Rietveld Refinement, Texture, and Residual Stress

From peak-by-peak to the whole pattern at once

Guide 3 taught you to read a powder pattern one peak at a time — position for the cell, intensity for the motif, width for the crystallite size. Guide 4 used those intensities to solve a first, rough structure, cracking the phase problem with direct methods or a Patterson map and building an electron-density map. But there is a nasty catch that guide 2 warned about: a powder squashes three dimensions of reflections down into one axis, so peaks pile on top of one another. Read them separately and heavily overlapped peaks blur into a single hump whose intensity you can only guess at. Extracting clean individual intensities from a real powder pattern is often simply impossible.

Hugo Rietveld's answer in 1969 was to stop fighting the overlap and embrace the whole curve. Rietveld refinement never tries to extract separate intensities. Instead it takes a structural model — the atoms you solved in guide 4, or one pulled from a database — and CALCULATES the entire pattern the model would produce, point by point across every value of 2-theta: the background, the position of every reflection, its height, and the exact shape of each peak. Then it compares that calculated curve to the one your diffractometer actually measured and nudges the model until the two curves lie on top of each other. Overlapping peaks are no longer a problem to be untangled; they are just part of the shape you are fitting.

What Rietveld actually adjusts

The whole method is one big least-squares fit: it varies a set of numbers to make the weighted sum of squared differences between observed and calculated counts as small as it can. The beauty is that the knobs it turns are not abstract — they are exactly the three physical stories of a peak you already know. Position parameters (the lattice parameters, plus a small specimen-displacement correction) slide the whole comb of reflections left or right. Intensity parameters (the atomic coordinates, site occupancies, and the thermal displacement factors) feed the structure factor and set every peak's height. Width parameters govern peak shape, and this is where the Scherrer story of size and microstrain lives.

Two more knob groups are pure housekeeping but matter a lot in practice: a smooth background curve soaks up scattering from air, sample holder, and any glassy content, while a scale factor sets how much of the pattern this phase contributes. You do not free all of these at once. Turning every parameter loose from the start lets the fit wander into nonsense, because many knobs are correlated — a wrong background can be faked by wrong thermal factors, for instance. The craft of Rietveld is the ORDER in which you release parameters, from the robust and global to the delicate and local.

1. Start from a solved or database structure and a well-calibrated instrument — never from a blank guess.
2. Free the background and the overall scale factor first, so the calculated curve sits at roughly the right level.
3. Free the lattice parameters and the specimen-displacement term to lock every peak onto its measured position.
4. Free the peak-shape width parameters so the calculated peaks are as broad or sharp as the real ones — this captures size and strain.
5. Only now free the delicate motif parameters — atomic coordinates, then occupancies, then thermal factors — and add a preferred-orientation correction if peaks are systematically off.

Did the fit succeed? Read the difference curve, not just the number

How do you know the refinement worked? Software reports R-factors: the weighted-profile R (Rwp) measures how far the calculated curve strays from the observed one, and the expected R (Rexp) is the best you could possibly do given the counting noise. Their ratio, the goodness-of-fit S = Rwp / Rexp (equal to the square root of chi-squared), ideally settles near 1 — you cannot beat the statistical noise floor, so an S of about 1.1 to 1.5 is a healthy fit. An S far below 1 usually means you over-modeled or the counting times were short, not that you did brilliantly.

THE PLOT THAT ACTUALLY JUDGES A REFINEMENT

  I |        + = observed        line = calculated
    |         +
    |        +++          +
    |       + | +        +++          +
    |   +++ +   + +     + | +   +++   +++
    |__+_____|___+_+___+___+_+_+___+_+___+__
    |  | | |  | |  | |   |  |  | | |  |     <- tick marks: allowed hkl
    |
  o-c|  difference (observed minus calculated) -- want flat noise
    |__/\____________/\______v____________  <- a spike here = trouble
    +----+----+----+----+----+----+----> 2theta (degrees)

Texture: when the crystals are not random after all

Everything so far leaned on guide 2's bargain: a powder holds millions of tiny crystals pointing every which way, so a fair share is correctly angled for each plane family and the measured intensities match the ideal random-powder values. Real materials often break that promise. Roll a sheet of metal, draw a wire, or grow a film and the grains line up — a preferred orientation crystallographers call texture. Think of floor tiles that were meant to be scattered at random angles but were instead nearly all laid the same way. When that happens the powder pattern lies: some peaks tower far above their fair height while others shrink, purely because too many or too few grains are oriented to reflect. That is exactly the distortion the preferred-orientation knob in the last section quietly corrects.

But texture is not only a nuisance to correct away — it is often the very thing you want to measure, because it makes a material directionally different: transformer steel is textured on purpose so it magnetizes easily along the rolling direction, and a badly textured aluminium sheet forms ugly ears when you draw it into a can. To map it, you fix the detector on a single reflection at its Bragg angle, then tilt and spin the sample through every orientation, recording how the intensity rises and falls. The result, plotted on a stereographic net, is a pole figure: a map of where that plane's normal prefers to point. A companion inverse pole figure asks the reverse — which crystal directions align with a chosen sample axis, such as the rolling direction.

A single pole figure is only a two-dimensional shadow of the full picture, though — it shows one plane family projected onto the sphere. Measure several pole figures from different reflections and you can reconstruct the complete three-dimensional orientation distribution function (ODF), the probability density of every possible grain orientation. Pole figures are the shadows; the ODF is the solid object casting them. (Electron backscatter diffraction, from another rung, gets at the same orientations grain by grain in a microscope, but pole figures from an ordinary diffractometer average over millions of grains at once — cheap, fast, and bulk.)

Residual stress and thin films: the lattice as a built-in ruler

Here is a lovely idea: the spacing between atomic planes is a ruler that lives inside the material, and Bragg's law reads it for free. Stretch the lattice and every d-spacing grows, so each peak slides to a lower angle; squeeze it and the peaks march to higher angle. So a peak's exact POSITION measures strain, and with the material's elastic constants that strain converts to stress — the internal, self-balancing residual stress locked into a part by machining, welding, quenching, or shot-peening, with no external load applied at all. It is completely nondestructive: you shine X-rays on the surface and let the atoms report their own strain.

The standard recipe is the sin-squared-psi method. You measure the d-spacing of one chosen reflection at several sample tilts (the angle psi between the surface normal and the diffracting plane), then plot d against sin^2(psi). For a surface under biaxial stress the plot comes out beautifully straight, and its slope is proportional to the stress: a positive slope means the surface is in tension, a negative slope means compression. Because X-rays only penetrate a few microns, this reads the near-surface stress specifically — which is exactly what matters for fatigue and coatings. Keep one distinction crisp: a uniform macrostress SHIFTS the peak, while grain-to-grain microstrain BROADENS it (the Scherrer/strain width of guide 3). Position for the average stress, width for the scatter around it.

One last frontier: thin films and coatings, where the interesting layer may be only nanometres thick and a normal scan drowns in signal from the substrate beneath. Two tricks rescue it. In grazing-incidence diffraction you hold the incoming beam at a tiny angle so it skims through the film and barely reaches the substrate, boosting the film's contribution for phase identification. And X-ray reflectivity, taken at even smaller angles, ignores the crystal structure entirely: the gentle interference fringes of the reflected beam reveal a film's thickness, density, and roughness — it even works on amorphous layers. For epitaxial films, mapping a patch of reciprocal space around a substrate reflection exposes how the film is strained or relaxed and whether misfit dislocations have crept in. When the sample is truly tiny or the effect truly faint, this is where the tunable brilliance of a synchrotron earns its keep — the same instrument, pushed to its limit.