When Beer-Lambert lies

The Beer-Lambert law is one of the simplest quantitative expressions in all of chemistry: absorbance is proportional to concentration. That is the textbook version. In practice, the linear relationship holds only within a narrow window of conditions, and stepping outside that window is one of the most common ways students (and professionals) get systematically wrong analytical numbers. This guide walks the five ways the line stops being a line, what each failure mode looks like on a calibration curve, and what to do about it.

The law in one line

$A = \varepsilon \cdot b \cdot c$

Absorbance $A$ (unitless) equals molar absorptivity $\varepsilon$ (L·mol⁻¹·cm⁻¹) times path length $b$ (cm) times molar concentration $c$ (mol/L). All three of those factors have to hold constant for the linearity to work. The derivation from first principles (Skoog 10e §24-1, Mayerhöfer 2020) assumes:

1. The analyte molecules absorb independently — no chemical interaction between neighbours.
2. The incident radiation is monochromatic at a fixed wavelength.
3. The sample is a homogeneous, non-scattering medium.
4. The instrument sees only the transmitted analytical beam — no stray light at other wavelengths.

Every deviation below comes from one of those four assumptions breaking. Real spectrophotometry always has all four breaking to some degree; the question is whether the deviation is smaller than your required precision.

Deviation 1 — high concentration

What it looks like. The calibration plot is linear at low concentration and curves downward above some threshold (A > 1.0 is a common practical cutoff; the exact threshold depends on the analyte).

Why it happens. Two effects, both traceable to inter-molecular interaction at high $c$:

• Refractive-index change. Concentrated solutions have a measurably different $n$ from the pure solvent. Because $\varepsilon$ depends on $n$ (through the Lorentz-Lorenz relationship), the “constant” molar absorptivity drifts as concentration rises.
• Chemical association/dissociation. At high $c$, dimers, complexes, or hydrogen-bonded aggregates form; each species has its own spectrum and its own $\varepsilon$. What you measure is a weighted sum. Methylene blue dimerises noticeably above ≈10⁻⁵ M — its calibration goes non-linear at concentrations that seem trivially dilute.

Workaround. Dilute the sample. Beer-Lambert almost always recovers linearity below A ≈ 1.0. For samples that must be measured at high concentration (kinetics runs, in-line monitoring), calibrate over the actual concentration range instead of extrapolating a low-concentration fit — the polynomial fit is defensible if you cannot avoid the regime.

Deviation 2 — polychromatic radiation

What it looks like. The plot is linear at low $c$ but bends downward at high $A$, with the bend showing up sooner for sharper absorption bands than for broad ones.

Why it happens. No monochromator is truly monochromatic. It passes a wavelength window of width $\Delta\lambda$ (typically 1–2 nm for a mid-range UV-Vis). Within that window, $\varepsilon$ is not constant; on the shoulder of an absorption band, $\varepsilon$ varies rapidly with $\lambda$. What the detector actually measures is

$A_{observed} = -\log_{10}\left(\frac{\int I_0(\lambda) \cdot 10^{-\varepsilon(\lambda) bc}\ d\lambda}{\int I_0(\lambda)\ d\lambda}\right)$

which reduces to $\varepsilon b c$ only when $\varepsilon(\lambda)$ is constant across the window. Sharp bands violate this assumption first.

Workaround. Measure at the peak of the absorption band, not the shoulder — the top of a peak is where $\varepsilon(\lambda)$ has zero slope, so the polychromatic averaging is minimised. Use a narrower slit width if your instrument allows it; that reduces $\Delta\lambda$ at the cost of light throughput and precision.

Deviation 3 — stray light

What it looks like. The plot curves downward, with a hard asymptote at high $c$ that the sample cannot climb past no matter how concentrated it gets. Often mistaken for Deviation 1 or 2.

Why it happens. All spectrophotometers have some stray light — light at wavelengths other than the analytical wavelength that reaches the detector through scattering off internal optics, imperfect gratings, or fluorescence. Call the stray-light fraction $s$ (typically 10⁻³ to 10⁻⁵ of $I_0$ for a decent instrument, ≪10⁻⁴ for a research-grade). Because the analyte does not absorb at the stray wavelengths, they pass through nearly unattenuated, and the true absorbance formula becomes

$A_{observed} = -\log_{10}\left(\frac{10^{-\varepsilon bc} + s}{1 + s}\right)$

At high $\varepsilon b c$, the $10^{-\varepsilon bc}$ term goes to zero and $A_{observed} \to -\log_{10}(s/(1+s)) \approx -\log_{10} s$. For $s = 10^{-4}$, that asymptote is $A \approx 4$; you literally cannot measure an absorbance above 4 on that instrument regardless of the actual concentration.

Workaround. Dilute below the stray-light-limited regime (A < 2 for most benchtop instruments; A < 3 for research-grade). Test the stray light of your instrument with a known highly-absorbing solution (KCl at 220 nm is a standard test — Skoog 10e §24E-2).

Deviation 4 — chemical speciation

What it looks like. The calibration curve is not just non-linear; its slope changes depending on some other condition (pH, temperature, ionic strength, added ligand). The apparent $\varepsilon$ shifts under conditions that should not matter.

Why it happens. The analyte exists as multiple species in solution, each with its own spectrum. Common example: bromothymol blue is yellow in the protonated form and blue in the deprotonated form, with distinct spectra and distinct $\varepsilon$ values. Total absorbance is the sum weighted by the mole fraction of each species — which depends on pH. If you calibrate at one pH and measure at another, you will read the wrong concentration.

Workaround. Buffer the sample to force complete conversion to one species (usually 2 pH units past the $\mathrm{p}K_a$ in either direction, so 99% of one form is present). Alternatively, measure at an isosbestic point — a wavelength where both species have the same $\varepsilon$ — so the reading is independent of the ratio. Every acid-base indicator’s spectrum has an isosbestic point; find it by overlaying the pure-form spectra and looking for the crossing.

Deviation 5 — scattering

What it looks like. Non-zero baseline at wavelengths where the analyte should not absorb; sloping baseline that rises toward shorter wavelengths; irreproducibility between replicates.

Why it happens. Suspended particles, colloids, aggregates, and bubbles scatter light out of the analytical beam. The detector sees the scattering as an apparent absorbance — but it is not real absorption, so it does not follow Beer-Lambert. Rayleigh scattering rises as $1/\lambda^4$, so the baseline tilt is steeper at shorter (UV) wavelengths.

Workaround. Filter the sample (0.45 μm or 0.22 μm membrane, depending on particle size). Centrifuge if the aggregates re-form after filtering. For samples that inherently scatter (protein solutions, cell suspensions), subtract a baseline measured at a non-absorbing reference wavelength, or use a scattering-corrected instrument geometry (integrating sphere).

Worked example — an ambiguous non-linear plot

You measure a calibration curve for a dye and get A = 0.20, 0.42, 0.65, 0.85, 1.02, 1.13 at concentrations 1, 2, 3, 4, 5, 6 μM. The plot is clearly curving toward an asymptote near A ≈ 1.2. Which deviation is at play?

A note on the multi-species case

When the analyte is a mixture of species with overlapping spectra (protein at multiple wavelengths, dye equilibrium, chromophore + degradation product), you can still use Beer-Lambert, but as a matrix equation instead of a scalar one:

$\mathbf{A} = \mathbf{\varepsilon} \cdot \mathbf{c} \cdot b$

where $\mathbf{A}$ is the vector of absorbances at $n$ wavelengths, $\mathbf{c}$ is the vector of concentrations for $m$ species, and $\mathbf{\varepsilon}$ is the $n \times m$ molar absorptivity matrix. This is how UV-Vis spectroscopy quantifies protein concentration in the presence of a nucleic-acid contaminant (the classic 260/280 ratio is a 2-wavelength case of the same equation, Harris 10e §19-4). As long as you know $\mathbf{\varepsilon}$ for each species independently and each species obeys Beer-Lambert in isolation, the linear algebra reconstructs individual $c$ values from the combined spectrum.

Practice problems

Where to go next

Once you can spot which deviation is at play, the Beer-Lambert calculator becomes a working tool rather than a black box — the number it returns is only as good as your assumption that the linear law applies. When you get an unexpected concentration, the first diagnostic move is to look at the calibration plot’s shape, not to reach for the calculator again. And when the deviation is Deviation 4 (chemical speciation), a pH calculator helps you check whether you are on the right side of the analyte’s pKa to have a clean spectrum.