6
ARTICLES PUBLISHED ONLINE: 8 DECEMBER 2013 | DOI: 10.1038/NPHYS2815 Consistent thermostatistics forbids negative absolute temperatures Jörn Dunkel 1 * and Stefan Hilbert 2 Over the past 60 years, a considerable number of theories and experiments have claimed the existence of negative absolute temperature in spin systems and ultracold quantum gases. This has led to speculation that ultracold gases may be dark-energy analogues and also suggests the feasibility of heat engines with efficiencies larger than one. Here, we prove that all previous negative temperature claims and their implications are invalid as they arise from the use of an entropy definition that is inconsistent both mathematically and thermodynamically. We show that the underlying conceptual deficiencies can be overcome if one adopts a microcanonical entropy functional originally derived by Gibbs. The resulting thermodynamic framework is self-consistent and implies that absolute temperature remains positive even for systems with a bounded spectrum. In addition, we propose a minimal quantum thermometer that can be implemented with available experimental techniques. P ositivity of absolute temperature T , a key postulate of thermodynamics 1 , has repeatedly been challenged both theoretically 2–4 and experimentally 5–7 . If indeed realizable, negative temperature systems promise profound practical and conceptual consequences. They might not only facilitate the creation of hyper-efficient heat engines 2–4 but could also help 7 to resolve the cosmological dark-energy puzzle 8,9 . Measurements of negative absolute temperature were first reported in 1951 by Purcell and Pound 5 in seminal work on the population inversion in nuclear spin systems. Five years later, Ramsay’s comprehensive theoretical study 2 clarified hypothetical ramifications of negative temperature states, most notably the possibility to achieve Carnot efficiencies η> 1 (refs 3,4). Recently, the first experimental realization of an ultracold bosonic quantum gas 7 with a bounded spectrum has attracted considerable attention 10 as another apparent example system with T < 0, encouraging speculation that cold-atom gases could serve as laboratory dark-energy analogues. Here, we show that claims of negative absolute temperature in spin systems and quantum gases are generally invalid, as they arise from the use of a popular yet inconsistent microcanonical entropy definition attributed to Boltzmann 11 . By means of rigorous derivations 12 and exactly solvable examples, we will demonstrate that the Boltzmann entropy, despite being advocated in most modern textbooks 13 , is incompatible with the differential structure of thermostatistics, fails to give sensible predictions for analytically tractable quantum and classical systems, and violates equipartition in the classical limit. The general mathematical incompatibility implies that it is logically inconsistent to insert negative Boltz- mann ‘temperatures’ into standard thermodynamic relations, thus explaining paradoxical (wrong) results for Carnot efficiencies and other observables. The deficiencies of the Boltzmann entropy can be overcome by adopting a self-consistent entropy concept that was derived by Gibbs more than 100 years ago 14 , but has been mostly forgotten ever since. Unlike the Boltzmann entropy, Gibbs’ entropy fulfils the fundamental thermostatistical relations and produces sensible predictions for heat capacities and other ther- modynamic observables in all exactly computable test cases. The 1 Department of Mathematics, Massachusetts Institute of Technology, 77 Massachusetts Avenue E17-412, Cambridge, Massachusetts 02139-4307, USA, 2 Max Planck Institute for Astrophysics, Karl-Schwarzschild-Straße 1, Garching 85748, Germany. *e-mail: [email protected] Gibbs formalism yields strictly non-negative absolute temperatures even for quantum systems with a bounded spectrum, thereby invalidating all previous negative temperature claims. Negative absolute temperatures? The seemingly plausible standard argument in favour of negative absolute temperatures goes as follows 10 : assume a suitably designed many-particle quantum system with a bounded spectrum 5,7 can be driven to a stable state of population inversion, so that most particles occupy high-energy one-particle levels. In this case, the one-particle energy distribution will be an increasing function of the one-particle energy . To fit 7,10 such a distribution with a Boltzmann factor exp(-β ), β must be negative, implying a negative Boltz- mann ‘temperature’ T B = (k B β ) -1 < 0. Although this reasoning may indeed seem straightforward, the arguments below clarify that T B is, in general, not the absolute thermodynamic temperature T , unless one is willing to abandon the mathematical consistency of thermostatistics. We shall prove that the parameter T B = (k B β ) -1 , as determined by Purcell and Pound 5 and more recently also in ref. 7 is, in fact, a function of both temperature T and heat capacity C . This function T B (T , C ) can indeed become negative, whereas the actual thermodynamic temperature T always remains positive. Entropies of closed systems When interpreting thermodynamic data of new many-body states 7 , one of the first questions to be addressed is the choice of the appropriate thermostatistical ensemble 15,16 . Equivalence of the microcanonical and other statistical ensembles cannot—in fact, must not—be taken for granted for systems that are characterized by a non-monotonic 2,4,7 density of states (DOS) or that can undergo phase-transitions due to attractive interactions 17 —gravity being a prominent example 18 . Population-inverted systems are generally thermodynamically unstable when coupled to a (non-population- inverted) heat bath and, hence, must be prepared in isolation 5–7 . In ultracold quantum gases 7 that have been isolated from the environment to suppress decoherence, both particle number and energy are in good approximation conserved. Therefore, NATURE PHYSICS | VOL 10 | JANUARY 2014 | www.nature.com/naturephysics 67 © 2014 Macmillan Publishers Limited. All rights reserved.

Temperatura Negativa Fisica

Embed Size (px)

DESCRIPTION

Temperatura Negativa Fisica

Citation preview

  • ARTICLESPUBLISHED ONLINE: 8 DECEMBER 2013 | DOI: 10.1038/NPHYS2815

    Consistent thermostatistics forbids negativeabsolute temperaturesJrn Dunkel1* and Stefan Hilbert2

    Over the past 60 years, a considerable number of theories and experiments have claimed the existence of negative absolutetemperature in spin systems and ultracold quantum gases. This has led to speculation that ultracold gases may be dark-energyanalogues and also suggests the feasibility of heat engines with efficiencies larger than one. Here, we prove that allprevious negative temperature claims and their implications are invalid as they arise from the use of an entropy definitionthat is inconsistent both mathematically and thermodynamically. We show that the underlying conceptual deficiencies canbe overcome if one adopts a microcanonical entropy functional originally derived by Gibbs. The resulting thermodynamicframework is self-consistent and implies that absolute temperature remains positive even for systems with a boundedspectrum. In addition, we propose a minimal quantum thermometer that can be implemented with available experimentaltechniques.

    Positivity of absolute temperature T , a key postulate ofthermodynamics1, has repeatedly been challenged boththeoretically24 and experimentally57. If indeed realizable,negative temperature systems promise profound practical andconceptual consequences. They might not only facilitate thecreation of hyper-efficient heat engines24 but could also help7 toresolve the cosmological dark-energy puzzle8,9. Measurements ofnegative absolute temperature were first reported in 1951 by Purcelland Pound5 in seminal work on the population inversion in nuclearspin systems. Five years later, Ramsays comprehensive theoreticalstudy2 clarified hypothetical ramifications of negative temperaturestates, most notably the possibility to achieve Carnot efficiencies > 1 (refs 3,4). Recently, the first experimental realization of anultracold bosonic quantum gas7 with a bounded spectrum hasattracted considerable attention10 as another apparent examplesystem with T < 0, encouraging speculation that cold-atom gasescould serve as laboratory dark-energy analogues.

    Here, we show that claims of negative absolute temperaturein spin systems and quantum gases are generally invalid, as theyarise from the use of a popular yet inconsistent microcanonicalentropy definition attributed to Boltzmann11. By means of rigorousderivations12 and exactly solvable examples, we will demonstratethat the Boltzmann entropy, despite being advocated in mostmodern textbooks13, is incompatible with the differential structureof thermostatistics, fails to give sensible predictions for analyticallytractable quantum and classical systems, and violates equipartitionin the classical limit. The general mathematical incompatibilityimplies that it is logically inconsistent to insert negative Boltz-mann temperatures into standard thermodynamic relations, thusexplaining paradoxical (wrong) results for Carnot efficiencies andother observables. The deficiencies of the Boltzmann entropy canbe overcome by adopting a self-consistent entropy concept thatwas derived by Gibbs more than 100 years ago14, but has beenmostly forgotten ever since. Unlike the Boltzmann entropy, Gibbsentropy fulfils the fundamental thermostatistical relations andproduces sensible predictions for heat capacities and other ther-modynamic observables in all exactly computable test cases. The

    1Department of Mathematics, Massachusetts Institute of Technology, 77 Massachusetts Avenue E17-412, Cambridge, Massachusetts 02139-4307, USA,2Max Planck Institute for Astrophysics, Karl-Schwarzschild-Strae 1, Garching 85748, Germany. *e-mail: [email protected]

    Gibbs formalism yields strictly non-negative absolute temperatureseven for quantum systems with a bounded spectrum, therebyinvalidating all previous negative temperature claims.

    Negative absolute temperatures?The seemingly plausible standard argument in favour of negativeabsolute temperatures goes as follows10: assume a suitably designedmany-particle quantum system with a bounded spectrum5,7 canbe driven to a stable state of population inversion, so that mostparticles occupy high-energy one-particle levels. In this case, theone-particle energy distributionwill be an increasing function of theone-particle energy . To fit7,10 such a distributionwith a Boltzmannfactor exp(), must be negative, implying a negative Boltz-mann temperature TB = (kB)1 < 0. Although this reasoningmay indeed seem straightforward, the arguments below clarify thatTB is, in general, not the absolute thermodynamic temperature T ,unless one is willing to abandon the mathematical consistency ofthermostatistics. We shall prove that the parameter TB = (kB)1,as determined by Purcell and Pound5 and more recently also inref. 7 is, in fact, a function of both temperature T and heat capacityC . This function TB(T ,C) can indeed become negative, whereas theactual thermodynamic temperatureT always remains positive.

    Entropies of closed systemsWhen interpreting thermodynamic data of newmany-body states7,one of the first questions to be addressed is the choice of theappropriate thermostatistical ensemble15,16. Equivalence of themicrocanonical and other statistical ensembles cannotin fact,must notbe taken for granted for systems that are characterizedby a non-monotonic2,4,7 density of states (DOS) or that can undergophase-transitions due to attractive interactions17gravity being aprominent example18. Population-inverted systems are generallythermodynamically unstable when coupled to a (non-population-inverted) heat bath and, hence, must be prepared in isolation57.In ultracold quantum gases7 that have been isolated from theenvironment to suppress decoherence, both particle numberand energy are in good approximation conserved. Therefore,

    NATURE PHYSICS | VOL 10 | JANUARY 2014 | www.nature.com/naturephysics 67 2014 Macmillan Publishers Limited. All rights reserved.

    http://www.nature.com/doifinder/10.1038/nphys2815mailto:[email protected]://www.nature.com/naturephysics

  • ARTICLES NATURE PHYSICS DOI: 10.1038/NPHYS2815

    barring other physical or topological constraints, any ab initiothermostatistical treatment should start from the microcanonicalensemble. We will first prove that only the Gibbs entropy providesa consistent thermostatistical model for the microcanonical densityoperator. Instructive examples will be discussed subsequently.

    We consider a (quantum or classical) system with microscopicvariables governed by the Hamiltonian H = H ( ;V ,A),where V denotes volume and A = (A1, ...) summarizes otherexternal parameters. If the dynamics conserves the energy E , allthermostatistical information about the system is contained in themicrocanonical density operator

    ( ;E,V ,A)=(EH )

    (1)

    which is normalized by the DOS

    (E,V ,A)=Tr[(EH )]

    When considering quantum systems, we assume, as usual,that equation (1) has a well-defined operator interpretation, forexample, as a limit of an operator series. For classical systems, thetrace simply becomes a phase-space integral over . The average ofsome quantity F with respect to is denoted by FTr[F], andwe define the integrated DOS

    (E,V ,A)=Tr[(EH )]

    which is related to the DOS by differentiation,

    =

    E

    Intuitively, for a quantum system with spectrum {En}, the quantity(En,V ,A) counts the number of eigenstates with energy lessthan or equal to En.

    Given the microcanonical density operator from equation (1),one can find two competing definitions for the microcanonicalentropy in the literature1214,17,19,20:

    SB(E,V ,A)= kB ln(),

    SG(E,V ,A)= kB ln()

    where is a constant with dimensions of energy, required tomake the argument of the logarithm dimensionless. The Boltzmannentropy SB is advocated by most modern textbooks13 and usedby most authors nowadays2,4,5,7. The second candidate SG is oftenattributed toHertz21 butwas in fact already derived byGibbs in 1902(ref. 14, Chapter XIV). For this reason, we shall refer to SG as Gibbsentropy.Hertz proved in 1910 that SG is an adiabatic invariant21. Hiswork was highly commended by Planck22 and Einstein, who closeshis comment23 by stating that he would not have written some of hispapers had he been aware of Gibbs comprehensive treatise14.

    Thermostatistical consistency conditionsThe entropy S constitutes the fundamental thermodynamicpotential of the microcanonical ensemble. Given S, secondarythermodynamic observables, such as temperature T or pressure p,are obtained by differentiation with respect to the natural controlvariables {E,V ,A}. Denoting partial derivatives with respect toE by a prime, the two formal temperatures associated with SBand SG are given by

    TB(E,V ,A)=(SBE

    )1=

    1kB

    =

    1kB

    (2)

    TG(E,V ,A)=(SGE

    )1=

    1kB

    =

    1kB

    (3)

    Note that TB becomes negative if < 0, that is, if the DOS isnon-monotic, whereas TG is always non-negative, because isa monotonic function of E . The question as to whether TB orTG defines the thermodynamic absolute temperature T can bedecided unambiguously by considering the differential structure ofthermodynamics, which is encoded in the fundamental relation

    dS =(SE

    )dE+

    (SV

    )dV +

    i

    (SAi

    )dAi

    1TdE+

    pTdV +

    i

    aiTdAi (4)

    All consistent thermostatistical models, corresponding to pairs(,S) where is a density operator and S an entropy potential, mustsatisfy equation (4). If one abandons this requirement, any relationto thermodynamics is lost.

    Equation (4) imposes stringent constraints on possible entropycandidates. For example, for an adiabatic (that is, isentropic)volume change with dS= 0 and other parameters fixed (dAi = 0),one finds the consistency condition

    p=T(SV

    )=

    (EV

    )=

    HV

    (5)

    More generally, for any adiabatic variation of some parameterA {V ,Ai} of the Hamiltonian H , one must have (Supple-mentary Information)

    T(SA

    )E

    =

    (EA

    )S

    =

    HA

    (6)

    where T (S/E)1 and subscripts S and E indicate quantitieskept constant, respectively. The first equality in equation (6) followsdirectly from equation (4). The second equality demands correctidentification of thermodynamic quantities with statistical expec-tation values, guaranteeing for example that mechanically mea-sured gas pressure agrees with abstract thermodynamic pressure.The conditions in equation (6) not only ensure that the thermo-dynamic potential S fulfils the fundamental differential relation(equation (4)). For a given density operator , they can be used toseparate consistent entropy definitions from inconsistent ones.

    Using only the properties of themicrocanonical density operatoras defined in equation (1), one finds24

    TG

    (SGA

    )=

    1

    ATr[(EH )

    ]=

    1Tr[

    A(EH )

    ]= Tr

    [(HA

    )(EH )

    ]=

    HA

    (7)

    This proves that the pair (,SG) fulfils equation (6) and, hence,constitutes a consistent thermostatistical model for the mi-crocanonical density operator . Moreover, because generallyTB(SB/A) 6=TG(SG/A), it is a trivial corollary that the Boltz-mann entropy SB violates equation (6) and hence cannot be athermodynamic entropy, implying that it is inconsistent to insertthe Boltzmann temperature TB into equations of state or efficiencyformulae that assume validity of the fundamental thermodynamicrelations (equation (4)).

    Similarly to equation (7), it is straightforward to show that, forstandard classical Hamiltonian systems with confined trajectoriesand a finite ground-state energy, only the Gibbs temperature TGsatisfies themathematically rigorous equipartition theorem12

    iHj

    Tr

    [(iHj

    )

    ]= kBTG ij (8)

    68 NATURE PHYSICS | VOL 10 | JANUARY 2014 | www.nature.com/naturephysics 2014 Macmillan Publishers Limited. All rights reserved.

    http://www.nature.com/doifinder/10.1038/nphys2815http://www.nature.com/naturephysics

  • NATURE PHYSICS DOI: 10.1038/NPHYS2815 ARTICLESfor all canonical coordinates = (1,...). The key steps of the proofare identical to those in equation (7); that is, one merely exploitsthe chain rule relation (EH )/=(H/)(EH ), whichholds for any variable in the Hamiltonian H . Equation (8) isessentially a phase-space version of Stokes theorem12, relatinga surface (flux) integral on the energy shell to the enclosedphase-space volume.

    Small systemsDifferences between SB and SG are negligible for most macroscopicsystems with monotonic DOS , but can be significant for smallsystems12. This can already be seen for a classical ideal gas in d-spacedimensions, where17

    (E,V )=EdN/2V N , =(2m)dN/2

    N !hd0(dN/2+1)

    for N identical particles of mass m and Planck constant h.From this, one finds that only the Gibbs temperature yieldsexact equipartition

    E =(dN21)kBTB, (9)

    E =dN2

    kBTG (10)

    Clearly, equation (9) yields paradoxical results for dN = 1, where itpredicts negative temperature TB< 0 and heat capacity CB< 0, andalso for dN = 2, where the temperature TB must be infinite. This isa manifestation of the fact that SB is not an exact thermodynamicentropy. In contrast, the Gibbs entropy SG produces the reasonableequation (10), which is a special case of the more generalequipartition theorem (equation (8)).

    That SG also is themore appropriate choice for isolated quantumsystems, as relevant to the interpretation of the experimentsby Purcell and Pound5 and Braun et al.7, can be readilyillustrated by two other basic examples: for a simple harmonicoscillator with spectrum

    En= h(n+

    12

    ), n= 0,1,...,

    we find by inversion and analytic interpolation = 1+ n = 1/2+ E/(h) and, hence, from the Gibbs entropySG= kB ln the caloric equation of state

    kBTG=h2+E

    which, when combined with the quantum virial theorem, yields anequipartition-type statement for this particular example (equipar-tition is not a generic feature of quantum systems). Furthermore,T =TG gives a sensible prediction for the heat capacity,

    C =(TE

    )1= kB

    accounting for the fact that even a single oscillator can serve asminimal quantum heat reservoir. More precisely, the energy of aquantum oscillator can be changed by performing work througha variation of its frequency , or by injecting or removing energyquanta, corresponding to heat transfer in the thermodynamicpicture. The Gibbs entropy SG quantifies these processes in asensible manner. In contrast, the Boltzmann entropy SB= kB ln()with = (h)1 assigns the same constant entropy to all energystates, yielding the nonsensical result TB = for all energyeigenvalues En and making it impossible to compute the heat

    capacity of the oscillator. The failure of the Boltzmann entropy SBfor this basic example should raise doubts about its applicability tomore complex quantum systems7.

    That SB violates fundamental thermodynamic relations not onlyfor classical but also for quantum systems can be further illustratedby considering a quantum particle in a one-dimensional infinitesquare-well of length L, for which the spectral formula

    En= an2/L2, a= h2 2/(2m), n= 1,2,..., (11)

    implies = n = LE/a. In this case, the Gibbs entropy

    SG = kB ln gives

    kBTG= 2E, pGTG

    (SGL

    )=

    2EL

    as well as the heat capacity C = kB/2, in agreement withphysical intuition. In particular, the pressure equation is consistentwith condition (equation (5)), as can be seen by differentiatingequation (11) with respect to the volume L,

    pEL=

    2EL= pG

    That is, pG coincides with the mechanical pressure as obtainedfrom kinetic theory19.

    In contrast, we find from SB= kB ln() with = L/(2Ea) for

    the Boltzmann temperature

    kBTB=2E < 0

    Although this result in itself seems questionable, unless one believesthat a quantum particle in a one-dimensional box is a dark-energycandidate, it also implies a violation of equation (5), because

    pBTB

    (SBL

    )=

    2EL6= p

    This contradiction corroborates that SB cannot be the correctentropy for quantum systems.

    We still mention that one sometimes encounters the adhoc convention that, because the spectrum in equation (11) isnon-degenerate, the thermodynamic entropy should be zerofor all states. However, such a postulate entails several otherinconsistencies (Supplementary Information). Focusing on theexample at hand, the convention S = 0 would again imply thenonsensical result T =, misrepresenting the physical fact thatalso a single degree of freedom in a box-like confinement can storeheat in finite amounts.

    Measuring TB instead of TFor classical systems, the equipartition theorem (equation (8))implies that an isolated classical gas thermometer shows, strictlyspeaking, the Gibbs temperature T = TG, not TB. When broughtinto (weak) thermal contact with an otherwise isolated system, agas thermometer indicates the absolute temperature T of the com-pound system. In the quantumcase, theGibbs temperatureT can bedetermined with the help of a bosonic oscillator that is prepared inthe ground state and then weakly coupled to the quantum system ofinterest, because (kBT )1 is proportional to the probability that theoscillator has remained in the ground state after some equilibrationperiod (Methods). Thus, the Gibbs entropy provides not only theconsistent thermostatistical description of isolated systems but alsoa sound practical basis for classical and quantum thermometers.

    It remains to clarify why previous experiments5,7 measuredTB and not the absolute temperature T . The authors of ref. 7,

    NATURE PHYSICS | VOL 10 | JANUARY 2014 | www.nature.com/naturephysics 69 2014 Macmillan Publishers Limited. All rights reserved.

    http://www.nature.com/doifinder/10.1038/nphys2815http://www.nature.com/naturephysics

  • ARTICLES NATURE PHYSICS DOI: 10.1038/NPHYS2815

    for example, estimate temperature by fitting a quasi-exponentialBoseEinstein function to their experimentally obtained one-particle energy distributions10. Their system contains N 1particles with Hamiltonian HN and DOS N . The formally exactmicrocanonical one-particle density operator reads

    1=TrN1[N ] =TrN1[(EHN )]

    N(12)

    To obtain an exponential (canonical) fitting formula, as usedin the experiments, one first has to rewrite 1 in the equiva-lent form 1= exp[ln1]. Applying a standard steepest descentapproximation13,19 to the logarithm and assuming discrete one-particle levels E`, one finds for the relative occupancy p` of one-particle level E` the canonical form

    p`'eE`/(kBTB)

    Z, Z =

    `

    eE`/(kBTB) (13)

    The key observation here is that the exponential approxima-tion (equation (13)) features TB and not the absolute ther-modynamic Gibbs temperature T = TG. This becomes obvi-ous by writing equation (12) for a given one-particle energyE` as p` = N1(E E`)/N (E) = exp[lnN1(E E`)]/N (E)and expanding lnN1(E E`) for small E`, which gives p` exp[E`/(kBTB,N1)], where kBTB,N1 N1(E)/N1(E), inagreement with equation (2). That is, TB in equation (13) is actuallythe Boltzmann temperature of the (N1)-particle system.

    Hence, by fitting the one-particle distribution, one determinesthe Boltzmann temperature TB, which can be negative, whereas thethermodynamic Gibbs temperature T =TG is always non-negative.The formal definitions ofTG andTB imply the exact general relation(Supplementary Information)

    TB=TG

    1kB/C(14)

    where C = (TG/E)1 is the total thermodynamic heat capacityassociated with T =TG. As evident from equation (14), differencesbetween TG and TB become relevant only if |C | is close to or smallerthan kB; in particular, TB is negative if 0

  • NATURE PHYSICS DOI: 10.1038/NPHYS2815 ARTICLEStary Information). Such constraints lead to a strong effectivecoupling between the spin degrees of freedom, thereby invalidatingbasic assumptions in the derivation of canonical distributions,such as equation (13).

    In summary, for systems with a bounded spectrum, the effectiveBoltzmann temperature TB differs not only quantitatively butalso qualitatively from the actual thermodynamic temperatureT =TG> 0. Unfortunately, the measurement conventions adoptedby Braun et al.7, and similarly those by Purcell and Pound5, aredesigned to measure TB instead of TG.

    Carnot efficiencies > 1 ?The above arguments show that the Boltzmann entropy SB isnot a consistent thermodynamic entropy, neither for classical norfor quantum systems, whereas the Gibbs entropy SG providesa consistent thermodynamic formalism in the low-energy limit(small quantum systems), in the high-energy limit (classicalsystems) and in between. Regrettably, SB has become so widelyaccepted nowadays that, even when its application to exoticstates of matter7 leads to dubious claims, these are rarelyquestioned. One example are speculations2,4,7 that population-inverted systems can drive Carnot machines with efficiency>1. Toevaluate such statements, recall that a Carnot cycle, by definition,consists of four successive steps: isothermal expansion; isentropicexpansion; isothermal compression; isentropic compression. Thetwo isothermal steps require a hot and cold bath with temperaturesTH and TC, respectively, and the two isentropic steps canbe thought of as place-holders for other work-like parametervariations (changes of external magnetic fields, and so on). Theassociated Carnot efficiency

    = 1TC

    TH(15)

    owes its popularity to the fact that it presents an upper bound forother heat engines19. To realize values > 1, one requires either TCor TH to be negative. At least formally, this seems to be achievableby considering systems as in Fig. 1 and naively inserting positive andnegative Boltzmann temperature valuesTB0 into equation (15).

    Speculations2,4,7 of this type are unsubstantiated for severalreasons. First, TB is not a consistent thermodynamic temperature,and, if at all, one should use the absolute temperature T =TG>0 inequation (15), which immediately forbids > 1. Second, to changeback and forth between population-inverted states with TB < 0and non-inverted states with TB > 0, work must be performednon-adiabatically26, for example, by rapidly switching a magneticfield. As the thermodynamic entropy is not conserved duringsuch switching processes, the resulting cycle is not of the Carnottype anymore and requires careful energy balance calculations3.In particular, such an analysis has to account for the peculiar factthat, when the heat engine is capable of undergoing populationinversion, both a hot and cold bath may inject heat into the system.Properly defined efficiencies of thermodynamic cycles that involvesystemswith lower and upper energy bounds are, in general, not justsimple functions ofTG orTB. Naive application of equation (15) canbe severely misleading in those cases.

    On a final note, groundbreaking experiments such as those byPurcell and Pound5 and Braun et al.7 are essential for verifying theconceptual foundations of thermodynamics and thermostatistics.Such studies disclose previously unexplored regimes, therebyenabling us to test and, where necessary, expand theoreticalconcepts that will allow us to make predictions and are essentialfor the development of new technologies. However, the correctinterpretation of data and the consistent formulation of heatand work exchange15 under extreme physical conditions (forexample, at ultracold or ultrahot27 temperatures, or on atomic orastronomical scales) require special care when it comes to applying

    the definitions and conventions that constitute a specific theoreticalframework. When interpreted within a consistent thermostatisticaltheory, as developed by Gibbs14 more than a century ago, neitherthe work of Purcell and Pound5 nor recent experiments7 provideevidence for negative absolute temperatures. Unfortunately, thisalsomeans that cold atomgases are less likely tomimic dark energy.

    MethodsMinimal quantum thermometer. A simple quantum thermometer for measuringthe thermodynamic Gibbs temperature T =TG can be realized with a heavy atomin a one-dimensional harmonic trap. The measurement protocol is as follows:before coupling thermometer and system, one must prepare the isolated system ina state with well-defined energy E = ES and the thermometer oscillator with smallangular frequency in the ground state ET= h/2. After coupling the thermometerto the system, the total energy remains conserved, but redistribution of energymay take place. A measurement of the thermometer energy after a sufficiently longequilibration period will produce an oscillator eigenvalue E T= h(n

    +1/2), where

    n {0,...,b(EE0)/(h)c}, with E0 denoting the systems ground state and bxc theinteger part. If the total energy remains conserved and the thermometer oscillatoris non-degenerate, the probability P[E T|E] of measuring a specific oscillator energyE T is equal to the microcanonical probability of finding the system in a stateE S = E (E

    TET) E :

    P[E T|E] =g (E+ETE T)

    (E)

    where g (E S) is the degeneracy of the level E

    S of the system, and

    (E)=E SE

    g (E S)

    Assuming that the energy levels lie sufficiently dense ( 0) we can approximatethe discrete probabilities P[E T|E]'p(E

    T|E)dE

    T by the probability density

    p(E T|E)=(E+ETE T)

    (E)

    This distribution can be obtained by repeating the experiment many times, and asimple estimator for the (inverse) absolute temperatureT >0 is (equation (3))

    1kBT=(E)(E)

    = p(ET|E) (16)

    In practice, one would measure p(E T|E) for E

    T > ET = h/2 and extrapolate toE T= ET. The thermometer equation (16) is applicable to systems with and withoutpopulation inversion. The precision of this minimal thermometer is set by theoscillator frequency and the number of measurements.

    Received 10 July 2013; accepted 18 October 2013; published online8 December 2013

    References1. Callen, H. B. Thermodynamics and an Introduction to Thermostatics

    (Wiley, 1985).2. Ramsay, N. F. Thermodynamics and statistical mechanics at negative absolute

    temperatures. Phys. Rev. 103, 2028 (1956).3. Landsberg, P. T. Heat engines and heat pumps at positive and negative absolute

    temperature. J. Phys. A 10, 17731780 (1977).4. Rapp, A., Mandt, S. & Rosch, A. Equilibration rates and negative absolute

    temperatures for ultracold atoms in optical lattices. Phys. Rev. Lett. 105,220405 (2010).

    5. Purcell, E. M. & Pound, R. V. A nuclear spin system at negative temperature.Phys. Rev. 81, 279280 (1951).

    6. Hakonen, P. & Lounasmaa, O. V. Negative absolute temperaturehot spins inspontaneous magnetic order. Science 265, 18211825 (1994).

    7. Braun, S. et al. Negative absolute temperature for motional degrees of freedom.Science 339, 5255 (2013).

    8. Peebles, P. J. & Ratra, B. The cosmological constant and dark energy.Rev. Mod. Phys. 75, 559606 (2003).

    9. Loeb, A. Thinking outside the simulation box. Nature Phys. 9, 384386 (2013).10. Carr, L. D. Negative temperatures? Science 339, 4243 (2013).11. Sommerfeld, A.Vorlesungen ber Theoretische Physik (Band 5): Thermodynamik

    und Statistik 181183 (Verlag Harri Deutsch, 2011).12. Khinchin, A. I.Mathematical Foundations of Statistical Mechanics

    (Dover, 1949).13. Huang, K. Statistical Mechanics 2nd edn (Wiley, 1987).14. Gibbs, J. W. Elementary Principles in Statistical Mechanics (Dover, 1960)

    (Reprint of the 1902 edition).

    NATURE PHYSICS | VOL 10 | JANUARY 2014 | www.nature.com/naturephysics 71 2014 Macmillan Publishers Limited. All rights reserved.

    http://www.nature.com/doifinder/10.1038/nphys2815http://www.nature.com/naturephysics

  • ARTICLES NATURE PHYSICS DOI: 10.1038/NPHYS2815

    15. Campisi, M., Talkner, P. & Hnggi, P. Fluctuation theorem for arbitrary openquantum systems. Phys. Rev. Lett. 102, 210401 (2009).

    16. Campisi, M. & Kobe, D. Derivation of the Boltzmann principle. Am. J. Phys.78, 608615 (2010).

    17. Dunkel, J. & Hilbert, S. Phase transitions in small systems: Microcanonical vs.canonical ensembles. Physica A 370, 390406 (2006).

    18. Votyakov, E V., Hidmi, H. I., De Martino, A. & Gross, D. H. E. Microcanonicalmean-field thermodynamics of self-gravitating and rotating systems.Phys. Rev. Lett. 89, 031101 (2002).

    19. Becker, R. Theory of Heat (Springer, 1967).20. Campisi, M. Thermodynamics with generalized ensembles: The class of dual

    orthodes. Physica A 385, 501517 (2007).21. Hertz, P. ber die mechanischen Grundlagen der Thermodynamik. Ann. Phys.

    (Leipz.) 33 225274; 537552 (1910).22. Hoffmann, D. ... you cant say anyone to their face: your paper is rubbish. Max

    Planck as Editor of Annalen der Physik. Ann. Phys. (Berlin) 17, 273301 (2008).23. Einstein, A. Bemerkungen zu den P. Hertzschen Arbeiten:ber die

    mechanischen Grundlagen der Thermodynamik. Ann. Phys. (Leipz.)34, 175176 (1911).

    24. Campisi, M. On the mechanical foundations of thermodynamics: Thegeneralized Helmholtz theorem. Stud. Hist. Philos. Mod. Phys. 36,275290 (2005).

    25. Stanley, R. P. Enumerative Combinatorics 2nd edn, Vol. 1 (Cambridge Studiesin Advanced Mathematics, Cambridge Univ. Press, 2000).

    26. Tremblay, A-M. Comment on Negative Kelvin temperatures: Some anomaliesand a speculation. Am. J. Phys. 44, 994995 (1975).

    27. Dunkel, J., Hnggi, P. & Hilbert, S. Nonlocal observables and lightconeaveraging in relativistic thermodynamics. Nature Phys. 5, 741747 (2009).

    AcknowledgementsWe thank I. Bloch, W. Hofstetter and U. Schneider for constructive discussions. We aregrateful to M. Campisi for pointing out equation (14), and to P. Kopietz, P. Talkner,R. E. Goldstein and, in particular, P. Hnggi for helpful comments.

    Author contributionsAll authors contributed to all aspects of this work.

    Additional informationSupplementary information is available in the online version of the paper. Reprints andpermissions information is available online at www.nature.com/reprints. Correspondenceand requests for materials should be addressed to J.D.

    Competing financial interestsThe authors declare no competing financial interests.

    72 NATURE PHYSICS | VOL 10 | JANUARY 2014 | www.nature.com/naturephysics 2014 Macmillan Publishers Limited. All rights reserved.

    http://www.nature.com/doifinder/10.1038/nphys2815http://www.nature.com/doifinder/10.1038/nphys2815http://www.nature.com/reprintshttp://www.nature.com/naturephysics

    Consistent thermostatistics forbids negative absolute temperaturesNegative absolute temperatures?Entropies of closed systemsThermostatistical consistency conditionsSmall systems Measuring instead of TQuantum systems with a bounded spectrumCarnot efficiencies >1 ?MethodsMinimal quantum thermometer.

    Figure 1 Non-negativity of the absolute temperature in quantum systems with a bounded spectrum.ReferencesAcknowledgementsAuthor contributionsAdditional informationCompeting financial interests