Appendix A

Thermodynamic Functions and Legendre

Transforms

Background

Chapter 2 prepared a brief thermodynamic basis for understanding transitions from

crystalline solids to their conjugate melt phases, and vice versa. The internal energy,
enthalpy, and Gibbs free energy played important roles in that development. Certain
system parameters, such as the temperature, pressure, and component concentra-
tions, were selected as independent intensive variables when discussing state func-
tion changes under carefully prescribed conditions. The relationships among the
many thermodynamic state functions and a system’s intensive and extensive param-
eters can seem arbitrary, and occasionally confusing. Students often ask why and
how the selection of state functions relates to these variables, and most of those who
do not, have merely memorized the connections. This appendix provides a rigorous
yet manageable alternative.

Thermodynamic Functions and Variables

Internal energy, U, is a fundamental thermodynamic state function providing the ten-

dency of isolated, or ‘closed’ material systems to approach equilibrium. Moreover,
U is a linear, homogeneous function, U(S, V, Ni , . . . ), of all the system’s so-called
‘natural’ extensive variables, including its entropy, S, volume, V , component molar
masses, Ni , etc. Intensive parameters, by contrast, such as temperature, T , pressure,
P, and chemical potential, μi , respectively, are also system variables that often are
more conveniently measured under practical solidification conditions than would the
‘natural’ extensive quantities. With care, as we shall show, extensive thermodynamic
variables may be easily substituted by alternative independent parameters of the
system, each replacing their associated extensive variable. Substitution of intensive
variables for extensive variables generates different thermodynamic functions, or
potentials, that can equivalently determine the state of equilibrium for solidifying
systems maintained under widely varying circumstances, such as melts communi-
cating with a heat bath, a pressure source, or exchanging mass with a reservoir of
chemical potential.

M.E. Glicksman, Principles of Solidification, DOI 10.1007/978-1-4419-7344-3, 449


C Springer Science+Business Media, LLC 2011
450 A Thermodynamic Functions and Legendre Transforms

Interestingly, all such ‘practical’ intensive variables, often imposed by the sys-
tem’s environment, can be mathematically related to conjugate extensive thermody-
namic variables. One, therefore, can reformulate the internal energy function, U, by
replacing one or more of its extensive variables by intensive ones. A good example
of needing such an alternative function is recognizing that a solidifying system’s
temperature, T , is both easily measured using thermometry and controlled in an
oven or furnace with a wide variety of standard laboratory instruments, such as ther-
mocouples or pyrometers. By contrast, temperature’s conjugate extensive variable is
the entropy, which, unfortunately, lacks practical devices for its direct measurement
or control. How, therefore, might one transform the internal energy function, U ,
to provide an alternative thermodynamic function by replacing one, or more, of its
extensive variables, viz., (S, V, Ni , . . . ), by its respective intensive variables, viz.,
(T, P, μi , . . . )?
Such a substitution may be accomplished using Legendre transforms—a standard
mathematical procedure. Legendre transforms, among other things, lead directly to
important thermodynamic ‘potentials’ that will be used throughout this book. These
potentials provide alternatives for the internal energy. Some of these potentials
might seem more familiar than others, as they remain in common use, but the formal
procedure outlined next of how they are created remains identical, irrespective of the
variables chosen.

The Formal Math Problem

Given a function Y defined as

Y = Y(X 0 , X 1 , X 2 . . . X n ), (A.1)

one desires to replace its independent variables, X i , by one or more of their partial
derivatives, Pi , where
 
∂Y
Pi ≡ . (A.2)
∂ X i 1,2,...n =i

The subscripts placed outside the parentheses in Eq. (A.2) denote the variables that
are held fixed during partial differentiation. In fact, the Pi s represent the geometric
‘slopes’ of the function Y plotted with respect to the selected variable X i , holding
all other variables fixed.
Consider the case of a function of just one independent variable, Y(X ). This
function represents a curve in X -Y space, composed of Cartesian coordinate pairs,
(X, Y ). The derivative, P = ddYX , is the tangent to the curve Y(X ) at any point. One
may replace the independent variable, X , by P. Simply replacing the coordinate
X by the function’s slope at that point to yield a new function, Y(P), however,
doesn’t quite work, because, as will be shown, some information is lost by this
direct substitution. Notice that for a single variable, Eq. (A.2) becomes an ordinary
A Thermodynamic Functions and Legendre Transforms 451

differential equation, which, when integrated, yields the solution, Y(X ). Indeed,
this result occurs, but the solution so obtained is established only to within an as yet
unknown constant of integration! See Fig. A.1. Thus, by using direct substitution
some information is lost.

Fig. A.1 Loss of information occurs by replacing the independent variable, X , in Y(X ) by the
slope, P, to form a new transformed function Y(P). Clearly, knowing the slope at every point
along the curve is not equivalent to knowing the coordinates, X, Y . The curves Y1 , Y2 , . . . all
have identical P(X ) values, but are distinctly different functions

The original function Y(X ) may, nonetheless, be represented without loss using
the substitution of slopes via Plücker line geometry. Plücker line geometry shows
that a ‘continuous’ envelope of tangents lines would suffice to define the curve, Y,
as suggested in Fig. A.2. Individual tangents contribute an infinitesimal line segment
that join together smoothly to form the curve Y(P) without any loss of information.
This procedure is, therefore, equivalent to ordinary point geometry, where an infinite
number of coordinate pairs connect to form a curve. The difference, however, is that
each tangential segment contributing to the curve now fundamentally represents
both a slope, or derivative, P, plus some positional information, or locus, obtained
from the derived function, P(X ). Full specification of all the tangent lines is gained
by establishing the relationship between the Y -intercepts, Ψ , and their associated
slopes, P. In short, these become the transformed new independent variables. Thus,
Ψ (P) is mathematically equivalent to the original Y representation, namely, Y(X ),
with all information contained in that original function fully retained in the new Ψ
representation. Thus,

Ψ = (P). (A.3)
452 A Thermodynamic Functions and Legendre Transforms

Fig. A.2 Shown here schematically is an equivalent representation of a function, Y(X ), using the
envelope of all its tangent lines, Pn (X ). This type of Plücker line geometry requires the relation-
ship, (P), which are the functions connecting the slopes of the tangent lines to their Y -intercepts,
i.e., the Ψ s. No loss of information results from this representation

Now the formal mathematics problem reduces to how one calculates (P) given
Y(X ). A well-known mathematical manipulation called the Legendre transforma-
tion provides the appropriate procedure.

Legendre Transformations

One-Variable Transforms
As indicated in Fig. A.2, the slope of the tangent line touching the curve Y(X ) at
the general coordinates X, Y , is

Y −Ψ
P= . (A.4)
X −0

The linear form, Eq. (A.4), provides the definition of the new function, Ψ , which
we term the Legendre transform of the function Y, with respect to X. Operationally,
one writes Ψ = Y[X ], which notes that Ψ is the function Y with its independent
variable X replaced by the slope P = dY/d X , or

Ψ = Y(X ) − P X. (A.5)
A Thermodynamic Functions and Legendre Transforms 453

The inverse transform, i.e., recovering Y(X ) given the function (P), is the sym-
metrical inversion,

dΨ = dY(X ) − Pd X − X d P. (A.6)

Recalling Eq. (A.2), which shows that dY = Pd X , Eq. (A.6) reduces to

dΨ = −X d P, or, equivalently,

dΨ
X =− . (A.7)
dP

Equations (A.5) and (A.7) together allow elimination of the variables Ψ and P
from the transform function (P), thereby inverting the transform and returning
the original function Y(X ). The Legendre transform and its inverse are shown as a
series of operational steps in the table below.

Transform → ←Inverse transform

Y = Y(X ) Ψ = (P)
P = ddY
X −X = dΨdP
Ψ = −P X + Y Y = XP +Ψ
Eliminate X and Y → Eliminate P and Ψ →
Ψ = (P) Y = Y(X )

Multivariate Transforms
The generalization of the Legendre transform method to multivariate functions, such
as U(S, V, Ni , . . . ), which are commonly encountered in thermodynamics, requires
a straightforward extension.
Starting with the fundamental equation Y = Y(X 0 , X 1 , . . . X n ) portrayed in an
n + 2 dimensional space, and represented by the coordinates Y, X 0 , X 1 , X 2 , . . . X n .
The partial derivative, Pk , is defined in the usual way as
 
∂Y
Pk ≡ , (A.8)
∂ Xk 0,1,2,...,n =k

and represents the partial slope of the hypersurface, Y. The total differential of the
function Y is given as the usual sum,


n
dY = Pk d X k . (A.9)
k=0

In an analogous manner to the one-variable case, the hypersurface can be rep-

resented either as the locus of all points satisfying the fundamental equation,
454 A Thermodynamic Functions and Legendre Transforms

Y = Y(X 0 , X 1 , . . . X n ), or as the envelope of all its tangent hyperplanes. Each

hyperplane may be characterized by its intercept, Ψ , which is a function
of all the slopes P0 , P1 , P2 . . . Pn and their associated independent variables,
X 0 , X 1 , X 2 , . . . X n , namely


n
Ψ =Y− Pk X k . (A.10)
k=0

Taking the differential of Eq. (A.10) yields


n
dΨ = − X k d Pk , (A.11)
k=0

and thus,
∂Ψ
Xk = − . (A.12)
∂ Pk

Thermodynamic Potentials

The general function, Y(X 0 , X 1 , . . . ), may now be chosen as the internal energy,
U = U(S, V, Ni , . . . ). For simple chemical systems, such as an alloy melt, com-
posed of n components, not subject to external fields affecting their magnetic, elec-
tric, or strain energies, the internal energy may be expressed in the linear, or Eule-
rian form U (S, V, Ni ) = T S − P V + μi Ni , where repeated subscripts here imply
summation. The intensive conjugate parameters, T, −P, and μi , that multiply each
of the extensive parameters are the temperature, (minus) pressure, and chemical
potential, respectively. These ‘physical’ quantities correspond precisely to the par-
tial derivatives of U, namely, P0 , P1 , · · · . One tends to overlook the simple mathe-
matical relationships between the extensive and conjugate intensive thermodynamic
variables because of one’s familiarity with temperature and pressure as ‘physical’
quantities. However, one should recall that for every extensive variable adding to
a system’s internal energy, there exists a companion, or conjugate, intensive quan-
tity. The product of intensive and extensive variable pairs yields some independent
contribution to the internal energy.
The Legendre transform procedure merely substitutes properly an intensive vari-
able for its conjugate extensive variable. The functions based on U(S, V, Ni ) that
result from taking partial Legendre transforms with respect to these variables are
called thermodynamic potentials, some of which the reader will find familiar, others
perhaps less so.

Helmholtz Potential
If the temperature, T , replaces the entropy, S, as an independent variable, the result-
ing Legendre transform is called the Helmholtz potential, or Helmholtz free energy.
A Thermodynamic Functions and Legendre Transforms 455

Both terms occur in common useage, and are denoted by F(T, V, Ni ). The oper-
ational math notation used here for the partial Legendre transform, indicating the
replacement of a particular extensive thermodynamic variable, (in this case replac-
ing the entropy, S, by its conjugate, the temperature, T ), is,

F(T, V, Ni ) ≡ U[T ], (A.13)

where the square brackets enclose the ‘active’ transform parameter or parame-
ters. The steps to find the partial Legendre transformation and its inverse for the
Helmholtz potential are outlined the following table:

Transform S → T S ← T inverse
U = U(S, V, Ni ) F = F(T, V, Ni )
T = (∂U/∂ S)V,Ni −S = (∂F/∂ T )V,Ni
F =U −TS U = F +TS
Eliminate S and U → Eliminate F and T →
F = F(T, V, Ni ) U = U(S, V, Ni )

One concludes that the Helmholtz potential has the following total differential:

dF = −SdT − Pd V + μi dNi . (A.14)

Enthalpy
If the pressure replaces the volume as an independent variable, the partial trans-
form is called the ‘enthalpy’, or heat content. Both names remain in use, and are
denoted by H(S, P, Ni ). The notation for the partial Legendre transform, indicating
the replacement of the system’s volume, V , by its conjugate, the (minus) pressure,
−P, is,

H(S, P, Ni ) ≡ U[−P]. (A.15)

Steps to obtain the partial Legendre transformation and its inverse for the
enthalpy are outlined the next table:

Transform V → P V ← P inverse
U = U(S, V, Ni ) H = H(S, P, Ni )
−P = (∂U/∂ V ) S,Ni V = (∂H/d P) S,Ni
H = U + PV U = H − PV
Eliminate U and V → Eliminate H and P →
H = H(S, P, Ni ) U = U(S, V, Ni )
456 A Thermodynamic Functions and Legendre Transforms

One concludes that the enthalpy has the following total differential:

dH = TdS + VdP + μi dNi . (A.16)

Gibbs Free Energy

If both the temperature replaces entropy, and the pressure replaces volume, the par-
tial transform is called the ‘Gibbs potential’, or chemical free energy. Both terms
are in common use, and are denoted by G(T, P, Ni ). The notation for the partial
Legendre transform, indicating replacement of both the system’s entropy, S, and
volume, V , by their conjugate variables, the temperature, T , and (minus) pressure,
−P, is,

G(T, P, Ni ) ≡ U[T, −P]. (A.17)

The steps to obtain the partial Legendre transformation and its inverse for the
Gibbs potential are outlined the next table:

Transform S, V → T, P S, V ← T, P inverse
U = U(S, V, Ni ) G = G(T, P, Ni )
T = (∂U/∂ S)V,Ni −S = (∂G/∂ T ) P,Ni
−P = (∂U/∂ V ) S,Ni V = (∂G/d P)T,Ni
G = U − T S + PV U = G + T S − PV
Elimin. U, S and V → Elimin. G, T and P →
G = G(T, P, Ni ) U = U(S, V, Ni )

One concludes that the Gibbs function has the following total differential:

dG = −SdT + VdP + μi dNi . (A.18)

Note, for constant T , constant P, and a closed system, d Ni = 0, dG = 0. Thus, it is

the Gibbs potential that reaches a minimum when say thermochemical equilibrium
is achieved under these typical laboratory conditions.

Grand Canonical Potential

If both the temperature replaces entropy, and the chemical potentials replace mole
numbers, the partial transform is called the ‘Grand Canonical potential’, denoted
by χ (T, V, μi ). The notation for the partial Legendre transform, indicating replace-
ment of both the system’s entropy, S, and mole numbers, Ni , by their conjugate
variables, the temperature, T , and the chemical potentials μi , is,

χ (T, P, μi ) ≡ U[T, Ni ]. (A.19)

A Thermodynamic Functions and Legendre Transforms 457

The steps to obtain the partial Legendre transformation and its inverse for the
Grand Canonical potential are outlined the next table:

Transform S, Ni → T, μi T, μi ← S, Ni inverse
U = U(S, V, Ni ) χ = χ (T, V, μi )
T = (∂U/∂ S)V,Ni −S = (∂χ /∂ T )V,μi
μi = (∂U/∂ Ni ) S,V Ni = (∂χ/dμi )T,V
χ = U − T S − μi Ni U = χ + T S + μi Ni
Elimin. U, S and Ni → Elimin. χ , T and μi →
χ = χ (T, V, μi ) U = U(S, V, Ni )

One concludes that the Grand Canonical potential has the following total differ-
ential:

dχ = −SdT − Pd V − Ni dμi . (A.20)

Note, for constant T , constant V , and for an open system providing constant μi ,
dχ = 0. Thus, it is the Grand Canonical potential that reaches a minimum when
say thermochemical equilibrium is achieved under constrained conditions where the
surroundings set the chemical potentials of the grain boundaries and microscopic
internal defects such as dislocations.