aiwiki.page
English
Science / Physics / temperature

Temperature

Temperature is the physical quantity that determines the direction of heat flow between bodies and reflects the average energy of microscopic motion. Its SI unit is the kelvin.

79 keywords957 linked from4 not yet writtenWritten by AI
ThermodynamicsHeatKelvinAbsolute ZeroBoltzmann Consta…Climate ChangeEntropyThermodynamic Eq…Temperatur…

Temperature is a physical quantity that indicates how hot or cold a body is. In thermodynamics, it determines the direction in which heat flows between bodies in contact: heat moves from the hotter body to the colder one until both reach equal temperature. At the microscopic level, temperature is closely tied to the average energy of random particle motion. In the International System of Units, it is measured in kelvin, and the scale is anchored to absolute zero and to the Boltzmann constant.

Temperature matters because it governs a wide range of phenomena. It determines when matter melts, boils, or becomes superconducting; it sets the rates of chemical reactions and the activity of enzymes; it constrains the efficiency of engines; and it drives weather and climate change. Measuring it accurately has also shaped the development of metrology and of the entropy-based understanding of energy.

This entry first defines temperature through thermal equilibrium and its thermodynamic and statistical-mechanical meanings, then distinguishes it from heat and links it to the ideal gas. It then traces the historical development of temperature scales and thermometry, and the modern SI definition. The entry continues with how temperature shapes radiation, phase change, chemical kinetics, living systems, and the Earth's climate, and ends with engineering uses, the role of "temperature" as a parameter in statistics and machine learning, and the limits of the concept in nonequilibrium and negative-temperature systems.

Thermal equilibrium and the zeroth law

When two bodies are placed in thermal contact, energy can pass between them through a shared boundary or direct contact, but no work is done and no matter is exchanged. Energy flows from the body with the higher temperature to the body with the lower one, and the flow continues until the net transfer stops. At that point the bodies are in thermal equilibrium and share the same temperature. The energy exchanged during the process is heat, which is treated in a later section. The point here is that the exchange ends only when temperatures match. Equilibrium is therefore a condition of zero net energy flow, and temperature is the property the bodies share when that condition holds. The energy involved belongs to each body's internal energy, which changes as the bodies approach equilibrium.

The zeroth law of thermodynamics states that if body A is in thermal equilibrium with body C, and body B is also in equilibrium with C, then A and B are in equilibrium with each other. This property is called transitivity. Because equilibrium behaves this way, it divides bodies into classes whose members share a common temperature, and a numerical label can be attached to each class. Any two bodies can then be compared without being placed in direct contact with each other. A thermometer relies on this logic: it acts as the intermediary body C, first brought into equilibrium with one system and then with another. The law was named by Ralph Fowler in the 1930s, after the first and second laws had already been numbered, because it was recognized as a logical foundation for those laws and for the concept of temperature.

Temperature is an intensive property, meaning it does not depend on how much matter is present. Consider two identical containers of water, each at 20 °C. When they are combined, the mixture is still at 20 °C, but its internal energy is twice that of either container. Doubling the amount of matter doubles the internal energy, which is an extensive quantity, yet leaves the temperature unchanged. For this reason temperature cannot be found by adding up contributions from the parts of a system. Size also plays no role in the equilibrium condition itself. A small body placed in contact with a much larger one ends up at the larger body's temperature, which is why a thermometer must have a small heat capacity relative to the system it measures, so that it does not significantly alter that system.

Physicists often describe temperature operationally. A temperature reading is a statement about the outcome of a hypothetical contact: if the measured body were placed in thermal contact with the reference system, no net heat would flow between them. A thermometer is calibrated so that its indicated value corresponds to this condition. This is why the reading does not depend on the particular material inside the instrument. A mercury column, a resistor, and a gas bulb will give the same temperature for the same body, provided each is calibrated against the same reference. The zeroth law guarantees that such comparisons are consistent, because if two instruments each agree with a third, they agree with each other. The operational definition applies to bodies that are internally in equilibrium. In strongly non-equilibrium states, a single temperature may not be meaningful, a topic addressed later in this entry. The numerical value of a temperature reading then depends on the chosen scale, and the SI unit for thermodynamic temperature is the kelvin.

Microscopic meaning: kinetic theory and statistical mechanics

For an ideal gas of point-like molecules, temperature can be traced directly to molecular motion. In kinetic theory, which Daniel Bernoulli sketched in 1738 and which Maxwell and Boltzmann put on a quantitative footing in the 1860s and 1870s, pressure arises from the momentum that molecules transfer when they strike the walls of a container. Combining this with the empirical gas law pV=NkTpV = NkT yields ⟨Etrans⟩=32kT\langle E_{\text{trans}}\rangle = \tfrac{3}{2}kT for the mean translational kinetic energy of each molecule, where kk is the Boltzmann constant. Temperature is therefore proportional to average translational kinetic energy, and the constant of proportionality is fixed by kk alone. The same relation gives a root-mean-square molecular speed of vrms=3kT/mv_{\text{rms}} = \sqrt{3kT/m}, about 517 m/s for nitrogen at 300 K. The full spread of speeds around this value is described by the Maxwell–Boltzmann distribution.

The relation holds exactly only for translational motion in a monatomic gas. When molecules also rotate and vibrate, the mean translational energy no longer captures the whole thermal energy, and temperature tracks the total energy distributed among all modes. Because the result is an average, temperature is a statistical quantity: an individual molecule has an energy but no temperature of its own.

Statistical mechanics generalizes this picture to any system in thermal equilibrium with a large reservoir. The probability of finding the system in a microstate of energy EiE_i is proportional to e−Ei/kTe^{-E_i/kT}, and normalizing these weights by the partition function ZZ gives the Boltzmann distribution:

Pi=e−Ei/kTZP_i = \frac{e^{-E_i/kT}}{Z}

For two levels, the ratio of populations is N2N1=g2g1 e−(E2−E1)/kT\frac{N_2}{N_1} = \frac{g_2}{g_1}\,e^{-(E_2-E_1)/kT}, where gg counts degenerate states. Temperature thus sets the relative populations of states at a given energy spacing. When kTkT is small compared with the gap, the ground state dominates; when kTkT is large, the populations approach a distribution weighted only by degeneracy. This dependence is why spectroscopists can read temperatures from line intensities. Temperature also has an entropy-based definition, 1/T=(∂S/∂E)V,N1/T = (\partial S/\partial E)_{V,N}, where entropyentropy is given by Boltzmann's formula S=kln⁡WS = k\ln W. For dilute gases, Bose–Einstein and Fermi–Dirac statistics reduce to the Boltzmann form when occupation numbers are small.

Classical statistical mechanics also gives the equipartition theorem: each independent quadratic term in the energy, or each degree of freedom, carries an average energy of 12kT\tfrac12 kT. A monatomic gas has three translational terms, so its internal energy is 32NkT\tfrac32 NkT and its molar heat capacity at constant volume is 32R\tfrac32 R, where RR is the gas constant. A diatomic gas near room temperature adds two rotational terms, giving 52R\tfrac52 R, which matches measurements on many gases. Vibrational modes contribute two quadratic terms each, a kinetic and a potential one, but at ordinary temperatures they are largely frozen out, a quantum effect that classical equipartition cannot reproduce. In solids, each atom behaves as three oscillators, giving a molar heat capacity near 3R3R, the Dulong–Petit value. The deviations at low temperature, which Einstein and Debye explained in 1907 and 1912 with quantized vibrations, showed the limits of the classical theory.

Temperature is also subject to fluctuations. A system held at fixed temperature does not have a sharply defined energy: the variance of its energy is ⟨(ΔE)2⟩=kT2CV\langle(\Delta E)^2\rangle = kT^2C_V, so the relative spread shrinks roughly as 1/N1/\sqrt{N} with the number of particles NN. For a sample of NN ideal monatomic molecules, the relative fluctuation of kinetic energy is 2/(3N)\sqrt{2/(3N)}. That amounts to a few parts in a thousand for a million molecules and about 10−1110^{-11} for a mole, which is why macroscopic thermometers appear to read a definite value. For nanoparticles or single biomolecules, however, the apparent temperature visibly fluctuates. The same statistical behavior underlies the jittering of microscopic particles in Brownian motion, which Einstein analyzed in 1905.

Thermodynamic temperature, absolute zero and the third law

A scale independent of any thermometric substance

Early thermometers depended on the behavior of a chosen fluid or gas, so their readings carried that substance's properties. Thermodynamic temperature was built to remove this dependence, starting from the heat engine operating in a reversible cycle. The thermal efficiency of a Carnot engine operating between two heat reservoirs depends only on their temperatures, not on the nature of the working substance. For such an engine, the efficiency is η=1−Tc/Th\eta = 1 - T_c/T_h, where ThT_h and TcT_c are the hot and cold reservoir temperatures, so the relation involves no material constants at all.

Kelvin used this idea to fix a scale. Taking the Carnot cycle as the reference, the ratio of heat absorbed at the hot reservoir to heat rejected at the cold one is set equal to the ratio of the two absolute temperatures:

QhQc=ThTc\frac{Q_h}{Q_c} = \frac{T_h}{T_c}

Kelvin found that Carnot's assumption that heat is a conserved quantity had to be abandoned, yet his fundamental work on heat engines was still correct. The theory took roughly three decades to mature. Carnot's discovery of the temperature function was developed gradually over about thirty years by Clapeyron, Helmholtz, Joule, Rankine, Thomson (Kelvin), and Clausius. In practice the scale is anchored to a reference point. Before 2019 that point was the triple point of water, to which the value of 273.16 K was assigned.

Absolute zero and the third law

The zero of the Kelvin scale is the point where the system's energy is at its minimum. At absolute zero the system has its minimum total internal energy, counting kinetic plus potential energy. Quantum effects complicate the classical picture, because at T = 0 molecular motion does not entirely stop.

Absolute zero is a limit that can be approached but not attained. The third law of thermodynamics makes this precise. Its most widely accepted form, the unattainability principle, is usually attributed to Nernst. It was stated in 1912 and holds that no thermodynamic process can reach the temperature of absolute zero in a finite number of steps. A general proof was long missing: the only evidence was that particular cooling methods become less efficient as the temperature decreases. A 2017 derivation by Masanes and Oppenheim covers arbitrary cooling processes, and it finds that the lowest reachable temperature can scale as an inverse power of the cooling time.

The underlying mechanism concerns entropy. In the Nernst form, as the temperature approaches absolute zero, the entropy difference between any two equilibrium states must approach zero. Each cooling stage then removes less and less entropy, so reaching zero would require an infinite number of these diminishing steps. A related consequence is that the heat capacity of a substance must vanish at absolute zero, since otherwise the entropy would diverge.

Absolute temperature, entropy and reversible heat

Absolute temperature is tied to entropy through reversible heat exchange. For a reversible process in which a small amount of heat δQrev\delta Q_{\text{rev}} flows into a body at temperature TT, the entropy change is

dS=δQrevTdS = \frac{\delta Q_{\text{rev}}}{T}

This relation makes temperature the factor that converts heat into entropy. Some treatments take it as the definition of absolute temperature, which makes the Carnot-engine construction and the entropy approach two routes to the same scale.

Laboratory approaches to the lowest temperatures

The lowest temperatures reached in laboratories are for ultracold atomic gases. Two methods do most of the work. In laser cooling, an atom absorbs a photon and re-emits it in a different direction, and the difference in photon energy is extracted from the atomic motion, so the atoms slow down. Evaporative cooling takes the next step. The highest-energy atoms are bled off from the trap, and elastic collisions among the remaining atoms re-establish equilibrium at a lower temperature. By combining these methods, temperatures below one nanokelvin have been achieved. In 2003, an MIT group cooled a sodium gas to only half-a-billionth of a degree above absolute zero, and a later experiment reached a kinetic temperature of 450 ± 80 picokelvin by decompressing and evaporatively reducing a partially condensed vapor.

Lower values have been reported with other techniques. Matter-wave lensing has produced effective temperatures of tens of picokelvin, which is the lowest kinetic temperature measured to date. Spin subsystems can be colder still in a formal sense: spin distributions at positive and negative effective spin temperatures of 50 pK have been prepared by spin-gradient demagnetization in an optical lattice. These spin temperatures describe internal degrees of freedom, not the motion of atoms, so they are not directly comparable with kinetic temperatures.

Heat versus temperature

Heat and temperature are distinct quantities, although everyday speech often treats them as interchangeable. Temperature is a property of the state of a body: it describes how the microscopic motion and energy of its particles are distributed, and it determines whether two bodies are in thermal equilibrium. Heat is a process. It is the energy transferred from one body to another because of a temperature difference. Once the transfer stops, neither body contains "heat" as a stored quantity. Each body keeps its internal energy, and the transfer has only changed the amount it holds. For this reason, phrases such as "heat content" are misleading in thermodynamics.

The distinction took a long time to establish. In the eighteenth century, chemists such as Lavoisier treated heat as an imponderable fluid called caloric. In 1798, Benjamin Thompson (Count Rumford), who was supervising the boring of cannon barrels in Munich, argued that the heat produced seemed to come from mechanical work rather than from a material substance. In the 1840s, James Prescott Joule measured the mechanical equivalent of heat, showing that work and heat are both forms of energy that can be converted into one another. Heat flows spontaneously from a hotter region to a colder one, a direction fixed by the second law of thermodynamics.

Mechanisms of heat transfer

Thermal energy moves by three mechanisms, each governed by its own law.

Conduction carries energy through solids and through fluids at rest. Energy passes between neighbouring particles through lattice vibrations, through the motion of free electrons in metals, and through molecular collisions. The rate is described by Fourier's law, in which the heat flux is proportional to the temperature gradient, q=−k∇T\mathbf{q} = -k\nabla T, where kk is the thermal conductivity. Combining this with energy conservation yields the heat-equation, whose characteristic diffusion time over a distance LL scales as L2/αL^2/\alpha, with α=k/(ρc)\alpha = k/(\rho c) as the thermal diffusivity. Materials differ widely: copper conducts heat at roughly several hundred watts per metre-kelvin, while wood conducts at a fraction of a watt per metre-kelvin. A metal object therefore feels colder than a wooden one at the same temperature, because it draws heat from the skin faster.

Convection transfers energy by the bulk motion of a fluid. Natural convection is driven by buoyancy arising from density differences, while forced convection is produced by a pump or fan. At a solid surface, the transfer rate is often approximated by Newton's law of cooling, Q˙=hA(Ts−T∞)\dot{Q} = hA(T_s - T_\infty), where hh is a heat-transfer coefficient. Unlike conductivity, hh is not a fixed material property; it depends on flow geometry, fluid properties and velocity.

Radiation transfers energy through electromagnetic waves and needs no medium. The net power exchanged between surfaces is governed by the Stefan–Boltzmann law, which scales with the fourth power of absolute temperature. Radiation therefore becomes dominant at high temperatures and is the only mechanism by which solar energy crosses the vacuum to reach Earth.

Heat capacity and the rate of temperature change

How much a body's temperature rises when it receives a given amount of energy depends on its heat capacity. The specific heat capacity cc is the energy needed to raise the temperature of unit mass by one kelvin, so that, without a phase change, Q=mc ΔTQ = mc\,\Delta T. Water has a specific heat of about 4186 J/(kg·K), which is high compared with most common materials; iron is about 450 J/(kg·K) and copper about 385 J/(kg·K). Consequently, a small cup of water and a bathtub of water can both be at 40 °C, yet the bathtub holds far more thermal energy because its mass is much larger. Equal temperatures thus say nothing about the total energy stored, which also depends on mass and specific heat.

The speed at which a temperature changes depends on both heat capacity and the rate of exchange with the surroundings. In a simple lumped model, a body that exchanges heat with its surroundings across area AA approaches the ambient temperature with a time constant τ=mc/(hA)\tau = mc/(hA). A large mass or high specific heat means a sluggish response, while high conductance to the surroundings speeds it up. This is why sand on a beach heats and cools quickly, whereas a large body of water moderates the temperature of nearby coasts.

Ideal gases and equations of state

The ideal gas law

The ideal gas law is the simplest equation of state linking the pressure, volume, and temperature of a gas. In molar form it reads PV=nRTPV = nRT, where nn is the amount of substance in moles, TT is the absolute temperature in kelvins, and RR is the universal gas constant, approximately 8.314 J·mol⁻¹·K⁻¹. The same constant applies to every gas. Because RR equals the Boltzmann constant multiplied by the Avogadro constant, the law can also be written per molecule as PV=NkBTPV = Nk_\mathrm{B}T, which connects it directly to the kinetic picture of molecules in motion. Engineering texts often replace RR with a substance-specific value, R/MR/M, where MM is molar mass, so that the law is applied per unit mass.

The temperature in this relation must be absolute. Pressure and volume are proportional to TT only when TT is measured from the zero point of the kelvin scale, which is why the law fixes the physical meaning of the zero in absolute zero.

Historical derivation

The law was assembled from separate experimental relations over about two centuries. In 1662 Robert Boyle trapped a fixed quantity of air in a J-shaped glass tube with mercury and showed that its pressure and volume are inversely related at constant temperature; this is regarded as one of the earliest equations of state. Jacques Charles studied the expansion of gases with temperature in the 1780s, but he left the work unpublished. In 1801 John Dalton reported that a range of gases and vapours expanded by the same amount between fixed temperature points, and in 1802 Joseph Louis Gay-Lussac presented the volume–temperature result, crediting Charles's earlier work. Gay-Lussac also derived a pressure–temperature relation at constant volume while building an air thermometer. Using his data for air, he estimated the extrapolated zero of volume to lie at roughly −266.66 °C, close to the modern value of −273.15 °C, though he noted that the law breaks down near the condensation point of vapours.

Boyle's, Charles's, and Gay-Lussac's laws combine into the combined gas law, P1V1/T1=P2V2/T2P_1V_1/T_1 = P_2V_2/T_2, for a fixed amount of gas. The missing ingredient was the amount of gas. In 1811 Amedeo Avogadro proposed that equal volumes of gases at the same temperature and pressure contain equal numbers of particles, and combining this with the earlier relations yielded the general form PV=nRTPV = nRT.

Deviations in real gases

Real gases follow the ideal law closely under ordinary conditions, typically within about 5 percent near room temperature and atmospheric pressure. The deviations become significant at high pressure and low temperature. Two assumptions of the kinetic-molecular model fail under these conditions: molecules occupy a non-negligible volume, and they attract one another. Attraction is strong enough at low temperature that gases condense into liquids, and it can make the measured pressure lower than the ideal prediction.

A common diagnostic is the compressibility factor Z=PV/nRTZ = PV/nRT (see compressibility factor), which equals 1 for an ideal gas. Values below 1 indicate that attractions dominate, and values above 1 indicate that repulsion and excluded volume dominate.

In 1873 Johannes Diderik van der Waals introduced a corrected equation, (P+an2/V2)(V−nb)=nRT(P + an^2/V^2)(V - nb) = nRT (see van der Waals equation). The term aa corrects the pressure for intermolecular attraction, and bb subtracts the volume occupied by the molecules themselves. The constants aa and bb differ from gas to gas and are fitted to experiment. The equation gave a far better description of real behaviour over a wider range of pressures, at the cost of losing the universality of the ideal law. Later equations, such as the virial expansion and cubic models of the Redlich–Kwong and Peng–Robinson type, extend the same approach.

Temperature in equations of state for fluids and solids

More generally, an equation of state is a relation among pressure, volume (or density), and temperature that is characteristic of a substance. For a fluid, it gives pressure as a function of temperature and density, from which other thermodynamic properties such as enthalpy and entropy can be derived by differentiation. Temperature is therefore one of the independent state variables that fix the state of the substance, and the form of its dependence determines how heat capacities and phase behaviour follow.

For solids, equations of state describe how volume changes under compression and with heating. Compression is commonly modelled with forms such as the Birch–Murnaghan equation, while thermal pressure is often described with the Mie–Grüneisen framework, in which the temperature dependence enters through a material-specific parameter. These relations allow temperature and pressure effects in solids and liquids to be separated and compared with measurements.

Historical development

Ancient Greek natural philosophy treated hot and cold as qualities rather than measurable quantities. In the scheme later systematized by Aristotle, each element combined two of four primary qualities (hot, cold, wet, dry), so fire was hot and dry while water was cold and wet. Heat and cold were therefore a matter of kind rather than degree. Devices from antiquity had already shown, through pneumatic effects, that heated air could displace a liquid.

The first instruments that registered heat were thermoscopes: an air-filled bulb joined to a tube whose liquid level shifted as the air expanded or contracted. A 1638 account by Benedetto Castelli places such a device in Galileo's hands around 1603, but no dated instrument survives. Several other names, among them Santorio, Robert Fludd and Cornelis Drebbel, appear in attributions of the design. The earliest written mention of the thermoscope is found in a 1612 medical commentary by Santorio. Sources disagree on when a graduated version first appeared, giving about 1612 in one case and 1625 in another. Air-based instruments also drifted with the pressure of the surrounding atmosphere, which made their readings unreliable. Sealed liquid-in-glass designs followed; Ferdinando II, Grand Duke of Tuscany, is credited with one in 1654. By the close of the 1600s, liquid-filled Florentine instruments had largely superseded open thermoscopes.

The decisive change came in the eighteenth century, when instrument makers began fixing reference points and defining numerical scales. Ole Rømer, working in 1701, anchored zero to a brine mixture and assigned 60 to boiling water. Isaac Newton had earlier proposed a scale based on heated linseed oil. The Royal Society in London received Daniel Gabriel Fahrenheit's scale in 1724. Fahrenheit used a mercury thermometer, set zero at a mixture of ice, water and ammonium chloride, and placed the freezing point of water at 32 and body temperature at 96. René-Antoine Ferchault de Réaumur's 1730 scale, built around an alcohol thermometer, put freezing at 0 and boiling at 80, and it was adopted most widely in France, Germany and Russia. Anders Celsius, an astronomer at Uppsala, presented a two-point thermometer to the Royal Swedish Academy in 1742. His original assignment placed boiling at 0 and freezing at 100, an orientation that was reversed in the 1740s or 1770s, with credit variously given to Linnaeus and Christin. Fixed reference points such as the freezing and boiling of water allowed different laboratories to produce comparable readings, which earlier handmade instruments had not achieved.

In the nineteenth century, temperature moved from a calibrated reading toward a quantity with physical content. Sadi Carnot's 1824 analysis of heat engines showed that the maximum efficiency depended only on the temperatures of the hot and cold reservoirs. William Thomson, later Lord Kelvin, drew on this in 1848 to define an absolute scale independent of any particular thermometric fluid. James Prescott Joule's experiments in the 1840s established the mechanical equivalent of heat, which tied heat to energy. Rudolf Clausius, James Clerk Maxwell and Ludwig Boltzmann then linked temperature to molecular motion: in kinetic theory, the average translational energy of gas molecules rose with temperature. Clausius's work on the second law led to the concept of entropy, which later became a bridge between thermal and statistical descriptions.

Temperature scales

A temperature scale assigns numbers to thermal states by fixing two reference points and dividing the interval between them. Most of the variety in today's scales comes from different choices of those reference points, made at different times for different purposes.

The Fahrenheit scale is the oldest of the scales still in use. The earliest of these is the Fahrenheit Scale invented by the German-Dutch scientist Gabriel Fahrenheit (1686-1736). His zero was tied to a mixture of salt and ice rather than to water alone, and the familiar water points became 32 °F and 212 °F. His scale was set as 0 deg for the freezing point of a water-salt solution and 212 ºF for the boiling point of water. Fahrenheit's proposal in 1724 drew on the earlier Rømer scale, and he multiplied its divisions to avoid fractions. The Fahrenheit scale remains in everyday use mainly in the United States, while most of the world uses Celsius.

The Réaumur scale, proposed in 1730, is now obsolete. The Réaumur scale (°Ré, °Re, °R) sets the freezing and boiling points of water at 0 and 80 degrees respectively. Because the water interval was divided into 80 parts instead of 100, one Réaumur degree equals 1.25 Celsius degrees, so T°C=54 T°ReˊT_{\text{°C}} = \tfrac{5}{4}\,T_{\text{°Ré}}. This scale was popular in France, Germany, and Russia, but it was replaced by Celsius by the 1790s. The Réaumur scale is mainly of historical interest.

The Celsius scale was proposed by the Swedish astronomer Anders Celsius in 1742, but the original version ran in reverse. When Celsius introduced his scale in 1747, it was the reverse of today's scale, with the boiling point of water being zero degrees and the freezing point being one hundred degrees. A year later the two constants were switched, creating the temperature scale used today. Accounts differ on the timing of the final inversion. One source credits it to Linnaeus shortly after Celsius's death in 1744, and another dates the renaming to 1948, while a third gives 1954. The scale's 100 divisions between the fixed points of water are what gave it the name "centigrade," which was used for a long time.

The Rankine scale, proposed in 1859 by the Scottish engineer W. J. M. Rankine, is an absolute scale whose degree matches the Fahrenheit interval. As on the Kelvin scale, which was first proposed in 1848, zero on the Rankine scale is absolute zero, but a temperature difference of one Rankine degree (°R or °Ra) is defined as equal to one Fahrenheit degree, rather than the Celsius degree used on the Kelvin scale. It persists mainly in some engineering fields where heat calculations are carried out in Fahrenheit units.

The kelvin scale is the absolute scale used in physics. Its degree is the same size as the Celsius degree, so conversion between the two is a pure offset: TK=T°C+273.15T_{\text{K}} = T_{\text{°C}} + 273.15 Absolute zero therefore corresponds to −273.15 °C. Conversion to Fahrenheit is different because the degree sizes differ by a factor of 9/5: T°F=95 T°C+32T_{\text{°F}} = \tfrac{9}{5}\,T_{\text{°C}} + 32 and the Rankine value is T°R=T°F+459.67T_{\text{°R}} = T_{\text{°F}} + 459.67.

Scales with arbitrary reference points remain in use because they suit everyday needs. The Celsius scale's 0-to-100 range aligns neatly with the metric system and makes calculations simpler, and it is the standard in science and in most countries. Fahrenheit's finer degrees, which are 5/9 the size of a Celsius degree, give more resolution across the range of daily air temperatures, which is one reason it persists in weather reporting. Many thermometers in Canada, which officially uses Celsius, display both scales for the same reason.

Thermometers and measurement principles

Expansion-based instruments

Liquid-in-glass thermometers use the fact that a liquid expands more than the glass that contains it, so the liquid column rises along a calibrated stem as temperature increases. Bimetallic thermometers use the mismatch between two metals instead. Two strips with different expansion coefficients are bonded together, so the composite bends toward the metal with the lower expansion when heated. That bending moves a pointer or closes a switch, which makes the device suitable for simple dial gauges and thermostats. Both types are robust and need no power, but their resolution and response time are limited by the mass and geometry of the sensing element.

Electrical sensors

Resistance temperature detectors (RTDs) are wire-wound or thin-film devices whose resistance rises with temperature because metals have a positive temperature coefficient of electrical resistance. Standard platinum elements are specified at 100 Ω at 0 °C and change by about 0.39 Ω for each Celsius degree. Platinum dominates the field because among the metals used, platinum produces the most linear, repeatable, and stable RTDs. Platinum thermometers can measure temperatures in the range of 10 to 1200 K, and they are among the most precise temperature sensors available, with resolution and measurement uncertainties of ±0.1 °C. Because the reading depends only on resistance, no special extension cables or cold-junction compensation are required for RTDs, which distinguishes them from thermocouples.

A thermocouple is built from two dissimilar conductors joined at one end. It produces a temperature-dependent voltage through the Seebeck effect, and this voltage can be interpreted as a temperature. The voltage is generated along the wires where a temperature gradient exists, not at the junction itself, and the junction only defines the measuring point. The voltage is generally in the microvolt range, so care must be taken to obtain a usable measurement. Because the output is a difference, a thermocouple reads the temperature between its two ends. Ideally one junction is held at a known temperature, the cold junction, while the other, the hot junction, is placed where the temperature is to be determined, and the measured potential indicates the difference between them. In practice it is often impractical to keep the cold junction at a fixed known temperature, so instruments add an electronic compensation measurement. Common pairings include copper/constantan, iron/constantan, and n- and p-doped bismuth telluride.

Gas thermometers and primary thermometry

A primary thermometer infers temperature from a known physical law, without calibration against another thermometer. The most developed example is acoustic gas thermometry. It exploits the relationship between the speed of sound in a dilute gas and the thermodynamic temperature of that gas. For a monatomic gas, the squared speed of sound is given by u2=53kTmu^2 = \frac{5}{3}\frac{kT}{m}, where kk is the Boltzmann constant and mm is the average molecular mass. The thermodynamic temperature is deduced from absolute speed-of-sound measurements that are traceable to the metre, the kilogram and the second. Since the establishment of ITS-90, spherical and quasi-spherical cavity resonators have been used to realize primary acoustic gas thermometry in the range 7 K to 552 K. Constant-volume gas thermometers are the classical forerunners of this approach, and gas-based methods remain central to checking the scale that instruments use in practice. After the 2019 redefinition of the kelvin in terms of the Boltzmann constant, primary thermometry techniques were developed for direct measurement of thermodynamic temperature. These include acoustic gas thermometry, Johnson noise thermometry, refractive-index gas thermometry, and dielectric-constant gas thermometry.

Noncontact measurement

Pyrometers infer temperature from the thermal radiation a body emits, without touching it. An infrared or optical instrument focuses this radiation onto a detector and converts the signal into a temperature by assuming a relationship between emitted power and temperature, typically the Stefan–Boltzmann law for an ideal emitter. Real surfaces emit less than a blackbody, so the result depends on the emissivity assumed for the target. Errors arise when that value is wrong, when reflected radiation from surroundings is picked up, or when dust, steam, or a window absorbs part of the signal. Pyrometers therefore suit moving, hazardous, or very hot objects, such as molten metal or rotating parts, where contact sensors cannot be used reliably.

Sources of error and calibration

Every contact thermometer measures its own temperature, so the reading is accurate only when the sensor is at the temperature of the body it probes. Thermal lag arises because the sensor has heat capacity and thermal resistance at its surface, so its reading trails rapid changes. Self-heating occurs when the measuring current dissipates power in the sensing element, raising its temperature above that of its surroundings; RTDs are commonly operated at low excitation currents to limit this effect. Immersion error appears when heat conducts along the sheath or stem of a probe that is not fully immersed. Calibration against reference standards corrects systematic offsets: a platinum resistance thermometer is often used as a calibrating device for other instruments, and calibrations are traceable to fixed points and national laboratories.

Fixed points, ITS-90 and the 2019 SI redefinition

Practical thermometry has always depended on reference states that can be reproduced anywhere. A fixed point is an equilibrium condition in which a pure substance changes phase at a temperature that is reproducible for a given pressure. Triple points are the most useful, because one pressure fixes the temperature exactly. The triple point of water, where ice, liquid water and vapour coexist, is the best known example. The melting and freezing points of high-purity metals fill the same role at higher temperatures. A thermometer is calibrated by placing it in one of these states and comparing its reading with the accepted value.

The International Temperature Scale of 1990 (ITS-90), specified by the International Committee for Weights and Measures (CIPM) and adopted at the 1989 General Conference on Weights and Measures, turns this principle into a working standard. It replaced the 1968 scale and covers 0.65 K to 1357.77 K, a lower range that was later extended to 0.9 mK by a supplementary scale, PLTS-2000. Its defining points include the triple points of equilibrium hydrogen (13.8033 K), neon (24.5561 K), oxygen (54.3584 K), argon (83.8058 K), mercury (234.3156 K) and water (273.16 K), together with the freezing points of tin, zinc, aluminium and silver. Between the hydrogen triple point and the silver freezing point, platinum resistance thermometers are interpolated by prescribed procedures. Above the silver point, the scale relies on radiation thermometry and the Planck radiation law. ITS-90 is intended as an approximation of thermodynamic temperature, chosen so that laboratories in different countries obtain comparable results.

Before 2019, the kelvin itself was tied to a fixed point. Since 1954 it had been defined as 1/273.16 of the thermodynamic temperature of the water triple point, so that this point was exactly 273.16 K. The water triple point was realized in a sealed glass cell evacuated to about 611.657 Pa, in which ice, water and vapour coexist. This approach worked well near ordinary temperatures, but the realization of the reference point carried a non-negligible uncertainty, and the metrology community noted that the definition made measurements at very high and very low temperatures less certain. The uncertainty of the triple-point realization itself was on the order of tens of microkelvin, which set a practical floor on the accuracy of the unit.

On 20 May 2019, the redefinition of the SI came into force, and the kelvin became defined by fixing the numerical value of the Boltzmann constant at exactly 1.380649 × 10⁻²³ J K⁻¹. The unit now rests on a fundamental constant, and the kilogram, metre and second are defined through other exact constants. The water triple point is no longer a defining quantity; it has become a measured value that remains very close to 273.16 K, which was chosen deliberately so that the redefinition would not change the readings of previously calibrated thermometers. Because the change was small compared with the uncertainties of ITS-90, that scale remained the recommended practical standard without major revision.

The new definition shifted the work of realizing the kelvin toward primary thermometry, in which temperature is derived from a physical law and known constants rather than from comparison with another thermometer. Acoustic gas thermometry infers temperature from the speed of sound in a monatomic gas, which scales with the square root of temperature, measured in a resonant cavity. Its determinations of the Boltzmann constant reached relative uncertainties below 1 × 10⁻⁶. The CIPM's Consultative Committee on Thermometry required that a second, independent method reach a relative uncertainty below 3 × 10⁻⁶ before the redefinition could proceed, and dielectric-constant gas thermometry supplied a supporting result at about 1.9 × 10⁻⁶.

Johnson noise thermometry (JNT) is a purely electronic primary method. It compares the thermal voltage noise of a resistor at the unknown temperature with a pseudo-random reference noise generated by a quantum-based voltage-noise source, which is itself tied to electrical standards. The method reaches an uncertainty of about 1 mK near 300 K, and its operating range is commonly quoted as roughly 4 K to 700 K, or up to 1000 K in some reviews. National laboratories have used it to determine deviations from ITS-90 in the range of about 505 K to 933 K, where it overlaps with acoustic and radiation-based results. Other primary approaches, such as Coulomb blockade thermometry and low-temperature acoustic gas thermometry, have been developed to extend realizations of the kelvin below about 25 K.

Blackbody radiation and temperature of radiating bodies

A blackbody is an idealized body that absorbs all incident radiation. In thermal equilibrium it emits radiation whose spectrum depends on nothing but its temperature. The spectral distribution is given by Planck's law:

Bλ(T)=2hc2λ5 1ehc/(λkBT)−1B_\lambda(T)=\frac{2hc^{2}}{\lambda^{5}}\,\frac{1}{e^{hc/(\lambda k_{\mathrm B}T)}-1}

Here hh is the Planck constant, cc the speed of light, λ\lambda the wavelength, and kBk_{\mathrm B} the Boltzmann constant. Classical treatments, which assumed that energy is exchanged continuously between radiation and the cavity walls, could not reproduce the observed spectrum and predicted unbounded emission at short wavelengths, the so-called ultraviolet catastrophe. Planck resolved this by assuming that energy exchange occurs in discrete quanta, which is the origin of the photon picture of electromagnetic radiation.

Wien's displacement law follows from Planck's formula. The wavelength at which the spectral radiance peaks is inversely proportional to absolute temperature, λmax⁡=b/T\lambda_{\max}=b/T, where b≈2.898×10−3 m⋅Kb \approx 2.898\times10^{-3}\ \text{m·K}. Wilhelm Wien discovered this relation several years before Max Planck developed the more general equation. In practice, doubling the absolute temperature halves the peak wavelength, so a heated object shifts from dull red through orange toward bluish white as it gets hotter. Yet the rule must be applied with care. The Planck curve is broad, and across the narrow visible band the spectrum near its peak is nearly flat, so a body's perceived colour depends on the overall shape of its spectrum rather than on the peak position alone.

The total power radiated is described by the Stefan–Boltzmann law. The power emitted per unit area of a blackbody is proportional to the fourth power of its absolute temperature, j∗=σT4j^{*}=\sigma T^{4}, with σ≈5.67×10−8 W⋅m−2⋅K−4\sigma \approx 5.67\times10^{-8}\ \text{W·m}^{-2}\text{·K}^{-4}. Josef Stefan established the relation experimentally, and Boltzmann then derived it theoretically in 1884. The theoretical derivation consists of integrating Planck's spectrum over all wavelengths, which is why the constant can be expressed in terms of fundamental constants. The fourth-power dependence is steep: raising a blackbody from 1000 K to 2000 K multiplies its emitted power per unit area by a factor of sixteen.

These laws make blackbody radiation a practical thermometer for distant objects. Treating a star as a blackbody, its temperature is inversely proportional to the wavelength of its peak intensity. Stefan himself applied his law to determine the temperature of the surface of the Sun. Modern estimates place the solar photosphere near 5,800 K, for which Wien's law gives a peak near 500 nm, in the green part of the spectrum. The same logic applies to the sky-wide background radiation. The cosmic microwave background is a thermal quasi-uniform black body radiation which peaks at 2.725 K in the microwave regime at 160.2 GHz, corresponding to a 1.9 mm wavelength. The FIRAS instrument aboard the COBE satellite later confirmed the spectrum to be a blackbody to better than one part in 10,000. The radiation was emitted when the early universe became transparent at roughly 3,000 K, and its wavelength has since been stretched by cosmic expansion, cooling it to its present temperature.

Phase transitions and critical temperatures

The melting and boiling temperatures of a substance depend on pressure, and the coexistence curves between solid, liquid and gas form the boundaries of a phase diagram. The curves meet at the triple point, the only combination of temperature and pressure at which all three phases coexist in equilibrium. The liquid–vapor curve does not extend indefinitely. It ends at a critical temperature Tc and a critical pressure pc, and this endpoint is the critical point. For water this endpoint lies at 647.096 K and 22.064 megapascals. Above that temperature, a gas cannot be liquefied no matter how much pressure is applied. The fluid in that region is a supercritical fluid, whose properties lie between those of a liquid and a gas.

Phase changes at a fixed pressure take place at a nearly constant temperature, yet they still exchange large amounts of energy. The latent heat absorbed during melting or released during condensation is used to break or form the bonds that hold molecules in their phase, not to raise the temperature. Melting ice at 0 °C and 1 atm, for example, absorbs about 334 J per gram while the mixture stays at 0 °C. The same logic explains why boiling water holds at 100 °C under standard pressure while the burner keeps supplying heat.

Near the critical point, the difference between liquid and vapor shrinks. At the critical point, the properties of the liquid and gas phases become identical. The density difference between the two phases can therefore serve as an order parameter, a quantity that is nonzero on one side of a transition and vanishes at the transition. Transitions of this kind, in which the order parameter goes to zero continuously rather than jumping, are called continuous or second-order transitions, and they carry no latent heat. For ferromagnets, the Curie temperature represents a critical temperature, at which the disordered paramagnetic phase and the spontaneously ordered ferromagnetic phase coincide, and the system undergoes a second-order phase transition. Near such points many quantities follow power laws with critical exponents, and systems exhibiting phase transitions can be divided into a few universality classes depending on the spatial dimension, the kind of order parameter and the symmetry of the Hamiltonian.

Magnetic ordering gives the clearest example of a critical temperature. The Curie temperature of iron is about 1043 K. Below the Curie temperature a ferromagnet has spontaneous magnetization, and above it the material behaves as a paramagnet. In the paramagnetic region the susceptibility follows the Curie–Weiss law χ=CT−TC\chi = \frac{C}{T - T_C}, where TCT_C is the Curie temperature. The same concept applies to ferroelectrics, where the order parameter is the electric polarization that goes from a finite value to zero when the temperature is increased above the Curie temperature.

Superconductivity follows a similar pattern, with a critical temperature below which electrical resistance disappears. Superconductivity is the phenomenon of zero electrical resistance that occurs when some materials are cooled below a critical temperature. Early superconductors had low critical temperatures. Among known superconducting materials, the Nb3Ge compound showed the highest Tc, which could not exceed 23.2 K. The discovery of cuprate ceramics in 1986 raised the figure substantially, and a material with a critical temperature on the order of 90 K was reported in 1987. Hydrides under extreme pressure have pushed the record further, although the reported values differ between studies. A 2015 result placed hydrogen sulfide superconducting at about 200 K, while a 2018 report described lanthanum hydride as superconducting at 250 K. Both results require pressures of millions of times atmospheric pressure, which is why the practical significance of these records remains limited.

Chemical kinetics and equilibrium

Temperature affects both how fast a chemical reaction proceeds and where a reaction mixture settles. In chemical kinetics, warming a mixture usually speeds reactions up. A larger fraction of molecules has the necessary energy to react as temperature rises. The reactants must collide with enough energy to overcome the activation energy barrier, and a modest rise in temperature increases the share of collisions that do so, which is why the effect is so pronounced.

The quantitative form of this dependence was given by Svante Arrhenius. He developed the equation in the late 19th century. In the Arrhenius relation, the rate constant is written as k=A e−Ea/RTk = A\,e^{-E_a/RT}, where the frequency factor AA is also called the pre-exponential factor or Arrhenius factor, and the activation energy EaE_a is the energy barrier. Because temperature appears in the exponent, small temperature changes produce large changes in k. Taking logarithms gives ln⁡k=ln⁡A−Ea/(RT)\ln k = \ln A - E_a/(RT), so a plot of ln k against 1/T is a straight line, and this form is particularly useful when analyzing such a plot. Its slope is related to the activation energy and is given by the formula −EaR-\frac{E_a}{R}. In practice, the same relation is used in reverse: measuring k at two temperatures gives the activation energy through ln⁡(k2/k1)=(Ea/R)(1/T1−1/T2)\ln(k_2/k_1) = (E_a/R)(1/T_1 - 1/T_2). The Arrhenius treatment also explains why catalysts matter more when reactions are slow. The rate of an uncatalyzed reaction is more affected by temperature than the rate of a catalyzed reaction, because the catalyzed pathway has a lower barrier.

Temperature also shifts the position of chemical equilibrium. The van 't Hoff equation, proposed by Dutch chemist Jacobus Henricus van 't Hoff in 1884, links the equilibrium constant to temperature through the standard reaction enthalpy. Its differential form is dln⁡KeqdT=ΔrH⊖RT2\frac{d \ln K_{\mathrm{eq}}}{dT} = \frac{\Delta_r H^\ominus}{RT^2} The sign of the enthalpy change determines the direction of the shift. For a positive, endothermic ΔH° the equilibrium constant increases with temperature. Exothermic reactions behave in the opposite way, so their equilibrium constants fall as temperature rises. In practice, the equation is usually integrated under the assumption that the standard reaction enthalpy is constant, and this is only approximate because the standard reaction enthalpy and entropy do vary with temperature for most processes. A van 't Hoff plot, which plots ln K against 1/T, is therefore a standard experimental method for determining reaction enthalpy. This temperature dependence has direct consequences for industry. Many large-scale syntheses are exothermic, so raising the temperature to speed up a reaction tends to lower the equilibrium yield, and plant operators must balance rate against yield.

The temperature dependence of spontaneity is captured by the Gibbs free energy. The Gibbs free energy change is written as ΔG = ΔH − TΔS, which can be grouped into an enthalpy term and an entropy term. American scientist Josiah Willard Gibbs proposed this function in 1873. The enthalpy term does not depend on temperature, whereas the entropy term is multiplied by T, so temperature can change which term dominates. Temperature can be the deciding factor when the enthalpy and entropy terms have opposite signs. For example, a reaction with negative ΔH and negative ΔS is spontaneous only at low temperature, because it is driven by enthalpy. The reverse case, with positive ΔH and positive ΔS, is spontaneous only at high temperature, where the entropy term wins.

The link between Gibbs energy and equilibrium connects these ideas. The standard free energy change is related to the equilibrium constant by ΔG⊖=−RTln⁡K\Delta G^\ominus = -RT\ln K. Substituting ΔG⊖=ΔH⊖−TΔS⊖\Delta G^\ominus = \Delta H^\ominus - T\Delta S^\ominus gives the integrated form of the van 't Hoff relation, so the temperature dependence of K reflects both the enthalpy and the entropy of the reaction.

Temperature in biology and living systems

Biological molecules work within narrow thermal limits. An enzyme usually has an optimum temperature at which its catalytic rate peaks. Rates rise at first because molecules move faster and collide more often, and then fall as the protein loses its folded structure. Human enzymes often work best near 37 °C, while enzymes from heat-loving microorganisms have much higher optima. Denaturation is only part of the explanation for the decline above the optimum: some researchers argue that changes in heat capacity of the enzyme–substrate complex account for much of it. The measured optimum also depends on the experiment. In one study, the apparent optimum of a single enzyme shifted by about 5 °C when only the assay length or enzyme concentration changed, because denaturation during the measurement altered the activity curve.

Organisms are commonly sorted by how they manage body temperature. Endotherms such as birds and mammals generate most of their heat through metabolism and keep their core temperature within a fairly narrow range, which requires a substantial energy budget. Ectotherms, including most fish, amphibians, reptiles and insects, depend more on external sources of heat. They often regulate their temperature through behavior, such as basking in sunlight, seeking shade, or moving between microhabitats. The division is a spectrum rather than a strict dichotomy, and many species combine both strategies at different times of day or stages of life.

In humans, thermoregulation is coordinated by the hypothalamus, which compares core temperature with a set point and triggers corrective responses. When the body is too warm, blood vessels in the skin dilate and sweat evaporates to remove heat. When it is too cold, vessels near the skin constrict to reduce heat loss, and shivering, the involuntary contraction of skeletal muscles, produces additional heat. Fever is a different kind of response. Substances released during infection act on the hypothalamus and raise the set point, so the body conserves and generates heat until it reaches the new value. Once the set point falls again, sweating and vasodilation resume. These mechanisms form part of wider homeostasis, in which several feedback loops work together.

Temperature also governs the pace of life across organisms. For many biological processes, the rate increases by a factor of roughly two to three for each 10 °C rise within the normal physiological range, a relationship often summarized by the Q10 coefficient. This is closely related to the temperature dependence of reaction rates more generally, and it means that the development time of ectotherms such as insects can be predicted from accumulated thermal exposure, a method known as degree-day modeling. Microorganisms are grouped by the temperature ranges in which they grow, from psychrophiles in cold habitats to thermophiles in hot springs and hydrothermal vents. Plants and animals likewise have thermal tolerance limits, and these limits help determine where species can survive. As climate shifts, those boundaries move, which is why thermal tolerance is central to studies of species distribution.

Earth's temperature and climate system

Global surface temperature records

Global mean surface temperature is not read from one instrument. It is assembled from land weather stations, ship and buoy measurements at sea, and satellite data, and each input has its own sampling problems. HadCRUT5 combines the CRUTEM5 land record with the HadSST4 sea-surface record, and it is offered in an unfilled version that avoids the smoothing introduced by interpolation and a spatially complete version. GISTEMP, from NASA's Goddard Institute for Space Studies, takes its land station input from NOAA's GHCN-M and its ocean input from NOAA's ERSSTv5, and it uses nighttime satellite lights to identify urban areas so that urban warming can be removed. Each product expresses temperature as an anomaly relative to a base period rather than as an absolute value, because absolute averages would add uncertainty of their own.

Uncertainty arises from several sources: gaps in spatial coverage (which were severe before about 1950), interpolation schemes used to estimate unobserved regions, the merging of land, ship, buoy and satellite data, changes in instruments and station locations, and local land-use change. HadCRUT5 handles this with a 200-member ensemble, and GISTEMP v4 offers a similar ensemble on a 2° grid. Errors at neighboring stations and in successive years tend to be correlated, so uncertainty in a global trend is larger than a naive estimate based on independent readings would suggest. Satellite sea-surface records are calibrated against drifting buoys, and their per-pixel uncertainty is typically a fraction of a kelvin. Satellite estimates of atmospheric temperature trends have historically carried larger uncertainty than surface records.

Energy budget and radiative balance

The climate is governed by a radiative budget at the top of the atmosphere. Averaged over the globe and over a year, incoming solar radiation amounts to roughly 340 W/m², of which about 30 percent is reflected by clouds, aerosols and the surface, leaving roughly 240 W/m² absorbed. The planet balances this by emitting the same energy back to space as thermal infrared (longwave) radiation. Because emitted power scales with the fourth power of absolute temperature, the surface and atmosphere need only modest temperature changes to restore balance. Clouds, water vapor and trace gases lower the effective emissivity of the planet to a value of roughly 0.78 in simple models, so that the surface sits well above the temperature a bare planet at the same distance from the Sun would have.

Greenhouse gases alter this balance mainly by raising the altitude from which infrared radiation escapes to space, where the air is colder. This reduces outgoing longwave radiation at the top of the atmosphere, creating a net energy surplus. The surplus warms the oceans, land and atmosphere until outgoing radiation again matches absorbed sunlight. Estimates of the present planetary imbalance are on the order of hundreds of terawatts, a figure that is small compared with the total incoming solar flux but large enough to drive sustained warming. For the mechanism of the greenhouse effect itself, see greenhouse effect.

Vertical structure of the atmosphere

In the troposphere, temperature generally falls with height at an average rate of about 6.5 K per kilometre, down to the tropopause, which lies near 10–12 km in mid-latitudes and higher in the tropics. Above it, in the stratosphere, temperature rises with altitude because ozone absorbs ultraviolet radiation. Vertical profiles are measured mainly with radiosondes, which are balloon-borne instruments invented in the late 1920s independently by Vilho Vaisala in Finland and Pavel Molchanov in the Soviet Union. Radiosondes provide high-resolution profiles of temperature, pressure, humidity and wind from the surface into the middle stratosphere, and their records are now supplemented by reference-quality stations in the GCOS Reference Upper Air Network.

Satellites provide the global view. Microwave sounding units, flown from 1979 onward, do not measure temperature directly. They detect microwave emission from oxygen molecules in broad atmospheric layers, so each channel mixes information from several levels, and stratospheric cooling can bias tropospheric channels unless it is corrected. Radio occultation, which uses signals from navigation satellites, retrieves near-vertical profiles with high vertical resolution and is not affected by clouds. Combined observations for 1979 to 2018 show a stratosphere that cooled by roughly 1–3 K and a troposphere that warmed by roughly 0.6–0.8 K, with model studies attributing the stratospheric cooling largely to rising carbon dioxide and declining ozone.

Urban effects and paleoclimate

Urban heat islands arise because cities replace vegetation and moist soils with asphalt, concrete and dark roofs, which store more heat and evaporate less water, while tall buildings reduce nocturnal cooling by trapping radiation and waste heat from traffic and buildings adds further energy. The effect is usually strongest at night and in calm, clear weather. Because many station records come from or near cities and airports, climate datasets apply homogenization and urban-adjustment procedures, and the remaining residual uncertainty is one reason why station siting and metadata matter. Local temperatures can therefore differ markedly from the regional mean even within a few kilometres, depending on land use, elevation, and proximity to water.

Temperatures before the instrumental era are reconstructed from proxies. Ice cores preserve isotope ratios in ice and gas bubbles that record past air composition, while marine and lake sediments hold microfossils and chemical signatures whose composition depends on water temperature. Tree rings, cave deposits and borehole profiles provide additional records. Each proxy must be calibrated against modern observations, and its resolution and dating precision decline the further back a reconstruction reaches, so paleotemperature estimates carry wider uncertainty than the recent instrumental record. For the broader field, see paleoclimatology.

Engineering and everyday technology

Most practical temperature control relies on negative feedback. In a building, a sensor reads the air in a room and switches heating on when the room falls below the target, then switches it off once the target is passed. A common example is the thermostat in your home. It senses the air temperature in the room and turns the heat on if it falls too low. When the air warms up, it turns the heat source off. Heating, ventilation and air-conditioning (HVAC) systems use the same principle in both directions, switching heating or cooling depending on the sign of the error. The loop has a built-in delay: the air temperature supplies the feedback, so the controller never measures the heat source directly, which produces a long control loop time and a lot of lag. If the heater is on the opposite side of the room from the thermostat, the room tends to overheat, because the warm air takes time to reach the sensor. Simple on-off controllers, such as the bimetallic-strip thermostats of older heating systems, make the room temperature swing between two values around the set point, which is often acceptable for residential heating. Commercial and modern systems usually replace this with proportional-integral-derivative controllers that adjust valves, dampers or fan speeds continuously.

Industrial processes depend on temperature control for product quality rather than comfort. Heat treatment of metals such as steel is a clear case: annealing, quenching and tempering work because the microstructure changes with both temperature and time, so furnaces must hold set temperatures and ramp rates within narrow tolerances. In food processing, pasteurization is defined by specific time–temperature combinations, and refrigerated storage and transport form a cold chain in which deviations can spoil the product. In pharmaceutical manufacturing, reaction temperature affects the rate, selectivity and yield of chemical synthesis, and many biological products and vaccines must be kept within narrow temperature ranges from production to use. Freeze-drying, another common method, depends on controlled shelf temperature and pressure.

Electronics is one of the most demanding areas of thermal control. The relevant quantity is the semiconductor device's junction temperature, which is the temperature in its active region. It is the critical value for reliability and service life, because it is higher during operation than the case or ambient temperature. As a rule of thumb, a rise of just 10 Kelvin already halves the expected lifetime. Junction temperatures above 160 °C will almost always result in early failure of the semiconductor device. Manufacturers therefore use thermal management to keep junction temperature low. Options include better thermal interface materials, larger contact areas, shorter heat paths and, above all, more capable cooling, such as liquid cooling instead of air cooling. Production testing uses the same physics in reverse: by applying increased voltages and temperatures to a semiconductor, the weaker semiconductors will fail during testing.

Heat engines and refrigerators are limited by the temperatures of their reservoirs. For an engine working between a hot reservoir at ThT_h and a cold reservoir at TcT_c, both in kelvin, the maximum possible efficiency is ηCarnot=1−TcTh\eta_{\text{Carnot}} = 1 - \frac{T_c}{T_h} This bound is set by the second law of thermodynamics rather than by engine design, and it is the basis of the Carnot cycle benchmark associated with Sadi Carnot. For example, an engine running between 500 K and 300 K cannot convert more than 40 % of the heat it absorbs into work. A 100% efficient heat engine would require a cold reservoir at absolute zero or an infinitely hot source. Real plants fall further short, partly because real equipment always departs from the ideal model, so the Carnot efficiency should be treated as an upper bound rather than a target. The same temperature limits apply to heat-pump and refrigeration performance: a refrigerator's coefficient of performance is Tc/(Th−Tc)T_c/(T_h - T_c), and running the cycle backwards explains why modern heat pumps can deliver several times more heat energy into a home than the electrical energy they consume.

Statistics and machine learning

In statistics and machine-learning, "temperature" is a scalar hyperparameter that rescales raw scores before they are converted into probabilities. Its most common home is the softmax function, which maps a vector of scores ziz_i to a probability distribution through pi=exp⁡(zi/T)/∑jexp⁡(zj/T)p_i = \exp(z_i/T) / \sum_j \exp(z_j/T). Dividing every score by T before exponentiation means that T alone governs how strongly the output concentrates on the largest score. At T equal to 1 the ordinary softmax is recovered. A sharper, peakier distribution results when T is below 1, and a flatter one when T is above 1. As the parameter grows without bound, the output approaches a uniform distribution, while at zero the highest-scoring element receives all of the probability.

The parameter matters most in text generation with a language model. The softmax output over the vocabulary gives the probability of each candidate next token, and sampling from it introduces randomness. Values of T below 1 concentrate most of the probability mass on the most likely words, whereas values above 1 give less likely words a larger share. In practice, common settings fall roughly between 0 and 2, and a setting of zero means greedy selection of the single most probable token. Higher temperatures raise the entropy and perplexity of the output, which produces more randomness and uncertainty in generation. The same parameter is also used for calibration rather than sampling. In that setting a single value is fitted on a validation set so that the model's stated confidence better matches its observed accuracy, and that value is then kept at deployment.

The term's lineage in stochastic models is usually traced to the 1985 work of Ackley, Hinton and Sejnowski on the Boltzmann machine, which used a temperature to regulate randomness during sampling. The softmax form itself is closely related to the Boltzmann distribution of statistical mechanics: if the scores are read as negative energies, the probability of each state is proportional to an exponential of its energy divided by T.

Two optimization methods carry the same idea further. Simulated annealing searches for a minimum of an objective function by proposing random changes. A change that lowers the objective is always accepted. A change that raises it by ΔE\Delta E is accepted with probability P(accept)=exp⁡(−ΔET)P(\text{accept}) = \exp\left(-\frac{\Delta E}{T}\right) This rule is known as the Metropolis criterion and is motivated by the Boltzmann distribution from statistical mechanics. At high T, many uphill moves are accepted, which lets the search escape local minima. The temperature is then lowered over the run, so that the probability of accepting worse solutions reduces and the search settles near a good solution. The schedule matters: a slow reduction in temperature normally improves the quality of the solution. The method takes its name from the metallurgical process of annealing.

Markov chain Monte Carlo methods use a related but different scheme. For a system of this kind, the probability density of observing a state is given by the Boltzmann distribution, also called the Gibbs distribution when the state space is not finite. Rather than computing that distribution directly, a Markov chain is run at a fixed temperature, and its asymptotic distribution of sampled states approaches the target distribution. Here the temperature is a fixed setting that determines how concentrated the samples are on low-energy regions.

The link to physics is mathematical rather than physical. Scores or objective values play the role of negative energies, and T is a tunable parameter with no measured heat content behind it. In simulated annealing, the Boltzmann constant is typically set to 1, which underlines that no thermometer or thermal bath is involved. The analogy is nonetheless useful. It provides a single dial that trades exploration against exploitation: a higher temperature makes the differences between class probabilities smaller, which can help when options should be explored more evenly, while a lower temperature favors confident, exploitative choices. This makes the parameter useful in optimization and in generative modeling, where it shapes the diversity of outputs.

Negative temperatures and nonequilibrium limits

Definition and population inversion

Negative absolute temperatures follow from the statistical definition of temperature, 1T=(∂S∂E)V,N\frac{1}{T} = \left(\frac{\partial S}{\partial E}\right)_{V,N}, where SS is entropy, EE is energy, VV is volume and NN is particle number. Most familiar systems gain entropy as their energy rises, which keeps their temperatures positive. A negative value requires that adding energy shrink the number of accessible microstates, and this is possible only when the energy spectrum has an upper bound. Free particles, whose kinetic energy has no ceiling, therefore cannot reach negative temperature, which confines the phenomenon to special systems such as nuclear spins, optical lattices and certain quantum gases.

In a two-level system, the Boltzmann distribution gives the population ratio N2N1=e−(E2−E1)/kBT\frac{N_2}{N_1} = e^{-(E_2 - E_1)/k_\mathrm{B}T}. At positive temperature the lower level is always more populated. When TT is negative, the same expression makes the upper level the more populous one, a state known as population inversion. Such an inverted distribution can be stable and in thermal equilibrium, provided that the relaxation processes within the subsystem are much faster than its coupling to the environment. The Boltzmann distribution and the Boltzmann constant are the basic tools for this description.

Early history

The first laboratory realization came in 1951, when Edward Purcell and Robert Pound studied the nuclear spins of lithium-7 in a lithium fluoride crystal, using nuclear magnetic resonance. By rapidly reversing the applied magnetic field, they left the spins in a configuration in which the higher-energy orientation was more heavily populated. The inversion persisted because spin–spin relaxation, which establishes internal equilibrium among the spins, is far faster than spin–lattice relaxation, which exchanges energy with the crystal. The idea had earlier roots in Lars Onsager's 1949 statistical treatment of large-scale vortices in two-dimensional turbulence. In 1956, Norman Ramsey treated these states as genuine thermodynamic equilibrium states and placed them on a formal footing within statistical mechanics.

Heat flow and the debate over interpretation

A system at negative temperature is, in one sense, hotter than any system at positive temperature. When it is placed in thermal contact with a positive-temperature body, heat flows from the negative-temperature system into the positive one, opposite to the usual direction. This reversal is the reason the concept requires care: the ordering of "hotness" is not the naive one, and the second law of thermodynamics has to be applied with the sign of TT in view.

The interpretation has been disputed. Dunkel and Hilbert (2014) argued that, under the Gibbs entropy, the experiments and several proposed quantum-gas realizations do not establish negative absolute temperatures, because the effective temperature defined from Boltzmann entropy differs from the thermodynamic temperature in these regimes. Other thermodynamic analyses conclude that the Purcell–Pound results do fit a negative-temperature description. The disagreement concerns which entropy definition is appropriate, not the measured population inversions themselves.

Motional degrees of freedom and effective temperatures

In 2013, Braun and coworkers reported a negative temperature for the motional degrees of freedom of ultracold bosonic atoms held in an optical lattice. The lattice supplies the upper energy bound that the motional spectrum otherwise lacks. This experiment showed that a negative temperature can apply to one subset of degrees of freedom while others remain at positive values, which is a clear case of temperature differing among degrees of freedom.

Many authors therefore distinguish negative effective temperatures, which describe non-equilibrium population states, from negative thermodynamic temperatures of true equilibrium. Such effective negative temperatures are realized experimentally through population inversion, and theoretical work has examined heat engines and refrigerators in which one reservoir operates at negative effective temperature. More generally, most physical bodies are described by local thermodynamic equilibrium, in which each small region has its own temperature, and this approximation is less secure for very small quantum systems, strongly interacting matter, and plasmas, where the very definition of temperature remains an open question.

References

  1. THE KELVIN TEMPERATURE SCALE Physics Homework Help, Physics Assignments and Projects Help, Assignments Tutors onlinephysicsmax.com
  2. Kelvin and the absolute temperature scaleresearchgate.net
  3. www.arxiv.orgarxiv.org
  4. A derivation (and quantification) of the third law of thermodynamicsarxiv.org
  5. Unattainability of Absolute Zerobohrium.com
  6. Third law of thermodynamicsen.wikipedia.org
  7. How are temperatures close to absolute zero achieved and measured?scientificamerican.com
  8. Introduction to light forces, atom cooling, and atom trappingarxiv.org
  9. MIT Team Achieves Coldest Temperature Everphotonics.com
  10. Cooling of Bose-Einstein condensates below 500 Picokelvinrle.mit.edu
  11. Matter wave lensing to picokelvin temperaturesarxiv.org
  12. Cooling through quantum criticality and many-body effects in condensed matter and cold gasesarxiv.org
  13. A general derivation and quantification of the third law of thermodynamicsnature.com
  14. Absolute zeroen.wikipedia.org
  15. Charles's lawen.wikipedia.org
  16. Gay-Lussac%27s lawen.wikipedia.org
  17. Gas Laws and Clinical Application - StatPearls - NCBI Bookshelfncbi.nlm.nih.gov
  18. Equation of stateen.wikipedia.org
  19. Deviations from the Ideal Gas Lawchemed.chem.purdue.edu
  20. Johannes van der Waalschemed.chem.purdue.edu
  21. Gas Laws: Boyle's Law, Charle's Law, Gay-Lussac's Law, Avogadro's Lawtoppr.com
  22. Van der Waals Equation: Derivation, Relation with Ideal Gas Law %%page%% %%sep%% %%sitename%% - GeeksforGeeksgeeksforgeeks.org
  23. Ideal Gas Law — Use ideal gas law and specific gas constant...varsitytutors.com
  24. Ideal Gas Equation - an overviewsciencedirect.com
  25. The History of Thermometers - Utmelutmel.com
  26. The Galileo Projectgalileo.library.rice.edu
  27. The Thermometer & the Scientific Revolution - World History Encyclopediaworldhistory.org
  28. Thermometeren.wikipedia.org
  29. Thermometry History & The Invention of Thermometers - HVAC Schoolhvacrschool.com
  30. History of the Thermometerlearn.weatherstem.com
  31. A Brief History of Temperature Scalescalcflux.com
  32. Temperature Scales Explainedtutco.com
  33. WHAT ARE THE FOUR BASIC TEMPERATURE SCALES?web.mae.ufl.edu
  34. Temperature Scale - an overviewsciencedirect.com
  35. Anders Celsiusencyclopedia.com
  36. Rankine scaleen.wikipedia.org
  37. What Is Celsius Based On? Water, History & Kelvin - ScienceInsightsscienceinsights.org
  38. Fahrenheiten.wikipedia.org
  39. Anders Celsius - Magnet Academynationalmaglab.org
  40. How Was the Celsius Temperature Scale Developed?allthescience.org
  41. Temperature Sensors based on Semiconducting Oxides: An Overviewarxiv.org
  42. Platinum Resistance Temperature Detectors - an overviewsciencedirect.com
  43. Resistance Temperature Detector - an overviewsciencedirect.com
  44. Platinum Resistance Thermometer - an overviewsciencedirect.com
  45. Thermocoupleen.wikipedia.org
  46. Temperature measurementimage-ppubs.uspto.gov
  47. Thermoelectric method and apparatus for charging superconducting magnetsimage-ppubs.uspto.gov
  48. Acoustic Gas Thermometrynist.gov
  49. Acoustic gas thermometry - ADSui.adsabs.harvard.edu
  50. Practical cylindrical acoustic gas thermometer of constant performance for space exploration - ScienceDirectsciencedirect.com
  51. International Temperature Scale of 1990en.wikipedia.org
  52. ITS-90 Fixed Points - Defining Temperature Pointsits-90.com
  53. The International Temperature Scale of 1990 (ITS-90) H. Preston-Thomaslef.mec.puc-rio.br
  54. The International Temperature Scale of 1990 (ITS-90)dwyeromega.com
  55. ITC - Triple Point of Water Cellitc.gov.hk
  56. Kelvin: Present Realizationnist.gov
  57. kelvin (K) - NPLnpl.co.uk
  58. What Is the Triple Point of Water?fluke.com
  59. The Boltzmann Constant for the Definition and Realization of the Kelvinresearchgate.net
  60. Practical realisation of the kelvin by Johnson noise thermometrynist.gov
  61. (PDF) Practical realisation of the kelvin by Johnson noise thermometryresearchgate.net
  62. johnson noise thermometer: Topics by Science.govscience.gov
  63. Spectral model selection in the electronic measurement of the Boltzmann constant by Johnson noise thermometryarxiv.org
  64. An improved electronic determination of the Boltzmann constant by Johnson noise thermometry - PubMedpubmed.ncbi.nlm.nih.gov
  65. Photon-Atom Granularity Noise Thermometryarxiv.org
  66. Towards realising the redefined kelvin - ScienceDirectsciencedirect.com
  67. The Puzzling of Stefan-Boltzmann Law: Classical or Quantum Physicsarxiv.org
  68. Wien%27s displacement lawen.wikipedia.org
  69. Ludwig Edward Boltzmannarxiv.org
  70. Minimal Length Phenomenology and the Black Body Radiationarxiv.org
  71. Blackbody Radiationpressbooks.online.ucf.edu
  72. Cosmic Microwave Backgroundastro.kias.re.kr
  73. What is Cosmic Microwave Background?rfessentials.com
  74. Wien's displacement lawen.wikipedia.org
  75. From Planck's Law to Stefan–Boltzmannoptris.com
  76. A better presentation of Planck's radiation lawarxiv.org
  77. Planck's blackbody radiation law: Presentation in different domains and determination of the related dimensional constantsarxiv.org
  78. Lecture 38: The Cosmic Microwave Backgroundastronomy.ohio-state.edu
  79. Did the Big Bang create the Cosmic Microwave Background?astronomy.com
  80. Critical point (thermodynamics)en.wikipedia.org
  81. 11.5 Phase Diagramspressbooks.online.ucf.edu
  82. Phase Transformations in Van der Waals Fluidfarside.ph.utexas.edu
  83. A brief account of the Ising and Ising-like models: Mean-field, effective-field and exact resultsarxiv.org
  84. Curie Temperature of Iron - The Physics Factbookhypertextbook.com
  85. Curie temperatureen.wikipedia.org
  86. The record for high-temperature superconductivity has been smashed againtechnologyreview.com
  87. Process for preparing high Tc superconducting materialimage-ppubs.uspto.gov
  88. Superconducting critical temperature under pressure - ScienceDirectsciencedirect.com
  89. Superconductivity temperatures on the risenews.cnrs.fr
  90. Relationship between Reaction Rates, Temperature, and Activation Energy; Arrhenius Equation, Orientation Factor (M13Q8)wisc.pb.unizin.org
  91. Arrhenius Activation Energy for Two Temperaturevcalc.com
  92. Arrhenius Equation Explained: Definition, Examples, Practice & Video Lessonspearson.com
  93. 6.2.3.1: Arrhenius Equation - Chemistry LibreTextschem.libretexts.org
  94. Van 't Hoff equationen.wikipedia.org
  95. Van’t Hoff Equation - Definition, Derivation, and Applicationssciencenotes.org
  96. 9.6: Temperature Dependence of Equilibrium Constants - the van ’t Hoff Equation - Chemistry LibreTextschem.libretexts.org
  97. Spontaneity: Free Energy and Temperaturecourses.lumenlearning.com
  98. Chemistry Tutorial - Thermodynamics - Spontaneity and Gibbs Free Energy - Predicting Spontaneity with Gibbs Free Energyphysicsclassroom.com
  99. 18.5: Second Law and Gibbs Free Energy - Chemistry LibreTextschem.libretexts.org
  100. Arrhenius Equation - an overviewsciencedirect.com
  101. Activation Energy Calculatorpearson.com
  102. Arrhenius Equation Calculatorpearson.com
  103. The Van 't Hoff Equation — How Temperature Bends an Equilibrium Constantunseel.com
  104. ΔG°, K and Van’t Hoff Plots (M17Q6)wisc.pb.unizin.org
  105. Optimum temperature may be a misleading parameter in enzyme characterization and applicationncbi.nlm.nih.gov
  106. On the Temperature Dependence of Enzyme-Catalyzed Ratespubs.acs.org
  107. Effect of Temperature on Enzymatic Reaction - Creative Enzymescreative-enzymes.com
  108. Optimal Temperature in Biological Chemistry Ifiveable.me
  109. Global land-ocean surface temperature data: HadCRUT5climatedataguide.ucar.edu
  110. Global surface temperature data: GISTEMP: NASA Goddard Institute for Space Studies (GISS) Surface Temperature Analysisclimatedataguide.ucar.edu
  111. Estimating trends in the global mean temperature recordarxiv.org
  112. Solar Cycle as a Distinct Line of Evidence Constraining Earth's Transient Climate Responsearxiv.org
  113. Satellite-based time-series of sea-surface temperature since 1981 for climate applicationsnature.com
  114. Earth's energy budgeten.wikipedia.org
  115. Radiation Budget (Earth Energy Balance)studyguides.com
  116. Chapter 7: The Earth’s Energy Budget, Climate Feedbacks, and Climate Sensitivityipcc.ch
  117. The Earth's Radiation Budget - NASA Sciencescience.nasa.gov
  118. Observed Temperature Changes in the Troposphere and Stratosphere from 1979 to 2018 in: Journal of Climate Volume 33 Issue 19 (2020)journals.ametsoc.org
  119. Radiosondes - an overviewsciencedirect.com
  120. Stratospheric Temperature Trendsacom.ucar.edu
  121. Which is a more reliable measure of global temperature: thermometers or satellites?skepticalscience.com
  122. Satellite measurements of warming in the troposphereskepticalscience.com
  123. Tropospheric warming and stratospheric cooling in the 21st centuryuser.eumetsat.int
  124. Upper Air Temperature Validationremss.com
  125. Feedback and Feedforward Control: Explainedirisdynamics.com
  126. principles-of-automatic-controls - Introduction to Feedback Control Systemsandreamunafo.github.io
  127. Calculate junction temperature (Tj): online semiconductor calculatorcooltec.de
  128. Thermal testing apparatus and methodimage-ppubs.uspto.gov
  129. Semiconductor burn-in thermal management systemimage-ppubs.uspto.gov
  130. How Efficient Can a Heat Engine Be? The Carnot Cycle Explainedphysicsfundamentals.org
  131. Carnot Cycle Explained: Efficiency, Diagrams & Formulaturn2engineering.com
  132. How Temperature Affects Semiconductor Reliabilityjetcool.com
  133. Exploring the Impact of Temperature Scaling in Softmax for Classification and Adversarial Robustnessarxiv.org
  134. Calibrated Large Language Models for Binary Question Answeringarxiv.org
  135. Is Temperature the Creativity Parameter of Large Language Models?arxiv.org
  136. Introduction to Simulated Annealing • RAWraw.org
  137. Sampling from a set spins with clampingimage-ppubs.uspto.gov
  138. Stackless Ray-Object Intersections Using Approximate Minimum Weight Triangulations: Results in 2D That Outperform Roped KD-Trees (And Massively Outperform BVHs)arxiv.org
  139. Why should we use Temperature in softmax?codemia.io
  140. What Is and Why Use Temperature in Softmax?baeldung.com
  141. Investigation of Algorithms for Highly Nonlinear Model Fitting on Big Datasetsarxiv.org
  142. Optimization by linear kinetic equations and mean-field Langevin dynamicsarxiv.org
  143. Negative Absolute Temperature and the Dynamics of Quantum Phase Transitionsedoc.ub.uni-muenchen.de
  144. Consistent thermostatistics forbids negative absolute ...kleinex.mit.edu
  145. (PDF) Physics of negative absolute temperaturesresearchgate.net
  146. Negative temperature is cool for coolingarxiv.org
  147. Isotropic-Nematic Phase Transitions in Gravitational Systemsarxiv.org
  148. Inconsistent thermostatistics and negative absolute temperaturesarxiv.org