Equilibrium Conditions Between Subsystems

Put a hot brick against a cold one and wait: their temperatures equalize. Connect two gas chambers with a piston: their pressures equalize. Join two solutions through a membrane: their concentrations — more precisely, their chemical potentials — equalize. Three different equilibrations, three different intensive variables coming to a common value. It is no coincidence that these look alike. They are three instances of a single principle: an isolated composite system reaches equilibrium at the state that maximizes its total entropy, and maximizing entropy under each kind of exchange forces the matching intensive variable to be equal across the parts. This is the master theorem of equilibrium thermodynamics. Take two subsystems that can exchange a conserved quantity — energy, volume, or particles — at fixed total. The second law says the total entropy climbs to a maximum, and the first-order condition for that maximum equalizes the corresponding entropy derivative: $(\partial S/\partial U)$ (giving equal temperature), $(\partial S/\partial V)$ (equal pressure at equal $T$ ), or $(\partial S/\partial N)$ (equal chemical potential). One entropy-maximum principle generates all three equilibrium conditions at once. But being an extremum is not enough — equilibrium must be a stable maximum, or the smallest fluctuation would tip the system away. This is where the second-order conditions come in, and they carry real physical content. Requiring the total entropy to be concave (its second derivatives negative-definite) forces three inequalities: the heat capacity $C_V > 0$ , the isothermal compressibility $\kappa_T > 0$ , and a mixed condition. These are not decorative — they say that adding heat raises temperature and squeezing raises pressure, and their violation marks the onset of instability, where a homogeneous phase spontaneously separates into two. So the full structure of equilibrium is captured by first and second derivatives of one function. The first derivatives give the three equalities — the conditions for equilibrium. The second derivatives give the three inequalities — the conditions for that equilibrium to be stable. This is the complete first-and-second-order theory, and it is the bridge from the abstract entropy-maximum postulate to the concrete, observable facts that hot flows to cold, that stable matter resists compression, and that phase transitions occur exactly where stability fails.

Key takeaways

An isolated composite system reaches equilibrium by maximizing total entropy under the exchange of conserved quantities between its parts.
First-order conditions (the maximum): exchanging energy gives $T_1 = T_2$ , exchanging volume gives $P_1 = P_2$ , exchanging particles gives $\mu_1 = \mu_2$ .
All three equilibrium conditions come from one principle — maximizing $S$ equalizes the matching entropy derivative.
Second-order conditions (stability): requiring the maximum to be stable forces $C_V > 0$ , $\kappa_T > 0$ , and a mixed Hessian condition — entropy must be concave.
Where stability fails (concavity breaks), the homogeneous phase is unstable and phase separation occurs — the link to phase transitions.
Key idea: entropy maximization under exchange gives three first-order equalities (equal $T$ , $P$ , $\mu$ ) and three second-order stability inequalities ( $C_V, \kappa_T > 0$ ), the complete conditions for stable equilibrium.

The core insight

An isolated composite system maximizes its total entropy under exchange of conserved quantities. The first-order conditions equalize the intensive variables — $T_1 = T_2$ (energy exchange), $P_1 = P_2$ (volume), $\mu_1 = \mu_2$ (particles) — while the second-order conditions (concavity of $S$ ) give the stability inequalities $C_V > 0$ and $\kappa_T > 0$ , whose failure signals phase separation.

Where this lives today

The entropy-maximum principle and its stability analysis are the working foundation of any field that computes equilibrium states. In computational thermodynamics and materials science, the CALPHAD method finds the equilibrium phases of a multicomponent alloy by minimizing the total Gibbs free energy (equivalently maximizing entropy) subject to mass conservation, then checks the second-order (convexity) conditions to determine which regions are stable, metastable, or unstable — the spinodal and binodal curves that this article's stability conditions define. Spinodal decomposition, the spontaneous phase separation that occurs when a material is quenched into the unstable region, is a central process in metallurgy, polymer science, and the microstructure of everything from stainless steel to battery electrodes. In geophysics and planetary science, the same equilibrium conditions ( $T$ , $P$ , $\mu$ equal across phase boundaries) determine the mineral assemblages stable at each depth in a planet's interior, turning a rock into a barometer and thermometer for conditions hundreds of kilometers down. In chemical engineering, phase-equilibrium calculations — which distillation columns, extraction processes, and reactor designs all depend on — are exactly the solution of $\mu_i^\alpha = \mu_i^\beta$ for every component across every phase, with stability analysis flagging where a mixture will spontaneously split. And conceptually, the stability conditions are the rigorous statement of Le Chatelier's principle and the reason ordinary matter is robust: a stable phase resists being pushed, and the study of where that resistance fails — critical points, spinodals, instabilities — is the entire modern theory of phase transitions.

The physical picture

Consider an isolated box divided into two parts by an internal wall. If the wall conducts heat but is rigid and impermeable, the two parts can exchange energy but nothing else. The total energy is fixed, and the second law says the total entropy $S_1(U_1) + S_2(U_2)$ will rise to its maximum. Sliding energy from part 2 to part 1 changes the total by $[(\partial S_1/\partial U_1) - (\partial S_2/\partial U_2)]\,\mathrm{d}U_1$ , which keeps rising until the two energy-slopes match. That common slope is $1/T$ , so the equilibrium condition is $T_1 = T_2$ : thermal equilibrium is equal temperature. Now make the wall movable as well: it can also exchange volume. Maximizing entropy over the volume split adds a second condition, $(\partial S_1/\partial V_1) = (\partial S_2/\partial V_2)$ , which (at the equal temperature already established) is $P_1 = P_2$ : mechanical equilibrium is equal pressure. Make the wall permeable, and a third condition appears, equalizing $(\partial S/\partial N)$ , i.e. $\mu_1 = \mu_2$ : diffusive equilibrium is equal chemical potential. One principle — maximize total entropy — generates all three, each triggered by the kind of exchange the wall permits.

Figure

Entropy maximization under three exchanges gives three equalities of intensive variables

But an extremum could be a maximum, a minimum, or a saddle. For equilibrium to be stable, the total entropy must be a genuine maximum — its second derivatives must be negative, making $S$ concave. This has teeth. Concavity in energy is exactly $C_V > 0$ (adding heat raises temperature); concavity in volume is exactly $\kappa_T > 0$ (squeezing raises pressure). If either failed, a tiny fluctuation would grow: a slightly hotter region would draw more heat and run away, or a slightly denser region would collapse. The plot below shows what stability looks like — a concave entropy surface with a single stable peak, versus a non-concave one whose "equilibrium" is unstable and where the system phase-separates instead.

Figure

Stability = concavity: a concave S has a stable max; a non-concave region phase-separates

The core equations

The equilibrium of an isolated composite system is the maximum of its total entropy. For two subsystems exchanging energy at fixed total $U = U_1 + U_2$ , the first-order condition is:

\frac{\partial S_1}{\partial U_1} = \frac{\partial S_2}{\partial U_2} \;\Longrightarrow\; \frac{1}{T_1} = \frac{1}{T_2} \;\Longrightarrow\; T_1 = T_2 \qquad (1)

where: $S_i(U_i)$ — subsystem entropies; the equalized energy-slope is $1/T$ , so thermal equilibrium is equal temperature (the zeroth law, derived from entropy maximization). Allowing volume exchange at fixed total $V$ adds a second condition, which at equal temperature gives equal pressure:

\frac{\partial S_1}{\partial V_1} = \frac{\partial S_2}{\partial V_2} \;\Longrightarrow\; \frac{P_1}{T_1} = \frac{P_2}{T_2} \;\Longrightarrow\; P_1 = P_2 \qquad (2)

where: the equalized volume-slope is $P/T$ ; combined with $T_1 = T_2$ , mechanical equilibrium is equal pressure. Allowing particle exchange adds a third condition — equal chemical potential:

\frac{\partial S_1}{\partial N_1} = \frac{\partial S_2}{\partial N_2} \;\Longrightarrow\; \frac{\mu_1}{T_1} = \frac{\mu_2}{T_2} \;\Longrightarrow\; \mu_1 = \mu_2 \qquad (3)

where: the equalized particle-slope is $-\mu/T$ ; diffusive equilibrium is equal chemical potential. Particles flow from high $\mu$ to low $\mu$ until they match. Stability requires the extremum to be a maximum — the total entropy must be concave, so its second derivative is negative. For energy exchange:

\frac{\partial^2 S}{\partial U^2} = -\frac{1}{T^2 C_V} < 0 \;\Longleftrightarrow\; C_V > 0 \qquad (4)

where: $C_V$ — heat capacity at constant volume; thermal stability means adding heat raises temperature. Similarly, mechanical stability requires $-(\partial^2 S/\partial V^2) > 0$ , i.e. $\kappa_T > 0$ (compressibility positive). The full stability condition is that the Hessian matrix of second derivatives of $S(U,V)$ be negative-definite:

S_{UU} < 0, \quad S_{VV} < 0, \quad S_{UU}S_{VV} - S_{UV}^2 > 0 \qquad (5)

where: the three inequalities together guarantee a genuine, jointly stable maximum; the mixed (determinant) condition couples thermal and mechanical stability. Their failure marks the spinodal — the boundary of instability where a phase separates.

Derivation AThe three equilibrium conditions from one entropy-maximum principle

The goal is to derive all three equilibrium conditions — equal temperature, pressure, and chemical potential — as first-order conditions of a single variational principle: maximizing the total entropy of an isolated composite system under the exchange of the corresponding conserved quantity. This unification is the article's owned result: it shows the three equilibria are not three separate laws but three faces of the second law. We carry it through for all three exchanges and verify the stationarity conditions symbolically.

Step 1. Two subsystems with entropies $S_1, S_2$ can exchange energy, volume, and particles, subject to fixed totals $U_1 + U_2 = U$ , $V_1 + V_2 = V$ , $N_1 + N_2 = N$ . The total entropy is $S = S_1(U_1, V_1, N_1) + S_2(U_2, V_2, N_2)$ , and equilibrium is its maximum over the allowed splits. Why: the composite is isolated, so the second law dictates that the total entropy is maximized at equilibrium — the single governing principle.

Step 2. Energy exchange: $\dfrac{\partial S}{\partial U_1} = \dfrac{\partial S_1}{\partial U_1} - \dfrac{\partial S_2}{\partial U_2} = 0$ (since $\mathrm{d}U_2 = -\mathrm{d}U_1$ ), giving $\dfrac{\partial S_1}{\partial U_1} = \dfrac{\partial S_2}{\partial U_2}$ , i.e. $\dfrac1{T_1} = \dfrac1{T_2}$ , hence $T_1 = T_2$ . Why: energy removed from one subsystem is added to the other, so the two slopes appear with opposite signs, and the maximum equalizes them.

Step 3. Volume and particle exchange, by the identical argument: $\dfrac{\partial S_1}{\partial V_1} = \dfrac{\partial S_2}{\partial V_2}$ gives $\dfrac{P_1}{T_1} = \dfrac{P_2}{T_2}$ , so $P_1 = P_2$ at equal $T$ ; and $\dfrac{\partial S_1}{\partial N_1} = \dfrac{\partial S_2}{\partial N_2}$ gives $\dfrac{\mu_1}{T_1} = \dfrac{\mu_2}{T_2}$ , so $\mu_1 = \mu_2$ . Why: each independent conserved exchange gives one independent stationarity condition, and each equalizes the matching intensive variable — the same logic three times.

Step 4. Verify on ideal gases: with $S_i \propto a_i\ln U_i$ (thermal), maximizing $a_1\ln U_1 + a_2\ln(U - U_1)$ gives $a_1/U_1 = a_2/U_2$ , i.e. energy shared in proportion to $N_i$ — equal energy per particle, hence equal temperature. The volume split similarly gives equal pressure. Why: the ideal-gas check confirms the abstract conditions reduce to the familiar equalizations, and shows energy/volume are shared in proportion to particle number. *Verified: the thermal condition $1/T_1 = 1/T_2$ , the mechanical condition $P_1/T_1 = P_2/T_2$ , and the ideal-gas equal-energy-per-particle split are all checked symbolically in verify_aq-sm-equilibrium-conditions-subsystems.py.*

Go deeper · Derivation B · 2 minStability: the second-order conditions and the birth of phase separation

Derivation BStability: the second-order conditions and the birth of phase separation

The goal is to derive the stability conditions — the second-order requirements that turn an entropy extremum into a stable maximum — and to show that they are exactly $C_V > 0$ , $\kappa_T > 0$ , and a mixed Hessian condition, whose violation is the onset of phase separation. This completes the equilibrium theory: the first derivatives locate equilibrium, the second derivatives decide whether it can exist. We derive the conditions from the negative-definiteness of the entropy Hessian and verify the signs symbolically.

Step 1. For the entropy maximum to be stable, moving a bit of energy $\delta U$ (or volume) from one subsystem to the other must decrease the total entropy — otherwise the fluctuation would grow. Expanding to second order, this requires $\dfrac{\partial^2 S}{\partial U_1^2}\Big|_{\text{tot}} = \dfrac{\partial^2 S_1}{\partial U_1^2} + \dfrac{\partial^2 S_2}{\partial U_2^2} < 0$ . Why: a negative second derivative means the entropy curves downward away from the peak, so any fluctuation lowers $S$ and is suppressed — the definition of stability.

Step 2. For a single homogeneous phase this becomes the condition that each subsystem's entropy is concave in energy: $\dfrac{\partial^2 S}{\partial U^2} < 0$ . Using $\partial S/\partial U = 1/T$ and $\partial^2 S/\partial U^2 = -\dfrac{1}{T^2}\dfrac{\partial T}{\partial U} = -\dfrac{1}{T^2 C_V}$ , this is exactly $C_V > 0$ . Why: thermal stability is the physically obvious statement that adding heat raises temperature; a negative $C_V$ would mean adding heat cools the system, which runs away.

Step 3. The analogous volume condition gives $\dfrac{\partial^2 S}{\partial V^2} < 0$ , equivalent to the isothermal compressibility $\kappa_T > 0$ — squeezing raises pressure. Why: mechanical stability means a denser region pushes back harder; negative $\kappa_T$ would let a density fluctuation collapse, causing phase separation.

Step 4. Jointly, the full Hessian of $S(U,V)$ must be negative-definite: $S_{UU} < 0$ , $S_{VV} < 0$ , and $S_{UU}S_{VV} - S_{UV}^2 > 0$ . For the Sackur–Tetrode entropy all three hold, confirming the ideal gas is stable. Where they fail — inside the spinodal of a van der Waals fluid — the homogeneous phase is unstable and separates into liquid and vapor; the boundary $S_{UU}S_{VV} - S_{UV}^2 = 0$ is the spinodal curve. Why: the mixed determinant condition couples thermal and mechanical stability, and its vanishing is precisely where a phase transition begins — the deep link between stability and phase transitions. *Verified: $\partial^2 S/\partial U^2 < 0$ (thermal stability, $C_V > 0$ ), $\partial^2 S/\partial V^2 < 0$ (mechanical, $\kappa_T > 0$ ), and the negative-definite Hessian ( $S_{UU} < 0$ , $S_{VV} < 0$ , $S_{UU}S_{VV} - S_{UV}^2 > 0$ ) for the Sackur–Tetrode entropy are all checked symbolically in the verify script.*

Limiting cases

How the equilibrium and stability conditions specialize:

Case	Result	Meaning
Heat-conducting rigid wall	$T_1 = T_2$	Thermal equilibrium only; $P$ , $\mu$ can differ.
Movable wall	$T_1 = T_2$ , $P_1 = P_2$	Mechanical equilibrium added.
Permeable wall	$T_1 = T_2$ , $P_1 = P_2$ , $\mu_1 = \mu_2$	Full equilibrium.
Thermal stability	$C_V > 0$	Adding heat raises temperature.
Mechanical stability	$\kappa_T > 0$	Squeezing raises pressure.
Stability failure (spinodal)	$S_{UU}S_{VV} - S_{UV}^2 = 0$	Onset of phase separation.

Common trap“Equilibrium just means the intensive variables are equal — stability is a separate technicality”

Stability is not a technicality; it is half the theory, and it carries the physics of phase transitions. The equality conditions ( $T_1 = T_2$ , etc.) only locate an extremum of the total entropy — but an extremum can be a maximum, a minimum, or a saddle. Only a stable maximum is a real equilibrium; at a minimum or saddle, the tiniest fluctuation grows and the state falls apart. The stability conditions $C_V > 0$ and $\kappa_T > 0$ are what distinguish a genuine equilibrium from an unstable extremum. And they are where the most interesting physics lives: a system driven into a region where stability fails does not sit there — it phase-separates, splitting into two coexisting phases (liquid and vapor, say). The spinodal curve, where the Hessian determinant vanishes, marks exactly this boundary. So "equilibrium = equal intensive variables" is only the first half; the stability conditions complete it and open the door to phase transitions.

Worked examples

1 · Two blocks share energy and reach a common temperature (confidence builder)

Two blocks with entropies $S_1 = a_1\ln U_1$ and $S_2 = a_2\ln U_2$ (ideal-gas-like, $a_i = \tfrac32 N_i k_B$ ) share fixed total energy $U = U_1 + U_2$ . Find the equilibrium split, verify it is a stable maximum, and interpret.

Step 1. Maximize $S_{\text{tot}} = a_1\ln U_1 + a_2\ln(U - U_1)$ : set $\dfrac{\mathrm{d}S_{\text{tot}}}{\mathrm{d}U_1} = \dfrac{a_1}{U_1} - \dfrac{a_2}{U - U_1} = 0$ , giving $\dfrac{a_1}{U_1} = \dfrac{a_2}{U_2}$ , i.e. $\dfrac{U_1}{N_1} = \dfrac{U_2}{N_2}$ — equal energy per particle.

Step 2. Since $1/T_i = \partial S_i/\partial U_i = a_i/U_i$ , the condition $a_1/U_1 = a_2/U_2$ is exactly $T_1 = T_2$ . The blocks reach a common temperature, sharing energy in proportion to their particle numbers.

Step 3. Numerical case: equal blocks ( $a_1 = a_2$ ), $U = 3$ . Then $U_1 = U_2 = \mathbf{1.5}$ — energy splits evenly, equal temperatures. Unequal blocks ( $a_1 = 2a_2$ , block 1 has twice the particles), $U = 3$ : solving $2/U_1 = 1/(3 - U_1)$ gives $U_1 = \mathbf{2}$ , $U_2 = 1$ — block 1 holds twice the energy, so $U_1/N_1 = U_2/N_2$ and $T_1 = T_2$ .

Step 4. Stability: $\dfrac{\mathrm{d}^2 S_{\text{tot}}}{\mathrm{d}U_1^2} = -\dfrac{a_1}{U_1^2} - \dfrac{a_2}{U_2^2} < 0$ everywhere — the extremum is a genuine maximum, so the equilibrium is stable ( $C_V > 0$ ). The blocks equilibrate to a common temperature and stay there; a fluctuation that made one block hotter would be suppressed because it would lower the total entropy.

2 · Testing the stability of an ideal gas (exam level)

Verify that the Sackur–Tetrode entropy $S(U,V,N)$ satisfies all three stability conditions, confirming the ideal gas is a stable phase, and identify what would change near a phase transition.

Step 1. From $S = Nk_B[\ln(\tfrac{V}{N}(\tfrac{U}{N})^{3/2}C) + \tfrac52]$ , compute the thermal second derivative: $S_{UU} = \partial^2 S/\partial U^2 = -\tfrac{3N k_B}{2U^2} < \mathbf{0}$ . Thermal stability holds: $C_V = \tfrac32 Nk_B > 0$ . ✓

Step 2. Mechanical: $S_{VV} = \partial^2 S/\partial V^2 = -\tfrac{Nk_B}{V^2} < \mathbf{0}$ . Mechanical stability holds: $\kappa_T > 0$ . ✓ Squeezing the gas raises its pressure, as it must for a stable substance.

Step 3. Mixed condition: $S_{UV} = \partial^2 S/\partial U\partial V = 0$ (the Sackur–Tetrode entropy separates in $U$ and $V$ ), so the Hessian determinant is $S_{UU}S_{VV} - S_{UV}^2 = \tfrac{3N^2k_B^2}{2U^2V^2} > \mathbf{0}$ . ✓ The full Hessian is negative-definite: the ideal gas is jointly stable in energy and volume.

Step 4. Near a phase transition (a real gas, van der Waals): inside the spinodal the mechanical condition fails, $\kappa_T < 0$ (the isotherm has a region where $\partial P/\partial V > 0$ ), so $S_{VV} > 0$ and the Hessian is no longer negative-definite. There the homogeneous phase is unstable and separates into liquid and vapor. The ideal gas never does this — it has no attractive interactions to drive condensation — which is exactly why it is stable at all densities. Stability analysis distinguishes a gas that can condense from one that cannot.

Origins · 1 minequilibrium and how this was discovered

equilibrium

concept · the state of maximum total entropy

The concept of thermodynamic equilibrium as the state of maximum entropy — and the systematic derivation of the equilibrium and stability conditions from it — is due above all to Josiah Willard Gibbs, whose 1875–78 On the Equilibrium of Heterogeneous Substances laid out the complete variational theory: equilibrium as the maximum of entropy (or minimum of the appropriate free energy) subject to constraints, with the equalities of temperature, pressure, and chemical potential as first-order conditions and the convexity/concavity requirements as stability conditions. Herbert Callen, in his influential 1960 textbook, recast this as a clean axiomatic structure built on the entropy-maximum postulate and its extremum principle, making the "maximize entropy, then check the Hessian" procedure the modern standard. The stability conditions connect directly to Le Chatelier's principle (1884): a stable system responds to a perturbation in the direction that opposes it, which is precisely what $C_V > 0$ and $\kappa_T > 0$ encode. Equilibrium is thus not a static concept but a variational one — the summit of the entropy landscape — and its stability is the curvature at that summit.

History

1865

— Clausius formulates entropy and the tendency toward maximum entropy in an isolated system.

1875

— Gibbs derives the equilibrium and stability conditions from entropy maximization.

1876

— Gibbs states the equalities of T, P, μ as equilibrium conditions and the phase rule.

1884

— Le Chatelier states his principle, equivalent to the stability conditions.

1929

— Born and others clarify the convexity/concavity structure of the potentials.

1960

— Callen recasts thermodynamics on the entropy-maximum postulate and stability analysis.

Going further

References. Callen, Thermodynamics and an Introduction to Thermostatistics, 2nd ed., chs. 2–3 (equilibrium conditions) and chs. 8–9 (stability) · Pathria & Beale, Statistical Mechanics, 3rd ed., ch. 1 · Reif, Fundamentals of Statistical and Thermal Physics, ch. 3 · Kardar, Statistical Physics of Particles, ch. 1 · Landau & Lifshitz, Statistical Physics Part 1, §§21–24 (stability) · Gibbs, On the Equilibrium of Heterogeneous Substances (1875–78).

Provenance. The equilibrium conditions ( $T$ , $P$ , $\mu$ equal, from maximizing total entropy under exchange) and the stability conditions ( $C_V > 0$ , $\kappa_T > 0$ , negative-definite Hessian) are standard (Callen chs. 2–3, 8–9; Reif ch. 3). A verifier should confirm the first-order conditions $1/T_1 = 1/T_2$ , $P_1/T_1 = P_2/T_2$ , and the second-order signs $S_{UU} < 0$ , $S_{VV} < 0$ , $S_{UU}S_{VV} - S_{UV}^2 > 0$ for the Sackur–Tetrode entropy. Numerical instances: equal blocks split $U_1 = U_2 = 1.5$ ; unequal ( $a_1 = 2a_2$ ) split $U_1 = 2$ , $U_2 = 1$ (energy proportional to $N$ , equal $T$ ). All in verify_aq-sm-equilibrium-conditions-subsystems.py.

Builds on: Temperature, Pressure, and Chemical Potential as Entropy Derivatives · Second Law of Thermodynamics and Entropy · The Microcanonical Ensemble and the Postulate of Equal a Priori Probabilities · Thermodynamic Potentials and the Fundamental Relation. Leads to: Thermodynamic Stability and Convexity Conditions · Canonical Ensemble · Equivalence of Ensembles.

Why does maximizing entropy give the equilibrium, rather than minimizing energy?

It depends on what is held fixed, and for an isolated system it is entropy maximization. An isolated composite system has fixed total energy, volume, and particle number — it cannot lower its energy (energy is conserved), so the only thing that can change is how the fixed quantities are distributed among its parts, and the second law says that distribution evolves to maximize total entropy. That is the entropy-maximum principle. The energy-minimum principle is the equivalent statement in the energy representation, where you fix the entropy instead and minimize energy — appropriate for a system in contact with an entropy reservoir. And for systems in contact with a thermal (or thermal+pressure) bath, the equilibrium minimizes the appropriate free energy ( $F$ or $G$ ), which is entropy maximization of system-plus-bath in disguise. So there is one underlying principle (maximize total entropy of the isolated whole), and "minimize energy" or "minimize free energy" are its forms under different constraints. For an isolated system, entropy maximization is the direct statement.

Why does each type of exchange (energy, volume, particles) equalize a different intensive variable?

Because each intensive variable is defined as the entropy slope with respect to the corresponding conserved quantity. Temperature is $1/T = \partial S/\partial U$ , so equalizing $\partial S/\partial U$ (which is what energy exchange does at the entropy maximum) equalizes $1/T$ and hence $T$ . Pressure enters through $P/T = \partial S/\partial V$ , so volume exchange equalizes $P/T$ (and thus $P$ at equal $T$ ). Chemical potential enters through $-\mu/T = \partial S/\partial N$ , so particle exchange equalizes $\mu$ . The pattern is exact and universal: for every conserved quantity a system can exchange, there is a conjugate intensive variable (its entropy slope), and the entropy-maximum condition equalizes exactly that slope. This is why the three equilibria look so similar — they are the same variational condition applied to three different conserved quantities, and it generalizes to any additional conserved quantity (magnetization, charge) with its own conjugate field.

What physically goes wrong if the heat capacity C_V were negative?

A negative heat capacity means adding energy to a system lowers its temperature — and this leads to a runaway instability for a system in thermal contact with others. Suppose a small region fluctuates to slightly higher energy. With $C_V < 0$ it would become colder, so heat would flow into it from its warmer surroundings, making it even more energetic and even colder, and the fluctuation would grow without bound — the homogeneous state is unstable. This is exactly why the stability condition $C_V > 0$ (equivalently $\partial^2 S/\partial U^2 < 0$ ) is required for a stable equilibrium: it ensures a hot fluctuation loses heat and returns to equilibrium. Remarkably, systems with long-range interactions — most famously self-gravitating ones like star clusters — genuinely can have negative heat capacity (a star that loses energy gets hotter), and precisely because of this they do not have stable homogeneous equilibria and require a completely different, non-extensive thermodynamic treatment. Negative $C_V$ is the signature of a system that violates the ordinary stability conditions.

How does the stability analysis connect to phase transitions?

Intimately — phase transitions are exactly where stability fails. As long as the entropy is strictly concave (the Hessian negative-definite, $C_V > 0$ , $\kappa_T > 0$ ), a single homogeneous phase is stable. But if you compress a gas below its critical temperature, you eventually enter a region where the mechanical stability condition fails: $\kappa_T < 0$ , meaning the isotherm has a stretch where pressure would increase as you expand ( $\partial P/\partial V > 0$ ). No homogeneous phase can exist there — the system spontaneously separates into a dense (liquid) and dilute (vapor) phase. The boundary of this unstable region, where the Hessian determinant $S_{UU}S_{VV} - S_{UV}^2$ first vanishes, is the spinodal curve. The broader coexistence region (the binodal) is set by the equal-chemical-potential condition ( $\mu_{\text{liq}} = \mu_{\text{vap}}$ ) via the common-tangent/Maxwell construction. So the equilibrium conditions ( $\mu$ equal) locate the coexistence, and the stability conditions (Hessian sign) locate the instability — together they are the complete framework of first-order phase transitions.

Do these conditions still hold if the two subsystems are very different sizes?

Yes — the equilibrium conditions ( $T$ , $P$ , $\mu$ equal) hold regardless of the relative sizes of the subsystems, because they come from equalizing intensive (size-independent) quantities. What changes with size is how much of the conserved quantity each subsystem ends up with, not the equilibrium condition itself. If one subsystem is enormous compared to the other (a "reservoir"), it barely changes its own intensive variables while imposing them on the small system: the small system simply takes on the reservoir's temperature, pressure, and chemical potential. This is exactly the setup that defines the canonical ensemble (large thermal reservoir fixes $T$ ) and the grand canonical ensemble (large reservoir fixes $T$ and $\mu$ ). So the size-asymmetric limit is not an exception — it is the physically most important case, where the large subsystem acts as a bath that clamps the intensive variables of the small one. The equilibrium conditions of this article are precisely what justify treating a reservoir as fixing $T$ , $P$ , or $\mu$ .

Why must the equilibrium be a maximum of entropy specifically — could it be a minimum or a saddle?

The direction of the second law forces it to be a maximum. The second law says the entropy of an isolated system increases until it can increase no more, so equilibrium is where entropy stops rising — the top of the entropy landscape, a maximum. A minimum would be the opposite: the system would spontaneously move away from it (entropy would rise by leaving), so a minimum is not an equilibrium at all but a point of maximal instability. A saddle is intermediate — stable against some perturbations, unstable against others — and any fluctuation along the unstable direction grows, so a saddle is also not a true equilibrium. This is exactly why the first-order condition (zero gradient, which just says "extremum") is not sufficient: you must also check the second-order condition (negative-definite Hessian) to confirm you have a maximum rather than a saddle or minimum. Physically: at a genuine maximum, every possible fluctuation lowers the entropy and is therefore suppressed by the second law, so the state is stable. That is what "stable equilibrium" means, and it is why the concavity conditions ( $C_V > 0$ , $\kappa_T > 0$ ) are indispensable — they are the mathematical guarantee that you are at the summit, not on a slope or in a valley.