Skip to content
LumenaryAcademy
Methods 75 min

Spectroscopy and thermal analysis

Finding out what a solid is made of and what it does when heated — X-ray microanalysis, photoelectron and Raman spectroscopy, calorimetry, thermogravimetry and dilatometry, and how to choose among them.

Code tested · 6AI-draftedAwaiting review by a materials scientist

Your guides: Chloe · The Materials Scientist, Margaret · The Tester, Gideon · The Failure Analyst

MotivationH

What is it made of, and what does it do when heated?

A supplier changes nothing, it says, yet a new batch of polymer clips cracks during assembly while the old batch did not. An aluminium panel that bonded well for years suddenly lets its adhesive peel. A nickel alloy forging fails an inspection for a small, hard particle sitting at the origin of a crack. In each case microscopy, from the last lesson, shows where the problem is. The next questions are chemical and thermal: which elements are in that particle, what is the chemical state of the aluminium's surface, how crystalline is that polymer, what happens to it on heating?

This lesson covers the instruments that answer them. Three spectroscopies tell what a solid is made of: energy-dispersive X-ray spectroscopy (EDS) gives elements in a micrometre-sized spot under the electron microscope; X-ray photoelectron spectroscopy (XPS) gives elements and their chemical state in the top few nanometres; Raman spectroscopy gives the vibrations of bonds, which fingerprint phases and stresses. Three thermal methods tell how it behaves: differential scanning calorimetry (DSC) measures heat absorbed or released, thermogravimetric analysis (TGA) measures mass lost or gained, and dilatometry measures length. Lumenary Chemistry's spectroscopy lesson covers the molecular methods (UV–visible, infrared, NMR, mass spectrometry) and analytical chemistry covers calibration, uncertainty and trace analysis; here the subject is solids.

IntuitionH

Probe in, signal out

Every technique in this lesson sends something into the sample and measures what comes out. What goes in and what comes out decides what the technique can see and from how deep.

EDSXPSRamanelectrons inX-rays outX-rays inelectrons outlaser inshifted light out~1 µm deeptop few nm~1 µm spot
Three spectroscopies as probe and signal. In EDS an electron beam makes atoms emit characteristic X-rays from a volume about a micrometre deep. In XPS X-rays eject photoelectrons, and only those from the top few nanometres escape without losing energy. In Raman spectroscopy a laser is scattered with its colour shifted by the energy of a vibration, from a spot about a micrometre across.
TechniqueInOutTells youFrom
EDS (and WDS)electronscharacteristic X-rayselements, about 0.1 wt% and upabout 1 µm³ in an SEM
XPSX-raysphotoelectronselements and chemical statestop 5–10 nm
Ramanlaser lightinelastically scattered lightbonds, phases, stress, disorderabout 1 µm spot
DSCheatheat flowtransitions, enthalpies, heat capacitywhole sample, mg
TGAheatmassdecomposition, moisture, filler, oxidationwhole sample, mg
Dilatometryheatlengthexpansion, transformations, sinteringwhole sample, mm
ComputationC

EDS: elements under the electron beam

When a beam electron knocks an inner-shell electron out of an atom, an electron from a higher shell drops into the hole and the energy difference leaves as an X-ray photon. Because inner-shell energies depend almost entirely on the nuclear charge, each element emits X-rays at its own characteristic energies: the Kα\alpha line (L shell to K shell), Kβ\beta, Lα\alpha and so on. In 1913 Henry Moseley, at Manchester, measured them for a series of elements and found that the square root of the frequency rises in a straight line with atomic number, which settled the order of the periodic table and showed where elements were missing. A simple model, a K-shell hole screened by the one remaining K electron, gives Moseley's law for the Kα\alpha energy, E≈34×13.6 eV×(Z−1)2E \approx \tfrac34 \times 13.6\ \text{eV} \times (Z-1)^2.

An energy-dispersive detector, today a silicon drift detector on the side of an SEM or TEM, converts each arriving photon into a pulse of charge proportional to its energy and sorts the pulses into a spectrum. It records all elements from boron or carbon upwards at once, but its energy resolution is limited, about 125 to 130 eV at 5.9 keV. A wavelength-dispersive spectrometer (WDS) instead diffracts the X-rays from a crystal by Bragg's law and counts one wavelength at a time: slower, but with a resolution of a few eV. The block checks Moseley's law against tabulated energies and shows what resolution means for a famous overlap.

python
import math
import numpy as np

# 1. Moseley's law: K-alpha photon energy ~ (3/4) x 13.606 eV x (Z - 1)^2,
#    compared with tabulated K-alpha1 energies (X-ray Data Booklet), in keV.
tabulated = {"Al": (13, 1.487), "Si": (14, 1.740), "S": (16, 2.308), "Ti": (22, 4.511),
             "Fe": (26, 6.404), "Ni": (28, 7.478), "Cu": (29, 8.048), "Mo": (42, 17.479)}
print("element   Z   Moseley (keV)   tabulated (keV)   difference")
for el, (Z, E) in tabulated.items():
    E_m = 0.75 * 13.606 * (Z - 1) ** 2 / 1000
    print(f"{el:>7} {Z:3d}   {E_m:13.3f}   {E:15.3f}   {100 * (E_m - E) / E:+8.1f} %")

# 2. Sulfur K-alpha (2308 eV) next to molybdenum L-alpha (2293 eV), seen by an EDS detector
#    and by a wavelength-dispersive (WDS) spectrometer.  The EDS peak width grows with energy:
#    FWHM^2 = FWHM(Mn K-alpha)^2 + 2.5 (E - 5895 eV), all in eV, for a detector rated at 130 eV.
def gaussian_lines(E, fwhm, centres=(2293.0, 2308.0)):
    s = fwhm / 2.3548
    return sum(np.exp(-0.5 * ((E - c) / s) ** 2) for c in centres)

w_eds = math.sqrt(130.0**2 + 2.5 * (2300 - 5895))
w_wds = 8.0
fine = np.arange(2200.0, 2400.0, 0.25)
for name, w in [("EDS", w_eds), ("WDS", w_wds)]:
    y = gaussian_lines(fine, w)
    maxima = int(np.sum((y[1:-1] > y[:-2]) & (y[1:-1] > y[2:])))
    print(f"\n{name}: peak width {w:.0f} eV; maxima in the summed spectrum: {maxima}")

print("\nenergy (eV)   EDS                    WDS")
for e in np.arange(2272.0, 2330.0, 4.0):
    a = gaussian_lines(e, w_eds) / gaussian_lines(fine, w_eds).max()
    b = gaussian_lines(e, w_wds) / gaussian_lines(fine, w_wds).max()
    print(f"{e:11.0f}   {'#' * round(20 * a):20s}   {'#' * round(20 * b)}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Moseley's crude model is within 2 per cent for every element in the table, and within half a per cent from sulfur to nickel, good enough to identify a peak on sight. The second part is the problem every microanalyst meets. Sulfur's Kα\alpha and molybdenum's Lα\alpha lines are only 15 eV apart, while the EDS peak at that energy is 89 eV wide; the two merge into a single peak, and a sulfide inclusion in a molybdenum-bearing steel can be misread. Lead's Mα\alpha line, at 2.35 keV, joins the same crowd. With 8 eV peaks, WDS shows two clean maxima. Software deconvolves overlaps by fitting reference peak shapes, but only if the analyst has thought to include the right elements.

Turning peak heights into concentrations needs ZAF corrections for how the matrix affects the X-rays: atomic number (how far electrons penetrate and how many backscatter), absorption on the way out, and fluorescence excited by other X-rays. With standards and a flat polished specimen, an electron probe microanalyser with WDS reaches about 1 per cent relative accuracy for major elements; "standardless" EDS on a rough surface is often off by 10 per cent relative or more.

ComputationC

XPS: the top few nanometres, and chemical states

In X-ray photoelectron spectroscopy, monochromatic X-rays, usually aluminium Kα\alpha at 1486.6 eV, eject electrons from the atoms of the sample, and an analyser measures their kinetic energy. Energy conservation gives the binding energy of the level each came from:

EB=hν−EK−ϕ,E_\text{B} = h\nu - E_\text{K} - \phi,

where ϕ\phi is the work function of the spectrometer. Core-level binding energies identify the element, as X-ray energies do in EDS. The decisive difference is that they also shift by a few electronvolts with the atom's chemical state. Kai Siegbahn's group in Uppsala developed the method in the 1950s and 1960s, called it ESCA (electron spectroscopy for chemical analysis), and Siegbahn shared the 1981 Nobel Prize in Physics for it.

Electrons travelling through a solid lose energy in collisions; only those that escape without any loss contribute to the sharp peak. The chance of escaping from depth zz falls as e−z/λe^{-z/\lambda}, where the inelastic mean free path λ\lambda is a few nanometres for electrons of about 1 keV. That is what makes XPS a surface technique, and it also lets you measure the thickness of very thin films. The block works through three calculations for a silicon wafer with an oxide on it.

python
import math

# 1. XPS: binding energy from the measured kinetic energy (Al K-alpha source).
h_nu, phi = 1486.6, 4.5                       # photon energy and spectrometer work function, eV
for label, KE in [("peak A", 1382.7), ("peak B", 1378.6)]:
    print(f"{label}: kinetic energy {KE} eV -> binding energy {h_nu - KE - phi:.1f} eV")

# 2. Where does the signal come from?  Electrons from depth z escape without losing energy
#    with probability exp(-z / (lambda cos theta)), theta measured from the surface normal.
lam = 3.0                                     # nm, a typical inelastic mean free path here
print("\nfraction of the signal from the top ...    1 nm    3 nm    6 nm    9 nm")
for theta in (0, 60, 75):
    L = lam * math.cos(math.radians(theta))
    fr = [1 - math.exp(-t / L) for t in (1, 3, 6, 9)]
    print(f"take-off {theta:2d} deg from the normal:      " + "  ".join(f"{f:6.0%}" for f in fr))

# 3. Oxide thickness from how much it dims the silicon signal underneath.
ratio = 0.38                                  # Si(0) 2p intensity: oxidised / freshly cleaned wafer
print(f"\nsubstrate signal at {ratio:.0%} of the clean value -> oxide {-lam * math.log(ratio):.2f} nm thick")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The two Si 2p peaks have binding energies of 99.4 and 103.5 eV, matching reference values for elemental silicon and silicon dioxide in the NIST database: the wafer and its oxide, told apart in one spectrum. Looking straight down, 63 per cent of the signal comes from the top 3 nm and 95 per cent from the top 9 nm, three mean free paths. Tilting the sample so the analyser looks at 75° from the normal squeezes 72 per cent of the signal into the top nanometre: angle-resolved XPS uses this to separate a surface layer from what lies under it. If the substrate signal falls to 38 per cent of a clean wafer's, the oxide is 2.9 nm thick.

In practice XPS needs ultra-high vacuum, detects every element except hydrogen and helium down to about 0.1 to 1 atomic per cent, and analyses a spot typically tens to hundreds of micrometres across. Depth profiles are made by sputtering with argon ions, or with clusters of thousands of argon atoms for polymers; insulators charge up and are corrected with a low-energy electron flood and a reference peak. The peeling adhesive in the motivation is a classic XPS problem: a few nanometres of silicone release agent are invisible to EDS and obvious to XPS.

Worked examplesH

Raman spectroscopy: listening to the lattice

Shine a laser on a solid and almost all the scattered light keeps its colour. A tiny fraction, around one scattered photon in a million or fewer, comes out with a little less energy, having left it behind as a vibration of the lattice or a molecule (Stokes scattering), or with a little more, having picked up a vibration's energy (anti-Stokes). The shift, quoted in wavenumbers, cm⁻¹, is the vibration's frequency and does not depend on the laser. C. V. Raman and K. S. Krishnan reported the effect from Calcutta in 1928, and Grigory Landsberg and Leonid Mandelstam found it in quartz in Moscow the same year; Raman received the 1930 Nobel Prize in Physics.

A vibration appears in the Raman spectrum if it changes the material's polarisability, a different rule from infrared absorption, which needs a changing dipole (see Lumenary Chemistry's spectroscopy lesson), so the two are complementary. For solids, Raman spectra are fingerprints:

  • Carbon. Diamond has a single sharp line at 1332 cm⁻¹. Graphite has its G band near 1580 cm⁻¹; disorder and crystallite edges activate a D band near 1350 cm⁻¹, and the D/G ratio measures how disordered a carbon is. The shape of the 2D band near 2700 cm⁻¹ counts the layers in graphene.
  • Polymorphs. Anatase and rutile, two forms of TiO₂ with the same composition, give entirely different spectra (anatase's strongest line is near 144 cm⁻¹), so Raman distinguishes them where EDS cannot.
  • Stress. Compressing a crystal stiffens its bonds and raises the frequency of its lines. Silicon's line near 520 cm⁻¹ shifts by a few cm⁻¹ per gigapascal, which is how the strain built into transistor channels is mapped.

Metals have no useful Raman spectrum, because their free electrons screen the light, but their oxides, corrosion products and coatings do.

Example— Where the lines are, and the laser as a heater

With a 532 nm green laser, whose wavenumber is 107/532=18 79710^7/532 = 18\,797 cm⁻¹, silicon's 520 cm⁻¹ Stokes line appears at 107/(18 797−520)=547.110^7/(18\,797 - 520) = 547.1 nm and the anti-Stokes line at 107/(18 797+520)=517.710^7/(18\,797 + 520) = 517.7 nm. Anti-Stokes scattering needs a vibration that is already excited, so its strength relative to the Stokes line follows the Boltzmann factor e−hcν~/kBTe^{-hc\tilde\nu/k_\text{B}T}, multiplied by a factor for the different frequencies of the two lines. At 295 K the ratio works out at 0.099; at 400 K, 0.192; at 600 K, 0.359. A measured ratio of 0.120 under a tightly focused laser therefore means the spot is at about 320 K: the laser is heating the sample by 25 K. That matters, because heat shifts Raman lines too and can transform delicate phases under the beam.

Raman microscopes focus the laser to a spot about a micrometre across and work in air, through windows and in liquids, ideal for watching corrosion or battery electrodes in operation. The lasers can harm eyes, so instruments are enclosed and interlocked.

ComputationC

Differential scanning calorimetry

A differential scanning calorimeter heats a few milligrams of sample in a small pan, alongside an empty reference pan, at a steady rate, typically 10 K/min. A heat-flux DSC measures the small temperature difference between the two pans and converts it into a heat flow; a power-compensated DSC heats the two pans with separate heaters and measures the difference in power needed to keep them at the same temperature. Either way the output is the extra heat flow into the sample, in milliwatts, against temperature. Away from any transition it is the sample's heat capacity times the heating rate. Transitions show as features:

  • a glass transition as a step, because the heat capacity rises when an amorphous polymer or glass becomes able to rearrange (polymers, glasses and nanomaterials);
  • crystallisation, a release of heat, as an exothermic peak;
  • melting, which absorbs heat, as an endothermic peak whose area is the enthalpy of fusion.

The block simulates the curve of a PET-like polyester that was cooled quickly from the melt, so that it is mostly amorphous, then analyses it as a thermal analyst would: the midpoint of the glass-transition step, and the peak areas above straight baselines. The crystallinity is the enthalpy of melting, less any that came from crystals formed during the scan itself, divided by the enthalpy of fusion of a fully crystalline polymer; about 140 J/g is the value commonly used for PET.

python
import math
import numpy as np

rng = np.random.default_rng(42)

# A simulated DSC run on 8.0 mg of a PET-like polymer that was quenched from the melt,
# heated at 10 K/min.  Heat flow in mW, endothermic up.
mass_g, beta = 8.0e-3, 10 / 60                     # g, K/s
T = np.arange(30.0, 290.01, 0.1)                   # deg C

def gauss(T, centre, width):                       # area 1 when integrated over T
    return np.exp(-0.5 * ((T - centre) / width) ** 2) / (width * math.sqrt(2 * math.pi))

cp = 1.15 + 0.0015 * (T - 30) + 0.30 / (1 + np.exp(-(T - 78) / 1.8))   # J/(g K): step at Tg
true_cc, true_m = -27.0, 39.0                      # J/g: cold crystallisation (exo), melting (endo)
dH = true_cc * gauss(T, 133, 5.0) + true_m * gauss(T, 251, 4.5)       # J/(g K)
heat_flow = 1e3 * mass_g * beta * (cp + dH) + rng.normal(0, 0.004, T.size)   # mW

# --- Analysis, as a thermal analyst would do it ---
def line_fit(lo, hi):
    sel = (T >= lo) & (T <= hi)
    return np.polyfit(T[sel], heat_flow[sel], 1)

# Glass transition: the midpoint, where the curve is halfway between the baselines before and after.
before, after = line_fit(50, 68), line_fit(90, 110)
window = (T > 68) & (T < 90)
half = (np.polyval(before, T) + np.polyval(after, T)) / 2
Tg = T[window][np.argmin(np.abs(heat_flow[window] - half[window]))]
dcp = (np.polyval(after, Tg) - np.polyval(before, Tg)) / (1e3 * mass_g * beta)
print(f"glass transition (midpoint): {Tg:.1f} C, step in heat capacity {dcp:.2f} J/(g K)")

# Peaks: integrate above a straight baseline joining the curve on either side of the peak.
def peak_enthalpy(lo, hi):
    sel = (T >= lo) & (T <= hi)
    Ts, q = T[sel], heat_flow[sel]
    base = q[0] + (q[-1] - q[0]) * (Ts - Ts[0]) / (Ts[-1] - Ts[0])
    excess = q - base
    area_mJ = np.sum((excess[1:] + excess[:-1]) / 2 * np.diff(Ts)) / beta   # mW s = mJ
    peak_T = Ts[np.argmax(np.abs(excess))]
    return area_mJ / (mass_g * 1e3), peak_T                                 # J/g

H_cc, T_cc = peak_enthalpy(112, 155)
H_m, T_m = peak_enthalpy(228, 272)
print(f"cold crystallisation: peak {T_cc:.1f} C, {H_cc:+.1f} J/g (exothermic)")
print(f"melting:              peak {T_m:.1f} C, {H_m:+.1f} J/g (endothermic)")

H100 = 140.0                                       # J/g, a commonly used value for 100 % crystalline PET
print(f"\ncrystallinity as received: ({H_m:.1f} - {abs(H_cc):.1f}) / {H100:.0f} = {(H_m - abs(H_cc)) / H100:.0%}")
print(f"crystallinity just before melting: {H_m / H100:.0%}")
print(f"(the simulation was built with {(true_m + true_cc) / H100:.0%} and {true_m / H100:.0%})")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The analysis recovers what was put in: a glass transition at 78.0 °C with a heat-capacity step of 0.30 J/(g K), a cold-crystallisation peak at 133 °C releasing 26.9 J/g, and melting at 251 °C absorbing 39.1 J/g. The trap is in the crystallinity. The melting peak alone suggests 28 per cent, but most of those crystals grew during the scan, in the cold-crystallisation peak, when the warmed glass gained enough mobility to order. The material as received was only 9 per cent crystalline. That is the kind of difference that separates a clip that flexes from one that cracks.

024650100150200250temperature (°C)heat flow (mW), endo upglass transitioncold crystallisation(exothermic)melting(endothermic)
The simulated DSC curve from the block, without the noise: a step at the glass transition, an exothermic dip as the warmed glass crystallises, and an endothermic peak as the crystals melt. Instruments differ in which way they plot exothermic; always check the label.

DSC is just as useful for metals and composites: it finds an alloy's solidus and liquidus, the precipitation reactions of an age-hardening aluminium alloy (phase transformations) and the cure of an epoxy matrix. Chip-based fast calorimeters heat and cool at thousands of kelvin per second, fast enough to mimic injection moulding or additive manufacturing.

ComputationC

Thermogravimetric analysis

A thermogravimetric analyser hangs or rests a sample on a microbalance, resolving fractions of a microgram, inside a furnace, and records its mass while heating it in a chosen gas. Losses mean something left: moisture, solvent, a decomposing polymer, carbon dioxide from a carbonate. Gains mean something arrived, usually oxygen. The derivative of the mass curve, the DTG curve, turns each step into a peak at the temperature where the loss is fastest. A gas analyser (a mass spectrometer or an infrared spectrometer) on the exhaust identifies what left.

Calcium oxalate monohydrate is the classic test material because it decomposes in three clean steps: it loses its water, then carbon monoxide, then carbon dioxide, leaving calcium oxide. The block computes the step sizes from stoichiometry, simulates a TGA curve with illustrative first-order kinetics that place the steps near where they are usually observed at 10 K/min, and then demonstrates Kissinger's method: the DTG peak moves to higher temperature at faster heating rates, and plotting ln⁡(β/Tp2)\ln(\beta/T_\text{p}^2) against 1/Tp1/T_\text{p} gives a straight line of slope −E/R-E/R, from which the activation energy EE follows.

python
import math
import numpy as np

R = 8.314                                         # J/(mol K)

# 1. Stoichiometry: calcium oxalate monohydrate loses H2O, then CO, then CO2, leaving CaO.
M = {"Ca": 40.078, "C": 12.011, "O": 15.999, "H": 1.008}
M_start = M["Ca"] + 2 * M["C"] + 5 * M["O"] + 2 * M["H"]          # CaC2O4.H2O
gases = [("H2O", 2 * M["H"] + M["O"]), ("CO", M["C"] + M["O"]), ("CO2", M["C"] + 2 * M["O"])]
left = 100.0
print(f"CaC2O4.H2O, M = {M_start:.2f} g/mol")
for gas, m in gases:
    left -= 100 * m / M_start
    print(f"  loses {gas:4s} = {100 * m / M_start:5.2f} % of the starting mass -> {left:6.2f} % left")

# 2. A simulated TGA run.  Each step is first order with an Arrhenius rate constant
#    k = A exp(-E/RT); at a constant heating rate beta its conversion is
#    alpha(T) = 1 - exp(-(A / beta) * integral of exp(-E/RT) dT).
#    Illustrative activation energies; each A is chosen so that at 10 K/min the step is
#    fastest at the temperature given (the peak condition E beta / (R Tp^2) = A exp(-E/(R Tp))).
plan = [(90e3, 180.0), (180e3, 495.0), (200e3, 745.0)]          # (E in J/mol, peak in deg C)
steps = []
for (gas, m), (E, Tp_C) in zip(gases, plan):
    Tp = Tp_C + 273.15
    steps.append((m / M_start, E, E * (10 / 60) / (R * Tp**2) * math.exp(E / (R * Tp))))

T = np.arange(25.0, 950.0, 0.05) + 273.15         # K

def tga(beta):
    """Mass (%) and DTG (-d mass / dT, % per K) at heating rate beta in K/s."""
    mass = np.full_like(T, 100.0)
    for fraction, E, A in steps:
        k = np.exp(-E / (R * T))
        integral = np.concatenate(([0.0], np.cumsum((k[1:] + k[:-1]) / 2 * np.diff(T))))
        mass -= 100 * fraction * (1 - np.exp(-A / beta * integral))
    return mass, -np.gradient(mass, T)

mass, dtg = tga(10 / 60)
print("\nat 10 K/min:  T (C)   mass (%)")
for t in range(100, 901, 100):
    print(f"             {t:5d}   {mass[np.argmin(np.abs(T - 273.15 - t))]:7.2f}")

# 3. Kissinger: the DTG peak of the CO2 step shifts with heating rate, and
#    ln(beta / Tp^2) against 1/Tp is a straight line of slope -E/R.
print("\nbeta (K/min)   CO2 step fastest at (C)")
x, y = [], []
for b in (2, 5, 10, 20):
    m, d = tga(b / 60)
    sel = T > 873.15                               # look only at the third step
    Tp = T[sel][np.argmax(d[sel])]
    x.append(1 / Tp)
    y.append(math.log((b / 60) / Tp**2))
    print(f"{b:12d}   {Tp - 273.15:22.1f}")
slope = np.polyfit(x, y, 1)[0]
print(f"Kissinger activation energy: {-slope * R / 1e3:.1f} kJ/mol (the simulation used 200.0)")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The steps land where stoichiometry says they must: 12.33, 19.17 and 30.12 per cent of the starting mass, leaving 38.38 per cent as calcium oxide. That is the point of a test material, and the same reasoning reads real samples. A glass-fibre-reinforced nylon heated in air burns off its polymer and leaves the glass: a residue of 30 per cent means 30 per cent glass by mass. A carbonate that loses 44 per cent of its mass was pure calcium carbonate. A metal that gains mass in air is oxidising, and the rate of gain at a fixed temperature measures its oxidation resistance. In the kinetics part, the CO₂ step's fastest point moves from 685 °C at 2 K/min to 773 °C at 20 K/min, and Kissinger's plot recovers exactly the 200 kJ/mol the simulation was built with.

Worked examplesH

Dilatometry: measuring length

A dilatometer holds a specimen, typically a rod or cylinder 10 to 25 mm long, in a furnace and measures its length with a push rod and a displacement sensor, resolving a few nanometres. It measures the coefficient of thermal expansion, α=(1/L0) dL/dT\alpha = (1/L_0)\,\mathrm{d}L/\mathrm{d}T, which matters wherever materials are joined: a chip on a circuit board, a ceramic coating on a turbine blade, a glass sealed to a metal. Typical values near room temperature span two orders of magnitude, from about 0.5 × 10⁻⁶ K⁻¹ for fused silica and about 1 × 10⁻⁶ K⁻¹ for the iron–nickel alloy Invar to about 12 × 10⁻⁶ K⁻¹ for steel and 23 × 10⁻⁶ K⁻¹ for aluminium.

Phase transformations show up as changes in slope or sudden steps, because different crystal structures pack atoms at different densities.

Example— Seeing austenite form in a steel

Near 912 °C, X-ray diffraction gives lattice parameters of about 0.2904 nm for body-centred cubic iron (2 atoms per cell) and 0.3647 nm for face-centred cubic iron (4 atoms per cell). The volume per atom is 0.29043/2=0.0122450.2904^3/2 = 0.012245 nm³ in the BCC form and 0.36473/4=0.0121270.3647^3/4 = 0.012127 nm³ in the FCC form: the close-packed structure is 0.97 per cent denser, so each unit length shrinks by about 0.32 per cent when the iron transforms.

A 10.000 mm steel specimen heated from 20 to 700 °C, with a mean expansion coefficient of 14 × 10⁻⁶ K⁻¹, grows by 10.000×14×10−6×680=0.09510.000 \times 14 \times 10^{-6} \times 680 = 0.095 mm. When it then transforms to austenite it shrinks by about 0.032 mm while still being heated, a feature impossible to miss on an instrument that resolves nanometres. On cooling, the reverse transformations expand the specimen, and the temperatures at which the expansion starts and finishes mark them.

Quenching dilatometers heat small specimens by induction and cool them with gas jets at controlled rates; a series of such runs, with micrographs and hardness tests, is how the continuous-cooling-transformation diagrams of phase transformations are measured, with the martensite start MsM_\text{s} showing as a sharp expansion. Dilatometers also follow the shrinkage of sintering ceramics.

ExplanationH

Choosing a technique for a question

The skill that matters most is matching the instrument to the question. Start from what you need to know, not from the instrument you have.

QuestionFirst choiceWhy
Which elements are in a 5 µm inclusion?SEM with EDSelements at a micrometre scale, quickly
Is there a trace of sulfur next to molybdenum, or boron in a steel?WDS in an electron proberesolves overlaps; better detection limits and light elements
What is the oxidation state of chromium in a 3 nm passive film?XPS, with depth profiling or angle resolutionchemical state in the top nanometres
Which phases are present, and how much of each?X-ray diffractioncrystal structures, quantitative by Rietveld refinement
Is this carbon coating diamond-like or graphitic? Is this TiO₂ anatase or rutile?Ramanbonding and polymorphs, at micrometre scale
How crystalline is this polymer? What is its glass transition?DSCheat of fusion and heat-capacity step
How much filler, moisture or residual solvent is in it?TGA, ideally with evolved-gas analysismass lost at each temperature
At what temperature does this steel transform at 5 K/s?quenching dilatometrylength changes at controlled rates
How much hydrogen is in this weld?thermal desorption or hot extractionEDS and XPS cannot see hydrogen
Parts per million of lead in a solder?ICP-MS on a dissolved sampletrace analysis, as in Lumenary Chemistry's analytical chemistry

The cases in the motivation resolve this way. The cracking clips: DSC shows the new batch was moulded into a colder tool and is less crystalline, and TGA shows the filler content is unchanged. The peeling adhesive: XPS finds a few nanometres of silicone on the aluminium, transferred from a new release spray, which EDS could never have seen. The hard particle in the nickel forging: SEM backscattered imaging finds it, EDS shows titanium and nitrogen, and WDS confirms the nitrogen, which EDS measures poorly: a titanium nitride inclusion from the melt.

ExplanationH

Practice and standards

Results from these methods are only comparable when they are done the same way, so most have international standards: the ISO 11357 series for DSC of plastics and ISO 11358 for their thermogravimetry, for example, set specimen sizes, heating rates and how to construct baselines. Accredited laboratories calibrate against certified reference materials and take part in interlaboratory comparisons, as Lumenary Chemistry's analytical chemistry describes.

The research frontier pushes three ways. Operando measurements follow materials while they work: Raman and X-ray spectroscopy of battery electrodes during charging, near-ambient-pressure XPS of catalysts in gas, calorimetry of cells as they age. Synchrotrons add X-ray absorption spectroscopy, which gives oxidation states and local structure for any element in any phase, crystalline or not, and hard-X-ray photoemission, which sees tens of nanometres deep. And atom probe tomography field-evaporates a needle-shaped specimen atom by atom and reconstructs the positions and identities of tens of millions of atoms in three dimensions, the most direct view yet of clusters and segregation in alloys. Reference data are shared openly: the NIST XPS database, the X-ray Data Booklet and the RRUFF Raman database are in the references.

Common misconceptionsH

Common misconceptions

Misconception

"EDS gives the surface composition." EDS in the SEM samples a volume about a micrometre deep, thousands of atomic layers. For the surface, the top few nanometres, use XPS.

Misconception

"No peak means the element is absent." It means the element is below the detection limit, or is hydrogen or helium (invisible to both EDS and XPS), or has its peak hidden under another element's. Raman is silent for vibrations that do not change the polarisability and for metals.

Misconception

"A DSC peak temperature is a property of the material." Peak and onset temperatures move with heating rate, sample mass and thermal history; Kissinger's method relies on exactly that. Report the heating rate with every temperature, and remember that a polymer's crystallinity depends on how it was made.

Misconception

"Mass lost in a TGA means the material decomposed." It may be moisture or solvent leaving at low temperature, and in air a metal or carbon-containing material can gain mass by oxidation. Identify the gas before naming the reaction.

Exercises

Exercises

Molar masses: H 1.008, C 12.011, O 15.999, Ca 40.078 g/mol.

1. Spectroscopy and thermal analysis

Warm-upHuman

An XPS instrument uses aluminium Kα\alpha X-rays (hν=1486.6h\nu = 1486.6 eV) and has a spectrometer work function of 4.5 eV. A carbon 1s peak appears at a kinetic energy of 1196.1 eV. What is its binding energy, in eV?

2. Spectroscopy and thermal analysis

Warm-upHuman

Use Moseley's law, EKα≈34×13.6 eV×(Z−1)2E_{\text{K}\alpha} \approx \tfrac34 \times 13.6\ \text{eV} \times (Z-1)^2, to estimate the Kα\alpha energy of copper (Z=29Z = 29). Give the answer in eV, rounded to the nearest whole eV.

3. Spectroscopy and thermal analysis

CoreHuman

A stainless-steel fastener has discoloured in service. You need to know whether its passive film, a few nanometres thick, is still rich in chromium oxide or has become mostly iron oxide. Which technique answers this most directly?

Choose one

4. Spectroscopy and thermal analysis

CoreHuman

Magnesium hydroxide, Mg(OH)₂, is used as a flame-retardant filler in cables because it decomposes on heating to MgO and water vapour, absorbing heat. In a TGA run on pure Mg(OH)₂, what percentage of the starting mass is lost in this step? Give the nearest whole per cent. (Mg 24.305, O 15.999, H 1.008 g/mol.)

5. Spectroscopy and thermal analysis

Warm-upHuman

In the Raman spectrum of a graphitic carbon, which band, near 1350 cm⁻¹ with a green laser, grows as the material becomes more disordered? Give its usual one-letter name.

6. Spectroscopy and thermal analysis

CoreHuman

A dilatometer heats a specimen of plain low-carbon steel steadily from room temperature to 1000 °C. What does its length do as the ferrite transforms to austenite?

Choose one

7. Spectroscopy and thermal analysis

CoreComputational

Write peak_enthalpy(T, heat_flow_mW, mass_mg, beta_K_per_min, T_lo, T_hi) that returns the enthalpy of a DSC peak in J/g. Take the points with T_lo <= T <= T_hi, draw a straight baseline from the first to the last of them, integrate the heat flow above the baseline over temperature with the trapezoidal rule, convert from an integral over temperature to one over time using the heating rate, and divide by the mass. An exothermic (downward) peak should give a negative number.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

8. Spectroscopy and thermal analysis

StretchComputational

Write kissinger_energy(betas_K_per_min, peak_temps_C) that returns the activation energy in kJ/mol from Kissinger's method: fit a straight line to ln⁡(β/Tp2)\ln(\beta/T_\text{p}^2) against 1/Tp1/T_\text{p}, with TpT_\text{p} in kelvin, and return E=−R×slopeE = -R \times \text{slope}, with R=8.314R = 8.314 J mol⁻¹ K⁻¹. (Expressing β\beta in K/min rather than K/s changes only the intercept, not the slope.)

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Further reading and resources