📖 Core Concepts
Core Concepts — CHEM302: Analytical Chemistry: Principles of Chemical Measurement
0. What Is This Subject Really About?
Most of your chemistry so far has asked what happens: what bonds form, which way equilibrium shifts, how fast a reaction goes. Analytical chemistry asks a different question: how do we know, and how well do we know it? It is the epistemology of chemistry turned into a laboratory practice. Every number reported in a paper, such as a Ka, a yield, a purity, or a binding constant, passed through some analytical measurement. If that measurement was poorly designed or poorly understood, everything built on it is suspect.
Analytical chemistry is often caricatured as "titrations and instruments." A better way to see it is as the meeting point of nearly everything you have already learned:
- Statistics (MATH201) tells you what a measurement means. A number without an uncertainty is not a result. It is a rumor.
- Equilibrium and thermodynamics (CHEM102, CHEM301) tell you what species are actually present in your flask, which is rarely what the label on the bottle says.
- Electromagnetism and quantum physics (PHYS202) explain how light and electrons can be made to report on concentration.
- Linear algebra (MATH301) lets you untangle overlapping signals from mixtures.
- Kinetics and transport (PHYS201, CHEM102) govern how molecules spread out as they move through a chromatography column.
A unifying mental model you can carry through the whole course is this:
Every analytical method is a transducer. It converts a chemical quantity (moles, concentration, activity) into a physical signal (mass, volume, voltage, current, photons, time) through a relationship we understand well enough to invert.
Gravimetry converts moles into mass. Titrimetry converts moles into volume. Potentiometry converts activity into voltage. Spectrophotometry converts concentration into attenuated light. Chromatography first converts identity into time, separating components, and then uses one of the other transducers to quantify each. For every method, ask three questions:
- What is the physical law connecting the signal to the amount of analyte? (Its calibration function.)
- Under what conditions does that law break down? (Its deviations and interferences.)
- What limits how precisely I can read the signal? (Its noise and uncertainty.)
Harris and Lucy frame the field around the full "analytical process": formulating the question, selecting procedures, sampling, sample preparation, analysis, reporting, and drawing conclusions [Harris & Lucy, 2020]. Notice that the instrument is only one step. In practice the biggest errors usually come from sampling and sample preparation, not from the instrument.
1. Statistics and Quality Assurance: The Grammar of Measurement
1.1 Accuracy, precision, and the two kinds of error
Every measurement carries two kinds of error:
- Systematic (determinate) error is a consistent bias, such as a miscalibrated buret or an impure standard. It affects accuracy (closeness to the true value). Repeating the measurement does not reduce it. You can only find it by comparison with a standard reference material, an independent method, or a blank.
- Random (indeterminate) error is unpredictable scatter from electronic noise, reading a meniscus, or thermal fluctuations. It affects precision (reproducibility). Averaging reduces it.
The analogy everyone uses is a dartboard, and it is worth keeping. A tight cluster far from the bullseye is precise but inaccurate. A loose scatter centered on the bullseye is accurate but imprecise. Statistics can only describe the scatter. It is blind to the offset.
1.2 The Gaussian distribution
When many small, independent random errors add together, the central limit theorem (which you met in MATH201) tells us their sum tends toward a Gaussian distribution:
y = (1 / (σ√(2π))) · exp[ −(x − μ)² / (2σ²) ] μ = population mean σ = population standard deviation ±1σ contains 68.3% ±2σ contains 95.5% ±3σ contains 99.7%
In the lab we never know μ and σ. We estimate them from a finite sample:
x̄ = (Σ xᵢ) / n s = √[ Σ(xᵢ − x̄)² / (n − 1) ]
The n − 1 (degrees of freedom) appears because x̄ was computed from the same data. One piece of information has been "used up" in locating the center, so the scatter about x̄ is systematically a bit smaller than the scatter about the true μ. Dividing by n − 1 corrects this bias in s².
1.3 The Student t-distribution and confidence intervals
Here is a subtle and beautiful point. When n is small, s is itself an uncertain estimate of σ. W. S. Gosset, publishing as "Student" while working for the Guinness brewery, worked out the distribution that accounts for this extra uncertainty [Student, 1908]. The t-distribution has heavier tails than the Gaussian, and it converges to the Gaussian as n → ∞.
Confidence interval for the mean:
μ = x̄ ± t · s / √n
t depends on the confidence level (e.g., 95%) and degrees of freedom (n − 1)
Notice the √n. Halving the confidence interval requires four times as many replicates. This square-root law of diminishing returns will reappear in chromatography (resolution ∝ √N) and in signal averaging (S/N ∝ √number of scans). It is the same mathematics each time: independent random fluctuations add in quadrature.
Interpret a 95% confidence interval carefully. It does not mean "there is a 95% probability that μ lies in this particular interval." It means that if you repeated the whole experiment many times, 95% of the intervals constructed this way would contain μ [Harris & Lucy, 2020].
The t-statistic also drives hypothesis tests. Comparing a mean with a known value, comparing two means (after an F-test checks whether their variances can be pooled), and paired t-tests for comparing two methods on the same samples are all your daily tools for deciding whether a difference is real.
1.4 Propagation of uncertainty
Most results are computed from several measured quantities. How do their uncertainties combine? From MATH201 and calculus, the general first-order formula for independent variables is:
For F = f(x, y, z, …): σ_F² ≈ (∂F/∂x)² σ_x² + (∂F/∂y)² σ_y² + (∂F/∂z)² σ_z² + …
From this, Harris derives the practical rules [Harris & Lucy, 2020]:
Addition / subtraction: e_y = √(e₁² + e₂² + e₃²) (absolute uncertainties) Multiplication / division: %e_y = √(%e₁² + %e₂² + %e₃²) (relative uncertainties) Powers y = xᵃ: %e_y = a · %e_x Logarithm y = log₁₀ x: e_y = 0.43429 · (e_x / x) Antilog y = 10ˣ: e_y / y = 2.3026 · e_x
The logarithm rule has a practical consequence. Since pH = −log aH⁺, an uncertainty of ±0.02 in pH corresponds to roughly a ±4.6% relative uncertainty in hydrogen-ion activity. This is why pH is reported to two decimal places, and why those decimals correspond to only two significant figures in [H⁺]: the digits before the decimal point merely locate the power of ten.
Caution: these formulas assume the errors are independent. If you use the same pipet twice, or compute both a numerator and denominator from the same calibration, the errors are correlated and the simple rules can mislead you. In that case you need the full covariance terms, which take the form of a matrix product: Jᵀ Σ J, familiar from MATH301.
1.5 Outlier tests: Grubbs and Dixon
Sometimes one value sits suspiciously far from the rest. Two classical tests address this:
Grubbs test: G_calc = |x_questionable − x̄| / s
(x̄ and s computed including the suspect point)
Reject if G_calc > G_table [Grubbs, 1969]
Dixon Q test: Q = gap / range
gap = |x_questionable − nearest neighbor|
range = x_max − x_min
Reject if Q > Q_table [Dean & Dixon, 1951]
Harris and Lucy recommend the Grubbs test (current editions present it in place of the older Q test). The deeper lesson is philosophical. An outlier test is not permission to delete inconvenient data. First look for a physical cause, such as a spilled sample or an air bubble. With only 3 to 4 points these tests have very low statistical power, and the best response to a suspect point is often to collect more data.
1.6 Calibration and quality assurance
Linear least-squares calibration (y = mx + b) is the workhorse of quantitation. You already know the mathematics from MATH301: it is the projection of the data vector onto the column space of the design matrix. Analytical chemistry adds operational vocabulary:
- Sensitivity is the slope m.
- Detection limit is commonly taken as the signal at 3s above the blank, where s is the standard deviation of low-concentration samples. The concentration detection limit is 3s/m.
- Lower limit of quantitation is commonly taken as 10s/m.
- Linear range and dynamic range describe the span over which the calibration holds.
- Blanks (method blanks, reagent blanks, field blanks), spikes, and control charts are how a lab demonstrates that its numbers deserve trust.
These definitions follow Harris and Lucy's treatment [Harris & Lucy, 2020].
2. Activity: When Concentration Lies
2.1 Why activity exists
In CHEM301 you learned that the chemical potential of a species in solution is:
μᵢ = μᵢ° + RT ln aᵢ , aᵢ = γᵢ [i] / c°
Here is the key insight. Equilibrium constants are rigorously defined in terms of activities, not concentrations. The tidy expressions you wrote in CHEM102, such as Ksp = [Ag⁺][Cl⁻], are approximations that work only when γ ≈ 1, which means very dilute solutions.
Physically, each ion in solution is surrounded by an ionic atmosphere, a diffuse cloud enriched in ions of opposite charge. This cloud partially screens the ion's charge and lowers its effective "eagerness" to react. Picture a celebrity in a crowded room. Surrounded by an entourage, the celebrity is less available to interact with anyone else. The more crowded the room (the higher the ionic strength), the thicker the entourage.
2.2 Ionic strength
μ = ½ Σ cᵢ zᵢ²
Because charge enters as z², a 0.01 M solution of MgSO₄ (both ions doubly charged) has ionic strength 0.04 M, four times that of 0.01 M NaCl.
2.3 The extended Debye–Hückel equation
Debye and Hückel derived activity coefficients by combining the Poisson equation from electrostatics (PHYS202) with Boltzmann statistics for ion distributions (CHEM301) [Debye & Hückel, 1923]. The extended form used in Harris, valid at 25 °C up to about μ ≈ 0.1 M, is:
−0.51 · z² · √μ
log γ = ──────────────────────────
1 + (α · √μ / 305)
z = ion charge
μ = ionic strength (M)
α = effective hydrated ion size (pm)
305 pm·M^(1/2) arises from the Debye screening length at 25 °C
Tabulated α values trace back to Kielland's classic compilation [Kielland, 1937]. Several consequences follow:
- γ decreases as μ increases in the typical range, and does so much faster for highly charged ions because of the z².
- Small, highly charged ions have large hydrated radii. Li⁺ and Mg²⁺ hold their water tightly, so they behave as "bigger" ions than Cs⁺ or Ba²⁺. This is counterintuitive and worth remembering.
- "Salting in." Adding an inert salt such as KNO₃ increases the solubility of a sparingly soluble salt such as CaSO₄, because lower γ values require larger concentrations to reach the same activity product.
- Neutral molecules have γ ≈ 1 at low ionic strength, although at high ionic strength many show "salting out," with γ rising above 1.
At higher ionic strength (up to roughly 0.5 M), the empirical Davies equation is often used instead. Beyond that, specialized models such as Pitzer's are required.
3. Systematic Equilibrium Calculation
3.1 The method
In CHEM102, you probably solved equilibrium problems with ICE tables and clever approximations. Those approximations break for dilute solutions, polyprotic systems, and multiple coupled equilibria. Harris lays out an approach that always works because it is simply the complete set of physical constraints [Harris & Lucy, 2020]:
- Write all pertinent reactions.
- Write the charge balance: the solution is electrically neutral.
- Write the mass balance(s): atoms are conserved.
- Write the equilibrium expression for each reaction, using activities if needed.
- Count equations and unknowns. They must be equal.
- Solve, numerically if necessary.
Charge balance: Σ nᵢ [Cᵢ] = Σ mⱼ [Aⱼ]
(each concentration weighted by the magnitude of its charge)
Example, H₃PO₄ in water:
[H⁺] = [H₂PO₄⁻] + 2[HPO₄²⁻] + 3[PO₄³⁻] + [OH⁻]
Mass balance, F M H₃PO₄:
F = [H₃PO₄] + [H₂PO₄⁻] + [HPO₄²⁻] + [PO₄³⁻]
Note: the charge balance cannot include activity coefficients. Electroneutrality is a statement about real charge, so it is written in concentrations.
This approach is how professional speciation software works. It is also a linear-algebra-flavored problem, a system of coupled nonlinear equations. Spreadsheet solvers or Newton–Raphson iteration (MATH101/102) handle it easily.
3.2 Fractional composition (α) of polyprotic acids
For a diprotic acid H₂A, the fraction in each form depends only on [H⁺] and the Ka values, not on total concentration:
[H⁺]²
α(H₂A) = ─────────────────────────
[H⁺]² + K₁[H⁺] + K₁K₂
K₁[H⁺]
α(HA⁻) = ─────────────────────────
[H⁺]² + K₁[H⁺] + K₁K₂
K₁K₂
α(A²⁻) = ─────────────────────────
[H⁺]² + K₁[H⁺] + K₁K₂
α(H₂A) + α(HA⁻) + α(A²⁻) = 1
The pattern generalizes elegantly. The denominator for HnA has n + 1 terms, each a successive product of K's times a decreasing power of [H⁺], and each α is "its own term over the sum." This is a partition function in disguise. Compare it with the Boltzmann-weighted state populations from CHEM301: each protonation state has a "weight," and α is that weight divided by the sum of all weights.
Where pH = pKa, adjacent forms are present in equal amounts. For the intermediate form HA⁻ of a diprotic acid, a useful approximation (valid when the concentration F is not too small) is:
[H⁺] ≈ √[ (K₁K₂F + K₁K_w) / (K₁ + F) ] → pH ≈ ½ (pK₁ + pK₂)
This formula is exactly how you estimate the isoelectric point of a simple amino acid. That fact makes it one of the most important equations in this course for anyone heading toward peptide chemistry.
4. Gravimetric Analysis: The Balance as Ultimate Arbiter
Gravimetry is the oldest quantitative method and, done well, one of the most accurate. It is a primary method: it connects directly to the SI unit of mass without any calibration curve. The logic is simple. Convert the analyte quantitatively into a pure compound of known composition, then weigh it.
Gravimetric factor (example: Cl⁻ weighed as AgCl): mass Cl = mass AgCl × (M_Cl / M_AgCl) = mass AgCl × (35.453 / 143.32)
The hard part is producing a precipitate that is pure, filterable, and of exact stoichiometry. This is where the physical chemistry of nucleation enters:
Relative supersaturation = (Q − S) / S Q = actual solute concentration S = equilibrium solubility
- High supersaturation favors nucleation, producing many tiny particles: colloids that pass through filters and have huge surface areas for adsorbing impurities.
- Low supersaturation favors particle growth, producing fewer, larger, purer crystals.
Practical strategies follow directly from this [Harris & Lucy, 2020; Skoog et al., 2018]:
- Precipitate from hot, dilute solution.
- Add reagent slowly with stirring.
- Use homogeneous precipitation, in which the precipitant is generated slowly in situ. For example, urea hydrolyzing in hot water slowly raises the pH.
- Digest the precipitate: let it stand in hot mother liquor so small crystals dissolve and redeposit on large ones (Ostwald ripening), expelling trapped impurities.
Co-precipitation is the major source of error. Impurities can be carried down by adsorption on surfaces, inclusion (substituting into the crystal lattice), or occlusion (trapped pockets of solution). Inclusion is the hardest to remove, because the impurity is part of the crystal itself.
Colloidal precipitates such as AgCl are washed with a volatile electrolyte (dilute HNO₃) rather than pure water. Pure water would strip away the counter-ion layer that holds the coagulated particles together, causing peptization: the solid redisperses as a colloid and passes through the filter. The term has nothing to do with peptides, despite the name. Finally, drying or ignition converts the precipitate to a weighing form of definite composition.
5. Titrimetry: Counting Molecules with a Buret
5.1 The logic of a titration
A titration counts moles of analyte by measuring the volume of a standard reagent that reacts with it stoichiometrically. Distinguish two points:
- The equivalence point is the theoretical point at which stoichiometric amounts have reacted.
- The end point is what you actually observe: an indicator color change, a jump in potential, a break in a curve.
The difference between them is the titration error, and minimizing it is the art of the method.
5.2 Acid–base titration curves
Every titration curve can be computed with the systematic approach from Section 3. Harris shows that a single master equation covers all regions [Harris & Lucy, 2020], but the regional approximations build intuition. For a weak acid HA titrated with OH⁻:
- Before base is added: a weak-acid problem.
- Buffer region: Henderson–Hasselbalch applies, pH = pKa + log([A⁻]/[HA]). At the half-equivalence point, pH = pKa. This is the experimental route to measuring pKa.
- Equivalence point: the solution contains only A⁻, a weak base, so the pH is above 7.
- Past equivalence: excess OH⁻ dominates.
For polyprotic acids, each proton gives its own buffer region and break, provided successive pKa values are well separated. A common rule of thumb is a difference of at least 3–4 pK units. Where the pKa's overlap, the breaks merge.
5.3 Gran plots: linearizing the end point
Locating an end point from a sigmoidal curve depends on the steepest part of the curve, which is exactly where data are noisiest. Gran found a way around this [Gran, 1952]. Rearrange the equilibrium expression so that data before the end point fall on a straight line that extrapolates to Ve:
Weak acid HA titrated with strong base:
V_b · 10^(−pH) = (γ_HA / γ_A⁻) · K_a · (V_e − V_b)
Plot V_b·10^(−pH) vs. V_b → straight line
x-intercept = V_e
slope = −(γ_HA/γ_A⁻)·K_a
Here is the beauty of it. You use the well-behaved buffer-region data, fit a line with least squares, and extrapolate. No steep curve is needed. The method also gives you Ka from the slope.
6. EDTA Complexometric Titrations
EDTA (ethylenediaminetetraacetic acid, abbreviated H₄Y, with additional protonation possible on the nitrogens, up to H₆Y²⁺) is a hexadentate chelator. It wraps around a metal ion, binding through two nitrogens and four carboxylates, and forms 1:1 complexes with nearly every metal. That universal 1:1 stoichiometry is what makes it so useful.
6.1 Conditional formation constants
Only the fully deprotonated Y⁴⁻ binds the metal, so the effective binding strength depends on pH. Harris handles this with the fraction of EDTA present as Y⁴⁻ [Harris & Lucy, 2020]:
M^(n+) + Y⁴⁻ ⇌ MY^(n−4) K_f = [MY^(n−4)] / ([M^(n+)][Y⁴⁻]) [Y⁴⁻] = α_Y⁴⁻ · [EDTA] ([EDTA] = total uncomplexed EDTA) Conditional formation constant: K′_f = α_Y⁴⁻ · K_f = [MY^(n−4)] / ([M^(n+)][EDTA])
At pH 10, αY⁴⁻ ≈ 0.30. At pH 6 it is smaller by many orders of magnitude. This is a tunable selectivity knob. Metals with very large Kf (such as Fe³⁺) can be titrated at low pH, where weakly bound metals (such as Ca²⁺) are effectively invisible.
6.2 Auxiliary complexing agents
The problem is that many metals precipitate as hydroxides at the high pH needed for good EDTA binding. The solution is to add an auxiliary complexing agent, such as NH₃ for Zn²⁺. It binds the metal weakly enough to keep it in solution, but EDTA can still displace it:
α_M = [M^(n+)] / C_M = 1 / (1 + β₁[L] + β₂[L]² + β₃[L]³ + …) K″_f = α_M · α_Y⁴⁻ · K_f
This is the same "partition function" structure as the α fractions in Section 3, now applied to metal–ligand speciation. Once you see it, you will find it everywhere. Metal-ion indicators such as Eriochrome Black T work by binding the metal less strongly than EDTA does, so the color changes when EDTA strips the metal from the indicator at the equivalence point.
7. Redox Titrations, Formal Potential, and Electroanalytical Chemistry
7.1 The Nernst equation and formal potential
From CHEM102:
For aOx + ne⁻ ⇌ bRed:
E = E° − (RT/nF) ln( a_Red^b / a_Ox^a )
= E° − (0.05916 V / n) log( a_Red^b / a_Ox^a ) at 25 °C
Real solutions contain complexing anions, varying pH, and non-unit activity coefficients. Rather than tracking all of these, analysts use the formal potential E°′. This is the measured potential when the formal concentrations of Ox and Red are equal, under a specified medium. For example, the standard potential of Fe³⁺/Fe²⁺ is 0.771 V, but its formal potential is about 0.68 V in 1 M H₂SO₄. The difference arises because Fe³⁺ is complexed more strongly than Fe²⁺ by sulfate, which stabilizes the oxidized form and lowers E [Harris & Lucy, 2020]. E°′ folds activity coefficients and side reactions into one empirically useful number. It plays the same role as the conditional constant K′f for EDTA.
7.2 Redox titration curves
For Fe²⁺ titrated with Ce⁴⁺ (one electron each):
- Before equivalence: use the Fe³⁺/Fe²⁺ couple. At half-equivalence, E = E°′(Fe).
- At equivalence: E = ½[E°′(Fe) + E°′(Ce)]. For unequal electron numbers, use the weighted average (n₁E₁° + n₂E₂°)/(n₁ + n₂).
- After equivalence: use the Ce⁴⁺/Ce³⁺ couple.
Key reagents:
- Permanganate (MnO₄⁻) is intensely purple and serves as its own indicator. It is a powerful oxidant but not a primary standard; it must be standardized, for example against sodium oxalate.
- Dichromate (Cr₂O₇²⁻) is a primary standard, stable in solution, and somewhat weaker than permanganate. Its toxicity has reduced its use.
- Iodine methods. Iodimetry titrates a reducing analyte directly with I₃⁻. Iodometry lets an oxidizing analyte generate I₃⁻ from excess I⁻, then titrates that I₃⁻ with thiosulfate. Starch forms a deep blue complex with I₃⁻ and serves as the indicator.
- Redox indicators (for example, ferroin) are themselves redox couples. They change color over a range of about E°′In ± 0.05916/n V, so you choose one whose transition range brackets the equivalence potential.
7.3 Potentiometry: reference electrodes, indicator electrodes, and ISEs
A potentiometric cell pairs a reference electrode (constant potential, such as Ag/AgCl or saturated calomel) with an indicator electrode whose potential responds to the analyte. The measurement is made at essentially zero current, so the solution composition is not perturbed. You are reading an equilibrium property.
An unavoidable complication is the liquid junction potential. Ions diffuse across the boundary between the reference electrolyte and the sample at different rates, creating a small, uncertain potential that limits absolute accuracy, typically to a few millivolts.
Ion-selective electrodes (ISEs) are one of the most surprising ideas in this course. An ISE does not involve electron transfer at all. A membrane that selectively binds or transports one ion develops a phase-boundary potential through charge separation, which is the same physics as a biological membrane potential (BIOL101). The glass pH electrode is the classic example: ion exchange of H⁺ at hydrated silicate surfaces on either side of a thin glass membrane.
Glass pH electrode: E = constant − β (0.05916 V) · pH (25 °C; β ≈ 0.98–1.00, electromotive efficiency) General ISE with interfering ions X (Nikolsky–Eisenman form): E = constant ± (β · 0.05916 / n_A) · log[ 𝒜_A + Σ k_A,X · 𝒜_X^(n_A/n_X) ] 𝒜 = activity, k_A,X = selectivity coefficient (smaller = better)
A pH electrode measures hydrogen-ion activity, not concentration. Every pH reading you have ever taken is a thermodynamic measurement. That is precisely why the operational definition of pH rests on standard buffers.
The 59.16 mV per decade means a 1 mV error corresponds to about a 4% error in activity. This is the same logarithmic compression you saw in Section 1.4. Potentiometry is superb for wide dynamic range and only modest for precision. Glass electrodes also show alkaline (sodium) error at high pH, where Na⁺ begins to respond, and acid error at very low pH.
7.4 Voltammetry and coulometry
Potentiometry draws no current. Voltammetry deliberately drives electrochemistry by applying a potential and measuring the resulting current. The current is controlled by how fast analyte diffuses to the electrode, which links this topic to Fick's laws:
Cottrell equation (potential step, planar electrode, diffusion control): I(t) = n F A C √D / √(π t) Limiting current under steady-state diffusion: I_lim ∝ C
Because current is proportional to concentration, voltammetry is a quantitative method. In addition, the potential at which a wave appears identifies the species, so it also gives qualitative information. Cyclic voltammetry, which you may meet later, diagnoses reversibility and mechanism.
Coulometry is the electrochemical counterpart of gravimetry: it is absolute. You count electrons, and Faraday's law converts the count into moles with no calibration curve:
moles of analyte = Q / (n F) = (∫ I dt) / (n F) F = 96 485 C/mol
The classic application is coulometric Karl Fischer titration for trace water. Iodine is generated electrochemically from iodide and reacts stoichiometrically with water in the presence of SO₂ and a base [Fischer, 1935; Harris & Lucy, 2020]. Keep this in mind for the peptide section below.
8. Spectrophotometry: The Beer–Lambert Law and Its Limits
8.1 The law
A = −log₁₀ T = log₁₀(P₀ / P) = ε b c A = absorbance (dimensionless) T = transmittance = P/P₀ ε = molar absorptivity (M⁻¹ cm⁻¹) b = path length (cm) c = concentration (M)
8.2 Derivation, two ways
(a) Microscopic, molecular view. Consider a thin slab of thickness dx containing absorbers at number density N, each with capture cross-section σ. The fraction of photons removed is proportional to the fraction of the area "blocked":
dP / P = −σ N dx ⟹ P = P₀ exp(−σ N b) Converting to log₁₀ and molar units: A = (σ N_A / (2.303 × 1000 cm³/L)) · b · c ≡ ε b c
This is first-order decay, the same differential equation as radioactive decay or first-order kinetics (MATH102). Every layer of sample absorbs the same fraction of the light reaching it.
(b) From Maxwell's equations (PHYS202). In an absorbing medium, the refractive index becomes complex, ñ = n + iκ. A plane wave propagating in z becomes:
E(z,t) = E₀ exp[ i(ω ñ z / c − ω t) ]
= E₀ exp(−ω κ z / c) · exp[ i(ω n z / c − ω t) ]
└── decaying envelope ──┘
Intensity I ∝ |E|²:
I(z) = I₀ exp(−2ωκz / c) = I₀ exp(−4πκz / λ₀)
The absorption coefficient 4πκ/λ₀ is proportional to the concentration of absorbers, as long as each absorber contributes independently to the imaginary part of the medium's electric susceptibility. Beer's law then follows.
The Maxwell view also exposes a hidden assumption. The local field acting on a molecule depends on the medium's refractive index, so ε is not a pure molecular constant. Skoog notes that at high concentration ε should be replaced by εn/(n² + 2)², because the refractive index itself changes with concentration [Skoog et al., 2018]. Beer's law is a limiting law, exact only as c → 0, in the same spirit as the ideal gas law.
8.3 Sources of deviation
- Fundamental: at high concentration (roughly above 0.01 M), absorbers interact, which alters ε, and the refractive index changes.
- Chemical: the absorber participates in an equilibrium (acid–base, dimerization, complexation) whose position shifts with dilution. The analyte is "following Beer's law," but the species you think you are measuring is not the one actually present.
- Instrumental, polychromatic radiation: Beer's law holds strictly only for monochromatic light. If ε varies across the bandpass, the calibration curve bends toward the concentration axis. Measuring at an absorption maximum, where ε is flat, minimizes this.
- Instrumental, stray light: light reaching the detector without passing through the sample sets a ceiling on measurable absorbance.
- Photometric precision: uncertainty in transmittance propagates nonlinearly into concentration. Precision is poor at very low and very high A, and best at intermediate values. Classical analyses place the optimum somewhere around A ≈ 0.4–0.9, depending on the dominant noise source.
These categories follow Skoog's treatment [Skoog et al., 2018].
8.4 Multicomponent analysis: linear algebra in a cuvette
Absorbances are additive. At each wavelength, the total absorbance is the sum of each component's contribution:
A(λᵢ) = b · Σⱼ εⱼ(λᵢ) cⱼ Matrix form: A = K c (K_ij = b · εⱼ(λᵢ)) Square system (n wavelengths, n components): c = K⁻¹ A Overdetermined (more wavelengths than components): c = (KᵀK)⁻¹ Kᵀ A ← least-squares normal equations (MATH301)
This is your linear algebra course in action. Choose wavelengths where the component spectra differ most, so that K is well-conditioned and its columns are far from collinear. An isosbestic point, a wavelength where all spectra in a two-component interconversion cross, signals that exactly two absorbing species are interconverting with no intermediates. It is a fingerprint of a clean equilibrium.
9. Fluorimetry: Measuring Light Against Darkness
In fluorescence, a molecule absorbs a photon, relaxes vibrationally within the excited singlet state, and emits a lower-energy (red-shifted) photon. This shift is the Stokes shift. The quantum yield Φ is the fraction of absorbed photons re-emitted as fluorescence. A Jablonski diagram ties together everything from your quantum chemistry and photophysics background.
F = k · Φ · P₀ · (1 − 10^(−ε b c)) For low absorbance (εbc ≲ 0.01–0.05), expand 10^(−x) ≈ 1 − 2.303x: F ≈ 2.303 · k · Φ · P₀ · ε b c → linear in c
Why is fluorescence often 10–1000× more sensitive than absorbance? This is a beautiful point that is often stated imprecisely. In absorbance, you measure a small difference between two large signals, P₀ and P. Detecting a 0.1% dip in a bright beam is hard. In fluorescence, you measure emitted light, usually at 90° to the excitation beam and at a different wavelength, against an essentially dark background. It is like trying to spot one candle going out in a stadium full of floodlights versus spotting one candle lit in a dark stadium. A second advantage is that F scales with P₀, so a brighter source increases the signal, whereas absorbance is a ratio that brighter light does not improve [Skoog et al., 2018].
Fluorescence also offers two wavelengths, excitation and emission, which adds selectivity. Few molecules fluoresce strongly, so interferences are fewer.
Inner filter effects
At higher concentration, linearity fails for two reasons:
- Primary inner filter: the front of the cuvette absorbs so much excitation light that the observed volume in the center is underexcited.
- Secondary inner filter: emitted light is reabsorbed by the sample, wherever the absorption and emission spectra overlap.
Combined with quenching and self-association, these effects cause a fluorescence calibration curve to bend over and eventually decrease with concentration. A concentrated sample can then appear less fluorescent than a dilute one. Always keep absorbance low, and dilute if in doubt.
10. Chromatography: Separation as a Problem in Kinetics and Transport
10.1 Retention
Chromatography separates components by their differential partitioning between a mobile phase and a stationary phase. A solute that spends more time in the stationary phase moves more slowly.
Retention factor: k = (t_R − t_M) / t_M = (time in stationary) / (time in mobile)
= K_D · (V_s / V_m) (K_D = partition coefficient)
Separation factor: α = k₂ / k₁ (α > 1, by convention)
Retention is thermodynamics: it is set by the ΔG of transfer between phases (CHEM301). Peak width is kinetics and transport. Good separation needs both: peaks far apart and peaks narrow.
10.2 Plate theory and efficiency
Martin and Synge borrowed the idea of theoretical plates from distillation [Martin & Synge, 1941]. They imagined the column as N discrete equilibrium stages. Real columns have no plates, but N remains a useful measure of peak sharpness:
N = 16 (t_R / w)² = 5.55 (t_R / w_½)² (w = baseline width, w_½ = width at half height) N = L² / σ² (σ = standard deviation of band, length units) Plate height (HETP): H = L / N = σ² / L
The deeper picture is a random walk. Each molecule takes a random sequence of steps (adsorbing, desorbing, diffusing), so band variance grows linearly with distance traveled. This is the statistics of Section 1 again: independent variance contributions add, σ² ∝ L.
10.3 The van Deemter equation
H = A + B/u + C·u [van Deemter, Zuiderweg & Klinkenberg, 1956] u = linear flow velocity A = multiple paths (eddy diffusion): different routes around particles B/u = longitudinal diffusion: band spreads along the axis; B ∝ D_m C·u = mass transfer: equilibration between phases lags behind the moving band
The B term dominates at low flow, because the slower you go, the longer molecules have to diffuse. The C term dominates at high flow, because molecules cannot equilibrate fast enough. There is an optimal velocity at the minimum of H.
A practical consequence: smaller stationary-phase particles reduce both A and C, and they flatten the curve at high u. This is the reason behind modern sub-2 μm particles (UHPLC) and core–shell particles, which allow fast separations without much loss of efficiency. The cost is high back-pressure. In gas chromatography, diffusion coefficients are about 10⁴ times larger than in liquids, so the B term matters much more there.
10.4 Resolution: the master equation
R_s = Δt_R / w_av (R_s ≥ 1.5 ≈ baseline resolution)
R_s = (√N / 4) · ((α − 1) / α) · (k₂ / (1 + k₂))
efficiency selectivity retention
This equation is a strategy guide.
- Efficiency enters as √N. Doubling resolution through efficiency requires 4× the plates, which means 4× the column length and 4× the run time. Brute force is expensive.
- Retention saturates. Increasing k beyond about 5–10 gains little and costs time.
- Selectivity is the most powerful lever. Changing the chemistry (mobile-phase solvent, pH, stationary phase, temperature) to move α from 1.05 to 1.10 nearly doubles (α − 1)/α. A small change in chemistry often beats a large investment in hardware.
10.5 HPLC and reversed-phase chromatography
In reversed-phase HPLC, the stationary phase is nonpolar, typically C18 (octadecyl) chains bonded to silica. The mobile phase is polar: water mixed with acetonitrile or methanol. Hydrophobic solutes are retained longer. Increasing the organic fraction makes the mobile phase a "stronger" eluent and decreases k. For many solutes, log k varies roughly linearly with the organic fraction.
Isocratic elution faces the general elution problem. Conditions that resolve early peaks leave late peaks broad and slow, and conditions suitable for late peaks crush early peaks together. Gradient elution, which increases solvent strength over time, solves this. It also sharpens late peaks, because the trailing edge of a band sits in a stronger solvent than the leading edge and catches up.
For quantitation, peak area rather than height is usually used. Area is proportional to analyte amount and is less sensitive to peak broadening. An internal standard corrects for injection-volume variation [Harris & Lucy, 2020; Skoog et al., 2018].
10.6 Ion exchange and ion chromatography
Ion-exchange resins carry fixed charges: sulfonate groups on cation exchangers, quaternary ammonium groups on anion exchangers. Analyte ions compete with eluent ions for these sites. Selectivity generally increases with ion charge and, for a given charge, with decreasing hydrated radius.
Ion chromatography uses conductivity detection, but the high-conductivity eluent would swamp the signal. The clever fix is a suppressor, a membrane device placed after the column. For anion analysis, it converts the eluent (for example, KOH) into water, and converts analyte salts into their more conductive acid forms. Background drops and signal rises [Harris & Lucy, 2020].
Donnan exclusion also matters. Co-ions (same charge as the fixed sites) are excluded from the resin phase by an equilibrium membrane potential. This is the same physics behind ISE membranes and biological membrane potentials.
11. Sample Preparation and the Method of Standard Additions
"The analysis is only as good as the sample." Real samples such as blood, soil, cell lysate, or a crude synthesis mixture must be brought into a form the instrument can handle. This usually means dissolution, removal of interferences, and concentration.
11.1 Digestion
Inorganic analysis often requires destroying the organic matrix. This can be done with acid digestion (HNO₃, often in sealed microwave vessels), dry ashing, or fusion with molten flux for refractory materials. For proteins and peptides, a closely related step is acid hydrolysis (classically 6 M HCl, ~110 °C, ~24 h), which breaks peptide bonds before amino acid analysis.
11.2 Liquid–liquid extraction
Fraction of solute remaining in aqueous phase (volume V₁) after n extractions with fresh organic solvent (volume V₂ each), partition coefficient K = [S]_org/[S]_aq: q = V₁ / (V₁ + K V₂) Remaining after n extractions: qⁿ
A counterintuitive result follows. For a fixed total volume of extracting solvent, many small extractions beat one large one, because qⁿ falls faster than a single q computed with the combined volume. For ionizable analytes, the effective distribution coefficient D = αneutral·K depends on pH. Only the neutral form extracts into the organic phase. This is the fractional composition of Section 3 again, now acting as a "switch" you can flip with a buffer.
Solid-phase extraction (SPE) applies chromatographic principles to cleanup and preconcentration.
11.3 Standard addition
When the matrix changes the sensitivity, for example by quenching fluorescence or suppressing ionization in a mass spectrometer, a calibration curve prepared in pure solvent is wrong. In standard addition, known amounts of analyte are spiked into the sample itself, so the standard experiences exactly the same matrix as the unknown:
Single addition: I_X / I_(S+X) = [X]_i / ([S]_f + [X]_f) [X]_i = initial unknown concentration [X]_f = [X]_i · (V₀/V) [S]_f = [S]_i · (V_S/V) Multiple additions (Harris form): Plot I_(S+X)·(V/V₀) vs. [S]_i·(V_S/V₀) → x-intercept = −[X]_i
Know its limitations. Standard addition corrects for multiplicative matrix effects, which change the slope. It does not correct for additive interferences, such as a co-absorbing species that contributes a constant offset. It also assumes a linear response all the way down to zero.
12. Connections to Peptide Chemistry
Analytical chemistry is the hidden infrastructure of peptide science. Every claim in the field, such as "this peptide is 98% pure," "the IC₅₀ is 50 nM," or "the dose was 1 mg," rests on the measurement principles above. Here is where each topic leads:
- Polyprotic speciation, and charge as a function of pH. A peptide is a polyprotic acid with many ionizable groups: the N-terminus (pKa ~8–9), the C-terminus (~2–3), and side chains of Asp, Glu, His, Cys, Tyr, Lys, and Arg. Its net charge at any pH is the sum, over groups, of charge × α fraction, using exactly the formulas of Section 3.2. The isoelectric point is the pH where that sum equals zero. This single calculation predicts solubility minima, ion-exchange behavior, and electrophoretic mobility. Local environment shifts pKa values in folded structures, so this is the point where "textbook" pKa's meet real conditions.
- Reversed-phase HPLC is the peptide purity assay. Peptides are typically analyzed and purified on C18 columns using water/acetonitrile gradients containing ~0.1% trifluoroacetic acid. TFA acts as an ion-pairing agent: it protonates carboxylates and pairs with basic residues, which improves retention and peak shape. Peptides diffuse slowly (small D), so the van Deemter C term is large. Wide-pore (~300 Å) and core–shell particles are favored for larger peptides for exactly this reason. "Purity by HPLC" is usually area-percent at 214–220 nm, where the peptide bond absorbs. Be aware that this assumes all components absorb similarly per unit mass, which is a hidden approximation.
- UV quantitation via Beer–Lambert. For peptides and proteins containing Trp, Tyr, or cystine, ε280 can be estimated from composition. A widely used formula is ε280 ≈ 5500·nTrp + 1490·nTyr + 125·ncystine M⁻¹ cm⁻¹ [Pace et al., 1995]. A peptide with no aromatic residues is nearly invisible at 280 nm, which forces you to measure in the far UV or use other methods.
-
"Net peptide content" combines gravimetry, Karl Fischer, ion chromatography, and amino acid analysis. A lyophilized peptide powder is not 100% peptide. It contains bound water and counterions (often TFA or acetate from purification). Weighing 1.00 mg of powder may deliver only ~0.7–0.8 mg of actual peptide. Determining the true content requires a full mass balance:
- coulometric Karl Fischer titration for water (Section 7.4),
- ion chromatography for counterions (Section 10.6),
- amino acid analysis after acid hydrolysis (Section 11.1) for the absolute amount of peptide.
- Fluorescence. Tryptophan's intrinsic fluorescence is sensitive to its local environment, which makes it a probe of peptide folding and binding. Primary amines can be derivatized with reagents such as fluorescamine or o-phthalaldehyde (OPA) for sensitive quantitation. Inner filter corrections matter in binding titrations, where an absorbing ligand is added to a fluorescent peptide.
- Ion-exchange chromatography separates charge variants of peptides, such as deamidation products (Asn → Asp introduces a negative charge) and C-terminal processing variants. These are critical quality attributes in peptide therapeutics.
- Statistics and validation. Regulatory analytical methods for peptide drugs require documented accuracy, precision, linearity, LOD/LOQ, and robustness, which is Section 1 formalized into method-validation guidelines.
- Activity and ionic strength. Measured binding constants for charged peptides to charged targets depend strongly on ionic strength, through Debye screening. Reporting a Kd without the buffer composition is reporting half a result.
Looking ahead, mass spectrometry (LC-MS) will eventually become your primary identity tool for peptides. It builds directly on the chromatography in this course, and its quantitation relies on standard addition and stable-isotope internal standards, because electrospray ionization suffers severe matrix effects.
13. Common Misconceptions to Watch Out For
- "Precise means accurate." Tight replicates tell you nothing about bias. A beautifully reproducible method with a contaminated standard is reproducibly wrong. Accuracy can only be verified against an independent reference.
- "Averaging more replicates fixes everything." Averaging reduces random error as 1/√n and does nothing for systematic error. Past a certain point, more replicates are wasted effort.
- "An outlier test tells me which data to throw out." It flags statistical improbability. It does not establish a physical cause. With small n these tests have little power. Discarding data to improve agreement is a form of scientific misconduct if done without justification.
- "Adding an inert salt doesn't change anything." It changes the ionic strength, and with it every activity coefficient, solubility, and effective equilibrium constant. "Inert" salts are not inert in equilibrium.
- "A pH meter measures [H⁺]." It measures activity, subject to liquid-junction uncertainty. At high ionic strength, −log[H⁺] and measured pH can differ noticeably.
- "The equivalence point is where the indicator changes color." That is the end point. Choosing an indicator is precisely the task of making the end point coincide with the equivalence point. Also, the equivalence point of a weak acid titrated with strong base is not at pH 7.
- "E° is the potential that matters." In real media, the formal potential E°′ governs behavior, and it can differ from E° by tenths of a volt.
- "Beer's law is a law." It is a limiting law. It holds for dilute solutions, monochromatic light, and chemically stable absorbers. Always verify linearity over your working range.
- "Fluorescence is more sensitive because fluorophores absorb more strongly." The advantage comes from measuring light against a dark background, plus the ability to raise the signal with source power. Also, higher concentration does not always mean more fluorescence: inner filter effects can make the curve turn over.
- "To get better resolution, use a longer column." It works, but slowly, because Rs ∝ √N. Changing selectivity (α) through mobile phase, pH, or stationary phase is usually far more effective.
- "Faster flow always gives worse separation." The van Deemter curve has a minimum. Flowing too slowly also broadens peaks, through longitudinal diffusion.
- "Plates are physical things in the column." Theoretical plates are a mathematical abstraction. N is just another way of expressing band variance relative to retention.
- "Standard addition corrects for all interferences." It corrects slope-changing (multiplicative) matrix effects, not additive background from co-absorbing or co-eluting species.
- "A milligram of peptide powder is a milligram of peptide." Water and counterions can account for 20–30% of the mass. Always ask how "amount" was determined.
- "Uncertainty propagation rules always apply." The simple quadrature rules assume independent errors and small relative uncertainties (first-order Taylor expansion). Correlated measurements or large uncertainties need more careful treatment.
Closing Reflection
If you step back, a handful of ideas recur throughout this course, disguised in different notation:
- Fractional speciation, an "each term over the sum" partition function, governs acid–base α's, EDTA conditional constants, auxiliary complexation, and pH-dependent extraction.
- Logarithmic transduction, in pH, the Nernst equation, and absorbance, compresses huge dynamic ranges but amplifies relative error.
- Random walks and √n scaling govern confidence intervals, band broadening, and resolution.
- Linear models and least squares underlie calibration, Gran plots, multicomponent spectra, and standard addition.
- Activities, not concentrations, are what nature actually responds to.
Master these five patterns and the individual techniques become variations on a theme, rather than a list to memorize. That is the real goal of this course: not to know how to run an HPLC, but to know why it gives the number it gives, and how far to trust it.
References
- Dean, R. B.; Dixon, W. J. (1951). Simplified statistics for small numbers of observations. Analytical Chemistry, 23, 636–638.
- Debye, P.; Hückel, E. (1923). Zur Theorie der Elektrolyte. Physikalische Zeitschrift, 24, 185–206.
- Fischer, K. (1935). Neues Verfahren zur maßanalytischen Bestimmung des Wassergehaltes von Flüssigkeiten und festen Körpern. Angewandte Chemie, 48, 394–396.
- Gran, G. (1952). Determination of the equivalence point in potentiometric titrations. Part II. The Analyst, 77, 661–671.
- Grubbs, F. E. (1969). Procedures for detecting outlying observations in samples. Technometrics, 11, 1–21.
- Harris, D. C.; Lucy, C. A. (2020). Quantitative Chemical Analysis, 10th ed. W. H. Freeman.
- Kielland, J. (1937). Individual activity coefficients of ions in aqueous solutions. Journal of the American Chemical Society, 59, 1675–1678.
- Martin, A. J. P.; Synge, R. L. M. (1941). A new form of chromatogram employing two liquid phases. Biochemical Journal, 35, 1358–1368.
- Pace, C. N.; Vajdos, F.; Fee, L.; Grimsley, G.; Gray, T. (1995). How to measure and predict the molar absorption coefficient of a protein. Protein Science, 4, 2411–2423.
- Skoog, D. A.; Holler, F. J.; Crouch, S. R. (2018). Principles of Instrumental Analysis, 7th ed. Cengage Learning.
- Student [Gosset, W. S.] (1908). The probable error of a mean. Biometrika, 6, 1–25.
- van Deemter, J. J.; Zuiderweg, F. J.; Klinkenberg, A. (1956). Longitudinal diffusion and resistance to mass transfer as causes of nonideality in chromatography. Chemical Engineering Science, 5, 271–289.
⚙️ Mechanisms & Applications
Mechanisms & Applications: Turning Chemical Behavior into Defensible Measurements
Skipper’s central insight: analytical chemistry is not simply the chemistry of detecting substances. It is the design of a chain of inference: a sample produces a chemical or physical response; a model connects that response to an amount; controls establish whether the inference is trustworthy. The instrument never directly reports “the truth.” It reports a signal whose meaning depends on equilibria, transport, calibration, and sample history.
This builds directly on thermodynamics, organic chemistry, cell biology, and statistics. Thermodynamics tells us which species should exist; kinetics and transport determine whether the experiment reaches the assumed state; spectroscopy and electrochemistry convert molecular behavior into signals; statistics tells us how confidently we can interpret them. Harris and Lucy emphasize this quantitative chain, while Skoog, Holler, and Crouch develop the physical operation of the instruments. [1, 2]
1. The Measurement Process: Start with the Claim, Not the Instrument
- Define the measurand. Specify exactly what is being measured: total calcium, free Ca2+ activity, intact peptide concentration, or chromatographic purity. These are different quantities.
- Obtain a representative sample. A precisely measured aliquot cannot repair an unrepresentative blood draw, tissue specimen, or heterogeneous pharmaceutical powder.
- Preserve and prepare it. Prevent degradation, remove interfering material, and transfer the analyte into a measurable form.
- Generate a selective response. Precipitate it, titrate it, bind it, oxidize it, absorb light with it, or separate it from other components.
- Calibrate the response. Establish the relationship between signal and amount using appropriate standards.
- Challenge the result. Use blanks, duplicates, recovery experiments, reference materials, and independent checks.
- Report the result with its uncertainty and chemical meaning. A concentration without a stated basis, method, or relevant uncertainty may be misleading. [1]
Pharmaceutical example: “98% peptide purity by HPLC” usually means that a specified fraction of the integrated detector response belongs to the assigned main peak under that method. It does not automatically mean 98% peptide by mass. Water, counterions, residual solvents, nonabsorbing impurities, and unequal detector response factors can make the mass fraction substantially different.
2. Statistics and Quality Assurance: Distinguishing Scatter from Wrongness
Random error produces scatter; systematic error shifts results. Repeated measurements can make the mean very precise while leaving a calibration bias untouched. This is why replication and quality assurance solve different problems.
For n independent replicate measurements: x̄ = (Σxᵢ)/n s = [Σ(xᵢ − x̄)²/(n − 1)]¹ᐟ² Estimated standard uncertainty of the mean: u(x̄) = s/√n Two-sided confidence interval for a normal population mean: x̄ ± t₁₋α/₂, n₋₁ s/√n
The Gaussian model describes many accumulated random errors. The t-distribution accounts for estimating the population standard deviation from limited data. Its wider tails are the mathematical price of not knowing the true variance. A confidence interval describes the long-run coverage of the procedure; it does not assign a frequentist probability to the fixed true mean after measurement. [1]
Example: five independent assay results give a mean of 10.00 mg/L and a standard deviation of 0.10 mg/L. Using t = 2.776 for four degrees of freedom, the 95% interval is approximately 10.00 ± 0.12 mg/L. That interval represents replicate variability under the model—not necessarily uncertainty from stock-standard purity, recovery, or calibration bias.
First-order propagation for y = f(x₁, x₂, ...):
u²(y) ≈ Σ[(∂f/∂xᵢ)²u²(xᵢ)]
+ 2Σᵢ<ⱼ[(∂f/∂xᵢ)(∂f/∂xⱼ)cov(xᵢ,xⱼ)]
For C = Cstock Vpipette / Vflask,
with independent inputs:
[u(C)/C]² ≈ [u(Cstock)/Cstock]²
+ [u(Vpipette)/Vpipette]²
+ [u(Vflask)/Vflask]²
The covariance terms connect directly to linear algebra: uncertainty is a geometric object, not merely a list of error bars. If several standards share the same stock solution, their stock-concentration error is shared. Preparing more dilutions does not average that error away.
- Grubbs test: compares the largest standardized deviation, G = max|xi − x̄|/s, against an appropriate critical value. The usual single-outlier test assumes approximately normal data.
- Dixon tests: use gap-to-range statistics; the correct variant and critical value depend on sample size and test design.
- Good practice: investigate transcription errors, contamination, integration failures, and instrument faults. Never discard an inconvenient value merely to improve precision. Repeated testing without adjustment also increases false rejection.
- Quality-control samples: a reagent blank detects contamination; a matrix spike tests aspects of recovery and matrix response; a certified reference material tests agreement with an independently assigned value. None alone establishes universal method validity. [1]
Historical connection: William Sealy Gosset published the t-distribution in 1908 under the name “Student,” arising from practical small-sample problems in industrial work at Guinness. Rigorous inference was not born only from abstract mathematics; it also came from needing to make reliable decisions with expensive, limited experiments. [3]
3. Equilibrium Calculations: Count Everything, Then Impose Thermodynamics
A biological buffer or pharmaceutical formulation is a coupled system. Protonation changes metal binding; metal binding changes free ligand concentration; ionic strength changes activities. The reliable strategy is to write conservation laws before making approximations.
- List chemically plausible species.
- Write independent mass balances. Atoms or conserved chemical components cannot disappear.
- Write charge balance. A macroscopic bulk solution is electrically neutral to excellent approximation.
- Write equilibrium expressions using activities.
- Solve the coupled equations. Check that any neglected species are actually negligible.
For a solution containing only a monoprotic acid HA and water: Mass balance: C_A = [HA] + [A⁻] Charge balance: [H⁺] = [A⁻] + [OH⁻] Equilibria: K_a = a(H⁺)a(A⁻)/a(HA) K_w = a(H⁺)a(OH⁻) For a concentration standard state: aᵢ = γᵢ[i]/c° c° = 1 mol/L
The mass balance uses concentrations because it counts material. The equilibrium constant uses activities because it describes chemical potentials. This distinction is the direct application of what you learned in physical chemistry:
μᵢ = μᵢ° + RT ln aᵢ
Activity is not “a corrected count of molecules”; it represents their thermodynamic effectiveness in a specified environment.
Ionic strength: I = ½Σ cᵢzᵢ² Extended Debye–Hückel equation: log₁₀γᵢ = −A zᵢ²√I / (1 + B aᵢ,eff√I) aᵢ,eff is an effective ion-size parameter. A and B depend on solvent, temperature, and unit convention.
Debye and Hückel’s 1923 theory explains why an ion’s surrounding ionic atmosphere changes its chemical potential. The squared charge is consequential: divalent ions often show stronger nonideality than monovalent ions. The extended equation is a low-to-moderate ionic-strength approximation, not a universal model for concentrated buffers or physiological media. Moreover, individual-ion activities require conventions; directly measurable thermodynamic combinations include mean ionic activities. [1, 4]
Polyprotic acid fractionation turns a complicated species distribution into a useful map. For a diprotic acid, under an ideal concentration approximation—or using consistent conditional constants at fixed ionic strength:
Let h = [H⁺] and D = h² + Kₐ₁h + Kₐ₁Kₐ₂. α₀ = [H₂A]/C_A = h²/D α₁ = [HA⁻]/C_A = Kₐ₁h/D α₂ = [A²⁻]/C_A = Kₐ₁Kₐ₂/D α₀ + α₁ + α₂ = 1
Application: these fractions help predict buffer behavior, drug extraction, and the availability of a particular ligand form for binding. A peptide’s charge distribution also changes with pH, but real peptide protonation sites can interact; treating every residue as an independent textbook amino acid is an approximation.
4. Gravimetry and Titrimetry: Make Stoichiometry Observable
These classical methods are powerful because their signals—mass and delivered reagent amount—connect directly to stoichiometry. But the reaction product or endpoint must genuinely represent the analyte.
Precipitation gravimetry proceeds through a controlled material transformation:
- Dissolve the sample and adjust pH and solution composition to favor selective precipitation.
- Add precipitant under conditions that control supersaturation. Slow addition, stirring, and suitable temperature often favor larger, more filterable particles.
- Digest the precipitate, allowing dissolution and redeposition to improve particle size and sometimes purity.
- Filter and wash without dissolving excessive analyte or dispersing colloidal particles.
- Dry or ignite to a chemically defined, stable weighing form.
- Cool appropriately and weigh to constant mass. [1]
Example: sulfate determined as BaSO₄ Ba²⁺ + SO₄²⁻ → BaSO₄(s) m(SO₄²⁻) = m(BaSO₄) × M(SO₄²⁻)/M(BaSO₄)
The competing errors are instructive: finite solubility causes analyte loss, whereas adsorption, occlusion, and inclusion can add foreign material. A common ion can lower solubility, but complex formation can raise it. “Add a large excess of precipitant” is therefore not automatically sound advice.
Acid–base titration replaces weighing a product with counting the reagent needed to transform the analyte.
For a monoprotic weak acid titrated with strong base: Before equivalence, in the useful buffer region: pH ≈ pKₐ + log₁₀(n_A⁻/n_HA) At half-equivalence: pH ≈ pKₐ At equivalence: A⁻ + H₂O ⇌ HA + OH⁻ K_b = K_w/K_a After equivalence: Excess added OH⁻ usually dominates the pH.
Each region is governed by different chemistry. There is no single shortcut valid across the whole titration curve. Polyprotic acids give distinguishable endpoints only when successive equilibria and experimental conditions provide sufficient separation. Closely spaced pKa values produce overlapping chemistry rather than clean stoichiometric steps.
Gran plots use a transformed signal to estimate an equivalence volume by extrapolation. For a strong acid titrated with strong base, before equivalence and where water autoionization is negligible:
V₀ = initial sample volume V = added base volume Vₑ = equivalence volume C_b = base concentration [H⁺] ≈ C_b(Vₑ − V)/(V₀ + V) If activity effects are negligible: (V₀ + V)c°10^(−pH) ≈ C_b(Vₑ − V) Plot the left-hand quantity against V. The extrapolated x-intercept estimates Vₑ.
A nearly constant hydrogen-ion activity coefficient changes the slope rather than the ideal intercept; changing activity coefficients and electrode errors can introduce curvature. Gran’s mid-twentieth-century contribution was to show how mathematical transformation can extract an endpoint without relying only on locating the steepest part of a curve. [1]
Applications: titration remains useful for raw-material assays, acid or base capacity of formulations, and buffer characterization. In peptide manufacturing, a titration may measure total proton-binding capacity rather than selectively measuring intact peptide—an important limitation in mixtures.
5. EDTA Titration: Strong Binding Is Conditional on the Solution
EDTA forms many metal complexes with 1:1 metal-to-ligand stoichiometry. Its apparent simplicity hides a competition: EDTA can bind protons, the metal can bind other ligands, and the metal may precipitate as a hydroxide.
Core equilibrium: Mⁿ⁺ + Y⁴⁻ ⇌ MY⁽ⁿ⁻⁴⁾ K_f = [MY]/([M][Y⁴⁻]) At fixed pH: α_Y = [Y⁴⁻]/C_Y,free If only ligand protonation modifies binding: K_f′ = K_f α_Y If α_M = [M]/C_M,uncomplexed accounts for metal side reactions: K_f″ = K_f α_Y α_M
Here the constants are written in concentration form for a specified medium, and the simplified model treats one principal metal–EDTA complex. The free-metal fraction excludes metal already bound to EDTA.
- Buffer the sample to obtain adequate EDTA binding without unwanted precipitation.
- Add a metallochromic indicator that binds the metal less strongly than EDTA does under those conditions.
- Add standardized EDTA. It lowers free-metal concentration as metal–EDTA complex forms.
- Near the endpoint, EDTA removes metal from the indicator complex, changing the indicator’s color.
- Calculate metal amount from the 1:1 reaction, while accounting for interfering metals. [1]
Metal titration curves are often plotted as pM = −log10[M]. The conditional formation constant controls how sharply free-metal concentration changes near equivalence. An auxiliary ligand can keep the metal soluble, but by competing for it, that same ligand can weaken the endpoint. Analytical method design is often the management of such tradeoffs.
Applications: water-hardness analysis and selected calcium or magnesium assays. EDTA also prevents clotting in blood collection by sequestering calcium. Consequently, EDTA-contaminated specimens are unsuitable for many calcium measurements: sample preservation has chemically changed the measurand.
6. Redox and Electroanalytical Methods: Measure Electron Transfer or Its Driving Force
The bridge from thermodynamics to measurement is:
ΔG = −nFE For Ox + ne⁻ ⇌ Red: E = E° − (RT/nF) ln[a(Red)/a(Ox)] At 25 °C: E = E° − (0.05916 V/n) log₁₀[a(Red)/a(Ox)]
Redox titration counts electron equivalents through a chemical reaction. Acidic permanganate is both oxidant and intensely colored endpoint reagent; dichromate provides a useful, stable oxidant but introduces serious Cr(VI) hazards. Proton concentration matters because many redox half-reactions consume or generate H+.
In sufficiently acidic solution: MnO₄⁻ + 8H⁺ + 5e⁻ → Mn²⁺ + 4H₂O Iodimetry: Direct titration of a reductant with iodine. Iodometry: An oxidant liberates iodine from iodide; the iodine is then commonly titrated with thiosulfate. I₂ + 2S₂O₃²⁻ → 2I⁻ + S₄O₆²⁻
Iodine methods are useful in selected antioxidant, oxidant, and pharmaceutical assays. Starch gives a sensitive iodine-associated color, but its addition and the solution conditions must follow the method; strongly bound iodine can cause troublesome endpoint behavior.
A formal potential, E°′, incorporates specified medium-dependent effects such as complexation and fixed pH into a practical potential expression. It is not necessarily the thermodynamic standard potential. A redox indicator changes its oxidation-state-dependent color over a potential range; that range should lie within the sharp endpoint region. [1, 2]
Potentiometry measures voltage with negligible current:
- A reference electrode, such as Ag/AgCl, supplies a reproducible reference potential.
- An indicator electrode responds to the analyte’s activity or to a redox couple.
- A high-input-impedance meter minimizes current draw and electrode polarization.
- Calibration and controlled conditions translate the measured cell potential into an analytical result.
Ideal ion-selective response, with a defined cell polarity: E = K + (RT/zF) ln aᵢ For a conventional pH-cell polarity: E = K − (2.303RT/F)pH pH = −log₁₀a(H⁺) Ideal pH slope magnitude at 25 °C: 59.16 mV per pH unit
A glass pH electrode develops ion-selective boundary potentials at hydrated glass surfaces. It does not determine pH by electrolyzing hydrogen ions. Calibration with buffers is essential because the measured cell voltage includes more than the ideal membrane response, including reference and junction contributions. Protein fouling, temperature mismatch, dehydration, and alkaline-ion interference can all matter.
Medicine makes the distinction concrete: ion-selective electrodes can measure ionized calcium, whereas a total-calcium method includes calcium bound to proteins and other ligands. Different results need not indicate instrument disagreement; the methods may be measuring different chemical pools.
Voltammetry deliberately varies electrode potential and measures current. Reaction at the surface changes nearby concentrations, so diffusion, convection, and migration become central. Supporting electrolyte suppresses much of the migration contribution; controlled hydrodynamics or timing makes transport interpretable.
Cottrell equation for a diffusion-controlled potential step: i(t) = nFAD¹ᐟ²C*/(πt)¹ᐟ² Assumptions include planar, semi-infinite diffusion, negligible convection, and sufficiently rapid electrode reaction.
Coulometry integrates current to count charge:
Q = ∫i(t)dt Amount of analyte = Q/(nF) This requires the specified electron stoichiometry and effectively quantitative current efficiency.
Coulometric Karl Fischer analysis generates iodine electrochemically to determine water in an appropriate reaction medium. It is especially valuable for low-moisture pharmaceutical materials, including peptide products, although interfering sample chemistry must be assessed.
Historical connection: Faraday’s laws of electrolysis established the quantitative relationship between charge and chemical transformation. Nernst connected electrode potential to chemical state. Heyrovský’s development of polarography made current–potential behavior a practical analytical tool. These are three distinct achievements: counting electrons, measuring their driving force, and using their transport-dependent flow as a signal. [1, 2]
7. Absorption Spectrophotometry: From Electromagnetic Waves to Concentration
Beer–Lambert behavior follows when each additional thin layer removes the same fraction of the surviving light, with absorbers behaving independently.
dI/dz = −βI I(b) = I₀ exp(−βb) A = log₁₀(I₀/I) = βb/ln 10 If β = ln(10) εc: A = εbc
The connection to electromagnetism is deeper than a memorized exponential. For a monochromatic wave in a homogeneous absorbing medium, Maxwell’s equations give a complex refractive index, ñ = n + iκ, with the sign chosen consistently with the wave convention:
Field-amplitude decay: |E(z)| ∝ exp(−κωz/c₀) Because intensity is proportional to |E|²: I(z) = I₀ exp(−2κωz/c₀) Thus: β = 2κω/c₀ = 4πκ/λ₀ c₀ = speed of light in vacuum λ₀ = vacuum wavelength
Maxwell’s equations establish propagation and attenuation once the material response is specified. They do not, by themselves, prove that attenuation is proportional to solute concentration. That additional step assumes a suitable dilute, independent-absorber regime. [2]
- Select a wavelength with useful analyte absorption and manageable interference.
- Use a blank to establish the solvent, reagent, and cell baseline.
- Measure standards within a validated range.
- Measure the sample under matched conditions.
- Check dilution linearity and whether chemical speciation changes during dilution.
- Chemical deviations: protonation, binding, aggregation, or reaction changes the absorbing species.
- Instrumental deviations: finite spectral bandwidth and stray light disrupt the ideal model; stray light commonly produces negative deviations at high absorbance.
- Physical interference: particles scatter light, and concentrated solutions may no longer behave as independent absorbers.
Multicomponent analysis connects directly to linear algebra:
At wavelength λᵢ: Aᵢ = b Σⱼ εᵢⱼcⱼ Matrix form: A = bEc Weighted least-squares estimate: ĉ = (EᵀWE)⁻¹EᵀWA/b Requires adequate rank and a suitable error model.
If two components have nearly proportional spectra, the matrix becomes poorly conditioned. Even excellent measurements may not distinguish them reliably. More decimal places do not create chemical selectivity.
Biochemical applications: NADH can be monitored near 340 nm to follow enzyme reactions; protein estimates near 280 nm depend strongly on aromatic residues and, to a lesser extent, disulfide contributions. Peptide bonds absorb strongly in the far UV, supporting peptide HPLC detection, but solvents and additives can also absorb there. A universal “protein” extinction coefficient is not appropriate for arbitrary therapeutic peptides.
Historical connection: Bouguer and Lambert developed attenuation relationships with path length; Beer established the concentration dependence in 1852. Their combined law matters because it transforms an optical ratio into chemical amount—within a defined physical and chemical regime. [1, 2]
8. Fluorimetry: Observe the Fate of an Excited Molecule
- A fluorophore absorbs light and enters an excited electronic state.
- Rapid vibrational relaxation usually dissipates some excitation energy.
- Fluorescence competes with nonradiative relaxation, intersystem crossing, and quenching.
- The instrument separates the emitted light from the excitation light, often using approximately right-angle collection geometry.
Fluorescence quantum yield: Φ_F = photons emitted / photons absorbed For a simple competing-rate model: Φ_F = k_F/(k_F + k_nr + k_ISC + ...) In an ideal geometry: F ∝ Φ_F I₀(1 − 10⁻ᴬ) For A ≪ 1: F ∝ 2.303 Φ_F I₀ εbc
Fluorescence can achieve lower detection limits than absorption because it measures emitted light against a relatively dark background, rather than detecting a small difference between two large light intensities. But high sensitivity does not imply universal applicability: many analytes do not fluoresce usefully, and labeling may alter their behavior.
Inner-filter effects occur when the sample substantially absorbs excitation light before it reaches the observed region, or reabsorbs emitted light. Dilution can therefore increase fluorescence per mole without changing the molecule’s intrinsic quantum yield. This is an optical artifact, not necessarily molecular quenching.
Applications: fluorescent peptide substrates, protease assays, binding measurements, and derivatization-based amino-acid analysis. In a fluorogenic enzyme assay, controls must establish that the signal change represents chemical conversion rather than pH shifts, compound autofluorescence, quenching, or inner-filter effects. Stokes’s nineteenth-century fluorescence studies established the characteristic shift between excitation and emission wavelengths that enables much of this optical discrimination. [1, 2]
9. Chromatography: Separate by Repeated Partitioning, Then Quantify
A chromatographic column acts like a sequence of repeated opportunities to enter and leave a stationary environment. Molecules spending more time in that environment travel more slowly. Selectivity creates separation; band spreading erodes it.
Retention factor: k = (t_R − t_M)/t_M Plate number for a roughly Gaussian peak: N = 16(t_R/w_b)² or N = 5.54(t_R/w₁/₂)² Plate height: H = L/N Van Deemter equation: H = A + B/u + Cu
- A term: different flow paths through a packed bed create dispersion.
- B/u term: longitudinal diffusion is more damaging at slow flow because molecules have more time to spread.
- Cu term: finite mass transfer prevents complete local equilibration at high flow.
There is an optimum velocity because going slowly improves equilibration but worsens diffusion. Going fast suppresses diffusion but worsens mass-transfer broadening. Smaller particles generally improve efficiency at a substantial pressure cost.
Approximate isocratic resolution relationship: R_s ≈ (√N/4) [(α − 1)/α] [k₂/(1 + k₂)] α = k₂/k₁ > 1
The square-root dependence is humbling: doubling resolution by efficiency alone requires roughly quadrupling the plate count. Changing selectivity through pH, stationary phase, or solvent may be much more effective. This approximation assumes suitably behaved peaks and is not a universal gradient-elution equation. [1, 2]
Reversed-phase HPLC of peptides:
- Dissolve the sample in a compatible solvent, avoiding conditions that cause precipitation or severe injection-band distortion.
- Inject onto a hydrophobic stationary phase such as an appropriate C18 or C8 material.
- Begin with a sufficiently aqueous mobile phase to retain the target compounds.
- Increase organic-solvent content in a gradient so differently retained peptides elute within a practical time.
- Detect and integrate peaks; use standards and response relationships for quantitative claims.
- Evaluate closely related impurities and, where needed, use orthogonal identification methods.
pH changes peptide charge and surface interactions; additives may change ion pairing, peak shape, and downstream detector compatibility. Retention is therefore not governed by hydrophobicity alone. Deletion sequences, oxidized products, and epimers can be difficult to resolve. A single symmetric peak is not proof of chemical identity or homogeneity.
Ion exchange uses charged stationary groups: cation exchangers retain cations, whereas anion exchangers retain anions. Salt competes for binding and screens electrostatic interactions; pH changes analyte charge. These mechanisms are useful for charged biomolecules and pharmaceutical counterions.
Ion chromatography often separates inorganic ions by ion exchange and detects them by conductivity. In suppressed-conductivity methods, the suppressor lowers eluent background and converts analytes into highly conductive forms. In a common anion method, hydroxide eluent becomes water while analyte salts are converted to their corresponding acids. [1, 2]
Historical connection: Mikhail Tswett demonstrated early column chromatography with plant pigments. Martin and Synge’s partition chromatography established a foundational separation framework; van Deemter, Zuiderweg, and Klinkenberg’s 1956 treatment made the sources of band broadening quantitatively understandable. Chromatography evolved from observing colored bands to engineering separation performance. [2, 5]
10. Sample Preparation and Standard Addition: The Matrix Is Part of the Experiment
Sample preparation is where chemical selectivity often begins—and where analyte loss can become invisible.
Liquid–liquid extraction: D = C_analyte,organic,total / C_analyte,aqueous,total Fraction remaining in the aqueous phase after one extraction: q = V_aq/(V_aq + D V_org) After n identical fresh-solvent extractions: qⁿ Assumes equilibrium and effectively constant D and phase volumes.
For ionizable compounds, the distribution ratio depends on pH because neutral and charged forms partition differently. Repeated smaller extractions often outperform one extraction using the same total solvent volume. For peptides, however, adsorption, aggregation, and interfacial losses may complicate this simple model.
- Solid-phase extraction: condition the sorbent, load the sample, wash away selected interferences, and elute the retained analyte.
- Protein precipitation: removes much of a biological matrix, but can also carry away bound analyte.
- Digestion: converts a complex matrix into a form suitable for total-element measurement. It usually destroys the original molecular identity or speciation.
- Recovery studies: compare known additions with observed increments. A freshly added spike may not behave like analyte that has long been bound or embedded in the original material. [1, 2]
Standard addition calibrates within the sample matrix when that matrix changes sensitivity:
For equal sample aliquots brought to the same final volume, after appropriate blank correction: S = k_matrix(C_x,final + C_added,final) Plot S against C_added,final. x-intercept = −C_x,final Correct for dilution to recover the original sample concentration.
This works when added and native analyte share the same response, the response is linear, and the additions do not materially change the matrix. It does not automatically remove an additive spectral interference, restore analyte lost before spiking, or reveal whether a spike equilibrated with strongly bound native analyte.
Closing Connection: Orthogonal Measurements Answer Different Questions
For a therapeutic peptide, a defensible analytical picture may combine HPLC for separation, calibrated UV detection for amount, Karl Fischer analysis for water, ion chromatography for counterions, and an appropriate biochemical assay for activity. Identity and structural integrity require additional evidence. These measurements are complementary because they probe different properties.
The habit to retain: before accepting a result, ask: “Which species produced this signal? What process connected that species to the original sample? Which assumptions make the calculation valid? What plausible failure would my controls fail to detect?” That is the difference between operating an instrument and practicing analytical chemistry.
References
- Harris, D. C.; Lucy, C. A. Quantitative Chemical Analysis, 10th ed.; W. H. Freeman, 2020. Principal source for analytical statistics, quality assurance, equilibrium calculations, gravimetry, titrimetry, electrochemistry, spectroscopy, separations, and sample preparation.
- Skoog, D. A.; Holler, F. J.; Crouch, S. R. Principles of Instrumental Analysis, 7th ed.; Cengage Learning, 2018. Principal source for electroanalytical instrumentation, optical methods, fluorescence, and chromatographic theory and instrumentation.
- Student. “The Probable Error of a Mean.” Biometrika 1908, 6(1), 1–25. doi:10.1093/biomet/6.1.1.
- Debye, P.; Hückel, E. “Zur Theorie der Elektrolyte. I. Gefrierpunktserniedrigung und verwandte Erscheinungen.” Physikalische Zeitschrift 1923, 24, 185–206.
- van Deemter, J. J.; Zuiderweg, F. J.; Klinkenberg, A. “Longitudinal Diffusion and Resistance to Mass Transfer as Causes of Nonideality in Chromatography.” Chemical Engineering Science 1956, 5(6), 271–289. doi:10.1016/0009-2509(56)80003-1.
🪬 Skipper's Reflection
Skipper's Reflection: Analytical Chemistry — Principles of Chemical Measurement
When I first looked at the syllabus for CHEM302, I thought it would be a straightforward review of Gen Chem lab techniques with a bit of statistics sprinkled on top. I was entirely wrong. If physical chemistry (CHEM301) is the underlying rulebook of the universe, analytical chemistry is the gritty, rigorous discipline of figuring out how to actually read the score. Moving from idealized models to the messy reality of chemical measurement has been one of the most humbling and enlightening steps in my education so far.
1. What Surprised Me Most
What surprised me most was the realization that "concentration" as I learned it in Gen Chem is essentially a lie—or rather, an idealized approximation. When we studied activity and activity coefficients (Harris & Lucy, Chapter 8), it blew my mind to learn that a 0.1 M solution of a salt doesn't actually behave like a 0.1 M solution. The concept of ionic atmosphere and shielding completely changes how molecules interact. The extended Debye-Hückel equation felt like lifting the veil on why real-world pH measurements or equilibrium calculations never perfectly matched my theoretical Gen Chem predictions.
I was equally surprised by how deeply statistical this course was. Having taken MATH201 (Probability and Statistics), I knew about Gaussian distributions, but applying propagation of uncertainty to every step of a multi-stage gravimetric analysis or standard addition protocol was eye-opening. Every time I measure a liquid in a volumetric flask, I am not creating a fixed value; I am creating a probability distribution.
2. Conceptual Breakthroughs
My biggest "click" moment came when we tackled systematic equilibrium calculation for polyprotic acids. In Gen Chem, we used ICE tables and made approximations to ignore the dissociation of water or minor species. Here, Harris teaches us to write out the mass balance and charge balance equations. Suddenly, I realized: this is just linear algebra. It’s a system of $n$ equations and $n$ unknowns! I could perfectly map this back to MATH301. By setting up the charge balance and fractional composition ($\alpha$) equations, the chemical system became a pure mathematical matrix that could be solved exactly. It was a beautiful fusion of math and chemistry.
Another major breakthrough was the van Deemter equation in chromatography theory (Skoog et al., Chapter 26). Before this, column chromatography just felt like pouring liquid through sand and hoping things separated. The van Deemter equation ($H = A + B/u + Cu$) mathematically deconstructs band broadening into physical phenomena: multiple flow paths (eddy diffusion), longitudinal diffusion, and the kinetics of mass transfer between the stationary and mobile phases. Realizing that there is an optimal flow velocity where these opposing physical forces minimize the theoretical plate height (HETP) was a massive conceptual victory. It turned an empirical lab chore into an applied physics problem.
3. How This Knowledge Changes How I Think About Peptides
This course completely rewired how I view our peptide operations, particularly regarding synthesis QC and biological testing for molecules like Tirzepatide, Retatrutide, BPC-157, and TB-500.
First, Reversed-Phase HPLC (RP-HPLC) is the absolute lifeblood of the peptide industry. Before this course, I just knew we ran peptides through a C18 column to check purity. Now I understand the thermodynamics behind gradient elution. Molecules as massive and hydrophobic as Retatrutide will bind so tightly to the nonpolar C18 stationary phase that an isocratic (constant) mobile phase would either never elute them or cause severe band broadening due to excessive retention times. By slowly increasing the organic solvent (like acetonitrile), we continuously alter the partition coefficient ($K$) during the run. This is exactly how we resolve a 39-amino acid target peptide from a failure sequence that is missing just one amino acid.
Second, analyzing peptides in biological matrices (like tracking BPC-157 in serum) makes total sense now in the context of the standard addition method. Blood is a complex matrix that can suppress or enhance analytical signals (matrix effects). By spiking the sample with known amounts of the standard, we calibrate the instrument in the presence of the interference.
Finally, viewing peptides as massive polyprotic acids has shifted my perspective. Using the systematic equilibrium models from this course, I can better understand the isoelectric point (pI) of these molecules, which dictates their solubility and how they must be formulated (e.g., utilizing zinc-chelation, similar to EDTA complexometric titrations, to stabilize hexameric structures in certain peptide formulations).
4. Questions This Course Raises for Future Investigation
Studying the physical limits of resolution ($R_s$) in chromatography leaves me with a burning practical question: How do we push HPLC to reliably resolve chiral stereoisomers of large peptides? If a batch of Tirzepatide has a single D-amino acid instead of an L-amino acid at one position out of 39, the physical properties are nearly identical. What stationary phase modifications or mobile phase additives are required to force a difference in the partition coefficient for something that subtle?
Additionally, while we touched on the Beer-Lambert law and spectrophotometry, I am eager to dive deeper into how this integrates with Mass Spectrometry (which I know is coming in later studies). How do we interface the liquid phase of HPLC with the vacuum phase of an MS system without destroying the quantitative rigor we established in this course?
5. Confidence Rating
Most Confident:
I feel highly confident in systematic equilibrium calculations, statistical analysis, and HPLC theory. My background in linear algebra and statistics made the math of error propagation and matrix solving feel natural, and the van Deemter equation clicked perfectly with my thermodynamics intuition from PHYS201 and CHEM301.
Least Confident:
I am least confident about electroanalytical chemistry and potentiometry. While I understand the Nernst equation on paper, the physical reality of junction potentials, the inner workings of ion-selective electrodes, and the nuances of the glass pH electrode still feel a bit like black magic. The idea that a hydration gel layer on a piece of glass creates a measurable boundary potential that corresponds perfectly to $H^+$ activity is something I need to spend more time visualizing to truly trust.
🗺️ Visual Aids & Key Takeaways
Diagram 1: The Analytical Process — From Sample to Result
╔══════════════════════════════════════════════════════════════════════════════════╗
║ THE ANALYTICAL MEASUREMENT PROCESS ║
╚══════════════════════════════════════════════════════════════════════════════════╝
┌─────────────────┐
│ REAL-WORLD │ ← the population you actually care about
│ SAMPLE │ (river water, blood serum, pharmaceutical tablet)
└────────┬────────┘
│ representative sampling — the hardest step; no math fixes a bad sample
↓
┌─────────────────┐
│ SAMPLE │ ← digestion (acid, fusion), extraction, dilution,
│ PREPARATION │ filtration, pre-concentration
└────────┬────────┘
│ matrix is simplified; analyte is accessible
↓
┌─────────────────┐
│ MEASUREMENT │ ← spectrophotometry, titrimetry, chromatography,
│ METHOD │ potentiometry, gravimetry, etc.
└────────┬────────┘
│ raw signal (absorbance, volume, current, mass)
↓
┌─────────────────┐
│ CALIBRATION │ ← external standards, standard addition,
│ & CALCULATION │ internal standards; Beer-Lambert, Nernst, etc.
└────────┬────────┘
│ concentration or amount with uncertainty
↓
┌─────────────────┐
│ STATISTICAL │ ← t-tests, confidence intervals, propagation of error,
│ EVALUATION │ outlier rejection (Grubbs/Dixon), quality control charts
└────────┬────────┘
│
↓
┌─────────────────┐
│ REPORTED │ e.g., Cu²⁺ = 0.142 ± 0.003 mg/L (95% CI, n=6)
│ RESULT │
└─────────────────┘
KEY QUALITY CONCEPTS:
┌──────────────────┬────────────────────────────────────────────────────────┐
│ ACCURACY │ closeness to true value — checked with certified │
│ │ reference materials (CRMs) or spike recovery │
├──────────────────┼────────────────────────────────────────────────────────┤
│ PRECISION │ reproducibility — quantified by standard deviation s │
├──────────────────┼────────────────────────────────────────────────────────┤
│ SENSITIVITY │ change in signal per unit concentration (slope of │
│ │ calibration curve) │
├──────────────────┼────────────────────────────────────────────────────────┤
│ DETECTION LIMIT │ 3s_blank / sensitivity — lowest detectable signal │
└──────────────────┴────────────────────────────────────────────────────────┘
Maps the complete journey from field sample to reported result, emphasizing that each step introduces its own error and that sampling error often dominates all others.
---Diagram 2: Titration Curve Landscape — Four Titration Types Compared
╔══════════════════════════════════════════════════════════════════════════════════╗
║ TITRATION CURVE SHAPES AND WHAT CONTROLS THEM ║
╚══════════════════════════════════════════════════════════════════════════════════╝
pH
14 │
│
12 │ STRONG ACID / STRONG BASE WEAK ACID / STRONG BASE
│ (e.g., HCl + NaOH) (e.g., acetic acid + NaOH)
10 │ ╭────── ╭──────
│ │ ───────╯
8 │ │ ←─ half-equivalence point
│ ──────╯ pH = pKa here
6 │ ────╯ ──────╯ ← buffer region
│ ────╯
4 │ ───╯
│
2 │
└──────────────────────────────────────────────────────────────────── V(mL)
0 Ve/2 Ve Ve/2 Ve
CONTROLS ON JUMP SIZE:
┌─────────────────────────────────────────────────────────────────────────────┐
│ Concentration ↑ → steeper jump at equivalence point │
│ Ka (or Kb) ↑ → sharper equivalence point, larger jump │
│ Ka very small → jump nearly disappears; titration not feasible │
│ (Ka < ~10⁻⁹ for 0.1 M solutions) │
└─────────────────────────────────────────────────────────────────────────────┘
POLYPROTIC ACID (e.g., H₃PO₄ + NaOH):
pH
12 │ ╭───────
│ ──────────╮╯
10 │ ────────╮╯ │ ← 3rd eq. pt
│ │ ↑ │
8 │ ────────╮╯ │ 2nd eq. pt │
│ │ │ │ │
6 │ ──────╮╯ │ │ │ │
│ │ │ │ │ │ │
4 │ │ │ │
│ │ ← 1st eq. pt
2 │
└───────────────────────────────────────────────────────────────── V(mL)
Ve 2Ve 3Ve
Each plateau is a buffer region where pH ≈ pKa(n).
Equivalence points only VISIBLE when ΔpKa between steps ≥ 3-4 units.
EDTA COMPLEXOMETRIC (metal M + EDTA Y⁴⁻):
pM
16 │ ─────────────────────────────────────────────╮
│ │ ← large jump if KMY big
10 │ │
│ ╰──────────────
4 │
└─────────────────────────────────────────────────────────────── V(mL)
Ve
pM = -log[M²⁺]; jump size governed by log KMY (conditional K'MY at real pH)
Compares four titration curve shapes side by side, showing how acid strength, number of protons, and metal-ligand stability constants determine whether a sharp equivalence point exists.
---Diagram 3: The Van Deemter Equation — Why There Is an Optimal Flow Rate in Chromatography
╔══════════════════════════════════════════════════════════════════════════════════╗
║ VAN DEEMTER EQUATION: H = A + B/u + Cu ║
║ H = height equivalent to a theoretical plate (smaller H = better) ║
╚══════════════════════════════════════════════════════════════════════════════════╝
H
(plate
height)
│
│ total H curve
│ ╲
│ ╲ ← A term dominates here at all u (eddy diffusion,
│ ╲ independent of flow rate — just a constant offset)
│ ╲___
│ ───╮_______
│ ╲ ─────────────────── ← C term (mass transfer)
│ ╲ rises linearly with u
│ B/u ↘ ╲___________
│ ↑
│ H_minimum
│ at u_optimal
│
└──────────────────────────────────────────────────────────────── u (linear
u_optimal flow rate)
┌───────────────┬──────────────────────────────────────────────────────────────┐
│ TERM │ PHYSICAL MEANING │
├───────────────┼──────────────────────────────────────────────────────────────┤
│ A (eddy │ multiple flow paths through irregular packed bed; │
│ diffusion) │ minimized by smaller, uniform particles (sub-2 µm in UHPLC) │
├───────────────┼──────────────────────────────────────────────────────────────┤
│ B/u │ longitudinal diffusion along column axis; │
│ (diffusion) │ matters at LOW flow — analyte spreads while drifting; │
│ │ B is larger in gas phase (GC) than liquid phase (HPLC) │
├───────────────┼──────────────────────────────────────────────────────────────┤
│ Cu (mass │ slow equilibration between mobile and stationary phases; │
│ transfer) │ matters at HIGH flow — analyte can't partition fast enough; │
│ │ C reduced by thin stationary phase films, small particles │
└───────────────┴──────────────────────────────────────────────────────────────┘
RESOLUTION between two peaks:
√N (α - 1) k₂
Rs = ─── · ────── · ──────
4 α 1 + k₂
N = number of plates = L/H (efficiency — make H small)
α = selectivity = K₂/K₁ (chemistry — change stationary phase or mobile phase)
k = retention factor (optimize via mobile phase strength)
To improve Rs: increase N (longer column, smaller particles),
increase α (change chemistry), or adjust k.
Illustrates the three competing contributions to peak broadening in chromatography and shows why an optimal flow rate exists where H is minimized and column efficiency is maximized.
---Diagram 4: Electroanalytical Methods — Reference Electrodes, the Nernst Equation, and Ion-Selective Electrodes
╔══════════════════════════════════════════════════════════════════════════════════╗
║ POTENTIOMETRIC CELL ARCHITECTURE ║
╚══════════════════════════════════════════════════════════════════════════════════╝
┌──────────────────────────────────────────────────────────────────────────────┐
│ │
│ REFERENCE ELECTRODE │ salt bridge │ INDICATOR ELECTRODE │
│ │ (KCl gel) │ │
│ ┌────────────────────────┐ │ │ ┌──────────────────────┐ │
│ │ Ag wire coated with │ │ ══════════ │ │ e.g., glass membrane│ │
│ │ AgCl in sat'd KCl │ │ KCl fills │ │ pH electrode, or │ │
│ │ Ag|AgCl|KCl(sat'd) │ │ junction │ │ ion-selective (ISE) │ │
│ │ │ │ │ │ │ │
│ │ E = +0.197 V vs SHE │ │ │ │ responds ONLY to │ │
│ │ (stable, non-toxic) │ │ │ │ target ion activity │ │
│ └────────────────────────┘ │ │ └──────────────────────┘ │
│ │
│ VOLTMETER measures E_cell = E_indicator − E_reference │
│ │
└──────────────────────────────────────────────────────────────────────────────┘
NERNST EQUATION (for indicator electrode responding to ion Mⁿ⁺):
RT 0.05916 V
E = E° + ── ln[a] = E° + ───────── log(a_{Mⁿ⁺}) at 25°C
nF n
where a = γ · [M] (activity = activity coefficient × molar concentration)
┌─────────────────────────────────────────────────────────────────────────────┐
│ KEY INSIGHT: the electrode responds to ACTIVITY, not concentration. │
│ At low ionic strength, γ ≈ 1 and a ≈ [M]. │
│ At high ionic strength, γ < 1; ignoring this gives systematic error. │
│ Fix: use ionic strength adjustment buffer (ISAB) to swamp & equalize μ. │
└─────────────────────────────────────────────────────────────────────────────┘
GLASS pH ELECTRODE — a special ISE:
┌───────────────────────────────────────────────────────────────────────────┐
│ │
│ inner ┌──────────┐ outer │
│ reference ────│ hydrated │──── sample solution │
│ solution │ glass │ │
│ (pH 7 HCl) │ membrane │ H⁺ ions exchange with Na⁺ in glass │
│ └──────────┘ creating a phase boundary potential │
│ │
│ E_measured = const − 0.05916 · pH (slope ≈ −59.16 mV per pH unit) │
│ │
│ Alkaline error: Na⁺ interferes at high pH (glass responds to Na⁺ too) │
│ Acid error: H₂O activity changes at very low pH │
└───────────────────────────────────────────────────────────────────────────┘
Shows the architecture of a complete potentiometric cell, the Nernst equation governing the indicator electrode response, and the specific mechanism of the glass pH electrode including its known error sources.
---Diagram 5: Beer-Lambert Law — Physical Basis and Sources of Deviation
╔══════════════════════════════════════════════════════════════════════════════════╗
║ BEER-LAMBERT LAW: A = εbc ║
╚══════════════════════════════════════════════════════════════════════════════════╝
Incident │←─────────── path length b ────────────→│ Transmitted
light I₀ ──────▶│ absorbing solution, concentration c │──────▶ I
│ (molar absorptivity ε, L·mol⁻¹·cm⁻¹) │
A = log(I₀/I) = ε · b · c T = I/I₀ = 10^(−A)
PHYSICAL BASIS (from Maxwell / quantum picture):
─────────────────────────────────────────────────────────────
Each absorbing molecule presents a cross-section to photons.
Probability of absorption is proportional to:
(number of absorbers per cm²) = c · b · Nₐ · σ
Taking the log of transmitted fraction gives the linear law.
ε encodes the quantum mechanical transition dipole moment.
SOURCES OF DEVIATION FROM LINEARITY:
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ REAL (chemical) deviations: │
│ ● Analyte associates/dimerizes at high c → ε changes with c │
│ ● Equilibrium shifts (e.g., acid-base) change absorbing species │
│ │
│ INSTRUMENTAL deviations: │
│ ● Polychromatic light: if ε varies across bandwidth Δλ, │
│ calibration curve bends downward at high A │
│ ● Stray light: extra photons reach detector without passing through │
│ sample; most serious at high A (A > ~2) — apparent A plateaus │
│ ● Mismatched cuvettes, dirty windows, scratches │
│ │
│ PRACTICAL CONSEQUENCE: │
│ Keep A between 0.1 and 1.0 for best accuracy. │
│ (Optimum: A = 0.434 minimizes relative concentration error) │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
MULTICOMPONENT ANALYSIS (two analytes X and Y, overlapping spectra):
A(λ₁) = εₓ(λ₁)·b·cₓ + εᵧ(λ₁)·b·cᵧ
A(λ₂) = εₓ(λ₂)·b·cₓ + εᵧ(λ₂)·b·cᵧ
Two equations, two unknowns — solve simultaneously.
Works because Beer's Law is LINEAR (absorbances add).
In matrix form: A⃗ = ε̂ · c⃗ → c⃗ = ε̂⁻¹ · A⃗
(You learned matrix inversion in MATH301 — this is it in practice.)
Derives the Beer-Lambert law from a physical picture, catalogues all real and instrumental sources of non-linearity, and shows how matrix algebra enables simultaneous determination of multiple absorbing species.
---Key Takeaways
- The t-distribution governs small samples, not the Gaussian. When you have fewer than ~30 measurements and the population standard deviation σ is unknown (which is always), confidence intervals must use the Student's t-statistic at ν = n − 1 degrees of freedom. The interval is x̄ ± t·s/√n. As n → ∞, t → z (the Gaussian z-value). Using z when n is small gives dangerously narrow confidence intervals. (Harris & Lucy, Chapter 4.)
- Propagation of uncertainty follows specific rules based on operation type. For addition and subtraction, absolute uncertainties add in quadrature: e² = e₁² + e₂². For multiplication and division, relative uncertainties add in quadrature: (e/y)² = (e₁/x₁)² + (e₂/x₂)². The general formula for any function f(x₁, x₂,...) is e_f² = Σ(∂f/∂xᵢ)²·eᵢ². This means a pipetting error in a dilution step can dominate over the instrument precision — always trace error to its largest source. (Harris & Lucy, Chapter 3.)
- Activity, not concentration, is the thermodynamically correct variable. The true equilibrium constant expression uses activities a = γ[M], where γ is the activity coefficient. For dilute solutions (ionic strength μ → 0), γ → 1. The extended Debye-Hückel equation gives log γᵢ = −0.509 zᵢ² √μ / (1 + α√μ/305), where α is the effective hydrated ion diameter in pm. This is why EDTA titrations and ISE measurements are sensitive to background salt — they change μ and therefore γ. (Harris & Lucy, Chapter 8.)
- The Grubbs test (not intuition) is the correct method for outlier rejection. To test whether a suspect value xₛ is an outlier, compute G = |xₛ − x̄| / s, then compare to critical G values tabulated for sample size n and confidence level. If G_calculated > G_critical, reject the outlier. The Dixon Q-test (Q = gap/range) is simpler but less powerful. You are not allowed to discard a data point simply because it disagrees with your hypothesis — that is scientific fraud. (Harris & Lucy, Chapter 4.)
- The alpha-fraction formalism gives the exact speciation of polyprotic acids at any pH. For a diprotic acid H₂A with constants Ka1 and Ka2, the fraction present as A²⁻ is α₂ = Ka1·Ka2 / ([H⁺]² + [H⁺]·Ka1 + Ka1·Ka2). These alpha fractions sum to 1, obey no approximations, and are exact solutions to the equilibrium system. They are essential for calculating conditional stability constants in EDTA chemistry: K'MY = KMY · α(Y⁴⁻), where α(Y⁴⁻) is the EDTA fraction in fully deprotonated form at the working pH. (Harris & Lucy, Chapters 11 and 12.)
- The equivalence point of a weak acid–strong base titration is above pH 7. At the equivalence point, all analyte has been converted to its conjugate base A⁻, which hydrolyzes water: A⁻ + H₂O ⇌ HA + OH⁻. The pH is set by Kb = Kw/Ka. This is not a subtle correction — for acetic acid (Ka = 1.8 × 10⁻⁵), the equivalence point pH is about 8.7. Using a phenolphthalein indicator (endpoint ~pH 9) is appropriate; methyl orange (~pH 4) would be disastrously wrong. (Harris & Lucy, Chapter 11.)
- EDTA titrations use conditional stability constants to account for pH. The metal-EDTA stability constant KMY can be enormous (e.g., log KFeY = 25.1 for Fe³⁺), but EDTA exists mostly as H₂Y²⁻ or HY³⁻ at acidic pH, not Y⁴⁻. The conditional constant K'MY = KMY · αY⁴⁻ gives the effective stability at working pH. A metal can only be titrated directly with EDTA at a pH where K'MY is large enough (K'MY > ~10⁶) to give a sharp endpoint. Auxiliary complexing agents (e.g., ammonia buffer for Zn²⁺) hold metal in solution without competing too strongly. (Harris & Lucy, Chapter 12.)
- The Nernst equation means potentiometry measures ion activity, with slope 59.16/n mV per decade at 25°C. For a divalent cation (n = 2), each 10-fold change in activity shifts the electrode potential by only 29.58 mV — much smaller than for a monovalent ion (59.16 mV). Electrode slope calibration must be done at the measurement temperature. The glass pH electrode measures proton activity, not proton concentration; pH buffers used for calibration are defined by their certified activity values, not their nominal concentration. (Skoog, Holler & Crouch, Chapter 23.)
- The van Deemter equation explains why there is an optimal mobile phase velocity in chromatography. H = A + B/u + Cu. The A term (eddy diffusion) is constant; B/u (longitudinal diffusion) falls with increasing flow; Cu (mass transfer resistance) rises with increasing flow. The minimum plate height H_min occurs at u_opt = √(B/C). In HPLC, B is small (liquids diffuse slowly), so u_opt is relatively high and the B/u term is rarely limiting; mass transfer (C term) dominates at high flow. Sub-2 μm particles in UHPLC reduce both A and C, enabling faster runs at equivalent or better efficiency. (Skoog, Holler & Crouch, Chapter 26.)
- Resolution in chromatography depends independently on efficiency N, selectivity α, and retention factor k. Rs = (√N/4)·((α−1)/α)·(k₂/(1+k₂)). Doubling column length only doubles N and improves Rs by a factor of √2 — at the cost of doubling analysis time and pressure. Changing selectivity α (different stationary phase, mobile phase modifier, or temperature) can improve Rs dramatically with no added cost. This is why method development prioritizes chemistry (α) over brute-force column lengthening. (Harris & Lucy, Chapter 23.)
- Fluorimetry is inherently more sensitive than absorption spectrophotometry by 2–3 orders of magnitude. In absorption, you measure a small decrease in a large signal (I vs I₀); the signal-to-noise limit is set by how precisely you can measure I₀. In fluorescence, you measure emission against a near-zero background — the signal rises from zero with concentration. Consequently, detection limits of 10⁻¹⁰–10⁻¹² M are routine for fluorescent compounds. The inner filter effect (re-absorption of emitted photons at high concentration) causes fluorescence intensity to plateau and then decrease, so standard addition or dilution is essential above ~10⁻⁵ M. (Skoog, Holler & Crouch, Chapter 15.)
- Standard addition corrects for matrix effects that external calibration cannot. Known amounts of analyte are added to aliquots of the actual sample. The signal is plotted versus added concentration, and the x-intercept (extrapolated, negative x-axis) gives the unknown concentration. This works because both the unknown and the spikes experience identical matrix suppression or enhancement. It fails if the matrix effect is non-linear or if the method is non-linear. It also cannot correct for a blank interference that scales with volume rather than concentration. (Harris & Lucy, Chapter 5.)
- Co-precipitation is the main source of error in gravimetric analysis. When a precipitate forms, foreign ions can be trapped (occlusion), adsorbed on the surface, or incorporated into the crystal lattice (mixed crystal formation). For example, BaSO₄ precipitated from a solution containing NO₃⁻ will co-precipitate Ba(NO₃)₂. Digestion (heating the precipitate in mother liquor) allows recrystallization and reduces co-precipitation by allowing impurities to re-dissolve. The common ion effect reduces solubility but excess precipitating agent can cause ion-pair or complex formation, paradoxically increasing solubility. (Harris & Lucy, Chapter 7.)
- Reversed-phase HPLC retains nonpolar analytes on a nonpolar stationary phase eluted by aqueous-organic mobile phase. The stationary phase (typically C18 — octadecylsilane bonded to silica) is hydrophobic; more nonpolar analytes elute last. Increasing the organic modifier fraction (acetonitrile or methanol) in the mobile phase decreases retention (shorter k). Gradient elution — starting with high water content and ramping up organic — compresses peaks for a wide polarity range of analytes, analogous to temperature programming in GC. Peak areas (not heights) are used for quantification because area is independent of small flow-rate fluctuations. (Harris & Lucy, Chapter 23.)
- The charge balance and mass balance are the rigorous constraints for any equilibrium calculation. The charge balance states that the sum of positive charge concentrations equals the sum of negative: Σ(nᵢ[cationᵢ]) = Σ(nᵢ[anionᵢ]). The mass balance states that the total analytical concentration of each element is conserved across all species. Together, with the equilibrium constant expressions, they form a closed algebraic system. Approximations (e.g., assuming [H⁺] >> [OH⁻]) must be checked for self-consistency. The systematic approach eliminates the need to memorize different formulas for different buffer cases — one framework handles all. (Harris & Lucy, Chapter 8.)
References
- Daniel C. Harris and Charles A. Lucy, Quantitative Chemical Analysis, 10th edition, W. H. Freeman and Company / Macmillan Learning, New York, 2020. ISBN 978-1-319-07815-1.
- Douglas A. Skoog, F. James Holler, and Stanley R. Crouch, Principles of Instrumental Analysis, 7th edition, Cengage Learning, Boston, 2018. ISBN 978-1-305-57721-3.
- IUPAC, "Compendium of Chemical Terminology" (the 'Gold Book'), compiled by A. D. McNaught and A. Wilkinson, 2nd edition, Blackwell Scientific Publications, Oxford, 1997; online version 2019. DOI: 10.1351/goldbook. [Referenced for definitions of activity, ionic strength, and standard electrode potential.]
- H. van Deemter, F. J. Zuiderweg, and A. Klinkenberg, "Longitudinal diffusion and resistance to mass transfer as causes of nonideality in chromatography," Chemical Engineering Science, vol. 5, pp. 271–289, 1956. DOI: 10.1016/0009-2509(56)80003-7. [Original derivation of the van Deemter equation.]
- W. E. Deming, Statistical Adjustment of Data, Wiley, New York, 1943; Dover reprint, 1964. [Foundational statistical methods underlying propagation of uncertainty.]
- Frank Grubbs, "Procedures for Detecting Outlying Observations in Samples," Technometrics, vol. 11, no. 1, pp. 1–21, 1969. DOI: 10.1080/00401706.1969.10490657. [Original Grubbs outlier test paper.]
- Peter W. Atkins and Julio de Paula, Physical Chemistry, 10th edition, Oxford University Press, Oxford, 2014. [Background on activity coefficients and the Debye-Hückel limiting law, complementing CHEM301 material.]