Ceramics — alumina, zirconia, silicon carbide, silicon nitride; properties at high T: Visual walkthrough

This is the picture-first companion to the parent note. We lean on Thermal Stress and Fracture Mechanics, Crystal Bonding — Ionic vs Covalent, and finish pointing at Thermal Barrier Coatings on Superalloys.

Step 1 — Two atoms on a spring: the bond well

WHAT. Zoom all the way in. Any two neighbouring atoms in a crystal sit at a comfortable distance apart. Push them closer and they shove back; pull them apart and they tug back. That "comfortable distance" is a valley in an energy landscape.

WHY. Everything about heat expansion, stiffness, and cracking starts here. If we understand how one bond stores energy, the whole plate follows — a plate is just trillions of these bonds side by side.

PICTURE (Figure 1, s01). In Figure 1, $x$ is how far an atom sits from its happy spot ( $x = 0$ ). The height of the curve $U (x)$ is stored energy. The bottom of the valley is where the atom rests when cold. Notice the red curve is steeper on the left (squeeze) than the right (stretch) — that lopsidedness is the whole engine of expansion.

Figure — Ceramics — alumina, zirconia, silicon carbide, silicon nitride; properties at high T

Figure 1 — The bond-well $U (x)$ : a lopsided valley, steeper on the squeeze side.

Step 2 — Heat wobbles the atom, and the lopsided well pushes it outward

WHAT. Temperature is jiggling. A hot atom doesn't sit still at $x = 0$ ; it rattles back and forth, spending time at both positive and negative $x$ . The key realisation: because the valley is lopsided (easier to stretch than to squeeze), the atom's average position drifts to positive $x$ . The bond gets longer on average. That is thermal expansion.

WHY. This is the whole reason materials grow when heated. If the well were a perfect symmetric bowl, the average position would stay at $x = 0$ and nothing would expand. Expansion exists only because of the $- g x^{3}$ lopsidedness.

PICTURE (Figure 2, s02). Figure 2 shows two horizontal "energy levels": at low $T$ the atom rattles in a narrow band near the bottom — its average stays near $0$ . At high $T$ it explores wider, and the gentle stretch-side wall lets it lean right. The red dot marks the average position $⟨ x ⟩$ climbing outward as $T$ rises.

Figure 2 — As $T$ rises the atom samples wider, and its average $⟨ x ⟩$ (red dot) drifts right.

Intuition WHERE the coefficient comes from (the Boltzmann average)

Statistical physics says the atom spends time at each displacement $x$ in proportion to the Boltzmann weight $e^{- U (x) / k_{B} T}$ — low-energy spots are visited more, high-energy spots less. So the average position is a weighted average: $⟨ x ⟩ = \frac{\int x e ^{- U (x) / k_{B} T} d x}{\int e ^{- U (x) / k_{B} T} d x} .$

top integral ::: adds up every position $x$ weighted by how often the atom is there.
bottom integral ::: the normaliser, so the weights sum to 1 (it's a proper average).

Now the trick: the jiggle is small, so treat the lopsided term as a tiny nudge. Split $U = \frac{1}{2} c x^{2} - g x^{3}$ and expand the exponential of the small cubic part: $e^{- U / k_{B} T} \approx e^{- c x^{2} /2 k_{B} T} (1 + \frac{g x ^{3}}{k _{B} T}) .$ The first factor is a symmetric bell curve (Gaussian) of variance $⟨ x^{2} ⟩ = k_{B} T / c$ ; the $(1 + \frac{g x ^{3}}{k _{B} T})$ is the small correction. Using the standard Gaussian moments $⟨ x^{2} ⟩ = \frac{k _{B} T}{c}$ and $⟨ x^{4} ⟩ = 3 (\frac{k _{B} T}{c})^{2}$ , the numerator picks up $\frac{g}{k _{B} T} ⟨ x^{4} ⟩$ and the denominator stays $\approx 1$ : $⟨ x ⟩ \approx \frac{g}{k _{B} T} ⟨ x^{4} ⟩ = \frac{g}{k _{B} T} \cdot 3 (\frac{k _{B} T}{c})^{2} = \frac{3 g k _{B} T}{c ^{2}} .$ The symmetric part gives $⟨ x ⟩ = 0$ ; only the tiny cubic correction survives.

Figure 3 — The Boltzmann weight $e^{- U / k_{B} T}$ : fatter on the stretch side, so its average (black line) sits right of $x = 0$ .

Step 3 — From one bond to a whole plate: the expansion coefficient $α$

WHAT. A plate is billions of these bonds in a row. If each bond of rest-length $x_{0}$ stretches its average by $⟨ x ⟩$ , the whole plate stretches by the same fraction. We give that fraction-per-degree a name: $α$ .

WHY. We need one material number that says "how much does this stuff grow per degree of heating?" — so a designer can predict size changes without tracking every atom.

PICTURE (Figure 4, s03). Figure 4 shows a row of atoms: cold and tight on top; heated and slightly spread on the bottom. The tiny gap opened per bond, summed over the whole length, is the plate's growth $α L_{0} Δ T$ .

Figure 4 — Each bond's average stretch, summed along the row, is the plate's thermal growth.

Step 4 — Heat one face fast: the hot skin wants to grow

WHAT. Now blast one face with a torch. That skin heats by $Δ T$ and, left alone, would grow by a strain $ε = α Δ T$ . But the cold interior beneath it has not heated yet and refuses to grow with it.

WHY. This clash — hot skin wanting to expand, cold core holding it back — is the birth of thermal stress. No temperature difference, no stress. (Heat the whole thing uniformly and it just grows freely, no drama.)

PICTURE (Figure 5, s04). Figure 5 shows two layers: a red hot skin trying to lengthen (outward red arrows), a black cold core clamping it (inward black arrows). The skin is stuck at the cold length, so it's squeezed — under compression.

Figure 5 — Hot skin wants to grow (red arrows) but the cold core clamps it (black arrows) → skin in compression.

Step 5 — Blocked stretch becomes stress: Hooke's law and where $1/ (1 - ν)$ is born

WHAT. A blocked strain doesn't vanish — it turns into force per area, i.e. stress. The skin is a flat surface, so it is blocked in two in-plane directions at once (call them $1$ and $2$ ), not just one. Working that biaxial Hooke's law through is exactly where the factor $1/ (1 - ν)$ appears.

WHY. We must convert the "wanted-but-blocked" expansion into an actual mechanical stress, because cracks respond to stress, not to temperature — and we must get the geometric factor right, not just wave at it.

PICTURE (Figure 6, s05). Figure 6 shows a little square patch of the skin. It wants to grow the same amount in both in-plane directions but is held on all four edges, so equal stresses $σ$ push inward on both pairs of edges. Nothing pushes on the free front face — that is the plane-stress assumption stated below.

Figure 6 — A skin patch held on all four in-plane edges, front face free ( $σ_{3} = 0$ ): the biaxial stress state.

Formula Plane stress, and deriving the biaxial factor

Which 3D state? The heated skin is a thin surface layer with a free outer face. Nothing presses on that face, so the through-thickness stress is zero: $σ_{3} = 0$ . This is the plane-stress condition (as opposed to plane-strain, $ε_{3} = 0$ , which would apply to a fully buried layer). We adopt plane stress because the crack starts at or just below a free surface. With $σ_{3} = 0$ , Hooke's law in-plane relates the strain in direction $1$ to both in-plane stresses (each also squeezes sideways via $ν$ ), plus the free thermal growth $α Δ T$ : $ε_{1} = \frac{1}{E} (σ_{1} - ν σ_{2}) + α Δ T .$

$σ_{1} / E$ ::: direct stretch from the stress along direction 1.
$- ν σ_{2} / E$ ::: the sideways effect — a stress in direction 2 pinches direction 1 by a fraction $ν$ .
$α Δ T$ ::: the free thermal growth the skin wants.

The skin is fully restrained in-plane, so $ε_{1} = ε_{2} = 0$ , and by symmetry $σ_{1} = σ_{2} \equiv σ$ . Substitute: $0 = \frac{1}{E} (σ - ν σ) + α Δ T ⟹ σ \frac{1 - ν}{E} = - α Δ T .$ Solving gives the magnitude $σ_{t h} = \frac{E α Δ T}{1 - ν}$

the $(1 - ν)$ ::: it is not a fudge — it is literally the $(σ - ν σ)$ that popped out of the two-direction Hooke's law under plane stress. A lone rod (blocked in one direction only) would have no $ν$ and give plain $E α Δ T$ ; the plate is stiffer because each direction helps hold the other.
Sign: the minus sign means the hot skin stress is compressive — squeezed, as Figure 5 showed.

Step 6 — Which side actually cracks: compression on top, tension below

WHAT. Here is the twist most people miss. The hot skin is in compression — and ceramics are enormously strong in compression, so the skin itself does not crack. But a plate cannot push its skin inward without the layer just beneath being pulled the opposite way: the cooler sub-surface is stretched into tension. Ceramics are weak in tension, so the crack is born there, on the tensile side, and runs.

WHY. The final criterion uses the tensile fracture strength $σ_{f}$ precisely because that is where failure starts. If we compared the compressive skin stress to a tensile strength we'd be mixing up two different things. The magnitudes balance (the plate has no net force), so the peak tensile stress beneath equals the thermal stress magnitude $σ_{t h}$ — which is why the same $σ_{t h}$ enters the test.

PICTURE (Figure 7, s06). Figure 7 is a cross-section: red hot skin squeezed (compression arrows pointing in), the layer beneath stretched (red tension arrows pointing out), and a crack sketched opening on the tensile side.

Figure 7 — Compression on the hot skin (safe), tension just beneath (weak) → crack starts on the tensile side.

Step 7 — Cracking condition: tensile stress meets tensile strength

WHAT. The tensile layer carries stress equal in magnitude to $σ_{t h}$ . It survives while $σ_{t h} < σ_{f}$ and cracks the instant $σ_{t h}$ reaches the tensile strength $σ_{f}$ .

WHY. This is the decision line — the whole point is to find how big $Δ T$ can get before we hit it.

PICTURE (Figure 8, s07). Figure 8 plots two things against $Δ T$ : the growing tensile stress (red line, rising with $Δ T$ ) and the fixed tensile-strength ceiling (black dashed line). The moment the red line touches the ceiling, a crack forms.

Figure 8 — Tensile stress climbs with $Δ T$ ; it cracks where the red line hits the strength ceiling $σ_{f}$ .

Step 8 — Solve for the survivable jump: the number $R$

WHAT. Set the generated stress equal to the tensile strength and solve for $Δ T$ . That critical jump is the thermal-shock figure of merit $R$ — the biggest temperature jump the material can eat without cracking.

WHY. This single number lets an engineer rank ceramics for shock duty with one glance. It is the parent note's central result.

PICTURE (Figure 9, s08). Figure 9 shows the algebra as a balance: take Step 5's stress equation, replace $σ_{t h}$ with the strength $σ_{f}$ , and rearrange until $Δ T$ stands alone.

Figure 9 — Setting $σ_{t h} = σ_{f}$ and solving for $Δ T_{ma x}$ gives the figure of merit $R$ .

WHAT. Check what happens at the limits — the scenarios a careless reader would trip over.

WHY. A formula you can't stress-test at its extremes is a formula you don't understand.

PICTURE (Figure 10, s09). Figure 10 has three mini-panels: (a) $α \to 0$ , (b) $Δ T = 0$ , (c) slow vs fast heating.

Figure 10 — Three limits: zero expansion ( $R \to \infty$ ), uniform heat ( $σ_{t h} = 0$ ), and slow vs fast heating.

Definition The conductivity factor — done dimensionally honestly

For a severe, fast shock, how much true $Δ T$ actually builds up depends on how quickly heat conducts inward. Define

$k$ ::: the thermal conductivity (units W·m⁻¹·K⁻¹) — how fast the material carries heat from the hot skin into the cold core. High $k$ = the interior warms with the skin, so the real temperature difference stays small.

Because $R$ (in kelvin) and $k$ (in W·m⁻¹·K⁻¹) have different units, you may not simply multiply them and still call it a temperature. The honest severe-shock figure of merit is a different, named quantity: $R^{'} = R \cdot k = \frac{σ _{f} ( 1 - ν ) k}{E α} [units: W\cdotm^{- 1}] .$

$R^{'}$ ::: not a temperature — it is a power-per-length rating for resisting fast shock. A material scores high by being strong ( $σ_{f}$ ), floppy ( $E$ small), low-expansion ( $α$ small), and conductive ( $k$ large, so it evens out its own temperature). This is why conductive SiC ( $k \approx 120$ ) beats insulating zirconia at sudden thermal shock even though zirconia is the better steady insulator.

The one-picture summary

Figure 11 — The whole chain compressed into one frame.

The whole chain in one frame: lopsided bond well → atom drifts outward with heat ( $α$ ) → hot skin blocked by cold core → skin compressed, layer beneath in tension → tensile stress $σ_{t h}$ hits tensile strength $σ_{f}$ → survivable jump $R$ . Read Figure 11 left to right and you've re-derived the flagship formula.

Recall Feynman retelling — say it back in plain words

Two atoms sit in a valley. The valley is lopsided — easier to pull them apart than to push them together (and it has a quartic floor so it can't dive to minus infinity; we only ever work near the bottom where wobbles are small, $k_{B} T ≪ c^{3} / g^{2}$ ). So when heat makes them jiggle, they spend more time (Boltzmann says: more time in low-energy, easy-stretch spots) on the stretched side, and on average the bond gets longer. Do the weighted-average integral and the lean-out works out to $3 g k_{B} T / c^{2}$ . That's why things expand, and we call the growth-per-degree $α$ . Ceramics have deep stiff valleys, so their $α$ is tiny.

Now torch one face. The hot skin wants to grow by $α Δ T$ , but the cold guts underneath won't let it, so the skin gets squeezed — compression. Because the skin is a thin free-faced surface blocked in two directions (plane stress, $σ_{3} = 0$ ), working Hooke's law through both directions gives $σ_{t h} = E α Δ T / (1 - ν)$ — the $(1 - ν)$ is literally the two-direction bookkeeping, not a fudge. But the skin doesn't crack: ceramics love compression. The layer just beneath gets pulled into tension with the same magnitude, and ceramics are weak in tension — so the crack starts there. It survives until that tensile stress reaches the tensile breaking stress $σ_{f}$ . Set them equal, solve for the jump, and you get the magic number $R = σ_{f} (1 - ν) / E α$ : strong and low-expansion and floppy is what survives shock — not just "high melting point." And if the shock is very sudden, high conductivity $k$ helps too, giving the separate power-rating $R^{'} = R k$ .

Recall Quick self-test

Which term must be small in $R$ for good shock resistance, and why? ::: $α$ (and $E$ ) must be small — a low expansion coefficient means the blocked strain $α Δ T$ is tiny, so little stress builds. $R \propto 1/ α$ . Why does uniform heating cause no thermal stress? ::: Every bond grows equally, so no region fights any other — thermal stress needs a temperature difference, giving $Δ T = 0 \Rightarrow σ_{t h} = 0$ . On which side of the plate does the crack actually start, and why? ::: On the tensile side — the sub-surface layer beneath the compressed hot skin is stretched, and ceramics are weak in tension, so the crack initiates there and $σ_{f}$ is the tensile strength. Where does the $1/ (1 - ν)$ come from? ::: From applying Hooke's law in two in-plane directions under plane stress ( $ε_{1} = \frac{1}{E} (σ_{1} - ν σ_{2}) + α Δ T = 0$ with $σ_{1} = σ_{2}$ ); the $(σ - ν σ)$ gives the $(1 - ν)$ . What breaks the small-expansion formula, and at what temperature? ::: The neglected quartic floor $+ h x^{4}$ ; the linear-in- $T$ result is only valid while $k_{B} T ≪ c^{3} / g^{2}$ . Does $R$ depend on melting point? ::: No — $R = σ_{f} (1 - ν) / E α$ contains none of $T_{m}$ .

Visual walkthrough — Ceramics — alumina, zirconia, silicon carbide, silicon nitride; properties at high T