Gibbs Paradox

Consider the following scenario—suppose we have a box divided into two equal halves.
The entire box is thermally isolated, but the partition separating the two parts allows the
exchange of energy. In the left half, we place helium gas; in the right half, an equal
quantity of argon gas.

helium argon

N1, V1 N2,V2

Now suppose we remove the barrier. Both gases will diffuse to fill the entire container;
and intuitively, it seems reasonable to expect the total entropy of the system to increase.
It’s not so hard to calculate the increase in entropy. If we use the microcanonical
ensemble, we have
B
V N ( 2mE )
3N / 2
Ω ( E ,V , N ) = 3N
N !h
We will absorb the mass term into B, and so obtain
B
Ω ( E ,V , N ) = 3N
V N E 3N / 2
N !h
The entropy is given by
S = k ln Ω( E ,V , N )
 B  3Nk
= k ln  3N 
+ Nk ln V + ln E
 N !h  2
3Nk
= const + Nk ln V + ln E
2
Notice that we could get the same result from purely thermodynamic reasoning—from
the first TdS equation, for example.
We pretend we can neglect the “constant” term involving N ! Then, assume that the final
volume for both gases is just twice the initial volume (that is, V1 = V2 = V , and hence
V1 f = V2 f = V f = 2V ). We also assume that the energy remains constant. Under these
circumstances,

1
 ∆S = Sfinal − Sinitial
= N1k ln V1 f + N 2 k ln V2 f − N1k ln V1i − N 2 k ln V2i
= Nk ln 2V + Nk ln 2V − Nk ln V − Nk ln V
= Nk ln 2 + Nk ln 2
= 2 Nk ln 2
This result seems reasonable. And in fact, it is identical to the result we found earlier
when we calculated the entropy change involved in an adiabatic free expansion, if we
assume that V f = 2Vi . (Recall that we calculated the entropy by doing a quasi-static
compression to return the gas to its original state.)
Now, suppose we have exactly the same process, except that both halves of the insulated
container initially contain helium. For this case, one expects intuitively that ∆S = 0 . But
when we do the calculation, we find
∆S = Sfinal − Sinitial
= (2 N )k ln V f − 2 ( Nk ln Vi )
= (2 N )k ln(2V ) − 2( Nk ln V )
= 2 Nk ln 2
which is exactly the same result we got in the first case—something must be wrong!!
Strange as it may seem, the problem lies in the neglect of the N! term. To see why, return
to our original expression for entropy; this time, we will be more careful.
S = k ln Ω( E ,V , N )
 B  3Nk
= k ln  3N 
+ Nk ln V + ln E
 N !h  2
3Nk
= Nk ln V + ln E − k ln N ! − 3Nk ln h − k ln B
2
Using the Stirling approximation, ln N ! ≅ N ln N − N , we obtain
3Nk
S = Nk lnV + ln E − Nk ln N + Nk − 3Nk ln h + k ln B
2
Now we combine the first and third terms to obtain
V 3
S = Nk ln + Nk ln E − Nk − 3Nk ln h + k ln B
N 2
The last three terms will still make no contribution when we calculate the entropy change
∆S . But replacing V by V/N will make a difference. We use this new expression to
calculate ∆S for the two cases discussed above. We assume that N1 = N 2 = N .

2
case 1—helium and argon
∆S = Sfinal − Sinitial
V1 f V2 f V1i V
= N1k ln + N 2 k ln − N1k ln − N 2 k ln 2i
N1 N2 N1 N2
2V 2V V V
= Nk ln + Nk ln − Nk ln − Nk ln
N N N N
= Nk ln 2 + Nk ln 2
= 2 Nk ln 2
This result is exactly the same as the one we obtained above.
case 2—helium on both sides
∆S = Sfinal − Sinitial
Vf V 
= (2 N )k ln − 2  Nk ln i 
2N  N
2V V
= (2 N )k ln( ) − 2( Nk ln )
2N N
V V
= 2 Nk ln − 2 Nk ln = 0
N N
This result is quite different from the result we obtained above, and agrees with our
intuitive expectation. Strangely enough, keeping the N! term leads to the correct result,
even in a system that does not seem in the least quantum-mechanical.

Connection to the canonical ensemble and Baierlein’s result

It is worth noting that the same result follows from the canonical ensemble. Take
Baierlein Eq. (5.39) as a starting point:

( 2π mkT )
N 3N / 2
3 V
S = Nk + k ln .
2 h3 N N !
First note that
3 3
S= Nk ln e + Nk lnV + Nk ln ( 2π mkT ) − k ln N !− 3Nk ln h
2 2
where I have used the identity ln e = 1 to put the equations into Baierlein’s form. Using
the Stirling approximation,
ln N ! = N ln N − N
= N (ln N − 1)
= N ( ln N − ln e )
N
= N ln
e

3
we have, on substituting,
3 3
S= Nk ln e + Nk lnV + Nk ln ( 2π mkT ) − k ( N ln N − N ln e) − 3Nk ln h
2 2
5 V 3
= Nk ln e + Nk ln + Nk ln ( 2π mkT ) − 3Nk ln h
2 N 2
To make a connection to our earlier results, note that we could rewrite this equation in the
form
V 3
S = Nk ln + Nk ln T + constant .
N 2
Thus, if we recall that we measure entropy changes, it should be evident that the
canonical ensemble, the microcanonical ensemble, and macroscopic thermodynamics (for
example, the first TdS equation all lead us to the same result!
This form is also identical to Baierlein’s Eq. (5.41), which if we write it out in full, reads

 V  2π mkT  
3/ 2
V / N 5 / 2  5/ 2 
S = Nk ln  3 e  = Nk ln    e .
 λth
2
  N  h  

Writing out the log terms leads us back to Eq. 5.39, and to the intermediate result given
above.