TEM troubleshooting FAQ

No. Ratios of reciprocal-space distances and the angle between spots are calibration-free: the camera constant cancels out. The indexer uses exactly this ratio + angle matching, so you can index a pattern without knowing camera length at all. See it worked out on SAED Zone-Axis Indexer →

Yes, more than you might expect, as long as you know the structure. Computing the ratio-and-angle signature of every symmetry-unique zone axis up to index 5 shows that no two low-index axes in FCC, BCC, simple cubic or diamond share a signature, and they stay unique even at 6% in ratio and 4° in angle. What actually makes an answer ambiguous is measurement error on a high-index axis, or not knowing the structure: the same numbers fit [3 1 1] in all four of those structures exactly. There is also one pair that no amount of extra spots resolves, [4 4 1] and [5 2 2], whose nets are identical; that one needs a tilt to a second zone axis. The explorer computes all of this live for whatever precision you claim. Test your own precision on the SAED Indexing Guide →

All three can put extra spots in the same positions: a twin mirrors the parent lattice across the twin plane, double diffraction generates sum vectors from overlapping crystals, and a second phase adds its own reciprocal lattice at the orientation relationship. The guide now scores all three against a pattern whose cause is hidden, so you can watch the ambiguity happen instead of taking it on trust. Two things it makes concrete: a second-phase hypothesis carries a free lattice ratio, so judge it by the spots it predicts that are not in your pattern, not only by the ones it hits; and in diamond cubic, double diffraction and a fitted second phase routinely explain every extra spot equally well. When that happens, position has run out of information and you need a tilt, a dark-field image, or a composition. Test the three hypotheses on the SAED Indexing Guide →

Streaked spots come from shape transforms of features thinner in one direction than the others: stacking faults streak perpendicular to the fault plane, plate-shaped precipitates streak along the plate normal, and small or irregularly shaped domains broaden spots in every direction. The streak direction in the pattern tells you the real-space direction the feature is thin in. See it worked out on the SAED Indexing Guide →

A blurry SAED pattern almost always means the diffraction-focus (intermediate lens) isn't set to image the back focal plane precisely. The ray diagram on this page shows exactly which lens current controls that, and why it's a different adjustment from image focus. See it worked out on Back Focal Plane & Diffraction Focus →

A reflection only appears when the Ewald sphere passes close enough to that reciprocal-lattice point, within the excitation error. Voltage flattens the sphere, tilt swings it, and specimen thickness grows the relrods each point sits on, all three change which spots you actually see, and the live simulator on this page lets you watch it happen. See it worked out on Reciprocal Space & the Ewald Sphere →

Structure-factor selection rules forbid certain reflections for a perfect, infinite crystal, but real specimens break those assumptions in four ways: finite thickness extends relrods until the Ewald sphere clips them, double diffraction in overlapping regions creates sum vectors at forbidden positions, planar faults and surface termination introduce streaks through forbidden sites, and chemical ordering lights up superlattice spots the disordered structure forbids. The Ewald guide's tilt and thickness sliders show the relrod mechanism directly. See it worked out on Reciprocal Space & the Ewald Sphere →

s_g is the distance, in reciprocal space, from the reciprocal-lattice point g to the nearest point on the Ewald sphere, measured along the beam direction. A reflection with s_g = 0 is exactly at the Bragg condition; as |s_g| grows, the diffracted intensity falls off as sinc-squared of pi times s_g times t, modulated by specimen thickness t. The Ewald sphere guide draws s_g live as you tilt, and the weak-beam guide shows what happens when you push s_g deliberately large. See it worked out on Reciprocal Space & the Ewald Sphere →

Because the Ewald sphere is nearly flat at TEM voltages (its radius is hundreds of reciprocal-lattice spacings), its intersection with the reciprocal lattice is very sensitive to tilt: a fraction of a degree swings the sphere across many reciprocal-lattice points, lighting up an entirely different set of reflections. The Ewald guide's tilt slider shows it directly. See it worked out on Reciprocal Space & the Ewald Sphere →

It comes down to the misorientation angle and axis between the two grains' reciprocal lattices: LAGBs are typically under a few degrees, HAGBs are larger and more random, and twins sit at specific, often coherent misorientations (Σ3 in cubic metals). This page's misorientation module walks through telling them apart. See it worked out on Grain Orientations & Reciprocal Space →

Three indicators tell you: the displacement of Kikuchi lines from the pattern center, the intensity asymmetry between Friedel pairs, and the radius of the Laue circle in your SAED pattern. A large Laue circle means you are well off-zone; Kikuchi lines converging toward the center means you are tilting on-zone. This page's beam-tilt controls show all three changing together. See it worked out on Grain Orientations & Reciprocal Space → To get from the zone you are on to the one you want, the Kikuchi Map Navigator names the band to follow, the tilt angle, and the double-tilt holder settings.

Convergent-beam discs carry Kossel–Möllenstedt fringes whose spacing depends on thickness; plotting (s_i/n_i)² against (1/n_i)² and extrapolating to the intercept gives 1/t² directly (the Kelly–Allen method). This page builds that fit live from simulated fringe positions. See it worked out on CBED & Lamella Thickness →

Pick the structure (604 are built in, or load a CIF) and the zone axis, and the pattern is drawn: the spot positions come from the reciprocal lattice, the intensities from the structure factors. Switch the dynamical calculation on when the foil is thick enough for intensity to move back and forth between beams, which changes how strong the spots look but not where they sit. To check a measured pattern, drop it underneath and fit the camera constant, rotation and ellipticity against it. Simulate a pattern →

Use the kinematic pattern to decide where the spots are and which ones are forbidden. Use the dynamical one to decide how bright they are. In a thin foil (a few tens of nm) and away from a strong zone axis the two agree well. On a zone axis, or in a thick foil, the kinematic intensities are wrong: a forbidden spot can appear by double diffraction, and a strong spot can go dark at certain thicknesses. Both are on the same page, on the same crystal, so you can watch the difference with a thickness slider. Compare the two →

Record a ring pattern from a polycrystalline standard at the camera length you are using, then fit the ring radii against the standard’s known d-spacings: r = K/d, and the slope K is the camera constant in Å·px. Two things decide whether the number is any good. The pattern centre has to be refined first, because an off-centre guess widens every ring and biases K. And the calibration belongs to one camera length on one instrument on one day: it is not a property of the standard. The page also converts K to a camera length in mm from your pixel size and binning, and measures the elliptical distortion at the same time. Calibrate from a ring pattern →

Read the symmetry of three things: the whole pattern, the bright-field disc on its own, and a ±G pair of dark-field discs. Those observations name one of the 31 diffraction groups, and the diffraction group narrows the crystal’s point group. One zone axis usually leaves several point groups standing, so the determination is finished by repeating it on a second zone axis and keeping what both allow. The tables are derived from the 32 point groups on the page rather than typed in, and you can work the problem in either direction. Look up a diffraction group →

That's diffraction contrast: a grain's brightness depends on how strongly it's diffracting at the current tilt, controlled by the objective aperture position (bright-field vs dark-field) relative to that grain's excited reflections. The bright-field/dark-field tilt simulator on this page shows the same polycrystal flipping contrast as you tilt. See it worked out on The Three Contrasts of TEM →

It depends on the scattering-angle range you want: BF and ABF collect low-angle, largely coherent scattering (ABF is useful for light-element/atomic-column contrast), while ADF and especially HAADF collect higher-angle, more incoherent (Z-sensitive) scattering. This page maps camera length and atomic number onto each detector ring so you can see what each geometry actually captures. See it worked out on STEM Detectors: BF, ABF, ADF & HAADF →

HAADF is often called Z-contrast imaging, but the Z-dependence (roughly Z to the power 1.5 to 2) only holds cleanly when columns are thin enough to avoid channeling, the inner collection angle is high enough to suppress coherent diffraction contrast, and the two column types do not differ too much in thickness. Thick specimens, low camera lengths, and zone-axis channeling can all break the monotonic Z relationship. This page's detector-geometry sandbox shows how collection angles and atomic number interact. See it worked out on STEM Detectors: BF, ABF, ADF & HAADF →

The honest answer usually depends on whether the feature is elastic (structural: diffraction/phase contrast) or inelastic (chromatic aberration, energy loss) in origin, and whether it survives a tilt or aperture change. This page's artifact clinic and signal-versus-artifact verdict table work through the common cases at 80–300 kV. See it worked out on Scattering in the TEM →

Weak-beam trades signal for resolution on purpose: you're imaging far from the strong Bragg condition, so intensity is inherently low. The practical checklist on this page (exposure, drift, the g–3g tilt itself) walks through what's normal versus what's actually wrong. See it worked out on Weak-Beam Dark Field →

The imaged intensity peak is offset from the true core position by an amount that depends on the diffraction conditions, not a measurement error you can just calibrate away. This page's Howie–Whelan simulator shows the offset directly. See it worked out on Weak-Beam Dark Field →

Rings usually mean your region of interest spans multiple, randomly oriented crystallites (each contributing spots at a random azimuth) or an amorphous region. Shrinking the ROI to one crystallite should collapse the ring back into a spot pair, this page's draggable-ROI FFT demonstrates it directly. See it worked out on HRTEM & FFTs →

Moiré fringes come from two overlapping lattices beating together, and they're much coarser than either real lattice, a classic trap is mistaking overlap between a grain and its oxide for a genuine superstructure. This page's FFT guide covers the giveaway signs. See it worked out on HRTEM & FFTs →

An FFT of a real HRTEM image can produce peaks that do not correspond to real lattice planes: drift during acquisition shifts peak positions, scan distortion in STEM adds non-crystallographic peaks, an ROI that clips a crystal edge introduces truncation streaks, and a miscalibrated pixel size scales every d-spacing wrong. This page's draggable ROI with the Hann-window toggle shows how windowing alone can move or remove artifactual peaks. See it worked out on HRTEM & FFTs →

You need two reflections, not parallel to one another, in which the dislocation is invisible: b then lies along g₁ × g₂. Which two depends on the crystal and on the zone axes your holder can actually reach, and some candidate vectors cannot be told apart by invisibility alone (vectors along the same direction of different length vanish together). Work that out before the session rather than at the microscope: the planner tabulates g·b for every candidate in every reflection of the zones you name, and picks the smallest set of images that separates them. Plan the experiment →

g·b = 0 kills the contrast of a screw dislocation. An edge or mixed dislocation also displaces atoms along b × u, and that part survives: the residual image is faint but real unless |g·(b × u)| is below about 0.64, which is Howie and Whelan’s rule of thumb. Two other things produce the same symptom: elastic anisotropy, which makes the invisibility inexact in the first place, and a nearby second defect. If you can see a faint line where you expected nothing, check the edge term before rejecting the candidate. See which candidates carry a residual → or compute the image itself →

From the inside-outside contrast, and only with the signs written down. Image the loop in +g and in −g at a known excitation error s, and note whether the image lies inside or outside the true loop position. The sense flips with the sign of g, with the sign of s, and with the loop nature, so a result quoted without all three is not a result. Simulating the same loop with the signs you can control is the safest way to fix the convention before you trust an experiment. Simulate a loop →

Take the FFT of a region, measure the distance from the centre of the FFT to a spot, and convert that distance with the image calibration: a spot at r pixels in an N-pixel FFT of an image with pixel size dx is a spacing of N·dx/r. Three things set how good the number is: the pixel size (a scale bar or a known lattice), the size of the region (a small region gives broad spots), and picking the spot centre rather than its brightest pixel. Measure several orders of the same spacing and the error falls. Measure from an FFT →

One division: d = K/R, where R is the distance from the direct beam to the spot and K = λL is the camera constant. At 200 kV the wavelength is 0.02508 Å, so a nominal camera length of 1000 mm gives K = 25.08 Å·mm and a spot 12 mm out is 2.09 Å. The catch is that the camera length shown on the microscope is a lens setting, not a measurement of your session, and it moves with the specimen height and the lens history: a 5% error there lands whole on every spacing you report and never shows up as disagreement between your own numbers. Calibrate K from a ring pattern of a standard in the same session, measure across the pattern from a spot to its opposite through the centre and halve, and quote the uncertainty. See it worked out step by step →

That is the normal situation, not a failure: the spacing axis is crowded, and a 2% error bar covers several allowed reflections almost everywhere on it. Two ways out. Narrow the bar, by calibrating in the same session, measuring through the centre and averaging equivalent reflections, which typically takes you from a few percent to a few tenths of a percent. Or add a constraint, which is usually cheaper: two spacings and the angle between them test the geometry as well as the size, and that pins a zone axis rather than a single plane. Remember that a solid solution shifts lattice parameters by around half a percent for a few atomic percent of solute, so candidates closer together than that were never separable by spacing alone. See the candidates on one axis →

Normalise the image to the detector itself. Record the detector with the beam off (the dark level) and with the beam on the detector (the probe), then map every pixel to (I − dark)/(probe − dark). The image is then a fraction of the incident beam, which is the same quantity a simulation computes, so the two can be compared directly. Without that step a HAADF image is in arbitrary units and cannot be compared between sessions, let alone against a simulation. Normalise an image →

The log-ratio method measures t/λ, but converting that to nanometers needs the inelastic mean free path λ, and the Malis, Iakoubovskii, and Kramers-Kronig models disagree on its value. For the same collection angle and voltage, the models can differ by 10 to 20%, and that spread is the honest uncertainty floor on any log-ratio thickness. See it worked out on the EELS Thickness Calculator →

The log-ratio λ depends on the collection semi-angle β through a correction factor, so the reported thickness shifts when you swap the spectrometer entrance aperture. The EELS thickness calculator lets you sweep β and see the sensitivity directly; it also warns when β is large enough that the correction becomes unreliable. See it worked out on the EELS Thickness Calculator →

Three steps, and each one can cost you a factor. Fit the background in a window before the edge (a power law) and subtract it. Integrate the edge over a window of your choosing. Divide by the partial ionisation cross-section computed for that same window and your collection angle, not by a tabulated total. The result is atoms per unit area, and a ratio of two of them is a composition. If the foil is not thin, plural scattering has to be removed first or every count is spread out of the window and the answer comes out low. Quantify an edge →

Cliff-Lorimer: the ratio of two elements is the ratio of their peak intensities times a k-factor. The k-factor depends on the elements, the detector and the voltage, not on your specimen, which is why the method works without standards. What it assumes is that the foil is thin enough for absorption to be negligible, and that assumption fails quietly: as the foil thickens the softer line is absorbed on its way out and the composition drifts. The correction needs the thickness and the density, which is why the zeta-factor method, which gets both from the same spectrum, is worth the extra care. Work through a quantification →