: running at $u_{opt}$ gives maximum efficiency per column length, but analysis time is $t_R = t_0(1+k) = (L/u)(1+k)$, and pushing $u$ well above $u_{opt}$ sacrifices only a little $N$ (the $Cu$ branch is fairly flat with smaller particles) while saving a lot of time. Modern method development therefore rarely runs exactly at the van Deemter minimum; it runs somewhat above it, deliberately, because the efficiency curve is asymmetric — steep on the low-$u$ side, shallow on the high-$u$ side.

Particle size shifts this whole picture. Reduced-variable theory shows that the C-term contribution to plate height scales roughly as $d_p^2$ while the eddy term scales as $d_p$, so shrinking $d_p$ compresses the entire curve horizontally and vertically: $u_{opt}$ moves to higher velocity and $H_{min}$ drops. That is the entire physical justification for sub-2-µm UHPLC particles — not simply "smaller is better" as a slogan, but a specific, calculable statement about where the minimum sits and how flat the high-velocity branch becomes.

The resolution equation, written to expose all three levers explicitly:

where:

• $N$ = number of theoretical plates (efficiency), $N = L/H = 16(t_R/w)^2$
• $\alpha = k_2/k_1$ = selectivity factor between the two adjacent peaks ($\alpha \ge 1$ by convention)
• $k_2$ = retention factor of the later-eluting peak, $k = (t_R - t_0)/t_0$

Reading: doubling $N$ only multiplies $R_s$ by $\sqrt{2} \approx 1.41$; increasing $\alpha$ from 1.02 to 1.05 multiplies the selectivity term by roughly 2.5; increasing $k$ from 1 to 5 multiplies the retention term by $(5/6)/(1/2) = 1.67$ but increasing $k$ from 5 to 20 only buys a further factor of $(20/21)/(5/6) \approx 1.14$. The retention term saturates; $\alpha$ almost never does over the usable range, which is why chromatographers always try to improve selectivity before anything else. The van Deemter equation governing $N$ through flow rate:

where:

• $H$ = height equivalent to a theoretical plate (HETP), $H = L/N$
• $A$ = eddy-diffusion term (packing quality; $A \approx 2\lambda d_p$), independent of $u$
• $B$ = longitudinal diffusion coefficient term ($B = 2\gamma D_m$), dominates at low $u$
• $C$ = mass-transfer resistance term ($C = C_s + C_m$, stationary- and mobile-phase contributions), dominates at high $u$
• $u$ = linear mobile-phase velocity

Reading: minimizing $H$ with respect to $u$ gives the optimum velocity and minimum plate height; operating away from that minimum is a deliberate, quantifiable choice, not a mistake, provided you know how much efficiency you are giving up. The optimum velocity and minimum plate height, from $dH/du = 0$:

where the symbols are as above. Reading: $u_{opt}$ rises when $B$ (diffusion coefficient, hence temperature) rises or when $C$ (mass-transfer resistance, hence particle size) falls — exactly why small particles let you run fast and efficient simultaneously, and why raising temperature (which raises $D_m$) pushes the optimum to higher flow as well. The pressure drop across a packed column, from the Kozy-Carman/Darcy relation used throughout LC:

where:

• $\phi$ = column flow resistance parameter (typically 500–1000 for packed beds, dimensionless in this form)
• $\eta$ = mobile-phase viscosity
• $u$ = linear velocity, $L$ = column length, $d_p$ = particle diameter

Reading: pressure scales as $1/d_p^2$, so halving particle diameter quadruples the pressure needed to maintain the same velocity — this is the direct cost of the UHPLC efficiency gain, paid in pump and hardware pressure rating rather than in time.

Example 1 (foundational): finding $u_{opt}$ and $H_{min}$ from van Deemter coefficients, then converting to plate count and analysis time. A reversed-phase column, 15.0 cm long, packed with 5.0 µm particles, has been characterized for a test solute with van Deemter coefficients $A = 6.0\times10^{-4}$ cm, $B = 2.0\times10^{-3}$ cm²/s, and $C = 0.020$ s (these are typical magnitudes for a well-packed HPLC column). Find $u_{opt}$, $H_{min}$, the resulting plate count $N$, and the retention time of a solute with $k = 4.0$ and $t_0 = 1.20$ min (i.e., $u$ chosen to equal $u_{opt}$). Step 1 — optimum velocity.

Step 2 — minimum plate height.

Step 3 — plate count. With $L = 15.0$ cm:

Round to 3 sig figs: $N \approx 1.13\times10^3$ plates. (This is a modest plate count, consistent with a short column at 5 µm particle size — realistic for a fast screening method, not a high-resolution one.) Step 4 — void time and retention time. The mobile phase travels the column length at $u_{opt}$, so

Note this is the computed void time consistent with the column and velocity — it need not match an arbitrarily assumed $t_0$; use the physically consistent value here. Then

So operating exactly at the van Deemter optimum on this short 5-µm column gives $N \approx 1130$ plates and elutes the $k=4$ solute at 3.96 min — fast, but with limited efficiency, illustrating why short/wide-particle columns are chosen for speed-focused screening rather than high-resolution separations.

Example 2 (applied): should you switch to smaller particles or a longer column to hit a resolution target? A method currently gives $R_s = 1.05$ between two closely eluting impurities on a 10.0 cm column packed with 5 µm particles, run at $u_{opt}$, where $N = 8000$ plates, $\alpha = 1.06$, and $k_2 = 3.0$. Regulatory guidance requires $R_s \ge 1.5$. Two options are proposed: (a) double the column length to 20.0 cm at the same particle size and velocity, or (b) switch to 3 µm particles in a column re-optimized so that $N$ scales as expected from the derivation above ($H_{min} \propto d_p$), keeping length at 10.0 cm. Which option reaches the target, and at what cost in analysis time? Step 1 — confirm the baseline with the resolution equation.

This does not match the stated 1.05 exactly because real methods include measurement scatter in $N$, $\alpha$, $k$; take the calculation itself, 0.949, as the internally consistent baseline value to compare against for the two options (a small discrepancy between a quoted lab value and a from-first-principles recompute is normal and is exactly why you always recompute from the fundamental quantities rather than trusting a remembered number). Step 2 — option (a), double the length. Since $N \propto L$ at fixed particle size and velocity (same $H$), doubling $L$ doubles $N$: $N_{new} = 16{,}000$. Selectivity $\alpha$ and $k_2$ are unaffected by column length.

This falls short of 1.5. Analysis time doubles too, since $t_0 \propto L$ at fixed $u$: $t_{R,a} = 2 \times t_{R,\text{original}}$. Step 3 — option (b), shrink particles to 3 µm at the same length. From the derivation, $H_{min} \propto d_p$, so at fixed $L = 10.0$ cm,

Still short of 1.5, and by more than option (a) numerically, though it was achieved without lengthening the column (so, per unit time, it is more efficient — see step 4). Step 4 — compare cost in time. Since $u_{opt} \propto 1/d_p$ (derived above), the 3 µm column can also be run about $5/3 = 1.67\times$ faster than the 5 µm column while still sitting at its own optimum, so at the same column length the analysis time actually decreases by roughly that factor rather than increasing. Option (b) gives less resolution gain than doubling length, but it does so while cutting analysis time — a genuinely favorable trade if throughput matters as much as the resolution target. Step 5 — what actually reaches $R_s = 1.5$? Neither single change suffices; solve for the $\alpha$ needed instead, holding $N=8000$, $k_2=3.0$, to see how much smaller a selectivity improvement would need to be to close the gap alone:

Raising $\alpha$ from 1.06 to about 1.098 — a change of less than 4% in selectivity, achievable with a small mobile-phase pH or organic-modifier adjustment — reaches the target with no change in $N$, no time penalty, and no pressure penalty. This is the textbook demonstration of why selectivity is attacked first: a tiny, cheap change in $\alpha$ outperforms an expensive doubling of column length or particle-size overhaul.

Example 3 (exam/competition level): kinetic performance limit — maximum N obtainable at a fixed analysis time and a fixed maximum pressure. A separations engineer has a pump rated to $\Delta P_{max} = 400$ bar and wants the maximum number of theoretical plates obtainable in an analysis time no longer than $t_0 = 3.00$ min (void time), for a solute with diffusion coefficient $D_m = 8.0\times10^{-6}$ cm²/s in the mobile phase, using van Deemter-type coefficients that scale as $A = 2\lambda d_p$ with $\lambda = 1.0$, $B = 2\gamma D_m$ with $\gamma = 0.60$, and $C = C_0 d_p^2/D_m$ with $C_0 = 0.060$ (dimensionless combination consistent with the earlier derivation), viscosity $\eta = 0.90$ cP $= 9.0\times10^{-4}$ Pa·s, and flow-resistance parameter $\phi = 700$. Compare 5 µm and 2 µm particles: which gives more plates within both the time and pressure constraints? Step 1 — express $N$ and $t_0$ in terms of $u$ and $d_p$. From the scaling laws:

and, with column length $L$ free to choose (that is the variable we solve for, subject to the two constraints), $N = L/H$ and $t_0 = L/u$, so

This says: for fixed $t_0$, $N$ is maximized by maximizing $u/H$ — a velocity-efficiency "figure of merit" independent of $L$ (since $L$ is chosen afterward to fit the time budget exactly). Step 2 — apply the pressure constraint to fix the maximum allowed $u$ for each $d_p$, at whatever $L$ is ultimately chosen. Rearranged Darcy's law:

This couples $u$ and $L$; combined with $t_0 = L/u$, eliminate $L = u\,t_0$:

Compute the constant $\sqrt{\Delta P_{max}/(\phi\,\eta\,t_0)}$ in consistent SI units. $\Delta P_{max} = 400\ \text{bar} = 4.0\times10^7$ Pa; $t_0 = 3.00$ min $= 180$ s; $\phi\eta t_0 = 700 \times 9.0\times10^{-4}\ \text{Pa}\,\cdot\,\text{s} \times 180\ \text{s} = 700 \times 0.162 = 113.4\ \text{Pa}\,\cdot\,\text{s}^2$.

So $u_{max} = d_p \times 594.0\ \text{s}^{-1}$, with $d_p$ in meters and $u_{max}$ in m/s. For $d_p = 5.0\ \mu\text{m} = 5.0\times10^{-6}$ m: $u_{max} = 5.0\times10^{-6}\times594.0 = 2.970\times10^{-3}$ m/s $= 0.297$ cm/s. For $d_p = 2.0\ \mu\text{m} = 2.0\times10^{-6}$ m: $u_{max} = 2.0\times10^{-6}\times594.0 = 1.188\times10^{-3}$ m/s $= 0.1188$ cm/s. Step 3 — compute $H$ at each $u_{max}$ using the scaling van Deemter equation, and then $N$. For $d_p = 5.0$ µm $= 5.0\times10^{-4}$ cm, $D_m = 8.0\times10^{-6}$ cm²/s:

Then $N_5 = u_{max}t_0/H_5 = (0.297\ \text{cm/s} \times 180\ \text{s})/1.589\times10^{-3}\ \text{cm} = 53.46/1.589\times10^{-3} = 3.365\times10^4$ plates. For $d_p = 2.0$ µm $= 2.0\times10^{-4}$ cm:

Then $N_2 = (0.1188\ \text{cm/s}\times 180\ \text{s})/5.164\times10^{-4}\ \text{cm} = 21.38/5.164\times10^{-4} = 4.141\times10^4$ plates. Step 4 — compare. $N_2 = 4.14\times10^4$ plates versus $N_5 = 3.37\times10^4$ plates: the 2 µm particle wins, delivering about 23% more plates within the same pressure ceiling and the same analysis time. This is the kinetic performance limit calculation in miniature — it is precisely the reasoning that drove the industry from 5 µm to sub-2-µm particles once pumps capable of the higher pressures became available. Note also that both numbers are far above the modest 1130-plate result of Example 1, because here $L$ is being chosen optimally (long columns, run more slowly than each column's own $u_{opt}$) rather than fixed arbitrarily at 15 cm. Limiting check. If $\Delta P_{max} \to \infty$ (no pressure limit), $u_{max} \to \infty$ for any $d_p$, and $N \to u\,t_0/(Cu) = t_0/C$, independent of $u$ — meaning without a pressure ceiling the optimal strategy is to shrink $d_p$ without bound (since $C \propto d_p^2$, so $t_0/C \to \infty$ as $d_p\to 0$), confirming that it is specifically the pressure constraint, not the van Deemter physics, that sets a finite optimal particle size in real instruments.

Jozef Jan van Deemter (1912–1984) was a Dutch chemical engineer working at Royal Dutch Shell's laboratory in Amsterdam in the 1950s, during the period when gas chromatography was transitioning from a laboratory curiosity into an industrial analytical tool for petroleum fractions. Trained in chemical engineering rather than as a physical chemist by background, van Deemter approached the broadening of chromatographic peaks the way a process engineer approaches any transport problem in a packed bed: as a superposition of independent rate processes — path heterogeneity, longitudinal diffusion, and finite-rate mass transfer between phases — each contributing additively to the total plate height. This engineering instinct, treating a subtle physical-chemistry phenomenon (band broadening) with the same tools used for heat exchangers and packed catalytic reactors, is precisely why the equation bearing his name has survived unmodified in its essential form for seven decades, extended (via reduced-variable theory and later refinements to the C-term for different particle morphologies) but never overturned. The 1956 paper, published with collaborators F. J. Zuiderweg and A. Klinkenberg in Chemical Engineering Science, analyzed gas chromatography specifically, modeling the column as a series of theoretical plates undergoing simultaneous diffusion and non-equilibrium mass transfer. The mathematics — a random-walk argument for eddy diffusion, Einstein's diffusion relation for the $B$-term, and a resistance-in-series argument for the $C$-term — translated with only parameter reinterpretation to liquid chromatography once James Lovelock, Csaba Horváth, and others began the serious quantitative development of LC column theory in the 1960s and 1970s. The equation's real triumph is conceptual: it gave chromatographers a falsifiable, minimizable function relating an experimentally controllable variable (flow rate) to an experimentally measurable outcome (plate height), turning column optimization from folklore into calculus.