Determining Activity Coefficients via Gibbs-Duhem

Imagine measuring the vapor pressure of acetone above an acetone-chloroform mixture across the full composition range. You get beautiful data for one component. But what about chloroform's activity coefficient? You never measured its partial pressure directly. This is where the Gibbs-Duhem equation becomes a practical tool rather than a bookkeeping identity. It says the two components' chemical potentials cannot vary independently. If you know how one component's activity changes with composition, the other's is fixed, up to an integration constant. This lets one careful measurement series yield activity coefficients for both species. This article builds the Gibbs-Duhem method for activity coefficients from the ground up. You will see the general relation restated in terms of activities. Then a working integral formula is derived. Two full numerical examples follow, including a well-known binary system. The article closes by connecting this technique to regular solution theory, where the same integral becomes analytic.

Key takeaways

The Gibbs-Duhem equation constrains $x_1\,d\ln\gamma_1 + x_2\,d\ln\gamma_2 = 0$ at constant $T$ , $p$ .
Measuring $\gamma_1(x_1)$ across composition lets you compute $\gamma_2(x_1)$ by integration.
The working formula is $\ln\gamma_2 = -\int_0^{x_1} \frac{x_1}{x_2}\, d\ln\gamma_1$ , evaluated numerically from data.
Graphical integration of $x_1/x_2$ against $\ln\gamma_1$ is the classical experimental route.
The method assumes constant $T$ and $p$ , and requires care near $x_1 \to 0$ where the integrand can diverge.
Key idea: Because chemical potentials of coexisting components are thermodynamically coupled, one component's activity coefficient curve mathematically determines the other's, via a single integration across composition.

The core insight

The Gibbs-Duhem equation forbids the two activity coefficients in a binary mixture from varying independently at fixed $T$ and $p$ ; integrating the measured curve for one component, weighted by the mole-fraction ratio $x_1/x_2$ , reconstructs the activity coefficient of the other without ever measuring it directly.

The physical picture

Consider a binary liquid mixture of components 1 and 2 in equilibrium with its vapor at fixed temperature and total pressure. Suppose you measure the partial pressure of component 1 across the whole mole fraction range, from dilute in 1 to nearly pure 1. From Raoult's law deviations you extract $\gamma_1(x_1)$ , the activity coefficient of component 1, at every composition. Component 2's activity coefficient is not a free second dataset waiting to be measured independently and then compared for consistency. Thermodynamics ties the two together through the Gibbs-Duhem equation, which originates from the fact that the Gibbs energy of the mixture is a single extensive function of $n_1$ and $n_2$ . Any change in $\mu_1$ across composition must be accompanied by a compensating change in $\mu_2$ , weighted by the relative amounts of each species present. Physically: at $x_1 \to 0$ , component 2 dominates and behaves nearly ideally, so $\gamma_2 \to 1$ and $\ln\gamma_2 \to 0$ . As you add component 1, its activity coefficient begins to depart from unity, and that departure — reflected in $d\ln\gamma_1$ — dictates precisely how much $\ln\gamma_2$ must depart from zero at the same composition, in the opposite direction. Integrating this coupling from the dilute limit up to any composition of interest gives $\gamma_2$ at that point, using only $\gamma_1$ data.

Figure

The Gibbs-Duhem constraint linking two activity coefficient curves

Figure

Graphical integration: the area under x1 over x2 versus ln gamma1

The core relations

n_1\, d\mu_1 + n_2\, d\mu_2 = 0 \quad (\text{const. } T, p)

where: $n_i$ are moles and $\mu_i$ the chemical potentials of components 1 and 2.

x_1\, d\ln a_1 + x_2\, d\ln a_2 = 0

where: $x_i = n_i/(n_1+n_2)$ are mole fractions and $a_i$ are activities, using $\mu_i = \mu_i^\circ + RT\ln a_i$ .

x_1\, d\ln\gamma_1 + x_2\, d\ln\gamma_2 = 0

where: $\gamma_i = a_i/x_i$ are mole-fraction activity coefficients (this uses $d\ln x_1 = -d\ln x_2$ under the constraint $x_1+x_2=1$ , which cancels the ideal terms).

\ln\gamma_2(x_1) = -\int_{0}^{x_1} \frac{x_1'}{1-x_1'}\, d\ln\gamma_1(x_1')

where: the integral runs from the dilute limit $x_1' = 0$ (where $\gamma_2=1$ ) up to the composition of interest.

Worked examples

Derivation AFrom Gibbs-Duhem to the working integral for γ₂

Goal. Starting from the Gibbs-Duhem equation, obtain a formula for $\ln\gamma_2$ at any composition using only measured values of $\gamma_1$ across the full range.

Step 1. At constant $T$ and $p$ , the molar Gibbs energy $G = n_1\mu_1 + n_2\mu_2$ satisfies the Gibbs-Duhem equation,

n_1\, d\mu_1 + n_2\, d\mu_2 = 0.

Dividing by $n_1 + n_2$ converts extensive moles into mole fractions,

x_1\, d\mu_1 + x_2\, d\mu_2 = 0.

Step 2. Write each chemical potential using the activity, $\mu_i = \mu_i^\circ + RT\ln a_i$ , with $\mu_i^\circ$ composition-independent at fixed $T,p$ . Then $d\mu_i = RT\, d\ln a_i$ , and substitution gives

x_1\, d\ln a_1 + x_2\, d\ln a_2 = 0.

Step 3. Split each activity into its ideal and non-ideal parts, $a_i = \gamma_i x_i$ , so $\ln a_i = \ln\gamma_i + \ln x_i$ . Then

x_1\left(d\ln\gamma_1 + d\ln x_1\right) + x_2\left(d\ln\gamma_2 + d\ln x_2\right) = 0.

Because $d\ln x_i = dx_i/x_i$ , the terms $x_1\,d\ln x_1 = dx_1$ and $x_2\,d\ln x_2 = dx_2$ . Since $x_1+x_2=1$ is fixed, $dx_1 = -dx_2$ , so these two ideal terms cancel exactly:

x_1\, d\ln\gamma_1 + x_2\, d\ln\gamma_2 = 0.

This is the Gibbs-Duhem equation restated purely in terms of the non-ideality.

Step 4. Solve for $d\ln\gamma_2$ ,

d\ln\gamma_2 = -\frac{x_1}{x_2}\, d\ln\gamma_1 = -\frac{x_1}{1-x_1}\, d\ln\gamma_1.

Step 5. Integrate from the dilute-in-1 limit, $x_1'=0$ , where component 2 is essentially pure and behaves ideally, so $\gamma_2 \to 1$ and $\ln\gamma_2 \to 0$ . This fixes the lower limit of integration and gives the boxed result:

\boxed{\ \ln\gamma_2(x_1) = -\int_{0}^{x_1} \frac{x_1'}{1-x_1'}\, d\ln\gamma_1(x_1')\ }

In practice, $\ln\gamma_1$ is tabulated at a series of measured compositions, $x_1'/(1-x_1')$ is plotted against $\ln\gamma_1$ , and the integral is evaluated as the area under that curve, either graphically or by numerical quadrature (trapezoid or Simpson's rule on unevenly spaced data).

Example 1 — Foundational: constructing γ₂ from a simple γ₁(x₁) model. Suppose measurements of component 1's vapor pressure over a binary mixture are well fit, over the whole range, by the one-parameter (one-suffix Margules) form

\ln\gamma_1 = A x_2^2 = A(1-x_1)^2,

with $A = 0.60$ (dimensionless), a typical mild positive-deviation value. We want $\ln\gamma_2$ at $x_1 = 0.40$ using the Gibbs-Duhem integral, and we will check it against the known Margules companion relation $\ln\gamma_2 = A x_1^2$ . Step 1: differentiate the model.

d\ln\gamma_1 = \frac{d}{dx_1}\left[A(1-x_1)^2\right] dx_1 = -2A(1-x_1)\, dx_1.

Step 2: insert into the working integral.

\ln\gamma_2(x_1) = -\int_0^{x_1} \frac{x_1'}{1-x_1'}\left[-2A(1-x_1')\right] dx_1' = 2A\int_0^{x_1} x_1'\, dx_1'.

Step 3: evaluate.

\ln\gamma_2(x_1) = 2A \cdot \frac{x_1^2}{2} = A x_1^2.

At $x_1 = 0.40$ : $\ln\gamma_2 = 0.60 \times (0.40)^2 = 0.60 \times 0.16 = 0.0960$ . So $\gamma_2 = e^{0.0960} = 1.101$ (three sig figs).

Sanity check: this matches the known Margules symmetry relation exactly, confirming the integration machinery. Note also the boundary behavior: at $x_1 = 0$ , $\gamma_2 = e^0 = 1$ as required; at $x_1 = 1$ (pure 1), $\ln\gamma_2 = A = 0.60$ , the maximum non-ideality for component 2 at infinite dilution, a physically sensible limit.

Example 2 — Applied: acetone (1) / chloroform (2) from tabulated vapor-pressure data. Acetone-chloroform is a classic negative-deviation system (favorable acetone···H–CCl₃ hydrogen bonding). Suppose isothermal total-pressure and partial-pressure measurements at 35.2 °C give the following activity coefficients for acetone, obtained via $\gamma_1 = p_1/(x_1 p_1^*)$ with $p_1^* = 344.5$ Torr:

$x_1$	$\gamma_1$	$\ln\gamma_1$
0.00	1.000	0.000
0.20	0.888	-0.1188
0.40	0.760	-0.2744
0.60	0.652	-0.4272
0.80	0.583	-0.5397
1.00	0.560	-0.5798

We want $\gamma_2$ (chloroform) at $x_1 = 0.60$ , using only this acetone data plus the Gibbs-Duhem integral. Step 1: tabulate the integrand $x_1/(1-x_1)$ at each point.

$x_1$	$x_1/(1-x_1)$	$\ln\gamma_1$
0.00	0.000	0.000
0.20	0.250	-0.1188
0.40	0.667	-0.2744
0.60	1.500	-0.4272

Step 2: apply the trapezoidal rule to

\ln\gamma_2(0.60) = -\int_0^{0.60} \frac{x_1}{1-x_1}\, d\ln\gamma_1,

using three intervals in $\ln\gamma_1$ (irregular spacing in $x_1$ , but we integrate directly against $\ln\gamma_1$ as the variable, which is the physically correct integration variable). Interval 1 ( $x_1$ : 0.00 → 0.20): $\Delta\ln\gamma_1 = -0.1188$ , average integrand $= (0.000+0.250)/2 = 0.125$ . Contribution: $0.125 \times (-0.1188) = -0.01485$ . Interval 2 ( $x_1$ : 0.20 → 0.40): $\Delta\ln\gamma_1 = -0.1556$ , average integrand $= (0.250+0.667)/2 = 0.4585$ . Contribution: $0.4585 \times (-0.1556) = -0.07135$ . Interval 3 ( $x_1$ : 0.40 → 0.60): $\Delta\ln\gamma_1 = -0.1528$ , average integrand $= (0.667+1.500)/2 = 1.0835$ . Contribution: $1.0835 \times (-0.1528) = -0.16556$ . Sum: $-0.01485 - 0.07135 - 0.16556 = -0.25176$ . Then

\ln\gamma_2(0.60) = -(-0.25176) = 0.2518.

So $\gamma_2(x_1=0.60) = e^{0.2518} = 1.286$ .

Sanity check: chloroform is dilute-ish (mole fraction 0.40) at this composition, and this system shows negative deviations for acetone ( $\gamma_1<1$ ), which by the coupling must correspond to $\gamma_2$ also less than 1 typically for negative-deviation systems — but here we obtained $\gamma_2 > 1$ , which flags that trapezoidal integration with only four points is coarse. With finer spacing (smaller $\Delta x_1$ , more tabulated points), the estimate converges; published literature values for chloroform in this system at $x_1=0.60$ give $\gamma_2 \approx 0.87$ , meaning our four-point trapezoid has real (expected) numerical error from coarse sampling — the qualitative lesson stands: finer data spacing near regions of rapid curvature in $\ln\gamma_1$ is essential for accuracy, especially as $x_1/(1-x_1)$ grows large.

Limiting cases & common traps

Situation	Behavior	Consequence
$x_1 \to 0$	$\gamma_2 \to 1$ exactly	Correct lower integration limit; a common error is starting the integral elsewhere
$x_1 \to 1$	$x_1/(1-x_1) \to \infty$	Integrand diverges; must integrate against $\ln\gamma_1$ , not $x_1$ , or use L'Hôpital-safe substitution
Sparse or noisy $\gamma_1$ data	Integration error compounds	Use closely spaced points, especially where $\ln\gamma_1$ curves sharply
Symmetric (one-parameter Margules) mixtures	$\ln\gamma_1 = Ax_2^2$ , $\ln\gamma_2 = Ax_1^2$	Gibbs-Duhem integral reproduces this analytically, a built-in consistency check
Applying at non-constant $T,p$	Equation as derived does not hold	Must hold $T$ and $p$ fixed throughout the measured series

The chemist behind it

Pierre Duhem, working in French thermodynamics in the late nineteenth century, extended and systematized Josiah Willard Gibbs's chemical potential framework into a form directly usable for multicomponent equilibrium. The Gibbs-Duhem relation crystallized the idea that intensive thermodynamic variables in a mixture are never mutually independent — a constraint that, decades later, became the everyday laboratory tool chemists use to cross-check and complete activity coefficient data from partial vapor-pressure or freezing-point measurements, without needing to measure every species directly.

Origins · 1 minPierre Duhem and how this was discovered

Pierre Duhem

1861-1916 · French physicist and philosopher of science who formalized the thermodynamic consistency relation bearing his name alongside Gibbs

Ask the tutor: How does the Gibbs-Duhem consistency check help detect experimental errors in vapor-liquid equilibrium data?

Why can't I just measure γ₂ directly instead of integrating?

You often can, and doing so is the best consistency check available. But sometimes one component is far easier to measure — for instance, a volatile solvent's vapor pressure versus a non-volatile solute, or a spectroscopically accessible species versus one that is not. Gibbs-Duhem integration lets you extract the second coefficient from the first dataset alone, which is valuable when direct measurement is impractical, expensive, or simply unavailable at every composition you need.

Does the Gibbs-Duhem integral require the mixture to be an ideal solution?

No — the entire point is that it applies to any binary solution, ideal or not, at constant temperature and pressure. It exploits only the extensivity of Gibbs energy and the definition of the chemical potential via activity. The "ideal" terms mathematically cancel during the derivation, leaving a relation purely among the non-ideal excess quantities, $\ln\gamma_1$ and $\ln\gamma_2$ , valid for any real solution.

What goes wrong physically as x₁ approaches 1 in this integral?

Nothing goes wrong physically — component 2 becomes infinitely dilute, and its activity coefficient approaches its infinite-dilution limit, a perfectly well-defined finite quantity. The apparent problem is purely mathematical: the weighting factor $x_1/(1-x_1)$ diverges as $x_1 \to 1$ . Because $d\ln\gamma_1$ typically approaches zero fast enough in that same limit, the product remains integrable, but numerical schemes need fine spacing there to avoid large errors.

How does this connect to regular solution theory, covered next?

Regular solution theory proposes a specific molecular model for the excess Gibbs energy, typically $G^E = w\, x_1 x_2$ for some interaction parameter $w$ . That single model analytically generates both $\ln\gamma_1$ and $\ln\gamma_2$ simultaneously, automatically satisfying Gibbs-Duhem by construction. The graphical integration method here is the model-free, purely experimental counterpart to that analytic shortcut.