Entropy Concavity and Stability

Stability in the entropy representation is equivalent to entropy being concave in the extensive variables.
Entropy Concavity and Stability

Definition (global concavity)

Let the be S=S(U,V,N)S=S(U,V,N) for a single-phase system, with (U,V,N)(U,V,N) being . The entropy is concave if for any two states x1=(U1,V1,N1)x_1=(U_1,V_1,N_1) and x2=(U2,V2,N2)x_2=(U_2,V_2,N_2) and any λ[0,1]\lambda\in[0,1],

S ⁣(λx1+(1λ)x2)λS(x1)+(1λ)S(x2). S\!\big(\lambda x_1 + (1-\lambda)x_2\big) \ge \lambda S(x_1) + (1-\lambda)S(x_2).

Physically, this expresses that “mixing” or composing subsystems at fixed total (U,V,N)(U,V,N) cannot reduce total ; it is the mathematical backbone of why the state of an isolated system is a stable entropy maximum, i.e. a core part of .

Local (differential) criterion

If SS is twice differentiable, concavity is equivalent to the Hessian (matrix of second ) with respect to (U,V,N)(U,V,N) being negative semidefinite. For example, restricting attention to (U,V)(U,V) at fixed NN, one requires

(2SU2)V,N0,det(SUUSUVSVUSVV)0, \left(\frac{\partial^2 S}{\partial U^2}\right)_{V,N} \le 0, \qquad \det \begin{pmatrix} S_{UU} & S_{UV}\\ S_{VU} & S_{VV} \end{pmatrix} \ge 0,

with SUUS_{UU} etc. denoting the corresponding second derivatives at fixed NN.

Consequences for response functions

Concavity translates into positivity of familiar :

  • Using the defining relation 1/T=(S/U)V,N1/T = (\partial S/\partial U)_{V,N} for the TT, one finds

    (2SU2)V,N=1T2(TU)V,N=1T2CV, \left(\frac{\partial^2 S}{\partial U^2}\right)_{V,N} ={} -\frac{1}{T^2}\left(\frac{\partial T}{\partial U}\right)_{V,N} ={} -\frac{1}{T^2\,C_V},

    where CVC_V is the . Thus SUU0S_{UU}\le 0 implies CV0C_V\ge 0 (strict concavity corresponds to CV>0C_V>0).

  • Through Legendre transforms to potentials such as the F(T,V,N)F(T,V,N) and G(T,P,N)G(T,P,N) (a thermodynamic instance of the ), entropy concavity enforces mechanical stability conditions like (P/V)T,N<0(\partial P/\partial V)_{T,N}<0, i.e. a positive .

  • A standard identity relating measurable susceptibilities,

    CPCV=TVα2κT, C_P - C_V = \frac{T V \alpha^2}{\kappa_T},

    shows (for T>0T>0) that positivity of the and the implies the satisfies CPCVC_P\ge C_V.