Thomas G. Mayerhöfer on The Forgotten Half of Beer's Law: How Refractive-Index and Complex-Valued Chemometrics Are Rewriting Quantitative Spectroscopy

For 170 years, spectroscopists have built quantitative analysis on only one number pulled from every spectrum they ever recorded, discarding its inseparable twin. New dispersion-theory and complex-valued chemometric methods recover that missing half, the refractive index, and in favorable systems cut calibration error by as much as an order of magnitude.

Abstract

Every absorbance spectrum a chemist has ever recorded is only half of a complex quantity. The other half, the refractive index, has been measured, calculated, and then discarded as a nuisance for a century and a half. In a recent Episode 47 of the Analytically Speaking podcast, Dr. Thomas G. Mayerhöfer of the Leibniz Institute of Photonic Technology and Friedrich Schiller University Jena laid out the case that the Beer-Lambert law is not wrong so much as incomplete. Dispersion theory shows that absorption and refraction are the imaginary and real parts of one causal, Kramers-Kronig-linked response function, that Beer's law is only the dilute, weakly polarizable limit of the more general Lorentz-Lorenz relation, and that the refractive index obeys a concentration law of its own, whose empirical counterpart, molar refraction, was known and in routine use by the 1920s but was never joined to the absorption half — and which, on Mayerhöfer's account, had already been turned on the structure of benzene, arguing for Kekulé's ring against the thermochemistry of the day. Drawing that older physics back into modern practice, Mayerhöfer and coworkers have shown that regressing on the refractive index, or on complex-valued spectra that carry both channels at once, can reduce prediction error relative to conventional partial least squares (PLS) applied to the absorption index alone by up to an order of magnitude in favorable binary-mixture systems; in the classical least-squares formulation, the imaginary part of the predicted concentration additionally flags, and partly corrects, its own error. This article walks through the physics, the history, and the practical chemometric payoff, and outlines what a working spectroscopist would need to change to use it. For the full story, listen to Episode 47 of the Analytically Speaking podcast series. https://www.spectroscopyonline.com/analytically-speaking-podcast

Introduction

Ask most spectroscopists what a spectrum is, and the answer is nearly automatic: a plot of how much light a sample takes out of a beam, band by band, converted to concentration by way of the Beer-Lambert law,

A = ε b c, [1]

where A is absorbance, ε is the molar absorptivity (extinction coefficient), b is the optical path length, and c is the molar concentration of the absorbing species. That equation is arguably the single most used relationship in analytical chemistry, yet it does not appear in this form in August Beer's original 1852 paper.¹ Beer himself set out the intensity attenuation law and coined the term extinction coefficient two years later, in his 1854 Grundriss des photometrischen Calcüles — a book that opens by crediting Bouguer, Euler, Smith and Kästner, and Lambert for the systematization of 1760. The man whose name the law now carries claimed none of it, and the textbook division in which Lambert supplies the path law and Beer supplies concentration does not survive the sources.² The now-familiar expression descends instead from a separate photochemical tradition, and Beer's own 1853 optics textbook explicitly set absorption aside to focus on refraction instead.³

Thomas G. Mayerhöfer, who studies the wave optics of infrared spectroscopy at the Leibniz Institute of Photonic Technology (Leibniz IPHT) in Jena, Germany, returned to the Analytically Speaking podcast, where he had discussed the history and theory of infrared spectroscopy in Episode 29, to lay out what he calls "the other half of Beer's law," a body of theory and data built up across a series of papers since 2019 and consolidated most recently in a 2024 treatise on the wave optics of infrared spectroscopy.⁴ His argument begins with a deceptively simple observation: a sample's full optical response is a single complex number, the complex refractive index, in which an imaginary part describes absorption and a real part describes refraction and dispersion. Beer's law is a statement about the imaginary part alone. The two parts are not independent measurements; they are locked together by physical causality, so that measuring one across a wide enough spectral range already determines the other. That means the refractive index was never a separate measurement analytical chemistry was spared from making. It was information already present in every spectrum, simply left on the table and mostly forgotten.

The practical payoff of picking that information back up again is what makes this conversation worth every spectroscopist's attention: refractive-index-based and complex-valued regression methods, tested on real binary mixtures and extended to spiked blood plasma, have delivered errors as low as a quarter to well under a tenth of those from conventional absorbance-based calibration, depending on the system and the method.

The Complex Response Hiding in Plain Sight

Every optical measurement of a scalar, homogeneous, non-magnetic medium can be described by a single complex refractive index,

ñ(ω) = n(ω) + i k(ω), [2]

where ñ is the complex refractive index, n is the ordinary (real) refractive index, k is the absorption index, i is the imaginary unit, and ω is angular frequency. Absorbance, the quantity every spectroscopist calibrates against, is proportional only to k. Because ñ is an analytic response function of a causal, linear system, its real and imaginary parts cannot vary independently. They are connected by the Kramers-Kronig relations, which in one common form read

n(ω) − 1 = (2/π) P∫₀^∞ [ω′k(ω′) / (ω′² − ω²)] dω′, [3]

where n(ω) is the refractive index evaluated at frequency ω, k(ω′) is the absorption index evaluated at the integration variable ω′, and P denotes the Cauchy principal value of the integral, taken over all frequencies. In practice this means that if k(ω) is known across a sufficiently wide range, the dispersion of n(ω) is already fixed by causality; nothing new is being measured by also recording the dispersion, since it was implicit all along. Mayerhöfer's central claim is that treating absorbance as the whole story, rather than as one projection of a two-dimensional quantity, has cost the analytical spectroscopy field decades of calibration accuracy it did not know it was giving up.

What Benzene and Toluene Expose About Beer's Law

The clearest demonstration is also the simplest system available: benzene and toluene, a near-ideal binary mixture in which volumes add, Raoult's law holds, and there is no hydrogen bonding, complex formation, or specific solvation between components. Simple Beer-Lambert addition predicts that the infrared band positions of such a mixture cannot shift with composition; only band heights should change. In measured spectra, however, the strong band maxima migrate visibly as the mixture composition changes, while simulations built from simple additive absorbance do not move at all.⁵ The same behavior appears, to varying degrees, in benzene with cyclohexane and in benzene with carbon tetrachloride, systems with no meaningful chemistry to invoke. What is being measured, in other words, is not a molecular fingerprint scaled by concentration. It is how a wave propagates through a medium whose optical response is inherently collective.

The Lorentz-Lorenz Relation: Why Optical Response Is Collective

The reason the response is collective rather than additive is the local-field problem. In a dilute gas, each molecule responds only to the externally applied field. In a condensed phase, each molecule also feels the field radiated by every neighboring molecule it has just polarized, and those neighbors feel the field radiated back. Working through that self-consistent feedback, first formulated by Lorentz in 1878, yields the Lorentz-Lorenz relation,

(n² − 1) / (n² + 2) = (4π/3) Σᵢ Nᵢαᵢ, [4]

where n is the refractive index of the medium, Nᵢ is the number density of species i, αᵢ is its molecular polarizability, and the summation runs over all species present in the medium.⁶ Because the left-hand side, rather than n itself, is the quantity that is additive, the correct mixing rule for a binary system whose volumes add is a volume-fraction-weighted sum of the pure-component Lorentz-Lorenz functions,

LLmix = φ₁LL₁ + φ₂LL₂, [5]

where LLmix is the value of the left-hand side of equation [4] for the mixture, LL₁ and LL₂ are the corresponding values for pure components 1 and 2, and φ₁ and φ₂ are their respective volume fractions, with φ₁ + φ₂ = 1. That result also explains, independent of any argument about units of measure, why volume fraction rather than weight fraction is the physically correct calibration variable in mixture spectroscopy, a point Mayerhöfer connected on the podcast to earlier work by podcast host Jerry Workman and Howard Mark arguing that spectroscopy responds to volume fractions.⁷ The Lorentz-Lorenz relation is not itself exact, however. Its local-field term assumes spherical symmetry around each molecule and isotropic molecular polarizability, neither of which holds strictly for liquids such as benzene or toluene, so a residual mismatch with experiment remains — a mismatch that is itself informative, being sensitive to near-range order and molecular anisotropy. Anisotropic and micro-heterogeneous samples, which need a tensor or an effective-medium treatment rather than a single index, were a main theme of Mayerhöfer's earlier appearance in Episode 29. Beer's law, on this view, is simply the first-order truncation of the Lorentz-Lorenz relation in the dilute limit, and it inherits volume-fraction linearity from that same truncation while losing everything the full function contains.

Beer's Law as a Limiting Case, and Its Forgotten Twin

Taking the Lorentz-Lorenz relation of equation [4] to the dilute, weakly polarizable limit, where local-field feedback vanishes and absorbers act essentially independently, recovers Beer's law of equation [1] directly for the imaginary channel.⁸ The same single derivation, carried through for the real part, yields an equally simple companion law for the refractive index: Away from resonance, n − 1 is proportional to the product of oscillator number density and oscillator strength,

n − 1 = C · N · f, [6]

where n is the refractive index, N is the number density of oscillators, f is the oscillator strength, and C is a proportionality constant.⁹ Because that same product also sets the integrated area of the corresponding absorption band, both laws fall out of one derivation; whoever derives Beer's law has implicitly derived its twin as well, whether or not they notice it. One consequence Mayerhöfer highlighted is that a liquid's refractive index measured in a transparent window reflects the sum of oscillator strength across all of that liquid's electronic transitions, so that a simple refractometer reading is, in effect, an unrecorded spectrum condensed into a single number. That also settles a practical limit of the Kramers-Kronig route: a transform over a finite measurement window fixes the dispersion of n but not its absolute level. The missing constant is closely approximated by the non-resonant index n∞, provided it is measured in the near-infrared transparency region, around 6000 cm⁻¹, rather than at the sodium D line — a little dispersion always remains. In a mixture series that constant is not a fixed nuisance parameter either: n∞ differs between the pure components, so it varies with composition, which is exactly why the largest gains reported below appear in the system whose components differ most in background index.

The derivation also exposes an uncomfortable fact about the molar absorptivity ε of equation [1]: because absorbance is proportional to the imaginary part of the molecular polarizability divided by the refractive index of the surrounding medium, times concentration, ε is never a purely molecular constant. It depends on the matrix. That dependence is, in Mayerhöfer's telling, the physical reason a calibration built in one matrix routinely fails to transfer to another matrix containing the identical analyte, a failure mode usually blamed on instruments or sample preparation rather than on the equation itself.

Why the Refractive Index Does Not Simply Track Density

Since Isaac Newton, refractive index has generally been treated as tracking density. That picture holds reasonably well for alkanes and alcohols, but breaks down sharply for carboxylic acids, where refractive index actually decreases as density increases across a homologous series.¹⁰ The resolution is that the carboxyl group adds substantial mass, raising density, while contributing comparatively little oscillator strength in the spectral region that governs refractive index. The oscillators being counted here are electronic, not vibrational: it is the far-ultraviolet transitions of the C–H bonds, not the mid-infrared C–H stretch, that set the refractive index in the visible and near-infrared, so the number of C–H bonds serves as a proxy for electronic oscillator strength. Plotting refractive index against the concentration of C–H oscillators instead of density collapses alkane, alcohol, and carboxylic-acid series onto common straight lines, because each methylene group contributes a fixed number of C–H bonds with essentially constant oscillator strength per bond. In effect, n − 1 is proportional to C–H oscillator concentration, a direct restatement of the real-part law of equation [6] with oscillators, rather than whole molecules, as the counted entity.

A Refractometer in Every Vineyard

One of the more vivid illustrations Mayerhöfer offered concerns handheld refractometers used throughout agriculture to estimate grape sugar content. Such an instrument measures only the refractive index of grape juice, which is proportional to oscillator concentration; because sugar dominates the oscillator budget in that grape juice, the sugar calibration works well, even though the instrument is, strictly, counting oscillators rather than sugar molecules directly. That everyday device, and the degrees-Oechsle and Brix-type scales built on it, represent what may be the most widespread quantitative optical measurement in agriculture, running entirely on the half of the optical response that spectroscopy has otherwise mostly ignored.

Molar Refraction and a Century of Physical Chemistry Before Spectrometers

Before roughly 1920, refractive-index measurement was itself mainstream physical chemistry, combined with density into an additive quantity called molar refraction,

Rm = [(n² − 1) / (n² + 2)] · (M/ρ), [7]

where Rm is the molar refraction, n is the refractive index, M is the molar mass, and ρ is the density of the substance. Landolt and others tabulated additive increments for individual atoms, double bonds, and rings in reference works such as the 1911 Refraktometrisches Hilfsbuch of Roth and Eisenlohr, allowing chemists to predict molar refraction from a proposed structure or to invert the calculation and ask which structures were consistent with a measured value.¹¹ That the bracketed expression in equation [7] is identical to the left-hand side of equation [4] meant nineteenth-century chemists were routinely using the local-field relation as a working tool, decades before it was reconnected to spectroscopy.

Beer, Landolt, and the Kekulé Benzene Story

Mayerhöfer's historical case rests substantially on a close reading of an 1862 paper by Hans Landolt, who reports that he undertook his refractive-index survey of homologous liquid compounds at the explicit instigation of his Bonn colleague, August Beer.¹² That places Beer on the far side of a division he is often assumed not to have noticed: having published his absorption law in 1852 and an optics textbook that explicitly set absorption aside a year later, he nonetheless commissioned the search for a refractive-index concentration law before he died in 1863. Landolt's first target, tellingly, was the carboxylic-acid series that would not be properly explained by oscillator-density arguments until 2023.¹⁰ Later refractometric increment analysis, using molar refraction rather than thermochemical measurements, helped chemist Julius Wilhelm Brühl argue for the Kekulé ring structure of benzene against the thermochemical arguments prevalent at the time, a historical episode documented in detail by historian of science Stephen G. Brush.¹³

Classical Least Squares, Partial Least Squares, and a Revealing Sentence from 1988

Mayerhöfer's discussion of modern chemometrics turns on a single sentence from the foundational 1988 paper by David Haaland and Edward Thomas describing classical least squares performed band by band, which states that the approach allows "a high degree of rejection of spectral bands which do not follow Beer's law, or which include the presence of major interfering components."¹⁴ Mayerhöfer reads that sentence as evidence the field recognized, in the very paper that helped establish full-spectrum partial least squares regression, that some spectral bands simply do not obey Beer's law, conflating that physical failure with the separate, chemical problem of interfering components. Where band-by-band classical least squares at least made a misbehaving band visible as a diagnosable symptom, full-spectrum methods such as partial least squares and principal component regression fold that same misbehavior invisibly into additional latent variables, a process Mayerhöfer describes as converting a visible failure into a quietly accommodated one, at the cost of models carrying more factors than the underlying chemistry justifies and calibrations that resist transfer between instruments or matrices.

Quantifying the Gains: From Marginal to an Order of Magnitude

Working with benzene-toluene and related binary mixtures as benchmarks, Mayerhöfer and coworkers converted absorbance spectra into true optical constants using Fresnel-equation-based attenuated total reflection (ATR) correction closed with a Kramers-Kronig transform, then ran identical classical least squares regressions on the resulting channels.¹⁵ Using the absorption index k, the quantity Beer's law actually governs, as a baseline value of 100, the refractive index alone reduced relative error to roughly 50, and complex least squares performed on the full complex spectrum, on its own, achieved a similar reduction; the improvement became substantial only once a self-correction based on the imaginary part of the predicted concentration was applied, bringing relative error down to roughly 25.¹⁶ Results were system-dependent: benzene-cyclohexane improved less, and benzene-carbon tetrachloride, the system showing the largest deviation from Beer's law, showed no improvement from classical least squares over the refractive index alone. That same system, however, is where partial least squares performed on the refractive index channel showed its largest gain of all, achieving errors more than an order of magnitude lower than conventional partial least squares applied to the absorption index alone, because the mismatch that defeats classical least squares there, a large difference in the neat components' background refractive index, is precisely what partial least squares exploits.¹⁷ The same complex-valued reformulation was subsequently extended to inverse least squares, benchmarked against its real-valued counterpart under identical leave-one-out validation protocols, with the improvement reported to persist when the approach was extended to a more realistic matrix, blood plasma spiked with glucose and urea, though Mayerhöfer cautioned that the magnitude of improvement should not be assumed to transfer untested to other process streams.

Complex-Valued Chemometrics as a Self-Diagnostic

Perhaps the most distinctive result concerns what happens when a chemometric regression is performed directly on the complex spectrum, rather than projecting the data onto a single real axis first. The predicted concentration itself comes out complex. There is, of course, no physical meaning to an imaginary quantity of a chemical substance, but Mayerhöfer's group found that the imaginary component correlates almost linearly with the model's own real-valued prediction error.¹⁸ Fitting that relationship on training folds and applying it as a correction reduced residuals by more than half in classical least squares, while complex-valued inverse least squares reduced mean error and mean absolute error by over 50% under leave-one-out validation on both benzene-toluene and benzene-cyclohexane systems, without any additional correction step. Because standard, real-valued chemometrics projects a genuinely complex response onto a single axis, the systematic dispersion-related bias that projection discards has nowhere to go except into the residuals, where it resembles noise. In the complex-valued formulation, that same bias becomes identifiable structure, giving practitioners, in effect, a sample-by-sample flag for when the underlying electrodynamics have become non-negligible, without requiring an independent reference measurement for that sample.

In other words, ordinary chemometrics takes a measurement that really has two parts — think of an arrow, which has both a length and a direction — and keeps only its shadow on a single axis. Whatever gets lost in that flattening doesn't just disappear; it shows up later as unexplained error, indistinguishable from random noise. When you instead keep both parts of the measurement, that same lost information stops looking like noise and becomes structure the model can identify. In practice, that means the model can tell you, sample by sample, when the optics of the measurement are distorting the result enough to matter, without needing to double-check that sample against an independent reference measurement.

Getting the Optical Constants: What It Takes in Practice

Mayerhöfer described three routes to obtaining the refractive index and absorption index from routine measurements, in order of increasing effort. The simplest requires no new instrumentation at all: a Kramers-Kronig transform of an already-measured absorbance spectrum, treating the measured intensities as the imaginary part and computing the real part from the integral relation of equation [3], a calculation the group has shown works for both infrared and Raman spectra and can be implemented in a few minutes using standard fast Fourier transform routines.¹⁹ The second route, appropriate for ATR measurements, requires a Fresnel-equation-based correction incorporating the internal reflection element's refractive index, the angle of incidence, polarization, and a reasonable estimate of the sample's non-resonant index. The third route is full dispersion analysis, fitting a causal oscillator model to the measurement. The choice between the two is not simply one of effort: the Fresnel inversion is model-free and so recovers even weak oscillators, but it passes the measurement noise through with them, and it wants a spectral window that begins and ends in transparency — and, in ATR, one whose ends are supercritical, although subcritical excursions in between are no obstacle. Dispersion analysis, being forward modeling rather than inversion, is bound by neither condition: it suppresses the noise and tolerates both a truncated window and subcritical ends, at the cost of identifying every band by hand. Kramers-Kronig-consistent by construction, it is the route of choice for reference-quality optical constants.²⁰ To lower the practical barrier for the ATR route specifically, Mayerhöfer's group has developed a free tool, ATR Workbench, built around treating ATR as an inverse reflection problem and offering a graded hierarchy of corrections from simple wavelength scaling through full Fresnel and Kramers-Kronig inversion. It also detects oscillators automatically, which removes much of the manual band-finding that otherwise makes dispersion analysis expensive.²¹ Notably, the group's benchmarking found that complex-valued inverse least squares, using only a handful of well-chosen wavenumber points, matched or exceeded complex-valued partial least squares performance, meaning the most elaborate full-spectrum treatment is not required to capture most of the available signal benefit.

Where Beer's Law Remains Perfectly Adequate

Mayerhöfer was careful to limit the claims. Beer's law remains the correct and useful limiting case for dilute solutions of weakly polarizable species in a matrix whose refractive index does not shift with analyte concentration, the regime that describes most ultraviolet-visible spectroscopy of micromolar chromophores in water. He offered four practical warning signs that a measurement has left that regime: band maxima that shift with composition or concentration, a condition Beer-Lambert explicitly forbids; residual calibration curvature that survives conventional preprocessing; a calibration that fails to transfer between instruments for an identical analyte; and, more structurally, any measurement made on neat liquids, at high concentration, or by ATR, where the refractive index is built into the measurement geometry itself.

Summary and Conclusions

The through-line of Mayerhöfer's argument is that Beer's law is not incorrect but incomplete, the dilute, weakly polarizable limit of the more general Lorentz-Lorenz relation linking absorption and refraction as two projections of one complex, causally constrained optical response.²² Benzene-toluene and related binary mixtures demonstrate directly that absorbance alone cannot account for measured spectral behavior even in chemically inert systems, while historical evidence, most notably Landolt's 1862 account of working at Beer's own instigation, indicates that Beer himself set the search for a refractive-index concentration law in motion, even as the absorption half was acquiring his name, only to be set aside as spectrometry displaced refractometry over the following century. Restoring the discarded channel, whether as a standalone refractive-index regression, a Kramers-Kronig-derived complex spectrum, or a full complex-valued model whose imaginary part, in the least-squares formulations, doubles as a built-in error diagnostic, produced measured error reductions ranging from marginal to more than an order of magnitude, depending on the pairing of system and method — benzene-carbon tetrachloride being the least rewarding case for classical least squares and the most rewarding for partial least squares, while benzene-toluene gives roughly fourfold, using data and validation protocols already standard in the field.

Future Outlook

The practical entry point Mayerhöfer emphasized most strongly is also the cheapest: a Kramers-Kronig transform of an absorbance spectrum a lab already has, requiring no new hardware. Wider adoption will likely depend on that calculation, and Fresnel-based ATR correction more generally, becoming a standard preprocessing step in commercial FTIR and chemometrics software rather than a specialist procedure, together with more systematic documentation of ATR measurement geometry in routine workflows. Mayerhöfer's group has already extended the approach beyond simple binary solvent mixtures to a spiked blood plasma matrix, and further validation across pharmaceutical, process-analytical, and biomedical matrices, together with a forthcoming second book volume on attenuated total reflection infrared spectroscopy and continued development of the free ATR Workbench software, will determine how broadly the reported order-of-magnitude gains generalize beyond the favorable model systems studied so far.²³ Next up, maybe a discussion on the implications of this work for solid and powdered samples.

References

1. Beer, A. Bestimmung der Absorption des rothen Lichts in farbigen Flüssigkeiten. Ann. Phys. Chem. 1852, 162 (5), 78–88. DOI: 10.1002/andp.18521620505
2. Beer, A. Grundriss des photometrischen Calcüles; Vieweg: Braunschweig, Germany, 1854.
3. Beer, A. Einleitung in die höhere Optik; Vieweg: Braunschweig, Germany, 1853.
4. Mayerhöfer, T. G. Wave Optics in Infrared Spectroscopy: Theory, Simulation and Modeling; Elsevier: Philadelphia, PA, 2024.
5. Mayerhöfer, T. G.; Dabrowska, A.; Schwaighofer, A.; Lendl, B.; Popp, J. Beyond Beer's Law: Why the Index of Refraction Depends (Almost) Linearly on Concentration. ChemPhysChem 2020, 21 (8), 707–711. DOI: 10.1002/cphc.202000018
6. Mayerhöfer, T. G.; Popp, J. Beyond Beer's Law: Revisiting the Lorentz–Lorenz Equation. ChemPhysChem 2020, 21 (12), 1218–1223. DOI: 10.1002/cphc.202000301
7. Workman, J.; Mark, H. Units of Measure in Spectroscopy, Part I: It's the Volume, Folks! Spectroscopy 2014, 29 (2); Workman, J.; Mark, H. Units of Measure in Spectroscopy, Part III: Summary of Our Findings. Spectroscopy 2015, 30 (2).
8. Mayerhöfer, T. G.; Popp, J. Beer's Law Derived from Electromagnetic Theory. Spectrochim. Acta, Part A: Mol. Biomol. Spectrosc. 2019, 215, 345–347. DOI: 10.1016/j.saa.2019.02.103
9. Mayerhöfer, T. G.; Pipa, A. V.; Popp, J. Beer's Law—Why Integrated Absorbance Depends Linearly on Concentration. ChemPhysChem 2019, 20 (21), 2748–2753. DOI: 10.1002/cphc.201900787
10. Mayerhöfer, T. G.; Spange, S. Understanding Refractive Index Changes in Homologous Series of Unbranched Organic Compounds Based on Beer's Law. ChemPhysChem 2023, 24 (19), e202300430. DOI: 10.1002/cphc.202300430
11. Roth, W. A.; Eisenlohr, F. Refraktometrisches Hilfsbuch; Veit: Leipzig, Germany, 1911.
12. Landolt, H. Ueber die Brechungsexponenten flüssiger homologer Verbindungen. Ann. Phys. Chem. (Poggendorff's Annalen) 1862, 117 (11), 353–385.
13. Brush, S. G. Dynamics of Theory Change in Chemistry: Part 1. The Benzene Problem 1865–1945. Stud. Hist. Philos. Sci. 1999, 30 (1), 21–79.
14. Haaland, D. M.; Thomas, E. V. Partial Least-Squares Methods for Spectral Analyses. 1. Relation to Other Quantitative Calibration Methods and the Extraction of Qualitative Information. Anal. Chem. 1988, 60 (11), 1193–1202. DOI: 10.1021/ac00162a020
15. Mayerhöfer, T. G.; Ilchenko, O.; Kutsyk, A.; Popp, J. Quantitative Chemometrics Using Refractive Index Spectra. Appl. Spectrosc. 2025, 79 (11), 1659–1664. DOI: 10.1177/00037028251345774
16. Mayerhöfer, T. G.; Ilchenko, O.; Kutsyk, A.; Popp, J. Complex-Valued Chemometrics in Spectroscopy: Classical Least Squares Regression. Appl. Spectrosc. 2025, 79 (12), 1768–1775. DOI: 10.1177/00037028251343908
17. Mayerhöfer, T. G.; Ilchenko, O.; Kutsyk, A.; Popp, J. Complex-Valued Chemometrics in Spectroscopy: Partial Least Squares Regression. Appl. Spectrosc. 2026, 80 (4), 416–427. DOI: 10.1177/00037028251401941
18. Mayerhöfer, T. G.; Ilchenko, O.; Kutsyk, A.; Popp, J. Complex-Valued Chemometrics in Spectroscopy: Inverse Least Squares Regression. Appl. Spectrosc. 2026, 80 (1), 100–108. DOI: 10.1177/00037028251358392
19. Mayerhöfer, T. G.; Ilchenko, O.; Kutsyk, A.; et al. Complex-Valued Chemometrics for Analyzing Absorbance or Raman Spectra. Anal. Chem. 2026, 98 (15), 10953–10965. DOI: 10.1021/acs.analchem.5c03662
20. Mayerhöfer, T. G.; Ivanovski, V.; Popp, J. Infrared Refraction Spectroscopy. Appl. Spectrosc. 2021, 75 (12), 1526–1531. DOI: 10.1177/00037028211036761
21. ATR Workbench. Free software for Fresnel-based ATR correction, optical-constant retrieval, and dispersion analysis. Release is planned to accompany ref. (23); a research beta is available from the author.
22. Mayerhöfer, T. G.; Noda, I.; Popp, J. Reducing Non-Linearity in Spectral Evaluation via a Modified Lorentz–Lorenz Relation. Appl. Spectrosc. 2026, published online ahead of print. DOI: 10.1177/00037028261454699
23. Mayerhöfer, T. G. Attenuated Total Reflection Infrared Spectroscopy: From Wave Optics to Correction and Quantitative Analysis; Elsevier: Philadelphia, PA, forthcoming 2027.