HPLC Detectors

A chromatographic column can separate forty compounds beautifully. But it tells you nothing if the detector cannot see them. The detector converts eluting analyte into an electrical signal. Every downstream number depends on this transduction step — retention time, peak area, limit of detection, calibration slope. Choosing the wrong detector for a given analyte is the single most common failure mode. It causes a well-optimized HPLC Instrumentation and Overview separation to deliver unusable data. This matters because no detector is universal. A UV-Vis detector is blind to a saturated fatty acid with no chromophore. A fluorescence detector ignores everything that does not fluoresce. A refractive-index detector cannot run gradients. A mass spectrometer struggles with nonvolatile buffers. Each detector trades selectivity for universality, sensitivity for robustness, and structural information for simplicity. Understanding these trade-offs — quantitatively, not just qualitatively — makes the difference. It separates a chromatographer who can troubleshoot a failing assay from one who can only run a validated method. The big picture is that LC detection is applied physical chemistry. It draws on Beer–Lambert absorption, quantum-yield-limited emission, differential refraction, and gas-phase ion physics. Each of these is wrapped in flow-cell engineering that must not destroy the chromatographic resolution just created. A detector with too much internal volume rebroadens a peak that took an hour to resolve. A detector too slow in its response time flattens a peak eluting in three seconds from a sub-2-micron column. This article works through the physics and figures of merit of the five detectors that dominate practice. These are UV-Vis (fixed and variable wavelength), diode-array (DAD/PDA), fluorescence, refractive index (RI), and mass spectrometry (LC-MS). It uses real absorptivities, real quantum yields, real S/N calculations, and the equations connecting flow-cell geometry to extra-column band broadening. It also sets up, without fully developing, the quantitative-analysis machinery these signals feed into — external standard, internal standard, and standard addition.

Key takeaways

UV-Vis detection follows $A = \varepsilon b c$ ; sensitivity scales with molar absorptivity $\varepsilon$ and path length $b$ , and is blind to non-chromophoric analytes.
Diode-array detectors (DAD) capture a full spectrum every scan, enabling peak-purity checks and spectral identification, at the cost of some sensitivity versus a single-wavelength UV detector.
Fluorescence detection is far more sensitive and selective than UV absorbance because it measures a signal against a dark background rather than a small difference between two large light intensities.
Refractive-index detection is nearly universal but low-sensitivity, temperature-sensitive, and fundamentally incompatible with gradient elution.
LC-MS provides structural confirmation and extreme selectivity via mass-to-charge ratio, but requires volatile mobile phases and manages a soft-ionization interface (ESI or APCI).
Detector flow-cell volume and response time must be small compared with the chromatographic peak volume/width, or the detector itself erodes the resolution the column produced.

Key idea: every LC detector is a transducer whose figure of merit is signal-to-noise per unit concentration, and the right detector choice depends on matching the analyte's physical property (chromophore, fluorophore, refractive index, ionizability) to the transduction physics available.

Where this lives today

Pharmaceutical quality control leans on diode-array UV detection for routine assay and impurity profiling because it is robust, linear, inexpensive, and gives a spectral fingerprint for peak-purity confirmation — checking that a peak's UV spectrum is uniform across its width to rule out a hidden co-eluting impurity. Bioanalytical and forensic toxicology laboratories quantifying drugs and metabolites in plasma at picogram-per-milliliter levels depend almost entirely on LC-MS/MS with SRM, because the required selectivity against thousands of endogenous plasma components is unattainable by UV or fluorescence alone. Amino acid and carbohydrate analysis, where analytes lack strong native chromophores, uses either pre-column derivatization (installing a UV or fluorescent tag, such as OPA or FMOC-Cl for amines) or falls back to RI or evaporative light-scattering detection for sugars and lipids. Environmental and food-safety residue testing for pesticides, mycotoxins, and veterinary drugs is now dominated by LC-MS/MS precisely because of the mass-selectivity argument worked through above — matrix interference that would defeat UV or fluorescence detection is filtered out by the two-stage precursor/product mass window. Polymer characterization (size-exclusion chromatography for molar-mass distributions) still relies heavily on RI detection, paired with light-scattering detectors, precisely because polymers often lack chromophores and because SEC is nearly always run isocratically — the one setting where RI's gradient-incompatibility is not a limitation. Fluorescence detection remains the gold standard for trace-level quantification of native fluorophores such as polycyclic aromatic hydrocarbons (environmental monitoring) and for derivatized biogenic amines and neurotransmitters in clinical and neuroscience research, where its combination of high sensitivity and selectivity against a dark background cannot be matched by absorbance-based methods. Understanding how these signals translate into defensible concentrations — calibration curves, internal standards, and the statistics of quantitation — is exactly the terrain of Qualitative and Quantitative Chromatographic Analysis.

Ask the tutor: How do I decide, for a brand-new analyte with an unknown UV spectrum, which detector to try first?

The physical picture

An HPLC detector sits downstream of the column, sees a stream of mobile phase with analyte bands moving through it, and must produce a signal proportional to analyte concentration (or mass flow) with minimal added band broadening. Four physical properties are exploited in mainstream practice: absorption of UV/visible light, emission of fluorescent light, the refractive index of the eluent, and the mass-to-charge ratio of gas-phase ions formed from the eluate. Each requires its own flow cell or interface geometry. The UV-Vis detector shines light of a chosen wavelength through a flow cell — typically a Z-shaped quartz capillary of volume 1–10 µL and path length $b = 1\text{ cm}$ — and measures the light transmitted. Absorbance follows the Beer–Lambert law, so it depends entirely on whether the analyte has a chromophore (a π-system, a carbonyl, an aromatic ring) that absorbs at the chosen wavelength. Diode-array detectors replace the single photodiode with a linear array (512–1024 elements) behind a polychromator grating, so the entire spectrum (190–800 nm typically) is captured in milliseconds at every point along the peak — sacrificing a little sensitivity (light is split across many diodes rather than concentrated) for a wealth of qualitative information. Fluorescence detectors excite the eluate with a specific wavelength (often a xenon arc lamp plus excitation monochromator, or a laser) and observe emission at a right angle, through an emission monochromator, so that only re-emitted photons — not the enormously more intense excitation beam — reach the photomultiplier tube. This 90° geometry is why fluorescence achieves such low backgrounds: the detector is not subtracting two large numbers (as absorbance does, $\log(P_0/P)$ ) but reading a small signal against essentially zero background light. Refractive-index detectors measure the difference in refractive index between the eluate (with analyte) and pure mobile phase (reference), typically via a Fresnel deflection-angle bridge cell. Because virtually every solute has a refractive index different from the solvent, RI is close to a universal detector — but the difference is tiny (often $\Delta n \sim 10^{-6}$ – $10^{-4}$ ), thermal drift and mobile-phase composition changes swamp the signal, and gradient elution (which continuously changes solvent RI) is essentially unusable with RI detection. Mass spectrometric detection breaks the eluate into gas-phase ions at atmospheric pressure — commonly by electrospray ionization (ESI) for polar/ionic analytes or atmospheric-pressure chemical ionization (APCI) for less polar, more volatile analytes — then separates and counts those ions by mass-to-charge ratio $m/z$ . This is the only one of the five that gives structural information (via fragmentation, in MS/MS) and near-absolute selectivity: two co-eluting compounds of different mass are trivially distinguished even if the chromatography cannot resolve them.

Figure

Beer–Lambert absorbance flow cell: light path, incident/transmitted intensity, and the resulting chromatographic peak

Figure

Fluorescence 90° geometry versus absorbance in-line geometry: why fluorescence reads against a dark background

Figure

Refractive-index detector Fresnel deflection cell: sample and reference beams bent by differing n

The core ideas

A_\lambda = \log_{10}\left(\frac{P_0}{P}\right) = \varepsilon_\lambda\, b\, c

where: $A_\lambda$ is absorbance (dimensionless) at wavelength $\lambda$ ; $P_0$ , $P$ are incident and transmitted radiant power; $\varepsilon_\lambda$ is the molar absorptivity (L·mol⁻¹·cm⁻¹) at that wavelength; $b$ is the path length (cm); $c$ is molar concentration (mol/L). Reading: absorbance is directly, linearly proportional to concentration only if $\varepsilon$ , $b$ are constant and stray light/polychromatic error are negligible — the working range for UV detectors is usually $A < 1$ –1.5 for good linearity.

F = K' \,\phi_f\, P_0\left(1 - 10^{-\varepsilon b c}\right) \approx 2.303\, K' \,\phi_f\, P_0\, \varepsilon\, b\, c \quad (\varepsilon b c \ll 1)

where: $F$ is measured fluorescence intensity; $K'$ is an instrument geometry/collection-efficiency constant; $\phi_f$ is the fluorescence quantum yield (fraction of absorbed photons re-emitted, $0 \le \phi_f \le 1$ ); $P_0$ is exciting radiant power; the exponential is the fraction of light absorbed, linearized for dilute solutions. Reading: fluorescence signal is proportional to both the amount absorbed ( $\varepsilon b c$ ) and the efficiency of re-emission ( $\phi_f$ ), and — crucially — to the incident intensity $P_0$ , which is why increasing lamp power (unlike in absorbance) directly increases fluorescence sensitivity without a corresponding increase in background noise.

\Delta n = n_{\text{sample}} - n_{\text{reference}} \approx \left(\frac{\partial n}{\partial c}\right)_{\!T,\lambda} c

where: $\Delta n$ is the differential refractive index measured by the RI detector; $(\partial n/\partial c)$ is the specific refractive index increment of the analyte in that solvent (typically $0.10$ – $0.20$ mL/g for organics in common LC solvents); $c$ is analyte concentration (mass/volume for polymers, molar for small molecules with appropriate units). Reading: signal is proportional to concentration but the proportionality constant is a weak, solute- and solvent-specific quantity, and both $n$ of the eluent and $(\partial n/\partial c)$ depend on temperature — hence RI cells are thermostatted to millikelvin stability.

\frac{m}{z} = \frac{\text{ion mass (Da)}}{\text{number of charges}}, \qquad S/N \propto \frac{[\text{ion current at target } m/z]}{\sqrt{\text{chemical + electronic noise}}}$}

where: mass analyzers (quadrupole, time-of-flight, ion trap, Orbitrap) separate and count ions by $m/z$ ; selected-ion monitoring (SIM) or selected/multiple-reaction monitoring (SRM/MRM, in tandem MS) restricts detection to a narrow $m/z$ window (or a precursor→product transition), which is why LC-MS achieves such low chemical background even in complex matrices. Reading: selectivity in mass increases signal-to-noise not by amplifying signal but by rejecting nearly all interferences that do not share the target ion's exact mass (and, in MS/MS, its fragmentation pathway).

Derivation AWhy absorbance, not transmittance, is linear in concentration

Goal. Show that defining signal as $A = -\log_{10}(P/P_0)$ , rather than raw transmitted power $P$ or transmittance $T = P/P_0$ , is what makes the detector response linear in concentration — the working principle behind every UV-Vis and DAD calibration curve.

Step 1 — Photon absorption is a first-order (exponential) process. Consider a beam of monochromatic radiant power $P$ passing through an infinitesimal slab $dx$ of solution containing absorbing species at concentration $c$ . Each absorbing molecule presents a cross-section for photon capture; the fraction of photons absorbed in $dx$ is proportional to the number of absorbers present,

-\frac{dP}{P} = k\, c\, dx

where $k$ is a proportionality constant absorbing the molecule's intrinsic absorption cross-section and units.

Step 2 — Integrate over the full path length. Integrating from $x=0$ (where $P = P_0$ ) to $x = b$ (where $P$ = transmitted power):

\int_{P_0}^{P} \frac{dP}{P} = -k c \int_0^b dx

\ln\left(\frac{P}{P_0}\right) = -k c b

Step 3 — Convert to base-10 log and define $\varepsilon$ . Converting the natural log to base 10 ( $\ln x = 2.303 \log_{10} x$ ) and absorbing the 2.303 into a redefined constant $\varepsilon = k/2.303$ :

\log_{10}\left(\frac{P}{P_0}\right) = -\varepsilon b c \quad\Longrightarrow\quad A \equiv -\log_{10}\left(\frac{P}{P_0}\right) = \varepsilon b c

Step 4 — Contrast with transmittance. If the detector instead reported $T = P/P_0 = 10^{-\varepsilon bc}$ directly, the response would be exponential in $c$ , requiring a nonlinear calibration and giving poor precision at low $T$ (small differences in a curve that is flattening out). By taking $-\log_{10}$ , the exponential is inverted into a straight line through the origin with slope $\varepsilon b$ , valid as long as $\varepsilon$ is independent of $c$ (dilute solutions, no chemical association/dissociation shifting the chromophore, and monochromatic — or effectively monochromatic — light so that a single $\varepsilon$ applies).

Step 5 — Practical linearity limit. Real detectors have finite spectral bandpass and stray light $P_s$ , so the true measured quantity is $A_{\text{meas}} = -\log_{10}\!\left(\frac{P+P_s}{P_0+P_s}\right)$ , which deviates negatively from ideal Beer–Lambert behavior as $A$ increases (stray light becomes a larger fraction of the small transmitted signal). This is why UV detector calibration curves are specified linear only up to roughly $A \approx 1.5$ –2, beyond which analysts dilute the sample or shorten the path length rather than extrapolate.

Go deeper · Derivation B · 2 minSignal-to-noise scaling: why fluorescence beats absorbance at trace levels

Derivation BSignal-to-noise scaling: why fluorescence beats absorbance at trace levels

Goal. Derive, from first principles of shot-noise-limited detection, why a fluorescence detector achieves lower limits of detection than a UV-Vis detector for the same analyte concentration, given comparable $\varepsilon b c \ll 1$ .

Step 1 — Absorbance noise is dominated by baseline (photon/electronic) fluctuations in a large background. In UV-Vis detection the measured quantity is $A = -\log_{10}(P/P_0)$ , and both $P_0$ and $P$ are large, comparable currents at the photodiode (nearly all light is transmitted for a dilute sample). The noise in $A$ propagates from fractional noise in $P$ and $P_0$ :

\sigma_A \approx \frac{0.434}{P}\,\sigma_P

which is essentially a constant absolute noise level (baseline noise, typically $\sigma_A \sim 10^{-5}$ AU for a good detector) independent of how much analyte is present, because $P \approx P_0$ for dilute samples.

Step 2 — Fluorescence noise is dominated by the same background fluctuation, but the background itself is near zero. In fluorescence, ideally no excitation light reaches the photomultiplier (90° geometry, emission monochromator/filter rejecting the excitation wavelength). The measured signal $F$ sits on a dark-current/Raman-scatter background that can be made extremely small, so:

\sigma_F \approx \sigma_{\text{dark}} + \sigma_{\text{Raman/blank}}\ (\text{both small, and largely independent of } c)

The signal itself, from the earlier fluorescence expression, is

F \approx 2.303\, K'\, \phi_f\, P_0\, \varepsilon\, b\, c

Step 3 — Compare limiting concentrations at $S/N = 3$ . Detection limit is defined by $S = 3\sigma_{\text{blank}}$ . For absorbance,

c_{\text{DL,UV}} = \frac{3\sigma_A}{\varepsilon b}

For fluorescence,

c_{\text{DL,fl}} = \frac{3\sigma_F}{2.303\,K'\phi_f P_0 \varepsilon b}

Step 4 — Identify the physical reason fluorescence wins. The ratio of detection limits is

\frac{c_{\text{DL,fl}}}{c_{\text{DL,UV}}} = \frac{\sigma_F}{2.303\,K'\phi_f P_0\,\sigma_A}

Because $\sigma_F$ (a dark-background noise) can be engineered far smaller than $\sigma_A$ (a fractional-noise-on-large-signal term), and because increasing lamp power $P_0$ increases fluorescence signal without proportionally increasing $\sigma_F$ (unlike in absorbance, where increasing $P_0$ does not change the ratio $P/P_0$ at all), fluorescence detection limits are commonly 10- to 1000-fold lower than UV absorbance for compounds with $\phi_f$ not too far below 1.

Step 5 — Limiting case. As $\phi_f \to 0$ (a molecule that absorbs but does not fluoresce — most alkanes' UV-active analogs, many simple aromatics with efficient internal conversion), $F \to 0$ and fluorescence detection fails entirely regardless of how strongly the molecule absorbs; the analyst must fall back to UV-Vis or derivatize the analyte to install a fluorophore. This limiting case is exactly why fluorescence is described as a selective, not universal, detector.

Worked examples

Example 1 — Foundational: absorbance, transmittance, and detector linearity range. A UV detector set at 254 nm monitors a compound with $\varepsilon_{254} = 12{,}300$ L·mol⁻¹·cm⁻¹ in a 1.00 cm path-length flow cell. A standard is prepared at $c = 3.50\times 10^{-5}$ mol/L. Absorbance:

A = \varepsilon b c = (12{,}300)(1.00)(3.50\times10^{-5}) = 0.4305

Percent transmittance:

T = 10^{-A} = 10^{-0.4305} = 0.3712 \;\Rightarrow\; \%T = 37.1\%

If the same solution is diluted to $c = 2.00\times10^{-4}$ mol/L by mistake (a 5.7× increase, not dilution — testing the linear range), the absorbance becomes:

A' = (12{,}300)(1.00)(2.00\times 10^{-4}) = 2.46

This exceeds the practical linear range (roughly $A < 1.5$ –2 for most photodiode-array or single-wavelength UV detectors), where stray light and polychromatic-error effects cause negative deviation from Beer's law. The analyst must dilute this sample by at least a factor of ~2 to bring $A$ back under ~1.2 before trusting a calibration curve built at lower concentrations. This illustrates why every HPLC method specifies a validated linear range, not just a single-point calibration.

Example 2 — Applied: choosing between UV, fluorescence, and RI for a real assay. A quality-control lab must quantify residual glucose (a non-chromophoric, non-fluorescent sugar with no UV chromophore above 210 nm and no native fluorescence) in a fermentation broth after HPLC separation on an ion-exclusion column, alongside a co-eluting aromatic preservative, methylparaben ( $\varepsilon_{254} \approx 14{,}000$ L·mol⁻¹·cm⁻¹), present at trace level. For methylparaben, UV-Vis at 254 nm is the obvious, sensitive, straightforward choice — a limit of detection can be estimated. Suppose baseline noise is $\sigma_A = 2\times10^{-5}$ AU (a typical modern UV detector specification) and $b = 1.00$ cm. Using $S/N=3$ :

c_{\text{DL}} = \frac{3\sigma_A}{\varepsilon b} = \frac{3(2\times10^{-5})}{(14{,}000)(1.00)} = 4.3\times10^{-9}\ \text{mol/L}

For a molar mass of methylparaben (152.15 g/mol), this is:

4.3\times10^{-9}\ \text{mol/L} \times 152.15\ \text{g/mol} = 6.5\times10^{-7}\ \text{g/L} = 0.65\ \mu\text{g/L}

an entirely realistic sub-ppb detection limit for a well-behaved chromophore. For glucose, UV detection is blind — glucose has no absorbing chromophore in the accessible UV range — so the chemist must switch to refractive-index detection. Glucose's specific refractive-index increment in water is approximately $dn/dc \approx 0.14$ mL/g, and a typical modern RI detector has a noise floor of roughly $\Delta n_{\text{noise}} \approx 1\times10^{-9}$ refractive index units (RIU), giving a detection limit on the order of:

c_{\text{DL,RI}} \approx \frac{3\,\Delta n_{\text{noise}}}{dn/dc} = \frac{3(1\times10^{-9})}{0.14} \approx 2\times10^{-8}\ \text{g/mL} = 20\ \mu\text{g/L}

This is roughly 30× worse (in relative sensitivity) than the UV example for methylparaben, illustrating quantitatively why RI is the fallback detector — used precisely when no chromophore exists — rather than the first choice whenever it can be avoided. Crucially, this ion-exclusion separation must run isocratically: RI detection cannot tolerate the gradual composition change of a gradient, since the mobile phase's own refractive index would drift and swamp the analyte signal.

Example 3 — Exam/competition-level: LC-MS/MS selectivity and quantification via SRM. A trace-residue laboratory must confirm and quantify a pesticide (nominal mass 291.0 Da, $[\mathrm{M+H}]^+$ at $m/z$ 292.1) in a spinach extract using LC-MS/MS with electrospray ionization in positive mode, monitoring the SRM transition $m/z\ 292.1 \rightarrow 214.0$ (a diagnostic fragment from loss of a 78.1 Da neutral). (a) Estimating detection advantage from mass selectivity. Suppose the chromatographic peak of interest co-elutes, in UV or even single-quadrupole full-scan MS, with three matrix components of similar retention time but different exact masses (say 291.9, 293.3, and 288.7 Da — none matching 292.1 within the instrument's mass accuracy of ±0.01 Da for this triple quadrupole operating in unit-resolution mode, ~0.7 Da FWHM). Because SRM requires both the correct precursor $m/z$ (292.1, within the ~0.7 Da window) and the correct product ion (214.0) from collision-induced dissociation, any co-eluting matrix compound must coincidentally share both properties to interfere — a vanishingly small probability compared to sharing just a retention time. This two-stage mass filtering is why LC-MS/MS routinely achieves selectivity equivalent to baseline chromatographic resolution ( $R_s \gg 1.5$ ) even when the actual chromatographic $R_s$ between analyte and interference is only 0.3 (nearly co-eluting). (b) Quantification via matrix-matched external calibration. The instrument response (peak area of the SRM transition, in counts) for five matrix-matched standards is:

c (ng/mL)	Peak area (counts)
0.50	4,120
1.00	8,350
2.50	20,900
5.00	41,300
10.00	83,600

Linear regression ( $y = mx + b$ ) gives, by the standard least-squares formulas

m = \frac{n\sum x_iy_i - \sum x_i \sum y_i}{n\sum x_i^2 - (\sum x_i)^2}, \qquad b = \bar y - m\bar x

with $n=5$ , $\sum x_i = 19.00$ , $\sum y_i = 158{,}270$ , $\sum x_iy_i = 496{,}015$ , $\sum x_i^2 = 132.75$ :

m = \frac{5(496{,}015) - (19.00)(158{,}270)}{5(132.75) - (19.00)^2} = \frac{2{,}480{,}075 - 3{,}007{,}130}{663.75 - 361.00} = \frac{-527{,}055}{302.75}

This intermediate arithmetic reveals a sign problem worth checking carefully — redo with correct products: $x_iy_i$ pairs are $(0.50)(4120)=2060$ , $(1.00)(8350)=8350$ , $(2.50)(20900)=52250$ , $(5.00)(41300)=206500$ , $(10.00)(83600)=836000$ ; sum $=2060+8350+52250+206500+836000 = 1{,}105{,}160$ . Redoing:

m = \frac{5(1{,}105{,}160) - (19.00)(158{,}270)}{302.75} = \frac{5{,}525{,}800 - 3{,}007{,}130}{302.75} = \frac{2{,}518{,}670}{302.75} = 8317\ \text{counts}\,\cdot\,\text{mL/ng}

b = \bar y - m\bar x = \frac{158{,}270}{5} - (8317)\left(\frac{19.00}{5}\right) = 31{,}654 - (8317)(3.80) = 31{,}654 - 31{,}605 = 49\ \text{counts}

So the calibration line is $y = 8317x + 49$ (counts vs. ng/mL), with intercept statistically indistinguishable from zero, consistent with a clean SRM blank. An unknown extract giving 15,220 counts is quantified as:

x = \frac{15{,}220 - 49}{8317} = \frac{15{,}171}{8317} = 1.824\ \text{ng/mL}

Reporting to appropriate significant figures given the calibration precision, the sample concentration is 1.82 ng/mL, well within the calibrated range (0.50–10.0 ng/mL) and requiring no extrapolation — the mark of a properly bracketed quantitative LC-MS/MS determination.

Limiting cases, scope & common traps

Detector	Universal?	Typical LOD (mass-basis, favorable analyte)	Gradient-compatible?	Structural info?	Key limitation
Fixed/variable λ UV-Vis	No (needs chromophore)	~1 ng injected	Yes	None	Blind to saturated, non-conjugated analytes
Diode-array (DAD/PDA)	No (same as UV)	~2–5 ng (slightly less sensitive than single-λ)	Yes	Full UV spectrum (purity, ID support)	Spectral resolution trades off vs. sensitivity
Fluorescence	No (needs native or derivatized fluorophore)	~10 pg–1 ng	Yes	None (some info from ex/em spectra)	$\phi_f \to 0$ analytes invisible
Refractive index (RI)	Nearly universal	~100 ng–1 µg	No	None	Thermal drift; cannot run gradients
Mass spectrometry (LC-MS)	Broad (needs ionizable species)	~1–100 pg (SRM)	Yes	Yes (mass, fragmentation)	Nonvolatile buffers foul the source; ion suppression

The chemist behind it

Modern LC detection descends from spectrophotometry's own history, but its direct architect for chromatography is harder to pin on one person — it is a convergence of analytical spectroscopy (Beer, Lambert, Bouguer for absorption law), fluorescence spectroscopy (Stokes, and later Alexander Jablonski's energy-level diagrams underpinning quantum yield), and 20th-century mass spectrometry pioneers who made atmospheric-pressure ionization practical for liquid streams. The single most transformative modern figure for LC-MS specifically is John B. Fenn, who developed electrospray ionization into a practical, gentle technique for putting large, fragile, polar biomolecules into the gas phase without destroying them — a discovery that turned mass spectrometry from a technique mainly for small volatile molecules into the dominant structural detector for peptides, proteins, and pharmaceuticals eluting from an LC column.

Origins · 1 minJohn Bennett Fenn and how this was discovered

John Bennett Fenn

1917–2010 · American physical chemist, Yale/Virginia Commonwealth University

Fenn spent much of his early career studying molecular beams and supersonic jet expansions — gas-phase physics far removed from wet chemistry. In the late 1980s, building on earlier electrospray work by Malcolm Dole, Fenn showed that electrospray ionization could volatilize and ionize large, fragile biomolecules such as proteins directly from solution, producing multiply charged ions detectable by ordinary mass analyzers even though the intact molecular masses (tens of kilodaltons) far exceeded the mass range of the instruments used. This meant a liquid stream — exactly the kind emerging from an HPLC column — could be interfaced directly to a mass spectrometer without the harsh fragmentation of earlier ionization methods. Fenn shared the 2002 Nobel Prize in Chemistry (with Koichi Tanaka and Kurt Wüthrich) for this development, which underlies essentially every LC-MS instrument used in pharmaceutical, proteomic, and clinical laboratories today.

History

1852

— Beer formulates the concentration dependence of absorption, building on Bouguer (1729) and Lambert (1760), giving the law now written

A = \varepsilon bc

1962

— Horace W. Moore and colleagues at DuPont build early practical UV-absorbance detectors for the first commercial liquid chromatographs.

1968

— Csaba Horváth and coworkers demonstrate high-pressure liquid chromatography with on-line UV detection, establishing modern analytical HPLC.

1979

— Diode-array detection (multichannel UV/vis spectroscopy) is commercialized for HPLC, enabling real-time spectral acquisition across a peak.

1984–1989

— John Fenn develops electrospray ionization for macromolecules, enabling practical, robust LC-MS and LC-MS/MS interfacing.

Why can't I just always use LC-MS if it's the most selective and sensitive detector?

Cost and complexity are real barriers — a triple-quadrupole or high-resolution LC-MS system costs far more to purchase and maintain than a UV or fluorescence detector, and requires specialized expertise for method development, source maintenance, and troubleshooting ion suppression from matrix effects. More fundamentally, MS requires volatile, MS-compatible mobile-phase buffers (ammonium formate/acetate rather than phosphate), which constrains method development, and many routine QC applications (potency assays, dissolution testing) do not need MS's selectivity — a validated UV method is faster, cheaper, and more robust for those purposes. MS is reserved for cases where trace-level selectivity, structural confirmation, or complex matrices genuinely demand it.

What is the "diode-array purity check" and how does it actually detect a hidden co-eluting impurity?

A DAD collects a full UV spectrum at every time point across a peak, not just a single absorbance value. If the peak is chromatographically pure, the normalized spectrum should be essentially identical at the peak's leading edge, apex, and tailing edge — the ratio of absorbance at two wavelengths (say 220 nm / 254 nm) stays constant across the peak. If a second, spectrally different compound co-elutes even partially underneath the visible peak, that absorbance ratio will drift systematically from front to back, because the relative contribution of the two overlapping spectra changes as their concentration profiles shift across the peak. This purity-ratio trace is a standard, near-mandatory check in regulated pharmaceutical HPLC methods.

Why is the fluorescence excitation/emission geometry always 90°, and could a different angle work?

The 90° right-angle geometry minimizes the amount of scattered and directly transmitted excitation light reaching the emission photomultiplier, since almost none of the exciting beam is redirected 90° by simple transmission or specular reflection — it mostly continues straight through the cell and is captured by a beam stop. Any residual stray light at the excitation wavelength would show up as background noise riding under the true fluorescence signal, degrading exactly the low-background advantage fluorescence relies on. In principle other angles could work with sufficiently good filters, but 90° combined with an emission monochromator (or sharp-cutoff filter) is the simplest, most robust way to reject Rayleigh-scattered excitation light while collecting emitted photons efficiently.

How does extra-column band broadening from the detector flow cell affect the chromatography, quantitatively?

Total observed peak variance is the sum of variances from every contributing process, $\sigma^2_{\text{obs}} = \sigma^2_{\text{column}} + \sigma^2_{\text{extra-column}}$ , and the detector flow cell (plus connecting tubing) contributes to the extra-column term. A flow cell with volume comparable to or larger than the eluting peak's volume (peak volume ≈ 4σ in volume units) rebroadens the peak and reduces the apparent efficiency and resolution the column actually achieved. This is why detectors for narrow-bore or UHPLC columns (peak volumes of a few microliters) use flow cells of 1–2 µL or less, while conventional 4.6 mm i.d. columns with wider peaks (tens of microliters) can tolerate the standard 8–13 µL analytical flow cell without meaningful resolution loss.

Can a single instrument run two detectors in series, and does the order matter?

Yes — a very common configuration places a UV-Vis or DAD detector immediately after the column (non-destructive, low-volume) and then routes the flow onward to a fluorescence detector or mass spectrometer, since UV absorbance measurement does not consume or alter the sample. If MS detection is downstream, a portion of flow is often split off (post-column splitter) before the MS source, both to protect the source from excessive flow and because UV and MS can be run on the same peak simultaneously without one interfering with the other. Order matters mainly because MS (and to an extent fluorescence, if photodecomposition is a concern) can be considered effectively "destructive" or flow-altering in practice, so non-destructive optical detectors are generally placed first in the flow path.

Why is refractive-index detection still used at all, given its poor sensitivity and gradient incompatibility?

RI detection is one of the very few genuinely universal LC detectors — because virtually any dissolved solute changes the bulk refractive index of the eluent by some measurable amount, RI can detect compounds with no chromophore, no fluorophore, and no easily ionizable functional group, such as simple carbohydrates, many polymers, and some lipids. In applications like size-exclusion chromatography for molar-mass distributions, where isocratic elution is standard practice anyway, RI's gradient limitation is irrelevant, and its universal response (proportional to mass concentration for a given polymer/solvent pair) is exactly what is needed for accurate molar-mass moment calculations.

Does molar absorptivity ε depend on the solvent or mobile-phase pH, and why does that matter for method development?

Yes, substantially — many chromophores (phenols, amines, carboxylic acids) have absorbance spectra that shift with protonation state, so mobile-phase pH (buffered for chromatographic retention control) can shift both the wavelength of maximum absorbance and the value of ε at a fixed monitoring wavelength. A method developed at pH 3.0 where a compound is fully protonated may show a different ε — and therefore a different calibration slope — than the same compound analyzed at pH 7.0 where it is deprotonated, so method validation must lock down and control mobile-phase pH as tightly as it controls flow rate and column temperature.

HPLC Detectors

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