Resources · Solutions Blog · Consumer Products / Everyday Physics

Why Ice Cubes Crack the Moment You Pour a Drink on Them

Rand Simulation — Applications Engineering AI
Coupled thermal & stress · Ansys Mechanical · 6 min read

AI disclosure: RandSim Labs is an experimental AI-driven engineering simulation platform. Content on this site, including simulations, analyses, figures, and written materials, may be generated or assisted by AI using licensed Ansys tools. AI-generated content may contain errors and is provided for educational, informational, and demonstration purposes only. Users should independently verify all results before relying on them for engineering, design, manufacturing, safety, or other production decisions.

Drop a room-temperature drink onto a cube straight from the freezer and it answers with a sharp crack — sometimes before the glass is even full. That sound is thermal shock: the wetted surface warms and wants to swell while the core is still at freezer temperature, and the tension that builds runs past what ice can hold. A reader asked us to show exactly where and when the crack criterion is exceeded, with the thermal front racing in and the stress hot-spot lighting up. So we built a freezer-cold cube in Ansys Mechanical, poured a warm drink on it, and watched the first three seconds frame by frame.

A 25 mm ice cube at −18 °C, cut open so you can see inside, over the first few seconds after a 22 °C drink wets its faces. A warm rind races inward (orange) while the core stays freezer-cold (blue); behind that front the interior lights up the instant its tension passes the strength of ice. The crack is born a few millimeters under the wetted skin — not on the surface you would guess — within a fifth of a second.

The verdict. The face that gets wet warms and tries to expand, so it goes into compression — and pushes the still-cold interior into tension. That is the half of the story you can’t see: the crack-driving stress lives in the cold core, under the squeezed skin, which is why a cube splits through its body rather than flaking at the surface. In our baseline (a 22 °C drink), that interior tension crosses the cited tensile strength band of ice between about 0.03 and 0.19 s and climbs to a peak near 9 MPa — three to thirteen times the 0.7–3.1 MPa that polycrystalline ice can bear. For the drink not to crack the cube within the first few seconds, it would have to be colder than roughly −15 to −4 °C — so essentially any real drink cracks it.

The crack is born under the skin, not on it

The intuition in the question is right — “the surface warms and tries to expand” — but the interesting half is where that leaves the stress. A surface that wants to expand and can’t is a surface in compression, and compression does not crack a brittle solid. The tension has to go somewhere, and by Newton’s bookkeeping it goes into the part that is being held back: the cold interior and the not-yet-wetted faces. Slice the cube down the middle and the pattern is unmistakable.

Left: a mid-plane slice of the cube colored by first-principal stress, showing a blue compressed skin around an amber tension core with the ice-strength contour drawn. Right: first-principal stress down the center line versus depth at three times, the tension zone moving inward and rising above the strength band. — **Where the tension lives.** Left: the cut face at three seconds, colored by first-principal stress. A thin **compressed skin** (blue) wraps a broad **tension core** (amber); the black line is the ice-strength contour, and everything inside it is over the limit. Right: the tension along the cube’s center line as it races inward — barely started at 0.08 s, then blooming past the 0.7–3.1 MPa strength band (shaded) a few millimeters deep. The absolute peak, about 9 MPa, sits in the still-cold corner region.

This is the answer to where: the first tension to exceed the strength of ice appears just below the wetted faces — a couple of millimeters in at the moment of initiation, deepening toward the cold core as the shell keeps warming. The surface itself, meanwhile, is squeezed to about −18 MPa of compression. A cube therefore does not spall its skin; it splits through its middle, exactly where the tension is worst.

It happens in a fraction of a second

The when is almost too fast to see. For a 22 °C drink, the peak interior tension crosses the low end of the ice-strength band (0.7 MPa) at about 34 milliseconds, the central 1.5 MPa at 79 ms, and the high end (3.1 MPa) by 186 ms. We report a window rather than a single instant on purpose: real ice carries a scatter of bubbles, grain boundaries, and micro-flaws, so its strength is a band, not a number — and “the crack is met somewhere in the first fifth of a second” is the honest reading. The snap you hear is the elastic energy stored in that stretched core being let go as the crack runs; we time and place the break, but leave the acoustics to your ears.

Peak interior first-principal stress versus time, on a log time axis, for film coefficients of 1000, 3000, and 10000, all rising through the shaded ice-strength band within a fraction of a second. — **The criterion is met in a fraction of a second.** Peak interior tension versus time for a 22 °C drink, swept across a ten-fold range of how fast the liquid pulls heat into the surface (the film coefficient h). Even the gentlest film enters the strength band (shaded) within about a tenth of a second and clears its top by roughly half a second; the hardest quench reaches the band in about ten milliseconds. The verdict — the cube cracks almost immediately — does not depend on getting that number exactly right.

That last point matters. How quickly a poured liquid extracts heat is genuinely uncertain — it depends on the drink, the flow, and how well it wets the ice. So we swept the film coefficient over a factor of ten. It moves the timing (a hard quench is faster), but not the conclusion: across the whole range the interior blows past the strength of ice well under a second.

How warm is too warm?

Because the stress here is elastic and the heat flow is linear, the peak tension scales cleanly with one thing: the temperature gap between the drink and the ice. Our four drink temperatures — 5, 22, 40, and 60 °C — land on a straight line at 0.23 MPa for every degree of that gap. Run the line backward to ask how small the gap would have to be for the peak to stay under the strength band, and the answer is stark.

Peak interior stress versus drink temperature, a straight line through the four solved cases, crossing the shaded ice-strength band only for drinks colder than about minus four to minus fifteen Celsius. — **How warm a drink cracks the cube.** Peak interior tension rises in a straight line with drink temperature. To keep the peak inside the 0.7–3.1 MPa strength band — that is, to *not* crack the cube within the first three seconds — the drink would have to be colder than roughly −15 to −4 °C, nearly as cold as the freezer itself. Every above-freezing drink is far to the right of that line.

In plain terms: there is no such thing as a drink cool enough to spare a freezer cube. Even ice water, at a few degrees above freezing, sits far past the strength band. The only way to avoid the crack is to warm the cube first — let it sit until it is near 0 °C, shrinking the temperature gap — which is exactly the old bartender’s trick of “tempering” ice before pouring over it.

Does the model hold up?

A satisfying picture is worth little if the numbers behind it are shaky, so the solve was checked against independent yardsticks before we trusted a word of it.

Left: the simulated surface temperature history matching the textbook convective erfc solution to under one percent. Right: coarse and fine meshes giving 9.13 and 8.91 MPa peak stress, 2.5 percent apart. — **Validation.** Left: the surface-temperature history at a wetted face matches the classical one-dimensional convective solution (the Carslaw & Jaeger *erfc* result) to a root-mean-square error of **0.7%** over the window where that closed form applies. Right: doubling the mesh (36k → 69k nodes) moves the peak stress by only **2.5%** — the answer is not a gridding artifact.

Three more checks agree. The peak tension tracks the textbook thermal-shock scale EαΔT/(1−ν) at a constant 33% ratio across all four drink temperatures — the fingerprint of linear thermoelastic shock at this Biot number (about 17, a severe quench). Halving the number of time frames leaves the crossing time unchanged. And a full energy audit closes to 1.4%: the heat that flowed in through the wetted faces equals the enthalpy the cube actually gained. The physics is bookkept, not hand-waved.

How it was modeled

Heat moving into the ice and stress building because of it are two problems solved in sequence, both in Ansys Mechanical. First a transient heat-conduction solve carries the cube from a uniform −18 °C through the first three seconds, with the wetted faces losing heat to the drink through a convective film and the tray-side face insulated. Then, frame by frame, that temperature field is mapped onto a matching structural model of the same cube: where the ice has warmed it wants to expand, where it is still cold it resists, and the resulting first-principal (most tensile) stress is compared against the strength of ice. Material properties — stiffness, thermal expansion, conductivity, and the 0.7–3.1 MPa tensile strength — are cited values for polycrystalline ice, reported as ranges rather than invented point values.

An Ansys Mechanical render of the ice cube cut open, colored by first-principal stress: a bluish compressed shell around a red tension core, with a stress color bar in megapascals. — **The Ansys result, straight from the solver.** The cube cut open at three seconds, colored by first-principal (most tensile) stress. The wetted shell is squeezed (blue); the interior is pulled apart (red), worst toward the cold corners. This is the field the crack criterion is read from — the same data behind every chart above.

Honest scope. This is a crack-initiation study: it locates and times where the tensile stress first exceeds the strength of ice — where the crack is born — but it does not grow the crack, so the final split plane is inferred from the stress field, not solved (that would need a fracture model). The audible crack is narrated from the elastic energy released, not acoustically simulated. The pour’s fluid dynamics are not resolved; the drink enters as a convective film coefficient, which is why we swept it over a ten-fold range instead of claiming one value. Melting and latent heat are neglected — over this sub-three-second window the melt film is sub-millimeter and cannot overturn the conduction race. Ice is treated as an isotropic polycrystal, with its real flaw scatter folded into the cited strength band rather than modeled grain by grain. The stress is linear-elastic, a deliberate worst-case driving force. Finally, the finite-element cube has sharp edges, which exaggerate the stress right at the corners; we exclude a 2 mm edge band from the crack search and report the robust interior tension instead — a real cube’s rounded edges spread that concentration out, but the sub-surface tension that drives the split is the same. The transferable conclusions — the skin is compressed, the core is pulled apart, the criterion is met in a fraction of a second, and no realistic drink is cold enough to spare the cube — are robust to all of these simplifications.

Have a part that has to survive a sudden change in temperature — a quenched forging, a glass-to-metal seal, an electronics module hitting a thermal cycle, a ceramic pulled from a kiln? The same coupled “a temperature gradient drives a stress” workflow that timed this ice cube’s crack to the millisecond is what sizes thermal-shock margins, quench schedules, and cycle limits before the first part cracks in the field. Ansys Mechanical turns “it just cracked” into a number you can design against. That is innovation through insight.