Thermal Radiation Effect in the Free Expansion of an Ideal

Gas and Gibbs’ Paradox in Classical Thermodynamics

A. Paglietti

Department of Structural Engineering

University of Cagliari, 09123 Cagliari, Italy
e-mail: paglietti@unica.it

Abstract: The standard theory of ideal gases ignores the interaction of the
gas particles with the thermal radiation (photon gas) that fills the
otherwise vacuum space between them. This is an unphysical feature of
the theory since every material in this universe, and hence also the
particles of a gas, absorbs and radiates thermal energy. The interaction
with the thermal radiation that is contained within the volume of the body
may be important in gases since the latter, unlike solids and liquids, are
capable of undergoing conspicuous volume changes. Taking this
interaction into account makes the behaviour of the ideal gases more
realistic and removes Gibbs’ paradox.

1. Introduction

A rigid vessel is composed of two equal chambers separated by a removable

partition. The two chambers are filled with two different ideal gases at the same
initial temperature T0. Let n1 and n2 be the number of moles of gas contained in the
first and in the second chamber, respectively. The gases in the vessel cannot
exchange heat with the surroundings as the vessel walls are adiabatic. As the
partition is removed, which can be done without performing any work on the gases,
the two gases diffuse through each other. This is a spontaneous irreversible process
and we are interested in determining the entropy increase it produces.
The problem is a standard one in Classical Thermodynamics. Its solution can be
found in many textbooks. In particular, the excellent treatise by Fast [1, pp. 41-43]
contains a clear approach to this problem, entirely within the realm of Classical
Thermodynamics. Being ideal, the two gases do not interact with each other during
the mixing process. This means, in particular, that as the partition is removed each of
them expands into the volume occupied by the other, as if it expanded in a vacuum.
Furthermore, since the free expansion of an ideal gas in a vacuum does not produce
any temperature change, the final temperature of the gas mixture must be the same as
the initial temperature T0. The entropy change ∆S due to the considered expansion
process is obtained directly from the classical expression of the entropy of an ideal
gas:

S = n (cv ln T + R ln v) + const. (1.1)

Here cv is the molar specific heat at constant volume, n the number of moles of the
gas, R the universal gas constant, T the absolute temperature, while v = V/n is the
volume V per mole. It is thus found that

V1 + V2 V +V
∆S = R (n1 ln + n2 ln 1 2 ). (1.2)
V1 V2

where V1 and V2 are the volumes of the two chambers. In the present case we have V1
= V2 = V, which simplifies eq. (1.2) to:

∆S = R (n1 + n2) ln 2. (1.3)

The so-called Gibbs’ paradox arises because the above formulae are independent
of the physicochemical nature of the two gases. This seems hardly acceptable.
Surely, the entropy change resulting from mixing together say one mole of helium
and one mole of ammonia should be different than that resulting from mixing two
similar amounts of two different isotopes of oxygen. In other words a change in the
properties of the mixing gases should produce a change in the entropy increase due
to the mixing, which is denied by the above formulae.
The same formulae apply in particular when n1 = n2 = 1. In this case eq. (1.3)
simplifies to

∆S = 2 R ln 2 >0. (1.4)

Now, if the two ideal gases are the same, the gas pressure will be the same in the two
chambers because they both contain one mole of the gas at the same volume and the
same temperature. In this case too the theory predicts the same entropy change (1.4)
following the partition removal. No entropy increase should however occur, simply
because no change in the state of the two gases takes place as the partition is
removed. This inconsistency in itself is often referred to as the Gibbs’ paradox.
Actually, it is a further consequence of the fact that eqs (1.2) and (1.3) do not depend
on the properties of the mixing gases.

2
 The existence of Gibbs’ paradox casts some shadows on the classical theory of
ideal gases. It suggests that that theory may somehow be flawed even in the range of
pressures and temperatures where it turns out to provide an otherwise superb
approximation of the behaviour of the real gases.
The same paradox is met when the theory of ideal gases is approached by the
methods of Statistical Mechanics. In that case the paradox is traditionally resolved in
various ways. The most convincing one introduces the notion of indistinguishability
of the various molecules of the same gas when calculating the number of the
available micro-states. Another, though more summary way to fix the paradox is to
dismiss as meaningless any calculation about the mixing of two identical gases, on
the ground that in this case there are not actually two different gases to mix. Both
ways only remove the paradox in the case in which the mixing gases are the same,
though. When it comes to two different gases, also the Statistical Mechanics
approach predicts that the entropy change due to their mixing should not depend on
the nature of the gases. A way out of this shortcoming has been proposed by Lin. For
a recent commentary of his proposal and reference to the original papers the reader is
referred to [2]. Lin’s approach, however, introduces the concept of information
entropy, which is foreign to Classical Thermodynamics.
In looking for a solution to this problem, one cannot help but observe that
Classical Thermodynamics and Statistical Mechanics are essentially two different
approaches, resting on entirely different bases. Each approach should therefore be
consistent within its own framework. It is only a matter of scientific rigour then, that
we should not use one approach to justify the other. For this reason, in the following
sections we shall attempt to resolve the various aspects of Gibbs’ paradox within the
realm of Classical Thermodynamics.
More precisely, we shall show that the origin of the paradox lies in the fact that
the notion of ideal gas, as introduced in Classical Thermodynamics or, for that
matter, also in Statistical Mechanics, cannot be entirely realistic. This is so because it
represents a material that does not radiate any energy. Every material in this universe
radiates energy, depending on its temperature. This fact is ignored by the theory of
the ideal gases. The latter are imagined as being made of volumeless particles,
endowed with mass, unable to interact at distance with each other and with the
radiation that always fills the very space in which they are moving about. As a matter
of fact, that space will exchange energy with them through thermal radiation until
thermal equilibrium is reached. Taking into account this phenomenon will make the
ideal gas more physical and also remove Gibbs’ paradox.
A similar interaction can be neglected in liquids and solids, since they undergo
only minor volume changes.

3
2. The photon gas that fills the vacuum amid the gas particles

Any cavity or otherwise empty space harbours electromagnetic radiation within it.
This radiation is often referred to as thermal radiation or photon gas. The latter
terminology is particularly appropriate because from the macroscopic standpoint
such a radiation behaves in many ways as a gas. In this section the main formulae
concerning the photon gas are recalled, as can be found in many books of Classical
Thermodynamics (see e.g. [1, pp.158-162] or [3, Sect. 13.16]). They apply in thermal
equilibrium conditions, that is when the walls of the cavity and/or any material
particle within it have reached the same equilibrium temperature. The latter will be
referred to as the temperature of the photon gas itself. Its value in the absolute
temperature scale will be denoted by T, as usual.
In thermal equilibrium conditions, a unit volume of space filled with a photon gas
stores the energy u given by

u = aT 4 . (2.1)

The constant a appearing here can be expressed as

4
a= σ, (2.2)
c

where c is the velocity of light while σ is the well-known Stefan-Boltzman constant

[σ = 56.697 nW/m2 deg4, which makes a=75.646 10−8 nJ/m3deg4 since c=2.998 108
m/sec]. The internal energy U of a volume V of photon gas is accordingly given by

U = aT 4 V . (2.3)

Under the same conditions, the photon gas exerts a pressure P on the walls of the
cavity, the value of which is

1 1
P= u= aT 4 . (2.4)
3 3

This pressure does work as the volume expands or contracts. The work supplied by
the photon gas to the cavity walls as its volume is increased by dV turns out to be

1 1
dWout = P dV = u dV = aT 4 dV . (2.5)
3 3

Thus if Q denotes the amount of heat absorbed by the photon gas, we can apply the
1st principle of Thermodynamics to state that

4
 dU = dQ − dWout . (2.6)

From this and from eqs (2.3) and (2.4) we get

4
dQ = 4 aT 3 V dT + aT 4 dV . (2.7)
3

By applying this equation to a reversible process we obtain

4
dS = 4 aT 2 V dT + aT 3 dV , (2.8)
3
since dQ = TdS for reversible processes. The total differential equation (2.8) can
easily be integrated to give the following expression for the entropy S of the photon
gas:

4
S= aT 3V , (2.9)
3

the integration constant being set equal to zero, since S = 0 for T = 0.

Finally, from eq. (2.7) specific (per unit volume) thermal capacity rv of the photon
gas at constant volume is immediately obtained:

1 dQ
rv = = 4 aT 3 . (2.10)
V dT

3. Influence of thermal radiation on the adiabatic free expansion of an ideal gas

According to classical theory, a free adiabatic expansion of an ideal gas should

leave its temperature unaltered. This follows from the fact that the particles of the
gas are free from long-range interactions, which means that they can only store
kinetic energy. In a free expansion no external work is done by the gas, since it
expands in a vacuum. No thermal energy is exchanged either, as the expansion is
adiabatic. The kinetic energy of the gas is therefore conserved, which in particular
means that the gas temperature in the final equilibrium state after the expansion must
be the same as the initial one.
The above explanation does not take into account that no material is physically
admissible if it does not emit and absorb thermal radiation. The ideal gas is lacking
in this respect, which may lead to some physical inconsistencies in its behaviour. In
the present section we shall show how the interaction with the ubiquitous photon gas,

5
which cannot be avoided by any real gas –not even in the ranges of temperatures and
pressures where its state equations coincide with those of an ideal gas– makes the
gas cool as a result of a free adiabatic expansion.
Let us first of all determine the amount of heat that a photon gas absorbs from the
surroundings as it expands at constant temperature. If ∆V is the volume increase of
the gas, the amount of heat needed to expand it isothermally is given by

4
∆Q = aT 4 ∆V . (3.1)
3

This immediately follows from eq. (2.7) once we set dT=0 and integrate the resulting
equation between the initial volume V and the final one V + ∆V.
If this amount of heat is not supplied, which is the case when the expansion takes
place adiabatically, then the volume change ∆V will produce a reduction in the
photon gas temperature. On the other hand, if the cavity walls are adiabatic but the
cavity itself also contains a gas, then the amount of heat ∆Q will be taken from the
particles of the gas, which will cool down as a result. In this case, the temperature
change can to a good approximation be calculated by assuming that the thermal
capacity of the photon gas is negligible with respect to that of the ideal gas within the
cavity. Keeping in mind that the molar specific heat cv of an ideal gas does not
depend on volume and referring to the case in which the cavity contains just one
mole of ideal gas, the temperature variation due to the expansion of the latter will be
given by

∆Q 4 aT 4 ∆V
∆T = − =− . (3.2)
cv 3 cv

If there are n moles of gas in the cavity, this temperature variation should be divided
by n. In writing eq. (3.2) we assumed that ∆T was small enough as to produce a
negligible change in cv during the expansion. This is certainly so in the vast majority
of cases. In any case, taking account of the dependence of cv on T does not appear to
pose any serious problem.
The entropy change of the system due to the considered expansion is the sum of
the contribution ∆Sg coming from the ideal gas and the contribution ∆Sr of the
photon gas. The former is readily obtained from eq. (1.2) and is given by

T + ∆T V + ∆V
∆Sg = cv ln + R ln (3.3)
T V

for each mole of gas. The other contribution is calculated from eq. (2.8) to be given
by:

6
 4
∆Sr = 4 aT 2 V ∆T + aT 3 ∆V . (3.4)
3
If the volume V is not too large, ∆Sr can be neglected with respect to ∆Sg. Moreover,
for small values of ∆T as the one involved in the present case, the first term in the
right-hand side of eq. (3.4) can be neglected too, as long as T is sufficiently far from
zero. Under these conditions we can, to a good approximation, calculate the total
entropy change of the system as

V + ∆V
∆S = R ln . (3.5)
V

In conclusion, we can say that as far as the considered process is concerned the
entropy change of the system is not essentially affected by the presence of the photon
gas. The latter, however, does produce a cooling effect in the ideal gas temperature
resulting after the expansion. The measure of this effect is given by eq. (3.2).

4. The entropy change of the two gases ensuing from the partition removal

Let us now refer to the two gases filling the two separate chambers of the
adiabatic vessel considered in Sect. 1. As the partition is removed, the two gasses
expand by flowing one through the other. Each gas expands as if it were in a
vacuum, since the two gases are supposed to be ideal. This enables us to calculate the
entropy change of the system of the two gases by referring to the following sequence
of processes:

(1) A free adiabatic expansion of each separate gas, bringing them to the same final
volume 2V. In this process, the gas does not supply/absorb any work to/from the
surroundings since the process can be assimilated to a free expansion in a
vacuum. The same process will produce a change in the entropy and temperature
of each expanding gas, though. Moreover, different gases will suffer different
temperature changes.

(2) A transfer of heat between the two expanded gases bringing them to the same
final temperature T*. Being a heat transfer from a hotter gas to a colder one, this
process will result in a further entropy change of the system. In the process, no
heat is exchanged with the surroundings as the vessel walls are adiabatic.

(3) Finally, the two gases are allowed to mix reversibly at the constant volume 2V
they reached at the end of phase (1). A process of this kind is admissible from
the physical standpoint and was first conceived by Plank [4, Sect. 236] (see also

7
 [1, p. 41] and [5, p. 729]). Being adiabatic and reversible, the process will not
produce any entropy change. It will not produce any change in the temperature
of the gases either, since the latter are ideal and the process does not bring about
any volume change. The system will thus reach the final state in which the two
gases are mixed together at temperature T* and volume 2V.

The details of each of the above three phases of the process are worked out below.

Phase 1. (Free adiabatic expansion)

One mole of an ideal gas fills the volume V of one chamber of the vessel. Let T0
be its initial temperature and cv′ its molar specific heat at constant volume. The gas
expands adiabatically in a vacuum until it occupies the volume 2V of the entire
vessel, thus increasing its volume by ∆V=V. If the presence of heat radiation is taken
into account, the equilibrium temperature T1 that results after the expansion can be
calculated from eq. (3.2) to be given by:

4 aT0 4 ∆V
T1 = T0 − . (4.1)
3 cv′

A similar adiabatic expansion of the gas filling the other chamber will bring it to the
final volume 2V and to a final equilibrium temperature T2 given by:

4 aT0 4 ∆V
T2 = T0 − , (4.2)
3 cv′′

where cv′′ is the molar specific heat at constant volume of this new gas.
In order to calculate the entropy change brought about by this phase of the
process, we first observe that if ∆V is not too large, both T1 and T2 will be
sufficiently near to T0 as to allow us to apply the approximate equation (3.5). This
process will therefore increase the entropy of each gas by the amount

∆e S = R ln 2. (4.3)

This will produce an overall increase in the entropy of the two gases by the amount

∆1S = 2 ∆e S = 2 R ln 2. (4.4)
It is apparent that such an entropy change coincides with that which would be
expected for an ideal gas in the absence of the photon gas. The presence of the latter,
however, will produce a cooling effect that is proportional to ∆V resulting from the
considered expansion, as predicted by eqs (4.1)-(4.2). Small as this effect may be (it
is proportional to ∆V, though!), it cannot be avoided not even by an ideal gas, since

8
every material must absorb and radiate thermal energy. Most importantly, the
amount of cooling depends on the specific heat of the gas, so that it will generally be
different for different ideal gases.

Phase 2. (Internal heat transfer)

The previous process brings the two gases to states (T1, 2V) and (T2, 2V),
respectively. To be definite, we shall assume that cv′ > cv′′ , which in view of eqs (4.1)
and (4.2) implies that T1 > T2. We now put the two gases in thermal contact with
each other while keeping them thermally insulated from the surroundings. In these
conditions the hotter gas will supply heat to the colder one until both gases reach the
same equilibrium temperature T*. The determination of T* is an elementary problem
of heat transfer. It is solved by equating the total amount of heat Q1* lost by the
hotter gas to the amount of heat Q2* absorbed by the colder one. Since we are
considering one mole of each gas, we have

Q1* = (T1 − T *) cv′ (4.5)

and

Q 2* = (T * − T2 ) cv′′ . (4.6)

We therefore get

T1 cv′ + T2 cv′′
T* = . (4.7)
cv′ + cv′′

Once the amounts of heat (4.5) and (4.6) are known, the entropy change caused by
the process can also be obtained. Since the temperatures T1, T2 and T* are close
together, we shall not introduce any serious mistake if we assume that the heat Q1* is
lost at the constant temperature T1 and that the heat Q2* is absorbed at constant
temperature T2. The entropy change due to the heat transfer is, accordingly:

Q1* Q 2* cv′ cv′′ (T1 − T2 ) 2

∆2S = − + = , (4.8)
T1 T2 cv′ + cv′′ T1 T2

which is clearly greater than zero. The important point to be noted here is that this
entropy change depends on the specific heats of the mixing gases. Such a
dependence is both explicit, as shown by equation (4.8), and implicit through T1 and
T2 via equations (4.1) and (4.2). In the case in which cv′ = cv′′ , the latter equations
yield T1 = T2, which makes ∆2S = 0. In particular, ∆2S vanishes if the two chambers
are filled with the same gas.

9
Phase 3. (Isothermal reversible mixing)
As previously observed, in this phase of the process the two perfect gases do not
suffer any change in entropy or temperature since they undergo a reversible
isothermal mixing. The presence of the photon gas does not alter this conclusion,
since this phase of the process takes place at a constant volume.

From eqs (4.4) and (4.8) the entropy change due to the whole process turns out to
be, therefore:

cv′ cv′′ (T1 − T2 )2

∆S = ∆1S + ∆2S = 2 R ln 2 + . (4.9)
cv′ + cv′′ T1 T2

The first term in the far right-hand side of this equation is due to the volume
expansion and has the same value no matter the ideal gases under consideration. The
last term attains different values for different mixing gases.
Incidentally, such a result helps to define precisely when two ideal gases are to be
treated as the same or as different as far as the thermodynamics of their mixing is
concerned. It shows that it all depends on whether they have the same specific heat
or not. Other differences in their properties are not relevant from the
thermodynamical standpoint. This fact should be taken into due account when a
Statistical Mechanics approach to this phenomenon is sought. It implies that some
otherwise different microstates of a gas mixture should be considered to be the same
if they are relevant to two different ideal gases which possess the same specific heat.

5. The solution of Gibb’s paradox

According to the analysis of the previous section, there are two contributions to
the entropy change ∆S following the removal of the partition between the two ideal
gases contained in the adiabatic vessel we considered in the Introduction. The first
contribution is ∆1S and is relevant to what we called Phase 1of the process. It is due
to the separate expansion of each gas and is practically independent of their
particular physicochemical properties [cf. eq. (4.3)]. The crucial point to be noted
here is, however, that if account of thermal radiation is duly taken, the final
temperature resulting from this phase of the process does depend on the properties of
the expanding gas; more precisely, on its specific heat. This means that it is different
for different gases [cf. eqs. (4.1) and (4.2)].
The other contribution to the entropy change, namely ∆2S, results from Phase 2.
At a variance with ∆1S, this contribution depends on the specific heats cv′ and cv′′ , of
the two gases. Moreover, as the difference between cv′ and cv′′ tends to zero, so does
∆2S. As already observed, this shows that as far as this phase of the process is

10
concerned, a quantitative measure of the difference between the two gases is
provided by the difference in their specific heats. Any property of the gases other
than their specific heats has no effect on the entropy change resulting from this part
of the process.
From the above results we can conclude that the total entropy change ∆S depends
on the difference in the specific heat of the gases, since ∆S = ∆1S + ∆2S. This
resolves the main part of Gibbs’ paradox, which objected to having the same entropy
change no matter the physicochemical properties of the two gases. In fact this is not
so, if the effect of thermal radiation is appropriately accounted for. The established
formulae also show that small variations in the difference of the specific heats of the
two gases produce small variations in the predicted value of ∆S, which complete the
answer to the main part of the paradox.
What remains to be considered is the part of Gibbs’ paradox that refers to the
particular case in which the gases in the two chambers are the same and are at the
same initial pressure and temperature. In this case the two gases will not expand as
the partition is removed. This is because the number of gas particles that go from
chamber 1 to chamber 2 after the partition removal will be equal to the number of
gas particles that go from chamber 2 to chamber 1, the two gases being in thermal
equilibrium with each other. Since the particles of the two gases are identical, the net
effect of this particle exchange will be the same as if the partition was not removed at
all. It is obvious then that in this case the partition removal cannot produce any
entropy change at all, which answers the same-gas part of the paradox.
One may still wonder, however, why should the entropy change vanish in the case
of two identical gases, whilst the slightest difference in their specific heat would
produce the finite entropy jump ∆S = ∆1S + ∆2S. There is no inconsistency here,
either. The point is that the volumetric contribution ∆1S always applies in the case of
two different gases −no mater how little their specific heats differ from each other, as
long as they are not the same. The same contribution, however, is altogether absent
in the case of two identical gases, since, as observed above, no expansion can take
place in that case. It should be obvious then that one should not compare the entropy
change due to a process that includes an expansion, with that of another process that
does not.
We can also observe that if we confine our attention to the contribution ∆2S,
which is not due to the expansion, then the entropy change relevant to the case of
two different gases will tend to zero as cv′ and cv′′ tend to the same value. This is
perfectly consistent with the fact that ∆S=0 in the limit case in which the two gases
are identical and answers the same-gas part of Gibbs’ paradox concerning the alleged
entropy jump inconsistency too.
It might still be objected that if we put a red gas in chamber 1 and a white gas in
chamber 2 we would get a pink mixture once the partition is removed, even if the
two gases are “the same” in that cv′ = cv′′ . This could be seen as the evidence that the

11
entropy of the system would increase, contrary to what we concluded above. Such a
change in colour, however, is not a thermodynamic process. “Redness” and
“whiteness” are not state variables of the system. Nor do they enter the state
equations of the gases. As a consequence, no internal energy or entropy change can
be produced in the system by the colour change of the two gases. No work or heat is
absorbed in the process either. The gas colour change process that would certainly
take place in the considered situation is a purely mechanical process; a consequence
of the disordered distribution of the velocities of the gas particles. It also could be
regarded as a demonstration of the thermal agitation of the gas particles, which is
always active in any gas, even in thermal equilibrium, provided that T≠0.
But, is the pink state of the two gases more disordered than the initial red and
white one? Perhaps. In some sense at least. However, as remarked in [6], disorder
and macroscopic entropy are not always related.

References

1. J.D. Fast: Entropy. Macmillan, London 1970.

2. S.-K. Lin: Gibbs Paradox and the Concepts of Information, Symmetry,
Similarity and Their Relationship. Entropy 10, 1-5, (2008).
3. M.W. Zemansky: Heat and Thermodynamics, 5th ed. McGraw-Hill, Tokyo 1968.
4. M. Plank: Treatise on Thermodynamics. Dover Publications, New York 1945.
5. A. Ben-Naim: Is mixing a thermodynamic process? Am. J. Phys. 55, 725-733
(1987).
6. A. Ben-Naim: On the so-called Gibbs’ paradox, and on the real paradox.
Entropy 9, 132-136. (2007).

12