Negative Temperature

What negative temperature really means

The everyday picture of temperature — a measure of how fast molecules jiggle — is a special case, not the definition. The rigorous statistical-mechanics definition is in terms of how a system's entropy S responds to a change in its energy E:

1 / T = ∂S / ∂E

For almost everything you have ever heated — a gas, a metal, water — pouring in energy opens up more microstates, so entropy rises: ∂S/∂E > 0 and T > 0. But there is no law saying entropy must always increase with energy. If a system reaches a point where adding energy reduces the number of accessible microstates, then ∂S/∂E < 0, and the temperature is negative.

The crucial requirement is a bounded energy spectrum — a ceiling on how much energy the system can hold. An ordinary gas has no such ceiling: a molecule can always move faster, so entropy climbs forever. But a set of magnetic spins in a field, or atoms confined to a single band of an optical lattice, has a finite maximum energy. As you push such a system toward that maximum, the spins or atoms have fewer and fewer ways to arrange themselves, entropy falls, and T flips sign.

The two-level system

The cleanest model is a collection of N independent particles, each of which can sit in only two states: a lower level at energy 0 and an upper level at energy ΔE. Let N_low and N_high be the populations. In thermal equilibrium they obey the Boltzmann distribution:

N_high / N_low = exp(−ΔE / kT)

Read this carefully — it is the heart of the whole subject:

Temperature	Boltzmann factor e^(−ΔE/kT)	Population	Entropy state
T = +0 K	0	All in lower level	Minimum entropy
T small positive	between 0 and 1	Mostly lower	Low entropy
T = +∞	1	Equal: N_high = N_low	Maximum entropy
T = −∞	1	Equal: N_high = N_low	Maximum entropy
T small negative	greater than 1	Mostly upper (inverted)	Low entropy
T = −0 K	∞	All in upper level	Minimum entropy

Notice that +∞ and −∞ describe the same physical state (exactly equal populations). They are two labels for the single point at the top of the entropy curve. Pushing past it — into population inversion, where N_high > N_low — requires the exponent ΔE/kT to be negative, which can only happen if T itself is negative.

Entropy versus energy

For the two-level system the total energy is E = N_high · ΔE. The entropy, from the Boltzmann counting of arrangements, is

S = k · ln( N! / (N_low! · N_high!) )

Plot S against E and you get a symmetric dome. At E = 0 (all low) entropy is zero. At E = N·ΔE/2 (half up, half down) entropy peaks. At E = N·ΔE (all high) entropy is zero again. The slope ∂S/∂E — and therefore 1/T — is positive on the rising left half and negative on the falling right half. The peak, where the slope is zero, corresponds to infinite temperature. The whole right half of the dome is negative-temperature territory.

Why negative is hotter than positive

The sign convention feels backwards until you switch variables. Define the coldness β = 1/(kT). As a system absorbs energy, β slides smoothly downward:

cold ----------- hot
β:  +large  →  +0  →  −0  →  −large
T:  +0 K  →  +∞  →  −∞  →  −0 K

β is the monotonic, well-behaved quantity; temperature is its awkward reciprocal that blows up and changes sign at the top. So the true ordering from coldest to hottest is: +0 K, ..., +300 K, ..., +∞ K, then −∞ K, ..., −300 K, ..., −0 K. A system at −0 K is the hottest thing imaginable. Put a negative-T body in contact with any positive-T body and heat flows out of the negative-T body until they reach a common β. The second law is perfectly happy: total entropy still increases.

The laser connection

A laser is the most famous machine that runs on a negative temperature. Its gain medium is "pumped" so that more atoms sit in the excited level than the ground level — exactly a population inversion. Apply the Boltzmann relation to the two lasing levels and you read off a negative temperature for that transition. With N_high > N_low, an incoming photon is more likely to trigger stimulated emission (adding a coherent photon) than to be absorbed, so the light amplifies. No inversion, no gain, no laser. The medium as a whole is not in global equilibrium — only the two pumped levels are characterized by the negative T — which is why a laser does not violate any thermodynamic law.

How it has been measured

Experiment	System	Method & result
Purcell & Pound, 1951	⁷Li and ¹⁹F nuclear spins in LiF	Sudden reversal of the magnetic field inverted the spin populations; the spin system relaxed through +∞ and into negative T, lifetime ~5 minutes.
Ramsey, 1956	Theory	Showed negative temperatures are thermodynamically consistent, defining the β ordering and the heat-flow direction.
Abragam & Proctor, 1958	Nuclear spins, calorimetry	Brought positive- and negative-T spin systems into thermal contact and watched them equilibrate, confirming negative T behaves as a true temperature.
Braun et al., 2013	~10⁵ ³⁹K atoms in an optical lattice	Inverted the motional energy distribution by flipping the sign of the lattice and interaction terms — the first negative absolute temperature for kinetic (not spin) degrees of freedom.

Putting numbers on it

Take a nuclear-spin-½ system in a field B. The two Zeeman levels are split by ΔE = 2μB. For protons (μ ≈ 1.41 × 10⁻²⁶ J/T) in a 1 T field, ΔE ≈ 2.8 × 10⁻²⁶ J. Suppose a field reversal produces a 60/40 inversion, N_high/N_low = 1.5. Then:

N_high/N_low = exp(−ΔE/kT) = 1.5
⇒ −ΔE/kT = ln(1.5) = 0.405
⇒ T = −ΔE / (k · 0.405)
     = −(2.8e−26) / (1.38e−23 · 0.405)
     ≈ −5.0 × 10⁻³ K = −5 mK

So a modest inversion corresponds to a temperature of a few millikelvin — but negative. The closer the inversion gets to complete (all spins up), the closer T creeps to −0 K, the hottest extreme.

JavaScript — populations and temperature

const k = 1.380649e-23; // Boltzmann constant, J/K

// Boltzmann populations of a two-level system at temperature T
function populations(deltaE, T, N = 1) {
  const ratio = Math.exp(-deltaE / (k * T)); // N_high / N_low
  const N_low = N / (1 + ratio);
  const N_high = N - N_low;
  return { N_low, N_high, ratio };
}

// Temperature implied by an observed inversion ratio r = N_high / N_low
function temperatureFromRatio(deltaE, r) {
  // r = exp(-deltaE / kT)  =>  T = -deltaE / (k * ln r)
  return -deltaE / (k * Math.log(r));
}

const dE = 2.8e-26; // proton Zeeman splitting at 1 T, joules

console.log(populations(dE, 300));        // ratio ~0.999993, barely below 1
console.log(populations(dE, 1e-3));       // cold: almost all in lower level
console.log(temperatureFromRatio(dE, 0.5)); // positive T (more low than high)
console.log(temperatureFromRatio(dE, 1.0)); // ratio 1 => T = ±Infinity
console.log(temperatureFromRatio(dE, 1.5)); // r > 1 => NEGATIVE T (~ -5 mK)
console.log(temperatureFromRatio(dE, 50));  // near-total inversion => T -> -0 K

// Coldness beta is the well-behaved ordering variable
function beta(T) { return 1 / (k * T); }
console.log(beta(300) > beta(-300)); // true: +300 K is COLDER than -300 K

Where negative temperature shows up

• Lasers and masers. Population inversion in the gain medium is, by definition, a negative-temperature distribution over the lasing levels.
• Nuclear magnetic resonance. Spin systems can be driven to negative spin temperatures by adiabatic field reversal; central to early NMR and to dynamic nuclear polarization.
• Ultracold atoms. Atoms in optical lattices with engineered bounded bands realize negative absolute temperatures for motion (Braun 2013), used to probe stability and equilibration.
• Spin systems and magnetism. Bounded-energy spin Hamiltonians (Ising-like) are the textbook setting for entropy domes and sign-flipping temperature.
• Cosmology and dark energy speculation. Negative-temperature states have been floated, controversially, as toy models for systems that behave anomalously under added energy.

Common mistakes

• Thinking negative T is colder than 0 K. It is the opposite. Negative-T systems sit beyond +∞ on the energy scale and are the hottest states there are.
• Trying to reach it in an unbounded system. A gas, a harmonic oscillator, or anything with limitless kinetic energy can never go negative; entropy keeps rising. You need a capped spectrum.
• Treating T as the fundamental variable. The smooth, monotonic quantity is β = 1/(kT). Reasoning in T leads to the apparent paradox of "jumping" from +∞ to −∞; reasoning in β makes it continuous.
• Assuming the whole laser is at negative T. Only the inverted lasing levels are; the medium is out of global equilibrium. Negative temperature is a property of the relevant degrees of freedom.
• Claiming it breaks the second law. It does not. Heat flowing from negative to positive T still increases total entropy, exactly as the second law demands once β ordering is used.
• Confusing population inversion with simply heating. Heating drives populations toward equality (T → +∞); only active pumping past equality produces inversion and negative T.