What is in this water, and how sure are we?
A family's tap water is tested for lead. The World Health Organization's guideline value is 10 micrograms per litre (WHO), ten parts per billion by mass, and the European Union's 2020 drinking-water directive lowers its own limit to 5 µg/L by 2036. The laboratory reports 8 µg/L. Is the water within the guideline?
The question cannot be answered from the number alone. If the result is 8 ± 1 µg/L, almost certainly yes. If it is 8 ± 5 µg/L, the water might be well over the guideline, and a responsible report would say the measurement cannot decide. An analytical result is a number with an uncertainty, produced by a method whose performance has been checked, in a laboratory that can show its checks. That is what this lesson is about.
Analytical chemistry answers two questions: what is in a sample (qualitative analysis) and how much (quantitative analysis). It underpins medicine (blood tests), food safety, environmental monitoring, forensic science, anti-doping, the release of every batch of medicine, and the research results of every other branch of chemistry. It is also where chemistry meets statistics most directly.
The importance of asking the right question was shown painfully in 2008, when milk products in China were found to contain melamine. Protein in milk was routinely estimated from its total nitrogen content, and melamine is two-thirds nitrogen by mass, so adding it made diluted milk look protein-rich to that test. Many thousands of infants fell ill. The measurements were not imprecise; they were measuring the wrong thing. Selectivity, the ability of a method to respond to the analyte and not to other substances, is as important as precision.
Accuracy, precision and uncertainty
Definition— accuracy, trueness and precision
Accuracy is how close a single result is to the true value. It has two components. Trueness is how close the average of many results is to the true value; its opposite is bias, caused by systematic errors. Precision is how close repeated results are to each other; its opposite is scatter, caused by random errors. Precision measured within one run is repeatability; across days, analysts and laboratories it is reproducibility, which is always worse.
Definition— measurement uncertainty
The uncertainty of a result is a parameter, such as a standard deviation, that describes the spread of values that could reasonably be attributed to the quantity measured. A standard uncertainty is expressed as a standard deviation; an expanded uncertainty , usually with coverage factor , gives an interval that contains the true value with about 95 per cent confidence. The international rules are in the GUM.
For replicate results , the best estimate is the mean , the scatter is described by the sample standard deviation , and the uncertainty of the mean is . Because is itself estimated from few results, a 95 per cent confidence interval for the mean uses Student's distribution: (see statistical inference).
When the result is calculated from several measured quantities, their uncertainties combine. For a product or quotient such as , the relative standard uncertainties add in quadrature:
valid when the inputs are independent and their uncertainties small. The program computes the values from scratch, then treats a common laboratory task: standardising a sodium hydroxide solution by titrating a weighed mass of potassium hydrogen phthalate (KHP, molar mass , purity ) to the end point, which uses a volume .
import math
import numpy as np
def t_critical(df, confidence=0.95):
"""Two-sided critical value of Student's t, found by integrating its density (Simpson) and bisecting."""
c = math.gamma((df + 1) / 2) / (math.sqrt(df * math.pi) * math.gamma(df / 2))
def central_area(t):
x = np.linspace(0.0, t, 2001)
y = c * (1 + x * x / df) ** (-(df + 1) / 2)
h = x[1] - x[0]
return 2 * h / 3 * (y[0] + y[-1] + 4 * y[1:-1:2].sum() + 2 * y[2:-1:2].sum())
lo, hi = 0.0, 100.0
for _ in range(60):
mid = (lo + hi) / 2
lo, hi = (mid, hi) if central_area(mid) < confidence else (lo, mid)
return (lo + hi) / 2
print("df: ", " ".join(f"{df:6d}" for df in range(1, 11)))
print("t95: ", " ".join(f"{t_critical(df):6.3f}" for df in range(1, 11)))
# 1. Random error: five replicate standardisations of a sodium hydroxide solution (mol/L)
c = np.array([0.1012, 0.1009, 0.1015, 0.1010, 0.1013])
n, mean, s = len(c), c.mean(), c.std(ddof=1) # ddof=1: divide by n - 1
half = t_critical(n - 1) * s / math.sqrt(n)
print(f"\nreplicates: mean = {mean:.5f} mol/L, s = {s:.5f}, RSD = {100 * s / mean:.2f} %")
print(f"95 % confidence interval: {mean:.5f} ± {half:.5f} mol/L")
# 2. Propagating uncertainty through c = m * P / (M * V) for one titration
m, u_m = 0.5105, 0.0002 # g of potassium hydrogen phthalate (KHP), balance uncertainty
P, u_P = 1.0000, 0.0005 # purity (mass fraction) from the certificate
M = 204.221 # g/mol, uncertainty negligible here
V, u_V = 24.70e-3, 0.03e-3 # L of NaOH delivered by the burette
c1 = m * P / (M * V)
rel = math.sqrt((u_m / m) ** 2 + (u_P / P) ** 2 + (u_V / V) ** 2)
print(f"\none titration: c = {c1:.5f} mol/L, standard uncertainty {c1 * rel:.5f} mol/L ({100 * rel:.2f} %)")
for name, r in [("mass", u_m / m), ("purity", u_P / P), ("volume", u_V / V)]:
print(f" share of the variance from {name:7}: {100 * r**2 / rel**2:5.1f} %")
# 3. The same by Monte Carlo: simulate the inputs and look at the spread of the result
rng = np.random.default_rng(2026)
draws = rng.normal(m, u_m, 200_000) * rng.normal(P, u_P, 200_000) / (M * rng.normal(V, u_V, 200_000))
print(f"Monte Carlo: mean {draws.mean():.5f} mol/L, standard deviation {draws.std():.5f} mol/L")The values grow sharply for very small samples (12.7 for two results, 2.78 for five), which is why chemists rarely report a mean of two. The Monte Carlo simulation, the method of the GUM's first supplement, agrees with the formula here; it earns its keep when the formula's assumptions fail, for example when an uncertainty is large or the calculation strongly non-linear.
Calibration curves and least squares
Most instrumental methods do not measure concentration directly. A spectrophotometer measures absorbance; a mass spectrometer counts ions. The analyst therefore measures standards of known concentration, fits a calibration function, and reads the unknown off it. For absorbance, the Beer–Lambert law of Spectroscopy predicts a straight line, .
Our example measures iron in well water. Iron(II) forms an intensely red-orange complex with three molecules of 1,10-phenanthroline, a chelating ligand from Coordination chemistry, which absorbs strongly at 510 nm. Standards from 0 to 5 mg/L of iron give the absorbances in the program below.
Theorem— Least-squares line
Among all lines , the one that minimises the sum of squared vertical deviations has
Proof
is a convex quadratic in and , so its minimum is where both partial derivatives vanish. gives , so the line passes through the centroid . Substituting into gives . Since and , we may replace by , and solving for gives the formula.
∎
Geometrically, the fitted values are the orthogonal projection of the measured signals onto the plane spanned by the vectors and , which is why the residuals sum to zero (see linear algebra).
Three more quantities turn a fitted line into an analytical result. The standard error of the regression measures the scatter of the points about the line (with because two parameters were fitted). An unknown measured times with mean signal has concentration with standard error
smallest near the centre of the calibration range. And the limit of detection (LOD), the smallest concentration that can be distinguished from zero with reasonable confidence, and the limit of quantification (LOQ), the smallest that can be measured with acceptable precision, are commonly estimated as and , where is the standard deviation of the blank or (ICH Q2(R2)).
The factor 3.3 is not arbitrary. Lloyd Currie's analysis of 1968 (paper) separates two questions. To decide whether a signal is above the blank, set a threshold above it, so that a blank crosses it only 5 per cent of the time (false positives). For a substance to be detected reliably, its signal must cross that threshold 95 per cent of the time (false negatives), which needs a true signal another higher: . The LOQ's factor of 10 corresponds to a relative standard deviation of 10 per cent at that level.
import math
import numpy as np
# Calibration standards for iron as its 1,10-phenanthroline complex, absorbance at 510 nm, 1 cm cell
conc = np.array([0.0, 0.5, 1.0, 2.0, 3.0, 4.0, 5.0]) # mg/L Fe
absb = np.array([0.004, 0.101, 0.203, 0.398, 0.597, 0.794, 0.992]) # measured absorbance
n = len(conc)
xbar, ybar = conc.mean(), absb.mean()
Sxx = np.sum((conc - xbar) ** 2)
b = np.sum((conc - xbar) * (absb - ybar)) / Sxx # slope
a = ybar - b * xbar # intercept
resid = absb - (a + b * conc)
s_yx = math.sqrt(np.sum(resid**2) / (n - 2)) # scatter about the line
s_b = s_yx / math.sqrt(Sxx)
s_a = s_yx * math.sqrt(np.sum(conc**2) / (n * Sxx))
r2 = 1 - np.sum(resid**2) / np.sum((absb - ybar) ** 2)
t = 2.571 # t for 95 %, n - 2 = 5 degrees of freedom
print(f"slope b = {b:.5f} ± {t * s_b:.5f} L/mg (95 %)")
print(f"intercept a = {a:.5f} ± {t * s_a:.5f}")
print(f"s(y/x) = {s_yx:.5f}, R² = {r2:.5f}")
print("residuals:", " ".join(f"{e:+.4f}" for e in resid))
# An unknown water sample, measured three times
y0 = np.array([0.462, 0.459, 0.465])
m = len(y0)
x0 = (y0.mean() - a) / b
s_x0 = s_yx / b * math.sqrt(1 / m + 1 / n + (y0.mean() - ybar) ** 2 / (b**2 * Sxx))
print(f"\nsample: {x0:.3f} ± {t * s_x0:.3f} mg/L Fe (95 % confidence)")
# Detection and quantification limits from the calibration (ICH Q2 convention)
print(f"LOD = 3.3 s(y/x)/b = {3.3 * s_yx / b:.3f} mg/L, LOQ = 10 s(y/x)/b = {10 * s_yx / b:.3f} mg/L")
# Cross-check the fit with NumPy's least-squares solver
X = np.column_stack([np.ones(n), conc])
(a2, b2), *_ = np.linalg.lstsq(X, absb, rcond=None)
print(f"np.linalg.lstsq agrees: a = {a2:.5f}, b = {b2:.5f}")The slope, 0.1977 L/mg, corresponds to a molar absorption coefficient of about 11 000 L mol⁻¹ cm⁻¹ once multiplied by iron's molar mass, typical of an intensely coloured complex. The sample contains 2.32 ± 0.01 mg/L of iron. The intercept's confidence interval does not include zero, a small positive blank that the fitted intercept handles correctly; forcing the line through the origin would have introduced a bias. Always look at the residuals: a curved pattern (positive at the ends, negative in the middle) would reveal non-linearity that a high hides.
Note
Matrix effects and standard additions. Standards made up in pure water may not behave like a real sample, whose other components (the matrix) can enhance or suppress the signal. In the method of standard additions the analyst adds known amounts of analyte to portions of the sample itself, fits a line, and extrapolates to zero signal; the matrix is then the same for every point.
Titrimetric and gravimetric analysis
The oldest quantitative methods are still among the most accurate, because they rest on stoichiometry and on two of the best measurements in the laboratory, mass and volume.
Titrimetric (volumetric) analysis adds a solution of accurately known concentration, the titrant, from a burette until the reaction with the analyte is just complete: the equivalence point. An indicator colour change or an electrode signal marks the end point, which should coincide with it as closely as possible. You met acid–base titrations in Acids, bases and buffers; other families use other reactions.
- Complexometric titrations use EDTA, which binds almost every metal ion 1 : 1. Water hardness, the total calcium and magnesium, is measured this way, with an indicator dye that changes colour when EDTA has taken the last metal ion from it.
- Redox titrations use oxidising or reducing titrants, such as cerium(IV) or iodine, and follow the changing electrode potential from Electrochemistry.
- Precipitation titrations use silver ions to measure chloride.
- Karl Fischer titration, based on a reaction of water with iodine, measures traces of water in medicines, solvents and oils, and is one of the most widely used titrations in industry.
Gravimetric analysis converts the analyte into a pure solid of known composition, which is filtered, washed, dried or heated, and weighed. Sulfate is precipitated as barium sulfate; nickel as a bright red complex with dimethylglyoxime, another chelate. The calculation uses the gravimetric factor, the ratio of molar masses that converts mass of precipitate into mass of analyte.
Example— Sulfate by gravimetry
A 0.5000 g sample of a fertiliser is dissolved, the sulfate is precipitated as BaSO₄ (molar mass 233.39 g/mol), and after filtering and heating to constant mass the precipitate weighs 0.6214 g. Moles of BaSO₄ mol, equal to the moles of sulfate. Mass of sulfate (96.06 g/mol) g, so the sample is per cent sulfate by mass. The gravimetric factor is . Nothing here needed calibration: mass, molar masses and stoichiometry alone give the answer, which is why gravimetry and titrimetry are used to certify reference materials.
Classical methods need more sample and more time than instruments, and they cannot reach trace levels, but they are primary in the metrological sense: their results are traceable to the SI through the kilogram and the mole without depending on another standard of the same substance.
Chromatography: separating before measuring
Real samples are mixtures, and most detectors cannot tell similar molecules apart. Chromatography separates them first. A mobile phase (a gas or a liquid) carries the sample through a stationary phase (a coating on the inside of a capillary, or fine particles packed in a column). Each compound spends a characteristic fraction of its time held by the stationary phase, so compounds travel at different speeds and reach the detector at different retention times. The Russian-Italian botanist Mikhail Tsvet separated plant pigments into coloured bands on a column of chalk in the early 1900s and named the method after the colours (Greek chroma); Archer Martin and Richard Synge's partition chromatography of 1941 (paper) and its plate theory earned them the 1952 Nobel Prize in Chemistry.
Definition— retention factor, selectivity, plate number and resolution
With dead time and retention times :
- the retention factor is the ratio of time spent in the stationary phase to time in the mobile phase;
- the selectivity measures how differently the column treats the two compounds;
- the plate number measures the column's efficiency (narrow peaks, large );
- the resolution measures how well two peaks are separated. means baseline separation for equal Gaussian peaks.
The Purnell equation links them: . It is the chromatographer's map. Efficiency enters only as , so doubling the column length (and the analysis time) improves resolution by only 41 per cent; selectivity, changed by choosing a different stationary phase, mobile phase or temperature, is usually the most powerful lever.
import numpy as np
t_M = 1.20 # min: an unretained marker reaches the detector
peaks = [(6.40, 0.200), (6.85, 0.210)] # (retention time, width at half height), min
(t1, w1), (t2, w2) = peaks
Rs = 1.18 * (t2 - t1) / (w1 + w2) # 1.18 ≈ sqrt(2 ln 2), for Gaussian peaks
N = 5.54 * (t2 / w2) ** 2 # plate number from the second peak (5.54 = 8 ln 2)
k1, k2 = (t1 - t_M) / t_M, (t2 - t_M) / t_M
alpha = k2 / k1
Rs_purnell = np.sqrt(N) / 4 * (alpha - 1) / alpha * k2 / (1 + k2)
print(f"k1 = {k1:.2f}, k2 = {k2:.2f}, alpha = {alpha:.3f}, N = {N:.0f} plates")
print(f"resolution from the peaks: {Rs:.2f}; from the Purnell equation: {Rs_purnell:.2f}")
need = (1.5 / Rs) ** 2
print(f"for Rs = 1.5 we need {need:.2f} times as many plates ({need * N:.0f}),")
print(f"for example a column {need:.2f} times as long, and a run {need:.2f} times as long")
# A text chromatogram: two Gaussian peaks of equal area
t = np.arange(6.10, 7.16, 0.05)
signal = sum(np.exp(-4 * np.log(2) * ((t - tr) / w) ** 2) * 0.200 / w for tr, w in peaks)
for ti, yi in zip(t, signal):
print(f"{ti:5.2f} min |{'#' * int(round(30 * yi / signal.max()))}")The two peaks are nearly, but not quite, baseline-separated (). The two estimates of resolution agree to within a few per cent; the Purnell equation assumes that both peaks share one plate number, which real neighbouring peaks only approximately do. Gas chromatography (GC) suits volatile compounds such as solvents, fuels and fragrances; high-performance liquid chromatography (HPLC), pumping liquid through columns of particles a few micrometres across, handles most medicines, proteins and pollutants. Thin-layer chromatography on a coated plate is the quick qualitative check that organic chemists use many times a day.
Mass spectrometry: weighing molecules
A mass spectrometer turns molecules into gas-phase ions, separates the ions by their mass-to-charge ratio , and counts them. It answers both of analysis's questions at once: the masses identify, the counts quantify.
Ionisation. Electron ionisation, used with GC, knocks an electron out of each molecule and breaks many of the resulting ions into fragments, whose pattern is a fingerprint that can be matched against large spectral libraries. Electrospray ionisation and MALDI (matrix-assisted laser desorption/ionisation) are gentle enough to lift whole proteins into the gas phase as ions; they earned John Fenn and Koichi Tanaka a share of the 2002 Nobel Prize in Chemistry. For elements, an inductively coupled argon plasma at several thousand kelvin atomises and ionises everything (ICP-MS), which is how lead in water is measured at parts per trillion.
Mass analysers. A quadrupole lets through only ions of one at a time, selected by oscillating electric fields; a time-of-flight analyser accelerates all ions to the same kinetic energy and times their flight, since lighter ions arrive first; an orbital trap (the Orbitrap) holds ions orbiting a spindle-shaped electrode and reads their from the frequency of their oscillation. Tandem mass spectrometry (MS/MS) selects one ion, breaks it and analyses the pieces, which gives great selectivity in complex samples like blood plasma.
Identifying and quantifying. You met electron-ionisation spectra, isotope patterns and exact masses in Spectroscopy. Three further calculations are everyday analytical work. The resolving power says whether two ions of almost equal mass give separate peaks, and mass accuracy, quoted in parts per million (ppm), says which formula a measured mass fits. Electrospray gives large molecules several charges, so a protein appears as a ladder of peaks whose spacing reveals the charge and hence the mass. And for quantitation, the best practice is to add to every sample a fixed amount of an internal standard: the analyte itself made with ¹³C, ¹⁵N or deuterium atoms, which behaves identically in the instrument but is a few mass units heavier.
import numpy as np
PROTON, ELECTRON = 1.007276, 0.000549 # u
mono = {"C": 12.0, "H": 1.0078250, "N": 14.0030740, "O": 15.9949146}
# 1. Resolving power and mass accuracy: the ions CO+ and N2+ both have nominal mass 28
m_CO = mono["C"] + mono["O"] - ELECTRON # an ion has lost an electron
m_N2 = 2 * mono["N"] - ELECTRON
print(f"CO+ {m_CO:.5f} u, N2+ {m_N2:.5f} u: resolving power needed m/dm = {m_CO / (m_N2 - m_CO):.0f}")
peak = 28.00555 # a measured peak
for name, m in [("N2+", m_N2), ("CO+", m_CO)]:
print(f" measured {peak:.5f} u is {1e6 * (peak - m) / m:+7.1f} ppm from {name}")
# 2. Electrospray: two neighbouring peaks of one protein, charges z (higher m/z) and z + 1
mz_a, mz_b = 1060.476, 998.154
z = round((mz_b - PROTON) / (mz_a - mz_b)) # from m/z = (M + z*PROTON) / z
M = z * (mz_a - PROTON)
print(f"\nthe peak at m/z {mz_a} carries charge {z}; protein mass M = {M:.1f} u")
print("predicted charge ladder:", " ".join(f"{q}+: {(M + q * PROTON) / q:.1f}" for q in range(z - 2, z + 4)))
# 3. Quantitation with an isotope-labelled internal standard (IS) in every vial
std = np.array([1.0, 5.0, 10.0, 20.0, 50.0]) # ng/mL of analyte, standards in clean solvent
area = np.array([1490, 7560, 15020, 29900, 75300]) # analyte peak areas
area_is = np.array([30900, 31200, 30800, 31100, 31000]) # IS peak areas (20 ng/mL in every vial)
b_ext = np.polyfit(std, area, 1) # external calibration: area against concentration
b_int = np.polyfit(std, area / area_is, 1) # internal-standard calibration: area ratio
# A vegetable extract: the matrix suppresses ionisation of BOTH compounds (they reach the source together)
sample_area, sample_is = 11200, 19300
c_ext = (sample_area - b_ext[1]) / b_ext[0]
c_int = (sample_area / sample_is - b_int[1]) / b_int[0]
print(f"\nexternal calibration: {c_ext:5.2f} ng/mL")
print(f"internal-standard calibration: {c_int:5.2f} ng/mL (IS signal {100 * sample_is / area_is.mean():.0f} % of normal)")A resolving power of about 2500 separates CO⁺ from N₂⁺; time-of-flight instruments reach tens of thousands, and Orbitrap and Fourier-transform ion cyclotron resonance instruments hundreds of thousands or more, enough to assign elemental formulas from masses measured to within a few ppm. The measured peak sits 1.7 ppm from N₂⁺ and 400 ppm from CO⁺, so the identification is not in doubt. Two neighbouring electrospray peaks fix the charge at 16 and the mass at 16 951.5 u, the average mass of horse-heart apomyoglobin, a protein often used to check instruments. The last example is the important one in practice: the vegetable extract suppresses ionisation, so the signal is only 62 per cent of what the same amount gives in clean solvent, and an external calibration reports 7.5 ng/mL instead of 12. The labelled internal standard is suppressed by exactly the same factor, so the ratio is unaffected. Using an isotopically labelled internal standard in this way, carried to its rigorous form, is isotope dilution mass spectrometry, one of the primary methods that national metrology institutes use to certify reference materials.
Quality control in real laboratories
A result that leaves a laboratory can decide whether a medicine is released, a patient treated, a river declared safe or an athlete banned. Laboratories therefore surround every measurement with a system of checks.
- Method validation. Before a method is used, the laboratory demonstrates its selectivity, linearity and range, trueness, precision (repeatability and intermediate precision), detection and quantification limits and robustness to small changes, following guidance such as ICH Q2(R2) for medicines or the Eurachem guides.
- Certified reference materials. Materials whose composition has been established by national metrology institutes such as NIST in the United States, the European Commission's Joint Research Centre or the National Institute of Metrology in China are analysed alongside samples to check trueness and to make results traceable to the SI.
- Control charts. A control sample of stable composition is analysed in every batch and plotted on a Shewhart chart with lines at the mean (warning) and (action). Rules such as "one point beyond " or "two consecutive points beyond on the same side", popularised for hospital laboratories by James Westgard and colleagues in 1981, flag a batch for investigation before its results are released.
- Blanks, duplicates and spikes in every batch catch contamination, check precision on real samples and measure recovery.
- Proficiency testing. Laboratories regularly analyse the same unknown samples sent by an independent organiser and are scored against the consensus or reference value; persistent outliers must find and fix the cause.
- Accreditation. An independent body assesses the whole system against ISO/IEC 17025, the international standard for the competence of testing and calibration laboratories; regulated industries add rules such as Good Laboratory Practice and Good Manufacturing Practice.
Solids bring their own methods, such as X-ray diffraction to identify crystalline phases and thermal analysis to follow decomposition and phase changes; Lumenary Materials covers them in X-ray diffraction and spectroscopy and thermal analysis.
Modern laboratories, as of 2026, run much of this automatically: laboratory information management systems record every sample, instrument and control result, flag control-chart violations, and keep the audit trail that regulators inspect. Automation and machine learning increasingly plan analyses and process the data, as Self-driving laboratories describes, but the statistical logic of this lesson is what makes their results trustworthy.
Something safe to try at home
Precision with a kitchen scale. With an adult, weigh the same spoon ten times, taking it off the scale between readings, and then weigh ten different spoonfuls of dry rice. Compute the mean and standard deviation of each set. The first measures the scale's repeatability; the second adds the variation in how you fill the spoon, which is almost always much larger: a lesson in where uncertainty really comes from.
Paper chromatography. With an adult, cut a strip from a white coffee filter, put a dot of water-soluble felt-tip ink or a drop of food colouring about 2 cm from the bottom, and stand the strip in a glass with a little water so that the water is below the dot. As the water climbs, the colours separate: many black and brown inks turn out to be mixtures of blue, red and yellow dyes. Use only water and food-safe colours, and wash your hands and the glass afterwards. Never taste or sniff the solutions, even at home.
Common misconceptions
Misconception
"More decimal places mean a more accurate result." Digits beyond those justified by the uncertainty are noise. Report a result as, say, 2.32 ± 0.01 mg/L, with the uncertainty given to one or two significant figures and the value rounded to match.
Misconception
"R² = 0.999 proves the calibration is good." A high correlation coefficient says the points lie near some line, not that a straight line is the right model or that the precision is adequate. Inspect the residuals, and judge the calibration by the uncertainty of the results it gives.
Misconception
"Below the detection limit means zero." It means the method cannot distinguish the signal from the blank with confidence. The substance may still be present at lower levels; report "less than the LOD" with the LOD stated.
Misconception
"Repeating a measurement many times makes it accurate." Replication shrinks random error as but leaves systematic error untouched. Bias is found only with reference materials, independent methods and careful thought about every input.
Exercises
1. Analytical chemistry
Warm-upHumanA laboratory checks its sodium hydroxide titration on a certified reference solution whose concentration is 0.1000 mol/L. Five replicates give 0.1031, 0.1029, 0.1032, 0.1030 and 0.1033 mol/L (mean 0.1031 mol/L, relative standard deviation 0.15 %). Which statement describes these results best?
2. Analytical chemistry
CoreHumanA 50.00 mL sample of tap water is titrated with 0.01000 mol/L EDTA, which binds Ca²⁺ and Mg²⁺ 1 : 1. The end point is reached after 12.40 mL. Water hardness is conventionally reported as if all of it were calcium carbonate (CaCO₃, 100.09 g/mol). What is the hardness in mg of CaCO₃ per litre? Give the nearest whole number.
3. Analytical chemistry
CoreHumanNickel in a 0.8000 g sample of steel is precipitated as the red complex bis(dimethylglyoximato)nickel(II), Ni(C₄H₇N₂O₂)₂, molar mass 288.92 g/mol. After filtering and drying, the precipitate weighs 0.1972 g. Nickel's molar mass is 58.693 g/mol. What mass of nickel, in milligrams, did the sample contain? Give the nearest whole number.
4. Analytical chemistry
CoreHumanA concentration is calculated as . The relative standard uncertainties are 0.10 % for the mass , 0.05 % for the purity , 0.20 % for the volume , and negligible for the molar mass . The inputs are independent. What percentage of the variance of comes from the volume? Give the nearest whole number.
5. Analytical chemistry
CoreHumanA method for lead in water has a blank standard deviation of 0.0021 (signal units) and a calibration slope of 0.0450 per µg/L, so its detection limit is µg/L and its quantification limit 0.47 µg/L. A sample's signal corresponds to 0.11 µg/L. What should the laboratory report?
6. Analytical chemistry
Warm-upHumanA river sample contains so much dissolved organic matter that it suppresses the instrument's response to copper, so standards made up in pure water would give a biased result. The analyst instead adds known, increasing amounts of copper to equal portions of the sample itself, measures each, fits a line and extrapolates to zero signal. What is this calibration method called?
7. Analytical chemistry
CoreComputationalWrite inverse_prediction(conc, signal, y0). It fits the least-squares line signal = a + b * conc to the calibration standards, then returns a tuple (x0, s_x0): the concentration of an unknown whose replicate signals are the list y0, and its standard error
where is the number of standards and the number of replicates of the unknown. Return plain floats.
8. Analytical chemistry
StretchComputationalWrite two functions.
resolution(t1, w1, t2, w2)returns the resolution of two Gaussian peaks from their retention times and widths at half height, , but use the exact constant in place of 1.18.length_for(L, rs_now, rs_target)returns the column length that would givers_target, if only the plate number changes and the plate number is proportional to the length.
Further reading and resources
- Evaluation of measurement data: Guide to the expression of uncertainty in measurement (GUM) and supplements · Joint Committee for Guides in Metrology (JCGM)The international guide to uncertainty, free from the BIPM, with Supplement 1 on Monte Carlo methods.
- EurachemFree guides for analytical laboratories, including "Quantifying Uncertainty in Analytical Measurement" and "The Fitness for Purpose of Analytical Methods".
- ICH quality guidelines (Q2(R2): Validation of analytical procedures) · International Council for HarmonisationHow the pharmaceutical industry validates methods, including the detection- and quantitation-limit conventions used here.
- ISO/IEC 17025:2017 General requirements for the competence of testing and calibration laboratories · ISO
- Guidelines for drinking-water quality · World Health OrganizationGuideline values for lead and many other substances, and the reasoning behind them.
- Standard Reference Materials · National Institute of Standards and TechnologyCertified reference materials, from blood lead to river sediment, that laboratories use to check their accuracy.
- A new form of chromatogram employing two liquid phases · A. J. P. Martin and R. L. M. Synge · 1941Partition chromatography and the plate theory, recognised by the 1952 Nobel Prize in Chemistry.
- A multi-rule Shewhart chart for quality control in clinical chemistry · James O. Westgard, P. Lorraine Barry, Marian R. Hunt and Torgny Groth · 1981The origin of the "Westgard rules" still used in hospital laboratories.
- Limits for qualitative detection and quantitative determination: application to radiochemistry · Lloyd A. Currie · 1968The paper that separated the decision threshold from the detection limit, with false-positive and false-negative rates.
- IUPAC Compendium of Chemical Terminology (the Gold Book) · IUPACAgreed definitions of terms such as limit of detection, selectivity and resolution.