Thermal Expansion of Solids and Liquids

Walk out of Howrah Station on a June afternoon and stand at the end of a platform. Look down at the tracks. You are looking at two parallel steel rails, each one a long I-beam ten or twelve metres long, bolted to wooden sleepers, with a noticeable gap between the end of one rail and the beginning of the next. Roughly a centimetre of empty air. Now come back in January, when the Kolkata afternoon is twenty degrees cooler — you will see the same rails but the gaps are wider. This is not carelessness on the part of the engineer who laid the track. It is physics, and the physics is what this article is about.

Every solid and every liquid you have ever touched gets bigger when you warm it. A brass utensil expands infinitesimally the first time you drop it into hot dal. A glass bottle of lemonade swells — fractionally — when you leave it in the summer sun. The iron girders of the Howrah Bridge are several centimetres longer on a June afternoon than they are on a January dawn. The mercury in a clinical thermometer rises because its volume grows with temperature, and the glass grows too — just less, which is why you can read a temperature off it at all. The expansion is almost always too small to notice on a single object: you put a warm vessel down on the counter and cannot tell it is larger. But pile enough of it up — a kilometre of rail, a five-hundred-metre steel bridge, a steam boiler running all day — and the millimetres become centimetres and the centimetres become problems that bring down structures if no engineer has planned for them.

This article takes the phenomenon apart. You should have read temperature and thermometers first — we use \Delta T throughout, and the distinction between Celsius and Kelvin matters for one sign question. You will also benefit from having seen Young's modulus for the thermal-stress section. We derive the three expansion laws from a single molecular picture, prove that the area and volume coefficients are simply 2\alpha and 3\alpha rather than being independent numbers you have to look up, work through bimetallic strips (the physics behind every old electric iron and every mechanical thermostat), and close with the startling anomaly of water — a single material whose ecological importance stems from breaking the rule.

The molecular picture — why things expand at all

Before you write any formula, you should know what is actually happening inside the material. A solid is a lattice of atoms held in place by interatomic bonds, each bond behaving — for small displacements — like a tiny spring. At absolute zero each atom sits at the minimum of its bond's potential energy curve. Give the lattice some thermal energy and each atom oscillates about that minimum.

If the potential were a perfect parabola U(r) = \tfrac{1}{2} k (r - r_0)^2 — an ideal spring — then the atom would swing equally far to the left and right of r_0, and the average separation would still be r_0. There would be no expansion. Heat the crystal all you like; the time-averaged bond length would not budge.

But real interatomic potentials are not symmetric. They have a short-range repulsion that rises very steeply and a long-range attraction that falls off gradually. The curve is steeper on the left (where atoms are crowding) than on the right (where they are separating). When an atom oscillates in this asymmetric well, it spends more time on the shallower — wider — side. Its time-averaged position drifts to the right as amplitude grows. That drift is thermal expansion.

Why: the asymmetry is not a detail — it is the whole cause of expansion. A purely harmonic solid would not expand. Real atoms are bound by potentials that rise sharply when you push them together (because inner electrons start to overlap) and fall off gently when you pull them apart; the gentle side dominates the time average and that is the drift you observe as \Delta L.

With this picture in mind, everything else becomes algebra. The fractional increase in bond length for a small temperature rise is small and — to first order — proportional to \Delta T. Call the proportionality constant \alpha. Then the bond length at temperature T_0 + \Delta T is r_0(1 + \alpha \Delta T). A macroscopic rod of length L_0 is a chain of roughly L_0/r_0 such bonds, so its length at the new temperature is

Every macroscopic expansion formula below is an echo of that single molecular fact.

Linear expansion — the workhorse formula

Start with a rod of length L_0 at temperature T_0. Raise the temperature to T_0 + \Delta T. The new length is L, and the change is \Delta L = L - L_0. Empirically, for most solids and for small \Delta T (up to a few hundred kelvins), \Delta L is directly proportional to both the original length and the temperature change:

The proportionality constant \alpha is the coefficient of linear expansion. Its units are \text{K}^{-1} or (\,^\circ\text{C})^{-1} — these are identical because one kelvin is exactly one degree Celsius of change; only the zero differs.

Why: \Delta T is a difference of temperatures. The size of one degree is the same on both scales. Only the zero is shifted (0{ }^\circ\text{C} = 273.15{ }K). So \Delta T_{\text{K}} = \Delta T_{{}^\circ\text{C}} always, and \alpha is numerically the same whether you quote it per kelvin or per degree Celsius. This is not true of absolute temperatures — 300 K is not 300\,^\circ\text{C}.

A few values worth committing to memory:

These numbers span a factor of roughly 25. Invar — an iron-nickel alloy — has been tuned so that its magnetostriction almost exactly cancels its thermal expansion near room temperature; it is the metal of choice for precision pendulum clocks and surveying rods. At the other end, aluminium aircraft panels grow and shrink visibly with altitude.

Worked estimate: the Howrah Bridge rail gap

A single steel rail in the Kolkata section of Indian Railways is 13 m long (the standard 1950s rail length that many stretches still use). In January the rail sits at around 10°C at dawn; in a May afternoon the steel temperature reaches 55°C in full sun. Then \Delta T = 45 K. With \alpha_{\text{steel}} = 11 \times 10^{-6}\,\text{K}^{-1},

If you lay the rails end to end in winter and leave no gap, come summer each rail needs 6 mm of extra room. If it cannot find that room, it will find it anyway — by buckling sideways, derailing the train on top of it. So rails are laid with a gap of around 10 mm in winter, chosen so that the gap is nearly closed (but never negative) at the hottest point of the year. The same logic fixes the expansion joints on every long bridge — the Howrah Bridge's main span includes toothed expansion joints that swallow tens of centimetres of summer growth. (Modern long-welded rail uses a different strategy, pre-stressing the steel so the whole rail is compressed and doesn't buckle; but for ordinary jointed track the gap is the visible engineering choice.)

Areal and volumetric expansion — and why \beta = 2\alpha, \gamma = 3\alpha

If a rod grows by \Delta L when heated, what happens to a plate? A plate is a rod in two directions. Both directions expand by the same fractional amount \alpha\Delta T, so if the plate starts as a rectangle of sides L_0 and W_0, after heating its area is

Expand the square:

Why: (1 + x)^2 = 1 + 2x + x^2, applied with x = \alpha\Delta T.

Now \alpha \approx 10^{-5}\,\text{K}^{-1} and \Delta T might be 100 K, so \alpha\Delta T \sim 10^{-3}. The square of that is 10^{-6} — a millionth. Drop it:

Why: to first order, the two-dimensional fractional expansion is exactly twice the one-dimensional one. We have derived the relation \beta = 2\alpha; we did not have to measure a separate area coefficient. The only assumption was that \alpha\Delta T \ll 1 — true for every engineering situation you will meet.

For a cube of side L_0, the volume after heating is

Expand:

Both correction terms are negligible. So

We write \gamma = 3\alpha for the volumetric coefficient. Same argument: to first order, \gamma is not a separate measurement — it is determined once \alpha is known. This is a quietly deep fact. It says that materials do not have three independent thermal-expansion constants; the shape of a heated object is isotropically scaled (for an isotropic material), and a single number suffices.

What changes for anisotropic materials? Wood, crystalline quartz, and graphite expand differently along different axes. A plank of teak swells much more across the grain than along it. In that case you have three linear coefficients \alpha_x, \alpha_y, \alpha_z, and the volumetric coefficient is \gamma = \alpha_x + \alpha_y + \alpha_z. For metals and glasses, however, the lattice is effectively isotropic and one \alpha is all you need.

Why a hole expands too — and at the same rate

Here is a question that trips almost every student the first time they meet it. You have a steel plate with a circular hole cut into it. You heat the plate. Does the hole get bigger or smaller?

Intuition pulls both ways. Doesn't the metal expand into the hole, closing it up? Or is the hole outlined by the surrounding metal, which also expands outward?

The answer is clean: the hole expands by exactly the same fractional amount as if it were filled with metal. If the hole has radius r_0, after heating its radius is r = r_0(1 + \alpha\Delta T).

Why: imagine filling the hole with a metal plug of the same material, so the plate is now a solid disk. Heat the disk. Every point in the disk scales outward from the centre by the factor (1 + \alpha\Delta T) — that is the molecular picture: every interatomic distance grows by that factor. Now imagine the plug simply coloured differently so you can pick it out. The plug is a perfect disk of radius r_0(1 + \alpha\Delta T); the surrounding plate has a perfect hole of that same radius. Remove the plug — nothing about the hole's geometry cares. So the hole's radius is r_0(1 + \alpha\Delta T), and the hole has grown.

This is not a trick or a paradox; it is the direct consequence of uniform scaling. Every metal ring, every rivet hole, every washer's inner diameter grows on heating. It is precisely the principle used to fit a steel tyre onto a wooden cartwheel in the old rural workshops of India: heat the steel ring red-hot so it expands enough to slide over the wheel, then cool it with water. As it contracts, it grips the wheel more tightly than any bolt could.

Expansion of liquids — and the apparent-versus-real trap

Liquids do not hold a shape, so "linear expansion of a liquid" means nothing. You quote only the volumetric coefficient, usually written \gamma_\ell for the liquid and \gamma_g for the glass vessel holding it. Typical values:

• Water at 20°C: \gamma_\ell \approx 207 \times 10^{-6}\,\text{K}^{-1}
• Mercury: \gamma_\ell \approx 182 \times 10^{-6}\,\text{K}^{-1}
• Ethanol: \gamma_\ell \approx 1100 \times 10^{-6}\,\text{K}^{-1}
• Coconut oil: \gamma_\ell \approx 740 \times 10^{-6}\,\text{K}^{-1}

Liquids expand about ten times as much as solids per degree, which is why thermometers use a liquid (mercury, coloured alcohol, or — in modern instruments — a thermocouple plus circuitry).

Here is the subtlety. When you measure the expansion of a liquid by watching its level rise in a glass capillary, you are not measuring the real expansion of the liquid — you are measuring the liquid's expansion relative to the glass vessel. The glass expands too; the inside of the vessel gets bigger, so some of the liquid's expansion is "swallowed" by the vessel's expansion.

The apparent coefficient of expansion — what you read off the rising column — is

Why: if the glass vessel's internal volume grows by \gamma_g V_0 \Delta T, that growth is room the liquid does not have to fill. Only the extra expansion of the liquid beyond the glass's is visible as a rising column. Subtract.

For mercury in ordinary glass: \gamma_{\text{Hg}} = 182 \times 10^{-6}, \gamma_{\text{glass}} \approx 3\alpha_{\text{glass}} = 27 \times 10^{-6}, so \gamma_{\text{apparent}} = 155 \times 10^{-6}\,\text{K}^{-1}. About 85% of the real mercury expansion shows up as a rising column; 15% is hidden inside the vessel's own growth. For a precision thermometer you need to know both numbers.

Thermal stress — what happens if you forbid the expansion

The expansion laws above assume the material is free to grow. Clamp it at both ends so it cannot grow, and something else must give. The material wants to expand by \Delta L = \alpha L_0 \Delta T but is forced to remain at L_0. Equivalently, you have imposed a compressive strain

(the negative sign says the strain is compressive: the material is shorter than it would like to be).

By Hooke's law, this strain generates a stress

where Y is Young's modulus and the sign convention has been absorbed so that \sigma > 0 is a compressive stress pushing outward on whatever is clamping the rod.

Why: Hooke's law says stress equals modulus times strain, \sigma = Y\epsilon. For a rod heated and clamped, the forced strain is -\alpha\Delta T (compressive), and the rod pushes back with a stress Y\alpha\Delta T trying to restore its natural length. The clamps must supply that stress — and if they cannot, something breaks.

For steel, Y \approx 2.0 \times 10^{11}\,\text{Pa} and \alpha = 11 \times 10^{-6}\,\text{K}^{-1}, giving

A 40 K rise generates roughly 90 MPa — already about a quarter of the yield stress of mild steel. Ignore thermal expansion in a constrained rail and by summer it will warp. This is why long rails are either laid with expansion gaps or (for modern continuous-welded rail) deliberately pre-stressed while hot so the residual tension swallows the expansion without warping.

Worked: stress in a clamped brass rod

A brass rod of length 0.8 m is clamped rigidly at both ends at 20°C and then heated to 80°C. Find the thermal stress. Y_{\text{brass}} = 1.0 \times 10^{11}\,\text{Pa}, \alpha_{\text{brass}} = 19 \times 10^{-6}\,\text{K}^{-1}.

Why: \Delta T = 60 K, the values of Y and \alpha multiply cleanly. Note the length cancelled out — thermal stress does not depend on the length of the rod, only on how hot you made it and on the material constants. A longer rod pushes harder on the clamps with a larger force, but the stress (force per unit area) is the same.

The force on each clamp, for a rod of cross-sectional area 1\,\text{cm}^2 = 10^{-4}\,\text{m}^2, is F = \sigma A = 11,400 N — more than a tonne of force per square centimetre of rod. Unclamped, the rod would simply grow by \Delta L = 19\times10^{-6}\cdot 0.8\cdot 60 \approx 0.91 mm. Clamped, it pushes with over a tonne.

Bimetallic strips — turning expansion into motion

Bond a strip of brass onto a strip of steel of the same length and hold the pair flat at room temperature. Now heat both. Brass expands twice as fast as steel (19 versus 11 per million per kelvin). The brass layer wants to be longer than the steel layer — but they are glued together, so they cannot separate. The only way for the composite to accommodate both expansions is to bend, with the brass on the outside (longer) of the curve and the steel on the inside.

Bimetallic strips are the mechanical heart of every old mechanical thermostat, every mercury-free room-temperature gauge, and many electric-iron cutouts: as the iron heats up, the strip curves and opens (or closes) an electrical contact, switching the heater off at a set temperature. Reverse the process on cooling and the circuit closes again. A coiled bimetallic strip — same principle, just wound up — drives the hand on the circular oven thermometers you see in small bakeries.

The bend formula (for curious readers)

If you want the radius of curvature R of a symmetric bimetallic strip of total thickness d after heating by \Delta T, the result is

for two layers of equal thickness and similar Young's moduli. A strip 2 mm thick made of brass and steel, heated by 50 K, curves to a radius of about 42 cm — a visible arc in a 10 cm strip. The derivation is given in the going-deeper section below; the intuition to carry away now is that thinner strips curve more sharply (higher 1/R for smaller d), which is why commercial thermostats use strips fractions of a millimetre thick.

Anomalous expansion of water — the exception that keeps fish alive

Everything above said: heat it, it expands. Water disobeys over one narrow range.

Between 0°C and 4°C, water contracts on heating. Its volume decreases as the temperature rises. Above 4°C, it expands normally — the curve returns to its usual upward slope. So the density of water is not a monotonic function of temperature: it rises from 0 to 4°C, reaches a maximum at 4°C, then falls at higher temperatures.

At the molecular level, this happens because water molecules form tetrahedrally-coordinated hydrogen bonds, and an ice-like open arrangement is less dense than liquid water at 4°C. As ice melts and the temperature begins to rise, the remaining ice-like clusters break up, pulling the molecules closer together and raising the density. Above 4°C, ordinary thermal agitation takes over, pushing molecules apart again. The two effects balance exactly at 4°C.

What this means for a Himalayan lake in winter

Put a tarn somewhere in the upper Himalayas and let December arrive. The air temperature drops below 0°C. The surface water cools; as long as the surface is warmer than 4°C, the cooled water is denser than what is below it, so it sinks. Warmer water rises to take its place and is cooled in turn. The lake overturns continuously — and the entire lake column cools toward 4°C.

Once the surface reaches 4°C and keeps cooling, a remarkable switch happens. Surface water below 4°C is less dense than the 4°C water beneath it, so it stops sinking. A thin, cold layer floats on top. Continue cooling and the surface forms ice, which is less dense still and floats decisively. From this point on, the bottom of the lake stays at 4°C — the densest water, insulated from above by its own ice cap and cold surface layer.

That 4°C bottom layer is why fish survive winters in Dal Lake, in Pangong Tso, in Gurudongmar. If water were an ordinary liquid — densest at 0°C, continuously sinking — the lake would cool uniformly down to freezing and then freeze from the bottom up, destroying everything in it. The peculiar bend in the water density curve between 0 and 4°C is the reason aquatic life exists in temperate and polar regions at all. A single exception to an otherwise universal rule has shaped half the planet's ecology.

Why glass jars crack when you pour in hot pickle

Another daily consequence of water's (and glass's) expansion behaviour: a thick glass jar, sitting at 25°C, when filled with hot pickle or hot chai at 80°C, often cracks. The culprit is thermal stress from differential expansion. Glass conducts heat poorly. When hot liquid hits the inside wall, the inner surface of the glass heats up and tries to expand — but the outer surface, still cool, refuses to give. The inner layer pushes outward against a cold, rigid outer shell. The resulting tensile hoop stress in the outer layer exceeds glass's yield stress (about 50 MPa for ordinary glass), and the jar cracks.

Borosilicate glass (brand-named Pyrex or Borosil in Indian kitchens) has \alpha of only 3.3\times 10^{-6}\,\text{K}^{-1} — about one-third that of ordinary glass. The thermal stress it generates per kelvin is a third as large; it can survive the same hot pickle without cracking. This is why pickle bottles, tea-flasks, and laboratory beakers sold for hot use are made of borosilicate and not soda-lime glass. The whole difference is that one number in the third column of the expansion table.

Worked examples

Explore the expansion yourself

The interactive below lets you drag a temperature-change slider and watch a steel rail (or brass, or aluminium) extend in real time. The numerical readouts update live so you can see which metals cause the biggest worry for bridge-design engineers.

Common confusions

• "The hole expands into the metal." It does not. A hole in a heated plate grows exactly as if it were filled with metal of the same material. The hole's radius, area, and volume (for 3D holes) all grow by the usual factors. The rule: every linear dimension scales by (1 + \alpha\Delta T), including the linear dimensions of holes, channels, and cavities.

• "\beta = 2\alpha and \gamma = 3\alpha are separate laws to memorise." They are not separate laws. They follow algebraically from expanding (1 + \alpha\Delta T)^n to first order. The only assumption is \alpha\Delta T \ll 1, which is always true for sub-melting temperature changes. If you ever see a table quoting \beta and \gamma as "independently measured" quantities, those are just 2\alpha and 3\alpha with tiny second-order corrections.

• "A thermometer reads the real expansion of mercury." It reads the apparent expansion — mercury's real expansion minus the glass vessel's expansion. The column height corresponds to (\gamma_{\text{Hg}} - \gamma_{\text{glass}}) V_0 \Delta T divided by the capillary cross-section area. A precision thermometer is calibrated against a water or gas standard to absorb this correction automatically.

• "Water expands as it freezes because water is weird." The freezing-from-liquid transition is a separate phenomenon; water expands by about 9% at the freezing point as it goes from liquid to solid ice. The anomalous expansion discussed in this article is the behaviour of liquid water between 0 and 4°C, where the liquid (not the phase transition) contracts on warming. Both phenomena exist, both are caused by hydrogen bonding, but they are physically distinct.

• "Brass thermostats work because brass is more sensitive to temperature." Brass alone cannot make a thermostat move — a free brass strip just grows in place. The bimetallic trick is differential expansion: brass bonded to steel, forcing the composite to bend. Both metals' coefficients matter; what drives the bending is the difference \alpha_{\text{brass}} - \alpha_{\text{steel}}.

• "Thermal expansion always means the object gets bigger." Over ordinary temperature ranges, yes — but there are materials with negative linear expansion in certain ranges. Zirconium tungstate (\text{ZrW}_2\text{O}_8) shrinks over nearly 1000 K of heating. Indium bismuth alloys used in reinforced composites similarly contract. These are exotic but exist — they are the engineering answer to the problem of matching the thermal expansion of glass and silicon in microelectronics.

Where this leads next

• Thermal Expansion of Gases — gases expand about a hundred times more than solids for the same \Delta T, and the linear law gives way to Charles's and Gay-Lussac's laws en route to the ideal-gas equation.
• Calorimetry and Specific Heat — how much heat you must supply to produce the temperature change \Delta T that drives thermal expansion in the first place.
• Conduction — the mechanism by which a rail, or a glass wall, actually gets hot when heat flows through it.
• Elasticity — Stress and Strain — Hooke's law and Young's modulus, the elastic framework behind the thermal-stress formula \sigma = Y\alpha\Delta T.
• Temperature and Thermometers — the temperature scales (Celsius, Kelvin) that make \Delta T a well-defined quantity in the first place, and the thermometers that use thermal expansion to measure it.