Theoretical Plates and Resolution: How Peaks Separate on a Column

The plate model: borrowing a still from the distillation column

The word plate is an inheritance from fractional distillation. In a distillation column, each physical tray (theoretical plate) is a stage where rising vapor and descending liquid reach equilibrium; the more plates, the sharper the separation between components of similar boiling point. In 1941, Archer Martin and Richard Synge — inventing partition chromatography in the process — imported this bookkeeping wholesale, treating a chromatographic column as a stack of hypothetical segments in each of which the analyte fully equilibrates between the mobile and stationary phases before moving on. Their paper (Biochem. J. 1941) earned them the 1952 Nobel Prize in Chemistry.

The crucial caveat, which every analytical chemist should internalize, is that these plates do not exist. Real chromatography never reaches equilibrium — the mobile phase is always sweeping analyte forward faster than partitioning can keep up. The theoretical plate is a mathematical abstraction: the height equivalent to a theoretical plate (HETP, symbol H) is simply defined so that a column of length L behaves as if it contained N = L/H perfectly-equilibrated stages. It is, honestly, a fudge factor that packages all the messy band-broadening physics into a single number.

Its value comes from what it lets you compute. A Gaussian elution peak has a variance σ² that grows linearly with migration distance. If the column contributes a constant plate height H, then σ_x² (in length units) = H·L, and the number of plates N = L/H = (L/σ_x)², where σ_x is the spatial band width. Because a peak eluting at retention time t_R has a temporal standard deviation τ = t_R·(σ_x/L), this rearranges into the working formulas below — the entire practical power of plate theory in one substitution.

Counting plates: N from peak width, and why Gaussian peaks are 4σ wide

Every chromatographer measures N the same way — from a peak on a chromatogram. For a symmetric Gaussian peak the standard forms are:

• N = 16 (t_R / w_b)² — using the baseline width w_b, obtained from tangents drawn to the inflection points, which spans 4σ.
• N = 5.54 (t_R / w₁/₂)² — using the width at half height w₁/₂, which spans 2.355σ; the constant is 8·ln2 ≈ 5.545.

The factor of 16 is not arbitrary: since w_b = 4σ (in time units, 4τ), then (t_R/w_b)² = (t_R/4τ)², and N = (t_R/τ)² gives N = 16(t_R/w_b)². The half-height version is preferred in practice because the baseline tangent method is error-prone on tailing peaks. For asymmetric peaks, laboratories increasingly report the USP tangent or the Foley–Dorsey equation, N = 41.7·(t_R/w₀.₁)² / (b/a + 1.25), which uses the 10%-height width and the asymmetry ratio b/a to correct for tailing.

A concrete count: suppose a compound elutes at t_R = 8.00 min with a half-height width of 0.100 min. Then N = 5.54·(8.00/0.100)² = 5.54·6400 = 35,500 plates. On a 15 cm column, H = L/N = 150 mm / 35,500 = 4.2 μm — a plate height of about twice the particle diameter for a 2 μm packing, which is roughly the theoretical floor. That single H value is the honest report card of the column's kinetic quality; the closer H/dₚ approaches ~2, the better packed and better operated the column.

Van Deemter and Golay: why plate height depends on flow rate

Plate theory tells you how many plates you have but is silent on why. That gap was filled by the rate theory, crystallized in the 1956 paper of J. J. van Deemter, F. J. Zuiderweg, and A. Klinkenberg (Chem. Eng. Sci.). They expressed plate height as a function of the mobile-phase linear velocity u:

H = A + B/u + C·u

Each term is a distinct broadening mechanism:

• A — eddy diffusion (multipath): analyte molecules take different-length paths through the packed bed. A ≈ 2λdₚ, scaling with particle diameter and packing quality λ. It is flow-independent in the simplest treatment.
• B/u — longitudinal (axial) molecular diffusion: the band spreads along the column axis by ordinary diffusion; B = 2γDₘ, where Dₘ is the analyte's mobile-phase diffusion coefficient and γ the obstruction factor. This term dominates at low flow — go too slow and the peak broadens by sitting still. It is the term that matters most in GC.
• C·u — resistance to mass transfer: finite time is needed to diffuse into and out of the stationary phase (and stagnant mobile phase in pores); the faster you flow, the further the band travels before equilibrating, so C·u grows linearly with u and dominates at high flow.

Because B/u falls and C·u rises with velocity, H passes through a minimum at u_opt = √(B/C), giving the fastest possible plates. For open-tubular capillary GC columns there is no packing, so the A term vanishes and the Golay equation (Marcel Golay, 1958) applies, with the C term split into gas-phase (C_M) and stationary-film (C_S) contributions. This is precisely why capillary columns achieve H values of a few tenths of a millimeter and plate counts in the hundreds of thousands.

The resolution equation: turning plates into peak separation

Plates are a means to an end; the end is resolution (Rs), the ability to distinguish two adjacent peaks. Its fundamental definition is the difference in retention times divided by the average baseline width:

Rs = 2(t_R₂ − t_R₁) / (w_b₁ + w_b₂)

At Rs = 1.0 the peaks are about 98% resolved (roughly 2% overlap, a visible valley); at Rs = 1.5 they are baseline-resolved, corresponding to 6σ of separation and roughly 99.7% purity of each band — the universally cited target for a quantitative assay. The genius of the field, formalized by Purnell in 1960, is recasting this into the master resolution equation, which factors resolution into three independently-tunable pieces:

Rs = (√N / 4) · ((α − 1) / α) · (k₂ / (1 + k₂))

• Efficiency term √N/4: resolution grows only with the square root of plate count. Doubling Rs by efficiency alone requires quadrupling N — i.e., a 4× longer column and a 4× longer run. This is the diminishing-returns trap.
• Selectivity term (α−1)/α: the separation factor α = k₂/k₁ measures the thermodynamic difference in partitioning. This is by far the most powerful knob — nudging α from 1.05 to 1.10 roughly doubles the term. It is controlled by chemistry: stationary phase, mobile-phase composition, pH, temperature.
• Retention term k₂/(1+k₂): the retention factor k = (t_R − t_M)/t_M. This term rises steeply from k = 0, reaching 0.83 at k = 5 and 0.91 at k = 10, then plateaus. The sweet spot is 1 < k < 10: enough retention to separate, not so much that peaks broaden and runs drag on.

Worked example: pushing two peaks to baseline

Consider a reversed-phase HPLC method resolving two structurally similar impurities. Initial conditions on a 15 cm column give N = 12,000, α = 1.05, and k₂ = 3.0. Plug into the master equation:

• √N/4 = √12000 / 4 = 109.5/4 = 27.4
• (α−1)/α = 0.05/1.05 = 0.0476
• k₂/(1+k₂) = 3.0/4.0 = 0.750

Rs = 27.4 × 0.0476 × 0.750 = 0.98 — the peaks touch but are not baseline-resolved. Now compare two strategies to reach Rs = 1.5.

Strategy 1 — brute-force efficiency. Hold α and k fixed and raise N. Since Rs ∝ √N, you need N scaled by (1.5/0.98)² = 2.34, so N ≈ 28,100 — meaning roughly 2.3× the column length (about 35 cm) and 2.3× the analysis time and backpressure. Expensive.

Strategy 2 — improve selectivity. Change the organic modifier (say methanol → acetonitrile) or drop the mobile-phase pH to shift the pKa-dependent partitioning, nudging α from 1.05 to only 1.08. Now (α−1)/α = 0.08/1.08 = 0.0741. Rs = 27.4 × 0.0741 × 0.750 = 1.52 — baseline resolution, same column, same run time. This is the enduring lesson of Purnell's factoring: when peaks won't separate, chase α (chemistry) before you chase N (hardware). Efficiency is the last resort, not the first.

Limits, subtleties, and where plate theory breaks down

Plate theory is a Gaussian idealization, and reality supplies several complications:

• Extra-column broadening: the observed variance is σ²_total = σ²_column + σ²_injector + σ²_tubing + σ²_detector. On short, high-efficiency UHPLC columns, dead volume in connectors and an oversized detector cell can silently halve your measured N. Efficient columns are unforgiving of sloppy plumbing.
• Asymmetry (tailing/fronting): the 16(t_R/w_b)² formula assumes a symmetric peak. Silanol interactions with basic analytes cause tailing (asymmetry factor As > 1.2), inflating apparent width and deflating N. The Foley–Dorsey correction or moment analysis (using true statistical moments μ₁, μ₂) is then mandatory for an honest count.
• Gradient elution: in a gradient the retention factor k changes continuously as the run proceeds, so a single k in the resolution equation is undefined. Peak-capacity metrics and the linear-solvent-strength (LSS) model of Snyder replace the isocratic plate framework.
• Overloading: at high sample mass the isotherm becomes nonlinear (Langmuir-type), producing shark-fin fronting peaks whose width depends on concentration — plate theory, built on a linear isotherm, simply does not apply. This is the border between analytical and preparative chromatography.

A deeper subtlety: the van Deemter A term's flow-independence is itself an oversimplification. The Giddings coupling theory (J. Calvin Giddings, Dynamics of Chromatography, 1965) showed that eddy diffusion and mobile-phase mass transfer are coupled, giving A a mild velocity dependence and yielding the more accurate Knox equation, h = Aν^(1/3) + B/ν + Cν, in reduced-parameter form (reduced plate height h = H/dₚ, reduced velocity ν = u·dₚ/Dₘ). Reduced parameters let you compare a 1.7 μm UHPLC column against a 5 μm HPLC column on equal footing — a well-packed column reaches a reduced-h minimum near 2 regardless of particle size, which is the number a column-quality zealot actually watches.

Plate (thermodynamic) theory versus rate (kinetic) theory of chromatographic band broadening

Feature	Plate theory (Martin–Synge)	Rate theory (van Deemter/Golay)
Core idea	Column = series of discrete equilibrium stages (plates)	Continuous mass-transfer and diffusion kinetics
Predicts N	Yes, empirically from peak width	Yes, mechanistically from H = A + B/u + Cu
Explains WHY N changes with flow	No — plate height is a fixed fudge factor	Yes — B/u, Cu terms give a minimum H at u_opt
Key output	N, H (plate count, plate height)	H vs. linear velocity curve, optimum flow
Physical realism	Plates don't physically exist (no true equilibrium)	Grounded in diffusion (Dm, Ds) and eddy dispersion
Best use	Reporting/comparing column efficiency	Optimizing flow rate, particle size, temperature